No title

COMMENT pubs.acs.org/est The Calculus of Unsustainability S ustainability is a squishy term—and diﬃcult to operation...

Author: Jerald L. Shnoor

8 downloads 2414 Views 109MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

COMMENT pubs.acs.org/est

The Calculus of Unsustainability

S

ustainability is a squishy term—and diﬃcult to operationalize for public policy. As I give talks to civic groups and students, I am often asked, “Exactly what is it that is not sustainable?” Perhaps the underlying problem is the public’s lack of understanding about material stocks and ﬂows, integrals and derivatives, i.e., the calculus of unsustainability. In the U.S., we simply do not seem to get it. On October 31 of this year, according to the United Nations, the global population reached 7 billion (http://www.unfpa.org/ public/home/news/pid/7597). What a milestone! When I was born there were 2.5 billion people on earth, so we’ve almost trebled population in one (old) editor’s lifetime. But emissions of greenhouse gases have increased even more, greater than 6-fold since that time. That rate of change in emissions (the derivative) makes for a huge accumulation of greenhouse gases in the atmosphere as shown in the ﬁgure (Source: Scripps-NOAA, 2010). Clearly, such accumulation (the integral—area under the curve) of a radiatively important trace gas is not sustainable over the long run. It is as simple as that. The good news is that the rate of population change is decreasing, and demographers tell us we may reach a plateau around 9 billion people sometime in the middle of the 21st century. The bad news is that the plateau is likely to not be sustainable either—it depends on how we use our resources. Resource utilization (land, labor, capital) deﬁnes our economics. My favorite economist is Herman Daly, of ecological economics fame at the University of Maryland. Daly is the one who wrote A Steady State Economy (http:// www.sd-commission.org.uk/publications.php?id=775) to which no one seems to have paid much policy attention. If economics really is the “dismal science”, then environmental science must be “depressing science on a suicide watch”. Contributing to our depression is the exploitation of natural capital like nonrenewable resources. Deaccumulation (an integral of a declining supply curve) can be perilous and also UNSUSTAINABLE. The static reserve index, that is how many years remain of proven reserves at current global usage rates, is only about 64 for petroleum, 13 for indium in LCD displays, 30 for antimony in drugs, and 29 years for silver (http:// en.wikipedia.org/wiki/Oil_reserves;http://img.labnol.org/ﬁles/ how_many_years.jpg). But that does not tell the whole story because nonrenewable resource usage is not static, rather the consumption rate is usually increasing exponentially. Fortunately, we are ﬁnding more “proven reserves” and recycling, reusing, and substituting for them. But can a mining company ever be sustainable? (See answer below). Erdmann and Graedel published a fascinating review article in ES&T arguing that static reserves are not the best way to view unsustainability of metals. Rather criticality, which considers both vulnerability and risk of supplies, is more meaningful. They show that the Rare Earth Elements (REE), Platinum Group Elements (PGE), niobium, cobalt, scandium, tungsten, gallium, and antimony are probably the critical ones we should be worrying about (Erdmann and Graedel, Environ. Sci. Technol. 2011, 45, 7620 7630, dx.doi.org/10.1021/es200563g). So many precious r 2011 American Chemical Society

metals are utilized in the exploding electronic consumer products of today. For example, the computer chip industry consumed about 11 metals in 1980 and now uses 60 elements today— almost 2/3 of the natural periodic table! (Schmitz and Graedel, Environment360, 2010, http://e360.yale.edu/content/feature. msp?id=2266) Likewise, the accumulation of greenhouse gases in the atmosphere is unsustainable in the absence of substitution for fossil energy. But U.S. Presidential candidate and Texas Governor Rick Perry said in a recent debate, “The science is not settled” on human-induced global warming. It reminds me of the principal error in The Skeptical Environmentalist—where Lomborg fails to appreciate that the accumulation of greenhouse gases (and climate change) will continue unimpeded for centuries in the absence of mitigation toward preindustrial levels (The Skeptical Environmentalist, 2001, Cambridge University Press). Perry and Lomborg fail the calculus test. We have warmed the earth so much already (in the early stages) that the ocean is absorbing an immense amount of heat. If every person on earth (7 billion) ran 40 industrial strength hair dryers of 1400 W each, it would be equivalent to the warming we add to the planet each year (http://www.climatestorytellers.org/stories/james-hansen-makikosato-perceptions-of-climate-change/). The U.S. could use some more math—the calculus of unsustainability. Our future depends upon it. (Answer: Yes, a mining company can be sustainable if they continuously discover new ways to recycle and/or substitute for nonrenewable minerals and avoid nonrecoverable, dissipative uses. Europe is approaching 100% recycle of lead. When that happens, it obviates the need for more lead mining and the company evolves into a recycling enterprise rather than an extractive industry.) Jerald L. Schnoor Editor-in-Chief

’ AUTHOR INFORMATION Corresponding Author

[email protected].

Published: December 05, 2011 10289

dx.doi.org/10.1021/es2038118 | Environ. Sci. Technol. 2011, 45, 10289–10289

LETTER pubs.acs.org/est

Comment on “Do Some NOx Emissions Have Negative Environmental Damages? Evidence and Implications for Policy” n their Viewpoint titled “Do Some NOx Emissions Have Negative Environmental Damages? Evidence and Implications for Policy”,3 the authors represent as novel the ﬁnding in two studies that marginal reductions in NOx emissions from particular sources in certain areas may increase PM2.5 levels and yield negative health impacts. The authors then entreat the EPA to “commission a review of these studies”, to be evaluated by the EPA Science Advisory Board and submitted as a report to Congress. We argue that the Viewpoint in general, and these recommendations in particular, incompletely consider atmospheric science and do not account for current best practices in air quality management. PM2.5 formation is governed by complex nonlinear chemistry, and so the impact of reducing precursor emissions diﬀer across receptors for three reasons: (1) levels of SO2, NOx, NH3; (2) meteorology (particularly temperature and relative humidity); and, (3) availability of ozone and related oxidants (e.g., OH, H2O2, etc.). Transient spatial and temporal increases in PM2.5 from decreases in NOx emissions are well documented in literature.1,4 Under certain atmospheric conditions, a reduction in NOx (particularly near-surface mobile NOx emissions under cold climate conditions) sometimes leads to increases in ozone and result in localized increases in PM2.5 concentrations, though this depends on baseline levels of NOx and VOC.1,4,5 For this reason, reductions of NOx within NOx-rich urban city centers (e.g., Chicago) may lead to localized increases in PM2.5 levels and may increase population exposure to PM2.5, thus resulting in the disbeneﬁt reported in Fann et al.2 Photochemical models used by EPA and State air agencies account for these nonlinear eﬀects. While there is clear policy relevance to the negative beneﬁt per ton estimates reported in Fann et al.2 and elsewhere, this eﬀect should be considered within the broader context of air quality management policy. In its 2004 report on air quality management in the U.S., the National Academies noted that air quality

I

attainment strategies should balance emission reductions across multiple pollutant precursors and spatial scales6. To this end, beneﬁt per ton estimates may inform the development of speciﬁc control strategies that aim to maximize human health beneﬁts while achieving air quality targets. Indeed, EPA recently completed a pilot project for the city of Detroit, in which it employed a risk-based and multiple pollutant approach to air quality management.2,7 This strategy weighed local and regional emission reductions among a variety of sources and across several PM2.5 and ozone precursors, and toxic air pollutants, to design an implementation plan that maximized human health beneﬁts and achieved a more equitable distribution of risk. Rather than reacting to incomplete evidence, approaches to air quality management such as these illustrate an approach to attainment plans that accounts for all of the best available science. Neal L. Fann* Air Quality Analysis Division, U.S. Environmental Protection Agency, Mail Drop C539-07 109 T.W. Alexander Drive Durham, North Carolina 27711, United States

(2) Fann, N.; Fulcher, C. M.; Hubbell, B. J. The inﬂuence of location, source, and emission type in estimates of the human health beneﬁts of reducing a ton of air pollution. Air Qual. Atmos. Health 2009, 2 (3), 169–176. (3) Fraas, A.; Lutter, R. Do some NOx emissions have negative environmental damages? Evidence and implications for policy. Environ. Sci. Technol. 2011, 45 (18), 7613– 7614. (4) Myslieiec, M. J.; Kleeman, M. J. Source apportionment of secondary airborne particulate matter in a polluted atmosphere. Envrion Sci. Technol. 2002, 36 (24), 5376–5384. (5) Pun, B. K.; Seigneur, C.; Bailey, E. M.; Gautney, L. L.; Douglas, S. G.; Haney, J. L.; Kumar, N. Response of atmospheric particulate matter to changes in precursor emissions: A comparison of three air quality models. Envrion Sci. Technol. 2008, 42 (3), 831–837. (6) U.S. National Research Council (NRC). Air Quality Management in the United States; National Academies: Washington, DC, 2004. (7) Wesson, K.; Fann, N.; Morris, M.; Fox, T.; Hubbell, B. A multipollutant, risk-based approach to air quality management: case study for Detroit. Atmos. Pollut. Res. 2010, 1, 296–304. (8) Fann, N.; Roman, H. A.; Fulcher, C. M.; Gentile, M. A.; Hubbell, B. J.; Wesson, K.; Levy, J. I. Maximizing health beneﬁts and minimizing inequality: incorporating local-scale data in the design and evaluation of air quality policies. Risk Analysis 2011, 6 (31), 908–922.

Dr. Sharon B. Phillips Air Quality Analysis Division, U.S. Environmental Protection Agency

Dr. Carey Jang Air Quality Analysis Division, U.S. Environmental Protection Agency

Dr. Farhan H. Akhtar Health and Environmental Impacts Division, U.S. Environmental Protection Agency

’ AUTHOR INFORMATION Corresponding Author

*Phone: (919) 541-0209; fax: (919) 5415315; e-mail: [email protected].

’ REFERENCES (1) Ansari, A. S.; Pandis, S. P. Response of inorganic PM to precursor concentrations. Envrion Sci. Technol. 1998, 32 (18), 2706–2714.

This article not subject to U.S. Copyright. Published 2011 by the American Chemical Society

10290

Received: October 19, 2011 Accepted: October 24, 2011 Published: November 11, 2011

dx.doi.org/10.1021/es203710m | Environ. Sci. Technol. 2011, 45, 10290–10290

LETTER pubs.acs.org/est

Speculation on the Origin of Monochloro-Nonabromodiphenyl Ethers. Letter to the Editor regarding Comment on “Identification of Monochloro-Nonabromodiphenyl Ethers in the Air and Soil Samples from South China” a Guardia et al.1 commented on the detection of three nonabromochlorodiphenyl ethers (NBCDEs) in air and soil samples from Guangzhou China and at an e-waste recycling area,2 and speculated in this and a prior publication3 that these compounds were impurities in Albemarle Corporation’s commercial decabromodiphenyl ether (DecaBDE) product. Albemarle would like to set the record straight. Yes, Albemarle ﬁled the patent application described in La Guardia et al. (2011). Albemarle has an active research and development program, ﬁles many patent applications as a result of this active research and development, and holds numerous patents. A substantial number of these patents and patent applications include processes that are never commercialized, including the process described in the cited patent application.4 In fact, Albemarle abandoned the cited patent application some time ago, and a simple check of the public patent databases would have revealed this. Our manufacturing process for DecaBDE does not use bromine chloride or mixtures of bromine and chlorine. Albemarle has never commercialized a bromochlorodiphenyl ether and has no intention to do so, nor do we do manufacture a “decahalodiphenyl oxide” product as indicated in the authors’ 2010 publication on page 4663. Albemarle does not manufacture DecaBDE in China. To our knowledge, the authors did not contact Albemarle to ascertain whether their speculation was based on fact. Rather, La Guardia et al. (2010, 2011) assumed a patent equated to a commercial product, a commercial product’s market introduction was that of the patent date, and a business presence in a country equaled manufacture of a product in that country. Those assumptions are incorrect. We recommend the authors, and the editors of Environmental Science & Technology, do a better job of fact checking prior to publication. Erroneous publications such as these divert attention and research dollars from meaningful pursuits.

Marcia L. Hardy,†,* Niomi L. Krystowczyk,† Steve W. LeVan,‡ and David W. Clary§

L

†

Health, Safety and Environment, Albemarle Corporation, 451 Florida Street, Baton Rouge, Louisiana

‡

Advocacy, Albemarle Corporation, 451 Florida Street, Baton Rouge, Louisiana

§

Chief Sustainability Oﬃcer, Albemarle Corporation, 451 Florida Street, Baton Rouge, Louisiana

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ REFERENCES (1) La Guardia, M.; Hale, R.; Harvey, E.; Harvey, E.; Chen, D. Comment on “Identiﬁcation of monochloro-nonabromodiphenyl ethers in the air and soil samples from South China. Environ. Sci. Technol. 2011, 45, 6707–6706. (2) Yu, Z.; Zheng, K.; Ren, G.; Wang, D.; Ma, S.; Peng, P.; Wu, M.; Sheng, G.; Fui, J. Identiﬁcation of monochloro-nonabromodiphenyl ethers in the air and soil samples from South China. Environ. Sci. Technol. 2011, 45 2619–2625. (3) La Guardia, M.; Hale, R.; Harvey, E.; Chen, D. Flame-retardants and other organohalogens detected in sewage sludge by electron capture negative ion mass spectrometry. Environ. Sci. Technol. 2010, 44, 4658–4664. (4) WIPO, WO/2008/027780, Preparation of decahalodiphenyl oxide. http://www. wipo.int/pctdb/en/wo.jsp?WO=2008027780 (as cited by La Guardia et al. 2011).

r 2011 American Chemical Society

Received: October 28, 2011 Accepted: October 31, 2011 Published: November 15, 2011 10291

dx.doi.org/10.1021/es203846d | Environ. Sci. Technol. 2011, 45, 10291–10291

VIEWPOINT pubs.acs.org/est

China Needs Forest Management Rather Than Reforestation for Carbon Sequestration Guanglei Gao, Guodong Ding,* Haiyan Wang, Yintong Zang, and Wenjun Liang Key Laboratory of Soil and Water Conservation & Desertiﬁcation Combating, Ministry of Education (Beijing Forestry University), Beijing 100083, P. R. China

s a “close-to-nature” approach for carbon sinks, planted forests (aﬀorestation and reforestation) have a priority to combat climate change in China. During the past decade, the Chinese government invested billions of dollars in a large-scale tree-planting (e.g., the Six Key Forestry Programs). At the U. N. Climate Summit (New York, 2009), Hu, China’s President, also committed that China should endeavor to increase forest coverage by 40 million ha to energetically increase forest carbon sinks by 2020 from the 2005 levels. Obviously, it is aﬀorestation that makes remarkable contributions to carbon sinks in China.1 However, excessive and monoculture aﬀorestation to implement China’s carbon sequestration programs may be ineﬃcient and cause unintended disastrous environmental consequences, especially in arid and semiarid regions.2 In fact, forest’s functions in carbon stock increasing are addressed by two Kyoto Protocol activities: aﬀorestation/reforestation and forest management. Aﬀorestation/reforestation has a top priority for carbon sink in China, whereas forest management has been almost thoroughly ignored. Further, from 2000 to 2010, although roughly 15 million ha of plantation were planted, which prompted the total forest coverage and forest stock being increased from 16.55% to 20.36% and 11.27 to 13.36 billion m3, respectively; China’s average forest stock, forest biomass carbon and forest carbon density still remain far less than international

A

r 2011 American Chemical Society

level. For example, China’s average woody forest volume was 85.88 m3/ha accounting for only 78% of the world, the plantation was even lower; meanwhile, the mean forest carbon stock in biomass (40.4 t/ha) was much lower compared to the global average of 71.6 t/ha. The gaps indicate the poor quality of China’s forest, which however implies the huge potential for carbon sink in the activities of forest management. Shao et al.1 estimated that if China followed forest management activities of the U.S., the increasing forest productivity would boost China’s forest carbon sequestration from 96 to 152 Tg C/yr without requiring additional forestland area (Table 1). In addition, if the existing carbon stock of forest biomass can be increased by 10% between 2010 and 2020, the cumulated carbon sinks will be much larger than 683 Mt of the Chinese oﬃcial aﬀorestation target; moreover, this amount is also much less than 65% of the latest international level. Scholars have been questioning that large-scale aﬀorestation eﬀorts in China have failed in the environmental restoration and carbon sequestration because of the negative chain ecological problems. For example, aﬀorestation with unmatched species in aﬀorestation regions may damage the local water balances, even exhaust limited groundwater resulting in trees death or dying; in arid or semiarid regions, it will ﬁnally lead to an enlarging desertiﬁcation.3 In addition, monoculture plantations or exotic species can also reduce biodiversity when it replaces natural ecosystems. Compared with this above, forest management may have many positive impacts on environmental recovery. Forest management emphasizes natural approaches instead of monoculture tree planting for environmental restoration. Aﬀorestation can be replaced by the native vegetation recovery. In northern China’s arid and semiarid regions, it is much more reasonable that small halophytic shrubs, savanna and steppe vegetation, and some herbaceous plants grow on aeolian sands and other land vulnerable to wind erosion.4 Meanwhile, a better mixture of plant species and appropriate measure of human activities will make a promotion to increasing biodiversity. In the socioeconomic aspect, with a cumulative aﬀorestation cost and decreasing suitable forestland, forest management can reduce the investment, as well as provide excessive jobs in a large area and promote rural development. Hence, forest management rather than the large-scale aﬀorestation meets the complex requirements of environmental restoration; and it is an eﬃcient approach for forestry carbon Received: October 12, 2011 Accepted: November 2, 2011 Published: November 16, 2011 10292

dx.doi.org/10.1021/es203897f | Environ. Sci. Technol. 2011, 45, 10292–10293

Environmental Science & Technology

VIEWPOINT

Table 1. The U.S.-China Comparisons in Forest Area, Carbon Stock in Forest Biomass and Their Annual Change5 forest area 2010 106 ha

carbon stock in living forest biomass

annual change/103 ha 1990 2000

1990

1995

2000

2010

106 t

2000 2010

Annual change/103 t 1990 2000

2000 2010

U.S.

304

386

383

16 951

17 998

18 631

19 308

105

131

China

207

1986

2986

4414

5295

5802

6203

88

91

In the last decade, both the U.S. and China implemented forestry carbon sequestration programs to reduce the carbon print. However, compared to China’s large-scale aﬀorestation, the U.S. has a much more carbon sinks with little additional forest area because of its emphasis on carbon sequestration activities of forest management. sequestration. The preservation and restoration of existing ecosystem should be the primary goal of carbon sequestration, and the destruction of these ecosystems by large-scale aﬀorestation may be counterproductive.2 However, in fact, it is the government who has authority over policy making. It is hard for government oﬃcials facing urgent tasks to give up the short-time beneﬁts which can show their merits and achievements. Thus, in China, although an increasing number of people have realized that monoculture aﬀorestation is not appropriate, government attitudes still changed slowly. Based on deep examinations, oﬃcials, scholars, managers, and citizens should have common understandings of China’s environmental restoration strategy. In a word, forest management has larger potential carbon sink ability than large scale aﬀorestation, and can avoid the potential large risk to ecosystem health. It is suggested that forest management should be a sustainable and sagacious choice for China’s forestry carbon sequestration.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 86-10-6233-7777; e-mail: [email protected].

’ ACKNOWLEDGMENT This work was supported by the Special Fund for Forestry Scientiﬁc Research in the Public Interest, State Forestry Administration of P. R. China (200804022A). We thank Prof. Dr. Xiangdong Lei (Beijing, China) for editing an early version of this paper. ’ REFERENCES (1) Shao, G. F.; Dai, L. M.; Dukes, J. S.; Jackson, R. B.; Tang, L. N.; Zhao, J. Z. Increasing forest carbon sequestration through cooperation and shared strategies between China and the United States. Environ. Sci. Technol. 2011, 45, 2033–2034. (2) Wang, Y.; Cao, S. Carbon sequestration may have negative impacts on ecosystem health. Environ. Sci. Technol. 2011, 45, 1759–1760. (3) Cao, S. Why large-scale aﬀorestation eﬀorts in China have failed to solve the desertiﬁcation problem. Environ. Sci. Technol. 2008, 42, 1826–1831. (4) Wang, X.; Chen, F.; Hasi, E.; Li, J. Desertiﬁcation in China: An assessment. Earth-Sci Rev 2008, 88, 188–206. (5) State of the World’s Forest 2011; Food and Agriculture Organization of the United Nations: Rome, 2011.

10293

dx.doi.org/10.1021/es203897f |Environ. Sci. Technol. 2011, 45, 10292–10293

VIEWPOINT pubs.acs.org/est

Controlling Air Pollution from Coal Power Plants in China: Incremental Change or a Great Leap Forward Zhang Bing,† Elizabeth Wilson,‡ and Bi Jun*,† † ‡

State Key Laboratory of Pollution Control & Resource Reuse, School of Environment, Nanjing University, Nanjing, 210046, P. R. China Humphrey School of Public Aﬀairs, University of Minnesota, MN, United States

C

hina is in the midst of the largest and fastest period of economic growth in global history. While this growth at 9 10% GDP per year has lifted hundreds of millions of people from poverty, the all out prioritization of economic prosperity has created serious air and water pollution that are aﬀecting the health of the local population, and also have increasing global climate impacts. A World Health Organization (WHO) report estimated that diseases triggered by indoor and outdoor air pollution kill 656 000 Chinese citizens each year.1 In addition to local criteria air pollutants like sulfur dioxide (SO2) and nitrogen oxide (NOx), China is now the largest emitter of greenhouse gases (GHGs) in the world. Strategies and policies to control air pollutants have been on the books for decades, but shifting local government focus away from the sole prioritization of economic growth and including environmental protection has proven very challenging. China has been implementing the single pollutant control strategy and focus on short-term main pollution control target. From the early to take soot as the main control object to total pollution control of SO2 since 9th ﬁve-year plan. After years of eﬀort, pollution control achieved some success. Thermal power plants have been eﬀectively controlled dust emission since 2000, while sulfur dioxide also reached its turning point of decline in 2007 and reduced 14.3% of SO2 during the 11th ﬁve-year plan period (see Figure 1). Although control of sulfur dioxide has made great progress, other kinds of air pollution and CO2 increased a lot in r 2011 American Chemical Society

the past decade. The NOx, mercury, and CO2 from thermal power sector in 2010 were 1.51 times that of 2005. The rapid development of China will not stop. In the next 10 years, China’s GDP will increase to 4 times that of 2010. Without new pollution control policies, the NOx, mercury, and CO2 from thermal power industry sector will be 1.28 times that of 2010. Since 12th ﬁveyear plan, China starts to control NOx and target to reduce 10% of NOx. In addition, the international pressure on carbon reduction push China to promised to reduce its carbon intensity— the amount of CO2 it emits for each dollar of economic output— by 45%. There are signiﬁcant advantages of a pollutant by pollutant approach, which allows operators and designers to target system design and hone operational features of new technologies. The single pollution control target will induce enterprises invest on end-of-pipe pollution control. In such condition, technology innovation and multipollution control technologies will be less cost-eﬀective than single pollution control. During the 11th ﬁveyear plan period (2005 2010), China has installed 500000 MW desulfurization that 86% units have ﬂue gas desulfurization (FGD) systems. However, the end-of-pipe pollutant controls will increase other pollutions, for example, controlling SO2 increases coal use, which increases associated CO2, mercury, and NOx. The used 600 MW unit in China should use average 1.58% auxiliary electricity for desulfurization. That is the removal of 1 kg SO2 will bring 6.1 kg CO2 and 0.015 kg NOx. In addition, the signiﬁcant reductions in SO2 emissions will reduce the cooling impact of reﬂective aerosols.2 Thus, the single pollution control strategy will press the bottle gourd, but played a gourd ladle. In addition, with the increasing pollution control types, resulting in continued expansion of pollutant puriﬁcation equipment, regardless of investment or operating costs, the complexity of puriﬁcation systems are facing great diﬃculties. Meanwhile, since air pollutants and GHGs often derive from the same sources—fossil fuel combustions—there is an opportunity to address the two problems simultaneously. Control technologies that are capable of simultaneously reducing emissions of multiple pollutants may also oﬀer the potential to achieve this at lower cost and reduced footprint when compared to conventional controls. Taking other air pollution and GHGs’ environmental impact into consideration, the total costs of coalﬁred plants will be not lower than nuclear power, hydropower, or wind power. Thus, one of most important implication of Received: October 24, 2011 Accepted: October 31, 2011 Published: November 23, 2011 10294

dx.doi.org/10.1021/es203766m | Environ. Sci. Technol. 2011, 45, 10294–10295

Environmental Science & Technology

VIEWPOINT

Figure 1. SO2 discharge in China and emission discharge from thermal power sector (NOx, mercury and CO2) from 1995 2020. The historical emissions discharge is from China’s Year Book. The trend of emission discharge is estimated based on IEA’s report.

co-control of air pollution and GHGs will make investors rethink about the cost-beneﬁt of diﬀerent energy sources. Co-control of air pollution and GHGs will reduce the total social cost and make more investment on new energy technology. In addition, cocontrol strategies will also make technologies which can improve energy eﬃciency more attractive, such as clean coal technology, etc. Thus, the co-control strategies by using new energy or technologies will be more cost-eﬀective than single pollution control strategy. Both the U.S. and Europe experience has shown that an integrated pollutants co-control strategy will be more cost-eﬀective than a single pollution control strategies.3 On the other hand, technology enables co-control of a variety of possible contaminants, such as catalytic reduction (SCR) or selective noncatalytic reduction (SNCR) technology to achieve combined desulfurization and denitriﬁcation, and the pulse corona plasma (PPCP) can removal of NOx and SO2 and PM together, electro-catalytic oxidation (ECO) technology can eﬀectively reduce SO2, NOx, PM2.5, and mercury emissions together.4 If taking into account the reduction of carbon dioxide, ﬂue gas desulphurization and denitriﬁcation with ultrasupercritical power generation technology (SC/USC + FGD + SCR), circulating ﬂuidized bed boiler (CFBC), pressurized ﬂuidized bed combined cycle (PFBC CC), integrated gasiﬁcation combined cycle (IGCC) will achieve higher desulfurization and denitriﬁcation rate, as well as provide coal-ﬁred power plants a more feasible way to treat CO2 and mercury. Therefore, cocontrol strategies can eﬀectively improve the eﬃciency of air pollution and GHGs reduction in a long-term perspective.5 It is easier to imagine Chinese policies continuing to pursue power plant eﬃciency and a continued push to fastrack P3 (SO2 + NOx + mercury) controls throughout the country, as has already been piloted in some provinces like Shanxi. However, without stricter environmental regulations and multipollutant control strategy, it appears unlikely that the P4 pathway (SO2 + NOx + mercury + CO2) will be the road taken. While economists in the U.S. have been proﬀering similar advice for decades, China may have a better chance of actually implementing the longer term planning and stable perspectives to guide power plant construction policies.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ REFERENCES (1) Platt, K. H., Chinese air pollution deadliest in world. National Geographic News 2007, http://news.nationalgeographic.com/news/ 2007/07/070709-china-pollution.html. (2) Zhao, Y.; McElroy, M. B.; Xing, J.; Duan, L.; Nielsen, C. P.; Lei, Y.; Hao, J. Multiple eﬀects and uncertainties of emission control policies in China: Implications for public health, soil acidiﬁcation, and global temperature. Sci. Total Environ. 2011, 409, 5177–5187. (3) McCarthy, J. E.; Parker, L. B. Costs and beneﬁts of clear skies: EPA’s analysis of multi-pollutant clean air bills, CRS Report for Congress 2005, http://www.policyarchive.org/handle/10207/bitstreams/2639.pdf. (4) Boyl, P. D., Multi-pollutant control technology for coal-ﬁred power plant. In Clean Coal and Power Conference, Washington, DC 2005. (5) Echeverri, D. P.; Fischbeck, P.; Krirgler, E. Economic and environmental costs of regulatory uncertainty for coal-ﬁred power plants. Environ. Sci. Technol. 2009, 43, 578–584.

10295

dx.doi.org/10.1021/es203766m |Environ. Sci. Technol. 2011, 45, 10294–10295

VIEWPOINT pubs.acs.org/est

Attraction of Night-Migrating Birds to Light-Blue Structures Causes Mass Bird Deaths Zhijiang Wang,† Aijun Lin,‡ Qipeng Yuan,‡ Wenbin Zhou,§ Wenbo Zhang,|| and X. Jin Yang‡,* †

Lushunkou Bureau of Agriculture, Forestry and Irrigation, Dalian, China; Beijing University of Chemical Technology, Beijing, China; § Nanchang University, Nanchang, China; Eco-humanity Alliance, Beijing, China.

)

‡

M

ass bird deaths have been occurring in numerous regions across the globe and have been raising signiﬁcant public health and ecological security concerns. Avian inﬂuenza virus has been identiﬁed as one of the major causes of mass migratory bird deaths1 and communication towers and guy wires are proving to be a deadly hazard to birds, in particular migratory birds.2 It is estimated that between 5 and 50 million migrating birds are killed in the U.S. each year by colliding and crashing with communication towers and guy wires during their night migration.2 However, the majority of mass bird deaths have been mysterious and unexplained. In the early morning of August 30, 2006, about 70 birds were found dead beside three buildings in the Port of Laotieshan, Dalian (121°440 -121°490 E, 39°010 -39°040 N), China. In the following days the tragedy of about a hundred birds death per day continued. The incidents caused signiﬁcant scares to the public as the bird ﬂu outbreak was epidemic across the world. The transmission of highly pathogenic H5N1 inﬂuenza A viruses to humans was proved3 and 12 cases of human infections had been reported in mainland China by the time of the incident. The dead birds were collected and sent to the laboratory. The laboratory examination ruled out the possibility of any disease r 2011 American Chemical Society

infections or food poisoning. Screening for bird ﬂu infections in the poultry industry around the area showed no signs of infections. It was found that the birds were all night-migrating birds and the cause of death was due to severe injuries on the head of the birds. Inspections of the building walls showed residues of bloods and feathers. We therefore concluded that the birds committed a mass “suicide” by crashing into the buildings. Building A experienced the majority of bird kills and is a size of 10 m length 6 m wide 4.5 m high, the smallest and lowest of the three incidental buildings. A 12 m high and 120 m long bridge is located only 6 m away on the northeast of Building A (Figure 1) and a ﬁve-ﬂoor building is just 50 m on the south of Building A. The bird crashes on Building A occurred on the south wall and it was surprising that none of the birds collided with the bridge and the ﬁve-ﬂoor building. Therefore, it was unlikely that the birds were accidentally colliding with the three buildings due to visibility reasons. The three buildings having bird deaths were all in light-blue color (see the insert of Figure 1) and the bridge and the ﬁve-ﬂoor building were concrete gray color. Around the three buildings there are 10 night lighting posts (25 m high). At night the area was very brightly lit and there were heavy mists on the days when the migratory bird deaths happened. Therefore, it was speculated that the birds ﬂew at lower altitude in mist, were attracted by the light-blue buildings and were disorientated. While there is currently no empirical evidence to support this hypothesis of structures color attraction, the eﬀect of the color of artiﬁcial night lighting on the attraction of migrating birds has been well documented.4,5 The duration and color of lighting at night are critical to whether birds are or are not attracted to lights. Solid white lights are more attractive to birds than colored or ﬂashing lights, but solid or pulsating red lights attract night-migrating birds at a much higher rate than white strobe lights. It was reported that the two television towers near Awendaw, South Carolina had substantial bird kills when they had red incandescent lighting and the mass bird kills stopped after the red lights were changed to white strobe lights. An average of 2300 birds was killed each year over a 10-year period at lighted smoke-stacks near Kingston, Ontario. The bird kills ended after the lights were changed to white strobes. To prove the hypothesis of the structure light-blue color attraction and disorientation to the Received: November 1, 2011 Accepted: November 3, 2011 Published: November 16, 2011 10296

dx.doi.org/10.1021/es203854q | Environ. Sci. Technol. 2011, 45, 10296–10297

Environmental Science & Technology

VIEWPOINT

Figure 1. Photo of the bird kill incident site. Dead birds were found on the ground beside Buildings A, B, and C and no dead birds were observed under the bridge. Building A is in 10 m length 6 m wide 4.5 m high, and B and C are 3- and 2-ﬂoors structures, respectively; the insert was the original color of the three buildings, which were all repainted in milk-yellow after the incident.

night-migrating birds, the three buildings were painted in milkyellow (Figure 1 A, B, C). Since the milk-yellow painting, no birds have ever been observed smashing into the buildings in the past four years. Scientiﬁc basis to establish policy regulation on the height, conﬁguration and lighting of communications towers to protect migratory birds has been developed.2 The ﬁnding here adds supporting evidence to regulate the coloring and nightlighting scheme of buildings and structures in migratory pathways of night-migrating birds to protect the migratory birds.

’ AUTHOR INFORMATION Corresponding Author

*Phone/fax: +86-10-64421030; e-mail: [email protected].

’ REFERENCES (1) Liu, J.; Xiao, H.; Lei, F.; Zhu, Q.; Qin, K.; Zhang, X. W.; Zhang, X. L.; Zhao, D.; Wang, G.; Feng, Y.; Ma, J.; Liu, W.; Wang, J.; Gao, G..F. Highly pathogenic H5N1 inﬂuenza virus infection in migratory birds. Science 2005, 309, 1206. (2) Longcore, T.; Rich, C.; Gauthreaux, S.A. 2005. Scientiﬁc basis to establish policy regulating, communications towers to protect migratory birds. http://www.abcbirds.org/newsandreports/special_reports/LPPtowerkill. pdf (accessed November 12, 2011). (3) Ferguson, N. M.; Cummings, Derek A.T.; Cauchemez, Simon; Fraser, Christophe; Riley, Steven; Meeyai1, Aronrag; Iamsirithaworn, Sopon; Burke, Donald S. Strategies for containing an emerging inﬂuenza pandemic in Southeast Asia. Nature 2005, 437, 209–214. (4) Gauthreaux, S. A. Jr.; Belser, C. Eﬀects of artiﬁcial night lighting on migrating birds. In Ecological Consequence of Artiﬁcial Night Lighting; Rich, C.; Longcore, T., Eds.; Island Press: Covelo, CA, 2005. (5) Gehring, J.; Kerlinger, P.; Manville, A. M. Communication towers, lights, and birds: successful methods of reducing the frequency of avian collisions. Ecol. Appl. 2009, 19, 505.

10297

dx.doi.org/10.1021/es203854q |Environ. Sci. Technol. 2011, 45, 10296–10297

VIEWPOINT pubs.acs.org/est

Check Dam in the Loess Plateau of China: Engineering for Environmental Services and Food Security. Yafeng Wang,*,† Bojie Fu,† Liding Chen,† Yihe L€u,† and Yang Gao‡ †

State Key Laboratory of Urban and Regional Ecology, Research Center for Eco-Environmental Sciences, CAS, Beijing 100085, P. R. China ‡ Key Laboratory of Water Cycle and Related Land Surface Processes, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, P. R. China

Figure 1. The amount of check dam in seven provinces of Loess Plateau.

I

n the world, dams and reservoirs are often controversial projects because of their social and environmental impacts, in addition to the high cost of construction and river diversion. But in Loess Plateau, check dams have attractive advantages because of unique environmental settings and regional food supply needs. Check dams are the most widespread structures for conserving soil and water in the Loess Plateau. The Loess Plateau covers an area of some 640 000 km2 in the upper and middle reaches of the China’s Yellow River. Over 60% of the land is susceptible to soil and water losses and the soil of this region has been called the “most highly erodible soil on earth”.1 Constructing check dams in the gullies is an eﬀective strategy for reducing sediment loss. More than 100 000 check dams have been built over the last 50 years in the Loess Plateau.2 After 50 years of construction, there are about 110 thousand check dams storing 21 billion m3 sediments in the Loess Plateau. The check dams distribute mainly in seven provinces including Shaanxi (36 816), Shanxi (37 820), Gansu (6630), Inner Mongolia (17 819), Ningxia (4936), Qinghai (3877), and Henan (4147) which account for 82.5% of the total 2 (Figure 1). It is highly signiﬁcant that the check dams curb sediments from ﬂowing into the Yellow River at a rate about 3 5 million tons r 2011 American Chemical Society

annually, and they have intercepted 28 billion tons sediment since 1950s in the Loess Plateau.2 Generally, a small watershed is used as a planning and construction unit for check dams. They are constructed step by step from downstream to upstream; the combination of large, medium, and small dams can eﬀectively mitigate ﬂood damage and sedimentation downstream. The check dams can essentially raise the base level of the controlled watershed and thus reduce gully and headward erosion. The check dams can potentially serve as carbon storage and sequestration structures. The average organic carbon content in sediment trapped by check dams is 3.4 g kg 1. If estimated at this rate, the carbon storage in check dams of the Loess Plateau can amount to 0.952 Gt (1Gt = 109 t = 1015 g), which amounts to 18 24% of the total carbon storage of forest vegetation in China.3 The existing reseach results indicated that carbon sequestration was 2.3 Tg (1 Tg = 1012 g) in the new plantations in the upper Yellow River Basin from 1998 to 2004.4 Therefore, we indicated Received: November 2, 2011 Accepted: November 8, 2011 Published: November 16, 2011 10298

dx.doi.org/10.1021/es2038992 | Environ. Sci. Technol. 2011, 45, 10298–10299

Environmental Science & Technology that the carbon sequestration is more than 400 times by the check dam than by the new plantations under the “Grain to Green” program in the Loess Plateau. Otherwise, this carbon will be transported to the Yellow River and carbon release to the atmosphere is inevitable during the transport process in the water environment to potentially contribute to global warming. When check dams are ﬁlled up with sediments, man-made plain lands come into being and can be reclaimed as high quality croplands because of the nutrient enrichment during soil erosion processes on hillslopes and improved water availability. By 2002, 3200 km2 of dam croplands had been created.2 According to the monitoring data from the Suide Soil and Water Conservation Experiment Station of the Yellow River Conservancy Commission, the soil water content in dam cropland is 1.86 times of that in slope cropland. The food production in dam cropland is 2 3 times higher than that in terrace cropland, and 6 10 times higher than that in slope cropland. The average yield is 45 000 kg ha 1 with some even up to 105 000 kg ha 1. Accordingly, dam cropland accounts for about 9% of the total cropland area in the Loess Plateau, whereas the food production amounts to 20.5% of the total food production.2 With the productive dam cropland, farmers can grow high proﬁt crops or developing fresh water aquaculture to raise family income. Therefore, farmers’ dependence on slope farming is largely reduced, which has facilitated the eﬀective implementation of the large scale vegetation restoration program known as Grain to Green that is motivated by the Chinese central government and recognized as the world's largest payment for ecosystem service initiative.4 The results indicated that the vegetation restoration eﬀorts had signiﬁcantly improved land coverage grass, scrub, and woods resulting in an eﬀective control of soil erosion.5 Accordingly, the check dam as a hydro-engineering approach for soil erosion control has actually brought about services for environmental conservation and human welfare in the Loess Plateau of China. To sustain these beneﬁts of check dams, participatory approach and adaptive management is required for planning, construction, use, and maintenance of the anthropogenic structure. The case of check dams in the Loess Plateau region veriﬁed the possibility toward building harmonious relationships between man and nature by well designed engineering systems. Therefore, ecologically and environmentally friendly design is promising to adapt to a changing world.

VIEWPOINT

(2) CMWR (Ministry of Water Resource of P.R. China). Programming for check dams in the Loess Plateau (Technical Report), 2003; pp 47 48. (In Chinese) (3) Zhao, M; Zhou, G. S. Carbon storage of forest vegetation in china and its relationship with climatic factors. Climate Change 2006, 74, 175–189. (4) Liu J, et al. Ecosystem Services Special Feature: Ecological and socioeconomic eﬀects of China’s policies for ecosystem services. Proc. Natl. Acad. Sci. U.S.A. 2008, 105, 9477-9482. (5) Fu, B. J.; Chen, L. D.; Ma, K. M.; Zhou, H. F.; Wang, J. The relationships between land use and soil conditions in the hilly area of the Loess Plateau in northern Shannxi, China. Catena. 2000, 1, 69–78.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 86-10-62849102; fax: 86-10-62849102; e-mail: yfwang@ rcees.ac.cn.

’ ACKNOWLEDGMENT This work was supported by the National Natural Science Foundation of China (No. 40901098) and the Knowledge Innovation Program of the Chinese Academy of Sciences (No. KZCX2-YW-QN408). We thank Geoﬀrey Hart (Montreal, Canada) for editing an early version of this paper. We are also grateful for the comments and criticisms of the journal’s anonymous reviewers and my colleagues. ’ REFERENCES (1) Laﬂen, J. M., Tian, J. L., Huang, C. H. Soil Erosion and Dryland Farming; CRC press: Boca Raton, FL, 2000; p 736. 10299

dx.doi.org/10.1021/es2038992 |Environ. Sci. Technol. 2011, 45, 10298–10299

POLICY ANALYSIS pubs.acs.org/est

Rationale for Control of Anthropogenic Nitrogen and Phosphorus to Reduce Eutrophication of Inland Waters William M. Lewis, Jr.* Cooperative Institute for Research in Environmental Sciences, Department of Ecology and Evolutionary Biology, University of Colorado, Boulder, Colorado 80309-0216, United States

Wayne A. Wurtsbaugh Department of Watershed Sciences and the Ecology Center, Utah State University, Logan, Utah 84322-5210, United States

Hans W. Paerl Institute of Marine Sciences, University of North Carolina at Chapel Hill, Morehead City, North Carolina 28557, United States ABSTRACT: Concentrations of phosphorus and nitrogen in surface waters are being regulated in the United States and European Union. Human activity has raised the concentrations of these nutrients, leading to eutrophication of inland waters, which causes nuisance growth of algae and other aquatic plants. Control of phosphorus often has had the highest priority because of its presumed leading role in limiting development of aquatic plant biomass. Experimental evidence shows, however, that nitrogen is equally likely to limit growth of algae and aquatic plants in inland waters, and that additions of both nutrients cause substantially more algal growth than either added alone. A dual control strategy for N and P will reduce transport of anthropogenic nitrogen through drainage networks to aquatic ecosystems that may be nitrogen limited. Control of total phosphorus in eﬄuents is feasible and is increasingly being required by regulations. The control strategy for nitrogen in eﬄuents is more diﬃcult, but could be made more feasible by recognition that a substantial portion of dissolved organic nitrogen is not bioavailable; regulation should focus on bioavailable N (nitrate, ammonium, and some dissolved organic nitrogen) rather than total N. Regulation of both N and P also is essential for nonpoint sources.

’ INTRODUCTION The United States and European Union are simultaneously moving toward nutrient regulation for inland waters with the goal of controlling eutrophication. The primary symptom of eutrophication is excessive growth of aquatic autotrophs, including suspended algae (phytoplankton), attached algae (periphyton), and aquatic vascular plants (macrophytes). Secondary symptoms include deep water anoxia in lakes, increased risk of harmful algal blooms, impairment of water treatment (taste and odor, ﬁltration problems), and changes in the composition of aquatic communities.1 Nutrient pollution has raised global algal biomass and photosynthesis in lakes by approximately 60% over background conditions;2 streams and rivers are similarly aﬀected. Within populated or agriculturally productive regions aquatic primary production and biomass often are many times greater than background.3 Two elements, phosphorus (P) and nitrogen (N), explain most of the experimentally diagnosed nutrient limitation of algal growth in inland waters under natural or human-modiﬁed conditions. Some research also suggests the potential for deﬁciencies r 2011 American Chemical Society

of other elements such as iron in inland waters,4,5 but this type of limitation is likely conﬁned to special situations. Although the scientiﬁc basis of nutrient regulation seemingly was settled in the 1970s with emphasis on phosphorus control, strong controversy now has emerged about the alternative possibilities for controlling one nutrient preferentially (P) or two nutrients with equal emphasis (P, N). We provide here a perspective on nutrient control as it applies to algae, ﬁrst for lakes and then for ﬂowing waters. Regulation of total P concentrations is a well established practice.6,7 Regulation of nitrogen for control of eutrophication has been a lower priority, but has developed in a few places by control of total nitrogen concentrations (e.g., New Zealand8). National and international organizations (U.S. Environmental Protection Agency, Organisation for Economic Co-operation and Development) Received: January 24, 2011 Accepted: November 9, 2011 Revised: October 17, 2011 Published: November 09, 2011 10300

dx.doi.org/10.1021/es202401p | Environ. Sci. Technol. 2011, 45, 10300–10305

Environmental Science & Technology

POLICY ANALYSIS

recognize the signiﬁcance of both elements, but current regulatory practice emphasizes phosphorus control. We describe lines of evidence showing that nutrient control based on both P and N oﬀers a broader range of strategies and reduces the potential for corollary damage caused by anthropogenic mobilization of N.

’ COMPARISONS OF PHOSPHORUS AND NITROGEN AS LIMITING NUTRIENTS IN LAKES The limiting nutrient concept (Liebig’s Law of the Minimum9,10) holds that nutrient deﬁciency at any given time in a photosynthetic organism can be traced to a single element, which is the element available in the least amount relative to the needs of the organism. Therefore, in controlling excessive algal growth, it is important to know which element limits the expansion of algal populations when their growth stops because of nutrient depletion. The limiting nutrient concept is more complex for an entire community or ecosystem than it is for a single organism. For example, species may diﬀer, even among organisms of similar type (e.g., algae), in their optimal internal N: P ratios11 13 and their ability to store critical nutrients or to take up a nutrient at low concentrations.14,15 Thus, it is possible in a mixed community of algae for some species to be limited by phosphorus and others to be limited by nitrogen. In addition, it is possible for an environment to be very near the nutrient limitation thresholds for N and P simultaneously. Thus, a slight enrichment with one element could cause the other element to become limiting (e.g., refs 16 18). A third possibility is that seasonal or spatially heterogeneous changes may occur in the relative availability of potentially limiting nutrients (19). All of these circumstances have been documented experimentally and in nature.20 Much more attention has been given to P limitation than to N limitation in inland waters for three reasons:20 (1) phosphorus is more easily removed from anthropogenic sources than nitrogen, (2) N2 ﬁxation by cyanobacteria (also known as blue-green algae) has been assumed to make N control ineﬀective, and (3) the correlation between chlorophyll (an index of algal abundance) and total P among lakes is stronger than the correlation between chlorophyll and total N.3 A high proportion of total phosphorus can be removed (to concentrations as low as 30 μg/L) from waste streams by ﬂocculation and sedimentation.21 Thus, phosphorus limitation can be induced even in a lake that is nitrogen limited by restricting the phosphorus supply to such an extent that phosphorus limitation overtakes nitrogen limitation.22,23 This is an eﬀective strategy when the main source of phosphorus is wastewater eﬄuent, which can be readily treated. It is less feasible where diﬀuse (nonpoint) sources are important, and may be entirely infeasible where background phosphorus concentrations are high.24 27 Nutrient enrichment experiments (bottle bioassays, mesocosms, whole lakes) for lakes from all parts of the world now show that nitrogen limitation is globally as common as phosphorus limitation (Figure 1, refs 28,18, and 20). The occurrence of nitrogen limitation in lakes globally raises questions about the presumption that nitrogen limitation is self-correcting through the growth of N2-ﬁxing cyanobacteria.29 Studies of the nitrogen ﬁxation rates for cyanobacteria show that they are unable to compensate fully for nitrogen limitation in lakes,30,31 most likely because the process of N2 ﬁxation can be inﬂuenced by factors other than nitrogen and phosphorus, including turbulence coupled

Figure 1. Growth response ratios (natural log of ratio of treatment to control, with standard error) of freshwater phytoplankton for worldwide bioassay studies (redrawn from ref 18; n = >500 for each treatment).

with low transparency, trace metal or iron deﬁciency, or organic matter availability.32 Eutrophic lakes that are nitrogen limited may even be dominated by cyanobacterial taxa that cannot ﬁx N2.33 Another important factor that works against N accumulation in lakes is microbial denitriﬁcation that converts nitrate, which is bioavailable, to N2 or N2O which are not. Denitriﬁcation is stimulated by nitrate enrichment of lakes.34 Thus, nitrogen ﬁxation and nitrogen limitation can coexist in lakes, and suppression of N availability may suppress total algal biomass even when cyanobacterial N2 ﬁxers are present. N2 ﬁxers may become a larger portion of the algal community if nitrogen availability is suppressed suﬃciently to cause N limitation, even if total biomass is reduced.35 The risk of inducing a shift in community composition favoring N2 ﬁxers is a possible undesirable byproduct of induced nitrogen limitation. Presence of N ﬁxers at moderate abundances is common over a wide trophic range,36 however, and is not exclusively a symptom of impairment. The correlation between phosphorus and mean or peak chlorophyll among lakes has been erroneously interpreted as showing cause and eﬀect. In fact, the correlation reveals little about nutrient limitation because phosphorus is a mandatory component of algal biomass, as is chlorophyll.20 Therefore, chlorophyll and phosphorus will always be present together (as will all other biomass components), whether phosphorus is limiting or not (Figure 2). Nutrient limitation cannot be inferred from such correlations. Algae excrete phosphatases at the cell surface and into the surrounding water that allow them to assimilate phosphorus derived from cleavage of phosphorus from organic matter.36 Algae also can take up 10 or more times the minimum amount of P needed for synthesis of protoplasm37 and store the excess P as polyphosphate. Thus, toward the end of the growing season, most of the phosphorus in the upper water column of lakes is incorporated into algal biomass, except in lakes that are so strongly polluted with P as to exceed algal capacity for P uptake.34 For nitrogen, a signiﬁcant portion of the dissolved fraction is refractory (not bioavailable, e.g., ref 39). Dissolved inorganic N (DIN, Table 1) typically is the main N source for algal growth in inland waters, but both unpolluted and polluted inland waters also contain substantial amounts of dissolved organic N (DON). Because DON persists even when phytoplankton show nitrogen stress, as indicated by very low concentrations of DIN, DON had until recently been considered entirely refractory, but experimental evidence now has shown that a signiﬁcant portion of DON is 10301

dx.doi.org/10.1021/es202401p |Environ. Sci. Technol. 2011, 45, 10300–10305

Environmental Science & Technology

POLICY ANALYSIS

Table 1. Concentrations of Total N and P (μg/L) in a Representative Municipal Eﬄuent with Secondary Treatment Plus 50% Nitriﬁcation and in Representative Unpolluted US Streams and Rivers68 71 nutrient

eﬄuent

unpolluted streamsa

range total P, μg/L fractionation, %

2000 4000

total P total dissolved P dissolved inorganic P dissolved organic P particulate P

10 30

100

100

96

63

88

30

8 4

33 37

10 000 15,000

100 500

range total N, μg/L fractionation, % total N

Figure 2. Simulation of the relationship between P and phytoplankton chlorophyll a among a hypothetical population of lakes (seasonal averages) for which P is not limiting (r2 = 0.70, from ref 20).

available to algal cells,38 40 including not only DON from natural sources but also anthropogenic DON such as urea, which is widely used in agriculture.41 Some algal taxa have exoenzymes (amino acid oxidases, proteolytic enzymes) at the cell surface or excreted from the surface so that ammonium or small organic molecules can be released from large organic molecules and enter the cell; some taxa also are able to take up organic nitrogen by pinocytosis or phagocytosis.42 In addition, some components of DON are converted to DIN by photodegradation, but other components of DON resist photodegradation.40 Thus, the persistence of DON in the absence of DIN indicates fractional turnover of the DON pool rather than complete unavailability of DON over time scales ranging from days to months during a growing season. Natural waters vary greatly in amount of refractory nitrogen in the DON pool. A study of rivers in the eastern U.S. showed two rivers with no detectable bioavailability and seven rivers with a mean of 23% ( 4% bioavailability as determined by change in DON concentrations in six-day incubations; an accompanying literature survey for 18 sites on rivers in Europe and the U.S. showed a mean of 30% ( 4 for the labile fraction as judged mostly by 14 day incubations.43 Thus, DON of natural waters must be viewed as potentially important nutritionally to algae under nitrogen stress, yet includes a signiﬁcant refractory component. Fractions of N and P diﬀer in their potential to predict experimentally diagnosed nutrient limitation in lakes. For phosphorus, total P and total soluble P are equally accurate indicators. For nitrogen, dissolved inorganic nitrogen (almost entirely composed of nitrate plus ammonium) is an indicator superior to total nitrogen or total dissolved nitrogen.44 This is not surprising, given the unavailability of a substantial portion of DON to algae.

’ CONTROL OF N, P, OR BOTH Sole focus on phosphorus as a means of controlling algal biomass may seem advantageous because it is much less expensive than control of both N and P.45 Some researchers also continue to argue that nitrogen control does not work because N2 ﬁxation can provide algae with labile nitrogen.46 According to this argument, lakes that are N deﬁcient will accumulate N over time, thus eventually reaching P limitation. Lake 227 of the Canadian Experimental Lake Area, which oﬀers the longest record of whole

total dissolved N dissolved inorganic N

100

96

79

77

29

NO3 -N

61

23

NH4+-N dissolved organic N

16 19

6 50

4

21

particulate N a

100

Unpolluted lakes will show lower DIP, DIN, PP.

lake manipulation, is cited as an example of evolving N suﬃciency under P enrichment,46 but a contrary interpretation of the data has been proposed.31 Multiyear whole lake enrichment experiments with P only document persistence of N limitation in lakes with substantial P and populations of N2 ﬁxers. For example, whole lake fertilization of several Swedish lakes with P only (multiple years), yielded no higher biomass or only slightly higher biomass than was found before fertilization.47 The same lakes developed biomass 15 - 60 times higher with P + N fertilization (refs 47 and 41 give other examples). Focus on phosphorus control presumes that phosphorus loading of a lake can be reduced suﬃciently to induce and sustain phosphorus control of algae. Where nonpoint phosphorus or background phosphorus sources are strong enough to sustain eutrophic conditions, phosphorus control measures may not provide enough phosphorus recovery to reduce algal biomass. In addition, allowing the balance between nitrogen and phosphorus to be strongly distorted over entire regions by selective control of phosphorus may change the species composition or diversity of aquatic communities,13,48 which often reﬂect a close balance between nitrogen and phosphorus availability.18,49 Finally, because nitrogen limitation is quite common in fresh waters and even more common in coastal waters and saline lakes,50 52 allowing nitrogen to be released indiscriminately from one water body to another through the drainage network could cause widespread stimulation of algal growth by providing nitrogen to algal communities downstream that otherwise would be nitrogen limited.53,54 Thus, dual nutrient control has multiple advantages.

’ STRATEGIES FOR LIMITING PHOSPHORUS AND NITROGEN IN THE ENVIRONMENT Use of total P as an index of P availability in lakes is defensible for lakes because most of the phosphorus in the growth zone of lakes is available to algae; it consists of total dissolved P (TDP) 10302

dx.doi.org/10.1021/es202401p |Environ. Sci. Technol. 2011, 45, 10300–10305

Environmental Science & Technology with its two components, dissolved inorganic P (DIP, often designated soluble reactive P, SRP) and dissolved organic P (DOP) plus particulate P (PP), which consists mostly of phytoplankton with their internal phosphorus stores. In lakes the particulate fraction of N also consists mainly of phytoplankton, and can be counted as bioavailable, as can DIN and some DON. Thus, the concept of bioavailability suggests that water quality standards for P in lakes can be based on total P, but for N they should be based on total N minus refractory DON. Regulating total N without adjusting for unavailable DON would be equally eﬀective, but would lower the feasibility and raise the cost of N control. For nutrient control we focus here on eﬄuents as nutrient sources because regulation of eﬄuents is feasible through established permitting processes and because the technological basis for regulation nonpoint of sources, which may be dominant nutrient sources in some cases,34 is weak.

’ EFFLUENT REGULATION THAT IS CONSISTENT WITH STANDARDS BASED ON BIOAVAILABILITY Point source eﬄuents, which are the main target for discharge permitting, are rich in bioavailable total dissolved P (Table 1). For the dominant treatment technologies (i.e., with the exception of oxidation ponds or ditches), particulate P is not a major concern because of the eﬃciency of particle removal during treatment. Thus, permits written on the basis of total phosphorus in eﬄuent typically will translate well into a limitation on bioavailable phosphorus in lakes. For nitrogen, the presence of dissolved organic N in eﬄuent is a complicating factor. DON in municipal eﬄuent is derived partly from the inﬂuent waste stream and partly from microbial metabolism that occurs during treatment.55 Eﬄuents appear to be similar to inland waters and nearshore marine waters in having both refractory and labile components. One study of a domestic treatment eﬄuent from a treatment facility with low nitrogen output attained by combined nitriﬁcation and denitriﬁcation showed a median labile component near 40% (range, 18 61%) based on 14-day bioassays.55 Other studies have shown a similar range for bioavailable N in municipal eﬄuent.56,57 If the total N limits are strict enough to be fully eﬀective in protecting lakes from enrichment with labile N, wastewater treatment facilities will ﬁnd that the limiting factor in their ability to produce low nitrogen eﬄuent is DON, which is more diﬃcult to remove than DIN. In fact, the ultimate baseline for DON concentration, as estimated by time course bioassays for a wastewater facility operating at low nitrogen output, may approach 1 mg/L.53 To regulate the refractory component of DON with stringency equal to that of DIN or labile DON overlooks the very diﬀerent potential eﬀects of the refractory and labile fractions of total dissolved nitrogen. A regulatory system that takes into account the relative abundance of refractory DON in setting eﬄuent limits for nitrogen would require a standardized analysis of refractory DON. Bioassays could be used for this purpose according to a rationale very similar to the long accepted CBOD5 (5 day) and CBODu (ultimate) analyses for organic carbon.55 For both nitrogen and carbon, improved technology also oﬀers new possibilities through the use of ﬂuorescence spectroscopy 58 60 which, if calibrated with bioassay, might allow rapid analysis of large numbers of samples for both DOC and DON.

POLICY ANALYSIS

Figure 3. Response of attached algae in streams to experimental enrichments with N, P, or N+P (n = 237; redrawn from ref 63).

Table 2. Summary of Three Possible Eﬄuent and TMDLRelated Regulatory Strategies for Nutrients basis of regulation

feasibility

cost

total P

high

moderate

allows N pollution

total P, Total N

low

high

may require removal

total Pa, total

high

high

of refractory N focuses on bioavailable

N

refractory N

comments

nutrients

a

TDP may be a better option for stream monitoring and lake loading limits where PP is mostly adsorbed onto mineral particles.

’ STREAMS Although rivers and slowly ﬂowing streams may produce phytoplankton populations comparable to those of lakes, periphyton (attached algae) also are important and may be dominant, especially in streams of small to intermediate size. Excessive growth of periphyton can be a byproduct of nutrient enrichment in streams or rivers. As in the case of lakes, extensive study at many sites has shown that phosphorus and nitrogen are about equally likely to be limiting to the growth of periphyton (Figure 3; refs 61 63). For stream periphyton, unlike lake phytoplankton, as much as half of experimentally tested locations show no nutrient limitation. As in the case of lakes, however the strongest responses to nutrient addition typically are for addition of both N and P. The stimulation threshold for nitrogen and phosphorus enrichment response in streams appears to be higher than in lakes.64 67 Thus, protective nutrient standard concentrations may justiﬁably be higher for streams than for lakes, but will diﬀer among distinct categories of streams. The arguments regarding fractions of phosphorus and nitrogen in lakes as given above are likely applicable to ﬂowing waters as well. One exception is the consistently greater proportion of mineral particulate phosphorus (there is no signiﬁcant mineral fraction for N) that is carried in suspension by ﬂowing waters (Table 1). It may be preferable to use total soluble phosphorus rather than total phosphorus as a basis for regulation of P in ﬂowing waters and for development of loading restrictions on lakes, given that mineral phosphorus is much less available to algae. Assessment of eutrophication in streams and rivers has lagged behind that of lakes. Additional research will be necessary to 10303

dx.doi.org/10.1021/es202401p |Environ. Sci. Technol. 2011, 45, 10300–10305

Environmental Science & Technology identify protective standards for them. Nevertheless, many of the issues surrounding nitrogen in streams and rivers are the same as for lakes. Regulation of eutrophication in ﬂowing waters should be based on N and P controls and recognition of refractory DON as a regulatory consideration.

’ CONCLUSION Restriction of the anthropogenic release of both N and P to inland waters is a means of controlling excessive algal growth. P regulation should be based on total P (for lakes) or total dissolved P (preferred for ﬂowing waters). N regulation should be based on bioavailable N rather than total N; regulation of total N will likely be infeasible or will require unrealistically high standards (Table 2). ’ AUTHOR INFORMATION Corresponding Author

*Phone: 303-492-6378; fax: 303-492-0928; e-mail: lewis@spot. colorado.edu.

’ REFERENCES (1) Kalﬀ, J. Limnology: Inland Water Ecosystems; Prentice Hall: NJ, 2002. (2) Lewis, W. M., Jr. Global primary production of lakes: 19th Baldi Memorial Lecture. Inland Waters 2011, 1, 1–28. (3) OECD. Eutrophication of Waters—monitoring, Assessment and Control; Paris, France: Organization for Economic Co-operation and Development. 1982. (4) North, R. L.; Guildford, S. J.; Smith, R. E. H.; Havens, S. M.; Twiss, M. R. Evidence for phosphorus, nitrogen, and iron colimitation of phytoplankton communities in Lake Erie. Limnol. Oceanogr. 2007, 52, 315–328. (5) Wurtsbaugh, W. A.; Horne, A. J. Iron in eutrophic nitrogen ﬁxation and growth. Can. J. Fish. Aquat. Sci. 1983, 41, 1419–1429. (6) Welch, E. B.; Lindell, T. Ecological Eﬀects of Wastewater: Applied Limnology and Pollutant Eﬀects; E & FN Spon: London, England, 2000. (7) Cooke, G. D.; Welch, E. B.; Peterson, S. A.; Nichols, S. A. Restoration and Management of Lakes and Reservoirs, 3rd ed.; Taylor & Francis Group, LLC: London, 2005. € (8) Abell, J. M.; Ozkundakci, D.; Hamilton, D. P. Nitrogen and phosphorus limitation of phytoplankton growth in New Zealand Lakes: Implications for eutrophication control. Ecosystems 2010, 13, 966–977. (9) Hutchinson, G. E. Eutrophication: The scientiﬁc background of a contemporary practical problem. Am. Sci. 1973, 61, 269–279. (10) Sterner, R. W.; Elser, J. J. Ecological Stoichiometry: The Biology of Elements from Molecules to the Biosphere; Princeton University Press: Princeton, NJ, 2002. (11) Smith, V. H. The nitrogen and phosphorus dependency of algal biomass in lakes: An empirical and theoretical analysis. Limnol. Oceanogr. 1982, 27, 1101–1112. (12) Suttle, C. A.; Harrison, P. J. Ammonium and phosphate uptake rates, N:P ratios, and evidence for N and P limitation in some oligotrophic freshwater lakes. Limnol. Oceanogr. 1988, 33, 186–202. (13) Pick, F. R. Species speciﬁc phytoplankton responses to nutrient enrichment in limnetic enclosures. Arch. Hydrobiol. Beih. Ergebn. Limnol. 1989, 32, 177–187. (14) Sterner, R. W.; Hessen, D. O. Algal nutrient limitation and the nutrition of aquatic herbivores. Ann. Rev. Ecol. Syst. 1994, 25, 1–29. (15) Burger, D. F.; Hamilton, D. P.; Hall, J. A.; Ryan, E. F. Phytoplankton nutrient limitation in a polymictic eutrophic lake: Community versus species-speciﬁc responses. Fund. Appl. Limnol. 2007, 169, 57–68. (16) Bergstr€om, A.; Jansson, M. Atmospheric nitrogen deposition has caused nitrogen enrichment and eutrophication of lakes in the northern hemisphere. Global Change Biol. 2006, 12, 635–643.

POLICY ANALYSIS

(17) Elser, J. J.; Andersen, T.; Baron, J. S.; Bergstrom, A.-K.; Jansson, M.; Kyle, M.; Nydick, K. R.; Steger, L.; Hessen, D. O. Shifts in lake N:P stoichiometry and nutrient limitation driven by atmospheric nitrogen deposition. Science 2009, 326, 835–837. (18) Elser, J. J.; Bracken, M. E. S.; Cleland, E. E.; Gruner, D. S.; Harpole, W. S.; Hillebrand, H.; Ngai, J. T.; Seabloom, E. W.; Shurin, J. B.; Smith, J. E. Global analysis of nitrogen and phosphorus limitation of primary producers in freshwater, marine and terrestrial ecosystems. Ecol. Lett. 2007, 10, 1135–1142. (19) Stoddard, J. L. Long-term changes in watershed retention of nitrogen its causes and aquatic consequences. Adv. Chem. Ser. 1994, 237, 223–284. (20) Lewis, W. M., Jr.; Wurtsbaugh, W. W. Control of lacustrine phytoplankton by nutrients: Erosion of the phosphorus paradigm. Int. Rev. Hydrobiol. 2008, 93, 446–465. (21) Stumm, W., Morgan, J. J. Aquatic Chemistry, Chemical Equilibria and Rates in Natural Waters, 3rd ed.; Wiley: New York, 1996. (22) Golterman, H. L. Physiological Limnology. an Approach to the Physiology of Lake Ecosystems; Elsevier: New York, 1975. (23) Ryding, S. O., Rast, W. The Control of Eutrophication of Lakes and Reservoirs, Man and the Biosphere series; Parthenon: New York, 1989; Vol. 1. (24) James, C.; Fisher, L. J.; Moss, B. Nitrogen driven lakes: The Shropshire and Cheshire meres?. Arch. Hydrobiol. 2003, 158, 249–266. (25) Bunting, L.; Leavitt, P. R.; Hall, V.; Gibson, C. E.; McGee, E. J. Nitrogen degradation of water quality in a phosphorus-saturated catchment: The case of Lough Neagh, Northern Ireland. Verh. Int. Verein. Limnol. 2005, 29, 1005–1008. (26) NRC. Endangered and Threatened Fishes in the Klamath River Basin: Causes of Decline and Strategies for Recovery; National Academies Press: Washington, DC, 2004. (27) Lewis, W. M., Jr.; Saunders, J. F., III; McCutchan, J. H., Jr. Application of a nutrient-saturation concept to the control of algal growth in lakes. Lake Res. Manag. 2008, 24, 41–46. (28) Elser, J. J.; Marzolf, E. R.; Goldman, C. R. Phosphorus and nitrogen limitation of phytoplankton growth in the freshwaters of North America: A review and critique of experimental enrichments. Can. J. Fish. Aquat. Sci. 1990, 47, 1468–1477. (29) Schindler, D. W. Evolution of phosphorus limitation in lakes: Natural mechanisms compensate for deﬁciencies of nitrogen and carbon in eutrophied lakes. Science 1977, 195, 260–262. (30) Howarth, R. W.; Marino, R.; Lane, J.; Cole, J. J. Nitrogen ﬁxation in freshwater, estuarine, and marine ecosystems: Rates and importance. Limnol. Oceanogr. 1988, 33, 669–687. (31) Scott, J. T.; McCarthy, M. J. Nitrogen ﬁxation may not balance the nitrogen pool in lakes over timescales relevant to eutrophication management. Limnol. Oceanogr. 2010, 55, 1265–1270. (32) Paerl, H. W. Physiological ecology and regulation of N2 ﬁxation in natural waters. Adv. Microb. Ecol. 1990, 11, 305–344. (33) Paerl, H. W.; Xu, H.; McCarthy, M. J.; Zhu, G.; Qin, B.; Li, Y.; Gardner, W. S. Controlling harmful cyanobacterial blooms in a hypereutrophic lake (Lake Taihu, China): The need for a dual nutrient (N & P) management strategy. Water. Res. 2011, 45, 1973–1983. (34) Downing, J. A.; McCauley, E. The nitrogen: Phosphorus relationship in lakes. Limnol. Oceanogr. 1992, 37, 936–945. (35) Smith, V. H. Low nitrogen to phosphorus ratios favor dominance by blue-green algae in lake phytoplankton. Science 1983, 221, 669–670. (36) Reynolds, C. S. Ecology of Phytoplankton; Cambridge: New York, 2006. (37) Healey, F. P. Characteristics of phosphorus deﬁciency in Anabaena. J. Phycol. 1973, 9, 383–394. (38) Berman, T.; Chava, S. Algal growth on organic compounds as nitrogen sources. J. Plankton Res. 1999, 21, 1423–1437. (39) Seitzinger, S. P.; Sanders, R. W.; Styles, R. Bioavailability of DON from natural and anthropogenic sources to estuarine plankton. Limnol. Oceanogr. 2002, 47, 353–366. (40) Bronk, D. A.; See, J. H.; Bradley, P.; Killberg, L. DON as a source of bioavailable nitrogen for phytoplankton. Biogeoscience 2007, 4, 283–296. 10304

dx.doi.org/10.1021/es202401p |Environ. Sci. Technol. 2011, 45, 10300–10305

Environmental Science & Technology (41) Finlay, K.; Patoine, A.; Donald, D. B.; Bogard, M. J.; Leavitt, P. R. Experimental evidence that pollution with urea can degrade water quality in phosphorus- rich lakes of the Northern Great Plains. Limnol. Oceanogr. 2010, 55, 1213–1230. (42) Bronk, D. A.; Steinberg, D. Nitrogen regeneration. In Nitrogen in the Marine Environment; Capone, D. G., Bronk, D. A., Mulholland, M., Carpenter, E., Eds.; Elsevier: New York, 2008; pp 385 468. (43) Wiegner, T. N.; Seitzinger, S. P.; Gilbert, P. M.; Bronk, D. A. Bioavailability of dissolved organic nitrogen and carbon from nine rivers in the eastern United States. Aquat. Microb. Ecol. 2006, 43, 277–287. (44) Morris, D. P.; Lewis, W. M., Jr. Phytoplankton nutrient limitation in Colorado mountain lakes. Freshwater Biol. 1988, 20, 315–327. (45) Wang, H. J.; Wang, H. Z. Mitigation of lake eutrophication: Loosen nitrogen control and focus on phosphorus abatement. Prog. Nat. Sci. 2009, 19, 1445–1451. (46) Schindler, D. W.; Hecky, R. E.; Findlay, D. L.; Stainton, M. P.; Parker, B. R.; Paterson, M. J.; Beaty, K. G.; Lyng, M.; Kasian, S. E. M. Eutrophication of lakes cannot be controlled by reducing nitrogen input: Results of a 37-year whole-ecosystem experiment. Proc. Natl. Acad. Sci. U. S. A. 2008, 105, 11254–11258. (47) Holmgren, S. K. Experimental lake fertilization in the Kuokkel Area, Northern Sweden. Phytoplankton biomass and algal composition in natural and fertilized subarctic lakes. Int. Rev. Gesamten Hydrobiol. Hydrogr. 1984, 69, 781–817. (48) Berman, T. The role of DON and the eﬀect of N:P ratios on occurrence of cyanobacterial blooms: Implications from the outgrowth of Aphanizomenon in Lake Kinneret. Limnol. Oceanogr. 2001, 46, 443–447. (49) Sterner, R. W. On the phosphorus limitation paradigm for lakes. Int. Rev. Hydrobiol. 2008, 93, 433–445. (50) Conley, D. J.; Paerl, H. W.; Howarth, R. W.; Boesch, D. F.; Seitzinger, S. P.; Havens, K. E.; Lancelot, C.; Likens, G. E. Controlling eutrophication: Nitrogen and phosphorus. Science 2009, 323, 1014–1015. (51) Paerl, H. W. Controlling eutrophication along the freshwatermarine continuum: Dual nutrient (N and P) reductions are essential. Estuarine Coasts 2009, 32, 593–601. (52) Hammer, U. T. Saline Lake Ecosystems of the World; Junk Publishers: Dordrecht, 1986, P. 616. (53) Elmgren, R.; Larsson, U. Nitrogen and the Baltic Sea: Managing nitrogen in relation to phosphorus. In The Scientiﬁc World, Special edition; Balkema Publishers: 2001; Vol. 1(S2), pp 371 377. (54) Paerl, H. W.; Valdes, L. M.; Piehler, M. F.; Lebo, M. E. Solving problems resulting from solutions: The evolution of a dual nutrient management strategy for the eutrophying Neuse River Estuary, North Carolina, USA. Environ. Sci. Technol. 2004, 38, 3068–3073. (55) Urgun-Demirtas, M.; Sattayatewa, C.; Pagilla, K. R. Bioavailablity of dissolved organic nitrogen in treated eﬄuents. Water Environ. Res. 2008, 80, 397–406. (56) Bronk, D. A.; Roberts, Q. N.; Sanderson, M. P.; Canuel, E. A.; Hatcher, P. G.; Mesﬁoui, R.; Filippino, K. C.; Mulholland, M. R.; Love, N. G. Eﬄuent organic nitrogen (EON): Bioavailability and photochemical and salinity-mediated release. Environ. Sci. Technol. 2010, 44, 5830–5835. (57) Filippino, K. C.; Mulholland, M. R.; Bernhardt, P. W.; Boneillo, G. E.; Morse, R. E.; Semcheski, M.; Marshall, H.; Love, N. G.; Roberts, Q.; Bronk, D. A. The bioavailability of eﬄuent-derived organic nitrogen along an estuarine salinity gradient. Estuarine Coasts 2011, 34, 269–280. (58) Chen, J.; Gu, B.; LeBouef, E. J.; Pan, H.; Dai, S. Spectroscopic characterization of the structural and functional properties of natural organic matter fractions. Chemosphere 2002, 48, 59–68. (59) Stedman, C. A.; Markager, S.; Bro, R. Tracing dissolved organic matter in aquatic environments using a new approach to ﬂuorescence spectroscopy. Mar. Chem. 2003, 82, 239–254. (60) Cory, R. M.; McKnight, D. M. Fluorescence spectroscopy reveals ubiquitous presence of oxidized and reduced quinines in dissolved organic matter. Environ. Sci. Technol. 2005, 39, 8142–8149. (61) Dodds, W. K.; Welch, E. B. Establishing nutrient criteria in streams. J. N. Am. Benthol. Soc. 2000, 19, 186–196.

POLICY ANALYSIS

(62) Tank, J. L.; Dodds, W. L. Nutrient limitation of epilithic and epixylic bioﬁlms in ten North American streams. Freshwater Biol. 2003, 48, 1031–1049. (63) Francoeur, S. N. Meta-analysis of lotic nutrient amendment experiments: Detecting and quantifying subtle responses. J. N. Am. Benthol. Soc. 2001, 20, 358–368. (64) Lewis, W. M., Jr.; McCutchan, J. H., Jr. Ecological responses to nutrients in streams and rivers of the Colorado mountains and foothills. Freshwater Biol. 2010, 55, 1973–1983. (65) Dodds, W. K.; Lopez, A. J.; Bowden, W. B.; Gregory, S.; Grimm, N. B.; Hamilton, S. K.; Hershey, A. E.; Marti, E.; McDowell, W. H.; Meyer, J. L.; Morrall, D.; Mulholland, P. J.; Peterson, B. J.; Tank, J. L.; Valett, H. M.; Webster, J. R.; Wollheim, W. N uptake as a function of concentration in streams. J. N. Am. Benthol. Soc. 2002, 21, 206–220. (66) Mulholland, P. J.; Steinman, A. D.; Elwood, J. W. Measurements of phosphorous uptake length in streams: Comparison of radiotracer and stable PO4 releases. Can. J. Fish. Aquat. Sci. 1990, 47, 2351–2357. (67) O’Brien, J. M.; Dodds, W. K.; Wilson, K. C.; Murdock, J. N.; Eichmiller, J. The saturation of N cycling in Central Plains streams: 15N experiments across a broad gradient of nitrate concentrations. Biogeochemistry 2007, 84, 31–49. (68) Lewis, W. M., Jr. Yield of nitrogen from minimally disturbed watersheds of the United States. Biogeochem. 2002, 57/58, 375–385. (69) Meybeck, M. Carbon, nitrogen, and phosphorus transport by world rivers. Am. J. Sci. 1982, 282, 401–450. (70) Dodds, W. K.; Oakes, R. M. A technique for establishing reference nutrient concentrations across watersheds aﬀected by humans. Limnol. Oceanogr.: Methods 2004, 2, 333–341. (71) Smith, R. A.; Alexander, R. B.; Schwarz, G. E. Natural background concentrations of nutrients in streams and rivers of the conterminous United States. Environ. Sci. Technol. 2003, 37, 3039–3046.

10305

dx.doi.org/10.1021/es202401p |Environ. Sci. Technol. 2011, 45, 10300–10305

POLICY ANALYSIS pubs.acs.org/est

Should We Pretreat Solid Waste Prior to Anaerobic Digestion? An Assessment of Its Environmental Cost Marta Carballa,* Cecilia Duran, and Almudena Hospido* Department of Chemical Engineering, School of Engineering, University of Santiago de Compostela, Rua Lope Gomez de Marzoa s/n, E-15782 Santiago de Compostela, Spain

bS Supporting Information ABSTRACT: Many studies have shown the eﬀectiveness of pretreatments prior to anaerobic digestion of solid wastes, but to our knowledge, none analyzes their environmental consequences/costs. In this work, seven diﬀerent pretreatments applied to two types of waste (kitchen waste and sewage sludge) have been environmentally evaluated by using life cycle assessment (LCA) methodology. The results show that the environmental burdens associated to the application of pretreatments prior to anaerobic digestion cannot be excluded. Among the options tested, the pressurize-depressurize and chemical (acid or alkaline) pretreatments could be recommended on the basis of their beneﬁcial net environmental performance, while thermal and ozonation alternatives require energy eﬃciency optimization to reduce their environmental burdens. Reconciling operational, economic and environmental aspects in a holistic approach for the selection of the most sustainable option, mechanical (e.g., pressurize-depressurize) and chemical methods appear to be the most appropriate alternatives at this stage.

’ INTRODUCTION Anaerobic digestion (AD) is a very promising option for the treatment of solid organic wastes due its ability to transform organic matter into biogas (with 60 70% CH4), with the concomitant reduction of the amount of ﬁnal solids to be disposed.1 Moreover, solid byproduct (digestate) can be further used for agricultural purposes.2 However, AD of solid wastes is often limited by long retention times (20 30 days) and/or low overall degradation eﬃciencies (30 50%), probably associated with the hydrolysis stage.3 Therefore, signiﬁcant eﬀort has been dedicated in recent years to ﬁnd alternatives to improve AD of solid wastes.4 Among the diﬀerent options, the use of pretreatments is the most studied and its operational eﬀectiveness has been demonstrated by many authors.1,5 8 All pretreatments entail the use of resources (chemicals and/or energy), thus deriving not only ﬁnancial but also environmental costs. Preliminary economical analyses of several pretreatments have been recently published;8,9 however, to the best of our knowledge, their environmental evaluation has not been addressed yet. Life cycle assessment (LCA) is one of the most widely known and internationally accepted methodologies to compare environmental impacts of processes and systems.10 Several LCAs have focused on examining sewage sludge (SS) treatment options and/or end uses,11 16 most of them concluding that the combination of AD and agricultural application is the preferable alternative from an environmental point of view. This methodology has been also applied as a decision support tool in the selection of the most adequate municipal solid waste (MSW) management strategy for several countries or regions.17 21 In most cases, the r 2011 American Chemical Society

recycling of valuable materials together with the AD of the organic fraction of MSW, mostly consisting of food waste or kitchen waste (KW), turned to be the best scenario in environmental terms. Other studies were only focused on the conversion and management options of the organic fraction of MSW,22 24 concluding that AD was environmentally more favorable than incineration or aerobic composting. So, up to now, much LCA has validated AD and agricultural application as the best options for treatment and ﬁnal disposal of organic solid wastes, respectively, but no information could be found on the environmental impact of the use of pretreatments prior to AD of solid wastes. In this study, we aim to fulﬁll this information gap by using LCA to evaluate the environmental consequences associated to the use of seven diﬀerent pretreatments before AD of two organic solid wastes (kitchen waste and sewage sludge), and consequently, add the environmental vector to the technical and the economic evaluation toward a more sustainable decision making process.

’ EXPERIMENTAL SECTION Functional Unit Definition. The functional unit (FU) is usually defined in terms of the system output;25 however, when dealing with waste management systems, the FU might Received: June 1, 2011 Accepted: October 31, 2011 Revised: September 26, 2011 Published: October 31, 2011 10306

dx.doi.org/10.1021/es201866u | Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology

POLICY ANALYSIS

Scenarios Description. The system included the different pretreatments applied, the anaerobic digestion process with energy recovery and the disposal of the digestate in agricultural land. Based on the different pretreatments and waste used (KW and SS), 16 scenarios (8 per type of waste) were considered for comparison (Table 2), including the reference scenario. The latter refers to the anaerobic treatment of 10 L of waste without being previously pretreated. A detailed description of the pretreatments applied to KW and SS can be found in Ma et al. (2011)8 and Carballa et al. (2006,5 20076), respectively. It is important to make clear that not all scenarios were experimentally tested; instead, estimations were conducted on the basis of the experimentally tested pretreatments. A detailed description of the assumptions required for those estimated alternatives can be found in the Supporting Information (SI). AD was carried out in a lab-scale continuously stirred tank reactor at thermophilic conditions (55 ( 2 °C) and with a sludge retention time (SRT) of 10 and 20 days for SS and KW, respectively. A detailed description of the equipment and their performance can be found elsewhere.8,27 However, due to the lack of information, the infrastructure related to lab-scale operation was excluded from the analysis. Final disposal on agricultural land was modeled on the basis of literature data (see the Inventory Analysis section). The generation of SS and KW was left beyond the boundaries of the system, and the common elements within the system boundaries, such as digested solids conditioning, transportation and the spreading procedure, were not included in the analysis for comparative reasons. Inventory Analysis. In this stage, the raw materials consumed, the energy used, the products and coproducts obtained, and the emissions to air, water and soil, were identified and quantified for each scenario. Lab-scale experimental data of the KW and SS characteristics for each scenario and AD performance were provided by Duong (2009),28 Ma et al. (2011),8 and Carballa et al. (2006;5 2007;6 200929), except for the levels of nutrients and heavy metals in KW that were taken from literature.30 Since

be defined in terms of the system input, that is, the waste to be managed.26 Accordingly, the management (i.e., pretreatment, anaerobic digestion, and agriculture application) of 10 L of solid waste has been chosen in this work. For KW, the 10 L consisted of food waste provided by Trans Vanheede Environmental Group (Belgium) diluted with wastewater coming from a sewage treatment plant (Ossemeersen, Belgium) in order to achieve the required organic loading rate (OLR) in the digester (for further information, see Ma et al., 20118). In the case of SS, 10 L of a mixture of primary and secondary sludge (70:30, v/v) collected from the two thickeners existing in a sewage treatment plant (Galicia, NW Spain) was considered (for a more detailed description, see Carballa et al., 200727). The physic-chemical parameters, nutrients and heavy metals content in the KW and SS are reported in Table 1. Table 1. Main Characteristics of Sewage Sludge (n = 20) and Kitchen Waste (n = 10, except Nutrients and Heavy Metals Which Come from Literature30) Used in the Experiments parameter

kitchen waste

sewage sludge

pH

3.8 ( 0.2

5.6 ( 0.2

CODt (g/kg KW or L SS)

268 ( 20

50 ( 18

CODs (g/kg KW or L SS) TS (g/kg KW or L SS)

75 ( 7 166 ( 14

4(2 53 ( 19

VS (g/kg KW or L SS)

155 ( 13

34 ( 13

N (g/kg TS)

31.6

21.5

P (g/kg TS)

5.2

30.5

Zn (mg/kg TS)

76

868

Cu (mg/kg TS)

31

293

Cd (mg/kg TS)

1

2

Cr (mg/kg TS) Pb (mg/kg TS)

2 4

167 93

Ni (mg/kg TS)

2

79

Table 2. Description of the Pre-Treatments Applied in the Diﬀerent Scenarios of KW and SS OLR scenariosa (kg COD m‑3d 1) KW1b

3.0

SS1

4.3

pretreatment

description Reference (nonpre-treated)

KW2b

4.0

alkaline

Addition of lime until pH 12, checking this value after 24 h, and neutralization with HCl (10 N)

SS2 KW3b

3.6 4.0

acid

before feeding the digester. Doses: 0.125 g CaO g 1 VSS and 0.058 g HCl g 1 VSS. Addition of HCl (10 N) until pH 2, checking this value after 24 h, and neutralization with NaOH (10 N)

SS3

4.0

KW4b

3.0

thermal

Heating and maintenance at 120 °C during 30 min. Cooling to room temperature before

thermo-acid

Acidiﬁcation with HCl (10 N) until pH 2 followed by thermal treatment at 120 °C during 30 min. Freezing at

SS4

8.7

KW5b

3.0

before feeding the digester. Doses: 0.026 g HCl g

1

VSS and 0.040 g “extra” NaOH g

3.0

KW6b

5.0

freeze thaw

SS6 KW7b

5.0 5.0

feeding the digester. pressurize depressurize Pressurization to 10 bar with airc and quick depressurization to atmospheric pressure.

SS7

5.0 5.0

SS8

6.6

VSS.

feeding the digester.

SS5

KW8b

1

Cooling to room temperature and neutralization with NaOH (10 N) before feeding the digester.

ozone

20 °C for 6 h and thawing at 55 °C for 30 min. Cooling to room temperature before

Ozonation in a 10 L bubble column operated in batch for 2 h at room temperature. The ozone dose was set approximately at 20 mg O3 g

1

TSS.

a

Names in italics indicate the scenarios that have not been experimentally tested but inventoried based on estimations. b NaOH (10 N) was used in all KW scenarios in order to neutralize the inﬂuent prior to feed the digester. c Air was considered instead of CO2 because no chemical eﬀect (e.g., increase in pH) was observed from the use of CO28. 10307

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology

POLICY ANALYSIS

Table 3. Summary of the Inventory Data for the 16 Scenarios Compared (see Table 2 for the Description of the Scenarios)a KW1 KW2 KW3 KW4 KW5

KW6

KW7

KW8

24

24

24

SS1

SS2

SS3

SS4

SS5

SS6

SS7

SS8

chemicals (g FU 1) NaOH

24

40

HCl

23

CaO

50

24

10

36 8

13

4

382

O2

355

22

energy used (kJ FU 1)

CH4 (%)

30

7

28

Na2S2O3 biogas production (L FU 1)

34

160

176

280

959

948

1,866

68

1,476

220

240

280

400

185

20

202 58.3

222 62.3

354 65.0

4,375

4,584

5,536

68

1,476

350

303

354

505

234

65.0

65.0

65.0

65.0

65.0

65.0

65.0

65.0

59.4

65.0

65.0

65.0

65.2

N

9.5

15.1

15.1

10.2

11.6

19.6

20.2

19.9

8.9

10.4

9.6

12.2

9.6

9.6

9.6

6.7

P

1.6

2.5

2.5

1.7

1.9

3.2

3.3

3.3

12.7

14.8

13.6

17.3

13.6

13.6

13.6

9.5

nutrients content (g FU 1)

a

Names in italics indicate that those scenarios have not been experimentally tested but inventoried based on estimations.

nutrients and heavy metals content in sludge varied during the experimental period, the average values reported in Table 1 were considered for all scenarios. One operational advantage of the use of pretreatments is that it allows operating the anaerobic digesters at higher OLR. Consequently, the OLR varied among the different scenarios, each corresponding to “optimal operational conditions”, that is, the highest OLR enabling steady state performance and conversion of at least 50% of the COD input into biogas (Table 2). Energy was not only used in some pretreatments but also during the AD process for stirring and heating. In this study, it was assumed that the biogas produced in the reference scenarios (KW-1 and SS-1) was sufficient to cover the heating and stirring requirements of the digester. The “extra” biogas produced as a consequence of the pretreatment application was considered to be burned in a cogeneration plant (gas engine) in order to produce both electricity and heat31 (power generation rate of 40% and a heat generation rate of 50%), and the associated air emissions of CO, CO2, CH4, and NMVOC (nonmethane volatile organic compounds) from the biogas combustion were included by using emission factors reported in literature.32 Regarding digestate application in agriculture, P and N were regarded as organic fertilizers that reduce the need of synthetic fertilizers by 70% and 50%, respectively.33 Nutrientrelated emissions (N to air as N2O and NH3 and P to water as PO43‑) were estimated by means of emission factors from literature.32 Background data related to the production of chemicals, energy, and fertilizers came from the Ecoinvent Database v2.34 36 Table 3 shows a summary of the input and output data collected or calculated for the 16 scenarios. A more detailed compilation of data can be found in the SI (Tables S1 and S2). Life Cycle Impact Assessment. This stage characterizes the environmental pressures related to the inventory by means of impact assessment models, and makes use of category indicators to condense and explain the inventory results. In this study, a well-established midpoint methodology was applied, the CML 2 baseline 2000 v2.05 implemented in the SimaPro 7.3 software (http://www.pre.nl/content/simapro-lca-software). Among the impact categories described by this method,37 the following were selected: abiotic resource depletion potential (ADP), eutrophication potential (EP), global warming potential (GWP), human toxicity potential (HTP) and terrestrial ecotoxicity potential (TTP).

’ RESULTS Table 4 displays the outcomes of the classiﬁcation and characterization steps of the impact assessment stage. The selected category indicators are separately presented since each one is expressed in its corresponding unit of reference. For a quick comparison of the diﬀerent alternatives examined, the relative performance of the individual scenarios within each category indicator (stated as the percentage ratio between the value of the scenario and the maximum value of that indicator within the same waste group) is also shown. Pressurize-depressurize (KW7, SS7) and chemical pretreatments (KW2, KW3, SS2, SS3) shared the top positions with minimum or even positive net impact on the environment, except for eutrophication, where the references (KW1, SS1) were the best scenarios. The latter is probably related to the OLR applied in each scenario, because taking into account that nutrients are hardly removed during AD (some precipitation can occur), the higher the OLR applied, the higher the amount of solids entering the digester, and thus, higher nutrients levels being discharged per FU. Among KW scenarios, KW-8 (ozone) was located at the end of the ranking in all categories, which can be explained by the high use of chemicals and energy. In the case of SS, the greatest impacts were observed in the most energydemanding scenarios, that is, SS4 (thermal), SS5 (thermo-acid) and SS6 (freeze thaw). Abiotic Resource Depletion Potential (ADP). ADP covers all potential impacts of the extraction of mineral and fossil fuels.37 Figure 1 shows the contribution of the different elements to this impact category (see Table S3 in the SI for detailed information). Taking into account that the Spanish energy profile is heavily reliant on fossil energy (80%), it was not surprising that the energy use was the main contributor to this category indicator. Therefore, the most damaging alternatives were those scenarios consuming large amounts of energy, that is, thermal (KW4, KW5, SS4, SS5) and freeze thaw (KW6, SS6) pretreatments. This fact was particularly remarkable in the SS scenarios because the volume of SS pretreated was higher (KW was pretreated without dilution). On the contrary, the consumption of chemicals had a minor impact, except for ozonation, with contributions of 62% and 57% in KW8 and SS8, respectively. This is mainly due to the higher amounts of oxygen consumed (Table 3) rather than the emission factor associated to its manufacture process (0.00301 g Sb-eq g 1 O2), which is 10308

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology

POLICY ANALYSIS

Table 4. Absolute and Relative (Within Each Waste Group) Results from the Impact Assessment Stage (See Table 2 for the Description of the Scenarios). Positive Values Mean a Negative Impact on the Environment, While Negative Values Mean a Positive Impact (Avoided Impact) on the Environmenta category indicator scenario

3‑

ADP (g Sb-eq/FU)

EP (g PO4 -eq/FU)

KW1

0.20

7.4%

1.27

35.8%

99.8

20.2%

38.1

KW2

0.16

5.7%

1.85

52.1%

106.7

21.6%

22.1

KW3

1.21

43.7%

2.06

58.0%

23.1

KW4

0.50

18.2%

1.65

46.4%

170.7

KW5 KW6

0.46 0.58

16.4% 20.9%

1.92 3.01

54.1% 84.7%

176.8 237.1

KW7

3.25

117%

2.48

KW8

2.77

100%

3.55

69.8% 100%

GWP (g CO2-eq/FU)

232.5 494.5

4.68%

HTP (g 1,4-PDB-eq/FU) 14.7% 8.53%

TTP (g 1,4-PDB-eq/FU) 1.8

22.1%

1.7

21.0%

60.8

23.2%

3.1

38.4%

34.5%

113.1

43.6%

4.6

56.8%

35.8% 48.0%

139.1 180.8

53.6% 69.6%

5.4 7.9

66.4% 97.9%

3.4%

2.7

47.0% 100%

8.7 259.6

100%

8.1

33.2% 100%

SS1

0.50

12.1%

1.71

46.1%

11.3

1.6%

548.1

49.4%

19.3

49.5%

SS2

0.66

15.8%

2.04

55.2%

21.8

3.1%

652.0

58.7%

22.7

58.4%

57.0%

22.2

SS3

2.03

49.1%

2.07

SS4

2.66

64.2%

3.70

SS5 SS6

4.15 3.97

100% 95.9%

3.53 3.54

SS7

4.67

113%

SS8

2.20

53.0%

56.1% 100% 95.5% 95.6%

128.4

18.4%

521.2

74.7%

1,110

633.1

698.0 697.5

100% 99.9%

1,029 1,047

100% 92.7% 94.3%

38.9 35.3 36.6

57.0% 100% 90.6% 94.0%

1.78

48.1%

420.7

60.3%

552.3

49.8%

20.7

53.1%

2.31

62.3%

388.6

55.7%

626.8

56.5%

19.8

50.9%

a

Names in italics indicate that those scenarios have not been experimentally tested but inventoried based on estimations. ADP: Abiotic Resource Depletion Potential; EP: Eutrophication Potential; GWP: Global Warming Potential; HTP: Human Toxicity Potential; TTP: Terrestrial Ecotoxicity Potential.

Figure 1. Contribution of diﬀerent elements to abiotic resource depletion (ADP).

the second lowest among all chemicals used in this study. Actually, oxygen was responsible for 66% and 85% of the impact associated to chemicals consumption in KW8 and SS8, respectively (SI Table S3). In most scenarios, the impact was compensated by the enhancement of the AD performance, which is reflected in the avoided products (energy and fertilizers). In this category, the avoided energy derived a greater benefit than the avoided fertilizers in most cases. Yet, it was not enough to balance the impact associated to the energy requirements of the pretreatments, except for the pressurize-depressurize method. In fact, the best results were achieved in these scenarios (KW7, SS7), which were net energy and fertilizers producers (Table 4).

Eutrophication Potential (EP). EP covers all potential impacts of excessively high environmental levels of macronutrients (mainly N and P), which may cause an undesirable shift in species composition and elevated biomass production.37 In all scenarios, nutrient-related direct emissions from the application of digestates on agricultural soil dominated the impact in this category (Figure 2 and SI Table S4). The poorest performance was observed in those scenarios where higher OLRs were applied, that is, higher solids entering the digester, and therefore, higher amounts of nutrients per FU. This impact is correlated with the benefit derived from the provision of a biosolids product that displaces the use of N- and P-based fertilizers, as both elements directly 10309

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology

POLICY ANALYSIS

Figure 2. Contribution of diﬀerent elements to eutrophication potential (EP).

Figure 3. Contribution of diﬀerent elements to global warming potential (GWP).

depend on the nutrient content (Tables 3, SI S1 and S2). The use of energy and chemicals shared the second position, each being more relevant in the energy-using (KW4, KW5, KW6, SS4, SS5, SS6) and chemical (KW2, KW3, KW8, SS2, SS3, SS8) pretreatments, respectively. In this category, the avoided products did not compensate the environmental impact, and consequently, the reference scenarios were the most preferable options for both types of waste (Table 4). Global Warming Potential (GWP). GWP is defined as the impact of human emissions on the radiative forcing of the atmosphere.37 In this work, the emissions of biogenic CO and CO2 have been disregarded (characterization factors equal to 0) according to Docka (2010).32 As expected, Figure 3 shows a relationship between energy use and the GWP, being the worst scenarios KW6, KW8, SS4, SS5, and SS6 (Table 4), all of them characterized by a high energy demand (>1.4 MJ FU 1). Behind the background emissions associated with energy production, this category is dominated by the direct emissions from biogas burning (CH4 and NMVOC) and digestate application in the soil (N2O) (SI Table S5). Only in KW8 and SS8 (ozonation), the emissions derived from the use of chemicals accounted for more

than 40% of total impact (Figure 3), once again due to the use of oxygen, whose production is highly energy intensive. In this category, the environmental benefit (negative values in figures) was always higher in the pretreatment scenarios than in the reference, and the rate of this benefit was mainly related to the increased biogas produced as a consequence of the application of the pretreatment (Figure 3 and Table 3). For both types of waste, pressurize-depressurize scenarios (KW7, SS7) performed the best, followed by the acid scenarios (KW3, SS3), all of them having net positive impact (i.e., negative category indicators) on global warming potential (Table 4). Human and Terrestrial Toxicity Potential (HTP and TTP). HTP covers the impacts of toxic substances present in the environment on human health, while TTP refers to impacts of toxic substances on terrestrial ecosystems.37 In these impact categories, there is a significant difference between both wastes (Figure 4), caused by the extremely high contribution of the direct emissions of heavy metals to soil in the SS scenarios (>63% in HTP, SI Table S6, and >58% in TTP, SI Table S7). This pattern can be explained by the nature of heavy metal pollution, mostly diffuse with industrial origin, and thus making very difficult 10310

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology

POLICY ANALYSIS

Figure 4. Contribution of diﬀerent elements to human toxicity potential (A) and terrestrial toxicity potential (B).

the control of heavy metals presence in sewage sludge.38 On the contrary, in order to protect human health, food products are very controlled and, consequently, the presence of heavy metals is not expected. This dominance of heavy metals has been found by other LCA studies,11,12 and consequently, the application of digestates in sectors where no risk of contaminating the animal or human food chain exists (e.g., floriculture) has been suggested. 13 In addition, the environmental burdens associated to HTP (Figure 4A) were more than 25-fold those of TTP (Figure 4B), which positions the human being in a more protective perspective than the ecosystem. Overall, the relative position of the different scenarios within these two categories was quite similar, being the pressure-depressure pretreatment (SS7, KW7) one of the best options, whereas thermal and freeze thaw methods showed the poorest performance due to the indirect emissions of toxic substances of the background processes associated (electricity production).

’ DISCUSSION Sixteen scenarios combining pretreatment, anaerobic digestion and agricultural disposal were environmentally modeled using life cycle assessment. The results show that the environmental burdens associated to the application of any pretreatment

prior to anaerobic digestion cannot be excluded. If for illustrative purposes, we value all the selected indicators equally, pressurizedepressurize and chemical (acid or alkaline) methods could be recommended on the basis of their net environmental performance. On the contrary, thermal, freeze thaw and ozonation alternatives would entail an environmental damage as the improvement of the AD process does not compensate the environmental burdens associated to the pretreatment. Energy vs Chemicals. Any pretreatment makes use of some form of energy (pressure, translational, rotational, thermal, or electrical) and/or chemicals and both resources can have a very diverse effect on the environment and humans. In this study, the scenarios using energy in their pretreatments possess a higher impact than those using chemicals. Therefore, though thermal pretreatments appear to be the most suitable for the improvement of waste stabilization, efforts must be done to enhance the energy balance by using the waste residual heat to maintain the temperature of the digester, by applying more efficient methods for waste disintegration such as microwave heating9 and for biogas utilization39 or by making use of a more sustainable energy production profile, that is, less dependent on fossil fuels. Among energy-using pretreatments, mechanical disintegration (i.e., pressurize-depressurize) is preferred over thermal methods due to the lower energy demand without compromising the increase in 10311

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology biogas production. Among chemical processes, category indicators invert the option preferences between acid and alkaline methods. The former performs better in terms of ADP and GWP, while the latter does in EP and toxicity potentials. Further research on alternative chemicals as well as on the optimization of the required dose are likely to entail a reduction of the environmental burdens of these scenarios. Biogas Production vs Digestate Quality. The two beneficial impacts derived from AD of solid waste are the biogas production (avoided energy) and a biosolids product (digestate) suitable for agricultural application (avoided fertilizers). When analyzing the operational effectiveness of the use of pretreatments, most researchers focus on the increase in biogas production and scarce information is provided on digestate quality. In terms of environmental performance, this study shows that the benefit derived from pretreatments application regarding biogas production is greater than in terms of improving the fertilizing capacity of digestates when both avoided products were measurable. This is probably related to the intrinsic purpose of pretreatments, that is, making organic matter available to boost biogas production. However, the more stringent limitations for biosolids application in agriculture would make practitioners to work hard to control digestate quality. Substrate Characteristics. Overall, no significant differences were observed between the analysis of kitchen waste and sewage sludge scenarios. Yet, there are some aspects that need to be addressed. Organic solids content in the waste is relevant not only because it determines the amount of nutrients and pollutants contained in the waste, affecting particularly eutrophication and toxicity potentials, but also because it establishes the loading rate at which the digesters can be run, and consequently, the biogas production rate. Moreover, substrate composition (i.e., biodegradability) will determine the effect of the pretreatments, since those wastes containing poorly digestable materials (e.g., cellulose or lignin), such as agroindustrial residues, would experience a more significant improvement in biogas production at the same cost (same energy and/or chemical consumption), thus yielding a more positive environmental impact. On the other hand, literature data were required to fill the gap of metals content in KW and differences are likely to exist between food wastes due to, for example, food legislation or habits of consumption. This demonstrates that general recommendations should be avoided and individual analyses should be conducted. Consideration of Offsets. One of the uncertainties of this study is the extent of potential offsets. The biogas produced in the reference scenarios was assumed to be sufficient to cover the energy needs of heating and stirring of the digester; however, Hospido et al. (2010)16 only compensated the heating demand and reported values for stirring of 1.68 and 0.84 kWh FU 1 for SRT of 20 and 10 days, respectively. Soda et al. (2010)40 showed that AD coupled to power generation processes resulted in excess energy production if high sludge-loading rates were applied. The inclusion of energy for stirring in this study would increase significantly the environmental burdens of all scenarios, especially those of KW due to the 2-fold SRT applied. Yet, since this study is focused on a comparative analysis, the ranking of the scenarios would not be affected. Methodology Aspects. In this study, biogenic CO2 and CO emissions have been excluded from the impact assessment. However, according to Griffith et al. (2009),41 around 20% of the total organic carbon found in wastewater has a fossil origin and therefore the figures obtained on GWP for SS scenarios would have

POLICY ANALYSIS

been underestimated. A sensitivity analysis was performed on the SS scenarios assuming the distribution reported,41 that is, 20% of fossil origin and 80% of biogenic C source, and the results revealed that, although the direct emissions associated to biogas combustion increased 3-fold, the general conclusions and the option preferences would not be reversed. Another aspect to be mentioned is the high significance that, in general, impact assessment methodologies give to heavy metals in comparison to other pollutants,42 44 probably as a combined result of the well-established toxicity models existing for these compounds and their bioaccumulative character. The FU has been here defined in terms of the system input as the function of our system was to guarantee the appropriate management of a waste stream rather than obtaining a particular product. In fact, our system had a combined objective: increase biogas production and provide a final substance with fertilizing value. Therefore, the definition of the FU in an output basis as recommended by other authors25 was not appropriate. Finally, we have used a “well to tank” approach when defining the boundaries related to the final use of energy (both heat and electricity) instead of a “well to wheel” system as suggested by other authors25 due to the comparative characteristics of this study. Sustainable Pretreatments Application. Reconciling operational, economic and environmental aspects for sustainable pretreatment application is not straightforward. Yet, this study in combination with financial and operational data from literature5,6,8,9,45,46 suggests that mechanical (e.g., pressurizedepressurize) and chemical pretreatments appear to be the most suitable at this stage. Further work may examine full-scale experience and a more integrated and energy-efficient scheme of waste management with the inclusion of subsequent digested solids treatment processes (dewatering, transportation, spreading) and biogas utilization pathways.

’ ASSOCIATED CONTENT

bS

Supporting Information. Assumptions made during the inventory analysis of nonexperimentally tested scenarios, additional inventory data for KW and SS scenarios and contribution of diﬀerent elements to abiotic resource depletion potential, eutrophication potential, global warming potential, human toxicity potential and terrestrial toxicity potential. This material is available free of charge via Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +34-881-816020; fax: +34-881-816015; e-mail: marta. [email protected] (M.C.), [email protected] (A.H.). Author Contributions

Both authors equally contributed to the work.

’ ACKNOWLEDGMENT This research was funded by postdoctoral contracts from the Xunta de Galicia for Dr. Marta Carballa (IPP-08-37) and Dr. Almudena Hospido (IPP-06-57). We also thank the Spanish Ministry of Education and Science (Project CTM2010-17196) and the Xunta de Galicia (Projects 09MDS010262PR and GRC2010/37) for its ﬁnancial assistance. Dr. Jingxing Ma and 10312

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology Thu Hang Duong are acknowledged for their well-done experimental work.

’ REFERENCES (1) Appels, L.; Baeyens, J.; Degreve, J.; Dewil, R. Principles and potential of the anaerobic digestion of waste-activated sludge. Prog. Energy Combust. Sci. 2008, 34 (6), 755–781. (2) Sludge into Biosolids. Processing, Disposal and Utilization, 1st, ed.; Spinosa, L, Vesilind, P., Eds.; IWA Publishing: London, 2001. (3) Lin, Y.; Wang, D.; Wu, S.; Wang, C. Alkali pretreatment enhances biogas production in the anaerobic digestion of pulp and paper sludge. J. Hazard. Mater. 2009, 170 (1), 366–373. (4) Dohanyos, M.; Zabranska, J.; Kutil, J.; Penicek, P. Improvement of anaerobic digestion sludge. Water Sci. Technol. 2004, 49 (10), 89–96. (5) Carballa, M.; Omil, F.; Alder, A. C.; Lema, J. M. Comparison between the conventional anaerobic digestion of sewage sludge and its combination with a chemical or thermal pre-treatment concerning the removal of pharmaceuticals and personal care products (PPCPs). Water Sci. Technol. 2006, 53 (8), 109–117. (6) Carballa, M.; Manterola, G.; Larrea, L.; Ternes, T.; Omil, F.; Lema, J. M. Inﬂuence of ozone pre-treatment on sludge anaerobic digestion: Removal of pharmaceutical and personal care products. Chemosphere 2007, 67, 1444–1452. (7) Carrere, H.; Dumas, C.; Battimelli, A.; Batstone, D.; Delgenes, J.; Steyer, J.; Ferrer, I. Pretreatment methods to improve sludge anaerobic degradability: A review. J. Hazard. Mater. 2010, 183, 1–15. (8) Ma, J.; Duong, T. H.; Smits, M.; Verstraete, W.; Carballa, M. Enhanced biomethanation of kitchen waste by diﬀerent pre-treatments. Biores. Technol. 2011, 102, 592–599. (9) Bordeleau, E. L.; Droste, R. L. Comprehensive review and compilation of pretreatments for mesophilic and thermophilic anaerobic digestion. Water Sci. Technol. 2011, 63 (2), 291–296. (10) Guinee, J. B.; Heijungs, R.; Huppes, G.; Zamagni, A.; Masoni, A.; Buonamici, R.; Ekvall, T.; Rydberg, T. Life cycle assessment: Past, present and future. Environ. Sci. Technol. 2011, 45, 90–96. (11) Suh, Y. J.; Rousseaux, P. An LCA of alternative wastewater sludge treatment scenarios. Resour. Conserv. Recycl. 2002, 35 (3), 191–200. (12) Hospido, A.; Moreira, T.; Rigola, M.; Feijoo, G. Environmental evaluation of diﬀerent treatment processes for sludge from urban wastewater treatments: anaerobic digestion versus thermal processes. Int. J. Life Cycle Assess. 2005, 13 (1), 57–64. (13) Tarantini, M.; Butol, P.; Marorino, L. An environmental LCA of alternative scenarios of urban sewage sludge treatment and disposal. Therm. Sci. 2007, 11 (3), 153–164. (14) Murray, A.; Horvath, A.; Nelson, K. Hybrid life-cycle environment and cost inventory of sewage sludge treatment and end-use scenarios: a case study from China. Environ. Sci. Technol. 2008, 42, 3163–3169. (15) Peters, M.; Rowley, H. Environmental comparison of biosolids management systems using life cycle assessment. Environ. Sci. Technol. 2009, 43 (8), 2674–2679. (16) Hospido, A.; Carballa, M.; Moreira, M.; Omil, F.; Lema, J. M.; Feijoo, G. Environmental assessment of anaerobically digested sludge reuse in agriculture: Potential impacts of emerging micropollutants. Water Res. 2010, 44, 3225–3233. (17) Chaya, W.; Gheewala, S. H. Life cycle assessment of MSW-toenergy schemes in Thailand. J. Clean Prod. 2007, 15 (15), 1463–1468. (18) Clearly, J. Life cycle assessments of municipal solid waste management systems: A comparative analysis of selected peer-reviewed literature. Environ. Int. 2009, 35, 1256–1266. (19) De Feo, G.; Malvano, C. The use of LCA in selecting the best MSW management system. Waste Manage. 2009, 29, 1901–1915. (20) Zhao, W.; van der Voet, E.; Zhang, Y.; Huppes, G. Life cycle assessment of municipal solid waste management with regard to greenhouse gas emissions: Case study of Tianjin, China. Sci. Total Environ. 2009, 407, 1517–1526.

POLICY ANALYSIS

(21) Bovea, M. D.; Iba~ nez-Fores, V.; Gallardo, A.; ColomerMendoza, F. J. Environmental assessment of alternative municipal solid waste management strategies. A Spanish case study. Waste Manage. 2010, 30, 2383–2395. (22) Lee, S. H.; Choi, K. I.; Osako, M.; Dong, J. I. Evaluation of environmental burdens caused by changes of food waste management systems in Seoul, Korea. Sci. Total Environ. 2007, 387, 42–53. (23) Khoo, H. H; Lim, T. Z.; Tam, R. B. H. Food waste conversion options in Singapore: Environmental impacts based on an LCA perspective. Sci. Total Environ. 2010, 408, 1367–1373. (24) Kim, M. H.; Kim, J. W. Comparison through a LCA evaluation analysis of food waste disposal options from the perspective of global warming and resource recovery. Sci. Total Environ. 2010, 408, 3998– 4006. (25) Singh, A.; Pant, D.; Korres, N. E.; Nizami, A. S.; Prasad, S.; Murphy, J. D. Key issues in life cycle assessment of ethanol production from lignocellulosic biomass: Challenges and perspectives. Biores. Technol. 2010, 101, 5003–5012. (26) McDougall, F.; White, P.; Franke, M.; Hindle, P. Integrated Solid Waste Management: A Life Cycle Inventory; Blackwell Science Ltd: Oxford, U.K., 2001. (27) Carballa, M.; Omil, F.; Ternes, T.; Lema, J. M. Fate of pharmaceutical and personal care products (PPCPs) during anaerobic digestion of sewage sludge. Water Res. 2007, 41, 2139–2150. (28) Duong, T. Pre-treatments to enhance the biomethanation of kitchen waste. Master Thesis Dissertation, Ghent University, Ghent, Belgium, 2009. (29) Carballa, M.; Omil, F.; Lema, J. M. Inﬂuence of diﬀerent pretreatments on anaerobically digested sludge characteristics: suitability for ﬁnal disposal. Water, Air, Soil, Pollut 2009, 119 (1 4), 311–321. (30) Zhang, R.; Hamed, M.; Hartman, K.; Gamble, P. Characterization of food waste as feedstock for anaerobic digestion. Biores. Technol. 2007, 98, 929–935. (31) Pertl, A.; Mostbauer, P.; Obersteiner, G. Climate balance of biogas upgrading systems. Waste Manage. 2010, 30 (1), 92–99. (32) Docka, G. Life cycle inventories of waste treatment services. In Ecoinvent Report No. 7; D€uberdorf: Switzerland, 2007. (33) Bengtsson, M.; Lundin, M.; Molander, S. Life Cycle Assessment of Wastewater Systems: Case Studies of Conventional Treatment, Urine Sorting and Liquid Composting in Three Swedish Municipalities, Report No. 9; Chalmers University of Technology: Gothenburg, Sweden, 1997. (34) Althaus, H.-J.; Chudacoﬀ, M.; Hischier, R.; Jungbluth, N.; Osses, M.; Primas, A. Life Cycle Inventories of Chemicals, Final report ecoinvent data v2.0 No. 8; Swiss Centre for Life Cycle Inventories: D€ubendorf, Switzerland, 2007. (35) Frischknecht, R.; Tuchschmid, M.; Faist-Emmenegger, M.; Bauer, C.; Dones, R. Strommix and Stromnetz, Ecoinvent Report No.6-XVI; Uster: Switzerland, 2007. (36) Nemecek, T.; K€agi, T. Life Cycle Inventories of Agricultural Production Systems, Ecoinvent Report No.15; Z€urich and D€ubendorf: Switzerland, 2007. (37) Guinee, J. B.; Gorree, M.; Heijungs, R.; Huppes, G.; Kleijn, R.; de Koning, A.; van Oers, L. A.; Weneger, A.; Suh, S.; Udo de Haes, H. A.; de Bruijn, H.; van Duin, R. y Huijbregts, M. Life Cycle Assessment: An Operational Guide to the ISO Standards; Leiden, The Netherlands, 2001. (38) S€orme, L.; Lindqvist, A.; S€oderberg, H. Capacity to inﬂuence sources of heavy metals to wastewater treatment sludge. Environ. Manage. 2003, 31 (3), 421–428. (39) P€oschl, M.; Ward, S.; Owende, P. Evaluation of energy eﬃciency of various biogas production and utilization pathways. Appl. Energy 2010, 87, 3305–3321. (40) Soda, S.; Iwai, Y.; Sei, K.; Shimod, Y.; Ike, M. Model analysis of energy consumption and greenhouse gas emissions of sewage sludge treatment systems with diﬀerent processes and scales. Water Sci. Technol. 2010, 61 (2), 365–373. (41) Griﬃth, D. R.; Barnes, R. T.; Raymond, P. A. Inputs of fossil carbon from wastewater treatment plants to U.S. rivers and oceans. Environ. Sci. Technol. 2009, 43, 5647–5651. 10313

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

Environmental Science & Technology

POLICY ANALYSIS

(42) Dreyer, L. C.; Niemann, A. L.; Hauschild, M. Z. Comparison of three diﬀerent LCIA methods: EDIP97, CML2001 and Ecoindicator 99. Does it matter which one you choose? Int J. Life Cycle Assess. 2003, 8 (4), 191–200. (43) Pizzol, M.; Christensen, P.; Schmidt, J.; Thomsen, M. Ecotoxicological impact of “metals” on the aquactic and terrestrial ecosystem: a comparison between eight diﬀerent methodologies for Life Cycle Impact Assessment (LCIA). J. Cleaner Prod. 2011, 19, 687–698. (44) Pizzol, M.; Christensen, P.; Schmidt, J.; Thomsen, M. Impacts of “metals” on human health: a comparison between nine diﬀerent methodologies for Life Cycle Impact Assessment (LCIA). J. Clean Prod. 2011, 19, 646–656. (45) M€uller, J. A. Prospects and problems of sludge pre-treatment processes. Water Sci. Technol. 2001, 44 (10), 121–128. (46) Perez-Elvira, S. I.; Nieto Diez, P.; Fdz-Polanco, F. Sludge minimisation technologies. Rev. Environ. Sci. Biotechnol. 2006, 5, 375–398.

10314

dx.doi.org/10.1021/es201866u |Environ. Sci. Technol. 2011, 45, 10306–10314

POLICY ANALYSIS pubs.acs.org/est

Beach Monitoring Criteria: Reading the Fine Print Meredith B. Nevers* and Richard L. Whitman U.S. Geological Survey, Great Lakes Science Center, Lake Michigan Ecological Research Station, 1100 N. Mineral Springs Road, Porter, Indiana 46304, United States ABSTRACT: Beach monitoring programs aim to decrease swimming-related illnesses resulting from exposure to harmful microbes in recreational waters, while providing maximum beach access. Managers are advised by the U.S. EPA to estimate microbiological water quality based on a 5-day geometric mean of fecal indicator bacteria (FIB) concentrations or on a jurisdiction-speciﬁc single-sample maximum; however, most opt instead to apply a default single-sample maximum to ease application. We examined whether re-evaluation of the U.S. EPA ambient water quality criteria (AWQC) and the epidemiological studies on which they are based could increase public beach access without aﬀecting presumed health risk. Single-sample maxima were calculated using historic monitoring data for 50 beaches along coastal Lake Michigan on various temporal and spatial groupings to assess ﬂexibility in the application of the AWQC. No calculation on either scale was as low as the default maximum (235 CFU/100 mL) that managers typically use, indicating that current applications may be more conservative than the outlined AWQC. It was notable that beaches subject to point source FIB contamination had lower variation, highlighting the bias in the standards for these beaches. Until new water quality standards are promulgated, more site-speciﬁc application of the AWQC may beneﬁt beach managers by allowing swimmers greater access to beaches. This issue will be an important consideration in addressing the forthcoming beach monitoring standards.

’ INTRODUCTION Passage of the U.S. BEACH Act1 required that all coastal recreational waters be monitored for fecal indicator bacteria (FIB) starting in 2004. With the initiation of numerous programs and expansion of existing programs, a wealth of data has been generated by monitoring agencies from coastal beaches, including Great Lakes beaches. With the increase in data generation, more instances of high FIB concentrations have been detected, resulting in a higher overall number of beach closures and the impression that beach water quality is universally declining,2 despite a lack of supporting information. Negative publicity, combined with known limitations of using FIB as an indicator—i.e., lengthy analysis time, natural sources3—has been a likely disincentive to expand beach monitoring beyond minimal requirements. While the existing ambient water quality criteria (AWQC) originally developed in 1986 are under revision by U.S. EPA, beach managers are obliged to monitor their beaches using the currently accepted FIB standards until new standards are promulgated; however, there is underused ﬂexibility already integrated into the existing AWQC that could beneﬁt the public and management. The AWQC developed by the U.S. EPA for freshwater were derived from epidemiological studies conducted in 19791982 at four beaches on two lakes directly inﬂuenced by point source contamination.4 The criteria deﬁne an acceptable illness rate of 8/1000 swimmers and an FIB standard; water with FIB concentrations in This article not subject to U.S. Copyright. Published 2011 by the American Chemical Society

excess of the criteria are out of compliance, and beach managers typically close the beach to swimming or issue a swimming advisory. The AWQC primarily recommend that beach management decisions be based on a geometric mean calculation of water quality of at least ﬁve samples collected over the previous 30-day period to estimate the steady-state mean; this is to prevent unnecessary closures/advisories due to day to day ﬂuctuations in FIB. Decisions about implementing the AWQC standards are the responsibility of the individual states, and most states opt to use a single sample maximum limit (ssmax) (e.g., refs 510), also presented in the AWQC, rather than the ﬁve-day mean despite the document’s urging that such a decision “may be erroneous”.11 Further, most states use the default ssmax (ssmaxEPA) of 235 colony-forming units (CFU) E. coli/100 mL water calculated by EPA and based on standard deviations derived in the epidemiological studies, disregarding the AWQC recommendation that “each jurisdiction should establish its own standard deviation for its conditions which would then vary the single sample limit”.11 FIB may exhibit high spatial-temporal variation in beach waters, the extent of which varies among beaches.1214 For this reason, Received: July 25, 2011 Accepted: November 7, 2011 Revised: October 27, 2011 Published: November 07, 2011 10315

dx.doi.org/10.1021/es202568f | Environ. Sci. Technol. 2011, 45, 10315–10321

Environmental Science & Technology the AWQC provide ﬂexibility in the form of establishing jurisdiction-speciﬁc ssmax (ssmaxCALC). An understanding of variation in local bacteria densities would beneﬁt swimmers and beach managers because a monitoring standard can be developed that may be as protective of public health risk as the default criteria outlined in the AWQC. In this study, we examine historical beach monitoring data for beaches along southern Lake Michigan, extending from Chicago through coastal Indiana, and we calculate ssmaxCALC, partitioning data in a number of diﬀerent spatial and temporal scales to highlight the potential variation in monitoring outcomes. Further, we explore the implications of this assessment on beach monitoring outcomes, expectation of health protection, and beach access. The results of this study emphasize typically overlooked details included in the AWQC and the adaptability of monitoring protocols, with implications for both the current and forthcoming ambient water quality criteria.

’ METHODS Beach monitoring data were obtained from several existing databases maintained by the Chicago Park District, the Indiana Department of Environmental Management, and the Indiana Dunes National Lakeshore;15,16 included in the analysis were 50 beaches that cover the majority of the Great Lakes coasts of Illinois and Indiana. These beaches are monitored for E. coli, one of the fecal indicator bacteria, as recommended for freshwater beaches in the AWQC. Frequency of monitoring at a given beach ranged from 1 to 7 days a week, and the available historical monitoring data ranged from 6 to 21 years. Beach monitoring is under the jurisdiction of numerous regulatory agencies, which use diﬀerent sampling replication and averaging protocols and have diﬀerent regulations regarding swimming restrictions when water quality is out of compliance.3 Although sampling depth, location, and time can have a signiﬁcant impact on E. coli results outcome,3 as well as analytical method (deﬁned substrate or membrane ﬁltration),17 we evaluate the data “as is”; that is, as submitted to regulatory agencies and U.S. EPA to satisfy requirements of the BEACH Act and therefore considered adequate for monitoring recreational water quality. For the time scale analysis, data from ﬁve beaches in Indiana were used that were collected during a period of 21 years (1990 2010). These beaches are directly inﬂuenced by the outfall of the Little Calumet River (Burns Ditch): from west (closest to the river outfall) to east, the beaches include Ogden Dunes (Ogden), West, Wells Street (Wells), Marquette Park (Marquette), and Lake Street (Lake) Beaches. In further analyses of spatial patterns, data collected from 2004 to 2010 at 50 subject beaches were used for more intensive analyses because of the higher collection frequency and presumed better characterization of overall water quality. Beaches in this region are aﬀected in diﬀerent ways by point and nonpoint sources of fecal contamination. Spatial analysis of Lake Michigan beaches included grouping the 50 beaches into 5 geographic regions: Chicago (22 beaches in Chicago, IL); Lake (7 beaches in western Lake County, IN); Burns Ditch (5 beaches inﬂuenced by the Burns Ditch outfall of the Little Calumet River in IN); Indiana Dunes (8 beaches in the Indiana Dunes State Park and National Lakeshore); and LaPorte (8 beaches in LaPorte County, IN). These designations are based jointly on geographic location, management jurisdiction, and speciﬁc point source inﬂuence.

POLICY ANALYSIS

Single-sample maxima (ssmax) for freshwater were calculated based on the AWQC11 developed in epidemiological studies conducted at two locations in the United States:4 ssmax ¼ 10∧ ðlog10 GM þ fZ SDgÞ or ssmax ¼ GM 10∧ ðZ SDÞ where ssmax = the single sample maximum limit; GM = 126, geometric mean E. coli concentration developed in the epidemiological study for acceptable illness rate of 8 per 1000; Z = 0.675, the 75% calculated one-sided conﬁdence level for a designated beach area; and SD is the calculated standard deviation of the singlesample log10 E.coli concentrations. In the AWQC, a geometric mean E. coli concentration for a minimum of 5 samples collected over a 30-day period in excess of 126 CFU/100 mL is considered to be out of compliance (GM = 126) and corresponds to an acceptable swimming-associated illness rate of 8 per 1000 swimmers. The AWQC also consider a single sample with a concentration in excess of 235 CFU/100 mL to be out of compliance. This is based on a control chart approach, with the upper control limit being the 75th percentile for a geometric mean of 126, and using a standard deviation of 0.4 log, as calculated from the epidemiological studies.4 With the recommendation of a beach- or jurisdiction-speciﬁc calculation, a beach that has a higher standard deviation in E. coli concentrations, indicative of higher variation overall, would have a higher ssmax, regardless of the water quality. Similarly, a beach with a lower standard deviation and lower variation in E. coli concentrations would have a lower ssmax. Data were analyzed using Systat 12.018 and SPSS 12.019 software. Overall FIB concentrations were compared across beaches in the spatial variance section using analysis of variance with P < 0.05. Standard deviation for monitoring data was calculated by bootstrapping a subset of the entire 20042010 data set. A total of 100 calculations of standard deviation were made using 100 randomly selected monitoring results without replacement, after this sample size was determined to be adequate to represent conﬁdence limits in a power analysis. The standard deviation and 95% conﬁdence interval are reported. The number of instances of FIB concentration exceeding the ssmaxCALC vs the ssmaxEPA were compared using the nonparametric McNemar test or the Fisher exact test where data distribution was uneven (P < 0.05).

’ RESULTS Temporal Variation. A review of data across time for a group of five Indiana beaches directly influenced by a point source (Burns Ditch beaches) indicates that variation in the ssmaxCALC depends on time range considered in the calculation. E. coli means and standard deviations calculated from all available monitoring data were highly variable year to year and did not follow a general trend, so ssmaxCALC fluctuated from a low of 253 in 1991 to a high of 430 in 2003 (Figure 1). Calculating across a 4-year, moving average smoothed the variation, resulting in a lower range of ssmaxCALC (296387; Figure 1). Cumulative calculation over the 21 years further smoothed variation, resulting in a narrower range but higher overall ssmaxCALC for any given year (304353; Figure 1). Regardless of calculation method, standard deviation and therefore ssmaxCALC for these beaches was consistently 10316

dx.doi.org/10.1021/es202568f |Environ. Sci. Technol. 2011, 45, 10315–10321

Environmental Science & Technology

POLICY ANALYSIS

Figure 1. Single-sample maxima (ssmaxCALC), as calculated from the EPA ambient water quality criteria,5 for ﬁve Indiana beaches along Lake Michigan using three methods of subdividing the data. Annual is an ssmaxCALC for each individual year; cumulative is the ssmaxCALC for each year using the target and all previous years of available data; 4 yr-running is the ssmaxCALC using the target year and previous three. The ssmaxEPA is the default 235 ssmax provided in the AWQC and is provided for reference. N for each individual year ranged from 33 (in 1991) to 383 (in 2004).

Figure 2. Calculated single-sample maxima (ssmaxCALC), as deﬁned in the EPA ambient water quality criteria5 for individual beaches along southern Lake Michigan and the entire area. N = 100 for each beach; box and lines indicate the ssmaxCALC with 95% conﬁdence limit.

higher than the ssmaxEPA currently used in the local beach monitoring programs. Spatial Variation. The increase in monitoring frequency in 2004 significantly increased the number of samples considered, therefore, our examination of spatial calculations of ssmaxCALC incorporated monitoring data from 2004 to 2010. Analysis of data across a range of spatial scales also showed high variation in E. coli concentrations and ssmaxCALC (Figure 2). Of the 50 beaches studied, four stood out with significantly higher E. coli concentrations: Jeorse Park, 63rd Street, Washington Park, and Buffington beaches (F = 64.694, df = 49, P < 0.01), all of which have a history of frequently elevated E. coli events.2,20 Only Washington Park, however, is situated immediately adjacent to a point source outfall; FIB sources at any of these beaches have not been definitively identified. For individual beaches, ssmaxCALC included a wide range: 261 CFU/100 mL at Washington Park, IN to 407

CFU/100 mL at Buffington Harbor, IN (Figure 2). Beaches with lower standard deviations/ssmaxCALC often included locations directly influenced by a point source outfall (Washington Park, Ogden, Lake). For a subset of beaches known to be directly inﬂuenced by a point source outfall,21 Burns Ditch, there was relatively less variation between ssmaxCALC for individual beaches (range 287302 CFU/100 mL). E. coli concentrations in general were signiﬁcantly higher at Lake and Marquette than at the other three beaches (impacted by Burns Ditch) according to an ANOVA (F = 40.373, df = 4, P < 0.01). Lake had the lowest standard deviation and therefore the lowest ssmaxCALC: 285 CFU/100 mL; the highest ssmaxCALC was associated with West (300 CFU/100 mL). When beaches were divided into ﬁve groupings, analysis of variance revealed a signiﬁcant diﬀerence across regions (F = 110.973, df = 4, P < 0.01), with Lake County and Chicago beaches having 10317

dx.doi.org/10.1021/es202568f |Environ. Sci. Technol. 2011, 45, 10315–10321

Environmental Science & Technology

POLICY ANALYSIS

Table 1. Values of ssmaxCALC for Each of Five Designated Regions along Southern Lake Michigan, Using Monitoring Data Collected 20042010a region Chicago

N

region ssmaxCALC

% exceeding ssmaxEPA

% exceeding region ssmaxCALC

diﬀerence in number of beach days

11248

394

17.4

11.3

681

Lake Burns Ditch

3951 2217

401 304

24.2 7.9

17.9 6.0

248 43

Indiana Dunes

2003

318

11.5

7.9

73

LaPorte

3149

328

9.2

5.9

102

a

Percent of beach days exceeding the default ssmaxEPA and exceeding ssmaxCALC speciﬁcally for a given region are provided. Using a region-speciﬁc ssmaxCALC would result in a higher number of beach days meeting the AWQC.

Figure 3. Outline of southern Lake Michigan beach locations showing variation in the standard deviation of E. coli concentrations (color gradations) and % diﬀerence in number of days exceeding the E. coli concentration ssmaxEPA vs ssmaxCALC (size gradations).

signiﬁcantly higher means (Figure 2). The lowest standard deviation and ssmaxCALC were associated with the Burns Ditch beaches (303 CFU/100 mL); this characteristic supports the application of the monitoring standard at point source locations, although the standard deviation was notably higher (0.567) than the 0.4 used in the AWQC. The highest standard deviation and ssmaxCALC was associated with the Lake County beaches (400 CFU/100 mL); these beaches are subject to frequently elevated E. coli concentrations and high ﬂuctuations overall, although no source of contamination has been adequately identiﬁed. Using data from the entire 50 beach data set, the ssmaxCALC for the entire southern Lake Michigan crescent was 376 CFU/100 mL (SD = 0.703).

Effect of Alternate ssmax on Number of Beach Advisories. Use of an ssmax based on local water quality would have resulted in an increase in beach access (open beaches) because in all instances ssmaxCALC was higher than the default 235 CFU/ 100 mL in the AWQC. Results of a nonparametric McNemar test indicated that use of ssmaxCALC for all regions (P < 0.01) and the southern Lake Michigan region (McNemar chi-squared = 1125.0, df = 1, P < 0.01) resulted in significantly fewer instances of beaches exceeding the ssmax (Table 1). Results for individual beaches also indicated that the higher ssmaxCALC resulted in signiﬁcantly fewer out-of-compliance events (P < 0.01) (Figure 3), according to the McNemar test. Beaches 10318

dx.doi.org/10.1021/es202568f |Environ. Sci. Technol. 2011, 45, 10315–10321

Environmental Science & Technology associated with natural, undeveloped areas without a nearby point source were more likely not to exhibit a signiﬁcant diﬀerence and included Central, Dunbar, Kemil, Long Beach, and Porter. Kemil had no samples collected with an E. coli count that exceeded either standard. Based on these historical monitoring data, use of a regional ssmaxCALC decreased the percent of instances when the beach water was out of compliance, when compared to the ssmaxEPA. This decrease ranged from 2 to 6%, which was the equivalent of 43681 beach days (Table 1). Using ssmaxCALC for individual beaches, the number of days out of compliance decreased in a range from 0 to 11% (Figure 3).

’ DISCUSSION Regardless of how the data were partitioned, temporally or spatially, all ssmaxCALC were higher than the standard ssmaxEPA currently applied by beach managers in southern Lake Michigan to determine when swimming water is out of compliance, according to the AWQC. This indicates that many current monitoring applications in this region and perhaps nationwide may be overly conservative relative to the recommendations of the AWQC. The AWQC are currently under revision by the U.S. EPA; until any upcoming changes are promulgated, beach managers are likely to monitor their beaches using the currently accepted FIB standards since retrospective analyses and applications for revisions are cumbersome. The AWQC, as we highlight, provide allowances for the wide range of waters subject to assessment; these could beneﬁt practical beach management and should be carefully considered in development of the new criteria. The great discrepancy between ssmaxEPA and ssmaxCALC may stem from the calculation used in epidemiological studies.4 The epidemiological studies were conducted at four freshwater beaches over two years, plus one additional beach/year, and included the collection of an unidentiﬁed number of days of indicator bacteria data along with interviews from thousands of beachgoers.4 These data were simpliﬁed to a single mean E. coli concentration and number of illnesses for each beach/year, and from the resulting nine data pairs, regression analysis was conducted.4 Further, the standard deviation from which the ssmaxEPA is calculated is based on the standard deviation of the log10 mean E. coli concentrations. This type of averaging will reduce variation, resulting in a lower standard deviation, and perhaps explaining the diﬀerence between those results and the analyses presented here. Calculation of ssmaxCALC must consider the temporal and spatial extent over which data are averaged. Neither of these factors are speciﬁcally mentioned in the AWQC, but Chawla and Hunter22 recommended periods of 34 years; this was subsequently included in the European Union directive.23 Using this approach, we calculated a 4-year running average, and it was notable that in the earlier years the standard deviation was higher (0.530.79) and the ssmaxCALC range was wider (335387 CFU/100 mL), prior to the added inﬂux of data starting in 2004 (range SD 0.510.66; ssmaxCALC 296324 CFU/100 mL). The increase in number of observations, in general, increases conﬁdence in the estimate of overall water quality.11 Use of a single year of data could result in highly variable ssmaxCALC due to interyear variation in FIB concentrations resulting from rainfall, sewer overﬂows, diﬀerences in sources, and hydrometeorological eﬀects.

POLICY ANALYSIS

Basing overall FIB estimates of variation on sources or similar geographic areas may provide a better estimate of overall variation in FIB. Research has indicated that background FIB ﬂuctuations are similar across beach regions as long as 35 km,20,24 but variation within and between beaches may warrant use of shorter lengths of coast to account for local sources and circulation patterns.13,14 In each beach group in this analysis, the ssmaxCALC was higher than 300 CFU/100 mL. Simplifying across the entire 50 beaches resulted in an ssmaxCALC well above the default 235 CFU/100 mL generally used (Figure 3). Use of an ssmaxCALC for speciﬁc regions resulted in increased access (i.e., fewer days out of compliance) to beaches at all regional divisions, with an increase, for example, of as many as 681 days of beach access overall for Chicago beaches. One of the ﬁve designated regions directly downcurrent of a point source (Burns Ditch) had the least overall beneﬁt because it had the lowest ssmaxCALC. The higher ssmaxCALC for all regions, even an ssmaxCALC for the entire southern Lake Michigan region, would allow for more beach access days. Previous estimates of regionspeciﬁc ssmaxCALC also indicated improvements in prediction success using empirical predictive models that can potentially provide results in less than an hour compared to current analysis techniques that can take 2448 h.25 Because of the high variation in E. coli among beaches, use of the default ssmaxEPA despite the recommendation for a sitespeciﬁc ssmaxCALC results in the application of diﬀerential illness risk across beaches. Two components of the ssmaxEPA are the acceptable illness rate set at 8/1000 and the 75% conﬁdence limits (around the geometric mean of 126 CFU/100 mL); use of ssmaxEPA when the standard deviation of a water body is diﬀerent from 0.4 aﬀects both of these assumptions. If the conﬁdence limits are controlled at 75%, use of ssmaxEPA in a situation where ssmaxEPA < ssmaxCALC results in the application of a lower level of risk tolerance, that is, management is “overprotective”. For example, at Buﬃngton, where logSD = 0.755 and ssmaxCALC = 407 CFU/100 mL, use of ssmaxEPA (235 CFU/100 mL) for beach management would eﬀectively protect for an acceptable illness rate of 5.74/1000. Alternately, if the acceptable illness rate is controlled at 8/1000, use of ssmaxEPA when ssmaxEPA < ssmaxCALC results in a reduction of the conﬁdence limit; from 75% in the AWQC standards to 64% in the case of Buﬃngton. The implications of the violation of these assumptions will need to be considered in the development of new standards because of the potential for uneven human health protection across beaches, regions, or states. Accuracy of risk estimates in the AWQC are hindered by the lack of data collected that correspond to high illness rates: because illness rate must be projected beyond a certain level, there is a high rate of error in estimates at higher bacteria/illness levels. Therefore, accuracy in presumptions of illness risk decreases at higher levels. Interpretation of higher illness rates and higher calculations of ssmaxCALC should take this into consideration. For this reason, studies that estimate risk based on concentrations of known pathogens (e.g, quantitative microbial risk assessment, QMRA26) may be a viable solution for more accurately estimating acceptable water quality. These evaluations determine the public health outcome with exposure to water contaminated with fecal pathogens (e.g., Norovirus, Cryptosporidium, Giardia). This technique may allow estimates of illness rates at higher levels of indicator bacteria, without extensive epidemiological studies. Although not considered in this analysis, the original AWQC also outlined the use of wider conﬁdence intervals for swimming 10319

dx.doi.org/10.1021/es202568f |Environ. Sci. Technol. 2011, 45, 10315–10321

Environmental Science & Technology areas with lower use, although use is not quantiﬁed in the criteria. The beaches considered in this analysis have a wide range of use, so those allowances may apply to some of these waters. Numerous beaches in Great Lakes locations could further consider these wider conﬁdence intervals in calculating and using an ssmaxCALC for monitoring. Use of the current FIB monitoring standard has been questioned due to inconsistencies in development and application: speciﬁcally, use of data from rain-free days, time period averaging, and the presence of a point source sewage impact. The low ssmaxCALC at the Burns Ditch beaches is reﬂected in a related study of E. coli variation conducted by U.S. EPA at one of these beaches (West);27 here, the ssmaxCALC, based on the standard deviation (0.538), was comparable to the results presented here: 291 CFU/100 mL. At the other freshwater beach studied by Wymer et al.,27 Belle Isle in Michigan, the resulting ssmaxCALC was 255 CFU/100 mL (SD = 0.453). The lowest ssmaxCALC for an individual beach in this study was Washington Park (261 CFU/100 mL; SD = 0.469), which is also aﬀected by a point source, Trail Creek. These comparisons support the strength of this statistic for point source impacted beaches. It should be noted, however, that all coastal beaches are required to incorporate these standards into a beach monitoring program, regardless of whether a point source is present. Our study illustrates that data averaging spatially and temporally is an important consideration in the development and implementation of new water quality criteria for both fresh and marine waters. Others have suggested the use of a more ﬂexible system of standards with more gradations to designate water quality,28 as opposed to the binary system inherent in the AWQC. Even with the adoption of new standards, however, implementation will likely take some time, and beach managers may re-evaluate the variable application of the current AWQC to allow greater access to beaches presumably without additional public health risk.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 219-926-8336 ext. 425; fax: 219-929-5792; e-mail: [email protected].

’ ACKNOWLEDGMENT This research was funded by the USGS Ocean Research Priorities Plan and the Great Lakes Restoration Initiative through USGS. Engaging discussions with Shannon Briggs (Michigan Department of Environmental Quality) and reviews by Samir Elmir (Florida Department of Health), Jean Adams (USGS), and anonymous reviewers helped us to improve this manuscript. Photo of North Avenue Beach, Chicago, Illinois; credit: Antonio Vernon. Any use of trade, product, or ﬁrm names is for descriptive purposes only and does not imply endorsement by the U.S. Government. This article is Contribution 1670 of the USGS Great Lakes Science Center. ’ REFERENCES (1) Beaches Environmental Assessment and Coastal Health Act, 33 USC 1251. Public Law 106284, 114 Stat. 870877, 2000; Vol. 33 USC 1251. (2) Dorfman, M.; Rosselot, K. S. Testing the Waters: A Guide to Water Quality at Vacation Beaches, 19th ed.; Natural Resources Defense Council: New York, 2009.

POLICY ANALYSIS

(3) Nevers, M. B.; Whitman, R. L. Policies and practices of beach monitoring in the Great Lakes, USA: A critical review. J. Environ. Monit. 2010, 12 (3), 581–590. (4) Dufour, A. P. Health Eﬀects Criteria for Fresh Recreational Waters; EPA-600/1-84-004; U.S. EPA: Cincinnati, OH, 1984. (5) New York State Department of Health. Public Health Law. Subpart 62 Bathing Beaches, Section 225 62.15, 2004. (6) State of Illinois, Swimming Pool and Bathing Beach Act 210 ILCS 125. 77: Public Health Section 820.400 minimum Sanitary Requirements for Bathing Beaches; Illinois General Assembly, 2004. (7) State of Indiana. Water Quality Standards. 327 IAC 21-6, Indiana Administrative Code, 2008. (8) State of Michigan. Michigan Natural Resources and Environmental Protection Act. Part 31, Water Quality Standard, Part 4 Rules, Rule 323.1062 (1), 1997; Vol. PA 451. (9) State of Minnesota. Minnesota Administrative Rule 7050.0222. Minnesota Oﬃce of the Revisor Statutes, 2008. (10) Wisconsin Department of Natural Resources. Water Quality Standards for Wisconsin Surface Waters; Department of Natural Resources: Madison, WI, 2001. (11) U.S. EPA. Ambient Water Quality Criteria for Bacteria; EPA 440/5-84-002; Oﬃce of Water Regulations and Standards: Washington, DC, 1986. (12) Boehm, A. B.; Grant, S. B.; Kim, J. H.; Mowbray, S. L.; McGee, C. D.; Clark, C. D.; Foley, D. M.; Wellman, D. E. Decadal and shorter period variability of surf zone water quality at Huntington Beach, California. Environ. Sci. Technol. 2002, 36 (18), 3885–3892. (13) Boehm, A. B. Enterococci concentrations in diverse coastal environments exhibit extreme variability. Environ. Sci. Technol. 2007, 41 (24), 8227–8232. (14) Whitman, R. L.; Nevers, M. B. Escherichia coli sampling reliability at a frequently closed Chicago beach: Monitoring and management implications. Environ. Sci. Technol. 2004, 38 (16), 4241–4246. (15) Indiana Department of Environmental Management. Accessed Sept. 15, 2011; https://extranet.idem.in.gov/beachguard/. (16) Illinois Department of Public Health. Accessed Sept. 15, 2011. http://app.idph.state.il.us/envhealth/ilbeaches/public/. (17) Eckner, K. F. Comparison of membrane ﬁltration and multipletube fermentation by the Colilert and Enterolert methods for detection of waterborne coliform bacteria, Escherichia coli, and enterococci used in drinking water and bathing water quality monitoring in southern Sweden. Appl. Environ. Microbiol. 1998, 64, 3079–3083. (18) Systat 12.0; SYSTAT Software, Inc.: Chicago, IL, 2007. (19) SPSS version 12; SPSS Inc.: Chicago, IL, 2003. (20) Whitman, R. L.; Nevers, M. B. Summer E. coli patterns and responses along 23 Chicago beaches. Environ. Sci. Technol. 2008, 42 (24), 9217–9224. (21) Nevers, M. B.; Whitman, R. L. Nowcast modeling of Escherichia coli concentrations at multiple urban beaches of southern Lake Michigan. Water Res. 2005, 39 (20), 5250–5260. (22) Chawla, R.; Hunter, P. R. Classiﬁcation of bathing water quality based on the parametric calculation of percentiles is unsound. Water Res. 2005, 39, 4552–4558. (23) EC (Commission of the European Communities). Directive 2006/7/EC of the European Parliament and of the council of 16 February 2006 concerning the management of bathing water quality and repealing Directive 76/160/EEC. Commission of the European Communities, 2006. (24) Nevers, M. B.; Whitman, R. L. Coastal strategies to predict Escherichia coli concentrations for beaches along a 35 km stretch of southern Lake Michigan. Environ. Sci. Technol. 2008, 42 (12), 4454– 4460. (25) Nevers, M. B.; Whitman, R. L. Eﬃcacy of monitoring and empirical predictive modeling at improving public health protection at Chicago beaches. Water Res. 2011, 45 (4), 1659–1668. (26) Schoen, M. E.; Ashbolt, N. J. Assessing pathogen risk to swimmers at non-sewage impacted recreational beaches. Environ. Sci. Technol. 2010, 44 (7), 2286–2291. 10320

dx.doi.org/10.1021/es202568f |Environ. Sci. Technol. 2011, 45, 10315–10321

Environmental Science & Technology

POLICY ANALYSIS

(27) Wymer, L. J.; Brenner, K. P.; Martinson, J. W.; Stutts, W. R.; Schaub, S. A.; Dufour, A. P. The EMPACT Beaches Project: Results from a Study on the Microbiological Monitoring of Recreational Waters; EPA 600/ R-04/023; U.S. EPA, Oﬃce of Research and Development: Cincinnati, OH, 2005. (28) Kim, J. H.; Grant, S. B. Public mis-notiﬁcation of coastal water quality: A probabilistic evaluation of posting errors at Huntington Beach, California. Environ. Sci. Technol. 2004, 38 (9), 2497–2504.

10321

dx.doi.org/10.1021/es202568f |Environ. Sci. Technol. 2011, 45, 10315–10321

POLICY ANALYSIS pubs.acs.org/est

Risk and Markets for Ecosystem Services Todd K. BenDor* Department of City and Regional Planning and UNC Institute for the Environment, University of North Carolina at Chapel Hill, New East Building, Campus Box #3140, Chapel Hill, North Carolina 27599-3140, United States

J. Adam Riggsbee RiverBank Ecosystems, Inc., Austin, Texas 78755, United States

Martin Doyle Nicholas School of the Environment, Duke University, Durham, North Carolina 27708 ABSTRACT: Market-based environmental regulations (e.g., cap and trade, “payments for ecosystem services”) are increasingly common. However, few detailed studies of operating ecosystem markets have lent understanding to how such policies aﬀect incentive structures for improving environmental quality. The largest U.S. market stems from the Clean Water Act provisions requiring ecosystem restoration to oﬀset aquatic ecosystems damaged during development. We describe and test how variations in the rules governing this ecosystem market shift risk between regulators and entrepreneurs to promote ecological restoration. We analyze extensive national scale data to assess how two critical aspects of market structure (a) the geographic scale of markets and (b) policies dictating the release of credits aﬀect the willingness of entrepreneurs to enter speciﬁc markets and produce credits. We ﬁnd no discernible relationship between policies attempting to ease market entry and either the number of individual producers or total credits produced. Rather, market entry is primarily related to regional geography (the prevalence of aquatic ecosystems) and regional economic growth. Any improvements to policies governing ecosystem markets require explicit evaluation of the interplay between policy and risk elements aﬀecting both regulators and entrepreneurial credit providers. Our ﬁndings extend to emerging, regulated ecosystem markets, including proposed carbon oﬀset mechanisms, biodiversity banking, and water quality trading programs.

’ INTRODUCTION Eﬀorts to improve environmental protection policy have sparked widespread interest in market-based environmental policies.1 These market structures take many forms, including publicly funded payments for ecosystem services (PES), voluntary environmental improvement programs (e.g., voluntary carbon markets), cap and trade programs, and regulated ecosystem oﬀset markets. The United States has begun moving toward “regulated oﬀset markets,” which induce demand for ecosystem services (see Chart 1) by requiring environmental conservation, preservation, or restoration (hereafter “conservation”) to oﬀset environmental destruction elsewhere. While many have been proposed, in reality, few ecosystem markets are operational, and most lessons for proposed markets are drawn from the well-established markets for aquatic ecosystems streams and wetlands in the United States.2 Because other ecosystem r 2011 American Chemical Society

markets include few genuine trades,1,3 aquatic ecosystem markets provide some of the best primary empirical data for evaluating ecological eﬀects of markets,4 landscape-scale market trading behavior,5 and regulatory behavior and decision-making capacity for overseeing ecosystem service markets.6 Since 1988, United States water policy has sought to attain “no net-loss” of aquatic ecosystems. Regulations have gradually evolved to require oﬀsets, usually through ecological conservation, for aquatic ecosystems impacted or destroyed during land development (for a review of policies creating this market, see ref 2). For example, if a land developer impacts 10 ha of wetlands Received: September 13, 2011 Accepted: November 1, 2011 Revised: November 1, 2011 Published: November 01, 2011 10322

dx.doi.org/10.1021/es203201n | Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology Chart 1

as part of a project, that developer must provide at least 10 ha of ecological conservation oﬀsets to fulﬁll the no-net-loss requirement. Developers can either provide ecological conservation themselves or purchase oﬀsets credits from a “mitigation bank”. Mitigation banks are private, entrepreneurial ﬁrms (but can be public entities) that speculatively conserve large tracts of aquatic ecosystems (largely through restoration), thus creating a bank of compensation ‘credits’. These credits can then be sold to multiple individuals seeking to impact aquatic ecosystems elsewhere. Compensatory mitigation, as this regulatory process is known, now comprises the largest tradable ecosystem service market in the United States.1 Aquatic ecosystem markets trade nearly $3 billion worth of wetland and stream conservation annually1,7 nearly 10 times that of the Endangered Species Act habitat programs conserving approximately 20,000 wetland ha (1999 2003 average), and over 73 km of streams annually.7 As aquatic ecosystem markets have grown nationwide, their regulation has begun to mirror ﬁnancial markets, as greater regulatory standards and outside investment have increased transparency and standardization of trades.8 It is important to draw a distinction between broader “pollution markets”. Pollution markets trade commodities based on pollution weight, volume, or concentration (e.g., water quality trading and the U.S. SO2 market), while ecosystem service markets trade environmental services measured through ecological assessment criteria (including point to nonpoint water pollution). Ecosystem service markets also tend to trade in commodities of area of entire, bundled ecosystems (e.g., area of wetlands or endangered species habitat, length of stream or riparian buﬀer) rather than particular pollutants (e.g., nitrogen), although this is not a ﬁrm distinction. One major ongoing debate concerns the extent to which traded ecosystems should rigidly mimic each other’s ecological functions. Trading ecosystems in this ‘in-kind’ manner creates trade-oﬀs between preserving speciﬁc functions and characteristics (e.g., replacing a cold-water stream with a cold-water stream) and inadvertently ‘thinning’ markets for certain ecosystems,9 since certain ecosystems (e.g., groundwater fed wetlands) become nonexchangeable due to their inherent uniqueness. We analyze the factors aﬀecting the prevalence of mitigation banking, which now forms the backbone of the compensatory mitigation industry.10 We collected data from regulators, industry associations, and performed the ﬁrst comprehensive survey of the national mitigation banking community. Our goal is to understand the risk considerations in this market and policies that modify risk (whether intentionally or not) and encourage mitigation banking by lowering market entry. Our results have important implications for proposed and emerging analogous ecosystem markets in the U.S. and worldwide.

’ PROGRAM IMPLEMENTATION: ECOLOGICAL RISK VS ENTREPRENEURIAL RISK Risk management is an important framework for understanding the success of environmental policy.11 Risk in aquatic

POLICY ANALYSIS

ecosystem markets is derived from two primary forms: regulatory risk and entrepreneurial (or ‘banker’) risk. Regulatory risk is the likelihood that the goal of no net-loss of ecosystem services will not be met. Given that regulators are enforcing environmental protection regulations, regulatory risk is very much a proxy for ecological risk. The task for regulatory agencies is to minimize ecological risk. Conversely, entrepreneurial risk is the likelihood that conservation activities (production of credits) will not be proﬁtable or worthwhile ﬁnancial investments. The primary regulator of aquatic markets is the Army Corps of Engineers (hereafter Corps), the federal agency administering Section 404 of the Clean Water Act.6 When the Corps permits aquatic ecosystem impacts, they encounter risk that an impact will not be fully oﬀset by the conservation provided by mitigation credits. Net ecosystem loss can result from three types of failure: a) failure to conserve ecosystems (including the same type of ecosystems) suﬃciently or altogether,2,12 b) failure to perform timely conservation,13 or c) failure to maintain long-term viability of a conserved site.14 Addressing these types of failures has been a goal of evolving federal policy, which recently adopted mitigation banking as a technique for reducing some of these ecological risk factors (ref 15 p 19594). Historically, compensation was provided by permittees (i.e., developers) themselves, known as ‘permitteeresponsible mitigation’,2 or governments, who typically run ‘inlieu fee’ programs, which collect and pool fees for aquatic impacts to fund future restoration projects.16 These approaches typically produced oﬀset sites that imposed substantial regulatory burdens and produced little ecological success (i.e., small, fragmented, and widely dispersed oﬀset sites2). Moreover, programs historically did little to ensure that the aquatic ecosystems services being lost were replaced by “equivalent” services (known as ‘in-kind’ mitigation, which is now an important component of mitigation programs nationwide due to substantial criticism6,17,18). Mitigation banks were initially proposed to solve these problems by creating ecosystem credits in advance of impacts,19 as opposed to a contract for future conservation, thus reducing or eliminating the ﬁrst two types of failure and associated ecological risk. However, risk reduction for the regulator shifts risk to the mitigation banker entrepreneurial risk. Mitigation bankers enter markets with heavy up-front capital investments, including substantial legal and planning work, land acquisition, design, and construction. Mitigation banks also rely on economies of scale, necessitating large, contiguous tracts of wetlands or stream reaches (typically measured in gross terms as hectares [wetlands] or linear meters [streams]), established years in advance, in order to produce credits for sale. Uncertainty around these investments and potential payoﬀs represent multiple sources of entrepreneurial risk. Mitigation bankers must weigh these investments against potential future demand for ecosystem credits, which are driven by urban, transportation, and land development, and are reliant on local, regional, and macroeconomic growth, i.e., other sources of uncertainty and entrepreneurial risk. Mitigation bank investments are also weighed against regulator behavior (which institutional economists might consider as “sovereign risk”), which can signiﬁcantly aﬀect credit demand. While federal policy establishes broad rules, a large degree of autonomy in interpreting and implementing the policy is left to local-level (district) staﬀ within the Corps, a source of variability in how mitigation banks are regulated. For example, regulators 10323

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology can slow the bank approval process, which has the eﬀect of driving up the legal and planning costs. They can also alter the ecological standards required of banks, thereby forcing ad hoc adjustments to restoration investments. Among the policies that vary by Corps districts is the credit release schedule in that regulators now typically disregard the original deﬁnition of banking as advance mitigation and, rather, allow scheduled credit releases whereby sale of a percentage of total bank credits is allowed prior to project completion. In fact, credits can be released by regulators prior to any veriﬁcation that a bank has met any ecological standards set by regulators (in order to get initial advance credit releases, however, bankers must obtain conservation easements, produce ﬁnancial assurances, and present detailed project designs and plans15). For example, policy may allow a bank to sell 30% of credits prior to achieving any ecological criteria thresholds and the rest in stages, as other criteria are achieved. This practice reduces entrepreneurial risks by increasing ecological risks. Regulations can also inﬂuence credit demand in several ways. The ﬁrst is through variations in geographic service areas, which are “...the geographic area[s] within which impacts can be mitigated at a speciﬁc mitigation bank...” [refs 10 and 15 Part 332.2]. Large geographic service areas increase potential demand for credits; small service areas reduce demand, and like the release schedule, each district has the ability to set its own service area policy. Large service areas allow conservation to be distant from impacts and thus higher ecological risk than small service areas. An interesting side note here is that recent work in Chicago has demonstrated the regulatory consequences of having few suppliers in a given service area, whereby spatial monopolies tend to form for banks.10 Finally, and perhaps most importantly, although regulators are directed under federal rules to prefer mitigation bank credits (ref 15 Part 332.3(b)(2 6)), they can instruct or allow the use of alternative forms of compensation, which can dramatically reduce the demand for mitigation bank credits. In sum, there is a distinct trade-oﬀ between regulatory risk and entrepreneurial risk. As originally conceived, mitigation banking practice involves substantial entrepreneurial risk. However, regulators’ ability to release credits before project completion and adjust geographic service areas has potentially reduced entrepreneurial risk but at the cost of increasing regulatory (i.e., ecological) risk.

’ METHODS AND DATA We developed comprehensive national statistics on supply and demand for aquatic ecosystem credits and its variation with market regulation. We collected national scale data on credit demand, including data on federal permitting behavior (FY2006 2008 permits/year; the only permit years for which data were available for each district in similar, comparable formats), and urban and transportation construction, which is measured as annual building permit (U.S. Census building permit data from 2005 to 2008) and transportation construction rates (total lane-km constructed 2000 2007). Road construction was ascertained using 2000 2007 data on road lane-km (a lane-kilometer is one lane-width for a linear km) from the Highway Performance Monitoring System,20 which includes 12 diﬀerent road types for each U.S. county, ranging in size from local rural roads to interstate highways.

POLICY ANALYSIS

On the supply side, we collected extensive data on prevalence of banking and wetlands, relative mitigation banking costs, advance credit release policies, and the geographic scale of markets (‘service areas’). Estimates of banking activity levels (number of banks and credits produced by banks) in districts were established using data from the RIBITS federal banking database21 in combination with additional data collected by Madsen et al.,1 who assembled the ﬁrst national census of the type, location, and size of stream and wetland mitigation banks. As of September, 2010, 11 districts (Albuquerque, Baltimore, Charleston, Ft. Worth, Los Angeles, Pittsburgh, San Francisco, St. Paul, Tulsa, Walla Walla, and Wilmington) had not been incorporated into the RIBITS regulatory database and therefore could not augment the Madsen et al.1 database. As of the date of data collection, RIBITS did not reliably account for stream banking at a national level. Our analysis augmented this database with additional, available regulatory data as RIBITS continues to expand into nationwide use. To our knowledge, this is the most comprehensive and representative mitigation banking database available. Wetland data were collected from the National Wetlands Inventory, established by the U.S. Fish and Wildlife Service.22 This database is somewhat incomplete in the western U.S. and does not exist for the State of Wisconsin (resulting in substantial incomplete data for the St. Paul District). Data were collected for all available advance credit release policies (as of mid-2009) in the 36 districts of the contiguous United States (not including districts in Hawaii and Alaska). Districts that contained banks, but had no formal advance release policies, were asked for at least four recent mitigation bank instruments (legal documents formally describing the bank and its operation). Here, individual bank early releases were averaged to approximate a de facto formal release policy. Data on service area size were previously collected for all Corps districts by Womble and Doyle.23 We also sought direct input on how regulations were interpreted by ecosystem market participants. Between April and May 2009, we administered a Web-based survey to mitigation banking professionals (N = 156, 47.7% response rate24) to better understand banker perceptions of recent regulations and the cost framework associated with mitigation banking. Bankers were asked to disaggregate banking costs into nine separate categories, including legal and site approval, land acquisition, baseline ecological monitoring, physical restoration (hydrological/stream channel construction), biological restoration (vegetation establishment), postrestoration ecological monitoring, and site maintenance. The Corps is divided into 36 regulatory districts in the contiguous United States, each of which operates largely autonomously through the direction of regulatory staﬀ. It is this autonomous rule interpretation that creates nationwide variability in mitigation bank regulation and, essentially, an experiment in how diﬀerent regulations aﬀect credit production and the number of mitigation bankers entering the market. As a result, data on permitting, credit release policies, and banking prevalence were collected (and only available) at the district level, while wetland prevalence (percentage of total land area) and building and road construction data were spatially aggregated to districts. The aggregation process for wetlands and building and road construction data involved allocating counties into districts, the analysis unit for this study. Counties divided by districts were placed into the district containing the larger part. This process was completed independently by two coders and compared for 10324

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology

POLICY ANALYSIS

Table 1. Regression Analyses on Bank Credits and Number of Banksa total bank credits (n = 30; R2= 0.80) coef. 1) % wetland area

129791.80

2) rigorous market area size (dummy)

2991.75

3) % advance release

a b

77.10

std. error

number of banks (n = 32; R2= 0.46) t d

coef.

std. error 73.28

4.02d

10.26

0.69

34714.36

3.74

294.43

4870.98

0.61

7.09

192.27

0.40 c

0.92

0.42

t

2.21c

4) road construction

0.89

0.41

2.19

0.00

0.00

5) building construction

0.23

0.11

1.97b

0.00

0.00

0.81

0.01

0.36

6) regulatory permitting

0.14

2.56

0.05

7) 8-digit HUC market area (dummy)

4298.42

5696.71

0.75

8) intercept

8309.08

9616.23

0.86

0.00 0.59 38.12

11.97 20.22

0.96

0.05 1.89b

Case-wise data on total bank credits and early release were only available for 30 districts, while data on bank counts were available for 32 districts. p < 0.1. c p < 0.05. d p < 0.01.

inter-rater reliability (97.13% match rate), whereby all inconsistencies were rectiﬁed. To understand the relationships between the factors discussed above, we implemented two ordinary least-squares (OLS) regression models to control for intervening eﬀects of building permitting rates, total road construction, regulatory permitting rates, wetland area as percentage of total land area, percentage of credits released before meeting ecological performance criteria, whether a district used HUC-8 boundaries (dummy variable), and whether a district had a strict policy deﬁning bank service areas (dummy variable). The ﬁrst model regressed these factors on total bank credits constructed in each district, while the second model regressed these factors on the number of individual banks constructed (see Table 1 for variable lists). No signiﬁcant collinearity was found between variables (VIF < 3.5). Aquatic ecosystem markets, which are an amalgamation of ecological, regulatory, and entrepreneurial interests, are diﬃcult to understand, partly because data are diﬃcult to acquire and unequivocal conclusions can rarely be drawn from the fragmented data sets. The unavailability of longitudinal data on policy, regulatory decisions, and permitting at the district level precludes the use of regression techniques that could causally link policies to bank and credit establishment (inability to establish or measure Granger causality).25 Indeed, data for mitigation are notoriously incomplete, and severe data collection and quality issues have hindered past evaluations.5,26,27 However, exploratory data analysis and simple linear regression were adequate for understanding broad relationships between market geography, phased credit sale policies, and banking prevalence at the district level. Thus, we utilized available information that could be used to indicate supply and demand sites of mitigation and how these responded to regulatory variability. Although our statistical analysis is noncausal, it represents a critical analysis for understanding market dynamics, which would otherwise necessitate years of wide-scale and costly posthoc data collection by the Corps.

’ RESULTS National Development Patterns. Average aquatic impact permitting rates ranged between 376 per year in the New Orleans district and 6,350 per year in the Jacksonville District (Figure 1A), both of which are wetland-dense regions (Figure 1B). New Orleans has historically had some of the highest permit counts in the nation, but development and/or permitting activities were

largely curtailed following Hurricane Katrina in August 2005 (immediately preceding FY 2006). In FY 2008, the New Orleans District granted 578 permits, signaling an upward climb in postKatrina permitting. Between 2000 and 2007, nearly 366,000 km of roads were constructed in the United States (Figure 1C and D), primarily in the Southwest (Los Angeles and Albuquerque Districts), the upper Great Plains (Omaha, St. Paul, and Kansas City Districts), and the Southeast (Jacksonville, Wilmington, and Savannah Districts). High growth regions, including the Jacksonville, Los Angeles, Sacramento, and Ft. Worth districts saw high rates of building construction, particularly single family housing, which accounted for the vast majority of building permitting in all measured years in these regions. Slow-growing areas in central-southern and midwestern districts, including Memphis, Little Rock, St. Louis, and Pittsburgh, had the lowest average building permit rates. Variability in Mitigation Banking. There were approximately 201 new banks since the publication of Madsen et al.,1 yielding a total of 994 banks containing 379,956 wetland credits within the contiguous U.S (Figure 1E; the Madsen et al.1 data set contained information on N = 809 total banks, 793 of which contained useful location information). Banking was particularly prevalent in the Southeast, and while some Southern districts (e.g., Ft. Worth, Galveston) do not contain a large number of banks, they have accrued extensive wetland credit inventories in large banks (i.e., few large banks). All districts with operating banks allowed early credit releases, ranging from 15 to 60% of total bank credits (mode = 30, mean = 36.7%; Figure 1F). As of August 2009 (when credit release data collection was ﬁnalized), the Albuquerque, Tulsa, Pittsburgh, and New England Districts did not contain any federally authorized mitigation banks (they still may contain state or locally authorized banks). All districts without formal policies allowed advance releases to multiple banks, often at very similar rates, thereby creating a de facto advance release policy. Although wording and the speciﬁc thresholds for staged credit release varied among districts, a common series of steps allowed incremental credit releases as banks increasingly achieved ecological standards. Even so, there was signiﬁcant variation between districts: the New Orleans District allowed the highest fraction of credits released prior to satisfying ecological performance criteria (to the left of the vertical dashed line), while other districts allowed only 30% (Figure 2 Left Panel). The Charleston District had a larger number of stages after which credits could be 10325

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology

POLICY ANALYSIS

Figure 1. Panel A: Total impact permits granted, by Regulatory District, Panel B: Relative Wetland Density (% of total district land area), Panel C: Building permits granted (Avg. 2005 2008), Panel D: Total lane km construction (2000 2007), Panel E: Number of mitigation banks and total credits (enumerated on map), Panel F: Advance release rates (% of total credits in a bank), Panel G: Geographic scale and rigor of policies determining bank market size (‘service area’), policy rigor is measured as rigorous, lenient, or a mix (‘divided districts’), Panel H: Costs to gain access to advance release credits (% of total bank construction costs).

released (incrementally over ﬁve years of monitoring and bank closeout), although the New Orleans District required 15 years of monitoring and successful ecological establishment in order to sell the ﬁnal 20% of bank credits. As shown in the Right Panel of Figure 2, there was also signiﬁcant variation between districts as

the required investments of each incremental step can incur very diﬀerent cost structures across districts. In the Forth Worth District, for example, a much higher percentage of the costs (83.8%) are generated prior to meeting ecological performance thresholds than in the Chicago District (35.9%). 10326

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology

POLICY ANALYSIS

Figure 2. Left Panel: Examples of Credit Release Stages for Mitigation Banks (‘Credit Release Schedules’) with pre-ecological threshold marked with a dashed line (percentage of total credits on vertical axis). For example, bankers can sell 30% of the units in a Chicago wetland bank to developers before any ecological threshold has been met. Right Panel: Comparison of Bank Creation Cost Trajectories for Two Example Districts.

The lowest advance credit release rates occurred in the Norfolk, Omaha, and St. Louis Districts, which allowed only 15% of credits to be sold prior to meeting ecological performance standards. One quarter of the districts (n = 9) granted a 30% advance credit release, and nearly one-half (n = 16) allowed 30 35%. The highest advance releases (60%) occurred in the New York and Rock Island Districts. Geographic service area regulation also varied from restricting transactions to a single watershed, basin, eco-region, or other government-deﬁned boundary, or any combination thereof

(Figure 1G). Most districts (∼70%) relied on 8-digit watersheds (HUC28) to deﬁne market sizes (HUC-8 watersheds are ∼1,800 km2). While ‘rigorous’ districts employed strict service area policies, most districts (∼64%) were more ‘lenient,’ allowing case-by-case variations. The Kansas City, Huntington, and Philadelphia Districts are divided regarding their enforcement of stringent service area boundaries. Many districts also use multiple geographic boundaries to determine service areas, including physiographic or EPA deﬁned eco-regions,29 state-deﬁned service areas (and watershed management or resource inventory areas), counties, or cities. 10327

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology Response rates for cost-related survey questions were low (18.7%), as mitigation bankers were reluctant to reveal bank construction cost information, even in conﬁdential, aggregate forms; this reluctance is consistent with other studies of banker investment costs (e.g., refs 30 and 31). Responses yielded information for 11 districts (N = 29; Panel H of Figure 1) in which 75.9% to 93.8% of total costs were expended on activities prior to reaching performance standards, which we deﬁned to include legal and planning costs, land acquisition, restoration design and implementation, and baseline ecological monitoring (see Right Panel of Figure 2 for examples of incremental cost structures in Forth Worth and Chicago). Although these results demonstrate substantial variation in cost structures throughout the country, due to the low response rate received, we must note that we did not use the cost-related survey data in our regression analyses. Integrating Demand and Policy with Bank Prevalence. Regression analyses (Table 1) showed that advance credit release rates had no significant relationship to total bank credits and an inverse relationship with the prevalence of individual banks (a proxy for number of bank firms). Additionally, there was no relationship between bank prevalence or credit production and policies rigorously enforcing a geographic service area size or mandating a common and fairly large service area (the 8-digit hydrological unit28). Again, we note that lack of time series data precludes causally focused regression analysis. However, studying the relationships between policy and outcomes is still meaningful for drawing lessons about landscape-scale market activity and incentives. The Environmental Paradox of Third-Party Offset Production. U.S. aquatic ecosystem markets give us some insight into how emerging markets might balance regulatory risk (a proxy for ecological risk) and entrepreneurial risk. If regulators seek to facilitate markets, they may begin by allowing advance credit sales or larger geographic market areas, thereby absorbing risk from entrepreneurs. The tension currently afflicting these ecosystem market policies lies between the goal of incentivizing credit supplier market entry versus ensuring that high quality offsets occurs well in advance of impacts and where they are needed most. However, our findings suggest that increasing ecological risk by allowing the early sale of credits, within a range of 15 60%, does not increase market activity and therefore cannot be justified for that reason alone. Early releases above or below this range may or may not have an effect. Our analysis was forced to consider the average credit release rates for each district and therefore does not allow us to determine if regulators consistently provide standardized releases that are independent of restoration project effort or quality. Economic theory suggests that if mitigation bankers encounter signiﬁcant market entry barriers (e.g., high investment costs, uncertain proﬁt margins, and credit demand32), and there is no way to overcome these barriers through advance credit sales, bankers will be less likely to locate in a given market area. In the case that credit suppliers fail to enter markets, credit purchasers would be forced to seek alternative techniques for creating conservation; systemic ecological and implementation failure has often plagued these alternative techniques.2 Our data show that regulators, in an attempt to attract more bankers, have typically adopted policies that allow bankers to sell a large fraction of their credits prior to demonstrating establishment of ecological functions and over a wide array of geographic service areas. However, our results do not demonstrate a

POLICY ANALYSIS

signiﬁcant link between these policies that attempt to incentivize market entry and actual rates of market entry, as measured by number of banks and credit production. Rather, market entry is primarily related to regional geography (the prevalence of aquatic ecosystems) and regional economic growth (construction rates), i.e., demand for oﬀsets. It appears that policies intended to increase market entry have not overcome the fundamental constraints created by the regional landscapes and economic dynamics that ultimately drive market demand. This study goes as far as possible to understand ecosystem markets in the U.S. with available national data. In order for further ecosystem market research to be possible, regulatory records must be more complete, understandable (e.g., few districts maintain high quality geo-spatial data), and contain time-indicated information on regulatory decision-making.5 Analyzing additional markets (e.g., carbon, endangered species habitat) in similar detail is not possible given the current scarcity of basic data. The lack of national, time-series market data (e.g., date of policy adoption, date of bank establishment) inhibits direct assertions about absolute causal linkages between individual district policies and market entry patterns. Our ﬁndings pose several questions that need to be addressed by any type of ecosystem service market regulatory structure: What are the trade-oﬀs of diﬀerent forms of risk and failure when using markets for environmental protection? If we discover and quantify these trade-oﬀs, what should regulators be willing to risk in order to enhance market entry? The crux of the matter in regulating ecosystem markets that rely on private investment in ecosystem conservation involves determining whether policies that incentivize market entry are irrelevant in comparison to broader economic and ecological forces determining market behavior. In fact, policies that incentivize market entry may distort market participation, providing divergent incentives to diﬀerent types of credit suppliers. This raises the more problematic issue of whether regulatory policies are actually incentivizing diﬀerent qualities of conservation. The actual functional quality of ecosystem credits produced is an aspect of ecosystem markets that we have not addressed, nor has it been systematically addressed elsewhere (the closest is perhaps the 2005 U.S. Government Accountability Oﬃce33 evaluation of seven districts, where major problems were found with permit evaluations and regulatory processes), but is of critical importance and interest to both regulators and the oﬀset industry.2 Restoration ‘quality’ can be thought of as the functional quality of ecological restoration in terms of gains to physical, chemical, or biological integrity; this is often diﬀerent from the deﬁnition used by regulators which more often measures conservation actions performed (process-driven) rather than functional uplift attained (outcome-driven; see ref 14 for an empirical study of this disconnect). Moreover, the time frame during which sales occur (particularly advance credit sales) and ecological function is fully established is often very diﬀerent.31 By allowing advance release, regulators sacriﬁce some precision in their ability to assess the quality of oﬀset projects in exchange for more bankers that enter markets and (hopefully) produce higher quality credits than would be created using other mitigation methods. However, it is possible that policies incentivizing banker entry could disproportionally beneﬁt mitigation bankers that create low quality credits. This process, known in economics as ‘adverse selection’,34 occurs when buyers and sellers have asymmetric information (i.e., bankers know more about their own abilities to produce credits than regulators). 10328

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology Under adverse selection, low quality credits producers will beneﬁt under an incentive structure that lowers market entry barriers established to limit ecological risk. Assuming that market entry barriers are much higher for the creation of high quality than for low quality credits, then low quality credit producers have the most to gain from policies that lower the cost of market entry (creating high quality oﬀsets involves greater expenditures in ﬁnding and obtaining ecologically valuable areas to conserve and elevated levels of expertise in designing and performing restoration). If regulators and credit purchases are unable to distinguish bankers based on their capability for creating high quality credits, or lack the ability to discriminate between bankers based on past conservation experience, then incentivizing market entry by decreasing entry costs may inadvertently incentivize low quality credit production.24 Example of the consequences of low quality credit production include bank failures, such as the Northlakes Park Bank in Florida, and Virginia’s Fort Lee Mitigation Bank, which sold nearly all total bank credits even though they both failed to establish proper hydrology.35 Given the increased use of market mechanisms for environmental management, scientists and policy makers need to increasingly view environmental conservation as a coupled ecologicaleconomic system. Thus, the future of conservation may be aﬀected less by species interaction and biogeochemical cycles than by local regulatory discretion, distorted incentives, market entry, and asymmetric information.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 919-962-4760. E-mail: [email protected].

’ ACKNOWLEDGMENTS We would like to thank Terry Chapin for his early comments on this article, as well as the National Mitigation Banking Association for their assistance with developing and distributing the national survey of mitigation bankers. ’ REFERENCES (1) Madsen, B.; Carroll, N.; Brands, K. M. State of Biodiversity Markets: Oﬀset and Compensation Programs Worldwide; Ecosystem Marketplace: Washington, DC, 2010. (2) NRC, Compensating for Wetland Losses Under the Clean Water Act; National Academy Press: Washington, DC, 2001. (3) EPA, EPA Water Quality Trading Evaluation Final Report; U.S. Environmental Protection Agency: Washington, DC, 2008. (4) Palmer, M. A.; Filoso, S. Restoration of Ecosystem Services for Environmental Markets. Science 2009, 325 (5940), 575–576. (5) BenDor, T.; Sholtes, J.; Doyle, M. W. Landscape Characteristics of a Stream and Wetland Mitigation Banking Program. Ecol. Appl. 2009, 19 (8), 2078–2092. (6) Hough, P.; Robertson, M. Mitigation under Section 404 of the Clean Water Act: where it comes from, what it means. Wetlands Ecol. Manage. 2009, 17 (1), 15–33. (7) ELI, Mitigation of Impacts to Fish and Wildlife Habitat: Estimating Costs and Identifying Opportunities; Environmental Law Institute: Washington, DC, 2007. (8) Nolles, K. Lessons on the Design and Implementation of Environmental Markets from the Financial markets (SEELab Working Paper: WP20060707); SIRCA Experimental Economics Laboratory: Sydney, Australia, 2006. (9) Salzman, J.; Ruhl, J. B. Currencies and the Commodiﬁcation of Environmental Law. Stanford Law Rev. 2000, 53, 607–694.

POLICY ANALYSIS

(10) Robertson, M. M.; Hayden, N. Evaluation of a Market in Wetland Credits: Entrepreneurial Wetland Banking in Chicago. Conserv. Biol. 2008, 22 (3), 636–646. (11) Yin, H.; Pfaﬀ, A.; Kunreuther, H. Can Environmental Insurance Succeed Where Other Strategies Fail? The Case of Underground Storage Tanks. Risk Anal. 2011, 31 (1), 12–24. (12) Cole, C. A.; Shafer, D. Section 404 Wetland Mitigation and Permit Success Criteria in Pennsylvania, USA, 1986 1999. Environ. Manage. 2002, 30 (4), 508–515. (13) Gutrich, J. J.; Hitzhusen, F. J. Assessing the Substitutability of Mitigation Wetlands for Natural Sites: Estimating Restoration Lag Costs of Wetland Mitigation. Ecol. Econ. 2004, 48, 409–424. (14) Reiss, K. C.; Hernandez, E.; Brown, M. T. Evaluation of Permit Success in Wetland Mitigation Banking: A Florida Case Study. Wetlands 2009, 29 (3), 907–918. (15) Corps, EPA Compensatory Mitigation for Losses of Aquatic Resources; Final Rule. http://www.epa.gov/owow/wetlands/pdf/ wetlands_mitigation_ﬁnal_rule_4_10_08.pdf (accessed 12/28/08). (16) Wilkinson, J. In-lieu fee mitigation: coming into compliance with the new Compensatory Mitigation Rule. Wetlands Ecol. Manage. 2008, 17 (1), 53–70. (17) Bedford, B. L. The Need to Deﬁne Hydrolic Equivalence at the Landscape Scale for Freshwater Wetland Mitigation. Ecol. Appl. 1996, 6 (1), 57–68. (18) Race, M. S.; Fonseca, M. S. Fixing Compensatory Mitigation: What Will it Take? Ecol. Appl. 1996, 6 (1), 94–101. (19) Corps, EPA Federal Guidance for the Establishment, Use and Operation of Mitigation Banks. http://www.epa.gov/owow/wetlands/ guidance/mitbankn.html (accessed 6/13/2005). (20) HPMS, Highway Performance Monitoring System (HPMS); U.S. Federal Highway Administration: Washington, DC, 2009. (21) Corps, Regional Internet Bank Information Tracking System (RIBITS); U.S. Army Corps of Engineers, Engineering Research and Development Center (Environmental Laboratory): Vicksburg, MS, 2008. (22) Tiner, R. NWI Maps: Basic Information on the Nation’s Wetlands. BioScience 1997, 47 (5), 269. (23) Womble, P.; Doyle, M. W. Setting Geographic Service Areas for Compensatory Mitigation Banking. Natl. Wetlands Newsl. In press. (24) BenDor, T.; Riggsbee, A. A survey of entrepreneurial risk in U.S. wetland and stream compensatory mitigation markets. Environ. Sci. Policy 2011, 14, 301–314. (25) Granger, C. W. J. Investigating causal relations by econometric models and cross-spectral methods. Econometrica 1969, 37 (3), 424–438. (26) BenDor, T.; Brozovic, N.; Pallathucheril, V. G. Assessing the Socioeconomic Impacts of Wetland Mitigation in the Chicago Region. J. Am. Plann. Assoc. 2007, 73 (3), 263–282. (27) Strand, M. Law and Policy: “Information, Please”. Natl. Wetlands Newsl. 2010, 32 (3), 24. (28) Seaber, P. R.; Kapinos, F. P.; Knapp, G. L. Hydrologic Unit Maps: U.S. Geological Survey; U.S. Geological Survey: Washington, DC, 1987. (29) EPA, Ecoregions of North America; U.S. Environmental Protection Agency: Washington, DC, 2010. (30) EEP, Ecosystem Enhancement Program 2004 2005 Annual Report; N.C. Ecosystem Enhancement Program: Raleigh, NC, 2005. (31) Robertson, M. M. Emerging Ecosystem Service Markets: Trends in a Decade of Entrepreneurial Wetland Banking. Frontiers Ecol. Environ. 2006, 4 (6), 297–302. (32) Hallwood, P. Contractual diﬃculties in environmental management: The case of wetland mitigation banking. Ecol. Econ. 2007, 63, 446–451. (33) GAO, Corps of Engineers Does Not Have an Eﬀective Oversight Approach to Ensure That Compensatory Mitigation Is Occurring: Report to the Ranking Democratic Member, Committee on Transportation and Infrastructure, House of Representatives; U.S. Government Accountability Oﬃce: Washington, DC, 2005. 10329

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

Environmental Science & Technology

POLICY ANALYSIS

(34) Akerlof, G. A. The Market for “Lemons”: Quality Uncertainty and the Market Mechanism. Q. J. Econ. 1970, 84 (3), 488–500. (35) McElﬁsh, J. M.; Nicholas, S. Structure and Experience of Wetland Mitigation Banks. In Mitigation Banking: Theory and Practice; Marsh, L. L., Porter, D. R., Salvesen, D. A., Eds.; Island Press: Washington, DC, 1996.

10330

dx.doi.org/10.1021/es203201n |Environ. Sci. Technol. 2011, 45, 10322–10330

ARTICLE pubs.acs.org/est

Human-Specific E.coli Single Nucleotide Polymorphism (SNP) Genotypes Detected in a South East Queensland Waterway, Australia Maxim S. Sheludchenko, Flavia Huygens,* and Megan H. Hargreaves Cell and Molecular Biosciences, Faculty of Science and Technology, Queensland University of Technology, Brisbane, Queensland, Australia

bS Supporting Information ABSTRACT: The World Health Organization recommends that the majority of water monitoring laboratories in the world test for E. coli daily since thermotolerant coliforms and E. coli are key indicators for risk assessment of recreational waters. Recently, we developed a new SNP method for typing E. coli strains, by which human-speciﬁc genotypes were identiﬁed. Here, we report the presence of these previously described speciﬁc SNP proﬁles in environmental water, sourced from the Coomera River, located in South East Queensland, Australia, over a period of two years. This study tested for the presence of human-speciﬁc E. coli to ascertain whether hydrologic and anthropogenic activity plays a key role in the pollution of the investigated watershed or whether the pollution is from other sources. We found six human-speciﬁc SNP proﬁles and one animalspeciﬁc SNP proﬁle consistently across sampling sites and times. We have demonstrated that our SNP genotyping method is able to rapidly identify and characterize human- and animal-speciﬁc E. coli isolates in water sources.

’ INTRODUCTION Water is a precious resource in Australia, which is the driest inhabited continent. Approximately every ten years in Australia there are three years with suﬃcient water supply and three years of drought, then water restrictions and other measures are put in place to ensure the supply.1 Environmentally conscientious residents are interested in eﬀective water management strategies and therefore their water authorities are as well. Microbial source tracking (MST) is one of the tools for the identiﬁcation of multiple sources of fecal pollution in water sources. Various microorganisms are known to be targets for MST.2 Some pathogenic E. coli have been transmitted via water.3 To our knowledge, there are no host-speciﬁc markers for E. coli, even among virulence genes and the recently discovered humanspeciﬁc serotype O81,4 that have been consistently detected in environmental water.5 Recently, we developed a new SNP method for typing E. coli strains, by which human-speciﬁc genotypes were identiﬁed.6 Typing methods used for MST rely on building a reference library from known host groups to identify sources in unknown water samples. The most common examples of these methods are repetitive extragenic palindromic (rep) PCR,7 ribotyping,8 and pulsed-ﬁeld gel electrophoresis (PFGE).9 They are diﬃcult to develop, interpret, and often are applied only in local geographical areas where the reference library was developed initially.10 Therefore, library-dependent methods are rarely used. Alternatively, there are library-independent methods which are usually based on polymerase chain reaction (PCR) only and do not require large reference libraries. These methods have targeted speciﬁc genes of microorganisms which are diﬃcult to r 2011 American Chemical Society

cultivate in normal conditions such as Bacteroides spp. 16S rRNA clone groups,11 F+ RNA coliphage diﬀerentiation,2 and enteric viruses including polyomaviruses and adenoviruses.12 The advantage of these molecular markers is that they appear to be host-speciﬁc,13 although, some studies have reported humanspeciﬁc markers from nonhuman sources. For instance, humanspeciﬁc Bacteroides marker was detected in other animals such as dogs14 and ﬁsh.15 In this study, using E. coli as the target organism, we chose a molecular typing approach that is a comparative method. This method simply determines whether isolates are the same or diﬀerent based on their SNP genotype proﬁle. The SNP-typing method developed recently6 and applied here is considered by the authors to be neither library-dependent nor -independent. In fact, the authors would describe this novel genotyping approach as a culture-dependent library-based method and once developed, can be done without culture. Library-based due to the origin of the SNPs identiﬁed from housekeeping genes described in the MLST database. We report the presence of these previously described speciﬁc SNP proﬁles in environmental water, sourced from the Coomera River, located in South East Queensland, Australia, over a period of two years. The sampling points were suggested by the Gold Coast City Council as being problematic sites with a history of large numbers of fecal coliforms. One potential source of fecal Received: May 11, 2011 Accepted: October 25, 2011 Revised: October 24, 2011 Published: October 25, 2011 10331

dx.doi.org/10.1021/es201599u | Environ. Sci. Technol. 2011, 45, 10331–10336

Environmental Science & Technology

ARTICLE

Table 1. Locations and Characteristics of Sampling Sites site name (GISa map reference)

code

site characteristics

site classiﬁcation

1

Coomera Marina ( 27.861672, 153.339089)

cattle/kangaroo grazing, house-boat mooring site

rural

2

Santa Barbara ( 27.855165, 153.350612)

park, BBQ, toilets and ﬁshing, private houses about 100 m away

recreational

3

Sanctuary Cove ( 27.851617, 153.362140)

canal estate, modern houses and apartments, modern infrastructure,

urban/suburban

commercial/light industrial area

a

4

Jabiru Island ( 27.879057, 153.380685)

disused sand mine, no houses, small park with sewered toilets

rural

5

Paradise Point ( 27.886359, 153.396596)

public swimming beach, mouth of river, much water traﬃc

recreational

6

Coombabah, the Estuary ( 27.896607, 153.366845)

established suburban area, uninhabited island opposite

suburban

Global information system.

Table 2. Weather Patterns on Days of Sampling variable

a

Autumn 2008

Winter 2008

Autumn 2009

Winter 2009

rainfall in mm, 24 h prior sampling rainfall in mm, 72 h prior sampling

0 0.4

1.3 33.9

3.2 15.2

1.4 1.4

tides (m)a

0.183

0.425

0.931

1.368

Personal communication (Daryl Metters, Maritime Safety Queensland, Department of Transport and Main Roads).

pollution may be the accidental sewage discharge from a large number of yachts and houseboats owned by residents who have boat moorings in the many canal estates. According to the Transport Operations (Marine Pollution) Act16 and Regulation,17 sewage discharge in canals and marinas is prohibited. Boat owners, however, may be unaware of the regulations, or noncompliant. As a result, sewage discharge into the Coomera estuary may be a continuing risk. The method described in our previous work6 allowed us to test for the presence of humanspeciﬁc E. coli in the current study.

’ EXPERIMENTAL SECTION Study Site. The Pimpana-Coomera watershed is located in South East Queensland, Australia. It is used intensively for agriculture and recreation and an anthropogenic effect can be observed in the watershed. The main water source is the Coomera River, which flows for 90 km from its headwaters in the Lamington National Park to the river mouth in the Pacific Ocean. The upper reaches of the river pass through mainly rural areas, where crop and cattle grazing are the common economic/land use activities. In the 1970s and 1980s, the river was widened 20 km upstream from the mouth as a consequence of sand and gravel extraction operations. The lower reaches of the Coomera River pass through highly developed areas including canal estates such as Santa Barbara, Hope Island, Sanctuary Cove, and the Coomera Mooring Marina. Most of the sewage collection system is gravity fed and follows natural catchment drainage lines until it reaches a centrally located wastewater treatment plant. After treatment, the water is released into the Gold Coast Seaway located south of the Coomera River estuary. Despite the existence of such an effective treatment system, large numbers of coliforms have been observed over a long period of time in the estuary by the Gold Coast City council. Environmental Water Sampling. Four seasonal trips to the Coomera catchment on the Gold Coast were undertaken to collect 24 river water samples (taken in duplicate) from May 2008 to July 2009. Sites selected for sampling included the following: Coomera Marina (1), Santa Barbara (2), Sanctuary Cove (3), Jabiru Island (4), Paradise Point (5), and Coombabah (6) (TOC Art Figure and Table 1). These were suggested by the

Gold Coast City Council as being problematic sites with a history of high concentrations of fecal coliforms. Two water samples of 600 mL each were collected in sterile bottles containing sodium thiosulphate (in case of chlorine residuals in the sample) and transported on ice to the laboratory. Samples were collected using a 1.8 m long dipper, to ensure that the sample was taken from the river rather than the edge where eddies and ephemeral contamination may have interfered with the results. Water samples were prepared for analyses immediately upon arrival at the laboratory, which was always within 6 h. Rainfall and tidal information for sampling days/times were retrieved from the Australian Bureau of Meteorology and are listed in Table 2 (http://reg.bom.gov.au/climate/data/index.shtmLand). Isolation of E. coli. The environmental water samples collected in duplicate were each mixed thoroughly, and 100 mL was filtered as described previously.18 As a high number of E. coli was expected based on previous studies of these sites (pilot study and Gold Coast Council data), samples were filtered both undiluted and diluted 1:10 to provide a countable number of colonies per filter pad. Testing according to the U.S. EPA standard19 was done on all water samples and dilutions, using 0.45-μm sterile gridded filter membranes (Millipore Corporation, Bedford, MA). The membranes were aseptically transferred to modified mTEC agar plates (BD, Sparks, MD) and were incubated aerobically at 35 °C for 2 h followed by an additional incubation at 44.5 °C for 24 h. Single magenta colonies demonstrating ß-D-galactosidase activity were selected and subcultured onto MacConkey agar #2 (Oxoid, UK) for DNA extraction as described previously.20 DNA Extraction. To prepare genomic DNA, single lactose fermenting pink colonies from MacConkey agar #2 were isolated and incubated overnight in 5 mL of nutrient broth (Oxoid, UK). A 500 μL portion of overnight culture was centrifuged at 10 000g for 1 min. Cell pellets were resuspended in 180 μL of DNAase/ RNAase-free water and used for DNA extraction on the Corbett X-tractorGene automated DNA extraction system (Corbett Robotics, Brisbane, Australia, Protocol no.141404 version 02). Quantity and purity of DNA extracts were tested on the DU 730 spectrophotometer (Beckman Coulter, USA). Identification of E. coli SNP Genotypes by Allele-Specific Real-Time PCR. The E. coli MLST database available at NIH 10332

dx.doi.org/10.1021/es201599u |Environ. Sci. Technol. 2011, 45, 10331–10336

Environmental Science & Technology

ARTICLE

Figure 1. Shades of red and black indicate human-speciﬁc SNP proﬁles. Shades of green indicate animal-speciﬁc SNP proﬁles. Blue indicates unique SNP proﬁles isolated from the Coomera River. The remaining colors are indicative of “mixed source” SNP proﬁles. Uncolored spaces indicate SNP proﬁles that have only been detected once or twice. Circles show values of E. coli in Colony Forming Units (CFU)/mL.

(http://www.shigatox.net) currently contains 668 E. coli strains that are grouped into 231 Sequence Types (STs). Informative SNP sets were identified by using the software program called “Minimum SNPs”.21 Eight SNPs, with a Simpson’s index of Diversity (D value) D of 0.96, were determined by the program for the differentiation of E. coli isolates.6 A method for highly discriminatory SNP interrogation of E. coli, by using allelespecific real-time PCR, was developed previously by our group6

and applied to all the E. coli isolates in this study. A number of SNP profiles resolved the E. coli population on the basis of unique eight character “barcodes” within a day. From each subculture/ McConkey plate at least 10 colonies, when possible, were selected for DNA extraction followed by SNP profile identification. Statistical Analysis. Binary logistic regression analysis was used to determine the relationship between human/nonhuman SNP profiles and seasons, rainfalls (24 h and 72 h prior sampling), tide 10333

dx.doi.org/10.1021/es201599u |Environ. Sci. Technol. 2011, 45, 10331–10336

Environmental Science & Technology

ARTICLE

Table 3. Summary of the Twenty Most Prevalent SNP Proﬁles Distributed Across Sampling Sites and Time-Pointsa

a

* indicates SNP proﬁle numbers as previously published in Sheludchenko et al.6

levels for estimating salinity at sampling times, land use categories (suburban, urban, rural, and recreational) and distance from the river mouth to the sampling site. All calculations were done using MiniTab 16.0 (Minitab Inc.) and p-values were calculated. To calculate the SNP diversity per sampling site, any SNP profile that was only found once or twice at a specific site (clear bars in Figure 1), was excluded from this analysis; instead, SNP profiles that were frequently found at each site were included (colored bars in Figure 1). Supporting Information Table S1 provides a summary of the numbers of SNP profiles found per season per sampling site. The total number of SNP profiles found at each sampling point was used in the binary logistic analysis.

’ RESULTS AND DISCUSSION Although it is well recognized that E. coli numbers in natural waterways ﬂuctuate in response to environmental factors, particularly rainfall events,22 there is limited research reporting the diversity of E. coli genotypes in environmental water in relation to similar hydrological conditions and land use.5,23 In the current study, a new highly discriminatory genotyping method based on SNP interrogation of E. coli6 was applied to detect host-speciﬁc E. coli genotypes in the Coomera watershed over a two-year period. In total, 165 isolates were grouped into 67 SNP proﬁles found at six sites collected between May 2008 and July 2009. A summary of all the SNP proﬁles observed in this study can be found in Table S2. A very low number of E. coli were detected in samples from site 3, which is therefore not discussed further in this paper.

Despite the fact that there was variation in E. coli numbers isolated from the various sampling sites over the period of two years (Figure 1), the SNP type diversity did not vary signiﬁcantly between sites, or times of sampling. Nor was a signiﬁcant relationship detected between SNP proﬁle diversity, and rainfall, seasons, tides, rural/suburban land use, or distance from the river mouth. From a SNP proﬁle perspective, we found the most diverse E. coli population at site 1, which also had the lowest number of E. coli. In addition, host-speciﬁc proﬁles were rarely detected at sites 4, 5, or 6, which had the largest number of E. coli. The 20 most prevalent SNP proﬁles were detected in a number of samples over the whole period (Figure 1 and Table S3). The remaining 47 SNP proﬁles were detected only once. Of the 20 most prevalent proﬁles, three human-speciﬁc proﬁles (29, 11, and 32) and one animal-speciﬁc proﬁle (7) corresponded to human- and animal-speciﬁc proﬁles published previously.6 Table 3 is a summary of the 20 most prevalent SNP proﬁles found at each sampling site and at each sampling time-point. Human- and animal-speciﬁc SNP proﬁles were distributed across all the sampling sites, irrespective of the seasons. Five of the top 20 SNP proﬁles (35, 70, 37, 80 and 40) could also be considered to be representative of human fecal contamination, since these proﬁles were previously characterized as “mainly human-speciﬁc”.6 Also included in the most prevalent cohort were SNP proﬁles previously identiﬁed as “nonspeciﬁc” (9, 21, 22, 23, 26, and 34) in terms of host assignation.6 SNP proﬁle 80, which in our previous study contained only two human isolates (and no animal), was identiﬁed in four of the six sampling sites.6 SNP proﬁles 14 and 82 were also detected, 10334

dx.doi.org/10.1021/es201599u |Environ. Sci. Technol. 2011, 45, 10331–10336

Environmental Science & Technology which have previously been characterized as cattle- and horsesourced, respectively.6 Interestingly, SNP proﬁle 45 was not found at any of the sites we tested, even though this proﬁle was shown by our previous study to be prevalent in local and international collections.6 In addition, we have not found SNP proﬁle 76 in this or our previous study,6 despite the fact that it is a proﬁle found in other countries. In conclusion, we found six human-speciﬁc (29, 11, 32, 70, 37, 80) SNP proﬁles and one animal-speciﬁc (7) SNP proﬁle consistently across sampling sites and times. SNP proﬁle 29 was found in the majority (44%) of samples tested in this study. SNP proﬁle 11 was the second-most commonly encountered proﬁle, being present in 28% of samples (Table S2). This study investigated SNP proﬁles, previously aligned with human or nonhuman sources,6 found in an E. coli population isolated from a natural waterway. The eﬀect of variables such as rainfall (24 and 72 h), tide height and time, general land use (rural, suburban), seasons, and distance from the river mouth as an estimate of salinity was investigated and it was found that none of the variables signiﬁcantly inﬂuenced the diversity of E. coli SNP proﬁles present in the water (p values >0.35). In addition, by applying our previously developed SNP genotyping method6 to genotype water-sourced E. coli, we were able to identify six human-speciﬁc E. coli SNP proﬁles, and four animal-speciﬁc E. coli SNP proﬁles in the Coomera River of South East Queensland, Australia.

’ ASSOCIATED CONTENT

bS

Supporting Information. Table S1 lists the number of SNP proﬁles found more frequently at each sampling point per season; Table S2 lists all the SNP proﬁles found in this study; Table S3 lists the abundance of SNP genotypes in the Coomera water catchment over a period of two years and four sampling events. This information is available free of charge via the Internet at http://pubs.acs.org/

’ AUTHOR INFORMATION Corresponding Author

*Phone: +61 7 3138 0453; fax: +61 7 3138 1534; e-mail: [email protected].

’ ACKNOWLEDGMENT We thank Melanie Robertson-Dean for providing assistance with statistical analyses. M.S.S. is in receipt of a postgraduate studentship from the Institute of Sustainable Resources, Queensland University of Technology. ’ REFERENCES (1) Department of the Environment, W., Heritage and the Arts. Australian weather and the seasons. http://www.culture.gov.au/articles/weather/. Visited April 2011. (2) Bernhard, A.; Field, K. Identiﬁcation of nonpoint sources of fecal pollution in coastal waters by using host-speciﬁc 16S ribosomal DNA genetic markers from fecal anaerobes. Appl. Environ. Microbiol. 2000, 66 (4), 1587–1594. (3) King, E. L.; Bachoon, D. S.; Gates, K. W. Rapid detection of human fecal contamination in estuarine environments by PCR targeting of Biﬁdobacterium adolescentis. J. Microbiol. Methods 2007, 68 (1), 76–8.

ARTICLE

(4) Scott, T. M.; Jenkins, T. M.; Lukasik, J.; Rose, J. B. Potential use of a host associated molecular marker in Enterococcus faecium as an index of human fecal pollution. Environ. Sci. Technol. 2005, 39 (1), 283–287. (5) Havelaar, A. H.; Furuse, K.; Hogeboom, W. M. Bacteriophages and indicator bacteria in human and animal faeces. J. Appl. Bacteriol. 1986, 60 (3), 255–262. (6) Sheludchenko, M. S.; Huygens, F.; Hargreaves, M. H. Highlydiscriminatory single nucleotide polymorphism interrogation of Escherichia coli using allele-speciﬁc Real-Time-PCR and eBURST analysis. Appl. Environ. Microbiol. 2010, 76 (13), 4337–4345. (7) Walk, S. T.; Alm, E. W.; Gordon, D. M.; Ram, J. L.; Toranzos, G. A.; Tiedje, J. M.; Whittam, T. S. Cryptic lineages of the genus Escherichia. Appl. Environ. Microbiol. 2009, 75 (20), 6534–6544. (8) Clermont, O.; Lescat, M.; O’Brien, C. L.; Gordon, D. M.; Tenaillon, O.; Denamur, E. Evidence for a human-speciﬁc Escherichia coli clone. Environ. Microbiol. 2008, 10 (4), 1000–1006. (9) Ahmed, W.; Tucker, J.; Bettelheim, K. A.; Neller, R.; Katouli, M. Detection of virulence genes in Escherichia coli of an existing metabolic ﬁngerprint database to predict the sources of pathogenic E. coli in surface waters. Water Res. 2007, 41, 3785–3791. (10) Ratajczak, M.; Laroche, E.; Berthe, T.; Clermont, O.; Pawlak, B.; Denamur, E.; Petit, F. Inﬂuence of hydrological conditions on the Escherichia coli population structure in the water of a creek on a rural watershed. BMC Microbiol. 2010, 10 (1), 222. (11) Smith, C. J.; Olszewski, A. M.; Mauro, S. A. Correlation of shiga toxin gene frequency with commonly used microbial indicators of recreational water quality. Appl. Environ. Microbiol. 2009, 75 (2), 316–321. (12) Kon, T.; Weir, S. C.; Howell, E. T.; Lee, H.; Trevors, J. T. Repetitive element (REP)-polymerase chain reaction (PCR) analysis of Escherichia coli isolates from recreational waters of southeastern Lake Huron. Can. J. Microbiol. 2009, 55 (1), 269–276. (13) D’Elia, T. V.; Cooper, C. R.; Johnston, C. G. Source tracking of Escherichia coli by 16S 23S intergenic spacer region denaturing gradient gel electrophoresis (DGGE) of the rrnB ribosomal operon. Can. J. Microbiol. 2007, 53, 1174–1184. (14) McLellan, S. L.; Daniels, A. D.; Salmore, A. K. Genetic characterization of Escherichia coli populations from host sources of fecal pollution by using DNA ﬁngerprinting. Appl. Environ. Microbiol. 2003, 69 (5), 2587–2594. (15) Seurinck, S.; Verstraete, W.; Siciliano, S. Microbial source tracking for identiﬁcation of fecal pollution. Rev. Environ. Sci. Biotechnol. 2005, 4 (1), 19–37. (16) Yan, T.; Sadowsky, M. J. Determining sources of fecal bacteria in waterways. Environ. Monit. Assess. 2007, 129 (1 3), 97–106. (17) Bernhard, A. E.; Field, K. G. A PCR assay to discriminate human and ruminant feces on the basis of host diﬀerences in BacteroidesPrevotella genes encoding 16S rRNA. Appl. Environ. Microbiol. 2000, 66 (10), 4571–4574. (18) Fong, T. T.; Lipp, E. K. Enteric viruses of humans and animals in aquatic environments: Health risks, detection, and potential water quality assessment tools. Microbiol. Mol. Biol. Rev. 2005, 69 (2), 357–361. (19) Ogorzaly, L.; Tissier, A.; Bertrand, I.; Maul, A.; Gantzer, C. Relationship between F-speciﬁc RNA phage genogroups, faecal pollution indicators and human adenoviruses in river water. Water Res. 2009, 43 (5), 1257–1264. (20) McQuaig, S. M.; Scott, T. M.; Harwood, V. J.; Farrah, S. R.; Lukasik, J. O. Detection of human-derived fecal pollution in environmental waters by use of a PCR-based human polyomavirus assay. Appl. Environ. Microbiol. 2006, 72 (12), 7567–7574. (21) Ahmed, W.; Goonetilleke, A.; Powell, D.; Gardner, T. Evaluation of multiple sewage-assoclated Bacteroides PCR markers for sewage pollution tracking. Water Res. 2009, 43 (19), 4872–4877. (22) Kildare, B. J.; Leutenegger, C. M.; McSwain, B. S.; Bambic, D. G.; Rajal, V. B.; Wuertz, S. 16S rRNA-based assays for quantitative detection of universal, human-, cow-, and dog-speciﬁc fecal Bacteroidales: A Bayesian approach. Water Res. 2007, 41 (16), 3701–3715. 10335

dx.doi.org/10.1021/es201599u |Environ. Sci. Technol. 2011, 45, 10331–10336

Environmental Science & Technology

ARTICLE

(23) McLain, J. E. T.; Ryu, H.; Kabiri-Badr, L.; Rock, C. M.; Abbaszadegan, M. Lack of speciﬁcity for PCR assays targeting human Bacteroides 16S rRNA gene: Cross-ampliﬁcation with ﬁsh feces. FEMS Microbiol. Lett. 2009, 299 (1), 38–43. (24) MSQ, Transport Operations (Marine Pollution) Act. Queensland, M. S., Ed. 1995. (25) MSQ, Transport Operations (Marine Pollution) Regulation. Queensland, M. S., Ed. 2008. (26) Eaton, A.; Clesceri, L. S.; Rice, E. W.; Greenberg, A. E. Standard Methods for the Examination of Water and Wastewater, 21st ed.; Washington, DC, 2005. (27) Oshiro, R. Method 1603: Escherichia coli (E.coli) in water by membrane ﬁltration using modiﬁed membrane-thermotolerant Escherichia coli agar (Modiﬁed mTEC); EPA 600-4-85-076; Oﬃce of Water Regulations and Standards, U.S. Environmental Protection Agency: Washington, DC, 2002. (28) Vogel, J. R.; Stoeckel, D. M.; Lamendella, R.; Zelt, R. B.; Domingo, J. W. S.; Walker, S. R.; Oerther, D. B. Identifying fecal sources in a selected catchment reach using multiple source-tracking tools. J. Environ. Qual. 2007, 36 (3), 718–729. (29) Robertson, G.; Huygens, F.; Giﬀard, G. Identiﬁcation and interrogation of highly informative single nucleotide polymorphism sets deﬁned by bacterial multilocus sequence typing databases. J. Med. Microbiol. 2004, 53, 35–45. (30) Kelsey, H.; Porter, D., E.; Scott, G.; Neet, M.; White, D. Using Geographic Information Systems and Regression Analysis to Evaluate Relationships between Land Use and Fecal Coliform Bacterial Pollution; Elsevier: Kidlington, Royaume-Uni, 2004; p 13. (31) Kleinheinz, G. T.; McDermott, C. M.; Hughes, S.; Brown, A. Eﬀects of rainfall on E. coli concentrations at Door County, Wisconsin beaches. Int. J. Microbiol. 2009, 2009, 9. (32) Ishii, S.; Hansen, D. L.; Hicks, R. E.; Sadowsky, M. J. Beach sand and sediments are temporal sinks and sources of Escherichia coli in Lake Superior. Environ. Sci. Technol. 2007, 41 (7), 2203–2209. (33) Lyautey, E.; Lu, Z.; Lapen, D. R.; Wilkes, G.; Scott, A.; Berkers, T.; Edge, T. A.; Topp, E. Distribution and diversity of Escherichia coli populations in the South Nation river drainage basin, Eastern Ontario, Canada. Appl. Environ. Microbiol. 2010, 76 (5), 1486–1496. (34) Wijesinghe, R. U.; Feng, Y.; Wood, C. W.; Stoeckel, D. M.; Shaw, J. N. Population dynamics and genetic variability of Escherichia coli in a mixed land-use watershed. J. Water Health 2009, 07 (3), 484–496.

10336

dx.doi.org/10.1021/es201599u |Environ. Sci. Technol. 2011, 45, 10331–10336

ARTICLE pubs.acs.org/est

Physicochemical Characterization of Particulate Emissions from a Compression Ignition Engine: The Influence of Biodiesel Feedstock N. C. Surawski,†,‡ B. Miljevic,†,|| G. A. Ayoko,†,§ S. Elbagir,§ S. Stevanovic,†,|| K. E. Fairfull-Smith,|| S. E. Bottle,|| and Z. D. Ristovski*,† †

)

ILAQH, Institute of Health and Biomedical Innovation, Queensland University of Technology, 2 George Street, Brisbane QLD 4001, Australia ‡ School of Engineering Systems, Queensland University of Technology, 2 George Street, Brisbane QLD 4001, Australia § Discipline of Chemistry, Queensland University of Technology, 2 George Street, Brisbane QLD 4001, Australia ARC Centre of Excellence for Free Radical Chemistry and Biotechnology, Queensland University of Technology, 2 George Street, 4001 Brisbane, Australia

bS Supporting Information ABSTRACT: This study undertook a physicochemical characterization of particle emissions from a single compression ignition engine operated at one test mode with 3 biodiesel fuels made from 3 diﬀerent feedstocks (i.e., soy, tallow, and canola) at 4 diﬀerent blend percentages (20%, 40%, 60%, and 80%) to gain insights into their particle-related health eﬀects. Particle physical properties were inferred by measuring particle number size distributions both with and without heating within a thermodenuder (TD) and also by measuring particulate matter (PM) emission factors with an aerodynamic diameter less than 10 μm (PM10). The chemical properties of particulates were investigated by measuring particle and vapor phase Polycyclic Aromatic Hydrocarbons (PAHs) and also Reactive Oxygen Species (ROS) concentrations. The particle number size distributions showed strong dependency on feedstock and blend percentage with some fuel types showing increased particle number emissions, while others showed particle number reductions. In addition, the median particle diameter decreased as the blend percentage was increased. Particle and vapor phase PAHs were generally reduced with biodiesel, with the results being relatively independent of the blend percentage. The ROS concentrations increased monotonically with biodiesel blend percentage but did not exhibit strong feedstock variability. Furthermore, the ROS concentrations correlated quite well with the organic volume percentage of particles a quantity which increased with increasing blend percentage. At higher blend percentages, the particle surface area was signiﬁcantly reduced, but the particles were internally mixed with a greater organic volume percentage (containing ROS) which has implications for using surface area as a regulatory metric for diesel particulate matter (DPM) emissions.

1. INTRODUCTION Alternative fuels, such as biodiesel, are currently being investigated not only to address global warming1 but also to reduce DPM emissions.2 While a considerable database exists describing the impact of diﬀerent transesteriﬁed biodiesel fuel types on regulated emissions (i.e., PM, NOx, CO, and HCs),3,4 limited information is available addressing the impact of diﬀerent biodiesel fuel types on other particle emission properties, such as particle number and size. Regulated emissions from compression ignition engines typically exhibit strong dependencies on both feedstock and blend percentage. With PM emissions (for example), animal fat based biodiesel gives greater PM reductions than soy based biodiesel, and the PM reductions exhibit a nonlinear reduction with respect to blend percentage.4 Given these results, it is quite likely that particle emissions will display similar dependencies. At present, a detailed database is not in r 2011 American Chemical Society

existence characterizing the unregulated physicochemical characteristics of DPM such as the following: particle number emission factors, particle size distributions, surface area as well as PAHs and ROS with diﬀerent biodiesel feedstocks and blend percentages. Consequently, a primary objective of this study was to explore the physicochemical properties of particle emissions from 3 biodiesel feedstocks tested at 4 diﬀerent blend percentages to shed light on their potential health impacts. A combination of physical and chemical factors inﬂuences the health eﬀects of DPM,5 where it is noted with biodiesel combustion that the particles have a much higher organic fraction.6 The organic Received: June 1, 2011 Accepted: October 31, 2011 Revised: September 2, 2011 Published: October 31, 2011 10337

dx.doi.org/10.1021/es2018797 | Environ. Sci. Technol. 2011, 45, 10337–10343

Environmental Science & Technology fraction of DPM includes many compounds that are deleterious to human health such as PAHs and ROS.7 Previous research has demonstrated a correlation between the semivolatile organic component (i.e., they partition between the gas and particle phase) of particles and their oxidative potential for DPM8 and also for wood smoke particles.9 Furthermore, a correlation has been demonstrated between the oxidative potential of particles and also PAH emission factors.10,11 Typically, the chemical properties of particulate emissions, such as PAHs and ROS are detected using oﬀ-line analytical chemistry techniques. The development of a near real-time technique enabling the detection of semivolatile organic compounds would be quite useful, given their importance in assessing the health eﬀects of DPM. As PAHs and ROS are both classed as semivolatile organic compounds, it is therefore possible that heating diluted exhaust within a TD will provide near real-time qualitative information on the presence of these components. As a result, a secondary objective of this work was to assess whether online measurements of the organic volume percentage (VORG) of DPM can provide information on genotoxic compounds on the surface of the particle that are usually measured using oﬀ-line analytical chemistry techniques. To achieve this objective, the relationship between VORG and ROS concentrations is explored. Historically, the regulation of DPM emissions has been achieved using a mass-based emissions standard;12 however, a particle number standard for heavy duty diesel engines will be introduced in the European Union at the Euro VI stage.12 While there have been studies suggesting that particle number emissions correlate with respiratory13 and cardiovascular14 morbidity from DPM more adequately than particle mass, toxicological studies have shown a strong inﬂammatory response from inert ultraﬁne particles in a size-dependent manner.15,16 Consequently, the toxicological literature suggests that particle surface area could be a relevant metric for assessing DPM health eﬀects. Given that DPM is quite often composed of a solid elemental carbon core with internally mixed semivolatile organics,17 a surface area based metric would provide information on the ability of toxic organic compounds to adsorb or condense on the surface of the particle. Consequently, a third objective of this work was to critically examine whether regulation of the DPM surface area emitted by a compression ignition engine has merit. All of the research objectives have been undertaken by investigating particle emissions from a nonroad diesel engine operated with various biodiesel feedstocks and blend percentages.

2. METHODOLOGY 2.1. Engine and Fuel Specifications. Particulate emissions testing was performed on a naturally aspirated 4 cylinder Perkins 1104C-44 engine with a Euro II (off-road) emissions certification. The engine investigated is typical of those used in underground mines in Australia and is the same engine used in Surawski et al.11 The engine was coupled to a Heenan & Froude water brake dynamometer (DPX 4) to provide a load to the engine. Ultralow sulfur diesel (denoted ULSD hereafter, <10 ppm sulfur) was used as the baseline fuel in this experiment, along with 13 biodiesel blends from 3 diﬀerent feedstocks, all of which were commercially available in Australia. All blends were prepared using calibrated graduated cylinders using a single batch of ULSD. The 3 biodiesel feedstocks investigated were soy, tallow, and canola, with each feedstock being investigated at 4 diﬀerent

ARTICLE

blend percentages, namely: 20%, 40%, 60%, and 80%. The opportunity arose during testing to undertake particle physical measurements with neat (i.e., 100%) soy biodiesel. The notation “BX” denotes that X% is the percentage (by volume) of the total blend made from biodiesel. In total, 14 diﬀerent fuel types were investigated in this study, all of which were undertaken at intermediate (i.e., 1400 rpm) speed full load. This test mode has the highest weighting in the ECE R49 test cycle introduced for Euro II engines and hence was selected for investigation in this study as it is the most representative mode from this test cycle.18 Particle physical measurements were made with all 14 fuel types, whereas particle chemical measurements were only made for ULSD, B20, and B80 blends made with each biodiesel feedstock. Further details on the engine speciﬁcations, the daily warm-up and oil changing procedure can be found in Surawski et al.11 2.2. Particulate Emissions Measurement Methodology. The methodology used for diluting the exhaust sample follows that of Surawski et al.19 and consists of a partial flow dilution tunnel followed by a Dekati ejector diluter. The methodology for measuring particle number size distributions follows that of Surawski et al.;19 however, a TSI 3010 condensation particle counter (CPC) was used instead of a TSI 3782 CPC. The methodology for measuring ROS is identical to that used in Surawski et al.19 Particle volatility was explored by passing the poly disperse size distribution through a TSI 3065 TD set to 300 °C. A correction for TD diffusional losses was performed using dried sodium chloride (NaCl) particles produced by an atomizer. The TD loss curve was obtained by measuring the NaCl particle number size distribution upstream and downstream of the TD (set to 300 o C), by switching the flow with a 3-way valve, and then calculating the proportion of particles lost (ηloss) via: ηloss = (1-(PNdownstream)/(PNupstream)), where PN denotes particle number concentration. PM10 measurements were obtained with a TSI 8520 DustTrak and were converted to a gravimetric measurement using the tapered element oscillating microbalance to DustTrak correlation for DPM obtained by Jamriska et al.20 The particle mass and number size distributions were all measured after the second stage of dilution. Measurements of particle phase and vapor phase PAHs were also performed. 2-Bromonaphthalene and the following US EPA priority PAHs in dichloromethane were quantiﬁed with a gaschromatography mass-spectrometry (GC-MS) system: naphthalene, acenaphthylene, acenaphthene, ﬂuorene, phenanthrene, anthracene, ﬂuoranthene, pyrene, benzo(a) anthracene, chrysene, benzo(b)ﬂuoranthene, benzo(a)pyrene, indeno(1,2,3-cd)pyrene, dibenzo(a,h)anthracene, and benzo(g,h,i)perylene. The methodology for sampling and quantiﬁcation following guidelines is presented in Lim et al.,21 and further information on the extraction procedure and the GC-MS system can be found in ref 11. Particle phase PAHs were collected on ﬁlters, and vapor phase PAHs were collected in tubes containing XAD-2 adsorbent prior to their quantiﬁcation using the GC-MS system. An in vitro cell-free assay was used to determine the oxidative capacity of particles, hereafter, referred to as ROS concentrations (inferred from ﬂuorescence measurements).22 For the ROS measurements, particles were bubbled through impingers (a test impinger and a HEPA ﬁltered control impinger) containing 20 mL of 4 μM BPEAnit solution, using dimethylsulfoxide (DMSO) as a solvent. More details on the ROS sampling and quantiﬁcation methodology such as the impinger collection eﬃciency, nitroxide probe theory, and its application to various 10338

dx.doi.org/10.1021/es2018797 |Environ. Sci. Technol. 2011, 45, 10337–10343

Environmental Science & Technology

Figure 1. Brake speciﬁc PM10 emission factors (g/kWh) for the 14 fuel types investigated in this study.

combustion sources can be found in Miljevic et al.9,22,23 All the ROS results were normalized to the gravimetric PM10 mass to give ROS concentrations in units of nmol/mg. Measurements of particle and vapor phase PAHs and ROS were made from the dilution to enable suﬃciently high concentrations for analysis. For the chemical measurements (i.e., PAHs and ROS) ﬁve replicates were used for ULSD and B80 soy, whereas for the other fuel types (B20 and B80 tallow and canola and B20 soy) three replicates were obtained. A diagram of the complete experimental setup can be found in the Supporting Information from Surawski et al.11 2.3. Data Analysis. Particles from biodiesel combustion usually exhibit a higher semivolatile organic fraction.6 As a result, heating biodiesel combustion particles with a TD should lead to a greater reduction in particle size compared with heating DPM. To quantify the volume reduction of particles upon heating with a TD, VORG (see Figure 7) was calculated from integrated particle volume size distributions obtained with a scanning mobility particle sizer (SMPS) via Vraw VTD ð1Þ VORG ð%Þ ¼ 100 Vraw where Vraw is the particle volume for unheated particles, and VTD is the particle volume for particles passed through a TD set to 300 o C. The assumption of spherical particles was made when performing calculations with eq 1. Raw results reporting the physicochemistry of DPM for all 14 fuel types along with dilution ratios can be found in Table S1. Standard error bars (i.e. ( standard error of the mean) are included on all ﬁgures to indicate variability in measurement precision for all measured quantities. Due to high measurement precision, the error bars are not visible on some graphs (e.g. Figures 1, 2, and 4). Note that error bars are not added to Figure 3 to avoid cluttering this ﬁgure. A full statistical analysis of the results using a two-way Analysis of Variance (ANOVA) appears in the supplementary information as well.

3. RESULTS AND DISCUSSION 3.1. PM10 Emission Factors. Figure 1 displays the brakespecific PM10 emission factors for all 14 fuel types investigated in

ARTICLE

Figure 2. Brake-speciﬁc particle number emissions (#/kWh) for the 14 fuel types investigated in this study.

this study. This figure shows that PM10 emission factors decrease in a monotonic fashion with respect to biodiesel blend percentage and that the PM10 emissions are also strongly dependent on biodiesel feedstock. For the soy feedstock, PM10 reductions range from 43% with B20 to 92% with B100, reductions in PM10 range from 58% for B20 to 88% for B80 for the tallow feedstock, whereas for the canola feedstock, the reductions range from 65% with B20 to 88% for B80. The observation of very large reductions in particulate matter emissions with biodiesel is a very commonly reported result in the biodiesel literature,3,4 with the results from this study confirming this general trend. 3.2. Particle Number Emission Factors. Figure 2 shows brake-specific particle number emission factors (#/kWh) for all 14 fuel types. The results show a strong dependency on both biodiesel feedstock and blend percentage. For the soy feedstock, particle number reductions range from 4% (B40) to 53% (B100), while for B20 a 12% particle number increase occurs. Particle number increases range from 71% (B20) to 44% (B80) for the canola feedstock. For the tallow feedstock, particle number increases range from 7% (B20) to 25% (B40), while a particle number reduction of 14% occurs for B80. A puzzling result to emerge from this study was the nonmonotonic trends in particle number emissions with respect to blend percentage. For all 3 feedstocks, a 20% blend increased particle number emissions, and for subsequent increases in blend percentage, the particle number emissions decreased. An exception to this trend was the tallow feedstock, which produced increased particle number emission for both 20% and 40% blends followed by subsequent decreases in particle number emissions with further increases in blend percentage. Nonmonotonic particle number emissions (relative to ULSD) with increasing blend percentages were observed by Di et al.,24 where the particle number increases were reduced as the diethylene glycol dimethyl ether blend (an oxygenated alternative fuel) percentage was increased. Di et al.24 suggested that particle oxidation kinetics were responsible for this result, with oxidation being suppressed at low blend percentages (giving particle number increases) and oxidation being promoted at high blend percentages (giving particle number reductions). This is a ﬁnding that should be investigated further with other biofuels. Given the absence of combustion-related diagnostic data, it is quite diﬃcult to provide a detailed mechanistic description of this result at this stage. 10339

dx.doi.org/10.1021/es2018797 |Environ. Sci. Technol. 2011, 45, 10337–10343

Environmental Science & Technology

ARTICLE

Figure 3. Particle number size distributions (corrected for dilution) for all fourteen fuel types (top panel: soy feedstock, middle panel: tallow feedstock, bottom panel: canola feedstock). TD denotes tests where diesel aerosol was passed through a TD set to 300 °C.

Figure 4. Count median diameter of particles (derived from a particle number size distribution) for all fourteen fuel types.

Variability in regulated emissions from compression ignition engines (i.e., PM, NOx, CO, and HCs) employing various biodiesel feedstocks is a topic that has been addressed fairly comprehensively in the diesel emissions literature.4,25,26 The variability of particle number emissions with diﬀerent biodiesel feedstocks, however, is a topic that has only been addressed recently.27 Fontaras et al.27 found that particle number emissions could be higher for biodiesel (by up to a factor of 3) due to the occurrence of nucleation with soy blends; however, reductions in particle number were achieved with other biodiesel feedstocks (such as palm and used frying oil methyl esters). The observation of variability in particle number emissions with diﬀerent biodiesel feedstocks has implications for conducting future biodiesel studies as this suggests that measurements should be conducted on an individual basis, rather than assuming generalizable trends with diﬀerent feedstocks. 3.3. Particle Number Size Distributions. Particle number size distributions for all 14 fuel types are shown in Figure 3; with all size distributions showing unimodality with a peak only in the accumulation mode. It can be seen from this graph that fuel type and blend percentage have varying effects on the observed particle number size distribution. While all fuel types display a shift to smaller particle diameters, the number of particles emitted is greater than that emitted by ULSD for all 4 canola blends, it is greater than ULSD for 2 tallow blends (less than ULSD for 2 blends), and it is greater than ULSD for only one soy blend (less than ULSD for the other 4 blends). Another feature evident from the particle number size distributions is that all biodiesel fuel types are particularly effective at reducing particle number concentrations at larger mobility diameters (>200 nm); however, for smaller mobility diameters (<50 nm) the number concentration of nanoparticles emitted is increased especially for the canola and tallow fuel types. Overall, the size distribution results presented here are quite different than those that are commonly reported, since increases in the accumulation mode particle concentrations are observed without the occurrence of nucleation. This effect is particularly evident for the canola blends, but also for the lower percentage tallow blends (B20B60), and also for one soy blend (B20). A signiﬁcant reduction in the count median diameter (CMD) of particles occurs as the biodiesel blend percentage is increased, which is a result that is commonly reported (but is certainly not a

universal trend) in the biodiesel literature.3 Canola blends (B20B80) exhibit the largest reduction in CMD (1933%), followed by tallow (1030%), with soy blends showing the smallest reduction in CMD (619%) (see Figure 4). Factors that could contribute to a reduced CMD with biodiesel include the following: the relative ease with which the biodiesel particle surface can be oxidized28 and also structural compaction of the particles.29 Structural compaction of particles (characterized by particles having a higher fractal dimension) would reduce the drag force on particles in a diﬀerential mobility analyzer which could reduce a particle’s transit time hence providing a reduction in the particle’s electrical mobility diameter. The particle number size distributions whereby diesel aerosol was passed through a TD (shown in Figure 3 for the B80 blends) can also oﬀer information on the mixing state of particles. Heating the particles with a TD led to a reduction in the median size of particles without a reduction in particle number for all feedstocks (except canola), which suggests that the semivolatile organic component of particles for the soy and tallow feedstocks are present as an internal mixture. Alternatively, for the canola feedstock, a reduction in particle number occurred in addition to a reduction in particle size which suggests that the presence of an external mixture of some purely volatile particles, in addition to some partially volatile particles for this fuel type. The presence of an external mixture containing some fully volatile organic compounds for the canola blends has implications for DPM health eﬀects, as inﬂammation and oxidative stress (precursors to some cardiovascular and respiratory diseases) are more heavily driven by the presence of organic compounds, rather than inert substances, such as soot.30,31 3.4. PAH Emission Factors and ROS Concentrations. Figure 5 displays the particle phase and vapor phase PAH emission factors. It can be observed that both particle and vapor phase PAHs are reduced for all 6 biodiesel fuel types (relative to the ULSD results), except for the B80 soy particle phase result. Particle phase PAH reductions range from a 3.5% increase for B80 soy to a decrease of about 60% for B80 canola. Vapor phase PAH reductions range from 33% for B80 soy to 84% for B20 tallow. Overall, very strong feedstock dependency can be observed for the PAH emissions factors, with the tallow feedstock generally providing the greatest reduction in particle and vapor phase PAHs (1684%), followed by the canola feedstock (no change 62% 10340

dx.doi.org/10.1021/es2018797 |Environ. Sci. Technol. 2011, 45, 10337–10343

Environmental Science & Technology

Figure 5. Brake-speciﬁc particle phase (top panel) and vapor phase (bottom panel) PAH emissions for the 7 fuel types where chemical analysis was performed. Error bars denote ( one standard error of the mean.

decrease), with the soy feedstock generally providing the smallest particle and vapor phase PAH reductions (4% increase 59% decrease). These results are consistent with the findings of Karavalakis et al.32 and Ballesteros et al.33 who both found vastly different PAH emission profiles when the biodiesel feedstock was changed. In terms of the PAH reductions with biodiesel, the USEPA4 states that the emissions of toxics (such as PAHs) should decrease with biodiesel. This is due to the correlation between emissions of toxics and emissions of hydrocarbons - which are generally reduced with biodiesel.3 Despite the reduction in particle and vapor phase PAHs with biodiesel, a concerning result is the phase distribution of the PAHs. PAHs with a greater number of aromatic rings (and hence higher molecular weight) exist in the particle phase and have a greater carcinogenicity than lower molecular weight, gas phase PAHs.34 The percentage of PAHs that are in the particle phase range from 44 to 75%, a result that is substantially higher than that reported by He et al.,35 who reported particle phase PAH percentages (i.e., of the total PAH emissions) ranging from 19 to 31% for a range of soy biodiesel blends. Another feature that may be observed from the PAH vapor phase results is how the emissions are independent of, or do not vary signiﬁcantly with, biodiesel blend percentage for the soy and canola fuel types. This experimental result was also observed by Ballesteros et al.,33 who noted that PAH reductions with rapeseed and waste cooking oil methyl esters did not exhibit a linear reduction with biodiesel blend percentage. ROS concentrations for the 6 fuel types where a ﬂuorescence signal was obtained (i.e., no data for B20 soy) are shown in Figure 6. From Figure 6, it can be observed that the ROS concentrations increase with biodiesel blend percentage, although there is not strong feedstock dependency, unlike some of the particle physical measurements presented thus far (e.g., particle number emission factors). Relative to neat diesel, ROS concentrations are reduced by 21% for B20 tallow and are increased by 16% for B20 canola. For the B80 tests, the tallow feedstock increased ROS concentrations by a factor of just over 9, for the soy feedstock an almost 10-fold increase was observed, while the B80 canola test increased ROS concentrations by a factor of approximately 7. 3.5. Particle Volatility and ROS Correlation. ROS are generally classed as “semivolatile” organic compounds that evaporate when

ARTICLE

Figure 6. ROS concentrations (nmol/mg) for the 6 fuel types where a ﬂuorescence signal was obtained.

Figure 7. A correlation between ROS concentrations and VORG for particles.

exposed to thermal treatment with a TD.9 Therefore, it is possible that qualitative information on ROS concentrations can be gained by investigating the volatility of particles. eq 1 demonstrated how VORG could be calculated from the integrated raw (i.e., non TD) and heated (i.e., with TD) particle volumes. Figure 7 represents an attempt to establish a correlation between VORG, or the volatility of particles, and their associated ROS concentrations. It can be observed from this graph that as the biodiesel blend percentage is increased, particles are internally mixed with more ROS (i.e., internal mixing present for soy and tallow feedstocks but not canola) and also have a higher VORG. Despite the presence of considerable scatter in the relationship, the Pearson correlation coefficient is quite strong (∼ 0.91). Consideration of the volatility of particles with a TD is, therefore, able to provide potentially useful information on ROS concentrations. 3.6. Particle Surface Area and Organic Volume Percentage of Particles. Toxicological studies, such as in ref 16, have pointed to the particle surface area as a potential metric for assessing the health effects of DPM. The surface area of a particle provides a measure of the ability of toxic compounds (such as PAHs or ROS) to adsorb or condense upon it. Therefore, a particle’s 10341

dx.doi.org/10.1021/es2018797 |Environ. Sci. Technol. 2011, 45, 10337–10343

Environmental Science & Technology

ARTICLE

’ AUTHOR INFORMATION Corresponding Author

*Phone: +617 3138 1129. Fax: +617 3138 9079. E-mail: z.ristovski@ qut.edu.au.

’ ACKNOWLEDGMENT The authors wish to acknowledge support and funding provided by SkillPro Services Pty Ltd and the Australian Coal Association Research Program for funding project C18014. Special thanks go to Mr. Julian Greenwood and Mr. Dale Howard, from SkillPro Services, for their technical expertise throughout testing and also for operating the dynamometer and providing the gaseous emissions and diagnostic test data. Figure 8. A graph showing the relationship between the heated particle surface area of DPM and VORG for all fuel types investigated.

surface area can be viewed as a “transport vector” for many compounds deleterious to human health. Figure 8 shows a relationship between the heated particle surface area (i.e., heated with a TD and assuming spherical particles) and VORG, plotted with respect to biodiesel blend percentage for all fuel types investigated. The heated particle surface area is employed in Figure 8 as this provides a good estimate of the total surface area that is available for adsorption or condensation. With increasing biodiesel blend percentage the heated particle surface area is reduced, with reductions ranging from no change to 74% with the soy fuel types, reductions of 1465% were achieved with tallow, and reductions of 1555% were achieved with canola fuel blends. Alternatively, as the biodiesel blend percentage is increased, the particles are composed of a greater VORG. Changes in VORG range from a 50% reduction to a 160% increase for soy fuel types, while for tallow, VORG increases are between 13 and 150%, while for canola fuel types, VORG ranges from a 19% decrease to a 190% increase. As was demonstrated in Figure 7, particles which contain a greater VORG display a concomitant increase in their ROS concentrations and hence the ability of these particles to induce oxidative stress. This is a particularly important result, as for alternative fuels to be a viable alternative to ULSD they must be able to deliver not only a reduction in the surface area of particles emitted (without a reduction in particle size) but also a reduction of semivolatile organics internally mixed within the particle surface. The results presented in Figures 7 and 8 naturally have implications for the regulation of DPM exhaust emissions using a surface area based metric. Regulating only the raw particle surface area emitted by a compression ignition engine would not be able to provide meaningful information on results such as those presented in Figure 8, as the surface chemistry of particles is not explicitly considered. Therefore, not only the raw surface area of particles but also the surface chemistry of particles is important for assessing the health impacts of DPM. These results suggest that the development of instrumentation (and standards) that enable the internal mixing status of particles to be determined (within a surface area framework) are potentially required.

’ ASSOCIATED CONTENT

bS

Supporting Information. Table S1-S7. This material is available free of charge via the Internet at http://pubs.acs.org.

’ REFERENCES (1) Agarwal, A. K. Biofuels (alcohols and biodiesel) applications as fuels for internal combustion engines. Prog. Energy Combust. Sci. 2007, 33 (3), 233–271. (2) McCormick, R. L. The impact of biodiesel on pollutant emissions and public health. Inhalation Toxicol. 2007, 19 (12), 1033–1039. (3) Lapuerta, M.; Armas, O.; Rodriguez-Fernandez, J. Eﬀect of biodiesel fuels on diesel engine emissions. Prog. Energy Combust. Sci. 2008, 34 (2), 198–223. (4) United States Environmental Protection Agency. A comprehensive analysis of biodiesel impacts on exhaust emissions. Draft technical report; 2002; pp 1126. (5) Giechaskiel, B.; Alfoldy, B.; Drossinos, Y. A metric for health eﬀects studies of diesel exhaust particles. J. Aerosol Sci. 2009, 40, 639–651. (6) Knothe, G.; Sharp, C. A.; Ryan, T. W. Exhaust emissions of biodiesel, petrodiesel, neat methyl esters, and alkanes in a new technology engine. Energy Fuels 2006, 20 (1), 403–408. (7) Sklorz, M.; Briede, J. J.; Schnelle-Kreis, J.; Liu, Y.; Cyrys, J.; de Kok, T. M.; Zimmermann, R. Concentration of oxygenated polycyclic aromatic hydrocarbons and oxygen free radical formation from urban particulate matter. J. Toxicol. Environ. Health, Part A 2007, 70 (21), 1866–1869. (8) Biswas, S.; Verma, V.; Schauer, J. J.; Cassee, F. R.; Cho, A. K.; Sioutas, C. Oxidative Potential of Semi-Volatile and Non Volatile Particulate Matter (PM) from Heavy-Duty Vehicles Retroﬁtted with Emission Control Technologies. Environ. Sci. Technol. 2009, 43 (10), 3905–3912. (9) Miljevic, B.; Heringa, M. F.; Keller, A.; Meyer, N. K.; Good, J.; Lauber, A.; Decarlo, P. F.; Fairfull-Smith, K. E.; Nussbaumer, T.; Burtscher, H.; Prevot, A. S. H.; Baltensperger, U.; Bottle, S. E.; Ristovski, Z. D. Oxidative Potential of Logwood and Pellet Burning Particles Assessed by a Novel Proﬂuorescent Nitroxide Probe. Environ. Sci. Technol. 2010, 44 (17), 6601–6607. (10) Cheung, K. L.; Ntziachristos, L.; Tzamkiozis, T.; Schauer, J. J.; Samaras, Z.; Moore, K. F.; Sioutas, C. Emissions of Particulate Trace Elements, Metals and Organic Species from Gasoline, Diesel, and Biodiesel Passenger Vehicles and Their Relation to Oxidative Potential. Aerosol Sci. Technol. 2010, 44 (7), 500–513. (11) Surawski, N. C.; Miljevic, B.; Ayoko, G. A.; Roberts, B. A.; Elbagir, S.; Fairfull-Smith, K. E.; Bottle, S. E.; Ristovski, Z. D. Physicochemical Characterization of Particulate Emissions from a Compression Ignition Engine Employing Two Injection Technologies and Three Fuels. Environ. Sci. Technol. 2011, 45 (13), 5498–5505. (12) Dieselnet, Emissions standards, European Union, heavy-duty truck and bus engines. In Dieselnet technology guide. Ecopoint Inc: 2009. http://www.dieselnet.com/standards/eu/hd.php. (accessed 2nd February 2011). (13) Peters, A.; Wichmann, H. E.; Tuch, T.; Heinrich, J.; Heyder, J. Respiratory eﬀects are associated with the number of ultraﬁne particles. Am. J. Respir. Crit. Care Med. 1997, 155 (4), 1376–1383. 10342

dx.doi.org/10.1021/es2018797 |Environ. Sci. Technol. 2011, 45, 10337–10343

Environmental Science & Technology (14) Pekkanen, J.; Peters, A.; Hoek, G.; Tiittanen, P.; Brunekreef, B.; de Hartog, J.; Heinrich, J.; Ibald-Mulli, A.; Kreyling, W. G.; Lanki, T.; Timonen, K. L.; Vanninen, E. Particulate air pollution and risk of STsegment depression during repeated submaximal exercise tests among subjects with coronary heart disease - The exposure and risk assessment for ﬁne and ultraﬁne particles in ambient air (ULTRA) study. Circulation 2002, 106 (8), 933–938. (15) Brown, D. M.; Wilson, M. R.; MacNee, W.; Stone, V.; Donaldson, K. Size-dependent proinﬂammatory eﬀects of ultraﬁne polystyrene particles: A role for surface area and oxidative stress in the enhanced activity of ultraﬁnes. Toxicol. Appl. Pharmacol. 2001, 175 (3), 191–199. (16) Oberdorster, G. Pulmonary eﬀects of inhaled ultraﬁne particles. Int. Arch. Occup. Environ. Health 2001, 74 (1), 1–8. (17) Maricq, M. M. Chemical characterization of particulate emissions from diesel engines: A review. J. Aerosol Sci. 2007, 38 (11), 1079–1118. (18) Dieselnet, Emission test cycles ECE R49. In Dieselnet technology guide. Ecopoint Inc: 2000. http://www.dieselnet.com/standards/ cycles/ece_r49.html. (accessed 18th August 2011). (19) Surawski, N. C.; Miljevic, B.; Roberts, B. A.; Modini, R. L.; Situ, R.; Brown, R. J.; Bottle, S. E.; Ristovski, Z. D. Particle Emissions, Volatility, and Toxicity from an Ethanol Fumigated Compression Ignition Engine. Environ. Sci. Technol. 2010, 44 (1), 229–235. (20) Jamriska, M.; Morawska, L.; Thomas, S.; He, C. Diesel bus emissions measured in a tunnel study. Environ. Sci. Technol. 2004, 38 (24), 6701–6709. (21) Lim, M. C. H.; Ayoko, G. A.; Morawska, L.; Ristovski, Z. D.; Jayaratne, E. R. Inﬂuence of fuel composition on polycyclic aromatic hydrocarbon emissions from a ﬂeet of in-service passenger cars. Atmos. Environ. 2007, 41 (1), 150–160. (22) Miljevic, B.; Fairfull-Smith, K. E.; Bottle, S. E.; Ristovski, Z. D. The application of proﬂuorescent nitroxides to detect reactive oxygen species derived from combustion-generated particulate matter: Cigarette smoke - A case study. Atmos. Environ. 2010, 44 (18), 2224–2230. (23) Miljevic, B.; Modini, R. L.; Bottle, S. E.; Ristovski, Z. D. On the eﬃciency of impingers with fritted nozzle tip for collection of ultraﬁne particles. Atmos. Environ. 2009, 43 (6), 1372–1376. (24) Di, Y.; Cheung, C. S.; Huang, Z. H. Experimental investigation of particulate emissions from a diesel engine fueled with ultralow-sulfur diesel fuel blended with diglyme. Atmos. Environ. 2010, 44 (1), 55–63. (25) Durbin, T. D.; Collins, J. R.; Norbeck, J. M.; Smith, M. R. Eﬀects of biodiesel, biodiesel blends, and a synthetic diesel on emissions from light heavy-duty diesel vehicles. Environ. Sci. Technol. 2000, 34 (3), 349–355. (26) Wang, W. G.; Lyons, D. W.; Clark, N. N.; Gautam, M.; Norton, P. M. Emissions from nine heavy trucks fueled by diesel and biodiesel blend without engine modiﬁcation. Environ. Sci. Technol. 2000, 34 (6), 933–939. (27) Fontaras, G.; Kousoulidou, M.; Karavalakis, G.; Tzamkiozis, T.; Pistikopoulos, P.; Ntziachristos, L.; Bakeas, E.; Stournas, S.; Samaras, Z. Eﬀects of low concentration biodiesel blend application on modern passenger cars. Part 1: Feedstock impact on regulated pollutants, fuel consumption and particle emissions. Environ. Pollut. 2010, 158 (5), 1451–1460. (28) Jung, H.; Kittelson, D.; Zachariah, M. Characteristics of SME biodiesel-fueled diesel particle emissions and the kinetics of oxidation. Environ. Sci. Technol. 2006, 40 (16), 4949–4955. (29) Smekens, A.; Godoi, R. H. M.; Vervoort, M.; Van Espen, P.; Potgieter-Vermaak, S. S.; Van Grieken, R. Characterization of individual soot aggregates from diﬀerent sources using image analysis. J. Atmos. Chem. 2007, 56 (3), 211–223. (30) Ayres, J. G.; Borm, P.; Cassee, F. R.; Castranova, V.; Donaldson, K.; Ghio, A.; Harrison, R. M.; Hider, R.; Kelly, F.; Kooter, I. M.; Marano, F.; Maynard, R. L.; Mudway, I.; Nel, A.; Sioutas, C.; Smith, S.; BaezaSquiban, A.; Cho, A.; Duggan, S.; Froines, J. Evaluating the toxicity of airborne particulate matter and nanoparticles by measuring oxidative stress potential - A workshop report and consensus statement. Inhalation Toxicol. 2008, 20 (1), 75–99. (31) Nel, A. Air pollution-related illness: Eﬀects of particles. Science 2005, 308 (5723), 804–806.

ARTICLE

(32) Karavalakis, G.; Bakeas, E.; Stournas, S. Inﬂuence of Oxidized Biodiesel Blends on Regulated and Unregulated Emissions from a Diesel Passenger Car. Environ. Sci. Technol. 2010, 44 (13), 5306–5312. (33) Ballesteros, R.; Hernandez, J. J.; Lyons, L. L. An experimental study of the inﬂuence of biofuel origin on particle-associated PAH emissions. Atmos. Environ. 2010, 44 (7), 930–938. (34) International Agency for Research on Cancer. Overall Evaluations of Carcinogenicity: An updating of IARC Monographs Volumes 142. IARC Monographs Evaluation of Carcinogenic Risks to Humans Supplement 7; 1987; pp 1440. http://www.iarc.fr/ (accessed 17th August 2011). (35) He, C.; Ge, Y. S.; Tan, J. W.; You, K. W.; Han, X. K.; Wang, J. F. Characteristics of polycyclic aromatic hydrocarbons emissions of diesel engine fueled with biodiesel and diesel. Fuel 2010, 89 (8), 2040–2046.

10343

dx.doi.org/10.1021/es2018797 |Environ. Sci. Technol. 2011, 45, 10337–10343

ARTICLE pubs.acs.org/est

Chemical Characterization and Source Apportionment of Fine and Coarse Particulate Matter Inside the Refectory of Santa Maria Delle Grazie Church, Home of Leonardo Da Vinci’s “Last Supper” Nancy Daher,† Ario Ruprecht,‡ Giovanni Invernizzi,‡ Cinzia De Marco,‡ Justin Miller-Schulze,§ Jong Bae Heo,§ Martin M. Shafer,§ James J. Schauer,§ and Constantinos Sioutas†,* †

Department of Civil and Environmental Engineering, University of Southern California, Los Angeles, California, United States LARS Laboratorio di Ricerca Ambientale SIMG/ISDE, Milan, Italy § Environmental Chemistry and Technology Program, University of Wisconsin-Madison, Madison, Wisconsin, United States ‡

bS Supporting Information ABSTRACT: The association between exposure to indoor particulate matter (PM) and damage to cultural assets has been of primary relevance to museum conservators. PM-induced damage to the “Last Supper” painting, one of Leonardo da Vinci’s most famous artworks, has been a major concern, given the location of this masterpiece inside a refectory in the city center of Milan, one of Europe’s most polluted cities. To assess this risk, a one-year sampling campaign was conducted at indoor and outdoor sites of the painting’s location, where time-integrated ﬁne and coarse PM (PM2.5 and PM2.5 10) samples were simultaneously collected. Findings showed that PM2.5 and PM2.5 10 concentrations were reduced indoors by 88 and 94% on a yearly average basis, respectively. This large reduction is mainly attributed to the eﬃcacy of the deployed ventilation system in removing particles. Furthermore, PM2.5 dominated indoor particle levels, with organic matter as the most abundant species. Next, the chemical mass balance model was applied to apportion primary and secondary sources to monthly indoor ﬁne organic carbon (OC) and PM mass. Results revealed that gasoline vehicles, urban soil, and wood-smoke only contributed to an annual average of 11.2 ( 3.7% of OC mass. Tracers for these major sources had minimal inﬁltration factors. On the other hand, fatty acids and squalane had high indoor-to-outdoor concentration ratios with fatty acids showing a good correlation with indoor OC, implying a common indoor source.

1. INTRODUCTION Damage to cultural assets has been of growing interest to museum conservators and curators. There is mounting evidence correlating indoor air pollution, biological contamination, mass tourism, and variability in microclimate conditions with material deterioration.1,2 A major concern is damage by particulate matter (PM) to masterworks of art displayed in museums. Potential hazards include “soiling” (perceptible degradation of visual qualities) due to deposition of airborne particles, particularly elemental carbon, and soil dust.3 Further damage can be induced by chemically reactive species, such as ammonium sulfate and organic acids.2,4 Typically, indoor PM consists of outdoor-inﬁltrating and indoor-emitted particles in addition to indoor-formed particles through reactions of gas-phase precursors emitted both indoors and outdoors.5,6 Moreover, the level and composition of indoor PM are governed by a myriad of factors. These mainly consist of the ventilation system, ﬁltration eﬀect of the building envelope, deposition rate of particles as well as the intensity of indoor and r 2011 American Chemical Society

outdoor sources.7,8 An accurate characterization of airborne PM in museums is therefore essential for conserving the exhibited artifacts. An emerging concern is with PM-induced damage to the “Last Supper” painting, one of Leonardo da Vinci’s most famous artworks, located in the refectory of Santa Maria delle Grazie Church in Milan, Italy. Although this painting has survived many challenges, including bombing during World War II, it is yet facing another challenge. The “Last Supper” painting, which was majorly restored in the 20th century, is at risk with air pollution arising from its surrounding Milan area. Milan is one of the most polluted areas in Western Europe9 with PM10 air quality standards frequently exceeded.10 In an attempt to protect the painting, a sophisticated heating, ventilation, and Received: August 5, 2011 Accepted: November 9, 2011 Revised: October 31, 2011 Published: November 09, 2011 10344

dx.doi.org/10.1021/es202736a | Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology air conditioning (HVAC) system equipped with particle ﬁltration has been installed. To assess the eﬀectiveness of this control measure, we conducted a one-year sampling campaign at indoor and outdoor sites of the refectory. At both locations, ﬁne and coarse PM (PM2.5 and PM2.5 10, respectively) samples were simultaneously collected then analyzed for their chemical properties. In the present article, the indoor-to-outdoor relationship of key tracers of PM sources is investigated in order to evaluate the impact of indoor and outdoor sources on indoor particle levels. Furthermore, the chemical mass balance model is applied to identify and estimate sources contributions to indoor PM2.5 concentration. Results of this study provide a quantitative understanding on the composition, origin and level of PM inside the refectory. Ultimately, these ﬁndings can be used as guidelines for the implementation of additional, and particularly source-speciﬁc, control strategies to mitigate the concentration of particle components potentially detrimental to the “Last Supper” painting. They can also be used as a benchmark in future studies aimed at protecting indoor artworks and antiquities.

2. MATERIALS AND METHODS 2.1. Sampling Description. To characterize PM inside the refectory, PM2.5 and PM2.5 10 were simultaneously sampled at indoor and outdoor sites of the refectory. The sampling campaign lasted from December 2009 to November 2010. During this period, 24-hour size-segregated PM samples were collected on a weekly basis by means of two sets of Sioutas personal cascade impactor samplers (Sioutas PCIS, SKC Inc., Eighty Four, PA11). Every set consisted of two collocated PCIS loaded with 37 and 25 mm filters for fine and coarse PM analyses, respectively. Each of the PCIS was placed at the indoor or outdoor site and operated at a flow rate of 9 lpm. For the purpose of chemical analysis, one set of the PCIS was loaded with Teflon filters (Pall Life Sciences, Ann Arbor, MI), whereas the other one was loaded with quartz microfiber filters (Whatman International Ltd., Maidstone, England). PM mass concentration was determined from the mass loadings of the weekly Teflon filters as described in the Supporting Information (SI). The indoor sampling location was inside the refectory of Santa Maria delle Grazie Church, where da Vinci painted the “Last Supper” on one of its walls. Samples were collected at approximately 1 m directly below the painting and a few centimeters from the wall surface. The site is equipped with a newly deployed HVAC system, supplying 4000 m3/h total air ﬂow, of which 2000 m3/h are external fresh air. This system is operated continuously. The air-ﬂow rate inside the refectory, whose volume is 3130 m3, is 3000 m3/h, resulting in an air exchange rate of roughly 1 h 1. This relatively low air change rate helps avoid convective air velocities on the painting to the degree possible. The remaining air ﬂow rate goes into two 130 m3 isolating zones, located at the entrance and exit of the refectory, through which visitors pass for isolation and decontamination from outdoor pollution. Furthermore, the air is ﬁltered with plane, pocket and absolute ﬁlters as well as chemical ﬁlters (Puraﬁl, Inc.); more details about these ﬁlters as well as the design and operation of the HVAC system can be found in the SI. The number of visitors and duration of visit are limited to 25 persons and 15 min at any time between 8:15 a.m. and 6:45 p.m. Visits are allowed each day, except for Monday, with number of visitors averaging 1000 visitors/day. The temperature and relative humidity are automatically controlled

ARTICLE

to maintain constant conditions. Large variations in the thermohygrometric factors in the refectory, enhanced by the presence of visitors, may lead to an increase in the deposition rate of airborne pollutants on the painting, as previously demonstrated by Camuﬀo and Bernardi.12 The temperature of the backside room is also maintained at about 2 °C higher than the refectory’s temperature to avoid PM deposition on the painting due to thermophoresis eﬀects.3,13 Finally, it should be noted that the HVAC system failed for few days during one week of April. PM concentration substantially increased during that week, compared to the remaining weeks with noninterrupted system functioning. This eﬀect was more noticeable for PM2.5 10, for which a nearly 9-fold increase was observed, as shown in SI Figure S3. This occurrence, however, has minor impacts on the results where monthly averages are reported throughout this study. 2.2. Chemical Analyses. To conduct the chemical analyses, the Teflon and quartz filters were sectioned into portions. The fractions used for elemental and organic carbon (EC and OC, respectively) quantification were grouped into weekly samples and quantified using the NIOSH thermal optical transmission method.14 All remaining fractions, with the exception of few PM2.5 10 sections, were composited monthly. Given their low mass loading, February/March, April/May, and October/November coarse samples were composited bimonthly. These monthly and bimonthly fractions were analyzed for water-soluble OC (WSOC) and ions using a Sievers 900 Total Organic Carbon Analyzer15 and ion chromatography, respectively. Total elemental content of these composites was also measured using high resolution magnetic sector inductively coupled plasma mass spectrometry (Thermo-Finnigan Element 2).16 Additionally, organic speciation was conducted on PM2.5 filter sections using gas chromatography mass spectrometry (GC-6980, quadrupole MS-5973, Agilent Technologies). PM2.5 10 lacked sufficient mass for this analysis. Details of these analyses are provided in the SI. 2.3. Source Apportionment. A molecular marker chemical mass balance model (MM-CMB) that was mathematically solved with the U.S. Environmental Protection Agency CMB (EPACMB8.2) software was used to estimate primary and secondary source contributions to indoor fine OC on a monthly basis. The effective variance weighted least-squares algorithm was applied to apportion the receptor data to the source profiles.17 MM-CMB was conducted using primary molecular source tracers that were quantiﬁed in the PM2.5 samples. Markers that are chemically stable and secondary organic aerosol (SOA) tracers that are unique to their precursor gases were selected as ﬁtting species.18 These included levoglucosan, αββ-20R&S C27-cholestane, αββ-20R&S-C29-sitostane, 17α(H)-22,29,30trisnorhopane, 17α(H)-21β(H)-hopane, 17β(H)-21α(H)-30norhopane, benzo(b)ﬂuoranthene, benzo(k)ﬂuoranthene, indeno[c,d]pyrene, benzo(ghi)perylene, EC, aluminum (Al), and titanium (Ti). The input source proﬁles were based on the observed molecular markers and assumed representative of sources in Milan. These proﬁles included wood-smoke,19,20 urban soil, gasoline vehicles,21 and diesel emissions.21 Biogenic-derived SOA was not included in the model but its contribution was estimated using ﬁxed tracer-to-OC ratios.22 Moreover, the selected urban soil proﬁle is not speciﬁc to Milan. However, the choice of this proﬁle is not critical for the overall apportionment of ﬁne OC as its contribution to total OC mass is small23. Its selection was 10345

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology

ARTICLE

Figure 1. Monthly average indoor concentration (compared to outdoor concentration) and bulk composition for (a, c) PM2.5 and (b, d) PM2.5 Error bars represent one standard error.

nonetheless based on a comparison of the elemental ratios of the measured data to those of available soil proﬁles, where the urban soil proﬁle of St. Louis (Missouri)23 provided a best ﬁt. In contrast, natural gas, coal soot and toluene-derived (anthropogenic) SOA sources were not considered in the model, as their molecular markers were not detected in the samples. Furthermore, contributions from vegetative detritus were not apportioned because n-alkanes (C29 C33) did not exhibit an odd-carbon preference indicative of modern plant material. Lastly, the CMB model results were considered valid if they met speciﬁc acceptance criteria as outlined in the SI.

10.

3. RESULTS AND DISCUSSION 3.1. Indoor Outdoor Relationship. 3.1.1. Particulate Mass and Composition. Indoor and outdoor monthly average PM2.5

and PM2.5 10 mass concentrations are shown in Figure 1(a, b). Indoor concentrations were substantially lower than those outdoors for both particle modes. This significant reduction in PM2.5 and PM2.5 10 concentration (88 ( 7% and 94 ( 3% on a yearly average ((standard deviation) basis, respectively) can be largely attributed to the efficacy of the HVAC system in removing infiltrating outdoor PM. Moreover, coarse PM exhibited 10346

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology extremely low indoor levels ranging between 0.12 and 0.83 μg/m3 with no specific seasonal trend. PM2.5 was the dominant indoor PM component with a concentration range of 1.7 4.9 μg/m3. It also followed a pattern dissimilar to that of its outdoor component, indicating that indoor sources may have major contributions to fine PM indoors. Finally, it is noteworthy that currently there are no regulations for PM levels in museums, galleries, and archives. However, the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE)24 as well as the Canadian Conservation Institute,25 provide recommendations for PM2.5. They suggest concentration limits of <0.1 and 1 10 μg/m3 for sensitive materials and general collections, respectively. PM2.5 levels in the refectory are within the limit values for general collections, but greater than those for sensitive materials. However, it is important to recognize that attaining such limits requires controls that may not be feasible and realistic. To determine the monthly bulk composition of indoor PM2.5 and PM2.5 10, a chemical mass balance was conducted as illustrated in Figure 1(c, d). PM chemical species were classiﬁed into water-insoluble organic matter (WIOM), water-soluble organic matter (WSOM), EC, ions, crustal material (CM), and trace elements (TE). WIOM and WSOM were determined by multiplying both WSOC and water-insoluble OC (WIOC = OCWSOC) concentrations by a factor of 1.7.26 28 Further details about the reconstruction are provided in the SI. WIOM was the dominant component of PM2.5 and PM2.5 10, accounting for an average ((one standard deviation) of 70.9 ( 16.4% and 61.3 ( 29.7% of total PM mass, respectively. PM2.5WIOM concentrations exhibited some variation, ranging between 1.7 μg/m3 in December and 3.1 μg/m3 in July. Conversely, coarse mode WIOM displayed lower concentrations, varying between 0.10 μg/m3 in December and 0.38 μg/m3 in April/May. WSOM was the next most abundant component of PM2.5 but only a minor fraction of OM, comprising 17.8 ( 4.4% and 19.9 ( 2.9% of their total mass, respectively. Moreover, WSOM only accounted for 8.8 ( 4.9% of coarse PM mass. EC, on the other hand, only contributed to PM2.5. Its concentration and relative proportion were however minimal (<0.05 μg/m3 and 1.5%). CM accounted for 8.2 ( 2.2% and 14.6 ( 4.7% of PM2.5 and PM2.5 10, respectively. Lastly, ions accounted for 3.1 ( 1.8% of PM2.5 and 5.4 ( 3.5% of PM2.5 10. The agreement between the reconstructed and gravimetric mass is overall good, averaging 104 ( 20% for PM2.5 and 88 ( 29% for PM2.5 10. The observed discrepancy could be related to uncertainties in the conversion factors from WSOC to WSOM, WIOC to WIOM, and metals to oxides as well as uncertainties in the measured mass, particularly for coarse PM. 3.1.2. Infiltration Ratios of Tracer Species. To investigate the influence of outdoor and indoor sources on PM levels inside the refectory, seasonal average indoor-to-outdoor (I/O) mass ratios and their standard deviations were determined for key tracers of major PM sources as summarized in Table 1 for PM2.5. I/O Spearman correlation coefficients (R) were also evaluated to determine whether an indoor species is attributable to infiltration from outdoors. This analysis, coupled with CMB results, provides a quantitative assessment of the infiltration of PM from specific outdoor sources. It should also be noted that concentrations that are below or comparable to the limit of detection (LOD), increase the uncertainty associated with the data, but should not cause a large overprediction. Concentration values that were below the LOD were assumed as half the detection limit. LODs of all measured species are listed in SI Table S1. The seasons were segregated as winter (December February), spring

ARTICLE

(March May), summer (June August), and fall (September November). Data for PM2.5 10, a minor component of indoor PM, is reported in SI Table S2. In consistence with its extremely low indoor mass, coarse PM exhibited small I/O ratios (e8%). For a given species, the combination of the I/O ratio and correlation provides an estimate of its I/O source relationship. The latter can be described by the following categories. First, a low I/O ratio accompanied by a good positive I/O correlation indicates a low but non-negligible inﬁltration factor and the lack of substantial indoor sources for a given species. Sulfate, which is a classic tracer of atmospheric outdoor aerosols with no known indoor sources29,30, presents a similar I/O source relationship, despite its usual association with high I/O ratios. However, in the current study, species that are usually considered to originate from outdoors (e.g., sulfate, EC in nonsmoking environments)29 31, have low I/O ratios due to their eﬀective removal by the HVAC system. Second, a species associated with a low I/O ratio and poor I/O correlation has a very low inﬁltration factor and no signiﬁcant sources impacting its indoor levels. Furthermore, a species with a high I/O ratio and a poor I/O correlation has a relatively low inﬁltration ratio but important indoor sources. For instance, OC, which is commonly associated with indoor sources29,31, exhibits a similar behavior. Lastly, a species with a high I/O ratio and good positive I/O correlation displays high inﬁltration eﬃciency and lacks important indoor sources. As can be inferred from Table 1, the I/O ratios were generally greater for PM2.5 than PM2.5 10 species, reﬂecting the lower inﬁltration eﬃciency and larger deposition velocity of coarse particles.32,33 Moreover, I/O ratios for ﬁne PM mass were below unity and presented some seasonality reaching a minimum (0.04 ( 0.03) in winter and a maximum (0.18 ( 0.06) in summer. These ratios were also accompanied by a negative I/O correlation (R = 0.31), indicating that indoor sources have major contributions to PM2.5. On the other hand, ionic species, sulfate, nitrate and ammonium, originated from outdoors (R = 0.66 0.76) but with very low inﬁltration factors (I/O ratio e2%). Similarly to ﬁne PM mass, OC and WIOC exhibited peak I/O ratios in summer (0.61 and 1.07, respectively) and were negatively correlated to their outdoor components, implying the existence of signiﬁcant indoor OM sources. Furthermore, WSOC, an indicator of SOA formation processes and biomass burning,34 showed a weak I/O association and slightly higher I/O ratios in spring and summer (∼0.20). These results reﬂect a possible formation of SOA indoors. EC, a key tracer for diesel exhaust,35 displayed a weak I/O correlation (R = 0.20) although it is expected to originate from outdoors given the prohibition of smoking inside the refectory. This could be related to its eﬃcient removal by ventilation, where EC was present at levels less than 0.05 μg/m3 with low I/O ratios (∼3%). These ﬁndings suggest a nominal inﬂuence of outdoor diesel emissions on indoor PM levels. Typical crustal metals, for example, Al, calcium (Ca), Ti and iron (Fe), exhibited I/O ratios ranging from 0.03 to 0.33 with similar peak occurrence in fall. Although these elements are expected to originate from outdoors, they were weakly to moderately related to their outdoor components (R = 0.17 0.45), which indicates their dependence on indoor sources, likely particle resuspension during visiting hours. Moreover, in spite of its weak I/O correspondence, potassium (K) was strongly associated with Ti and Al (R = 0.98 and 0.62, respectively) at the indoor site, indicating their common outdoor crustal source. Its poor I/O correlation may be attributed to its mixed outdoor origin. Marcazzan et al.36 reported that K is associated with motor vehicles in Milan. Conversely, Nickel (Ni), a marker of fuel oil 10347

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology

ARTICLE

Table 1. Indoor-to-Outdoor (I/O) Seasonal Average (One Standard Deviation) Mass Ratios and Spearman Correlation Coeﬃcients (R) for key species in PM2.5 average (standard deviation) I/O ratio

mass SO42‑ NO3 NH4+ OC WIOC WSOC EC Al Ca Ti Fe K Ni Cu Zn benzo(e)pyrene benzo(a)pyrene indeno(l,2,3-cd)pyrene benzo(ghi)perylene coronene picene 17β(H)-21α(H)-30-norhopane 17α(H)-21β(H)-hopane 22S-homohopane 22R-homohopane nonacosane triacontane hentriacontane dotriacontane tritriacontane squalane a tetradecanoic acid pentadecanoic acid hexadecanoic acid heptadecanoic acid octadecanoic acid nonadecanoic acid palmitoleic acid a oleic acid tributyl phosphate phthalic acid methylphthalic acid suberic acid azelaic acid 2-methylthreitol 2-methylerythritol 2-hydroxy-4-ispropyladipic acid pinonic acid 2,3-dihydroxy-4-oxopentanoic acid levoglucosan

winter

spring

summer

fall

R(I/O)

0.04(0.03) 0.01(0.003) <1% <1% 0.11(0.03) 0.11(0.04) 0.11(0.01) 0.03(0.007) 0.09(0.02) 0.17(0.04) 0.18(0.06) 0.03(0.01) 0.03(0.02) 0.04(0.01) 0.04(0.02) 0.02(0.02) 0.01(0.01) <1% 0.02(0.005) 0.01(0.004) 0.01(0.003) n.d 0.14(0.06) 0.22(0.14) 0.11(0.05) 0.10(0.05) 0.13(0.05) 0.20(0.08) 0.09(0.04) 0.24(0.17) 0.20(0.17) 167(71) 3.2(1.0) 1.1(0.35) 0.49(0.14) 0.42(0.09) 0.40(0.10) 0.15(0.05) 182(58) 2.2(0.8) 67(9.2) 0.04(0.02) 0.01(0.01) 0.06(0.06) 0.06(0.02) n.a n.a 0.26(0.21) n.a n.a 0.12(0.06)

0.14(0.10) 0.02(0.02) <1% <1% 0.38(0.25) 0.51(0.29) 0.20(0.05) 0.03(0.01) 0.13(0.07) 0.26(0.12) 0.27(0.11) 0.06(0.03) 0.14(0.09) 0.07(0.04) 0.06(0.02) 0.07(0.03) 0.03(0.03) 0.01(0.02) 0.03(0.06) 0.03(0.03) 0.02(0.03) n.d 0.23(0.01) 0.34(0.04) 0.15(0.02) 0.16(0.05) 0.17(0.03) 0.38(0.15) 0.11(0.04) 0.32(0.15) 0.14(0.04) 62(55) 5.8(0.62) 2.3(0.22) 1.0(0.11) 0.82(0.17) 0.62(0.08) 0.45(0.19) 315(34) 11(5.9) 32(8.5) 0.13(0.07) 0.03(0.03) 0.06(0.11) 0.13(0.04) n.d n.d 0.51(0.44) n.d n.d 0.09(0.10)

0.18(0.06) 0.01(0.001) 0.01(0.001) <1% 0.61(0.17) 1.07(0.45) 0.22(0.04) 0.03(0.01) 0.06(0.03) 0.15(0.06) 0.21(0.1) 0.05(0.02) 0.14(0.05) 0.04(0.01) 0.05(0.01) 0.10(0.05) n.d n.d n.d n.d n.d n.a 0.69(0.36) 0.85(0.7) 0.36(0.22) 0.32(0.11) 0.24(0.06) 0.76(0.25) 0.17(0.05) 0.65(0.21) 0.23(0.05) 58(6.4) 12(3.3) 6.1(1.7) 2.1(0.83) 1.3(0.23) 1.2(0.42) 0.73(0.17) 413(98) 37(17) 9.1(2.5) 0.07(0.02) n.d 0.17(0.02) 0.21(0.04) n.d n.d 1.07(0.48) n.d n.d 0.03(0.03)

0.11(0.07) <1% <1% <1% 0.25(0.12) 0.30(0.18) 0.15(0.03) 0.02(0.01) 0.19(0.03) 0.33(0.06) 0.31(0.12) 0.06(0.01) 0.12(0.07) 0.05(0.01) 0.04(0.01) 0.05(0.02) n.d n.d n.d <1% <1% n.a 0.13(0.07) 0.11(0.06) 0.08(0.03) 0.08(0.03) 0.14(0.05) 0.28(0.14) 0.08(0.02) 0.32(0.1) 0.18(0.10) 76(30) 11(5.0) 4.4(1.9) 0.94(0.52) 0.89(0.41) 0.52(0.22) 0.53(0.4) 346(55) 6.7(4.7) 37(36) 0.05(0.01) 0.03(0.01) 0.18(0.02) 0.14(0.05) n.a n.a 0.85(0.51) n.a n.a <1%

0.31 0.76 0.72 0.66 0.13 0.64 0.10 0.20 0.17 0.43 0.45 0.39 0.45 0.66 0.31 0.23 n.q n.q n.q n.q n.q n.q 0.61 0.73 0.53 0.62 0.38 0.30 0.59 0.47 0.69 0.68 0.88 0.66 0.57 0.20 0.20 -0.78 0.53 0.71 0.40 0.11 0.30 n.q n.q 0.43 n.q n.q 0.66

n.a, non detected indoor and outdoor compound. n.d, non detected indoor compound. n.q, Not quantiﬁed due to insuﬃcient number of data points (<6). R in bold is statistically signiﬁcant at a 0.05 level. a non detected outdoor concentration is replaced by 1/2 detection limit for I/O ratio computation. 10348

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology combustion,37,38 strongly correlated with its outdoor component (R = 0.66), signifying its mostly outdoor origin. Ni also displayed comparable I/O ratios (4 7%), indicating a stable but rather small outdoor inﬂuence on its indoor levels. Anthropogenic metals, copper (Cu) and zinc (Zn), displayed I/O ratios ranging from 0.02 to 0.10 and were weakly dependent on their outdoor components (R = 0.31 and 0.23, respectively), indicating a potential but small indoor inﬂuence such as their accumulation in indoor dust.39 This build up may in turn be determined by indoor and outdoor emissions.40 Among the listed polycyclic aromatic hydrocarbons (PAHs), picene, a molecular marker for coal soot,41 was not measured in the indoor samples despite its detection outdoors, implying that coal soot is not a source contributor to indoor PM levels. The remaining PAHs, common products of incomplete combustion including fossil fuel and biomass combustion,42 were mostly undetected indoors, especially in summer, and exhibited extremely low inﬁltration factors (e3%), indicating that their sources do not signiﬁcantly impact indoor PM levels. Hopanes, which are predominantly associated with engine lubricating oil of mobile sources,43 fairly correlated with their outdoor levels (R = 0.53 0.73), conﬁrming their outdoor origin. Excluding summer, during which high I/O ratios were observed for 17β(H)21α(H)-30-norhopane and 17α(H)-21β(H)-hopane (0.69 0.85), seasonal inﬁltration factors ranged from 0.08 to 0.36. These ratios highlight a year-long inﬂuence from vehicular sources on hopanes levels indoors. The peak summertime inﬁltration ratio may be a result of measurement uncertainties associated with the low outdoor hopanes levels (0.02 0.06 ng/m3). To investigate the origin of indoor n-alkanes (C29 C33), the carbon preference index (CPI), deﬁned as the concentration ratio of their odd-toeven numbered homologues, was estimated. A CPI about 1 indicates a dominance of anthropogenic sources, whereas a CPI greater than 2 indicates a prevalence of biogenic sources.44 These indoor compounds did not exhibit a discernible odd-to-even carbon number preference (CPI = 1.01 ( 0.14 on a yearly average basis), indicating their anthropogenic outdoor source, such as fossil fuel utilization and wood-smoke.45 Nonetheless, the low I/O correspondence for some of these n-alkanes can be related to the primarily biogenic nature of their outdoor components, which exhibited a CPI of 2.53 ( 0.61. Moreover, n-alkanes displayed highest I/O ratios (0.17 0.76) in summer, possibly related to condensation of inﬁltrating gas-phase n-alkanes onto indoor particles as a result of I/O temperature diﬀerences. In contrast, squalane existed in higher indoors than outdoor amounts. This undoubtedly suggests that indoor sources signiﬁcantly contribute to its presence indoors. Squalane is a naturally occurring compound in humans and plants as well as a compound used in skin care products,46,47 suggesting visitors as a possible source given the absence of plants in the refectory. n-alkanoic acids, C14 C19, were uncorrelated with their outdoor components and displayed relatively high I/O ratios with some greater than unity, indicating their predominantly indoor origin. Their I/O ratios also demonstrated a seasonal pattern with greatest ratios occurring in summer (0.73 12.4), in accordance with those of OC and WIOC. Moreover, these fatty acids displayed a yearly average CPI of 7.55 ( 0.92, indicative of their biogenic origin. For n-alkanoic acids, CPI is estimated as the concentration ratio of their even-to-odd numbered homologues. Potential indoor sources include skin emissions from visitors48 and wax49 emissions from the painting itself. Waxes were used during the restoration process of the painting.50 Furthermore,

ARTICLE

palmitoleic and oleic acids, normally associated with cooking,51 exhibited I/O ratios much greater than 1, also indicating their primarily indoor origin. However, given that cooking is prohibited inside the refectory, the most probable source is biogenic material45 potentially emitted from waxes.52,53 Emissions from skin surface lipids of visitors are also a likely source of oleic acid.48 These fatty acids are ubiquitous indoors and may be sorbed to indoor airborne particles or also settled dust54 that is subsequently resuspended by human/cleaning activities. Tributyl phosphate, a phosphate ester used in plasticizers and ﬂame retardants,55 persistently displayed I/O ratios exceeding unity, thereby indicating its indoor source, most likely wall or ceiling paint. In contrast, phtalic and methylphtalic acids, which exhibited low but non-negligible I/O ratios (e13 and 4%, respectively), inﬁltrated from outdoors (R = 0.71 and 0.40, respectively) with emissions from mobile sources and association with SOA formation45 as possible sources. The I/O ratios of suberic and azelaic acids, which are photo-oxidation products of biogenic unsaturated fatty acids,56 suggest the presence of SOA indoors. Nonetheless, the indoor levels of SOA are probably low, as WSOM only constitutes ∼20% of OM, as aforementioned. Tracers for biogenic-derived SOA include photo-oxidation products of α-pinene and isoprene. These comprise pinonic acid, 2-hydroxy-4-isopropyladipic acid, 2-methylthreitol, and 2-methylerythritol.22 Conversely, tracers for anthropogenicderived SOA include 2,3-dihydroxy-4-oxopentanoic acid, photo-oxidation product of toluene.22 Among these SOA tracers, only 2-hydroxy-4-isopropyladipic acid, derivative of α-pinene, was detected indoors with peak I/O ratio about unity (1.07 ( 0.48) in summer. This secondary compound also presented a poor I/O correlation (R = 0.43) suggesting its indoor formation. A likely pathway is gas-phase reactions involving α-pinene constituents and oxidants. The higher summertime I/O ratio reﬂects an enhanced production of SOA, possibly promoted by an increase in inﬁltrating oxidants. Lastly, the low I/O ratio and high I/O correlation for levoglucosan (R = 0.66), a tracer for biomass burning,57 indicates its low but non-negligible indoor intrusion, mainly in winter (12%). In summary, these ﬁndings show that key tracers of major outdoor sources generally have small inﬁltration factors. Additionally, it is particularly interesting that fatty acids were mainly of indoor origin with palmitoleic and oleic acids exhibiting I/O ratios >1. 3.2. CMB Results. 3.2.1. Source Apportionment of Fine OC. The monthly contributions of primary and secondary sources to indoor fine OC as estimated by the CMB model are shown in Figure 2a and summarized in SI Table S3a. Three sources, including wood-smoke, gasoline vehicles, and urban soil were identified. Contributions to OC from biogenic-derived SOA were not statistically significant (<2 standard error) with an utmost value of 0.015 μg/m3. Similarly, diesel emissions were not statistically significantly different from zero. The three sources collectively contributed to 6.3 20.7% of measured fine OC, with the remainder representing unidentified sources, likely including biogenic SOA. The largest contributor to OC mass consisted of gasoline vehicles, which accounted for 4.9 16.7% of OC. The largest percent contributions and source estimates occurred in cold months (December March) with highest average ((standard error) levels of 0.23 ( 0.024 μg/m3 attained in February. The next most contributing source was wood-smoke during winter, while urban soil during the 10349

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology

ARTICLE

Figure 2. Sources contribution to indoor (a) ﬁne organic carbon (OC) and (b) PM2.5 estimated using the chemical mass balance (CMB) model.

remaining seasons. Wood-smoke contribution was only statistically significant during December April and peaked in February to reach 0.039 ( 0.015 μg/m3 (i.e., 2.8% of OC mass). This seasonal pattern suggests wood burning for domestic heating during cold months. On the other hand, urban soil lacked any discernible seasonal trend and contributed to 1.14 ( 0.46% (0.018 ( 0.007 μg/m3) of OC mass on a yearly based average ((standard deviation). Lastly, unidentiﬁed source contributions, denoted as “other OC”, were estimated as the diﬀerence between measured OC and the contributions from the modeled sources. The percent of residual mass was higher during July October and could be attributed to uncharacterized primary and SOA sources. 3.2.2. Source Apportionment of Total PM2.5. Source contributions to total PM2.5 were assessed by converting CMB results for fine OC to PM2.5 apportionment using reported fine OC-to-PM mass ratios for each source.19 23 In addition to the sources identified in OC apportionment, “other OM” as well as sulfate, nitrate, and ammonium ion concentrations were considered in PM2.5 apportionment as displayed in Figure 2b and SI Table S3b. “Other OM” was estimated by multiplying “other OC” by a factor of 1.7.26 28 These sources collectively accounted for 96.2 ( 18.7% of the measured PM2.5 mass, on a yearly based average ((standard deviation). Some of the inconsistency in apportionment could be due to uncertainties associated with the conversion factor from OC to OM and geographical differences of the sources compositions. Finally, the most significant contributions were from “other OM” (80.5 ( 17.4%), followed by urban soil (6.9 ( 1.7%), gasoline vehicles (6.5 ( 2.8%), wood-smoke (1.2 ( 0.51%), then sulfate, nitrate and ammonium ions with contributions less than 1%. 3.3. Comparison of OC to Organic Acid Species in PM2.5. To elucidate the nature of uncharacterized OC, WSOC contribution to unapportioned OC was estimated as WSOC that is not associated with wood-smoke emissions. This calculation estimates wood-smoke contribution to OC as 71% water-soluble.58

Accordingly, unapportioned WIOC was determined as the difference between total WIOC and the sum of all primary source estimates (excluding wood-smoke) and WIOC from wood-smoke. These estimates are reported in SI Table S3a. As can be deduced, unattributed OC is largely water-insoluble (77.9 ( 3.2%), which indicates that uncharacterized OC sources are mostly primary. Given that major outdoor sources of OC have been included in the CMB model, and considering their low infiltration factors and contribution estimates, unknown primary sources are likely dominated by indoor sources such as dust of biogenic origin, or PM emissions from the visitors and the painting itself. The relative importance of ﬁne OM to indoor PM and the likelihood of its predominantly indoor source warrant further investigation of this aerosol component. Accordingly, monthly variations of indoor ﬁne WSOC, WIOC, fatty acids with I/O ratio >1 and CPI of n-alkanoic acids are examined as illustrated in Figure 3(a d). WSOC, attributed to SOA formation processes and biomass burning,34 mainly originated from indoors as previously noted. In fact, the contribution to WSOC from wood-smoke source was only signiﬁcant in winter and early spring when it averaged ((standard deviation) 0.02 ( 0.006 μg/m3. Total WSOC, on the other hand, maintained a stable average concentration of 0.31 ( 0.02 μg/m3 throughout the year, which further conﬁrms a continual and prevalent contribution to WSOC from indoor SOA formation processes. This contribution to overall OC was however minor, as WSOC only comprised 20 ( 3% of indoor OC across the year. WIOC, on the other hand, was a major component of indoor ﬁne OC, accounting for 80 ( 3% of its mass on a yearly average basis. It also follows closely unapportioned OC, which could not be assigned to outdoor sources, supporting the likelihood that OC is mostly insoluble of indoor primary origin. Furthermore, WIOC was present at a yearly average concentration of 1.25 ( 0.25 μg/m3, with higher levels observed during May October 10350

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology

ARTICLE

Figure 3. Monthly time series of (a) water-soluble organic carbon (WSOC), (b) water-insoluble organic carbon (WIOC) and un-apportioned organic carbon (OC), (c) palmitoleic and oleic acids, (d) carbon preference index of n-alkanoic acids (C14 C29), in ﬁne PM in the indoor environment.

and peak occurrence of 1.82 μg/m3 in July. These variations imply an increase in the source strength of indoor primary emissions during these months. To characterize potential indoor sources of OC, the monthly trend of palmitoleic and oleic acids, which were present in greater indoors than outdoor amounts, was investigated. These components exhibited a temporal distribution fairly similar to that of WIOC and unapportioned OC, reaching a collective peak of 41.9 ng/m3 in July and low of 14.4 ng/m3 in January. Particularly, palmitoleic and oleic acids, biogenic components sorbed to indoor airborne PM or dust as aforementioned, were signiﬁcantly and well correlated to WIOC (R = 0.75 and 0.76, respectively). Thus, this temporal correlation suggests their common source with biogenic material associated with indoor

dust, waxes used in the painting and human skin emissions as potential sources. Indoor n-alkanoic acids, C14 C29, exhibited a year-long strong even carbon preference, with an annual average CPI of 6.62 ( 0.55 and limited monthly variation. Carbon number maxima occurred at C16 and C14 (SI Table S4), which are commonly found indoors59 and associated with human skin surface lipids.48 These ﬁndings are indicative of the consistent biogenic source of these n-fatty acids such as waxes used in painting the “Last Supper” and skin emissions, as aforementioned. Moreover, these components highly correlate with WIOC (R = 0.73), a dominant component of OC, suggesting their shared origin with indoor dust of biogenic nature, emissions from the painting or human skin as likely sources. 10351

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology Overall, the results of this study showed that outdoor sources have small inﬁltration factors and contribution to PM2.5. OM, which could not be apportioned to any major outdoor sources, accounted for most of PM2.5 (80.5 ( 17.4%), and was largely water-insoluble (77.9 ( 3.2%) with indoor dust of biogenic origin as a potential source. Consequently, it can be concluded that controls to prevent inﬁltration of outdoor PM into the refectory, where the “Last Supper” painting is housed, are very eﬀective. However, additional measures targeting the reduction of ﬁne OM should be implemented. Particularly, these controls should address indoor sources of biological material that is likely associated with indoor dust. Lastly, we should note that ﬁndings of this study are characteristic of the speciﬁc site location, climatic conditions inside the refectory, visitors’ pattern and speciﬁcations of the HVAC system. These results, therefore, cannot be directly extrapolated to other exhibits.

’ ASSOCIATED CONTENT

bS

Supporting Information. Figures S1 S3 and Tables S1 S4. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT This research was supported by Southern California Particle Center, funded by US EPA and the University of Southern California (USC) Viterbi School of Engineering. We would like to thank the superintendent of ﬁne arts and culture in Lombardy for his willingness to accept this study. We also wish to acknowledge the support of USC Provost’s Ph.D. fellowship. We thank Jeﬀ DeMinter, Brandon Shelton and the staﬀ at the Wisconsin State Laboratory of Hygiene for their assistance with the chemical measurements. We also wish to thank SIMG-Italian College GPs and ISDE-International Doctors for the Environment for managerial support, Eng. Franco Gasparini, designer of the HVAC system for technical help, and the whole management and employees of the Sovrintendenza, particularly Arch. A. Artioli, Arch. G. Stolﬁ, Arch. Napoleone, Mr. G. Bonnet and Dr. L. Dall’Aglio ’ REFERENCES (1) Camuﬀo, D.; Van Grieken, R.; Busse, H.-J.; Sturaro, G.; Valentino, A.; Bernardi, A.; Blades, N.; Shooter, D.; Gysels, K.; Deutsch, F.; Wieser, M.; Kim, O.; Ulrych, U. Environmental monitoring in four European museums. Atmos. Environ. 2001, 35 (Supplement 1), S127–S140. (2) Brimblecombe, P. The composition of museum atmospheres. Atmos. Environ., Part B 1990, 24 (1), 1–8. (3) Nazaroﬀ, W. W.; Cass, G. R. Protecting museum collections from soiling due to the deposition of airborne particles. Atmos. Environ., Part A 1991, 25 (5 6), 841–852. (4) Alexandra, S.; Tunga, S.; Watts, S. F., Indoor pollutants in the museum environment. In Organic Indoor Air Pollutants, 2nd ed.; Salthammer, T., Uhde, E., Eds.; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, 2009; p 281. (5) Meng, Q. Y.; Turpin, B. J.; Korn, L.; Weisel, C. P.; Morandi, M.; Colome, S.; Zhang, J. J.; Stock, T.; Spektor, D.; Winer, A.; Zhang, L.; Lee, J. H.; Giovanetti, R.; Cui, W.; Kwon, J.; Alimokhtari, S.; Shendell, D.;

ARTICLE

Jones, J.; Farrar, C.; Maberti, S. Inﬂuence of ambient (outdoor) sources on residential indoor and personal PM2.5 concentrations: Analyses of RIOPA data. J. Exposure Anal. Environ. Epidemiol. 2005, 15 (1), 17–28. (6) Wallace, L. Indoor particles: A review. J. Air Waste Manage. Assoc. 1996, 46 (2), 98–126. (7) Salmon, L. G.; Nazaroﬀ, W. W.; Ligocki, M. P.; Jones, M. C.; Cass, G. R. Nitric acid concentrations in southern California museums. Environ. Sci. Technol. 1990, 24 (7), 1004–1013. (8) Nazaroﬀ, W. W. Indoor particle dynamics. Indoor Air 2004, 14 (Suppl 7), 175–183. (9) Putaud, J.-P.; Raes, F.; Van Dingenen, R.; Br€uggemann, E.; Facchini, M. C.; Decesari, S.; Fuzzi, S.; Gehrig, R.; H€uglin, C.; Laj, P.; Lorbeer, G.; Maenhaut, W.; Mihalopoulos, N.; M€uller, K.; Querol, X.; Rodriguez, S.; Schneider, J.; Spindler, G.; Brink, H. t.; Tørseth, K.; Wiedensohler, A. A European aerosol phenomenology—2: Chemical characteristics of particulate matter at kerbside, urban, rural and background sites in Europe. Atmos. Environ. 2004, 38 (16), 2579–2595. (10) Lonati, G.; Giugliano, M.; Butelli, P.; Romele, L.; Tardivo, R. Major chemical components of PM2.5 in Milan (Italy). Atmos. Environ. 2005, 39 (10), 1925–1934. (11) Misra, C.; Singh, M.; Shen, S.; Sioutas, C.; Hall, P. M. Development and evaluation of a personal cascade impactor sampler (PCIS). J. Aerosol Sci. 2002, 33 (7), 1027–1047. (12) Camuﬀo, D.; Bernardi, A. The Microclimate of Leonardo’s “Last Supper. Joint Edition European Cultural Heritage Newsletter on Research and Bollettino Geoﬁsico, Special Issue 1991, 14 (3), 1–123. (13) Nazaroﬀ, W. W.; Cass, G. R. Mass-transport aspects of pollutant removal at indoor surfaces. Environ. Int. 1989, 15 (1 6), 567–584. (14) Birch, M. E.; Cary, R. A. Elemental carbon-based method for occupational monitoring of particulate diesel exhaust: Methodology and exposure issues. The Analyst 1996, 121 (9), 1183–1190. (15) Stone, E. A.; Hedman, C. J.; Sheesley, R. J.; Shafer, M. M.; Schauer, J. J. Investigating the chemical nature of humic-like substances (HULIS) in North American atmospheric aerosols by liquid chromatography tandem mass spectrometry. Atmos. Environ. 2009, 43 (27), 4205–4213. (16) Zhang, Y.; Schauer, J. J.; Shafer, M. M.; Hannigan, M. P.; Dutton, S. J. Source apportionment of in vitro reactive oxygen species bioassay activity from atmospheric particulate matter. Environ. Sci. Technol. 2008, 42 (19), 7502–7509. (17) Watson, J. G. Overview of receptor model principles. Air Pollut. Control Assoc. 1984, 34 (6), 619. (18) Schauer, J. J.; Rogge, W. F.; Hildemann, L. M.; Mazurek, M. A.; Cass, G. R.; Simoneit, B. R. T. Source apportionment of airborne particulate matter using organic compounds as tracers. Atmos. Environ. 1996, 30 (22), 3837–3855. (19) Fine, P. M.; Cass, G. R.; Simoneit, B. R. T. Chemical characterization of ﬁne particle emissions from the ﬁreplace combustion of wood types grown in the Midwestern and Western United States. Environ. Eng. Sci. 2004, 21 (3), 387–409. (20) Sheesley, R. J.; Schauer, J. J.; Zheng, M.; Wang, B. Sensitivity of molecular marker-based CMB models to biomass burning source proﬁles. Atmos. Environ. 2007, 41 (39), 9050–9063. (21) Lough, G. C.; Christensen, C. G.; Schauer, J. J.; Tortorelli, J.; Mani, E.; Lawson, D. R.; Clark, N. N.; Gabele, P. A. Development of molecular marker source proﬁles for emissions from on-road gasoline and diesel vehicle ﬂeets. J. Air Waste Manage. Assoc. 2007, 57 (10), 1190–1199. (22) Kleindienst, T. E.; Jaoui, M.; Lewandowski, M.; Oﬀenberg, J. H.; Lewis, C. W.; Bhave, P. V.; Edney, E. O. Estimates of the contributions of biogenic and anthropogenic hydrocarbons to secondary organic aerosol at a southeastern US location. Atmos. Environ. 2007, 41 (37), 8288–8300. (23) Rutter, A. P.; Snyder, D. C.; Schauer, J. J.; Sheesley, R. J.; Olson, M. R.; DeMinter, J. Contributions of resuspended soil and road dust to organic carbon in ﬁne particulate matter in the Midwestern US. Atmos. Environ. 2011, 45 (2), 514–518. (24) American Society of Heating Refrigeration and Air Conditioning Engineers (ASHRAE). Museums, galleries, archives, and libraries. In Heating, 10352

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

Environmental Science & Technology Ventilating, and Air-Conditioning Applications; ASHRAE: Atlanta, 2011; Chapter 23. (25) Tetreault, J., Airborne Polluants in Museums, Galleries and Archives: Risk Assessment, Control Strategies and Preservation Management; Canadian Conservation Institute: Ottawa, 2003. (26) Turpin, B. J.; Lim, H. J. Species Contributions to PM2.5 Mass Concentrations: Revisiting Common Assumptions for Estimating Organic Mass. Aerosol Sci. Technol. 2001, 35, 602–610. (27) Russell, L. M. Aerosol Organic-Mass-to-Organic-Carbon Ratio Measurements. Environ. Sci. Technol. 2003, 37 (13), 2982–2987. (28) Sheesley, R. J.; Schauer, J. J.; Bean, E.; Kenski, D. Trends in secondary organic aerosol at a remote site in Michigan’s upper peninsula. Environ. Sci. Technol. 2004, 38 (24), 6491–6500. (29) Jones, N. Indoor/outdoor relationships of particulate matter in domestic homes with roadside, urban and rural locations. Atmos. Environ. 2000, 34 (16), 2603–2612. (30) Sarnat, J. A.; Long, C. M.; Koutrakis, P.; Coull, B. A.; Schwartz, J.; Suh, H. H. Using sulfur as a tracer of outdoor ﬁne particulate matter. Environ. Sci. Technol. 2002, 36 (24), 5305–5314. (31) Na, K.; Cockeriii, D. Organic and elemental carbon concentrations in ﬁne particulate matter in residences, schoolrooms, and outdoor air in Mira Loma, California. Atmos. Environ. 2005, 39 (18), 3325–3333. (32) Thatcher, T. L.; Layton, D. W. Deposition, resuspension, and penetration of particles within a residence. Atmos. Environ. 1995, 29 (13), 1487–1497. (33) Abt, E.; Suh, H. H.; Catalano, P.; Koutrakis, P. Relative contribution of outdoor and indoor particle sources to indoor concentrations. Environ. Sci. Technol. 2000, 34 (17), 3579–3587. (34) Weber, R. J.; Sullivan, A. P.; Peltier, R. E.; Russell, A.; Yan, B.; Zheng, M.; Gouw, J. d.; Warneke, C.; Brock, C.; Holloway, J. S.; Atlas, E. L.; Edgerton, E. A study of secondary organic aerosol formation in the anthropogenic-inﬂuenced southeastern United States. J. Geophys. Res. 2007, 112, 13 PP–13 PP. (35) Schauer, J. J. Evaluation of elemental carbon as a marker for diesel particulate matter. J. Expo Anal Environ. Epidemiol. 2003, 13 (6), 443–453. (36) Marcazzan, G. M.; Vaccaro, S.; Valli, G.; Vecchi, R. Characterisation of PM10 and PM2.5 particulate matter in the ambient air of Milan (Italy). Atmos. Environ. 2001, 35 (27), 4639–4650. (37) Cass, G. R.; McRae, G. J. Source-receptor reconciliation of routine air monitoring data for trace metals: An emission inventory assisted approach. Environ. Sci. Technol. 1983, 17 (3), 129–139. (38) Singh, M.; Jaques, P. A.; Sioutas, C. Size distribution and diurnal characteristics of particle-bound metals in source and receptor sites of the Los Angeles Basin. Atmos. Environ. 2002, 36 (10), 1675–1689. (39) Kim, N.; Fergusson, J. Concentrations and sources of cadmium, copper, lead and zinc in house dust in Christchurch, New Zealand. Sci. The Total Environ. 1993, 138 (1 3), 1–21. (40) Butte, W., Reference values of environmental pollutants in house dust . In Indoor Environment: Airborne Particles and Settled Dust; Morawska, L., Salthammer, T., Eds.; Wiley-VCH: Weinheim: 2003; p 416. (41) Oros, D. R.; Simoneit, B. R. T. Identiﬁcation and emission rates of molecular tracers in coal smoke particulate matter. Fuel 2000, 79 (5), 515–536. (42) Manchester-Neesvig, J. B.; Schauer, J. J.; Cass, G. R., The distribution of particle-phase organic compounds in the atmosphere and their use for source apportionment during the Southern California Children’s Health Study. (Technical Paper). J. Air Waste Manage. Assoc. 2003, pp 1065(15)-1065(15). (43) Schauer, J. J.; Fraser, M. P.; Cass, G. R.; Simoneit, B. R. T. Source reconciliation of atmospheric gas-phase and particle-phase pollutants during a severe photochemical smog episode. Environ. Sci. Technol. 2002, 36 (17), 3806–3814. (44) Simoneit, B. Characterization of organic constituents in aerosols in relation to their origin and transport: A review. Int. J. Environ. Anal. Chem. 1986, 23 (3), 207–237. (45) Rogge, W. F.; Mazurek, M. A.; Hildemann, L. M.; Cass, G. R.; Simoneit, B. R. T. Quantiﬁcation of urban organic aerosols at a

ARTICLE

molecular level: Identiﬁcation, abundance and seasonal variation. Atmos. Environ.,Part A 1993, 27 (8), 1309–1330. (46) Parente, M. E.; Gambaro, A.; Solana, G. Study of sensory properties of emollients used in cosmetics and their correlation with physicochemical properties. J.Cosmet.Sci. 2005, 56 (3), 175–182. (47) Kato, S.; Aoshima, H.; Saitoh, Y.; Miwa, N. Biological safety of LipoFullerene composed of squalane and fullerene-C60 upon mutagenesis, photocytotoxicity, and permeability into the human skin tissue. Basic Clinical Pharmacol.Toxicol. 2009, 104 (6), 483–487. (48) Nicolaides, N. Skin lipids: Their biochemical uniqueness. Science 1974, 186 (4158), 19–26. (49) Naik, D. V.; Weschler, C. J.; Shields, H. C., Indoor and outdoor concentrations of organic compounds associated with airborne particles: Results using a novel solvent system. In Indoor Air Pollution: Radon, Bioaerosols & VOCs; Kay, J. G., Keller, G. E., Miller, J. F., Eds.; Lewis Publishers: MI, 1991; p 67. (50) Mannucci, E.; Zerbi, G., Art and spectroscopy: Looking to paints and parchments. In GNSR 2001: State of Art and Future Development in Raman Spectroscopy and Related Techniques; Messina, G.,Santangelo, S., Eds.; IOS press: Amsterdam, 2002. (51) Robinson, A. L.; Subramanian, R.; Donahue, N. M.; BernardoBricker, A.; Rogge, W. F. Source apportionment of molecular markers and organic aerosol. 3. Food cooking emissions. Environ. Sci. Technol. 2006, 40 (24), 7820–7827. (52) Nazaroﬀ, W. W.; Weschler, C. J. Cleaning products and air fresheners: Exposure to primary and secondary air pollutants. Atmos. Environ. 2004, 38 (18), 2841–2865. (53) Kirk-Othmer, Encyclopedia of Chemical Technology, 4th ed.; Wiley: New York, 1998. (54) Ayoko, G. A.; Uhde, E., Organic compounds adsorbed on particles and settled house dust. In Indoor Environment: Airborne Particles and Settled Dust; Morawska, L., Salthammer, T., Eds.; Wiley-VCH: Weinheim: 2003; p 150. (55) Otake, T.; Yoshinaga, J.; Yanagisawa, Y. Exposure to phthalate esters from indoor environment. J. Exposure Anal. Environ. Epidemiol. 2004, 14, 524–528. (56) Kawamura, K.; Kasukabe, H.; Barrie, L. A. Source and reaction pathways of dicarboxylic acids, ketoacids and dicarbonyls in arctic aerosols: One year of observations. Atmos. Environ. 1996, 30 (10 11), 1709–1722. (57) Simoneit, B. R. T.; Schauer, J. J.; Nolte, C. G.; Oros, D. R.; Elias, V. O.; Fraser, M. P.; Rogge, W. F.; Cass, G. R. Levoglucosan, a tracer for cellulose in biomass burning and atmospheric particles. Atmos. Environ. 1999, 33 (2), 173–182. (58) Sannigrahi, P.; Sullivan, A. P.; Weber, R. J.; Ingall, E. D. Characterization of water-soluble organic carbon in urban atmospheric aerosols using solid-state 13C NMR spectroscopy. Environ. Sci. Technol. 2006, 40 (3), 666–672. (59) Ayoko, G. A.; Uhde, E., Organic compounds adsorbed on particles and settled house dust. In Indoor Environment: Airborne Particles and Settled Dust; Morawska, L., Salthammer, T., Eds.; Wiley-VCH: Weinheim, 2003; p 159.

10353

dx.doi.org/10.1021/es202736a |Environ. Sci. Technol. 2011, 45, 10344–10353

ARTICLE pubs.acs.org/est

Few-Layered Graphene Oxide Nanosheets As Superior Sorbents for Heavy Metal Ion Pollution Management Guixia Zhao, Jiaxing Li, Xuemei Ren, Changlun Chen, and Xiangke Wang* Key Laboratory of Novel Thin Film Solar Cells, Institute of Plasma Physics, Chinese Academy of Sciences, Hefei, 230031, P.R. China

bS Supporting Information ABSTRACT: Graphene has attracted multidisciplinary study because of its unique physicochemical properties. Herein, fewlayered graphene oxide nanosheets were synthesized from graphite using the modiﬁed Hummers method, and were used as sorbents for the removal of Cd(II) and Co(II) ions from large volumes of aqueous solutions. The eﬀects of pH, ionic strength, and humic acid on Cd(II) and Co(II) sorption were investigated. The results indicated that Cd(II) and Co(II) sorption on graphene oxide nanosheets was strongly dependent on pH and weakly dependent on ionic strength. The abundant oxygen-containing functional groups on the surfaces of graphene oxide nanosheets played an important role on Cd(II) and Co(II) sorption. The presence of humic acid reduced Cd(II) and Co(II) sorption on graphene oxide nanosheets at pH < 8. The maximum sorption capacities (Csmax) of Cd(II) and Co(II) on graphene oxide nanosheets at pH 6.0 ( 0.1 and T = 303 K were about 106.3 and 68.2 mg/g, respectively, higher than any currently reported. The thermodynamic parameters calculated from temperature-dependent sorption isotherms suggested that Cd(II) and Co(II) sorptions on graphene oxide nanosheets were endothermic and spontaneous processes. The graphene oxide nanosheets may be suitable materials in heavy metal ion pollution cleanup if they are synthesized in large scale and at low price in near future.

’ INTRODUCTION Heavy metal pollution due to the indiscriminate disposal of wastewater is a worldwide environment concern. Wastewaters from many industries such as metallurgical, mining, chemical manufacturing, and battery manufacturing industries contain many kinds of toxic heavy metal ions.1 Cadmium is among the toxic metals found in some surface and subsurface waters. It is wellknown that chronic cadmium toxicity has been the cause of Japan ItaiItai disease. The harmful eﬀects of Cd also include a number of acute and chronic disorders, such as renal damage, emphysema, hypertension, testicular atrophy, and skeletal malformation in fetus.2,3 Cobalt is a very toxic metal aﬀecting the environment. The increased use of Co(II) in nuclear power plants and in many industries has resulted in Co(II) ﬁndings its way to the environment. In high doses it causes bone defects, diarrhea, low blood pressure, lung irritations and paralysis, and may also cause mutations (genetic changes) in living cells.4 Thereby, it is necessary to eliminate the toxic heavy metal ions from wastewater before it is released into the environment. Traditional techniques for the elimination of heavy metal ions include precipitation, membrane ﬁltration, sorption, and ion exchange, etc.5,6 Among these methods, sorption technique has been used widely because it is simple, economical, and cost-eﬀective. Some sorbents, such as clay minerals, oxides, and carbon materials, have been studied extensively to remove heavy metal ions from aqueous solutions.79 However, these materials suﬀer from either low sorption capacities or eﬃciencies. Nanomaterials have gradually developed important roles to resolve this problem because of their high surface area, enhanced active sites, and abundant functional groups on the surfaces. r 2011 American Chemical Society

So far, a variety of nanomaterials such as carbon nanotubes,10,11 carbon nanotube based material composites,6 and graphene1214 have been studied in the removal of diﬀerent organic and inorganic pollutants from large volumes of aqueous solutions, and the results indicated that these carbon nanomaterials had high sorption capacity. Graphene, a kind of one or several atomic layered graphites, possesses special two-dimensional structure and excellent mechanical, thermal, and electrical properties.15,16 In our previous study,14 sulfonated graphene nanosheets were synthesized and used as sorbents to remove naphthalene and 1-naphthol. The sorption capacities of ∼2.32.4 mmol/g for naphthalene and 1-naphthol were the highest capabilities of today’s nanomaterials. Unlike carbon nanotubes, which require special oxidation processes to introduce hydrophilic groups to improve metal ion sorption, the preparation of graphene oxide nanosheets from graphite using Hummers method introduces many oxygen-containing functional groups such as COOH, CdO, and OH, on the surfaces of graphene oxide nanosheets. These functional groups are essential for the high sorption of heavy metal ions. Graphene oxides, which are considered as the oxidized graphene, contain oxygen-containing functional groups on the surfaces. Considering the oxygen-containing functional groups on the graphene oxide surfaces and high surface area (theoretical Received: September 29, 2011 Accepted: November 9, 2011 Revised: November 6, 2011 Published: November 09, 2011 10454

dx.doi.org/10.1021/es203439v | Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology

ARTICLE

Figure 1. Characterization of few-layered graphene oxide nanosheets: (a) TEM image; (b) AFM image; (c) XRD patterns of graphite and graphene oxide nanosheets; (d) XPS C1s spectrum; (e) Raman spectrum; (f) acidbase titration curve; (g) FT-IR spectrum; and (h) TG-TGA curves.

value of 2620 m2/g), the graphene oxide nanosheets should have high sorption capacity in the preconcentration of heavy metal ions from large volumes of aqueous solutions. However, the application of graphene oxide nanosheets as sorbents in the removal of heavy metal ions from aqueous solution is still scarce,17 especially in the

presence of humic substances, which present widely in the natural environment, and have strong complexation ability with metal ions because of their abundant oxygen-containing functional groups. It is therefore important to study the sorption behaviors of metal ions on graphene oxide nanosheets in the presence of humic substances. 10455

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology

ARTICLE

The objectives of this study were to (1) prepare few-layered graphene oxide nanosheets and apply them as sorbents to remove Cd(II) and Co(II) ions from aqueous solutions; (2) investigate the eﬀects of pH and ionic strength on Cd(II) and Co(II) sorption; (3) study the eﬀect of humic acid (HA), a kind of natural organic material, on Cd(II) and Co(II) sorption; (4) discuss the mechanism of Cd(II) and Co(II) interaction with graphene oxide nanosheets. This study demonstrated the broad applicability of this fascinating material in environmental pollution cleanup.

’ EXPERIMENTAL SECTION Materials. Few-layered graphene oxide nanosheets were prepared by using the modified Hummers method18 from the natural flake graphite (average particle diameter of 20 mm, 99.95% purity, Qingdao Tianhe Graphite Co. Ltd., China) using concentrated H2SO4 and KMnO4 to oxidize the graphite layer. With the aid of ultrasonication, the oxidized graphite layers were exfoliated from each other. Then 30% H2O2 was added in the suspension to eliminate the excess MnO4. The desired products were rinsed with deionized water. Detailed processes are described in the Supporting Information (SI). The prepared fewlayered graphene oxide nanosheets were used as sorbents to remove Cd(II) and Co(II) ions from aqueous solutions in the following experiments. All chemicals used in the experiments were analytical grade. HA was extracted from the soil of Hua-Jia county (Gansu province, China), and was characterized in detail.19 The surface site density was determined to be 6.46 103 mol/g by ﬁtting the potentiometric acidbase titration data with the aid of FITEQL 3.1. Characterization of Graphene Oxide Nanosheets. Graphene oxide nanosheets were characterized by transmission electron microscopy (TEM), atomic force microscopy (AFM), X-ray photoelectron spectroscopy (XPS), powder X-ray diffraction (XRD), Raman spectroscopy, potentiometric acidbase titration, Fourier transformed infrared spectra (FT-IR), and thermogravimetric analysis (TGA). The TEM images were obtained with a JEM-2010. The AFM images were obtained in air using a Digital Instrumental Nanoscope III in tapping mode. The XPS measurements were conducted with an ESCALab 220IXL system. The XRD patterns were measured on a D/max2500 with a Cu Kα source (λ = 1.541 Å). Raman spectra were recorded with a Renishew in Via Raman spectrometer (Renishaw plc, UK). The laser excitation was provided by a regular model laser operated at wavelength of 514 nm. The potentiometric acidbase titration was conducted using a computer-controlled titration system (DL50 Automatic Titrator, Mettler Toledo) in 0.01 M NaClO4 background electrolyte under argon conditions. The data sets of pH versus the net consumption of H+ or OH were used to calculate intrinsic acidity constants in diffuse-layer model with the aid of FITEQL 3.1. FT-IR spectroscopy measurements were mounted by using a Perkin-Elmer 100 spectrometer in KBr pellet at room temperature. TG and DTA curves were measured by using a Shimadzu TGA-50 thermogravimetric analyzer from room temperature to 800 °C with heating rate of 10 °C/min and an air flow rate of 50 mL/min. Sorption Experiments. The batch experiments of Cd(II) and Co(II) sorption on graphene oxide nanosheets were carried out at pH 6.0 ( 0.1 and in 0.01 M NaClO4 solutions in polyethylene test tubes. For most wastewaters in the environment, the pH is

Figure 2. Sorption of Cd(II) (A) and Co(II) (B) on graphene oxide nanosheets as a function of pH in diﬀerent NaClO4 concentrations. C[Cd(II)]initial = 20 mg/L, C[Co(II)]initial = 30 mg/L,T = 303K, m/V = 0.1 g/L. The vertical line on each panel indicates the pH of bulk solution precipitation for the metal ion at the total metal concentration employed.

∼6.0 because of the dissociation of CO2. The stock suspensions of graphene oxide nanosheets, Cd(II) or Co(II) solution, and NaClO4 solution were added in the polyethylene test tubes to achieve the desired concentrations of different components. The desired pH of the suspensions in each tube was adjusted by adding 0.01 mol/L HClO4 or NaOH solution. It was necessary to notice that the graphene oxide nanosheets were equilibrated with HA before the addition of Cd(II) or Co(II) solution in the presence of HA. After the suspensions were shaken for 24 h to achieve sorption equilibrium, the solid phase was separated from the solution using 0.22-μm membrane filters. The results of kinetic sorption suggested that Cd(II) and Co(II) sorption on graphene oxide nanosheets achieved equilibrium in several hours. The concentrations of Cd(II) or Co(II) in the filtrate were determined by atomic absorption spectroscopy. All the experimental data were the average of duplicate determinations, and the relative errors were about 5%. The amounts of Cd(II) or Co(II) ions adsorbed on graphene oxide nanosheets were calculated from the difference between the initial concentration 10456

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology (C0) and the equilibrium one (Ce) (Sorption % = (C0 Ce)/ C0 100%, and Cs = (C0 Ce)/m V, where Cs is the concentration of metal ion adsorbed on graphene oxide nanosheets, V is the volume of the suspension, and m is the mass of graphene oxide nanosheets).

’ RESULTS AND DISCUSSION Characterization of Graphene Oxide Nanosheets. The results of TEM, AFM, XRD, XPS, Raman, FT-IR, TGA, and acidbase titration characterization of the prepared graphene oxide nanosheets are shown in Figure 1. The synthesized graphene oxide nanosheets have lateral dimensions of several micrometers. The TEM image (Figure 1a) shows that fewlayered graphene oxides are formed, although the TEM image does not estimate the layer numbers of the graphene oxide nanosheets exactly. From the AFM image (Figure 1b), the thickness of graphene oxide nanosheets is about 2.98 nm, suggesting that few-layered graphene oxide nanosheets are formed.20,21 The thickness of one layer graphene oxide nanosheet is ∼0.81.0 nm.21 In the XRD patterns (Figure 2c) of graphene oxide nanosheets and graphite, the diﬀraction peak at 2θ = 26.40° (d = 0.34 nm), which corresponds to the normal graphite spacing (002) of graphite plane, disappears in the graphene oxide nanosheets. The broad and relatively weak diﬀraction peak at 2θ = 10.03° (d = 0.87 nm), which corresponds to the typical diﬀraction peak of graphene oxide nanosheets, is attributed to the (002) plane. The c-axis spacing increases from 0.34 to 0.87 nm after the graphite is modiﬁed to graphene oxide nanosheets, which is due to the creation of the abundant oxygen-containing functional groups on the surfaces of graphene oxide nanosheets.22,23 The C1s XPS spectrum (Figure 1d) indicates a considerable degree of oxidation with diﬀerent functional groups, i.e., the nonoxygenated ring C (284.5 eV, 71.4%), the C atom in CO bond (286.2 eV, 18.6%), the carbonyl C (287.8 eV, 9.8%), and the carboxylate carbon (OCdO) (289.0 eV, 0.2%).24 From the XPS analysis, it is clear that the graphene oxide nanosheets are highly oxidized by the oxidant. The speciﬁc peak area noted in Figure 1d shows that the main oxygen-containing groups are CO and CdO, which are expected to form strong surface complexes with metal ions on the solid surfaces. The synthesized graphene oxide nanosheets also have very high dispersion properties in aqueous solutions. The suspension did not form any aggregation even after several months of aging time with a brown yellow color. The high dispersion property of graphene oxide nanosheets in aqueous solution is favorable for the surface oxygenfunctional groups to freely form strong complexes with metal ions. In the Raman spectrum (Figure 1e), the G band at ∼1580 cm1 is associated with the vibration of sp2 carbon atoms in a graphitic 2D hexagonal lattice, and the D band at ∼1350 cm1 is related to the vibrations of sp3 carbon atoms of defects and disorder. The weak and broad 2D peak at ∼2700 cm1 is another indication of disorder as the result of an out-of-plane vibration mode. These strong G, D, and 2D bands are very similar to previous results of graphene oxide characterization.25,26 The pHpzc (point of zero charge) value is calculated to be 3.9 and the surface site density is 2.36 103 mol/g from the acid base titration (Figure 1f). The surface charge is positive at pH < pHpzc, and is negative at pH > pHpzc. The surface site density of graphene oxide nanosheets is about ﬁve times higher than that

ARTICLE

of oxidized multiwalled carbon nanotubes (MWCNTs) (4.2 104 mol/g).27 The oxygen-containing groups on the surfaces of graphene oxide nanosheets are characterized by FT-IR analysis (Figure 1g). Diﬀerent functional groups are found in the FTIR spectrum, i.e., CO group at 1220 cm1 and 1100 cm1, CdO group at 1730 cm1 and CdC at 1620 cm1, which indicates that large amounts of oxygen-containing functional groups exist on graphene oxide nanosheets. The TG curve (Figure 1h) shows signiﬁcant weight loss before 115 °C, which is due to desorption of absorbed water molecules on graphene oxide nanosheets. The graphene oxide nanosheets exhibit two steps of mass loss at 200 and 550 °C, which are attributed to the loss of CO and CO2 from the decomposition of oxygen functional groups and carbon oxidation, respectively.28,29 The graphene oxide nanosheets show the mass loss, starting at the temperature of 200 °C, illustrating a much lower thermal stability compared to the natural ﬂake graphite.28 Effect of pH and Ionic Strength. Figure 2 shows Cd(II) and Co(II) sorption on graphene oxide nanosheets as a function of pH in different NaClO4 solutions. The sorption of Co(II) increases slowly at pH < 6, quickly at pH 69, and then maintains high level at pH > 9. The sorption of Cd(II) increases with increasing pH at pH < 9, and about ∼98% Cd(II) is adsorbed on graphene oxide nanosheets at pH > 9. From the results of acidbase titration (Figure 1f), the pHpzc value of graphene oxide nanosheets is ∼3.9. At pH < pHpzc, the surface charge of graphene oxide nanosheets is positive because of the protonation reaction (SOH + H+ f SOH2+, where S represents the surface of graphene oxide nanosheets, and OH represents the oxygen-containing functional groups). The positive metal ions are difficult to adsorb on the positively charged surface of graphene oxide nanosheets because of the electrostatic repulsion. At pH > pHpzc, the surface charge of graphene oxide nanosheets is negative because of the deprotonation reaction ( SOH f SO + H+). The reaction scheme for hydroxide formation of metal ion (M2+) could be set out as below: M2þ T MðOHÞþ T MðOHÞ02 T MðOHÞ 3 T :::

ð1Þ M(II) could be combined with deprotonated surface sites, and the sorption of Co(II) and Cd(II) on graphene oxide nanosheets could be expressed by the following reaction: 2þ 2þ 2 SO ðsÞ þ MeðaqÞ T ½ð SO Þ2 Me ðsÞ

ð2Þ

The solution pH aﬀected the degree of deprotonation and the speciation of the surface functional groups. With increasing pH, the surface charge is more negative and the electrostatic interactions between the metal ions and graphene oxide nanosheets become stronger, and thereby result in the increase of metal ion sorption. The relative proportion of Cd(II) species is calculated from the stability constants (log β1 = 3.9, log β2 = 7.7, and log β3 = 8.7) and the results demonstrate that Cd(II) presents in the form of Cd2+, Cd(OH)+, Cd(OH)20, and Cd(OH)3 at various pH values (Figure S1). At pH < 8.0, the predominant Cd(II) species is Cd2+ and the removal of Cd(II) is mainly accomplished by sorption reaction. The precipitation curve of Cd(II) calculated from the precipitation constant of Cd(OH)2(s) (Ksp = 2.50 1014) and the initial Cd(II) concentration (i.e., 1.78 104 mol/L) is also shown in Figure 2A. Cd(II) begins to form precipitation at 10457

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology

Figure 3. Eﬀect of humic acid on the sorption of Cd(II) (A) and Co(II) (B) on graphene oxide nanosheets. C[HA]initial = 10 mg/L, C[Co(II)]initial = 30 mg/L, C[Cd(II)]initial = 20 mg/L, I = 0.01 M NaClO4, T = 303K, m/V = 0.1 g/L.

pH ∼9.1 in the absence of graphene oxide nanosheets. However, more than 90% Cd(II) is adsorbed on graphene oxide nanosheets at pH 9.0. Thereby, it is impossible to form precipitation because of the very low concentration of Cd(II) remained in solution. The relative proportion of Co(II) species is calculated from the stability constants (log β1 = 4.3, log β2 = 8.4, and log β3 = 8.4) and the results demonstrate that Co(II) presents in the form of Co2+, Co(OH)+, Co(OH)20, and Co(OH)3 at various pH values (Figure S2). At pH < 8.0, the predominant Co(II) species is Co2+ and the removal of Co(II) is mainly accomplished by sorption reaction. The precipitation curve of Co(II) calculated from the precipitation constant of Co(OH)2(s) (Ksp = 2.50 1016) and the initial Co(II) concentration (i.e., 5.08 104 mol/L) is also shown in Figure 2B. Co(II) begins to form

ARTICLE

Figure 4. Sorption isotherms of Cd(II) (A) and Co(II) (B) on graphene oxide nanosheets at diﬀerent temperatures. m/V = 0.1 g/L, pH = 6.0 ( 0.1, I = 0.01 M NaClO4. The solid lines are Langmuir model simulation, and the dashed lines are Freundlich model simulation.

precipitation at pH ∼8.2 in the absence of graphene oxide nanosheets. Therefore, at pH > 8.2, Co(II) sorption on graphene oxide nanosheets takes place partly though precipitation reaction. The sorptions of Cd(II) and Co(II) are weakly dependent on NaClO4 concentrations. The sorption curves shift to left at lower NaClO4 concentrations as compared to those at higher NaClO4 concentrations. This phenomenon can be attributed to the following: (1) The formed electrical double layer complexes between Cd(II)/Co(II) ions and graphene oxide nanosheets favor metal ion sorption when the concentration of NaClO4 is decreased. The sorption interactions between the functional groups and metal ions are mainly ionic interaction, which is in accordance with ion exchange mechanism. (2) The activity coeﬃcient of metal ions is aﬀected by NaClO4 concentrations, which then limits the 10458

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology

ARTICLE

Table 1. Parameters for Langmuir and Freundlich Models of Cd(II) and Co(II) Sorption on Graphene Oxide Nanosheets Langmuir

Freundlich

Qmax (mg/g)

KL (L/mg)

R2

KF (mg1n 3 Ln/g)

n

R2

T = 303 K

106.3

0.180

0.999

35.83

0.27

0.944

T = 313 K

153.6

0.089

0.997

30.70

0.377

0.997

T = 333 K

167.5

0.104

0.998

36.22

0.367

0.995

T = 303 K

68.2

0.088

0.993

12.11

0.417

0.983

T = 313 K

69.4

0.133

0.999

20.90

0.283

0.968

T = 333 K

79.8

0.133

0.999

22.41

0.298

0.928

experimental conditions Cd(II), pH 6.0

Co(II), pH 6.0

metal ion transfer from solution to solid surfaces. (3) At high ionic strength, the electrostatic repulsion is cut down and the particle aggregation increases, which then reduces the available sites to bind metal ions on graphene oxide surfaces.21,32 The eﬀect of ionic strength on Cd(II) and Co(II) sorption is more obvious at low pH as compared to that of ionic strength at high pH values. On the basis of the above theory, one can deduce that Cd(II) and Co(II) sorption on graphene oxide nanosheets is mainly attributed to outer-sphere surface complexation or ion exchange at low pH, and is attributed to inner-sphere surface complexation at high pH values.6,33 Effect of Humic Acid. Effect of HA on Cd(II) and Co(II) sorption on graphene oxide nanosheets is shown in Figure 3. The presence of HA reduces Cd(II) sorption on graphene oxide nanosheets. For Co(II) sorption, the presence of HA reduces Co(II) sorption at pH < 8, and at pH > 8 no obvious difference is found in the presence and absence of HA. For most materials, such as carbon nanotubes, clay minerals, and oxides, the presence of HA enhances metal ion sorption on solid phase at low pH values, and reduces metal ion sorption at high pH values.6,19,34,35 The increase of metal ion sorption at low pH is generally attributed to the strong complexation of metal ion with surface adsorbed HA on solid particles, whereas the decrease of metal ion sorption is interpreted by the formation of soluble MHA complexes in aqueous solution. Herein, the presence of HA decreases Cd(II) and Co(II) sorption on graphene oxide nanosheets at pH < 8, which may be attributed to the strong surface complexation and high surface site density of graphene oxide nanosheets. The surface site density of graphene oxide nanosheets is calculated to be 2.36 103 mol/g from the acidbase titration, whereas the surface site density of HA is 6.46 103 mol/g.19 The high surface site density of graphene oxide nanosheets assures the high sorption of Cd(II) and Co(II) ions on graphene oxide nanosheets. HA can be bound to graphene oxide nanosheets through strong ππ interactions. HA can interact with graphene oxide nanosheets in aquatic systems, thereby greatly changing their properties in such systems.6,36 Although the surface site density of graphene oxide nanosheets is lower than that of HA, the strong interaction of HA with graphene oxide nanosheets occupies parts of surface sites on graphene oxide nanosheets and also reduces the available binding sites of HA, and thereby results in the decrease of Cd(II) and Co(II) sorption on graphene oxide nanosheets. It is necessary to note that the effect of HA on Co(II) sorption at pH > 8 is not obvious, this is maybe attributed to the formation of Co(OH)2 precipitation at pH > 8.2.

Sorption Isotherms and Thermodynamic Data. Figure 4 shows Cd(II) and Co(II) sorption isotherms on graphene oxide nanosheets at three different temperatures. The experimental data are simulated with the Langmuir (Qs = Qs max 3 KL 3 Ce/1 + KL 3 Ce) and Freundlich (Qs = KF 3 Ce1/n) models, respectively (where Ce is the equilibrium concentration of metal ions in aqueous solution (mg/L), Qs is the amount of metal ions adsorbed on graphene oxide (mg/g), Qsmax is the maximum amount of metal ions adsorbed per unit weight of graphene oxide to form a complete monolayer coverage on the surface, KL represents enthalpy of sorption and should vary with temperature, and KF and n are the Freundlich constants related to the sorption capacity and sorption intensity, respectively. The relative parameters calculated from the two models are listed in Table 1. The sorption isotherms are fitted better by the Langmuir model than by the Freundlich model, suggesting that Cd(II) and Co(II) sorption on graphene oxide nanosheets are monolayer coverage. The Qsmax values of Cd(II) and Co(II) sorption on graphene oxide nanosheets are 68.2 and 106.3 mg/g, respectively. Comparing to Qsmax values of Cd(II) and Co(II) sorption on other sorbents, such as granular activated carbon (10.1 mg/g Cd(II) at pH 5 and T = 298 K),2 activated carbon fiber (13.6 mg/g Cd(II) at pH 5 and T = 298 K),2 activated carbon cloth (23.5 mg/g Cd(II) at pH 5 and T = 298 K),2 granular activated carbon oxide (30.8 mg/g Cd(II) at pH 5 and T = 298K),2 activated carbon fiber oxide (50.0 mg/g Cd(II) at pH 5 and T = 298 K),2 activated carbon cloth oxide (50.0 mg/g Cd(II) at pH 5 and T = 298 K),2 filtrasorb 400 (9.5 mg/g Cd(II) at pH 6 and T = 298 K),37 carbon aerogel (15.5 mg/g Cd(II) at pH 6 and T = 333 K),3 Indonesian peat (14.0 mg/g Cd(II) at pH 6 and T = 296 K),38 zeolite (14.4 mg/g Co(II) at pH 6 and T = 298 K),37 sepiolite (4.7 mg/g Co(II) at pH 7.8 and T = 293 K),38 Al-pillared bentonite (38.6 mg/g Co(II) at pH 6 and T = 303 K),39 lemon peel (25.6 mg/g Co(II) at pH 6 and T = 298 K),40 activated carbon (1.2 mg/g Co(II) at pH 6 and T = 303 K),41 one can see that the graphene oxide nanosheets have the highest sorption capacity of today’s materials. What’s more important, the abundant oxygen-containing functional groups on the surfaces of graphene oxide nanosheets make the adjacent oxygen atoms available to bind metal ions. From the XRD analysis, the c-axis spacing of graphene oxide nanosheets is ∼0.87 nm, which is large enough for the metal ions to enter into the interlayer space of graphene oxide nanosheets. Also, the graphene oxide nanosheets are the Lewis base and the metal ions are the Lewis acid.42 The delocalized π electron systems of graphene layer can act as Lewis base to form electron donor acceptor complexes with metal ions. Strong surface complexation 10459

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology

ARTICLE

K0, can be calculated by plotting lnKd versus Ce (Figures S3 and S4) and extrapolating Ce to zero. The standard enthalpy change (ΔH0) and the standard entropy change (ΔS0) are calculated from the following equation: ln K 0 ¼

Figure 5. Linear plot of lnK0 vs 1/T for the sorption of Cd(II) (A) and Co(II) (B) on graphene oxide nanosheets at 303, 313, and 333 K. m/V = 0.1 g/L, pH = 6.0, I = 0.01 M NaClO4.

between the graphene oxide nanosheets and metal ions occurs through the Lewis acidbase interaction, which also contributes to metal ion sorption on graphene oxide nanosheets. The sorption isotherm is the highest at T = 333 K and is the lowest at T = 303 K, indicating that Cd(II) and Co(II) sorption on graphene oxide nanosheets are promoted at higher temperature. The thermodynamic parameters (ΔH0, ΔS0, and ΔG0) for Cd(II) and Co(II) sorption on graphene oxide nanosheets can be calculated from the temperature dependent sorption isotherms. The standard free energy change (ΔG0) can be calculated from the following equation: ΔG0 ¼ RTln K 0

ð3Þ

where R is the universal gas constant (8.314 J 3 mol1 3 K1), T is the temperature in Kelvin. The sorption equilibrium constant,

ΔS0 ΔH 0 R RT

ð4Þ

Linear plots of lnK0 vs 1/T for Cd(II) and Co(II) sorption on graphene oxide nanosheets are shown in Figure 5. The thermodynamic parameters are calculated from the plot of lnK0 vs 1/T using eqs 3 and 4. The positive value of ΔH0 (10.49 kJ 3 mol1) for Co(II) sorption indicates that Co(II) sorption on graphene oxide nanosheets is an endothermic process. The interpretation to the endothermicity of ΔH0 is that Co(II) ions are well solvated in aqueous solution. For Co(II) ions to adsorb on graphene oxide nanosheets, they have to have their hydration sheath denuded to some extent, and this dehydration process needs energy. The energy of dehydration exceeds the exothermicity of Co(II) ions to attach to graphene oxide nanosheets. The removal of water molecules from Co(II) ions is essentially an endothermic process, and the endothermicity of the desolvation process exceeds that of the enthalpy of Co(II) sorption.1 The positive ΔS0 value (103.1 J 3 mol1 3 K1) of Co(II) also indicates that the sorption process is spontaneous with high aﬃnity. The negative ΔG0 values (20.73 kJ 3 mol1 at 303 K, 21.83 kJ 3 mol1 at 313 K, and 23.83 kJ 3 mol1 at 333 K) of Co(II) also indicate the spontaneous process of Co(II) sorption under the conditions applied. The decrease of ΔG0 with increasing temperature indicates more eﬃcient sorption at higher temperatures. At higher temperature, Co(II) ions are readily dehydrated, and therefore the sorption becomes more favorable. The thermodynamic parameters reﬂect the aﬃnity of graphene oxide nanosheets toward Co(II) ions in aqueous solutions and may suggest some structural changes in the sorbents.1,43 The thermodynamic parameters of Cd(II) sorption on graphene oxide nanosheets are ΔH0 = 7.39 kJ 3 mol1, ΔS0 = 100.1 J 3 mol1 3 K1, and ΔG0 = 2.97 kJ 3 mol1 at 303 K, 23.92 kJ 3 mol1 at 313 K and 25.97 kJ 3 mol1 at 333 K, respectively. The thermodynamic data of Cd(II) sorption on graphene oxide nanosheets also suggest that Cd(II) sorption is a spontaneous and endothermic process. The standard free energy change (ΔG0) values of Cd(II) sorption are more negative as compared to those of Co(II) sorption, suggesting that Cd(II) sorption on graphene oxide nanosheets is more spontaneous and easy than Co(II) sorption. The results are in good agreement with the pH-dependent sorption of Cd(II) and Co(II) as shown in Figure 2. At pH 6.0, about 50% Cd(II) is adsorbed on graphene oxide nanosheets, whereas only about 25% Co(II) is adsorbed from aqueous solution to graphene oxide nanosheets. From the literature query,2,3,3739 the prepared graphene oxide nanosheets have much higher sorption capacity for Cd(II) and Co(II) removal from aqueous solutions than any of today’s materials. This paper highlights the application of few-layered graphene oxide nanosheets as sorbents in environmental pollution management. The graphene oxide nanosheets may be suitable materials for ex situ environmental remediation of heavy metal ions. The solution pH and humic acid can aﬀect the removal of Cd(II) and Co(II) ions from aqueous solution to graphene oxide nanosheets. Although the graphene oxide nanosheets are relatively more expensive than other natural materials and other carbon materials such as active carbon, carbon nanotubes, the graphene 10460

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology oxide nanosheets will be synthesized in large scale and at low price in the near future with the development of technology. The graphene oxide nanosheets will be very suitable materials for the preconcentration and solidiﬁcation of heavy metal ions from large volumes of aqueous solutions in environmental pollution cleanup in the near future.

’ ASSOCIATED CONTENT

bS

Supporting Information. Preparation of graphene oxide nanosheets. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Tel: +86-551-5592788; fax: +86-551-5591310; e-mail: xkwang@ ipp.ac.cn.

’ ACKNOWLEDGMENT Financial support from 973 projects of MOST (2011CB933700), NSFC (20971126, 21071147, 21071107, 91126020), and open foundation of State Key Lab of Pollution Control and Resource Reuse is acknowledged. Thanks are also extended to Prof. D.A. Dzombak and the anonymous reviewers for their helpful comments to improve the quality of our manuscript. ’ REFERENCES (1) Chen, C. L.; Wang, X. K. Adsorption of Ni(II) from aqueous solution using oxidized multi-walled carbon nanotubes. Ind. Eng. Chem. Res. 2006, 45, 9144–9149. lvarez-Merino, M. A.; Lopez-Ramon, (2) Moreno-Castilla, C.; A M. V.; Rivera-Utrilla, J. Cadmium ion adsorption on diﬀerent carbon adsorbents from aqueous solutions. Eﬀect of surface chemistry, pore texture, ionic strength and dissolved natural organic matter. Langmuir 2004, 20, 8142–8148. (3) Goel, J.; Kadirvelu, K.; Rajagopal, C.; Garg, V. K. Cadmium(II) uptake from aqueous solution by adsorption onto carbon aerogel using a response surface methodological approach. Ind. Eng. Chem. Res. 2006, 45, 6531–6537. (4) Rengaraj, S.; Moon, S. H. Kinetics of adsorption of Co(II) removal from water and wastewater by ion exchange resins. Water Res. 2002, 36, 1783–1793. (5) Matlock, M. M.; Howerton, B. S.; Atwood, D. A. Chemical precipitation of lead from lead battery recycling plant wastewater. Ind. Eng. Chem. Res. 2002, 41, 1579–1582. (6) Yang, S. B.; Hu, J.; Chen, C. L.; Shao, D. D.; Wang, X. K. Mutual eﬀect of Pb(II) and humic acid adsorption onto multiwalled carbon nanotubes/poly(acrylamide) composites from aqueous solution. Environ. Sci. Technol. 2011, 45, 3621–3627. (7) Fonseca, B.; Figueiredo, H.; Rodrigues, J.; Queiroz, A.; Tavares, T. Mobility of Cr, Pb, Cd, Cu and Zn in a loamy sand soil: A comparative study. Geoderma 2011, 164, 232–237. (8) Tan, X. L.; Fang, M.; Chen, C. L.; Yu, S. M.; Wang, X. K. Counterion eﬀects of Ni2+ and sodium dodecylbenzene sulfonate adsorption to multiwalled carbon nanotubes in aqueous solution. Carbon 2008, 46, 1741–1750. (9) Tan, X. L.; Fan, Q. H.; Wang, X. K.; Grambow, B. Eu(III) sorption to TiO2 (anatase and rutile): Batch, XPS, and EXAFS study. Environ. Sci. Technol. 2009, 43, 3115–3121. (10) Long, R. Q.; Yang, R. T. Carbon nanotubes as superior sorbent for dioxin removal. J. Am. Chem. Soc. 2001, 123, 2058–2059.

ARTICLE

(11) Wang, X. L.; Lu, J. L.; Xing, B. S. Sorption of organic contaminants by carbon nanotubes: Inﬂuence of adsorbed organic matter. Environ. Sci. Technol. 2008, 42, 3207–3212. (12) Chandra, V.; Park, J.; Chun, Y.; Lee, J. W.; Hwang, I. C.; Kim, K. S. Water-dispersible magnetite-reduced graphene oxide composites for arsenic removal. ACS Nano 2010, 4, 3979–3986. (13) Chandra, V.; Kim, K. S. Highly selective adsorption of Hg2+ by a polypyrrole-reduced graphene oxide composite. Chem. Commun. 2011, 47, 3942–3944. (14) Zhao, G. X.; Jiang, L.; He, Y. D.; Li, J. X.; Dong, H. L.; Wang, X. K.; Hu, W. P. Sulfonated graphene for persistent aromatic pollutant management. Adv. Mater. 2011, 23, 3959–3963. (15) Liang, Y. Y.; Wu, D. Q.; Feng, X. L.; M€ullen, K. Dispersion of graphene sheets in organic solvent supported by ionic interactions. Adv. Mater. 2009, 21, 1679–1683. (16) Su, Q.; Pang, S.; Alijani, V.; Li, C.; Feng, X.; M€ullen, K. Composites of graphene with large aromatic molecules. Adv. Mater. 2009, 21, 3191–3195. (17) Zhao, G. X.; Ren, X. M.; Gao, X.; Tan, X. L.; Li, J. X.; Chen, C. L.; Huang, Y. Y.; Wang, X. K. Removal of Pb(II) ions from aqueous solutions on few-layered graphene oxide nanosheets. Dalton Trans. 2011, 40, 10945–40952. (18) Hirata, M.; Gotou, T.; Horiuchi, S.; Fujiwara, M.; Ohba, M. Thin-ﬁlm particles of graphite oxide 1: High-yield synthesis and ﬂexibility of the particles. Carbon 2004, 42, 2929–2937. (19) Tan, X. L.; Wang, X. K.; Geckeis, H.; Rabung, Th. Sorption of Eu(III) on humic acid or fulvic acid bound to alumina studied by SEMEDS, XPS, TRLFS and batch techniques. Environ. Sci. Technol. 2008, 42, 6532–6537. (20) Zhang, L.; Liang, J. J.; Huang, Y.; Ma, Y. F.; Wang, Y.; Chen, Y. S. Size-controlled synthesis of graphene oxide sheets on a large scale using chemical exfoliation. Carbon 2009, 47, 3365–3368. (21) Zhang, L.; Li, X.; Huang, Y.; Ma, Y. F.; Wan, X. J.; Chen, Y. S. Controlled synthesis of few-layered graphene sheets on a large scale using chemical exfoliation. Carbon 2010, 48, 2367–2371. (22) Fan, X. B.; Peng, W. C.; Li, Y.; Li, X. Y.; Wang, S. L.; Zhang, G. L.; Zhang, F. B. Deoxygenation of exfoliated graphite oxide under alkaline conditions: A green route to graphene preparation. Adv. Mater. 2008, 20, 4490–4493. (23) McAllister, M. J.; Li, J. L.; Adamson, D. H.; Schniepp, H. C.; Abdala, A. A.; Liu, J.; Herrera-Alonso, M.; Milius, D. L.; Car, R.; Prud’homme, R. K.; Aksay, I. A. Single sheet functionalized graphene by oxidation and thermal expansion of graphite. Chem. Mater. 2007, 19, 4396–4404. (24) Waltman, R. J.; Pacansky, J.; Bates, C. W. X-ray photoelectron spectroscopic studies on organic photoconductors evaluation of atomic charges on chlorodiane blue and p-(diethylamino)benzaldehyde diphenylhydrazone. Chem. Mater. 1993, 5, 1799–1804. (25) Valles, C.; Drummond, C.; Saadaoui, H.; Furtado, C. A.; He, M.; Roubeau, O.; Ortolani, L.; Monthioux, M.; Penicaud, A. Solutions of negatively charged graphene sheets and ribbons. J. Am. Chem. Soc. 2008, 130, 15802–15804. (26) Eda, G.; Chhowalla, M. Chemically derived graphene oxide: towards large-area thin-ﬁlm electronics and optoelectronics. Adv. Mater. 2010, 22, 2392–2415. (27) Chen, C. L.; Hu, J.; Xu, D.; Tan, X. L.; Meng, Y. D.; Wang, X. K. Surface complexation modeling of Sr(II) and Eu(III) adsorption onto oxidized multiwall carbon nanotubes. J. Colloid Interface Sci. 2008, 323, 33–41. (28) Wang, G. X.; Yang, J.; Park, J.; Gou, X. L.; Wang, B.; Liu, H.; Yao, J. Facile synthesis and characterization of graphene nanosheets. J. Phys. Chem. C 2008, 112, 8192–8195. (29) Park, S.; An, J. H.; Jung, I. W.; Piner, R. D.; An, S. J.; Li, X. S.; Velamakanni, A.; Ruoﬀ, R. S. Colloidal suspensions of highly reduced graphene oxide in a wide variety of organic solvents. Nano Lett. 2009, 9, 1593–1597. (30) Kragten, J. Atlas of Metal-Ligand Equilibria in Aqueous Solution; Ellis Horwood: Chichester, 1978. 10461

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

Environmental Science & Technology

ARTICLE

(31) Reddad, Z.; Gerente, C.; Andres, Y.; Cloirec, L. P. Adsorption of several metal ions onto a low-cost biosorbent: Kinetic and equilibrium studies. Environ. Sci. Technol. 2002, 36, 2067–2073. (32) Mercer, K. L.; Tobiason, J. E. Removal of arsenic from high ionic strength solutions: Eﬀects of ionic strength, pH, and preformed versus in situ formed HFO. Environ. Sci. Technol. 2008, 42, 3797–3802. (33) Bradbury, M. H.; Baeyens, B. Sorption of Eu on Na- and Ca-montmorillonites: Experimental investigations and modeling with cation exchange and surface complexation. Geochim. Cosmochim. Acta 2002, 66, 2325–2334. (34) Takahashi, Y.; Minai, Y.; Ambe, H.; Makide, Y.; Ambe, F. Comparison of adsorption behavior of multiple inorganic ions on kaolinite and silica in the presence of humic acid using the multitracer technique. Geochim. Cosmochim. Acta 1999, 63, 815–836. (35) Takahashi, Y.; Kimura, T.; Minai, Y. Direct observation of Cm(III)-fulvate species on fulvic acid-montmorillonite hybrid by laser induced ﬂuorescence spectroscopy. Geochim. Cosmochim. Acta 2002, 66, 1–12. (36) Hyung, H.; Fortnter, J. D.; Hughes, J. B.; Kim, J. H. Natural organic matter stabilizes carbon nanotubes in the aqueous phase. Environ. Sci. Technol. 2007, 41, 179–184. (37) Sanchez-Polo, M.; Rivera-Utilla, J. Adsorbent-adsorbate interactions in the adsorption of Cd(II) and Hg(II) on ozonized activated carbons. Environ. Sci. Technol. 2002, 36, 3850–3854. (38) Balasubramanian, R.; Perumal, S. V.; Vijayaraghavan, K. Equilibrium isotherm studies for the multicomponent adsorption of lead, zinc, and cadmium onto Indonesian peat. Ind. Eng. Chem. Res. 2009, 48, 2093–2099. (39) Marlohar, D. M.; Noeline, B. F.; Anirudhan, T. S. Adsorption performance of Al-pillared bentonite clay for the removal of cobalt(II) from aqueous phase. Appl. Clay Sci. 2006, 31, 194–206. (40) Bhatnagar, A.; Minocha, A. K.; Sillanp€a€a, M. Adsorptive removal of cobalt from aqueous solution by utilizing lemon peel as biosorbent. Biochem. Eng. J. 2010, 48, 181–186. (41) Subaymon, A. H.; Abid, B. A.; Al-Najar, J. A. Removal of lead copper chromium and cobalt ions onto granular activated carbon in batch and ﬁxed-bed adsorbers. Chem. Eng. J. 2009, 155, 647–653. (42) Huang, Z. H.; Zheng, X.; Lv, W.; Wang, M.; Yang, Q. H.; Kang, F. Adsorption of lead(II) ions from aqueous solution on low-temperature exfoliated graphene nanosheets. Langmuir 2011, 27, 7558–7562. (43) Genc-Fuhrman, H.; Tjell, J. C.; McConchie, D. Adsorption of arsenic from water using activated neutralized red mud. Environ. Sci. Technol. 2004, 38, 2428–2434.

10462

dx.doi.org/10.1021/es203439v |Environ. Sci. Technol. 2011, 45, 10454–10462

ARTICLE pubs.acs.org/est

Evaluation of Spatial Variability of Soil Arsenic Adjacent to a Disused Cattle-Dip Site, Using Model-Based Geostatistics Nabeel K. Niazi, Thomas F. A. Bishop, and Balwant Singh* Faculty of Agriculture Food and Natural Resources, The University of Sydney, Sydney, NSW 2006, Australia

bS Supporting Information ABSTRACT: This study investigated the spatial variability of total and phosphate-extractable arsenic (As) concentrations in soil adjacent to a cattledip site, employing a linear mixed model-based geostatistical approach. The soil samples in the study area (n = 102 in 8.1 m2) were taken at the nodes of a 0.30 0.35 m grid. The results showed that total As concentration (00.2 m depth) and phosphate-extractable As concentration (at depths of 00.2, 0.20.4, and 0.40.6 m) in soil adjacent to the dip varied greatly. Both total and phosphateextractable soil As concentrations signiﬁcantly (p = 0.0040.048) increased toward the cattle-dip. Using the linear mixed model, we suggest that 5 samples are suﬃcient to assess a dip site for soil (As) contamination (95% conﬁdence interval of (475.9 mg kg1 ), but 15 samples (95% conﬁdence interval of (212.3 mg kg1) is desirable baseline when the ultimate goal is to evaluate the eﬀects of phytoremediation. Such guidelines on sampling requirements are crucial for the assessment of As contamination levels at other cattle-dip sites, and to determine the eﬀect of phytoremediation on soil As.

’ INTRODUCTION Arsenic (As) is a highly toxic and carcinogenic element. Estimation of the spatial distribution of As is imperative to precisely quantify its concentration at contaminated sites.1,2 Such information can also assist in designing suitable remediation and management strategies of As-contaminated sites as it can delineate the level and spread of As contamination.3 Arsenical pesticides were used to control ticks in livestock from the early 1900s to 1955.2,4 This resulted in As contamination of the soil surrounding (now disused) cattle-dip sites in many countries including, Australia, the United States, Republic of South Africa, and New Zealand.2,5,6 The distribution of total As content in soils around disused cattle-dip sites in New South Wales (NSW), Australia, has been found to be high and extremely variable.79 Recently, Niazi et al.9 investigated the distribution of total As concentration in the topsoil (020 cm) adjacent to a cattle-dip site at the Wollongbar Research Institute in northern NSW, Australia. They reported that soil As concentrations varied over submeter scales, with total As concentration ranging from 313 to 1902 mg kg1, which is well above the ecological investigation level (EIL, 20 mg kg1) of total As in the soil in Australia.10 The authors observed higher As concentrations in soils closer to the cattle-dip (mean = 909 ( 351 mg kg1, 5 m away from the dip), which decreased with distance away from the dip bath (mean = 754 ( 284 mg kg1, 8 m away from the dip). This variability was attributed to the dipping process, pumping-out of the dipping ﬂuid from the cattle-dip, and the disposal of As-containing sludge material from the dip bath.7,9 r 2011 American Chemical Society

Hossain et al.11 found that the total As content was laterally heterogeneous in the surface soils (015 cm) and irrigationchannel water around a shallow tube-well (STW) command area (8 ha) at Paranpur in Faridpur sadar upazila (subdistrict), Bangladesh. They reported that the total soil As concentration in the topsoil varied between 11.41 and 61.04 mg kg1 and showed a positive and strong correlation (0.641, p < 0.01) with the amorphous (oxalate-extractable) iron in soil. They also showed that the total As concentration tended to decrease with distance from the STW. Dittmar et al.12 and Roberts et al.13 investigated the spatial variability of As in 18 paddy soils and irrigation water in Sreenagar upazila (Munshiganj, Bangladesh). They reported that As deposition in the soils from the As-rich irrigation water was spatially variable (1135 mg kg1, at 0 10 cm depth), with decreasing As concentrations away from the water inlet. This trend was described based on the formation and sedimentation of As-bearing hydrous ferric oxide aggregates, and sorption of As with the mineral surfaces across the ﬁeld.12 A cattle-dip is a point-source polluter, so greater As accumulation can be expected near the dip, conferring a spatial trend on observations. Although some studies have previously used geostatistics to describe the distribution of As in soil,9,11,12,14 none Received: May 19, 2011 Accepted: November 21, 2011 Revised: October 26, 2011 Published: November 21, 2011 10463

dx.doi.org/10.1021/es201726c | Environ. Sci. Technol. 2011, 45, 10463–10470

Environmental Science & Technology

ARTICLE

Figure 1. Schematic representation of the sampling design. The ﬁlled circles show the locations of the n = 102 soil cores.

have considered the issue of spatial trend (i.e., a coarse-scale deterministic eﬀect, not easily dealt with by conventional geostatistics).1,3,11 Model-based geostatistics, in the form of a linear mixed model, can be used to characterize the spatial pattern of the trend, as well as the spatial pattern of the ﬁne-scale ﬂuctuations in soil As content.3,1517 The parameters of the linear mixed model can then be used to guide baseline sampling recommendations. Such information can be useful to estimate baseline As concentrations around a contaminated site; the site can then be revisited to estimate the change in As concentration after remediation eﬀorts, e.g., phytoremediation using As-hyperaccumulating ferns.9 To our knowledge, model-based geostatistics has not been used to measure the spatial variation of a contaminant over a relatively small scale, such as As in soil in the vicinity of cattle-dip sites. Therefore the aims of this study were to (1) perform a ﬁneresolution survey of soil As in order to quantify its spatial variability using a linear mixed model, and (2) use the information on spatial variability to recommend an appropriate number of samples for estimating mean As concentration at a site.

’ MATERIALS AND METHODOLOGY Study Area and Data Set. The field site is a disused cattle-dip at Wollongbar in northern NSW, Australia (28° 490 12.000 S, 153° 230 49.200 E), where As-based pesticides were first used in 1953 in

the dipping solution to control cattle ticks.18 The area adjacent to the cattle-dip was selected for soil sampling, based on general information on the distribution of As around the cattle-dips from a previous study by Kimber et al.,7 who found that As concentrations declined below the EIL value (20 mg kg1) at a distance of 15 m away from the dip. The study area was immediately at the down-slope side of the dip where As contamination occurred due to the existence of entry and exit points on this side of the cattledip. According to the Australian Soil Classification,19 the soil at the site is Red Ferrosol. The soil has acidic pH of 4.82, and high concentrations of clay (44% on average) and free Fe (16% on average) (see Supporting Information (SI)). An intensive survey was performed to collect the soil samples in the study area (8.1 m2) adjacent to the cattle-dip. Samples (n = 102) were taken in a systematic design on a 0.30 m 0.35 m grid using a hand-driven stainless steel corer (diameter = 2.5 cm) (Figure 1). The soil cores were taken to a depth of 0.6 m and separated into three sections to obtain samples for the 00.2, 0.20.4, and 0.40.6 m depth intervals. Two soil cores (12 cm apart) were collected at each sampling point and respective sections for each depth interval were composited into a single sample (Figure 1). The surface samples (00.2 m) were dried at 105 °C overnight prior to digestion in a mixture of acids using the method described by Huang and Fujii,20 and analyzed for total As concentration (mg As kg1 soil dry weight) using an inductively 10464

dx.doi.org/10.1021/es201726c |Environ. Sci. Technol. 2011, 45, 10463–10470

Environmental Science & Technology

ARTICLE

coupled plasma atomic emission spectrometer (ICP-AES, Varian Vista AX CCD). Potassium dihydrogen phosphate solution (0.5 M KH2PO4, 1:25 soil/solution ratio, 4 h shaking time at room temperature) was used to extract the speciﬁcally sorbed As pool for all three depths.9 This fraction is considered to measure the bioavailable fraction of As.2 The concentration of phosphateextractable As (mg As kg1 soil dry weight) in the extracts was measured using a hydride-generation atomic absorption spectrometer (HG-AAS, Varian Spectraa 220Z). The relative standard deviation (RSD) was <2% for the ICP-AES analysis and <3% for HG-AAS analysis. The QA/QC details are given in the SI. Geostatistical Analysis. All statistical analyses were performed using R version 2.10.1, 21 and in particular the geoR package was used for the geostatistical analysis.22 For this study, a linear mixed model (commonly known as model-based geostatistics)15,17 was used to characterize the spatial variability of total As (at 00.2 m depth) and phosphate-extractable As (at three depths) in soil. A linear mixed model can be written as follows:23 z ¼ Xβ þ η

Mean error ðMEÞ ¼ 1=n

ð1Þ

where z is a length-n vector of observed As concentrations, and is regarded as the sum of a deterministic component (“fixedeffects” or “trend”) and a random component (“random effects”). The n p design matrix X contains values of the p explanatory variables, β is a length-p vector of parameters that (linearly) relates X to z, and η is a length-n vector of spatially correlated random effects, distributed as η ∼ N(0,V) and assumed to be second-order stationary (i.e., the mean and variance of the random effects are constant across the study site) (eq 1). The term V is a n n covariance matrix whose elements are determined by a variogram function, γ(h):23,24 V i, j ¼ σ2 γðhÞ

ð2Þ

where σ2 is the variance of the random effects and h is the spatial distance (m) between the ith and jth observations. In regard to the fixed effects, as we expected soil As concentrations to decrease with distance from the cattle-dip, we specified that the design matrix X comprise linear combinations of the spatial coordinates of the samples, i.e., the fixed effects would describe a trend surface. In regard to the random effects, we considered two forms of variogram. The first was a spherical function:25 8 ( 3 ) > 3h h : C0 þ C1 for h > j ð3Þ The second was an exponential function: γðhÞ ¼ C0 þ C1 f1 exp

variogram function was decided based on the Akaike Information Criterion (AIC) value, where the smaller the value the better the model.26 Once the best model was decided, the signiﬁcance of the ﬁxed eﬀects was tested with Wald statistics.17 Maps of As concentration were derived by using the observed data and the parameters of the linear mixed model in conjunction with the empirical best linear unbiased predictor (E-BLUP), an analogue of universal kriging.15 Linear mixed models require the data to be normally distributed; in cases where the residuals were not normally distributed the data were log-transformed. Prediction quality of the linear mixed model was assessed through leave-one-out-cross-validation (LOOCV), where each observation was removed from the data set in turn, and the E-BLUP was applied at each step to predict the value of the withheld observation as if it were never sampled. The following summary statistics were generated from LOOCV:

ðh=jÞ

g

ð4Þ

where C0 is the “nugget” variance (i.e., ﬂuctuations due to measurement error and variability that occurs at scales less than the minimum sampling interval), C1 is the autocorrelated variance, and j is the distance parameter. Following the assumption of second-order stationary (as mentioned above), σ2 = C0 + C1. In the case of the spherical function, j is the distance after which pairs of observations are not spatially correlated (i.e., the “range” of the variogram); for the exponential function the range is 3j, as it approaches a maximum semivariance asymptotically. The best

∑ ½^z ðsi Þ zðsi Þ

Mean square error ðMSEÞ ¼ 1=n

∑

h i2 ^z ðsi Þ zðsi Þ

Mean square deviation ratio ðMSDRÞ n o2 2 ^ ¼ 1=n z ðsi Þ zðsi Þ =^ σ ðsi Þ

∑

ð5Þ ð6Þ

ð7Þ

where, for location si, ^z (si) is the E-BLUP prediction obtained through LOOCV, z(si) is the observed value, and σ^2(si) is the E-BLUP estimation variance. The ME should be close to zero (0) for an unbiased estimate; the MSE is a measure of the accuracy of the predictions and it should be as small as possible. The MSDR measures how well the kriging variance represents the squared prediction error;1 it should be close to 1.0. Sample Size Estimation. We propose that simple random sampling can be used to determine the mean As concentration at a cattle-dip site. In this case, the parameters of the random effects of the linear mixed model can be used to determine the uncertainty about the mean for a given number of samples (eq 8):2729 rﬃﬃﬃﬃ 1 SEM ¼ γ ð8Þ ̅ ns where ns is the proposed number of samples, and γ ̅ is the dispersion variance for the study area.25 The latter was computed by Monte Carlo simulation, where we randomly selected a pair of locations in the study area, and calculated the semivariance based on the fitted semivariogram model.27 This was repeated 10 000 times and from this we calculated γ ̅ .

’ RESULTS AND DISCUSSION A slight positive skewness was observed for total As in the surface soil (00.2 m) and for phosphate-extractable As in the ﬁrst two depths (skewness = 0.580.81; SI Figure S1ac). For phosphate-extractable As at 0.40.6 m depth, relatively stronger positive skewness (2.60) occurred (SI Figure S1d) due to the presence of two outlying high values. The total As in the top layer (mean = 826 mg kg1) was well above the EIL limit of 20 mg kg1 for all samples (SI Figure S1a). Phosphate-extractable As (at 0 0.2 m depth) constituted e13.9% of the total As in soil. Spatial Variability of Soil As. The parameters for the random effects of the linear mixed models are presented in Figure 2. 10465

dx.doi.org/10.1021/es201726c |Environ. Sci. Technol. 2011, 45, 10463–10470

Environmental Science & Technology

ARTICLE

Figure 2. Variograms ﬁtted by residual maximum likelihood (REML). The parameters are C1 = autocorrelated variance (mg2 kg2); j = distance parameter (m); C0 = nugget variance (mg2 kg2). Filled circles represent the experimental variogram of the data, computed by conventional method-of-moments estimation.

For total As in the surface soil, the spherical model was the best variogram for the random effects and for phosphate-extractable As at three depths, the exponential function was best. Examination of the residual diagnostics showed that the assumption of normality was met in all cases except the phosphate-extractable As at 0.40.6 m depth (SI Figure S1d), which was subsequently log-transformed to stabilize its variance. For total As, range (j) was 1.169 m based on the spherical model and for the phosphateextractable As, j was between 0.259 and 0.491 m based on the exponential model (Figure 2), which is an effective range (3j) of 0.777 to 1.47 m. These results show that the As content in soil for total and phosphate-extractable As varied over short distances which could be efficiently described using geostatistical methods. The data also indicate that variation in phosphate-extractable As in the top two layers is less structured (i.e., more variable) than that at the lowest depth, as lower range (parameter) values were observed for the first two depths than at the lowest depth (see Figure 2). The parameters of the trend are presented in Table 1. The maps generated by E-BLUP interpolation of the linear mixed models are shown in Figure 3. The total and phosphateextractable As in the surface layer (00.2 m depth) increased signiﬁcantly (p = 0.004 and 0.038) from south to north direction (i.e., northings-trend, see Figure 1), moving toward the cattle-dip (Table 1; Figure 3a,b). The eastings-trend (i.e., moving toward east in a direction parallel to the cattle-dip, see Figure 1) also showed a signiﬁcant (p < 0.01) increase in the total and phosphateextractable soil As for the surface layer (Table 1). The increase in

northings-trend in total and phosphate-extractable As for the surface layer was by a factor of 192.9 and 7.7 mg kg1, respectively, with every 1 m approaching the cattle-dip. The eastings-trend indicated a relatively greater increase in As content (Table 1) with a unit distance parallel to the east of the cattle-dip. For the second depth (0.20.4 m), similar trends were evident for phosphateextractable As, however, the magnitude of increase was lower than that of the top-soil (00.2 m) (Table 1; Figure 3c). For phosphate-extractable As at 0.40.6 m, the only signiﬁcant eﬀect was in the northings-trend which showed As to decrease away from the cattle-dip which is similar to the top two layers (Table 1). These trends are also apparent in the As variability maps; for example, an increasing northings-trend for total As in the surface depth (i.e., darker hues in the area closer to the cattle-dip) (Figure 3a). A signiﬁcant increase in the northings-trend of phosphateextractable As at the lowest depth is also evident in the map (i.e., the blue color in proximity of the cattle-dip showing higher As concentration) (Figure 3d). This could be associated with the continuous deposition and outﬂow of dip solution in the area adjacent to the cattle-dip that could possibly result in the greater downward ﬂow of As.2,5,7,9 Kimber et al.7 surveyed over 25 cattle-dip sites in northern NSW and reported that As concentration was greater (>500 mg kg1) adjacent to cattle-dips and declined to <50 mg kg1 at a distance of >15 m from the dip bath. Increasing total and phosphate-extractable As concentration as indicated by the eastings-trend could be associated with the entrance direction 10466

dx.doi.org/10.1021/es201726c |Environ. Sci. Technol. 2011, 45, 10463–10470

Environmental Science & Technology

ARTICLE

Table 1. Fixed Eﬀect Parameters for Total and PhosphateExtractable As (mg kg1 Dry Weight) in Soil at the Diﬀerent Depths variable

estimatea standard error (()

P-value

total As (00.2 m) intercept

293.7

186.7

0.060

eastings

408.5

161.3

0.007

northings

192.9

68.7

0.004

112.8

59.5

0.031

eastings northings phosphate-extractable As 00.2 m intercept

27.1

11.71

0.012

eastings

28.5

9.8

0.002

7.7

4.3

0.038

7.8

3.6

0.016

intercept

35.7

10.7

>0.001

eastings northings

23.8 6.2

9.3 3.9

0.006 0.048

6.9

3.4

0.024

0.23

8.05 1024

northings eastings northings 0.20.4 m

eastings northings 0.40.6 mb intercept eastings northings eastings northings

3.00

of the prediction error variance.1 The results validate the models that described the spatial trends in the soil As content adjacent to the cattle-dip. Estimation of Sample Size. Total As content is used for soil contamination and remediation purposes,10 therefore, the total As (00.2 m) data were used to estimate sample size in this study. For estimating the sample size, the linear mixed models were refitted without the trend (SI Figure S2ad), and the sample variance was estimated from the semivariance of the fitted semivariogram models.28 The MSDR values were close to 1.0 for all models (SI Figure S3) which indicates that the kriging variance accurately reflects the prediction error and the semivariogram model correctly models the spatial variation.1 This is important when using the semivariogram to estimate the sample variance. The regression and reference lines overlapped which show a good relationship between the predicted and observed values (SI Figure S3ad). The results for total As in the surface layer based on eq 8 are shown graphically in Figure 5 where the SEM is shown to be dependent on the number of samples up to a certain point (20 samples). The SEM values calculated for a sample size of 5, 10, and 15 were 171.4, 121.1, and 99.0 mg kg1, respectively. Moreover, the SEM can be used to estimate the conﬁdence interval around the mean at 95% conﬁdence level which is as follows: y̅ ( tn0:025 1 SEM

c

ns

0.20

0.08

0.008

ns

a

The estimated parameters presented here correspond to the semivariogram models shown in Figure 2 . b The data were log-transformed, log(mg kg1). c No signiﬁcant trends.

of cattle into the dip bath. When cattle were forced to pass through the dip, splashing of the dipping solution in the region close to the cattle entrance side would be expected. The eastward increase in soil As content (total and phosphate-extractable) at the studied depths adjacent to the cattle-dip supports this explanation. In the top two layers, phosphate-extractable As concentrations were greater than that at the lowest depth (i.e., log-transformed As data were back-transformed)30 (Figure 3b,c,d). The presence of As in the 0.20.4 m layer could be due to the mixing of soil at the studied site. The restricted ﬂow below 0.4 m depth was possibly due to the presence of a clay hard-pan below this layer, recognized during soil sampling. The continuous movement of cattle over a long period in the vicinity of the dip bath possibly caused the development of the hard-pan.9 The limited leaching of As to deeper horizons could also be attributed to the presence of very high free Fe content (16%) in soil at this site; Fe oxides are known to have high adsorption capacity for As,2,9,31,32 which could minimize As leaching. Validation of the Linear Mixed Models. For all linear mixed models, the observed and predicted values (circles) for total and phosphate-extractable As data were uniformly distributed around the reference line (dotted line). The reference and regression lines overlapped (Figure 4ad), suggesting a good prediction using the model-based geostatistical approach. The LOOCV statistics show ME for all linear mixed models close to zero, suggesting unbiased estimates when the spatial trend was taken into account. The MSDR values were close to 1.0 for all models, which indicated no substantial overestimation and underestimation

ð9Þ

Based on the sample mean of 826 for total As in this study and at 95% conﬁdence level, we will obtain a conﬁdence interval of (475.9, (274.2, and (212.3 mg kg1 for a sample size of 5, 10 and 15, respectively. At 95% conﬁdence interval about mean (826 ( 475.9 mg kg1), a sample size of 5 could eﬃciently assess that the mean As value at the site is signiﬁcantly greater than the EIL limit (20 mg kg1) for total As in soil (see Figure 5). However, for further resampling to assess the eﬃcacy of phytoremediation eﬀorts a sample size of 15 (95% conﬁdence interval about mean = (212.3 mg kg 1 ) should be considered (Figure 5), since the variance of the diﬀerence in the spatial means between two sampling events could be quite large as it is equal to: 2 2 2 varðμ ̅ 1Þ ¼ σ 1 þ σ 2 σ 1, 2 ̅ 2μ

ð10Þ

where σ21 is the square of the SEM of the initial soil survey, σ22 is the square of the SEM of the subsequent soil survey, and σ21,2 is the covariance between both soil surveys. In this section, we have assumed that a simple random sampling scheme has been implemented but other options are possible which can improve the eﬃciency of the sampling scheme. Given the signiﬁcance of the trend parameters, i.e., decrease in As concentration in soil with distance from the dip bath and the highly variable nature of As, a stratiﬁed random sampling scheme can be employed where the strata are delineated based on distance from the dip bath. Our results found that the spatial variability of (total and phosphate-extractable) As in soil adjacent to the studied cattledip site was both large and varied over shorter distances; the As concentration in soil signiﬁcantly decreased away from the cattledip. Using the information on spatial variability, we recommend that a generic sampling scheme should have 5 samples which are good enough to statistically assess the contamination level with 10467

dx.doi.org/10.1021/es201726c |Environ. Sci. Technol. 2011, 45, 10463–10470

Environmental Science & Technology

ARTICLE

Figure 3. Maps showing the spatial variability of (a) total As at 00.2 m depth generated using a spherical model (untransformed data); and phosphateextractable As for (b) 00.2 m and (c) 0.20.4 m depths using an exponential model (untransformed data); (d) 0.40.6 m depth using an exponential model (log-transformed data were back-transformed).

respect to the given threshold value at the site; and 15 samples would be suﬃcient to determine the eﬀect of phytoremediation on As content in soil. However, further site-speciﬁc investigations are required to assess the level of contamination adjacent to the other As-contaminated cattle-dip sites using the sampling

guidelines proposed in this study. Geostatistical approach is recommended to estimate the variability in the contamination level at sites such as cattle-dip sites or other point-source contaminated sites. This methodology could also be useful for monitoring the changes in a contaminant level in the soil after 10468

dx.doi.org/10.1021/es201726c |Environ. Sci. Technol. 2011, 45, 10463–10470

Environmental Science & Technology

ARTICLE

Figure 4. Regression and reference lines showing the distribution of observed and predicted values for the total and phosphate-extractable As at the investigated depths, where the variogram models were ﬁtted based on the spatial (northings and eastings) trend (in Figure 2); (a,b,c) untransformed data (mg kg1 As), (d) log-transformed data (log, mg kg1 As). The leave-one-out cross validation (LOOCV) indices are also presented, ME = mean error; MSE = mean square error; MSDR = mean squared deviation ratio.

’ ASSOCIATED CONTENT

bS

Supporting Information. Total As digestion method, QA/QC procedure, and 3 ﬁgures. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +61 2 8627 1140; fax: +61 2 8627 1099; e-mail: Balwant. [email protected].

Figure 5. Number of samples (n) required to estimate the spatial mean of total As (mg kg1 dry weight) in soil with a known error and conﬁdence interval (CI) at 95% conﬁdence level (CL) around mean (untransformed data).

(phyto)remediation activities. It is also important to understand various processes which might have caused soil contamination from a particular activity at a given site, and remediation action undertaken at the site.

’ ACKNOWLEDGMENT N.K.N. gratefully acknowledges the Higher Education Commission of Pakistan for the award of a PhD scholarship. The project has been assisted by the NSW Government through its Environmental Trust. We are thankful to Dr. Lukas Van Zwieten, Stephen Kimber, George Nastase, Joshua Rust, Scott Petty, Victor Warren, and Desmond Cook at Wollongbar Agricultural Research Institute for ﬁeld assistance. N.K.N. is also thankful to Michael Nelson and Peter Geelan-Small (The University of Sydney) for useful advice regarding the data analysis in R, and Dr. Richard Murray Lark (Rothamsted Research) for his useful discussion and providing feedback. Finally, we are thankful to Dr Jorge L. Gardea-Torresdey (AE) 10469

dx.doi.org/10.1021/es201726c |Environ. Sci. Technol. 2011, 45, 10463–10470

Environmental Science & Technology and two anonymous referees for their constructive comments on the manuscript.

’ REFERENCES (1) Aelion, C. M.; Davis, H. T.; Liu, Y.; Lawson, A. B.; McDermott, S. Validation of Bayesian kriging of arsenic, chromium, lead, and mercury surface soil concentrations based on internode sampling. Environ. Sci. Technol. 2009, 43 (12), 4432–4438. (2) Smith, E.; Naidu, R.; Alston, A. M. Arsenic in the soil environment: A review. Adv. Agron. 1998, 64, 149–195. (3) Alary, C.; Demougeot-Renard, H. Factorial kriging analysis as a tool for explaining the complex spatial distribution of metals in sediments. Environ. Sci. Technol. 2010, 44 (2), 593–599. (4) Angus, B. M. The history of the cattle tick Boophilus microplus in Australia and achievements in its control. Int. J. Parasitol. 1996, 26 (12), 1341–1355. (5) Okonkwo, J. O. Arsenic status and distribution in soils at disused cattle dip in South Africa. Bull. Environ. Contam. Toxicol. 2007, 79 (4), 380–383. (6) Sarkar, D.; Makris, K. C.; Parra-Noonan, M. T.; Datta, R. Eﬀect of soil properties on arsenic fractionation and bioaccessibility in cattle and sheep dipping vat sites. Environ. Int. 2007, 33 (2), 164–169. (7) Kimber, S. W. L.; Sizemore, D. J.; Slavich, P. E. G. Is there evidence of arsenic movement at cattle tick dip sites? Aust. J. Soil Res. 2002, 40 (7), 1103–1114. (8) Niazi, N. K.; Singh, B.; Shah, P. Arsenic speciation and phytoavailability in contaminated soils using a sequential extraction procedure and XANES spectroscopy. Environ. Sci. Technol. 2011, 45 (17), 7135–7142. (9) Niazi, N. K.; Singh, B.; Van Zwieten, L.; Kachenko, A. G. Phytoremediation potential of Pityrogramma calomelanos var. austroamericana and Pteris vittata L. grown at a highly variable arsenic contaminated site. Int. J. Phytorem. 2011, 13 (9), 912–932. (10) National Environment Protection (Assessment of Soil Contamination) Measure (NEPM) Investigation Levels for Soil and Groundwater; National Environmental Protection Council Service Corporation: Adelaide, Australia, 1999; p 9. (11) Hossain, M. B.; Jahiruddin, M.; Panaullah, G. M.; Loeppert, R. H.; Islam, M. R.; Duxbury, J. M. Spatial variability of arsenic concentration in soils and plants, and its relationship with iron, manganese and phosphorus. Environ. Pollut. 2008, 156 (3), 739–744. (12) Dittmar, J.; Voegelin, A.; Roberts, L. C.; Hug, S. J.; Saha, G. C.; Ali, M. A.; Badruzzaman, A. B. M.; Kretzschmar, R. Spatial distribution and temporal variability of arsenic in irrigated rice ﬁelds in Bangladesh. 2. Paddy soil. Environ. Sci. Technol. 2007, 41 (17), 5967–5972. (13) Roberts, L. C.; Hug, S. J.; Dittmar, J.; Voegelin, A.; Saha, G. C.; Ali, M. A.; Badruzzanian, A. B. M.; Kretzschmar, R. Spatial distribution and temporal variability of arsenic in irrigated rice ﬁelds in Bangladesh. 1. Irrigation water. Environ. Sci. Technol. 2007, 41 (17), 5960–5966. (14) Chang, T. K.; Shyu, G. S.; Lin, Y. P.; Chang, N. C. Geostatistical analysis of soil arsenic content in Taiwan. J. Environ. Sci. Health, Part A 1999, 34 (7), 1485–1501. (15) Lark, R. M.; Cullis, B. R.; Welham, S. J. On spatial prediction of soil properties in the presence of a spatial trend: The empirical best linear unbiased predictor (E-BLUP) with REML. Eur. J. Soil Sci. 2006, 57 (6), 787–799. (16) Diggle, P. J.; Tawn, J. A.; Moyeed, R. A. Model-based geostatistics. Appl. Stat. 1998, 47 (3), 299–350. (17) Lark, R. M.; Cullis, B. R. Model-based analysis using REML for inference from systematically sampled data on soil. Eur. J. Soil Sci. 2004, 55 (4), 799–813. (18) Cattle Dip Site Locator. http://www.dpi.nsw.gov.au/agriculture/ livestock/health/speciﬁc/cattle/ticks/cattle-dip-site-locator (8 August, 2011). (19) Isbell, R. F. The Australian Soil Classiﬁcation, revised ed.; CSIRO: Collingwood, Victoria, 2002.

ARTICLE

(20) Huang, P. M.; Fujii, R. Selenium and arsenic. In Methods of Soil Analysis. Part 3, Chemical Methods; Sparks, D. L., Ed.; Soil Science Society of America: Madison, WI, 1996; pp 793831. (21) R Development Core Team, R: A language and environment for statistical computing; R Foundation for Statistical Computing: Vienna, Austria, ISBN 3-900051-07-0, 2008; http://www.R-project.org/. (22) Ribeiro, P., Jr.; Diggle, P. geoR: A package for geostatistical data analysis. R-NEWS 2001, 1 (2), 15–18. (23) Marchant, B. P.; Lark, R. M. Estimation of linear models of coregionalization by residual maximum likelihood. Eur. J. Soil Sci. 2007, 58 (6), 1506–1513. (24) Pringle, M. J.; Marchant, B. P.; Lark, R. M. Analysis of two variants of a spatially distributed crop model, using wavelet transforms and geostatistics. Agric. Syst. 2008, 98 (2), 135–146. (25) Webster, R.; Oliver, M. A. Geostatistics for Environmental Scientists; John Wiley & Sons: Chichester, UK, 2001. (26) Webster, R.; McBratney, A. B. On the Akaike information criterion for choosing models for variograms of soil properties. J. Soil Sci. 1989, 40, 493–496. (27) de Gruijter, J. J.; Brus, D. J.; Bierkens, M.; Knotters, M. Sampling for Natural Resource Monitoring; Springer Berlin Heidelberg: New York, 2006. (28) Domburg, P.; Degruijter, J. J.; Brus, D. J. A structured approach to designing soil survey schemes with prediction of sampling error from variograms. Geoderma 1994, 62 (13), 151–164. (29) Lark, R. M. Estimating the regional mean status and change of soil properties: Two distinct objectives for soil survey. Eur. J. Soil Sci. 2009, 60 (5), 748–756. (30) Zhou, X. H.; Gao, S. Conﬁdence intervals for the log-normal mean. Stat. Med. 1997, 16, 783–790. (31) Fendorf, S.; Eick, M. J.; Grossl, P.; Sparks, D. L. Arsenate and chromate retention mechanisms on goethite 0.1. Surface structure. Environ. Sci. Technol. 1997, 31 (2), 315–320. (32) Kim, J. Y.; Davis, A. P.; Kim, K. W. Stabilization of available arsenic in highly contaminated mine tailings using iron. Environ. Sci. Technol. 2003, 37 (1), 189–195.

10470

dx.doi.org/10.1021/es201726c |Environ. Sci. Technol. 2011, 45, 10463–10470

ARTICLE pubs.acs.org/est

Separating the Air Quality Impact of a Major Highway and Nearby Sources by Nonparametric Trajectory Analysis Ronald C. Henry*,† †

Department of Civil & Environmental Engineering, University of Southern California, 3620 South Vermont Avenue, Los Angeles California 90089-2531, United States

Alan Vette‡ ‡

U.S. EPA National Health and Environmental Eﬀects Research Laboratory, Research Triangle Park, North Carolina 27711, United States

Gary Norris§ and Ram Vedantham§ §

U.S. EPA National Exposure Research Laboratory, Research Triangle Park, North Carolina 27711, United States

)

Sue Kimbrough|| and Richard C. Shores|| U.S. EPA National Risk Management Research Laboratory, Research Triangle Park, North Carolina 27711, United States

bS Supporting Information ABSTRACT: Nonparametric Trajectory Analysis (NTA), a receptor-oriented model, was used to assess the impact of local sources of air pollution at monitoring sites located adjacent to highway I-15 in Las Vegas, NV. Measurements of black carbon, carbon monoxide, nitrogen oxides, and sulfur dioxide concentrations were collected from December 2008 to December 2009. The purpose of the study was to determine the impact of the highway at three downwind monitoring stations using an upwind station to measure background concentrations. NTA was used to precisely determine the contribution of the highway to the average concentrations measured at the monitoring stations accounting for the spatially heterogeneous contributions of other local urban sources. NTA uses short time average concentrations, 5 min in this case, and constructed local back-trajectories from similarly short time average wind speed and direction to locate and quantify contributions from local source regions. Averaged over an entire year, the decrease of concentrations with distance from the highway was found to be consistent with previous studies. For this study, the NTA model is shown to be a reliable approach to quantify the impact of the highway on local air quality in an urban area with other local sources.

’ INTRODUCTION The main objective of this paper is to estimate the near roadway impact of a major urban highway and other nearby sources on air quality by applying a new receptor based modeling technique, Nonparametric Trajectory Analysis (NTA), to observations of black carbon (BC), carbon monoxide (CO), nitrogen oxides (NOx), and sulfur dioxide (SO2). Sourceoriented dispersion models and methods can also be used to estimate the concentrations of air pollutants near a roadway, but these models require a priori information about the location and magnitude of emissions from all sources in the study domain. NTA and other receptor-oriented approaches use ambient data collected at monitoring sites to identify and quantify the impact of sources. Numerous studies have reported the eﬀects of major highways on local air quality for a wide range of pollutants.1 To isolate the r 2011 American Chemical Society

impact of the highway, it is necessary to account for the impact of other air pollution sources that could possibly cloud the interpretation of data collected in the vicinity of the highway. One methodology for accounting for other sources is the background subtraction method, the concentrations at a monitoring station upwind of the highway are subtracted from (or compared to) concentrations at one or more receptor monitoring stations downwind. An alternative approach is to measure pollutant concentrations at several distances downwind of the highway and observe the rate of decrease with distance; this is the gradient method.24 Of course, a major diﬃculty with both these Received: June 17, 2011 Accepted: November 1, 2011 Revised: October 19, 2011 Published: November 01, 2011 10471

dx.doi.org/10.1021/es202070k | Environ. Sci. Technol. 2011, 45, 10471–10476

Environmental Science & Technology methods is determining when the highway is upwind and when it is downwind of a monitor station, especially under light wind conditions. Wind speed and direction can vary on a time scale of minutes; periods of relatively constant winds that make the terms “upwind” and “downwind” meaningful must be carefully selected based on the variability of the wind direction. This often leads to the exclusion of most of the data, while NTA makes use of much more of the data. When properly applied, the background subtraction and gradient methods can be very useful. However, they cannot identify or quantify nonhighway sources as NTA does. These nonhighway sources may obscure the interpretation of data intended to determine the impact of a highway on local air quality. Unlike the gradient and related methods, instead of working with a relatively small number of selected periods with winds in the right direction relative to a speciﬁc, single source, NTA constructs back-trajectories and concentrations covering all available data to estimate the localized inﬂuence of all air pollution sources in the area. In this case, “local” refers to within ∼5 km of the sampling stations. Obviously, this degree of spatial resolution requires measurements with time resolution ﬁner than the standard 1-h averages. In this case, concentrations of species of interest and wind data were made on a 5-min average basis. Working with the totality of this data, NTA is able to determine the local geographical regions that are the source of high concentrations of pollutants observed at the monitor stations (i.e., receptors). In the past, 1-h was the shortest time resolution commonly available for meteorology and air quality data. So that analysis using 1 or 5-min averages was out of the question. 1-min and 5 min average meteorological data are becoming more common. The automated weather stations at major airports and other locations report 2-min running averages every minute. Routine air quality data is often logged every 1 or 2 min but only reported as 1-h averages. But special studies such as the subject of this paper are increasingly taking short-time averaged data. The NTA model and its application to measure the impact of a major highway on air quality in Las Vegas, NV are described below. Measurement data were collected at four sites transecting both sides of I-15, a major interstate in Las Vegas. The NTA estimates of the highway contribution are compared with the more conventional gradient method and are in reasonable agreement though the NTA model identiﬁed the presence of other sources of air pollution in addition to the interstate highway.

’ NTA METHODOLOGY Nonparametric Trajectory Analysis (NTA) is a receptororiented model that uses ambient measurement data to quantify the eﬀects of nearby sources on local air quality. The NTA model uses relatively short-time resolution data (1 to 5 min average) on pollutant concentrations and wind speed and direction to construct back-trajectories.5 Making use of nonparametric regression, NTA calculates point estimates of the conditional expected value of a pollutant at a receptor provided the air parcel has passed through a speciﬁc point prior to reaching the receptor site. Assume there are n back-trajectories with m points equally spaced in time along each trajectory arriving at a receptor. Let the points on the back-trajectories be given by (xij,yij) where i = 1,...,m and j = 1,...,n, further let Cj be the concentration at the receptor at the start of the jth back-trajectory, then the NTA value at point

ARTICLE

Figure 1. Back-trajectories (black lines) of air parcels arriving at a receptor site located at the origin (0, 0). Each point along a trajectory represents the location of the air parcel at 5-min intervals and is associated with the concentration observed when the trajectory arrives at the receptor site. The green line illustrates the two-dimensional trajectory of an air parcel arriving at the receptor site and the corresponding air pollutant concentration (32.1 ppb in this case). The NTA value for point (X,Y) is the weighted sum of the values associated with the trajectory points inside the red circle whose radius is the smoothing parameter (h).

(X,Y) is the expected value of concentration C given that air passes over point (X,Y) before reaching the monitoring station and is given by m

n

∑ ∑ CjWij i¼1 j¼1 m

n

∑∑

i¼1 j¼1

ð1Þ

Wij

where

X xij Y yij Wij ¼ K K h h

ð2Þ

and KðuÞ ¼ 0:75ð1 u2 Þ forjuj e 1 KðuÞ ¼ 0 otherwise

ð3Þ

Note that the weights Wij are all non-negative and have a maximum value of 0.752 = 0.5625. The smoothing parameter h is the radius of a circle centered at (X,Y) within which the expected value of concentration C is determined based on empirical data observed at a receptor (see Figure 1). The NTA value for point (X,Y) is the weighted sum of the values associated with the trajectory points within a radius of h (red circle in Figure 1). The weights for each trajectory point are based on the distance of the point from (X,Y). In this work, the smoothing parameter, h, is 0.5 km. This means that a back trajectory must pass within a radius of 0.5 km of the analysis point (X, Y) to be included in the NTA calculation. NTA results are generally not very sensitive to the value of the smoothing parameter. The value of 0.5 km was arrived at 10472

dx.doi.org/10.1021/es202070k |Environ. Sci. Technol. 2011, 45, 10471–10476

Environmental Science & Technology

ARTICLE

Table 1. Air Pollutants and Covariates Measured at Each Location Adjacent to I-15 pollutant or covariate

Figure 2. Schematic diagram showing the location and distances of monitoring stations from I-15. The annual wind rose shows the prevalence of winds from the south and west in the study area.

empirically as it gives results that are not too smooth or too rough. The smoothing parameter can be estimated by computationally intensive methods; however, these methods are timeconsuming and give smoothing parameters that are in agreement with the much faster empirical methods. The NTA analysis points (X, Y) are deﬁned on a 50 50 grid with the origin at the receptor and X and Y limits of 5 km and +5 km, thus the underlying grid resolution is 10 km ÷ 50 = 0.2 km or 200 m. The eﬀective resolution is twice this, or 400 m. Given the 500 m smoothing parameter, the eﬀective minimum resolution of the NTA results in this study is about 500 m. In addition to the smoothing parameter, the NTA model has the following major inputs. A matrix of x coordinates and a matrix of corresponding y coordinates of the points on the backtrajectories. These matrices are conﬁgured so that each column represents a trajectory. Associated with each of these columns is the concentration of the pollutant at the receptor at the time the trajectory reaches the receptor. The NTA model is unique in using constructed backtrajectories on the scale of a few kilometers and meteorological data on the time scale of minutes to identify local source-receptor relationships. These back-trajectories are constructed using wind speed and direction observations that have both measurement error and natural variability. The eﬀect of this uncertainty in wind speed and direction is increased uncertainty in the back-trajectories and an associated increase in the uncertainty of the NTA results. The errors in the NTA results in this paper include the eﬀects of errors in the trajectories. The eﬀect of the errors in the trajectories is to increase the overall error of NTA results by about 25 to 35%. These results and the exact methods for calculating the backtrajectories and associated errors are the subject of a paper now in preparation.

’ APPLICATION OF NTA TO LAS VEGAS DATA The NTA results given below use 5-min averaged observations from a study to determine concentrations and variations in concentrations of mobile source related air pollutants as a function of distance adjacent to I-15, a heavily traveled freeway in Las Vegas, NV.6 The annual average daily traﬃc (AADT) count is in excess of 150,000 vehicles. Data were collected at various distances on both sides of the freeway (Figure 2) to assess its impact on near-road air pollutant concentrations. Monitoring

method

CO

NDIR

NO, NO2, NOx SO2

chemiluminescence ﬂuorescence

black carbon

light absorption at 880 nm

wind speed/direction

sonic anemometer

station 4 is located on the predominantly upwind side of the freeway and used to compare concurrently measured concentrations at three sites located on the predominantly downwind side of the freeway. The study was performed from December 2008December 2009. Sample Collection and Analysis. Table 1 shows the measurement methods of the relevant data used in this paper. Description of the Las Vegas Site. The study site is located in south Las Vegas along the I-15 freeway, just north of the intersection of I-215 and south of the Russell Road intersection. The site was selected after carefully considering numerous sites in the Las Vegas area.7 The site is located in an area of mostly commercial properties. I-15 is mostly at grade in this area; however, the site is located adjacent to a cut section (i.e., below grade) with gentle slopes on each side of the freeway up to atgrade level where the monitors are located. The cut section allows a freight railroad spur line to pass over I-15. The railroad is located close to the site but is used infrequently (once per day). The McCarran International Airport is located about 1 km east and predominately downwind of the site.

’ NONHIGHWAY SOURCES IDENTIFIED BY NTA In addition to the highway, the NTA results given below show that the monitoring stations are signiﬁcantly impacted by nearby nonhighway sources. These are identiﬁed as a Local Industrial Area (LIA) west of the highway within 3 km of the monitoring stations and McCarran International airport with the main runway 15 km east of the highway and monitoring stations. The term “airport” is used here to represent all the sources east of the highway, but these are dominated by the emissions of aircraft, ground support vehicles, and general traﬃc associated with the airport. Like the “airport”, the LIA is also a complex source. It includes many small manufacturing and construction related businesses. Many of these are located within 1 or 2 km of the I-215 interstate highway that runs eastwest through the area. Thus, the LIA emissions include some I-215 emissions. Also, the LIA emissions will have some contribution from local vehicle traﬃc, especially diesel trucks. Thus, in the sections below, “nonhighway” sources means sources other than the I-15 highway, not exclusively nonvehicular sources. The following section describes how NTA identiﬁes the impact of these nonhighway sources and the method used to estimate the impact of the highway independent of these local sources. The NTA maps for BC at stations 2 and 4 are shown in Figure 3 and Figure 4 respectively. These two sites are chosen because each is approximately 100 m from the highway and each had a full complement of air quality measurements, including sulfur dioxide. Even though the two stations are separated by only about 200 m, the NTA maps are very diﬀerent with one showing high concentrations to the west of the highway 10473

dx.doi.org/10.1021/es202070k |Environ. Sci. Technol. 2011, 45, 10471–10476

Environmental Science & Technology

Figure 3. NTA map for black carbon at Station 2 (135 m east of the highway). The units of black carbon are μg m3. The gray contour lines are the 5-sigma errors in the NTA estimates. On this map, north is at the top, the units are km. The northsouth solid line is I-15 with intersections shown as circles. The eastwest solid line is I-215. The monitoring stations are located along the rail spur shown as a dashed line crossing I-15. The airport terminal is the star shown north of the primary and secondary runways. The small triangle in the upper left quadrant is the location of a cement plant; the diamond is the location of a large truck and taxi depot. The Local Industrial Area (LIA) is the area west of I-15 bounded by the cement plant to the north and the taxi depot to the west.

Figure 4. NTA map for black carbon at station 4, 115 m west of the highway. The large rectangular box is the area used to determine the NTA results in Table 2.

(Figure 3) and the other to the east (Figure 4). Figure 3 shows that high concentrations of BC (about 2 μg m3) are associated

ARTICLE

Figure 5. Contribution of I-15 to black carbon concentrations observed approximately 125 m downwind of I-15.

with back-trajectories spread widely over the area west of the highway. On the other hand, Figure 4 shows that the area of highest concentrations for station 4 (about 1.8 μg m3) is toward the east and centered near the main runway of the airport. The reason for the diﬀerence in these two maps is I-15 a major highway that runs northsouth between the two stations. Before arriving at station 2, back-trajectories from the west must pass over the highway; likewise, for station 4, the back-trajectories from the east must pass over the highway. Thus, air coming from the west over the LIA must pass over the highway before arriving at station 2. Air coming from the east over the airport must pass over the highway before arriving at station 4. Consequently, the highest average concentrations at station 2 are those associated with back-trajectories from the west (Figure 3) as these pass over two sources, the LIA and the highway. Similarly, the highest average concentrations at station 4 are associated with backtrajectories from the east that have passed over the highway and the airport (Figure 4). During periods of low wind speeds, air could meander back and forth over the highway. Low wind speeds were quite common in these data: the distribution of wind speeds at station 2 was highly skewed with a peak (mode) at 1.3 ms1 and a maximum of about 12 ms1. Even if the wind speeds are greater, variable wind directions along the back-trajectories may obscure simple upwinddownwind relationships. However, when averaged over tens of thousands of back-trajectories, these complications caused a minor increase in the NTA results. In Figures 3 and 4, the areas to the east of station 2 and to the west of station 4, respectively, are regions where air has not yet passed over the highway before reaching the receptor and represent the impact of the airport (Figure 3) and the LIA alone (Figure 4). On average, air passing over the airport has a black carbon concentration of about 1.3 to 1.4 μg m3 when it reaches station 2. Similarly, NTA shows that the LIA impact on station 4 varies from about 1.1 to 1.3 μg m3. These values may be taken as upwind or background values and used to estimate the impact of the highway separate from these local sources. Figure 3 shows 10474

dx.doi.org/10.1021/es202070k |Environ. Sci. Technol. 2011, 45, 10471–10476

Environmental Science & Technology

ARTICLE

that for station 2 the downwind concentration of BC is at most about 1.8 to 2 μg m3, subtracting 1.1 to 1.3 μg m3 background, gives an approximate average highway impact of 0.50.9 μg m3 black carbon at station 2 with transport from the west. The following section provides a discussion of the spatially variable impact of the highway in the presence of other local sources including the airport and the LIA.

’ HIGHWAY IMPACT IDENTIFIED BY NTA The NTA estimates of the highway contribution of BC concentrations observed at station 2 were calculated as the diﬀerence between the NTA results for station 2 (Figure 3) and station 4 (Figure 4). The results are shown in Figure 5. When

Figure 6. NTA map of sulfur dioxide in ppb at station 2.

station 2 is downwind of I-15, the impact of the highway is positive for back-trajectories coming from the west. The backtrajectories coming from the east of I-15 almost always had higher BC concentrations at station 4 than station 2 as it is downwind of the highway for these back-trajectories. Thus, the east side of the Figure 5 has mostly negative values; these are shown as shades of blue in the ﬁgure. The impact of the highway on station 4 seen in Figure 5 is determined by taking the absolute value of the negative values to the east of the highway and is about half of the impact of the highway on station 2. Note that the highway impact on station 4 is quite uniform, except for an area of anomalous positive values in the northeast. The 5-sigma error contours in the ﬁgure are about 1 μg m3, while the values are about 0.15 μg m3. Looking closely at the data, the cause is found to be few periods where BC at station 2 is much higher than all the other stations. These high BC numbers at station 2 seem to be real and not instrumental errors. Nonetheless, the NTA error analysis has correctly identiﬁed the eﬀect on NTA. In fact, a similar area of large errors is seen in the same place in Figure 3. The ability to use the errors in the NTA to identify areas aﬀected by outliers or questionable data is one of the major strengths of the method. The NTA map of BC from the highway is unusual in that the greatest impact of the highway is not associated with locations perpendicular to the highway near stations 2 and 4 as expected but with back-trajectories coming from southwest to south southwest (195245 degrees azimuth). Winds from this direction were especially common in the summer. The high concentrations may be associated with a vortex formed when the air from this direction meets the sunken highway and railway overpass described above in the site description section. The NTA maps were calculated for all the stations for BC, CO, and NOx. The results for NOx and CO are generally similar to those for BC. Only stations 2 and 4 had measurements for SO2. The NTA results for SO2 at station 2 are given in Figure 6. The highest concentrations of SO2 at station 2 occur with transport from the airport and are centered on the main and secondary runways. Since station 2 is east of the highway, the high

Table 2. Highway Impact from NTA Maps average/background average/background

std. dev.

composite of other studiesb

BC (μg m3) 0.0977

1.97

0.13

1.8

0.719

0.0943

1.64

0.11

1.6

0.288

0.1000

1.25

0.10

1.2

0.011

0.184

0.0157

1.52

0.06

21

0.455

0.0093

0.102

0.0145

1.29

0.05

3.8

station 3

0.439

0.0129

0.0864

0.0170

1.24

0.05

1.4

backgrounda

0.352

0.0111

station 1

83.80

2.81

39.55

3.64

1.89

0.12

1.8

station 2

64.54

3.23

20.29

3.98

1.45

0.11

1.6

station 3 backgrounda

59.80 44.25

3.39 2.32

15.55

4.11

1.35

0.10

1.3

average

std. dev.

highway

std. dev.

station 1

2.20

0.0715

1.08

station 2

1.84

0.0669

station 3

1.41

0.0747

backgrounda

1.12

0.0665

station 1

0.536

station 2

CO (ppm)

NOx (ppb)

a

Station 4. b Values from Figure 2 in Karner et al.1 10475

dx.doi.org/10.1021/es202070k |Environ. Sci. Technol. 2011, 45, 10471–10476

Environmental Science & Technology concentrations from the airport have no highway contribution. The NTA values west of the highway represent the impact of the LIA and the highway. Unlike BC, CO, and NOx, the airport is the dominant source of SO2 in the area. This is reasonable because the allowable sulfur content in jet fuel at the time of this study was still high (∼500 ppm) compared to the ultralow sulfur fuel used by cars and trucks (∼15 ppm). UpwindDownwind Analysis. A version of an upwind downwind analysis can be done with the NTA maps in Figures 3 and 4. The NTA analysis area is the 2 3 km box shown in black in Figure 4. The first column of Table 2 gives the average values of BC, CO, and NOx observed at each station when the air parcel has passed over the box before reaching that station. As explained above, this is a sum of the impact of the LIA and the highway. The value for station 4 is taken to be the impact of the LIA alone and is assumed to be the background. The second column is the standard deviation of the NTA values in the box, which is much greater than the standard deviation due to errors in the NTA. The highway column is the estimated impact of the highway, which is given by the average in the first column minus the background. The standard deviation of this number is given next. The ratio of the average to the background is given next along with its standard deviation. For comparison, the background-normalized values from a compilation of studies for the same distance are shown. For BC and NOx, the concentrations normalized to the background are very similar to those reported in the literature.1 CO, on the other hand, does not behave at all like CO in previously reported studies near highways. In this study the background CO was almost twice the amount of CO contributed by the highway at Station 1. It is not clear why the CO background was much higher in this study than the other highway related pollutants but likely the result of relatively high CO emissions from the LIA and other roadways Finally, the NTA results for BC were compared to the gradient method. The Supporting Information contains a brief description of the analysis and a graph comparing the gradient method, NTA, and literature values for BC. The NTA results are seen to be in good agreement with the literature values and the gradient method. The results for CO and NOx are similar to BC.

ARTICLE

(2) Clements, A. L.; Jia, Y.; Denbleyker, A.; McDonald-Buller, E.; Fraser, M. P.; Allen, D. T.; Collins, D. R.; Michel, E.; Pudota, J.; Sullivan, D. et al. Air pollutant concentrations near three Texas roadways, part II: Chemical characterization and transformation of pollutants. Atmos. Environ. 2009, 43 (30), 45234534, DOI: 10.1016/j.atmosenv.2009.06.044. (3) Zhu, Y.; Pudota, J.; Collins, D.; Allen, D.; Clements, A.; DenBleyker, A.; Fraser, M.; Jia, Y.; McDonald-Buller, E.; Michel, E. Air pollutant concentrations near three Texas roadways, part I: Ultraﬁne particles. Atmos. Environ. 2009, 43 (30), 45134522, DOI: 10.1016/j. atmosenv.2009.04.018. (4) Thoma, E. D.; Shores, R. C.; Isakov, V.; Baldauf, R. W. Characterization of near-road pollutant gradients using path-integrated optical remote sensing. J. Air Waste Manage. Assoc. 2008, 58 (7), 879890, DOI: 10.3155/1047-3289.58.7.879. (5) Henry, R. C. Locating and Quantifying the Impact of Local Sources of Air Pollution. Atmos. Environ. 2007, 42 (2), 358363, DOI: 0.1016/j.atmosenv.2007.09.039 (6) Kimbrough, S. Longterm continuous measurement of near road air pollution in Las Vegas: Seasonal variability in traﬃc emissions impact on local air quality. Manuscript in preparation. (7) Kimbrough, S; Vallero, D; Shores, R; Vette., A; Black, K; Martinez, V. Multi-criteria decision analysis for the selection of a near road ambient air monitoring site for the measurement of mobile source air toxics. Transp. Res. Part D-Transp. Environ. 2008, 13 (8), 505–515, DOI: 10.1016/j.trd.2008.09.009.

’ ASSOCIATED CONTENT

bS

Supporting Information. Text and Figure S1. This material is available free of charge via the Internet at http://pubs.acs. org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT The United States Environmental Protection Agency through its Oﬃce of Research and Development funded and managed the research described here. It has been subjected to the Agency’s administrative review and approved for publication. ’ REFERENCES (1) Karner, A. A.; Eisinger, D. S.; Niemeier, D. A. Near-Roadway Air Quality: Synthesizing the Findings from Real World Data. Environ. Sci. Technol. 2010, 44 (14), 53345344, DOI: 10.1021/es100008x. 10476

dx.doi.org/10.1021/es202070k |Environ. Sci. Technol. 2011, 45, 10471–10476

ARTICLE pubs.acs.org/est

Global SourceReceptor Relationships for Mercury Deposition Under Present-Day and 2050 Emissions Scenarios Elizabeth S. Corbitt,*,† Daniel J. Jacob,†,‡ Christopher D. Holmes,§ David G. Streets,|| and Elsie M. Sunderland‡,^ †

Department of Earth and Planetary Sciences, Harvard University, Cambridge, Massachusetts 02138, United States School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138, United States § Department of Earth System Sciences, University of California Irvine, Irvine, California 92697, United States Decision and Information Sciences Division, Argonne National Laboratory, Argonne, Illinois 60439, United States ^ Department of Environmental Health, School of Public Health, Harvard University, Boston, Massachusetts 02115, United States

)

‡

bS Supporting Information ABSTRACT: Global policies regulating anthropogenic mercury require an understanding of the relationship between emitted and deposited mercury on intercontinental scales. Here, we examine sourcereceptor relationships for present-day conditions and four 2050 IPCC scenarios encompassing a range of economic development and environmental regulation projections. We use the GEOS-Chem global model to track mercury from its point of emission through rapid cycling in surface ocean and land reservoirs to its accumulation in longer lived ocean and soil pools. Deposited mercury has a local component (emitted HgII, lifetime of 3.7 days against deposition) and a global component (emitted Hg0, lifetime of 6 months against deposition). Fast recycling of deposited mercury through photoreduction of HgII and re-emission of Hg0 from surface reservoirs (ice, land, surface ocean) increases the eﬀective lifetime of anthropogenic mercury to 9 months against loss to legacy reservoirs (soil pools and the subsurface ocean). This lifetime is still suﬃciently short that sourcereceptor relationships have a strong hemispheric signature. Asian emissions are the largest source of anthropogenic deposition to all ocean basins, though there is also regional source inﬂuence from upwind continents. Current anthropogenic emissions account for only about one-third of mercury deposition to the global ocean with the remainder from natural and legacy sources. However, controls on anthropogenic emissions would have the added beneﬁt of reducing the legacy mercury re-emitted to the atmosphere. Better understanding is needed of the time scales for transfer of mercury from active pools to stable geochemical reservoirs.

’ INTRODUCTION Human activities have caused at least a 3-fold increase in atmospheric mercury deposition to terrestrial and aquatic ecosystems over the past two centuries.16 Mercury bioaccumulates in freshwater and marine foodwebs with health consequences for exposed wildlife and humans.79 Anthropogenic emissions are mainly from coal combustion, waste incineration, and mining.10 Growing concern about elevated mercury in the environment has prompted negotiations under the United Nations Environment Programme (UNEP) toward a global treaty on anthropogenic mercury sources. Improving the understanding of sourcereceptor relationships linking mercury emissions to deposition ﬂuxes is critical in this context. Here, we use a global atmospheric model with coupled surface reservoirs (GEOS-Chem) to quantify source receptor relationships on continental scales for the present-day and for 2050 emission projections. Anthropogenic activities emit mercury in both elemental (Hg0) and divalent (HgII) forms. HgII is highly water soluble r 2011 American Chemical Society

and can be deposited close to sources. Hg0 is only sparingly soluble and has an atmospheric lifetime of months against oxidation to HgII, resulting in global-scale deposition. The speciation of anthropogenic mercury varies with source type and emissions control technology. Emission controls for other pollutants, such as ﬂue-gas desulfurization (FGD) in coal combustion, capture HgII as a cobeneﬁt. Greater capture can be achieved with injection of chemicals to oxidize Hg0 to HgII or with particles designed to adsorb mercury upstream of FGDs.11 Projections of future anthropogenic mercury emissions out to 2050 have been reported by Streets et al.10 on the basis of four IPCC SRES scenarios12 spanning a range of industrial growth and environmental regulation possibilities. They ﬁnd that global Received: July 22, 2011 Accepted: November 3, 2011 Revised: October 24, 2011 Published: November 03, 2011 10477

dx.doi.org/10.1021/es202496y | Environ. Sci. Technol. 2011, 45, 10477–10484

Environmental Science & Technology

ARTICLE

Figure 1. Global present-day budget of mercury as represented in GEOS-Chem. Blue arrows show primary and legacy sources of mercury to the atmosphere from long-lived deep reservoirs. Red arrows show the fate of mercury in surface (ocean, land, snow) reservoirs: recycling to the atmosphere or incorporation into more stable reservoirs (deep ocean, soils). Black arrows show deposition and redox ﬂuxes. Green arrows show processes not explicitly modeled in GEOS-Chem. Order-of-magnitude residence times in individual reservoirs are also shown.

anthropogenic mercury emissions may at worst double in the future (A1B scenario) or at best stay constant (B1). Coal combustion in developing countries is the largest driver of emission increases. We examine the implications of these future scenarios for global mercury deposition and comment on the major uncertainties. There have been no studies to date that quantify future deposition for long-range IPCC scenarios. This information is needed to evaluate the eﬀectiveness of global and national-level reductions in anthropogenic emissions on mercury deposition rates and to inform policy decisions such as the ongoing UNEP global treaty negotiations.

’ METHODS General Model Description. We use the GEOS-Chem global mercury model version 8-03-02 (http://acmg.seas.harvard.edu/ geos/), including a 3-D atmosphere coupled to 2-D slab ocean and terrestrial reservoirs.1315 We conduct simulations at 4 5 horizontal resolution, with 47 atmospheric levels in the vertical, using assimilated meteorological fields from the NASA Goddard Earth Observing System (GEOS-5). Following Selin et al.,13 we first initialize the model to steady state for preindustrial conditions, and this serves to equilibrate the 2-D terrestrial reservoir. We then update the model to present day by including anthropogenic emissions, increasing terrestrial concentrations

on the basis of anthropogenic deposition patterns, specifying subsurface ocean concentrations for different basins based on observations,1517 and conducting a simulation for 7 years to equilibrate the atmosphere. For the 2050 scenarios, we start from present-day conditions in the surface reservoirs and conduct a simulation for 7 years using future anthropogenic emissions. All results presented here are 3-year averages using 20052007 meteorological data. The model used here is as described by Holmes et al.14 with addition of a more mechanistic and resolved surface ocean model.15 Detailed comparisons of the model to observations are presented in these two references. The model tracks three mercury forms in the atmosphere: Hg0, HgII, and refractory particulate mercury (HgP). HgP makes a negligible (<1%) contribution to the total atmospheric burden, and we do not discuss it further. The atmospheric speciation of mercury deposited to the ocean is not relevant for aqueous chemistry as rapid reequilibration takes place in solution in open-ocean environments.18 Figure 1 shows the global cycling of mercury in the environment as represented by the model. “Primary” emission from mineral reservoirs through anthropogenic activities (coal combustion, industry, mining) and natural geogenic processes (weathering, volcanoes) initiates cycling between the atmosphere and surface reservoirs mediated by Hg0/HgII redox chemistry. Redox chemistry in the atmosphere includes oxidation of 10478

dx.doi.org/10.1021/es202496y |Environ. Sci. Technol. 2011, 45, 10477–10484

Environmental Science & Technology Hg0 to HgII by Br atoms and aqueous photoreduction of HgII to Hg0 in clouds. Dry deposition applies to both Hg0 and HgII and wet deposition only to HgII. Uptake on sea salt particles is a major sink for HgII in the marine boundary layer.19 Processes in the surface ocean include photochemical and biotic redox chemistry as well as sorption to particles. Mercury can be re-emitted to the atmosphere as Hg0 or transferred to deeper ocean waters by particle sinking and vertical entrainment.15 HgII deposited to land can be promptly photoreduced and re-emitted or bind to organic carbon and enter longer lived soil pools.20 HgII deposited to snow can be photoreduced and re-emitted or eventually transferred to the oceans or soils through meltwater. Here, we denote anthropogenic mercury transferred from surface reservoirs to subsurface reservoirs (subsurface/deep ocean, soils) as “legacy” mercury. The current model does not explicitly resolve the recycling of this legacy mercury and instead includes it in the speciﬁcation of boundary conditions.13 We include biomass burning in our simulation but treat it as a legacy emission. Total present-day emissions from all surface reservoirs in the model, 5200 Mg a1 including net ocean evasion of Hg0, are within the range of recent estimates (36006300 Mg a1 16,21). An innovation in the current model is the tagging of mercury from source to receptor including transit through the surface reservoirs. Tagged mercury tracers for particular source regions or source types maintain their identity through transport, chemical transformation, and cycling through surface terrestrial and ocean reservoirs. Anthropogenic emissions are divided geographically into 17 world regions based on Streets et al.10 Mercury upwelling from the subsurface ocean is divided among diﬀerent ocean basins (Supporting Information, Figures 1 and 2). Geogenic (volcanoes, mineral weathering), soil, and biomass burning emissions are also separated as individual tracers. Anthropogenic Emissions. Present-day anthropogenic emissions are based on a 1 1 gridded, speciated inventory for the year 2005,22 and are scaled to regional emission totals from Streets et al.10 The magnitude of global anthropogenic emissions has an estimated uncertainty of (30%, while chemical speciation has an uncertainty of (20%.10,23 Year 2050 simulations keep the fine spatial distribution of emissions the same but apply regional scaling factors projected by Streets et al.10,23 Scaling emissions at the regional level assumes a uniform increase or decrease in emissions across all sources within each region. The projections are based on four IPCC SRES scenarios (A1B, A2, B1, B2) distinguished by their assumptions regarding industrial growth, energy policy, and emissions control. The worst-case scenario (A1B) assumes heavy use of coal with limited emission control technology, while the best-case scenario (B1) assumes aggressive transition away from fossil fuel energy sources and implementation of efficient control technology (up to 70% mercury capture in developed countries). We call these “end-member” scenarios. Scenarios A2 and B2 are intermediate and have more spatially heterogeneous trends (Supporting Information Figure 3). The speciation of emissions varies by region due to differences in sector makeup and emissions controls, from <30% HgII in South America and Northern Africa, where artisanal gold mining is a large source of Hg0, to >60% HgII in Eastern Europe, Southern Africa, and South Asia, where power production is the largest source of emissions. Developing countries with less stringent environmental controls undergo the most growth in the future scenarios, especially in coal combustion, resulting in a greater fraction of global anthropogenic emissions as HgII in 2050 (5560% compared to 43% in the present). Streets et al.10 do

ARTICLE

not consider in their base projections the introduction of more advanced mercury control technology such as activated carbon injection, which is not currently commercially available, but they note that anthropogenic emissions could be as much as 30% lower in each future scenario with widespread adoption. See Supporting Information Table 1 for a summary of global emissions and deposition for the scenarios used in this study.

’ RESULTS AND DISCUSSION Time Scales for Mercury Deposition. Deposition to a given region consists of a locally sourced component from emitted HgII and a background component. The mean model lifetime of Hg0 against oxidation and deposition in the troposphere is 4 months, while the mean lifetime of boundary layer HgII against deposition is 3.7 days. One-third of emitted HgII in the model is photoreduced to Hg0, transferring from the local to the background deposition pool. The mean model lifetime of anthropogenic HgT (HgT t Hg0+HgII) against deposition is 5 months, while the lifetime of HgT from all sources is 6 months because emissions from natural processes are as Hg0. Re-emission of deposited mercury from surface reservoirs (Figure 1) increases the effective lifetime of anthropogenic mercury to 7 months (9 months for mercury from all sources) against incorporation into legacy organic soil and deep ocean reservoirs. The ability of GEOSChem to reproduce the observed atmospheric variability of Hg0 14 lends some confidence in these model time scales. We refer to gross deposition as the removal of atmospheric Hg to the surface reservoirs, including wet deposition of HgII and dry deposition of Hg0 and HgII. Some of that gross deposition is reemitted to the atmosphere as Hg0, and we refer to the remainder as net deposition, balanced by transfer to deeper reservoirs (Figure 1). We view net deposition as the metric for mercury enrichment in ecosystems, balancing primary emissions on a global scale. Our tracking of mercury through surface reservoirs in GEOS-Chem enables us to relate net deposition to the original emission source. The 9-month lifetime of atmospheric Hg0 against transfer to the legacy reservoirs (i.e., accounting for reduction and re-emission from the surface reservoirs) is shorter than the time scale for interhemispheric exchange (∼1 year24), which means that a strong hemispheric signature is to be expected in sourcereceptor relationships even for the background component of mercury. Figure 2 shows annual mean gross and net deposition ﬂuxes in the model for present-day conditions. Gross deposition peaks over polluted continents due to emitted HgII and over windy regions of the oceans due to high Br concentrations and fast seasalt deposition. The fraction of deposited mercury that is reemitted rather than transferred to the deeper reservoirs is 10% for land, 40% for the oceans, and 50% for snow. Most of the mercury deposited to land enters the soil pools where it has an estimated mean lifetime of 80 years against re-emission by soil respiration20 and is included here as a legacy source. By contrast, mercury deposited to the surface ocean has a lifetime of only 6 months against re-emission, competing with transfer to the subsurface ocean (lifetime of 5 months). Net deposition of mercury in the model thus tends to be higher over land than over oceans. Global SourceReceptor Relationships. We define the sourcereceptor influence function Iij for mercury deposition as

Iij ¼ Dij =Ei

ð1Þ

where Dij is the net deposition flux to receptor region j from emissions in region i and Ei is the total emission rate for region i. 10479

dx.doi.org/10.1021/es202496y |Environ. Sci. Technol. 2011, 45, 10477–10484

Environmental Science & Technology

ARTICLE

Figure 2. Annual mean mercury deposition and fast re-emission from surface reservoirs simulated by GEOS-Chem for present-day conditions. Fast re-emission from surface reservoirs competes with transfer to longer lived reservoirs. Net deposition is the balance between gross deposition and fast re-emission.

Figure 4. Sources of mercury deposited to aggregated world regions for the present-day and for the four 2050 IPCC scenarios of Streets et al.10 Numbers give annual net deposition ﬂuxes to the receptor region (gross deposition ﬂuxes in parentheses) and for 2050 represent the range of the IPCC scenarios. Pie charts show relative source contributions to deposition (average of the scenarios for 2050). “New anthropogenic” refers to mercury from primary emissions (coal combustion, waste incineration, mining) including recycling through surface reservoirs (ocean mixed layer, vegetation). “Legacy” refers to anthropogenic mercury recycled from intermediate reservoirs with a time scale of decades or longer and included in GEOS-Chem as boundary condition. Figure 3. Inﬂuence functions for anthropogenic mercury emitted from source regions in three latitudinal bands: extratropical Northern Hemisphere (Canada, United States, Europe, Russia, East Asia), northern tropics (Central America, Northern Africa, Middle East, South Asia, Southeast Asia), and Southern Hemisphere (South America, Southern Africa, Australia). Maps show the preferential locations for deposition of mercury emitted from each latitudinal band, normalized to the magnitude of emissions as given by eq 1.

This influence function enables us to evaluate where, gram-forgram, emissions reductions would be most effective to reduce deposition to a given region. Figure 3 shows influence functions for anthropogenic emissions in the extra-tropical Northern Hemisphere, the northern tropics, and the Southern Hemisphere. We find that extra-tropical sources make a particularly large contribution to deposition within their hemisphere. Emissions in the tropics have a more distributed influence. See Supporting Information Figure 4 for additional maps of influence functions by individual source regions. Supporting Information Figure 5 shows the fraction of total deposition attributed to anthropogenic sources from each region. Figure 4 shows the source attribution for mercury deposited to aggregated world regions under present-day and 2050 emissions.

Constraints from sediment and ice cores and from current anthropogenic emission inventories imply that deposition on a global scale is approximately one-third natural, one-third legacy anthropogenic, and one-third primary anthropogenic.1,5,6,21 Natural and legacy mercury emissions from terrestrial soils and oceans contribute the majority of net deposition in all regions except Asia, stressing the importance of better resolving the legacy component in future work. For example, it is thought that Hg0 evasion in the North Atlantic Ocean is presently enhanced due to enrichment of subsurface seawater by legacy anthropogenic sources.15,16 North America is likely the strongest contributor to this enrichment due to its high inﬂuence function and very high emissions from mining in the late 19th century.25,26 Mercury deposition in 2050 relative to present day is similar in the B1 scenario but increases in the other IPCC scenarios, reﬂecting the global trend in emissions.10 The increasing HgII fraction of total mercury emissions in the future results in an increasing relative domestic contribution to deposition. This is most apparent in Asia, where the fraction of mercury deposition from domestic anthropogenic sources increases from 54% in the present day to 5675% in 2050. Natural and legacy emissions are assumed here to stay constant between present day and 2050, 10480

dx.doi.org/10.1021/es202496y |Environ. Sci. Technol. 2011, 45, 10477–10484

Environmental Science & Technology

ARTICLE

Figure 5. Annual mean net mercury deposition ﬂuxes to Asia for the present-day and end-member IPCC 2050 scenarios.

Figure 6. (Top panels) Annual mean net mercury deposition ﬂux to the United States for present-day and 2050 A1B and B1 scenarios. (Bottom panels) Changes in the source contributions from anthropogenic emissions in the United States, Central America, and Asia in the A1B 2050 scenario relative to present day.

and as a result their relative contribution to deposition decreases in 2050 for all receptor regions. This is likely an incorrect assumption as legacy emissions should increase in concert with future increases in anthropogenic emissions. Asian emissions (mostly from China and India) account for over one-half of global anthropogenic emissions in all 2050 scenarios, and the magnitude of their projected change relative to present spans from near constant to a 240% increase. However, it is important to distinguish between China and India, as increases in India are much larger due to considerable growth in coal combustion. Figure 5 shows net deposition to Asia for the present-day and for the end-member 2050 scenarios. Deposition in China and downwind increases in the A scenarios but declines in the B scenarios due to emission controls. Deposition to India and downwind increases in all 2050 scenarios and is consistently the highest in the world. Even installation of FGD in 95% of Indian power plants in the B1 scenario is insuﬃcient to decrease deposition levels relative to present day. Decreasing deposition to South Asia would require emissions controls speciﬁcally targeting mercury capture. Mercury Deposition to the United States in 2050. Figure 6 shows present-day and 2050 simulated deposition fluxes of mercury to the contiguous United States. Components of present-day deposition include domestic anthropogenic emissions (17%), foreign anthropogenic emissions (23%), and natural and legacy terrestrial and ocean mercury (60%). This is similar to the previous GEOS-Chem source attribution of Selin and Jacob.27 In the 2050 A1B scenario, both the background and the local components of deposition increase as global

anthropogenic mercury emissions more than double and North American emissions increase by 60%. We find a mean 30% increase in mercury deposition rates for the United States, less than the increase in emissions because we assume no change in the natural and legacy components. The bottom panels of Figure 6 show the increase in source contributions to U.S. mercury deposition in the 2050 A1B scenario relative to present day. U.S. sources account for most of the increase in the Northeast, while Central American emissions (including Mexico) are important mainly in Texas. The increase in Asian emissions enhances net deposition more uniformly across the country but most strongly in the Southeast, reflecting both the vegetation density (enhancing dry deposition) and the deep convective precipitation scavenging of HgII from the upper troposphere.27,28 Though South Asian sources (mainly India) undergo the most dramatic growth in A1B, we find that their impact on U.S. deposition is less than that of East Asian sources (mainly China) because of their lower latitude. In the B1 scenario, U.S. anthropogenic emissions decrease by 38% for both Hg0 and HgII. Global emissions are similar in magnitude to the present day but shift southward and are therefore less efficient contributors to U.S. deposition. Thus, 2050 mercury deposition to the United States decreases by 10% on average and by up to 22% in the Northeast. East Asian emissions contribute to deposition in the United States primarily by elevating background concentrations29,30 rather than by direct intercontinental transport of short-lived HgII species. We ﬁnd that only 6% of present-day East Asian deposition to the United States is from direct trans-Paciﬁc transport 10481

dx.doi.org/10.1021/es202496y |Environ. Sci. Technol. 2011, 45, 10477–10484

Environmental Science & Technology of HgII, though the share can be up to 25% in the Paciﬁc Northwest and Alaska. East Asian total emissions increase by 47% in the A1B scenario, but the East Asian contribution to deposition in the United States only increases by 35% because most of the emissions increase is as HgII. Gram-for-gram, emissions from Russia and Eastern Europe are more eﬃciently transported to the United States because of re-emission of mercury deposited to snow during transport over the Arctic. Model Uncertainties. There are a range of uncertainties involved in global mercury modeling,3134 some of which are especially relevant to our understanding of global source receptor relationships. One important uncertainty is the atmospheric reduction of emitted HgII. On a global scale, the rate of HgII reduction must be relatively slow, as implied by constraints from the observed seasonal variation of Hg0 and the atmospheric variability of HgT.14,35 We conduct a sensitivity simulation with no HgII reduction and with Hg0 oxidation rates correspondingly adjusted to match observational constraints on Hg0 concentrations and find no major effects on the results reported here. More details are available in the Supporting Information. However, there is some evidence for fast HgII reduction taking place in coal combustion plumes.3639 This reduction would decrease the local component of regional deposition in our simulation. The associated error is difficult to quantify because the mechanism for HgII reduction in fresh plumes is unknown.36 Regardless of the fate of primary HgII, an important result of our work is the latitudinal structure of sourcereceptor relationships for mercury, i.e., emissions have the greatest eﬀect on deposition in their latitudinal band (Figure 3). This follows from the atmospheric lifetime of HgT against deposition, which is constrained by observation of HgT atmospheric variability.40,41 There is presently discussion in the literature as to whether atmospheric oxidation of Hg0, determining HgT deposition, involves Br atoms or OH and O3.14,42,43 Our standard simulations uses Br atoms, and we conduct a sensitivity simulation using OH and O3 as described by Holmes et al.14 We ﬁnd that in the base simulation net deposition to midlatitude regions is similar, while deposition is lower in the tropics and higher in polar regions. This is consistent with recent ﬁndings from an intercomparison of six mercury models for the Task Force on Hemispheric Transport of Air Pollution.31 Diﬀerences in modeled deposition are greatest where measurements are sparsest. Additional long-term monitoring stations in the Arctic and tropics would help constrain the atmospheric oxidant of Hg0. The source attribution of regional deposition remains essentially unchanged because deposition to a receptor region is most inﬂuenced by sources in the same broad latitudinal bands. Another issue is the fate of mercury in the surface reservoirs following deposition. Isotopic observations place constraints on the extent of fast recycling of mercury deposited to land,4446 but additional study is needed to characterize diﬀerences across multiple ecosystem types. The fraction of mercury deposited to oceans that is re-emitted to the atmosphere (40% in our standard simulation) depends on redox kinetics in the surface ocean and the size of the reducible HgII pool. Although redox kinetics for characterizing the net reduction of HgII to Hg0 in the surface ocean represent a major uncertainty,47 our simulation uses rate constants constrained by experimental data using stable Hg isotopes.48 The size of the reducible pool is highly uncertain and depends on partitioning to particulate organic carbon as well as formation of stable inorganic complexes in solution that are resistant to reduction.48 In our model parametrization 40% of

ARTICLE

dissolved HgII is available for reduction.15 This is the lower bound from measurements in freshwater ecosystems (4060%);48,49 however, no data are available for marine ecosystems. Ocean re-emissions increase or decline proportionally to the reducible HgII pool size. To address this uncertainty, evasion rates for the standard simulation have has been optimized to best match observational constraints for atmospheric and seawater Hg0 concentrations.15 We model HgII partitioning to particles and removal from the surface ocean based on variability in biological productivity and ocean export ﬂuxes. Modeled airsea exchange is also sensitive ((30%) to the evasion scheme employed.5052 Though the magnitude of evasion varies across schemes, the fraction of mercury from the subsurface ocean vs atmospheric deposition is unaﬀected, so source attribution is unchanged. Additional study of the redox kinetics of diﬀerent HgII complexes in marine waters as well as coupled cycling in association with organic carbon in the global oceans would improve our understanding of the lifetime of Hg in actively cycling reservoirs and the time scales for sequestering anthropogenic mercury in deep ocean reservoirs. Implications for Policy. Separation of source contributions to mercury deposition between a local component from emitted HgII and a background component from emitted Hg0 allows a simplified estimate of sourcereceptor relationships on continental scales. We used GEOS-Chem for the present-day and 2050 simulations to construct a best-fit linear regression model relating net deposition fluxes in a region i (Di) to the regional emission of HgII (Ei HgII) and to the global emission of Hg0 (EHg0). We find the following form (r2 = 0.91, Supporting Information, Figure 7) Di ¼ 0:39EiHgII þ 0:74EHg0 þ 10

ð2Þ

where all values are in μg m2 a1. The 0.39 coefficient for Ei HgII represents the average fraction of regional HgII emissions that deposits within the region and is not quickly re-emitted. The intercept of 10 μg m2 a1 represents the mean deposition from natural and legacy terrestrial sources. This linear regression assumes that all Hg0 emitted worldwide is equally efficient in contributing to deposition in a given receptor region, and this is not correct (see Figure 4 and related discussion). The simple regression equation still performs well in most regions, with a mean residual of 4 μg m2 a1. Supporting Information Figure 6 shows the major exporters of anthropogenic mercury by region. Humans are exposed to mercury through commercial ﬁsh caught in oceans worldwide.9 A combination of both decreases in deposition to local ecosystems and global oceans is therefore needed to most eﬀectively reduce exposures and risks. Asia presently contributes more than one-half of new anthropogenic deposition to all ocean basins (from 53% to the North Atlantic to 62% to the North Paciﬁc) because it represents such a large global source; its contribution is expected to further grow in the future. North American and European sources contribute 30% of new anthropogenic deposition to the North Atlantic and less in other ocean basins. However, two-thirds of present-day deposition to the ocean is from natural and legacy sources, and much of the legacy anthropogenic mercury is due to North American and European emissions from the past two centuries.25 Present-day primary anthropogenic emissions contribute only about one-third of global mercury deposition, and this has been used to argue that future emission controls would have relatively little impact. This perspective is ﬂawed in that it does not 10482

dx.doi.org/10.1021/es202496y |Environ. Sci. Technol. 2011, 45, 10477–10484

Environmental Science & Technology recognize that future emissions also increase the mercury stored in legacy pools. On time scales of decades to centuries, the legacy mercury presently in organic soils and subsurface ocean waters will enter more geochemically stable reservoirs in deep ocean sediments and recalcitrant soil pools.16,20 Thus, mercury currently in the legacy pools will decline over time unless new emissions restore it. The beneﬁt of decreasing primary anthropogenic emissions must therefore factor in the resulting decrease in re-emission of mercury from legacy pools. This is similar to the CO2 problem in that emitted CO2 has an atmospheric lifetime of only 5 years against uptake by the ocean and land but is reemitted multiple times from these surface reservoirs. The eﬀective legacy of emitted CO2 (expressed by the IPCC as global warming potential) is more than a century.53 In the same way, the eﬀect of anthropogenic mercury emissions should be viewed in terms of their long-term legacy. This calls for better understanding of the time scales associated with mercury in legacy pools and its transfer to geochemically stable reservoirs.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional information, including more information on emissions scenarios, sensitivity analysis, and additional ﬁgures. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: (617) 384-7835; e-mail: [email protected].

’ ACKNOWLEDGMENT We acknowledge ﬁnancial support for this study from the NSF Atmospheric Chemistry Program, the EPA STAR program, and Electric Power Research Institute (EPRI). E.S.C. acknowledges support from the NSF graduate fellowship program. E.M.S. acknowledges new investigator support from the HSPH-NIEHS Center for Environmental Health. We thank Helen Amos for helpful discussions. ’ REFERENCES (1) Lamborg, C. H.; Fitzgerald, W. F.; Damman, A. W. H.; Benoit, J. M.; Balcom, P. H.; Engstrom, D. R. Modern and historic atmospheric mercury ﬂuxes in both hemispheres: Global and regional mercury cycling implications. Global Biogeochem. Cycles 2002, 16 (4). (2) Fitzgerald, W. F.; Engstrom, D. R.; Mason, R. P.; Nater, E. A. The case for atmospheric mercury contamination in remote areas. Environ. Sci. Technol. 1998, 32 (1), 1. (3) Schuster, P. F.; Krabbenhoft, D. P.; Naftz, D. L.; Cecil, L. D.; Olson, M. L.; Dewild, J. F.; Susong, D. D.; Green, J. R.; Abbott, M. L. Atmospheric mercury deposition during the last 270 years: A glacial ice core record of natural and anthropogenic sources. Environ. Sci. Technol. 2002, 36 (11), 2303. (4) Roos-Barraclough, F.; Martinez-Cortizas, A.; Garcia-Rodeja, E.; Shotyk, W. A 14 500 year record of the accumulation of atmospheric mercury in peat: volcanic signals, anthropogenic inﬂuences and a correlation to bromine accumulation. Earth Planet. Sci. Lett. 2002, 202 (2), 435. (5) Fitzgerald, W. F.; Engstrom, D. R.; Lamborg, C. H.; Tseng, C. M.; Balcom, P. H.; Hammerschmidt, C. R. Modern and historic atmospheric mercury ﬂuxes in northern Alaska: Global sources and Arctic depletion. Environ. Sci. Technol. 2005, 39 (2), 557.

ARTICLE

(6) Hermanson, M. H. Historical accumulation of atmospherically derived pollutant trace-metals in the Arctic as measured in dated sediment cores. Water Sci. Technol. 1993, 28 (89), 33. (7) In Ecotoxicology of mercury, 2nd ed.; Wiener, J. G., Krabbenhoft, D. P., Heinz, G. H., Scheuhammer, A. M., Eds.; CRC Press: Boca Raton, FL, 2003. (8) Clarkson, T. W.; Magos, L. The toxicology of mercury and its chemical compounds. Crit. Rev. Toxicol. 2006, 36 (8), 609. (9) Sunderland, E. M. Mercury exposure from domestic and imported estuarine and marine ﬁsh in the US seafood market. Environ. Health Perspect. 2007, 115 (2), 235. (10) Streets, D. G.; Zhang, Q.; Wu, Y. Projections of Global Mercury Emissions in 2050. Environ. Sci. Technol. 2009, 43 (8), 2983. (11) Strivastava, R.. Control of mercury emissions from coal-ﬁred electric utility boilers: An update. U.S. Environmental Protection Agency: Research Triangle Park, NC, 2010. (12) In Special Report on Emissions Scenarios: A Special Report of Working Group III of the Intergovernmental Panel on Climate Change; Nakicenovic, N., Alcamo, J., Davis, G., Eds.; Cambridge University Press: Cambridge, U.K., 2000. (13) Selin, N. E.; Jacob, D. J.; Yantosca, R. M.; Strode, S.; Jaegle, L.; Sunderland, E. M. Global 3-D land-ocean-atmosphere model for mercury: present-day versus preindustrial cycles and anthropogenic enrichment factors for deposition. Global Biogeochem. Cycles 2008, 22 (2), GB2011. (14) Holmes, C. D.; Jacob, D. J.; Corbitt, E. S.; Mao, J.; Yang, X.; Talbot, R.; Slemr, R. Global atmospheric model for mercury including oxidation by bromine atoms. Atmos. Chem. Phys. 2010, 10 (24), 12037. (15) Soerensen, A. L.; Sunderland, E. M.; Holmes, C. D.; Jacob, D. J.; Yantosca, R. M.; Skov, H.; Christensen, J. H.; Strode, S. A.; Mason, R. P. An Improved Global Model for Air-Sea Exchange of Mercury: High Concentrations over the North Atlantic. Environ. Sci. Technol. 2010, 44 (22), 8574. (16) Sunderland, E. M.; Mason, R. P. Human impacts on open ocean mercury concentrations. Global Biogeochem. Cycles 2007, 21 (4), 15. (17) Sunderland, E. M.; Krabbenhoft, D. P.; Moreau, J. W.; Strode, S. A.; Landing, W. M. Mercury sources, distribution, and bioavailability in the North Paciﬁc Ocean: Insights from data and models. Global Biogeochem. Cycles 2009, 23. (18) Fitzgerald, W. F.; Lamborg, C. H.; Hammerschmidt, C. R. Marine biogeochemical cycling of mercury. Chem. Rev. 2007, 107 (2), 641. (19) Holmes, C. D.; Jacob, D. J.; Mason, R. P.; Jaﬀe, D. A. Sources and deposition of reactive gaseous mercury in the marine atmosphere. Atmos. Environ. 2009, 43 (14), 2278. (20) Smith-Downey, N. V.; Sunderland, E. M.; Jacob, D. J. Anthropogenic impacts on global storage and emissions of mercury from terrestrial soils: Insights from a new global model. J. Geophys. Res.Biogeosci. 2010, 115. (21) Pirrone, N.; Cinnirrella, S.; Feng, X.; Friedli, H.; Levin, L.; Pacyna, J.; Pacyna, E. G.; Streets, D. G.; Sundseth, K. Mercury: Emissions. HTAP 2010 Assessment Report, 2010. (22) Pacyna, E. G.; Pacyna, J. M.; Sundseth, K.; Munthe, J.; Kindbom, K.; Wilson, S.; Steenhuisen, F.; Maxson, P. Global emission of mercury to the atmosphere from anthropogenic sources in 2005 and projections to 2020. Atmos. Environ. 2010, 44 (20), 2487. (23) Pacyna, E. G.; Pacyna, J. M.; Fudala, J.; Strzelecka-Jastrzab, E.; Hlawiczka, S.; Panasiuk, D. Mercury emissions to the atmosphere from anthropogenic sources in Europe in 2000 and their scenarios until 2020. Sci. Total Environ. 2006, 370 (1), 147. (24) Jacob, D. J.; Prather, M. J.; Wofsy, S. C.; McElroy, M. B. Atmospheric distribution of Kr-85 simulated with a general-circulation model. J. Geophys. Res.-Atmos. 1987, 92 (D6), 6614. (25) Streets, D. G.; Devane, M. K.; Lu, Z.; Bond, T. C.; Sunderland, E. M.; Jacob, D. J. All-time releases of mercury to the atmosphere from human activities. Submitted for publication. (26) Pirrone, N.; Allegrini, I.; Keeler, G. J.; Nriagu, J. O.; Rossmann, R.; Robbins, J. A. Historical atmospheric mercury emissions and 10483

dx.doi.org/10.1021/es202496y |Environ. Sci. Technol. 2011, 45, 10477–10484

Environmental Science & Technology depositions in North America compared to mercury accumulations in sedimentary records. Atmos. Environ. 1998, 32 (5), 929. (27) Selin, N. E.; Jacob, D. J. Seasonal and spatial patterns of mercury wet deposition in the United States: Constraints on the contribution from North American anthropogenic sources. Atmos. Environ. 2008, 42 (21), 5193. (28) Guentzel, J. L.; Landing, W. M.; Gill, G. A.; Pollman, C. D. Processes inﬂuencing rainfall deposition of mercury in Florida. Environ. Sci. Technol. 2001, 35 (5), 863. (29) Jaﬀe, D.; Strode, S. Sources, fate and transport of atmospheric mercury from Asia. Environ. Chem. 2008, 5 (2), 121. (30) Lin, C. J.; Pan, L.; Streets, D. G.; Shetty, S. K.; Jang, C.; Feng, X.; Chu, H. W.; Ho, T. C. Estimating mercury emission outﬂow from East Asia using CMAQ-Hg. Atmos. Chem. Phys. 2010, 10 (4), 1853. (31) Travnikov, O.; Lin, C.-J.; Dastoor, A.; Bullock, O. R.; Hedgecock, I. M.; Holmes, C. D.; Ilyin, I.; Jaegle, L.; Jun, G.; Pan, L.; Pongrueksa, P.; Ryzhkov, A.; Seigneur, C.; Skov, H. Mercury: Global and Regional Modeling; HTAP 2010 Assessment Report, 2010. (32) Lin, C. J.; Pongprueksa, P.; Lindberg, S. E.; Pehkonen, S. O.; Byun, D.; Jang, C. Scientiﬁc uncertainties in atmospheric mercury models I: Model science evaluation. Atmos. Environ. 2006, 40 (16), 2911. (33) Ryaboshapko, A.; Bullock, O. R.; Christensen, J.; Cohen, M.; Dastoor, A.; Ilyin, I.; Petersen, G.; Syrakov, D.; Travnikov, O.; Artz, R. S.; Davignon, D.; Draxler, R. R.; Munthe, J.; Pacyna, J. Intercomparison study of atmospheric mercury models: 2. Modelling results vs. long-term observations and comparison of country deposition budgets. Sci. Total Environ. 2007, 377 (23), 319. (34) Bullock, O. R.; Atkinson, D.; Braverman, T.; Civerolo, K.; Dastoor, A.; Davignon, D.; Ku, J. Y.; Lohman, K.; Myers, T. C.; Park, R. J.; Seigneur, C.; Selin, N. E.; Sistla, G.; Vijayaraghavan, K. An analysis of simulated wet deposition of mercury from the North American Mercury Model Intercomparison Study. J. Geophy. Res.-Atmos. 2009, 114, 12. (35) Selin, N. E.; Jacob, D. J.; Park, R. J.; Yantosca, R. M.; Strode, S.; Jaegle, L.; Jaﬀe, D. Chemical cycling and deposition of atmospheric mercury: Global constraints from observations. J. Geophys. Res.-Atmos. 2007, 112 (D2), 14. (36) Lohman, K.; Seigneur, C.; Edgerton, E.; Jansen, J. Modeling mercury in power plant plumes. Environ. Sci. Technol. 2006, 40 (12), 3848. (37) Seigneur, C.; Karamchandani, P.; Vijayaraghavan, K.; Lohman, K.; Shia, R. L.; Levin, L. On the eﬀect of spatial resolution on atmospheric mercury modeling. Sci. Total Environ. 2003, 304 (13), 73. (38) Ter Schur, A.; Caﬀrey, J.; Gustin, M.; Holmes, C.; Hynes, A.; Landing, B.; Landis, M.; Laudel, D.; Levin, L.; Nair, U.; Jansen, J.; Ryan, J.; Walters, J.; Schauer, J.; Volkamer, R.; Waters, D.; Weiss-Penzias, P. An integrated approach to assess elevated mercury wet deposition and concentrations in the southeastern United States. 10th International Conference on Mercury as a Global Pollutant; Halifax, NS, Canada, 2011. (39) Edgerton, E. S.; Hartsell, B. E.; Jansen, J. J. Mercury speciation in coal-ﬁred power plant plumes observed at three surface sites in the southeastern US. Environ. Sci. Technol. 2006, 40 (15), 4563. (40) Slemr, F.; Schuster, G.; Seiler, W. Distribution, speciation, and budget of atmospheric mercury. J. Atmos. Chem. 1985, 3 (4), 407. (41) Lin, C. J.; Pehkonen, S. O. The chemistry of atmospheric mercury: a review. Atmos. Environ. 1999, 33 (13), 2067. (42) Dastoor, A. P.; Larocque, Y. Global circulation of atmospheric mercury: a modelling study. Atmos. Environ. 2004, 38 (1), 147. (43) Seigneur, C.; Lohman, K. Eﬀect of bromine chemistry on the atmospheric mercury cycle. J. Geophys. Res.-Atmos. 2008, 113 (D23). (44) Hintelmann, H.; Harris, R.; Heyes, A.; Hurley, J. P.; Kelly, C. A.; Krabbenhoft, D. P.; Lindberg, S.; Rudd, J. W. M.; Scott, K. J.; St Louis, V. L. Reactivity and mobility of new and old mercury deposition in a Boreal forest ecosystem during the ﬁrst year of the METAALICUS study. Environ. Sci. Technol. 2002, 36 (23), 5034. (45) Graydon, J. A.; St Louis, V. L.; Lindberg, S. E.; Hintelmann, H.; Krabbenhoft, D. P. Investigation of mercury exchange between forest canopy vegetation and the atmosphere using a new dynamic chamber. Environ. Sci. Technol. 2006, 40 (15), 4680.

ARTICLE

(46) Graydon, J. W.; Zhang, X. Z.; Kirk, D. W.; Jia, C. Q. Sorption and stability of mercury on activated carbon for emission control. J. Hazard. Mater. 2009, 168 (23), 978. (47) Qureshi, A.; MacLeod, M.; Hungerbuhler, K. Quantifying uncertainties in the global mass balance of mercury. Global Biogeochem, Cycles, in press. (48) Whalin, L.; Kim, E. H.; Mason, R. Factors inﬂuencing the oxidation, reduction, methylation and demethylation of mercury species in coastal waters. Mar. Chem. 2007, 107 (3), 278. (49) O’Driscoll, N. J.; Siciliano, S. D.; Lean, D. R. S.; Amyot, M. Gross photoreduction kinetics of mercury in temperate freshwater lakes and rivers: Application to a general model of DGM dynamics. Environ. Sci. Technol. 2006, 40 (3), 837. (50) Rolfhus, K. R.; Fitzgerald, W. F. Mechanisms and temporal variability of dissolved gaseous mercury production in coastal seawater. Mar. Chem. 2004, 90 (14), 125. (51) Sunderland, E. M.; Dalziel, J.; Heyes, A.; Branﬁreun, B. A.; Krabbenhoft, D. P.; Gobas, F. Response of a Macrotidal Estuary to Changes in Anthropogenic Mercury Loading between 1850 and 2000. Environ. Sci. Technol. 2010, 44 (5), 1698. (52) Andersson, M. E.; Gardfeldt, K.; Wangberg, I.; Sprovieri, F.; Pirrone, N.; Lindqvist, O. Seasonal and daily variation of mercury evasion at coastal and oﬀ shore sites from the Mediterranean Sea. Mar. Chem. 2007, 104 (34), 214. (53) Solomon, S. D.; Qin, M.; Manning, Z.; Chen, M.; Marquis, K. B.; Averyt, M. T.; Miller, H. L. IPCC Fourth Assessment Report (AR4) Climate Change 2007: The Physical Science Basis; Cambridge University Press: Cambridge, New York, 2007.

10484

dx.doi.org/10.1021/es202496y |Environ. Sci. Technol. 2011, 45, 10477–10484

ARTICLE pubs.acs.org/est

All-Time Releases of Mercury to the Atmosphere from Human Activities David G. Streets,†,* Molly K. Devane,† Zifeng Lu,† Tami C. Bond,‡ Elsie M. Sunderland,§ and Daniel J. Jacob|| †

Decision and Information Sciences Division, Argonne National Laboratory, Argonne, Illinois, United States Department of Civil & Environmental Engineering, University of Illinois at UrbanaChampaign, Urbana, Illinois, United States § Department of Environmental Health, Harvard School of Public Health, Boston, Massachusetts, United States School of Engineering and Applied Science, Harvard University, Cambridge, Massachusetts, United States

)

‡

bS Supporting Information ABSTRACT: Understanding the biogeochemical cycling of mercury is critical for explaining the presence of mercury in remote regions of the world, such as the Arctic and the Himalayas, as well as local concentrations. While we have good knowledge of present-day ﬂuxes of mercury to the atmosphere, we have little knowledge of what emission levels were like in the past. Here we develop a trend of anthropogenic emissions of mercury to the atmosphere from 1850 to 2008— for which relatively complete data are available—and supplement that trend with an estimate of anthropogenic emissions prior to 1850. Global mercury emissions peaked in 1890 at 2600 Mg yr1, fell to 700800 Mg yr1 in the interwar years, then rose steadily after 1950 to present-day levels of 2000 Mg yr1. Our estimate for total mercury emissions from human activities over all time is 350 Gg, of which 39% was emitted before 1850 and 61% after 1850. Using an eight-compartment global box-model of mercury biogeochemical cycling, we show that these emission trends successfully reproduce present-day atmospheric enrichment in mercury.

’ INTRODUCTION Mercury (Hg) is a natural element found everywhere in the Earth’s lithosphere. It is emitted naturally from the lithosphere to the atmosphere as gaseous elemental Hg0 through processes of erosion and volcanism. The atmospheric lifetime of Hg0 against deposition is ∼1 year, allowing transport on global scale.14 Deposited Hg cycles through the surface environment in oceans and soils, can be re-emitted to the atmosphere, and is eventually buried in ocean sediments or stable terrestrial reservoirs. Human activity, including fossil-fuel combustion, mining, and industrial production, has greatly augmented the natural ﬂux of Hg from the lithosphere to the atmosphere and from there to the surface environment. Present-day anthropogenic emissions are estimated to be about 2000 Mg yr1, as compared to 500 Mg yr1 for the natural geogenic emissions.59 The resulting accumulation of Hg in the environment has been documented from analysis of deposits in snow,4,10 ice cores,11 lake sediments,1215 and peat cores.16,17 Humans have disrupted the natural Hg cycle throughout their history in pursuit of riches, useful metals, and energy. It was 5000 years ago when humans ﬁrst began digging into the Earth’s crust to extract gold, silver, copper, coal, and other materials, all of which came laced with Hg. Since those ancient days, ever-increasing amounts of ever-varying materials have been extracted and reﬁned, and large quantities of Hg have been liberated in the process. The time scale for return of anthropogenic Hg to sediments has been estimated to be ∼2000 years.18,19 The accumulation of Hg in the r 2011 American Chemical Society

global environment thus represents the legacy of historical emissions, with continuous augmentation from present-day human activities. Balancing the sources, sinks, and ﬂuxes of Hg in diﬀerent parts of the world at diﬀerent times remains a modeling challenge, however,1825 and lack of knowledge of historical man-made contributions has inhibited progress.

’ DATA AND METHODS The purpose of this present work is to develop an estimate of the total amount of Hg released to the atmosphere due to human activities from the dawn of civilization until 2008, the most recent year for which activity data are available. For the period 18502008, we present detailed decadal trends for each major source type and each region of the world. Fourteen source types are included: copper, zinc, and lead smelting; artisanal and large-scale gold production; iron and steel manufacturing (separately); silver production; mercury production; cement manufacturing; caustic soda manufacturing (at chlor-alkali plants); and the combustion of coal, oil, and waste. We do not include emissions from biofuel or open biomass burning, as these represent a recycling of Hg previously deposited in the surface environment. Emissions are calculated for Received: August 8, 2011 Accepted: November 9, 2011 Revised: October 20, 2011 Published: November 09, 2011 10485

dx.doi.org/10.1021/es202765m | Environ. Sci. Technol. 2011, 45, 10485–10491

Environmental Science & Technology

ARTICLE

Figure 1. Time development of Hg emission factors for Cu smelters in ﬁve world regions. Each world region is comprised of countries with similar levels of technology development, ranging from most developed (Region 1) to least developed (Region 5). The composition of these regions is provided in Table S1 of the SI.

17 world regions (deﬁned in Table S1 of the Supporting Information (SI)). The method integrates our previous work on historical and future emissions2628 with our work on Hg emissions.9,2931 For the 11 industrial commodities, the assembly of historical production levels between 1850 and 2008 is documented in the SI. For the three combustion-related emissions, the activity levels are the same as used in previous work,2628 disaggregated into 144 sector/fuel/technology combinations (see Table 3 of ref 28.). A major challenge is to develop a representation of the timevarying Hg emission factors associated with each industrial activity. We have developed a dynamic representation of emission factors since 1850 to reﬂect the transition from old, smallscale, uncontrolled processes to modern, large-scale processes with emission controls. We know that materials production has grown year-by-year—sometimes dramatically—but at the same time process technology has improved and pollution controls have been adopted. So the resulting level of emissions at any given time reduces to a competition between production growth and technology improvement. We use the following transformed normal distribution function to estimate the variation of Hg emission factors over time: yr, p, t ¼ ðar, p br, p Þeð t

2

=2sr, p 2 Þ

þ br, p

where yr,p,t = emission factor in region r for process p in year t (g Mg1); ar,p = pre-1850 emission factor (g Mg1) in region r for process p; br,p = best emission factor achieved in region r for process p today (g Mg1); and sr,p = shape parameter of the curve for region r and process p. The use of such sigmoid curves to simulate the dynamics of technology change has been previously applied to energy and emission control technology,32 carbon sequestration,33 and

Figure 2. Trends in Hg emissions by (a) source type and (b) world region.

automobile technology.34 We have previously demonstrated the use of this technique in estimating both historical26 and future28 emissions. By selecting values of the parameters a, b, and s to correspond to the known or inferred time development pathway of relevant technologies, we can estimate the values of emission factor y at any point in time. Parameter values where this technique was used are provided in Table S2 of the SI. Figure 1 illustrates the use of this technique for estimating Hg emission factors for copper smelters. Note that we use ﬁve world regions to represent diﬀerent emission factor trajectories. Each world region is comprised of countries with similar levels of technology development, ranging from most developed (Region 1) to least developed (Region 5). The composition of these regions is provided in Table S1 of the SI. Emission factors for each world region were determined from a review of reported emission rates in representative countries6,8,29,3538 and used to anchor each trajectory. Table S3 of the SI presents and documents emission factor ranges for each industrial activity for 1850, 1930, and 2008 and provides citations for the studies used in the development of the emission factor curves. The procedure for combustion sources is diﬀerent and follows previous work26,28 in that unique emission factors are developed for each of the 144 sector/fuel/technology combinations, and transitions from simple to advanced systems are determined by technology shifts.

’ RESULTS AND DISCUSSION Historical anthropogenic Hg emission trends from 1850 to 2008 are shown in Figure 2, disaggregated by source type (upper) 10486

dx.doi.org/10.1021/es202765m |Environ. Sci. Technol. 2011, 45, 10485–10491

Environmental Science & Technology

ARTICLE

and world region (lower); here we aggregate emissions into seven super-regions (deﬁned in Table S1 of the SI). We calculate that global Hg emissions peaked in 1890 at 2600 Mg yr1. This was mainly due to extensive production of gold (Au) and silver (Ag) and production of the Hg that was used to extract them via amalgamation; our results for this period are in good agreement with previous work.39,40 Emissions then declined rapidly and remained relatively constant at 700800 Mg yr1 in the interwar years. After 1950, emissions grew again, driven mainly by growth in coal combustion and, lately, artisanal Au production, rising to 2000 Mg yr1 in 2008. Present-day values are consistent with previous work,69 as shown in Table S4 of the SI. The regional emission trends show that North America and Europe were the dominant emitting regions in the 19th century, but emphasis shifted initially to

Russia and then sharply to Asia after 1950. We estimate that Asia was responsible for 64% of global Hg emissions in 2008. We follow the methodology used previously for Hg emissions from power plants in China31 to determine uncertainties in the emission estimates. All input parameters and their corresponding probability distributions were incorporated into a Monte Carlo framework with Crystal Ball software and 10 000 simulations performed. For activity rates we assumed normal distributions with time-, region-, and sector-dependent uncertainties, based on previous work.26,30 Measurements of emission factors, if they existed, were applied directly in our model; and, if measurements were not available, probability distributions were based on expert judgment. For the emission estimates that follow, we present lower and upper bounds around the central estimate that correspond to an 80% conﬁdence interval (CI), meaning that, based on the underlying distributions, the probability of emissions being outside the stated range is less than 20%. Figure S1 of the SI shows the uncertainty range of the historical global Hg emissions. The range of uncertainty is large before 1910 (lower bounds varying between 30% and 45% and upper bounds varying between +65% and +110%, depending on the year), decreasing to a relatively stable level after 1920 (30% to +60%). Our modern-day uncertainty range is similar to estimates given by Pacyna et al.8 SI Figure S1 also shows that the largest contributors to total variance are the emission factor for Au/Ag production before 1940 and the emission factor for coal combustion after 1940. Because of the large contribution of the 19th century peak and its relatively high degree of uncertainty, we have investigated the situation in North America (the “Gold Rush”) in detail. The red line in Figure 3 shows our estimate of the amount of Hg needed to extract the reported amounts of Au and Ag produced, assuming that amalgamation began to be phased out after 1880 in favor of cyanidation. We also show the range of uncertainty (blue) caused by uncertainties in the Au/Ag production data and the implied Hg emission factor. Because considerable amounts of Hg were exported from Spain, Italy, and Slovenia to Mexico, we know that the amount of Hg available lies somewhere in the range (yellow) bounded by production in North America (N.A.) and the

Figure 3. Estimates of Hg consumed in North America in the 19th century to extract Au/Ag.

Table 1. Cumulative Commodity Production Amounts and Associated Hg Emissionsa pre-1850 material

emissions (Mg)

production (Tg)

all-time to 2008

emissions (Mg)

production (Tg)

emissions (Mg)

copper

45

1240

547

3410

592

4650

zinc

50

3750

403

6520

453

10 300

lead

55

2400

268

3590

323

5990

iron

330

20

32 900

1110

33 300

1130

∼0 0.229

∼0 41 700

44 200 0.720

388 53 300

44200 0.949

388 95 000 20 600

steel mercury

∼0

∼0

0.145

20 600

0.145

gold, artisanal

0.016

8200

0.025

10 200

0.041

18 400

silver

0.276

78 700

1.17

67 300

1.45

146 000 3000

gold, large-scale

cement

∼0

∼0

61 200

3000

61 200

caustic soda

∼0

∼0

1710

4240

1710

4240

coal

2900

868

319 000

33 900

322 000

34 800

oil waste

∼0 ∼0

∼0 ∼0

312 000 2310

2620 5310

312 000 2310

2620 5310

total a

production (Tg)

18502008

137 000

215 000

352 000

Values are rounded to no more than three signiﬁcant digits, consistent with the level of conﬁdence. 10487

dx.doi.org/10.1021/es202765m |Environ. Sci. Technol. 2011, 45, 10485–10491

Environmental Science & Technology combined production in N.A. and Europe. Our trend line lies for the most part within the overlapping range (green) of uncertainties in Hg availability and Hg requirement. Actual Hg consumption data for N.A. are not known with conﬁdence, but we show annual consumption values reported for N.A. and for N.A. minus the amounts of Hg that were used for purposes other than amalgamation in the U.S.41 (These other uses are not known for Canada and Mexico.) We believe that the consumption statistics are too low after 1865 compared to both production values and the inferred requirements for amalgamation; we attribute this to underestimation or even omission of imported Hg from Europe. Sources of uncertainty in these historical data are: other uses of mercury (medicine and paint), stockpiling of cheap Hg for later use when prices rise, and timing of the replacement of amalgamation with cyanidation in Mexico and Canada. Another issue is the amount of Hg lost per kg of Au/Ag produced, which we set at 1.3 kg. Previous recommendations for this ratio have ranged between 0.75 and 1.5,40,42 and Nriagu43 suggested a range of values from 0.854.1, depending on the richness of the ore (the richer the ore, the more Hg need for extraction). It is possible that in the early periods of Au/Ag mining in North America, richer ores were mined and more Hg was used. To test this possibility, we calculated the ratios of apparent Hg consumption to the amount of Hg required. Values are: 4.02 (1850s), 1.37 (1860s), 1.00 (1870s), 0.89 (1880s), 0.88 (1890s), and 1.15 (1900s). These values are thus consistent with declining ore richness.43 Our emission estimates for this period are in good agreement with previous estimates,39,40 and therefore we have conﬁdence in the size and timing of the 19th century peak in Hg emissions. Figure S2 of the SI shows the full trend of Hg emissions in North America. Integrating under the trend curve (Figure 2), we calculate that a cumulative total of 215 (34% to +74%) Gg of Hg were released to the atmosphere from human activities between 1850 and 2008. Table 1 shows that emissions during this period were dominated by Ag production (31%), Hg production (25%), and coal combustion (16%). The dominant regions were North America (32%), Europe (24%), and Asia (13%). We also conducted a review of materials production and fuel combustion for the period from when humans ﬁrst began to extract metals (ca. 3000 BC) until 1850. There are good estimates for the large contributions of Ag mining in South and Central America in the 16th18th centuries.4246 We also consulted a wide variety of other historical sources (see the SI). We estimate that about 137 Gg of Hg were released prior to 1850. It is not possible to do a formal uncertainty analysis for this value, but we believe conﬁdence is on the order of 50% to +300%. This pre-1850 amount of Hg is dominated by Ag production in Spanish America (58% of total) and production of the Hg needed to extract it (30%). Combining the pre-1850 and post-1850 values leads to the conclusion that cumulative, all-time releases of Hg to the atmosphere up to 2008 have been about 350 Gg. For anthropogenic emissions since 1850 we are able to develop speciated emission trends using speciation factors that we have documented in previous work.9,29,30 Figure 4 shows that the share of Hg0 in total Hg emissions has declined from 80% in 1850 to 55% today. Nevertheless, emissions of Hg0 have grown steadily in the modern era, from 420 Mg yr1 in 1950 to 1080 Mg yr1 in 2008, due to worldwide economic and industrial development. The emissions that we have calculated are direct releases, that is to say, injection of Hg directly into the atmosphere from thermal processes. In addition, there have been comparable quantities of anthropogenic Hg directly released from these

ARTICLE

Figure 4. Trends in speciated emissions of Hg: absolute magnitude (lines) and shares of total (shading).

same industrial activities as liquid and solid waste streams that have been deposited locally to land surfaces and water bodies. It is more diﬃcult to quantify these pools of Hg, but they may be amenable to estimation based on some of our results, such as the total commercial production of Hg over time, the amounts of raw ore and fuel processed, and the amounts of Hg captured by emission control devices. As one example, we estimate in this work that 720 Gg of Hg were produced commercially between 1850 and 2008 (Table 1), of which 340 Gg were used to extract Au/Ag. This implies that about 380 Gg of Hg have been used for other purposes—in other industrial processes or in the manufacture of products. This large pool of unaccounted Hg is likely sequestered in landﬁlls and other localized waste repositories. Nevertheless, it will eventually be released on time scales faster than the natural geogenic sources and therefore should be accounted for in biogeochemical models. Any changes to the availability of these nonatmospheric repositories of man-made Hg through activities such as dredging and waste combustion could liberate large quantities of Hg to the atmosphere and biosphere. Using the historical atmospheric emissions inventory developed here we are able to construct a temporally resolved simulation of how various biogeochemical reservoirs have changed in response to anthropogenically mobilized Hg. To do this, we developed an eightcompartment global box-model that is based on current estimates of global budgets.18,2224 The model includes three fast-cycling surface reservoirs (the atmosphere, the surface mixed-layer of the ocean, and a rapidly responding terrestrial compartment that includes vegetation, sea-ice, snowpacks, and labile organic carbon pools), two intermediate compartments that respond on the order of decades (intermediate ocean and slow terrestrial reservoir), two deep reservoirs that respond on the order of centuries (deep ocean and armored soil pool), and the large mineral reservoir of Hg. We assume that the mass ﬂows between compartments in this eight-box model are governed by ﬁrst-order processes having the same rate constants in the preindustrial period and the present. These rate constants are speciﬁed from global models of Hg cycling in the atmosphere,24 terrestrial systems,22 surface ocean,23 and intermediate and deep ocean.18 We ﬁrst run the model to steady state using a geogenic emissions value6 of 200 Mg yr1. We then conduct a time-dependent simulation between 1450 and 2008 using the historical emissions presented here. Biomass burning is speciﬁed 10488

dx.doi.org/10.1021/es202765m |Environ. Sci. Technol. 2011, 45, 10485–10491

Environmental Science & Technology

Figure 5. Modeled, time-dependent accumulation of anthropogenic Hg in global biogeochemical reservoirs.

as a constant value of 300 Mg yr1 between 1450 and the present. For anthropogenic emissions occurring before 1850, we divide the total releases (137 Gg) into a peak emissions period spanning 200 years (80% of total releases) to represent Ag mining in South and Central America between 1450 and 1650.43,46 We assume the remaining fraction of pre-1850 anthropogenic emissions (20%) was released at a constant rate between 1650 and 1850. Figure 5 shows the resulting modeled trends in the fast, intermediate, and slowly responding Hg reservoirs between 1450 and present. Results show that the present-day atmosphere has been enriched in Hg by a factor of 3.9 (range of 3.26.9 based on the low and high emission estimates shown in Figure S1 of the SI), which is in excellent agreement with archival records of deposition in ice and sediments.11,39 Fast-cycling terrestrial and ocean surface reservoirs closely track temporal trends in atmospheric concentrations and historical emissions, peaking in the early 1600s and 1800s due to Au and Ag mining and rising again after 1950 due to the large increases in coal combustion. Slow terrestrial and intermediate ocean Hg concentrations reﬂect a longer time history of anthropogenic inputs, because the lifetime of Hg in these reservoirs is on the order of decades (Figure 5). In slow terrestrial and subsurface ocean reservoirs it is possible to distinguish among peaks occurring in the 1600s due to Ag mining in South and Central America and the rise in emissions in 1850 to present day, though the local minimum in the early 1900s is not visible. Present-day enrichment of these reservoirs from anthropogenic sources is between 270280% relative to pre-1450 levels. This is higher than previous model

ARTICLE

estimates for the intermediate ocean18,25 that could not consider releases prior to 1850 and the fully resolved time history of anthropogenic emissions presented here. Considerable accumulation of historical anthropogenic Hg is evident in the deep ocean (>100 Gg since 1450) compared to <1 Gg in the armored soil pool because of the large sink of mercury from the surface and intermediate ocean through particle settling. Because of the long lifetime of Hg in the deep ocean (many centuries) and armored terrestrial pools, these reservoirs will continue to slowly rise over time, even with declines in future emissions. A recent analysis of worldwide trends in Hg concentrations in air and wet deposition47 showed a 2038% decline in atmospheric Hg concentrations from 1996 to 2009 at long-term monitoring stations in the Northern and Southern Hemisphere. A declining trend was also reported at the Mace Head baseline station in Ireland48 and at the Alert station in the Canadian Arctic,49 though the magnitude of the change at Alert was smaller. These measurement trends are in apparent conﬂict with the increasing trends in primary anthropogenic Hg0 emissions presented here (Figure 4). Slemr et al.47 postulated that the recent decline in atmospheric Hg could be due to declines in re-emission of Hg from surface ocean and soil reservoirs. Our geochemical box-model analysis shows a 19001950 decline of atmospheric concentrations following the late 19th century maximum in emissions, but a post-1950 increase driven by growing emissions (Figure 5). Although there is uncertainty regarding the time scales for re-emission and transfer to stable geochemical reservoirs, the trend of increasing anthropogenic emissions shown in Figure 2 is incompatible with a decline in Hg in the surface reservoirs. The observed atmospheric decline could possibly reﬂect a suppression of re-emission from the surface reservoirs or an increase in atmospheric oxidant concentrations, but there is no independent evidence for such eﬀects. One possible explanation for the observed atmospheric decline is the legacy of Hg in commercial products (batteries, thermometers, switches, etc.). Production of these Hg-containing products peaked in the late 20th century (ca. 1970) and has been declining since then. This Hg eventually enters the environment through incineration, wastewater, or leakage from landﬁlls and other solid waste repositories. Our inventory accounts for incineration but not the other modes of disposal. As pointed out above, the quantities involved are large, and there is little knowledge of how rapidly they could enter the environment and be emitted to the atmosphere. Better understanding is needed of the processes and time scales involved.

’ ASSOCIATED CONTENT

bS

Supporting Information. (1) Development of production data, 18502008; (2) Development of production data, pre-1850; (3) References used for production data; (4) Regional deﬁnitions (Table S1); (5) Parameter values used in the sigmoid function computation of time-varying Hg emissions factors (Table S2); (6) Hg emission factors for industrial processes (Table S3); (7) Comparison of Hg emission estimates for recent years (Table S4); (8) Uncertainty ranges (Figure S1); (9) Historical trend in North American Hg emissions (Figure S2). This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: (630) 252-3448; fax: (630) 252-6500; e-mail: dstreets@ anl.gov. 10489

dx.doi.org/10.1021/es202765m |Environ. Sci. Technol. 2011, 45, 10485–10491

Environmental Science & Technology

’ ACKNOWLEDGMENT Work at Argonne National Laboratory was supported by the U.S. Department of Energy. Argonne National Laboratory is operated by UChicago Argonne, LLC, under Contract No. DE-AC02-06CH11357 with the U.S. Department of Energy. Work at Harvard University was supported by the U.S. National Science Foundation. E.M.S. acknowledges new investigator support from the HSPH-NIEHS Center for Environmental Health. ’ REFERENCES (1) Fitzgerald, W. F.; Engstrom, D. R.; Mason, R. P.; Nater, E. A. The case for atmospheric mercury contamination in remote areas. Environ. Sci. Technol. 1998, 32, 1–7. (2) Durnford, D.; Dastoor, A.; Figueras-Nieto, D.; Ryjkov, A. Long range transport of mercury to the Arctic and across Canada. Atmos. Chem. Phys. 2010, 10, 6063–6086. (3) Temme, C.; Einax, J. W.; Ebinghaus, R.; Schroeder, W. H. Measurements of atmospheric mercury species at a coastal site in the Antarctic and over the South Atlantic Ocean during polar summer. Environ. Sci. Technol. 2003, 37, 22–31. (4) Loewen, M.; Kang, S.; Armstrong, D.; Zhang, Q.; Tomy, G.; Wang, F. Atmospheric transport of mercury to the Tibetan Plateau. Environ. Sci. Technol. 2007, 41, 7632–7638. (5) Mercury Fate and Transport in the Global Atmosphere; Pirrone, N., Mason, R., Eds.; Springer: New York, 2009. (6) Pirrone, N.; Cinnirella, S.; Feng, X.; Finkelman, R. B.; Friedli, H. R.; Leaner, J.; Mason, R.; Mukherjee, A. B.; Stracher, G. B.; Streets, D. G.; Telmer, K. Global mercury emissions to the atmosphere from anthropogenic and natural sources. Atmos. Chem. Phys. 2010, 10, 5951–5964. (7) Pacyna, E. G.; Pacyna, J. M.; Steenhuisen, F.; Wilson, S. Global anthropogenic mercury emission inventory for 2000. Atmos. Environ. 2006, 40, 4048–4063. (8) Pacyna, E. G.; Pacyna, J. M.; Sundseth, K.; Munthe, J.; Kindbom, K.; Wilson, S.; Steenhuisen, F.; Maxson, P. Global emission of mercury to the atmosphere from anthropogenic sources in 2005 and projections to 2020. Atmos. Environ. 2010, 44, 2487–2499. (9) Streets, D. G.; Zhang, Q.; Wu, Y. Projections of global mercury emissions in 2050. Environ. Sci. Technol. 2009, 43, 2983–2988. (10) Lahoutifard, N.; Sparling, M.; Lean, D. Total and methyl mercury patterns in Arctic snow during springtime at Resolute, Nunavut, Canada. Atmos. Environ. 2005, 39, 7597–7606. (11) Schuster, P. F.; Krabbenhoft, D. P.; Naftz, D. L.; Cecil, L. D.; Olson, M. L.; Dewild, J. F.; Susong, D. D.; Green, J. R.; Abbott, M. L. Atmospheric mercury deposition during the last 270 years: A glacial ice core record of natural and anthropogenic sources. Environ. Sci. Technol. 2002, 36, 2303–2310. (12) Phillips, V. J. A.; Louis, V. L., St.; Cooke, C. A.; Vinebrooke, R. D.; Hobbs, W. O. Increased mercury loadings to western Canadian Alpine lakes over the past 150 years. Environ. Sci. Technol. 2011, 45, 2042–2047. (13) Ribeiro Guevara, S.; Meili, M.; Rizzo, A.; Daga, R.; Arribere, M. Sediment records of highly variable mercury inputs to mountain lakes in Patagonia during the past millennium. Atmos. Chem. Phys. 2010, 10, 3443–3453. (14) Yang, H.; Battarbee, R. W.; Turner, S. D.; Rose, N. L.; Derwent, R. G.; Wu, G.; Yang, R. Historical reconstruction of mercury pollution across the Tibetan Plateau using lake sediments. Environ. Sci. Technol. 2010, 44, 2918–2924. (15) Mast, M. A.; Manthorne, D. J.; Roth, D. A. Historical deposition of mercury and selected trace elements to high-elevation National Parks in the western U.S. inferred from lake-sediment cores. Atmos. Environ. 2010, 44, 2577–2586. (16) Madsen, P. P. Peat bog records of atmospheric mercury deposition. Nature 1981, 293, 127–130. (17) Givelet, N.; Roos-Barraclough, F.; Shotyk, W. Predominant anthropogenic sources and rates of atmospheric mercury accumulation

ARTICLE

in southern Ontario recorded by peat cores from three bogs: Comparison with natural “background” values (past 8000 years). J. Environ. Monit. 2003, 5, 935–949. (18) Sunderland, E. M.; Mason, R. P. Human impacts on open ocean mercury concentrations. Global Biogeochem. Cycles 2007, 21, GB4022, [Online early access] DOI: 10.1029/2006GB002876. (19) Selin, N. E.; Jacob, D. J.; Yantosca, R. M.; Strode, S.; Jaegle, L.; Sunderland, E. M. Global 3-D land-ocean-atmosphere model for mercury: Present-day versus preindustrial cycles and anthropogenic enrichment factors for deposition. Global Biogeochem. Cycles 2008, 22, GB2011, [Online early access] DOI: 10.1029/2007GB003040. (20) Lamborg, C. H.; Fitzgerald, W. F.; Damman, A. W. H.; Benoit, J. M.; Balcom, P. H.; Engstrom, D. R. Modern and historic atmospheric mercury ﬂuxes in both hemispheres: Global and regional mercury cycling implications. Global Biogeochem. Cycles 2002, 16, 1104, [Online early access] DOI: 10.1029/2001GB001847. (21) Selin, N. E.; Jacob, D. J.; Park, R. J.; Yantosca, R. M.; Strode, S.; Jaegle, L.; Jaﬀe, D. Chemical cycling and deposition of atmospheric mercury: Global constraints from observations. J. Geophys. Res. 2007, 112, D02308, [Online early access] DOI: 10.1029/2006JD007450. (22) Smith-Downey, N. V.; Sunderland, E. M.; Jacob, D. J. Anthropogenic impacts on global storage and emissions of mercury from terrestrial soils: Insights from a new global model. J. Geophys. Res. 2010, 115, G03008, [Online early access] DOI: 10.1029JG001124. (23) Soerensen, A. L.; Sunderland, E. M.; Holmes, C. D.; Jacob, D. J.; Yantosca, R. M.; Skov, H.; Christensen, J. H.; Strode, S. A.; Mason, R. P. An improved global model for air-sea exchange of mercury: High concentrations over the North Atlantic. Environ. Sci. Technol. 2010, 44, 8574–8580. (24) Holmes, C. D.; Jacob, D. J.; Corbitt, E. S.; Mao, J.; Yang, X.; Talbot, R.; Slemr, F. Global atmospheric model for mercury including oxidation by bromine atoms. Atmos. Chem. Phys. 2010, 10, 12037–12057. (25) Strode, S.; Jaegle, L.; Emerson, S. Vertical transport of anthropogenic mercury in the ocean. Global Biogeochem. Cycles 2010, 24, GB4014, [Online early access] DOI: 10.1029/2009GB003728. (26) Bond, T. C.; Bhardwaj, E.; Dong, R.; Jogani, R.; Jung, S.; Roden, C.; Streets, D. G.; Trautmann, N. M. Historical emissions of black and organic carbon aerosol from energy-related combustion, 18502000. Global Biogeochem. Cycles 2007, 21, GB2018, [Online early access] DOI: 10.1029/ 2006GB002840. (27) Lamarque, J.-F.; Bond, T. C.; Eyring, V.; Granier, C.; Heil, A.; Klimont, Z.; Lee, D.; Liousse, C.; Mieville, A.; Owen, B.; Schultz, M. G.; Shindell, D.; Smith, S. J.; Stehfest, E.; Van Aardenne, J.; Cooper, O. R.; Kainuma, M.; Mahowald, N.; McConnell, J. R.; Naik, V.; Riahi, K.; van Vuuren, D. P. Historical (18502000) gridded anthropogenic and biomass burning emissions of reactive gases and aerosols: Methodology and application. Atmos. Chem. Phys. 2010, 10, 7017–7039. (28) Streets, D. G.; Bond, T. C.; Lee, T.; Jang, C. On the future of carbonaceous aerosol emissions. J. Geophys. Res. 2004, 109, D24212, [Online early access] DOI: 10.1029/2004JD004902. (29) Streets, D. G.; Hao, J.; Wu, Y.; Jiang, J.; Chan, M.; Tian, H.; Feng, X. Anthropogenic mercury emissions in China. Atmos. Environ. 2005, 39, 7789–7806. (30) Wu, Y.; Wang, S.; Streets, D. G.; Hao, J.; Chan, M.; Jiang, J. Trends in anthropogenic mercury emissions in China from 1995 to 2003. Environ. Sci. Technol. 2006, 40, 5312–5318. (31) Wu, Y.; Streets, D. G.; Wang, S. X.; Hao, J. M. Uncertainties in estimating mercury emissions from coal-ﬁred power plants in China. Atmos. Chem. Phys. 2010, 10, 2937–2947. (32) Gr€ubler, A.; Nakicenovic, N.; Victor, D. G. Dynamics of energy technologies and global change. Energ. Policy 1999, 27, 247–280. (33) Riahi, K.; Rubin, E. S.; Taylor, M. R.; Schrattenholzer, L.; Hounshell, D. Technological learning for carbon capture and sequestration technologies. Energy Econ. 2004, 26, 539–564. (34) Nakicenovic, N. The automotive road to technological change: Diﬀusion of the automobiles as a process of technological substitution. Technol. Forecast. Soc. 1986, 29, 309–340. 10490

dx.doi.org/10.1021/es202765m |Environ. Sci. Technol. 2011, 45, 10485–10491

Environmental Science & Technology

ARTICLE

(35) Mercury Study Report to Congress: An Inventory of Anthropogenic Mercury Emissions in the United States, Report EPA-452/R-97-004; U.S. Environmental Protection Agency: Washington, DC, 1997; Vol. II. (36) Assessment of Mercury Releases from the Russian Federation, Arctic Council Action Plan to Eliminate Pollution of the Arctic (ACAP); Danish Environmental Protection Agency: Copenhagen, Denmark, 2005. (37) Air Pollutant Emission Inventory Guidebook; European Environment Agency: Copenhagen, Denmark, 2009. (38) Hylander, L. D.; Herbert, R. B. Global emission and production of mercury during the pyrometallurgical extraction of nonferrous sulﬁde ores. Environ. Sci. Technol. 2008, 42, 5971–5977. (39) Pirrone, N.; Allegrini, I.; Keeler, G. J.; Nriagu, J. O.; Rossmann, R.; Robbins, J. A. Historical atmospheric mercury emissions and depositions in North America compared to mercury accumulations in sedimentary records. Atmos. Environ. 1998, 32, 929–940. (40) Strode, S.; Jaegle, L.; Selin, N. E. Impact of mercury emissions from historic gold and silver mining: Global modeling. Atmos. Environ. 2009, 43, 2012–2017. (41) Hylander, L. D.; Meili, M. The rise and fall of mercury: Converting a resource to refuse after 500 years of mining and pollution. Crit. Rev. Env. Sci. Technol. 2005, 35, 1–36. (42) Lacerda, L. D. Global mercury emissions from gold and silver mining. Water, Air, Soil Pollut. 1997, 97, 209–221. (43) Nriagu, J. O. Legacy of mercury pollution. Nature 1993, 363, 589. (44) Nriagu, J. O. Mercury pollution from the past mining of gold and silver in the Americas. Sci. Total Environ. 1994, 149, 167–181. (45) Camargo, J. A. Contribution of Spanish-American silver mines (15701820) to the present high mercury concentrations in the global environment: A review. Chemosphere 2002, 48, 51–57. (46) Cooke, C. A.; Balcom, P. H.; Biester, H.; Wolfe, A. P. Over three millennia of mercury pollution in the Peruvian Andes. Proc. Natl. Acad. Sci. U.S.A. 2009, 106, 8830–8834. (47) Slemr, F.; Brunke, E.-G.; Ebinghaus, R.; Kuss, J. Worldwide trend of atmospheric mercury since 1995. Atmos. Chem. Phys. 2011, 11, 4779–4787. (48) Ebinghaus, R.; Jennings, S. G.; Kock, H. H.; Derwent, R. G.; Manning, A. J.; Spain, T. G. Decreasing trends in total gaseous mercury observations in baseline air at Mace Head, Ireland from 1996 to 2009. Atmos. Environ. 2011, 45, 3475–3480. (49) Cole, A. S.; Steﬀen, A. Trends in long-term gaseous mercury observations in the Arctic and eﬀects of temperature and other atmospheric conditions. Atmos. Chem. Phys. 2010, 10, 4661–4672.

10491

dx.doi.org/10.1021/es202765m |Environ. Sci. Technol. 2011, 45, 10485–10491

ARTICLE pubs.acs.org/est

Evaluation of Hexavalent Chromium Extraction Method EPA Method 3060A for Soils Using XANES Spectroscopy Julien Malherbe,*,†,‡ Marie-Pierre Isaure,† Fabienne Seby,§ Russell P. Watson,‡ Pablo Rodriguez-Gonzalez,|| Paul E. Stutzman,^ Clay W. Davis,‡ Chiara Maurizio,# Nora Unceta,r John R. Sieber,‡ Stephen E. Long,‡ and Olivier F. X. Donard† †

)

Laboratoire de Chimie Analytique Bio-Inorganique et Environnement, IPREM, UMR CNRS 5254, Universite de Pau et des Pays de l’Adour, Helioparc Pau-Pyrenees, 2, avenue Pierre Angot, 64053 Pau Cedex 9, France ‡ Analytical Chemistry Division, Material Measurement Laboratory, National Institute of Standards and Technology, 100 Bureau Drive, Stop 8391, Gaithersburg, Maryland, 20899, United States § Ultra Traces Analyses Aquitaine (UT2A), Helioparc Pau-Pyrenees, 2, avenue Pierre Angot, 64053 Pau Cedex 9, France Departamento de Química Física y Analítica, Universidad Oviedo, c/Julian Clavería, 8. E-33006, Oviedo, Spain ^ Materials and Construction Research Division, Engineering Laboratory, National Institute of Standards and Technology, 100 Bureau Drive, Stop 8615, Gaithersburg, Maryland, 20899, United States # CNR-OGG c/o GILDA beamline, European Synchrotron Radiation Facility, 6 rue J. Horowitz, BP 220, 38043 Grenoble, France r Department of Analytical Chemistry, Faculty of Pharmacy, University of the Basque Country, Vitoria-Gasteiz, Spain

bS Supporting Information ABSTRACT: Hexavalent chromium (Cr(VI)) occurrence in soils is generally determined using an extraction step to transfer it to the liquid phase where it is more easily detected and quantiﬁed. In this work, the performance of the most common extraction procedure (EPA Method 3060A) using NaOHNa2CO3 solutions is evaluated using X-ray absorption near edge structure spectroscopy (XANES), which enables the quantiﬁcation of Cr(VI) directly in the solid state. Results obtained with both methods were compared for three solid samples with diﬀerent matrices: a soil containing chromite ore processing residue (COPR), a loamy soil, and a paint sludge. Results showed that Cr(VI) contents determined by the two methods diﬀer signiﬁcantly, and that the EPA Method 3060A procedure underestimated the Cr(VI) content in all studied samples. The underestimation is particularly pronounced for COPR. Low extraction yield for EPA Method 3060A was found to be the main reason. The Cr(VI) present in COPR was found to be more concentrated in magnetic phases. This work provides new XANES analyses of SRM 2701 and its extraction residues for the purpose of benchmarking EPA 3060A performance.

’ INTRODUCTION For solid samples, few analytical techniques enable direct determination of Cr speciation without converting samples to the liquid state. X-ray absorption near edge structure spectroscopy (XANES) is a technique of choice for environmental samples because it allows quantiﬁcation of the Cr(VI)/Cr(III) ratio with good sensitivity. For example, a limit of detection around 10 mg/kg has been reported for Cr(VI).1 This technique relies on the presence of electronic transitions sensitive to the valence, geometry, and distortion of molecules. In the case of Cr, pre-edge features are much more pronounced for Cr(VI) compounds, which have a tetrahedral coordination, than for Cr(III) compounds, which mostly have a centrosymmetric octahedral coordination.2,3 However, this technique is only available in synchrotron facilities, and thus, most speciation studies are performed using wet chemistry methods, which are more sensitive, but require r 2011 American Chemical Society

an extraction step to dissolve Cr(VI). Extraction is challenging because ideally it should be complete, and interconversion of Cr species (e.g., reduction of Cr(VI) to Cr(III)) must be prevented to avoid under- or overestimation of the Cr(VI) content. Diﬀerent extracting solutions have been used to perform the extraction of Cr(VI).4,5 In these studies, the yield of extraction, the possible interconversion of Cr species, and their recoveries were evaluated by spiking soils with exogenous Cr(III) and Cr(VI). The method yielding the best results was obtained using a combination of Na2CO3 and NaOH with continuous swirling and heating at 95 C. The pH of the Na2CO3/NaOH mixture Received: March 25, 2011 Accepted: November 3, 2011 Revised: October 6, 2011 Published: November 03, 2011 10492

dx.doi.org/10.1021/es201002g | Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology should be >12, above which Cr(VI) is stable in solution. This method was promoted in 1996 by the U.S. Environmental Protection Agency (US-EPA) with the reference EPA Method 3060A (hereafter called 3060A in this paper).6 Once the Cr(VI) is extracted, diﬀerent quantiﬁcation methods are available. The latest and most accurate one is performed using high-performance liquid chromatography coupled to inductively coupled plasma mass spectrometry (HPLC-ICP-MS) 7 and involves isotopic spiking of the sample to enable the correction of species inter conversion taking place during preparation and analysis steps.8,9 However, even if the precision of the quantiﬁcation step is an important parameter, a complete extraction step is still necessary to obtain accurate results. Using XANES, Szulczewski et al.10 recently showed that the phosphate buﬀer extraction procedure proposed by James et al.4 (50 mL K2HPO4/KH2PO4 5 mM + 30 min shaking) as a possible method for determining Cr(VI) was ineﬀective with 0.510% recovery of the total Cr(VI) in contaminated soils. Dermatas et al.11 and later Wazne et al.12 and Jagupilla et al.13 found discrepancies in Cr(VI) content of chromite ore processing residue (COPR) between 3060A extracted soils and XANES. They pointed out that the presence of reducing compounds resulted in artiﬁcially low Cr(VI) concentrations in the extraction solution. However, another explanation has recently been proposed by Chrysochoou et al.,14 who demonstrated that the Cr(VI) found in COPR was sometimes entrapped in nodules or bound to phases that are stable over a wide pH range, making it diﬃcult to extract Cr(VI) quantitatively. These studies suggest that it is diﬃcult to determine whether observed diﬀerences are due to interconversion of chromium species in the extracts or to a low extraction yield of EPA3060A or to both of these possibilities. To our knowledge, no in-depth studies have evaluated the actual extraction yield of Method 3060A on diﬀerent matrices. The aim of this study was to compare the amount of hexavalent Cr found using XANES to Method 3060A combined with stateof-the-art detection methods such as EPA Method 6800 (hereafter called 6800). Three solid samples with diﬀerent matrices were investigated by wet chemistry following the 3060A procedure. The raw samples and their extraction residues were also analyzed using XANES for comparison. Among the studied samples is a new standard reference material (SRM) 15 recently developed for extractable Cr(VI) content in soil using method 3060A.

’ MATERIALS AND METHODS Samples. Three Cr-containing samples with distinctive matrices and compositions were selected for this study (Table 1): SRM 2701 hexavalent chromium in contaminated soil (high level) 15 is a soil composed largely of COPR originating from New Jersey, a loamy soil from Belgium,16 and a paint sludge from Italy.16 The samples were used as received; i.e., no additional treatments were performed as indicated on the 3060A extraction procedure. However, for SRM2701 the effect of pregrinding prior extraction was tested (using XANES on extraction residues) but was not found to have a significant effect. The total Cr content in these samples (Table 1) was determined using Instrumental Neutron Activation Analysis (INAA) (total Cr for SRM 2701 was certiﬁed using three independent methods as described in the Certiﬁcate of Analysis), a solid state technique, because it has been shown that some samples could be

ARTICLE

Table 1. Sample Descriptions and Compositionsa SRM 2701

paint sludge

loamy soil

Mineralogy Identiﬁed by XRD (++) chromite

(++) calcite

(++) maghemite/magnetite

(++) maghemite

(++) quartz (+) calcite

(++) calcite

(++) chromite

(+) brucite

(+) quartz

(+) quartz

(+) biotite

(t) hydrocalumite

(t) larnite

(t) hydrocalumite

(t) hydrogarnet

(t) felspar

(t) alumina

(t) periclase

(t) biotite

(t) maghemite

(t) brownmillerite

(t) periclase

(t) chromite (t) hydrogarnet

Matrix Composition (mass %) Fe = 22.9

Ca = 11

Si = 15

Ca = 7.4

Fe = 5.8

Ca = 9.3

Mg = 7.4

C = 2.7

Fe = 6.5

C = 4.9

Si = 2.6

Al = 2.5

Si = 4.2 Total Cr (mg/kg) 42 600 ( 1200b

17 150 ( 640

12 160 ( 450

Cr(VI) by EPA3060A (mg/kg) 551 ( 35

11 400 ( 1300

2070 ( 250

a

(++) Major phase. (+) Minor phase. (t) Trace. Uncertainties are discussed in the text. b Certiﬁed value.

diﬃcult to totally digest using conventional liquid methods due to the presence of refractory phases.17 The Cr(VI) content in these samples was certiﬁed15 or determined by means of interlaboratory comparisons,16 using 3060A as an extraction procedure. EPA Method for Cr(VI) Extraction from Solid Samples and Speciation Analyses. The determination of Cr(VI) content in soils involves four steps: extraction, separation, detection, and quantification. The extraction step was performed following Method 3060A,6 which is the most widely used protocol. Briefly, approximately 0.1 g of a sample was weighed and placed into a microwave vessel. Then, 2 mL of the digestion solution consisting of 0.5 M NaOH and 0.28 M Na2CO3 was added into the vessel along with 16 mg of MgCl2 and 0.02 mL of a 1.0 M K2HPO4/KH2PO4 buffer. This combination of MgCl2 and buffer has been shown to reduce the oxidation of Cr(III) to Cr(VI). A modified Mehod 3060A was used for heating by using microwave-assisted extraction (CEM Discover, Matthews, NC, set to 95 C for 30 min with continuous stirring) instead of a hot plate. This modified method has been shown to give comparable results to those obtained with a hot plate.8,18 Extraction residues were separated by centrifugation, washed three times with deionized water, and prepared for XANES and X-ray diffraction (XRD) analyses as described in those sections. The diﬀerent Cr species present in the extracts were separated using HPLC with an anion exchange guard column (IONPAC AG 11, Dionex, Sunnyvale, CA), and detected using ICP-MS (Elan 6100 DRC, PerkinElmer, Waltham, MA) equipped with a Meinhard nebulizer ﬁtted to a cyclonic spray chamber. Methods Used for Cr(VI) Quantification after Extraction. Several methods exist to quantify Cr species19 following chemical extraction. To compare them in terms of precision and accuracy, 10493

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology four of these methods were used on the loamy soil and paint sludge: external calibration using HPLC-ICP-MS, standard additions using ICP-MS, isotope dilution mass spectrometry (IDMS), and Method 6800. Such a study15 has already been carried out on SRM 2701 in which determinative bias as high as 30% was observed, depending on the method used. Briefly, external calibration relies on the preparation of standards of known Cr(VI) concentrations in water,20 the standard addition method requires the addition of known quantities of the Cr(VI) to the sample,20 IDMS requires the addition of a known amount of isotopically enriched 50Cr(VI) to the sample (in our case this addition was performed before and after extraction),21 and Method 6800 requires the addition of known amounts of two isotopically enriched species 50Cr(VI) and 53Cr(III).6 Although exogenous spikes will not behave exactly as the analyte present within the matrix, EPA6800 is widely perceived as the most accurate quantification method because it can correct for conversions of Cr(VI) to Cr(III) and of Cr(III) to Cr(VI) during preparation and analysis of samples.22 External calibration and standard addition methods were carried out using Cr(VI) and Cr(III) stock solutions (Sigma-Aldrich, St. Louis, MO). Isotope dilution and Method 6800 methods were carried out using isotopically enriched Cr(VI) and Cr(III) synthesized from solid, 50 Cr-enriched Cr2O3 and 53Cr-enriched Cr2O3. For each detection method, the entire extractiondetection suite was repeated on three replicate specimens. X-ray Diffraction. XRD measurements were performed on raw soil samples before and after 3060A extraction. Residue from SRM 2701 was also magnetically separated, and both fractions were analyzed by XRD. XRD data were collected using a Bruker AXS D8 Advance unit using a θθ configuration with the source and a two-dimensional detector scanned with a 0.016 2θ step size for a total scan time of 35 min. The Cu X-ray tube was operated at 35 kV and 45 mA with a Ni filter to remove K-M (Kβ) radiation leaving just the K-L2,3 (Kα1,2) radiation (λ = 0.154 nm). XANES. Standards for XANES Calibration. Mixtures of Cr(VI) and Cr(III) were prepared to calibrate XANES. In the environment, Cr(VI) and Cr(III) are the most abundant forms of Cr because the other oxidation states (+II, +IV, +V) are less stable. Two sets of Cr-doped silica standards were prepared to contain 1% and 0.5% total Cr (by mass) and each set having Cr(VI)/ Crtotal ratios of 0%, 1%, 2%, 5%, 10%, 25%, 50%, 75%, and 100%. For silica standards containing 1% total Cr, aqueous Cr(III) or Cr(VI) was prepared by dissolving CrCl3 3 6H2O or K2Cr2O7 in purified water. For standards containing 0.5% total Cr, SRM 3112a Chromium (Cr) Standard Solution and SRM 2109 Chromium(VI) Speciation Standard Solution were used as sources of Cr(III) and Cr(VI), respectively. Aqueous Cr(III) or Cr(VI) solutions were added to pure solid silica in a solid/ liquid mass ratio of 1:3 in 50 mL flasks which were closed and stirred on a rotating table for 24 h, after which the solids were freeze-dried. The dried Cr(III)- and Cr(VI)-doped silica were manually ground. The final calibrants were prepared by mixing weighed amounts of these dried, Cr-doped silicas to the required Cr(VI)/Crtotal ratios. Pure silica was analyzed by synchrotron X-ray fluorescence, and no Cr was detected. Additional standard samples consisted of commercial powders of Cr2O3 (Baker) and K2CrO4 (Aldrich), and a chromite mineral containing FeCrIII2O4 provided by the mineral collection from Paul Sabatier University (Toulouse, France). Another reference sample was made by mixing pure powders of K2CrO4 and

ARTICLE

Table 2. Comparison of Cr(VI) Mass Fractions Obtained for the Loamy Soil and the Paint Sludge with Diﬀerent Quantiﬁcation Methodsa loamy soil

(mg/kg)

(mg/kg)

literature value

11 400 ( 1300

HPLC-ICPMS external calibration

10 652 ( 349

1985 ( 95

9460 ( 113

not determined

IDMS, spiked before extraction IDMS, spiked after extraction

10 628 ( 44 10 796 ( 411

1986 ( 40 1936 ( 35

EPA 6800, double spike IDMS

10 700 ( 38

1920 ( 31

HPLC-ICPMS standard addition

a

paint sludge

2070 ( 250

Uncertainties are discussed in the text.

CrK(SO4)2 3 12H2O (Fisher Scientiﬁc) to a Cr(VI)/Crtotal ratio of 5.3%. The standard additions method (detailed in ref 23) was applied to samples of SRM 2701 by spiking, grinding, and mixing sample aliquots with increasing amounts of solid K2Cr2O7. Instrumentation and Data Treatment. XANES measurements were performed on beamline BM08 at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France, using a Si(311) monochromator. Solid samples, solid residues from EPA extraction procedures, and calibration standards with 1% total Cr were prepared as pressed pellets and measured using a cryostat operating at 80 K to minimize potential speciation change of Cr under beam exposure. Cr K edge XANES spectra were collected in fluorescence mode using a 13-element Ge detector and an energy step size of 0.1 eV in the vicinity of the edge. Energy calibration was performed by measuring Cr foil between samples and defining the maximum of the first derivative of the absorption spectrum as 5989.0 eV. Additional measurements were performed at NIST beamline X23A2 at the National Synchrotron Light Source (NSLS) in Upton, NY, using a Si(311) monochromator. Unknown samples and silica standards with 0.5% total Cr were enclosed between two 2.5 μm polyester sheets (Somar International, Reno, NV) in a polyethylene holder. Samples were rotated during measurements to average out eﬀects of heterogeneity by passing a greater area of the specimen through the small primary X-ray beam (10 mm 0.5 mm). XANES spectra were recorded in ﬂuorescence mode using a four-element Si drift detector (Vortex-EX, SII Nanotechnology, Northridge, CA). Again, Cr foil was employed for energy calibration. For each specimen, collected spectra were pre-edge background subtracted using a linear ﬁt, postedge normalized using a quadratic polynomial and averaged in the Athena software.24

’ RESULTS AND DISCUSSION Comparison of the Cr(VI) Quantification Strategies after 3060A Extraction. Results for both the loamy soil and the paint

sludge show that the amount of Cr(VI) obtained with four out of five methods is within the uncertainty range determined by the interlaboratory study16 (Table 2). Only the standard additions method using ICP-MS gave a slightly underestimated result, which is difficult to explain as the Cr(VI) added to the sample should be stable at the pH condition (pH > 12). The IDMS method with single spiking of 50Cr-enriched Cr(VI) gave almost the same result whether the isotopic spike was added before or after 3060A extraction, meaning that the spike does not react 10494

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology

ARTICLE

Figure 1. (a) Correlation between the area of pre-edge peak areas and the mass fraction of Cr(VI) for doped-silica mixtures of 0.5% and 1% total Cr. (b) Calibration curve obtained using the height of pre-edge peaks when mixing SRM 2701 with increasing amount of K2Cr2O7 for the method of standard additions. The relative standard deviation of replicate area and height determinations was <3%.

much with the solid matrices of the two samples. Method 6800 appears to give the best repeatability although IDMS is similar when spiked before extraction. Also, Method 6800 yields almost the same accuracy as the other detection methods used for the two soils analyzed. In a previous study15 on SRM 2701, Method 6800 gave 30% higher Cr(VI) mass fractions than the other methods. This could be due to the presence of reducing agents (such as Fe(II) or organic matter8) in SRM2701 matrix that can have an impact on the conversion of Cr species during extraction, and Method 6800 can correct for interconversion of Cr species. Because the results for paint sludge and loamy soil agree well for a single extraction procedure followed by diﬀerent quantiﬁcation methods, there is no evidence of species interconversions or other phenomena resulting from chemical extraction and procedures in the quantitative methods. In contrast, prior results for SRM 270115 lead one to question whether the cause is related to the material and its behavior under extraction conditions. XANES results obtained on the original materials and XRD determinations of mineralogy are discussed below for the purpose of comparing the results of nondestructive analysis of the materials to results involving the extraction procedure. XANES Analysis of Samples and Their 3060A Residues. Pre-Edge Peak Calibration. The pre-edge and superimposed X-ray lines were fitted using an error function and Gaussian components, respectively, as previously described by Szulczewski et al.10 The distinctive peaks for the doped-silica standard end-members (0% and 100% Cr(VI)) were characterized in terms of position and width, and these data were applied to fit spectra from mixtures (see Supporting Information Figure S1). The two peaks surrounding the 1s3d transition of Cr(VI) result from moiety

distortion25 present in K2CrO4, which was the source of Cr(VI) in the Cr(VI)-doped silica standards. The area of the 1s3d peak corresponding to the Cr in tetrahedral configuration was determined and plotted versus the fraction of Cr(VI) in the calibrants with a high degree of correlation as shown in Figure 1a. The two calibrations performed at two separate beamlines compare well. For all spectra, the relative standard deviation of replicate (n = 3) area determinations was <3%. Limits of detection were not estimated in the normal, rigorous manner. Rough estimates were made by comparing the peak areas from low Cr(VI) calibrants. For the calibration performed with 1% total Cr, the 5% Cr(VI)silica could be unambiguously distinguished from the lower Cr concentrations, but it was not possible to diﬀerentiate the 0%, 1%, and 2% Cr(VI)silica standards. For the calibration performed with 0.5% total Cr, it was possible to distinguish 0% from 1% and 2% Cr(VI)silica but not 1% from 2%. The limit of detection for Cr(VI) must therefore be in the range 02% total Cr which represents 100 mg/kg of Cr(VI) in absolute value. The small diﬀerence between the slopes of these calibration curves performed in diﬀerent synchrotron facilities is likely due to diﬀerences in reference compounds and total chromium concentrations of the two calibration sets. Table 3 shows the mass fraction of Cr(VI) obtained by multiplying the ratio of Cr(VI)/Crtotal from XANES analyses and the total Cr. Uncertainty estimates were calculated by combining contributions from INAA and XANES. The uncertainties for INAA contribute approximately 2% relative standard uncertainty. The main source of uncertainty for XANES is related to diﬀerences observed among Cr K edge spectra of the model compounds used in the calibration and the spectra of the actual 10495

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology

ARTICLE

Table 3. Quantiﬁcation Results Using XANES sample SRM 2701

Cr(VI) XANES (mg/kg)a 3390 ( 373b

total Cr (mg/kg) 42 600 ( 1200

extraction yield (%)e

mass balance error (%)e

32.2

15.3

94.9

0.1

57.1

13.0

3312 ( 364c 3055 ( 137d 3060A extraction residue

2330 ( 256b

35 270 ( 190

2078 ( 229c 3060A extraction residue, nonmagnetic part 3060A extraction residue, magnetic part paint sludge 3060A extraction residue

270 ( 30b

10 380 ( 238

3592 ( 395b

48 820 ( 1120

12 177 ( 1339b 11 821 ( 1300c

17150 ( 640

567 ( 62b

10 560 ( 190

662 ( 73c loamy soil

4796 ( 528b

12 160 ( 450

4591 ( 505c 3060A extraction residue

2101 ( 231b

9460 ( 350

1922 ( 211c a

Estimated uncertainties, see text for details. b Using 1% Crtotal calibration. c Using 0.5% Crtotal calibration. d Using standard addition. e See text for details.

Cr compounds in the samples. It has been shown that, depending on the Cr(VI) compound used in the calibration, the K edge spectrum may present noticeable diﬀerences, particularly in the 1s3d region of Cr(VI).25 The samples being likely composed of more than two Cr species and their natures being for most cases unknown, it is diﬃcult to evaluate this uncertainty. However, its order of magnitude can be deduced in two ways. The ﬁrst one involves comparison of the calibration end-members’ (0% and 100% Cr(VI)) pre-edge peak heights to those of pure reference compounds such as Cr2O3, chromite, and K2CrO4 (see Supporting Information, Figure S2). The diﬀerence in height between K2CrO4 and silica doped with 100% Cr(VI) added to the diﬀerence in height between chromite and the silica doped with 100% Cr(III) represents a relative diﬀerence of approximately 6% in Cr(VI) content. The second way to evaluate the uncertainty is to analyze a reference material. This was done utilizing compounds having diﬀerent Cr atom environments than those used for the calibrations in a mixture having 5.3% Cr(VI). Quantiﬁcation of this reference sample using external calibration gave a value of 4.4% Cr(VI), which represents a relative error of 16%. About the same error (17%) was encountered when using a linear combination of the two pure compounds (K2CrO4 and CrK(SO4)2 3 12H2O) used to make the mixture with 5.3% Cr(VI) as done by Parsons et al.26 Similar errors were reported by Fandeur et al.3 The two diﬀerent methods used to evaluate uncertainties indicate that the true relative error must be between 6% and 16%. The intermediate value of 11% was therefore taken to calculate the relative error of external calibrations. Two recommendations can be made to minimize these uncertainties: (1) Cr2O3 or eskolaite should be avoided as a Cr(III) reference compound, if it is not present in the sample, because Cr in Cr2O3 is located in a distorted octahedral site. Its transition energy is close to the energy of the tetrahedral Cr(VI) feature and represents a spectral interference.3,25 If present here, its abundance is minor because it was not detected by XRD analyses (Table 1). (2) Cr(VI) compounds should be carefully chosen because the intensity and the peak width of the 1s3d transition of chromates is not unique and will be aﬀected by CrO42‑

moiety distortion; e.g., there will be diﬀerences between monochromate and dichromate.25 Choices can be made by comparing peak widths of sample and reference compounds.27 The method of standard additions applied to XANES, which to our knowledge has not been tested to date, proved to be suitable for Cr(VI) determination in SRM 2701 (Figure 1b). This approach allows easier spectrum ﬁtting because just one compound is added, and the compound can be chosen using data from published studies2,25,28 to have the approximate same peak width as the Cr(VI) found in the sample. No Cr(III) model compound is required; therefore, the uncertainty linked to the choice of compound used to model Cr(III) is eliminated. Figure 1b illustrates the advantages oﬀered by the standard addition approach for SRM2701. Because the Cr(III) chromite dominates the Cr speciation in the initial and spiked samples, the postedge portion of the spectrum remains similar among the original and spiked samples, facilitating normalization and deconvolution because all parameters, except the 1s3d peak height, can remain the same for treating all spectra. A relatively strong linear correlation was found between the height of the 1s3d peak and the amount of added Cr(VI). This approach gave a result of 3055 mg/kg which is about 10% lower than the external calibration, likely due to a diﬀerence in the Cr(VI) compound used. Standard addition depends on a good linear correlation between count rate and amount of analyte added. No X-ray ﬂuorescence calibration is truly linear because self-absorption29 causes a decrease in the ratio of counts per amount of analyte as the amount of analyte increases. In our case, spike samples containing up to 18 000 mg/kg of Cr were tested, and deviation from linearity was apparent at the highest amount of total Cr. Nonlinearity of this nature causes overestimation of the amount of Cr(VI). A second limitation of standard additions is the ability to completely mix the spike with the original sample. The implicit assumption is that perfect mixing has occurred and that X-rays will not be aﬀected by diﬀerential absorption between particles of soil and particles of spike compound, in this case, K2Cr2O7 powder. The assumption can be tested by observing the goodness of ﬁt of the unspiked sample to the calibration line. In our case, a test showed a potential bias of approximately +4% relative 10496

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology

ARTICLE

Figure 2. Superimposed Cr K edge spectra of doped-silica standards and (a) SRM 2701, (b) loamy soil, and (c) paint sludge. The pre-edge region used for quantiﬁcation purpose is featured in the zoom box.

between results from lines with and without the original sample. A standard additions calibration must be carefully designed to minimize the inﬂuence of self-absorption and limitations of mixing of spike and sample. Overall, the estimated uncertainty in this instance is a factor of 2 lower than for external calibration because just the errors linked to Cr(VI) addition must be included. As no reference material for total Cr(VI) in solids exists, further studies are needed to determine which method is the most accurate. Samples. Table 3 shows that the chemical extractions and XANES quantifications are reproducible because the results using the two sets of extractions are similar within the uncertainty intervals. Depending on the sample, the determined Cr(VI) mass fraction can vary significantly between XANES and EPA-based methods. The worst case is SRM 2701 where 3060A underestimates the mass fraction of Cr(VI) by a factor of 6 (see comparison with Table 1). The XANES analyses performed on the extraction residues show that an incomplete extraction step is responsible for Cr(VI) underestimation by 3060A because nonnegligible amounts of Cr(VI) were present in the residues (Figure 2). The masses of extraction residues were compared to the initial sample masses, and no significant differences were observed. Extraction yield Y was calculated as follows Y ð%Þ ¼

CrðVIÞsample CrðVIÞresidue CrðVIÞsample

100

ð1Þ

where Cr(VI)sample and Cr(VI)residue are the means of the XANES Cr(VI) values obtained using both calibrations. The extraction yields diﬀer signiﬁcantly among samples, indicating a possible inﬂuence of the sample matrix. The postedge part of the spectrum in Figure 2 shows that the SRM 2701 Cr-binding phases

have not been disturbed signiﬁcantly after extraction in contrast to the two other soils for which the extraction was more eﬃcient. As said earlier, this observation is most likely due to the fact that the postedge is dominated by the chromite phases, which are much more abundant than Cr(VI)-bearing phases. The INAA result for the paint sludge residue is likely overestimated, mainly due to the combined eﬀects of the small sample amount available and sample heterogeneity; however, it is not expected to have a signiﬁcant impact on extraction yield, which is already 95%. Other factors than a low-yield extraction could explain the underestimation of Cr(VI) by the 3060A method, for example, interconversion of chromium species during the detection step. To clarify that point, and considering that the amount of Cr(VI) determined by XANES in raw samples should be equal to the amount of Cr(VI) extracted by 3060A plus the amount of Cr(VI) left in the extraction residues, the mass balance error, ∂MB, was calculated as follows ∂MB ð%Þ ¼

XANES EPA CrðVIÞXANES sample ðCrðVIÞresidue þ CrðVIÞextract Þ

CrðVIÞXANES sample

100

ð2Þ XANES where Cr(VI)XANES sample and Cr(VI)residue are the averages of the XANES Cr(VI) values obtained using both calibrations. The relative errors in mass balance range from 0.1% to 15.3%, which is reasonably good considering the uncertainties associated with the results of each method. Therefore, it can be concluded that extraction yield is the main reason for underestimation of Cr(VI). Effect of 3060A Extraction on the Mineralogy of SRM 2701. The mineralogy of SRM 2701 (Figure 3) includes

10497

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology

ARTICLE

Figure 3. XRD patterns from SRM 2701 and its extraction residue before and after magnetic separations: Ct, calcite CaCO3; Cr, chromite FeCr2O4; Hc, hydrocalumite Ca2(Al,Fe)(OH)7 3 3H2O; P, periclase MgO; Hg, hydrogarnet Ca3(Al,Fe)2(OH)12; Mg, maghemite/magnetite Fe2O3/Fe3O4; Qz, quartz SiO2; Bw, brownmillerite Ca2(Fe,Al)2O5.

chromite, magnetite, maghemite, brownmillerite, calcite, and periclase and is characteristic of such samples.3032 The mineralogy of COPR can be influenced by the composition of initial chromite-bearing minerals, the processes used to extract chromium, and conditions to which the material is subjected. XRD analyses (Figure 3) performed on SRM 2701 subjected to different treatments showed that chromite, the main chromium(III)bearing mineral, was still present in the residue of extraction as expected. In addition, maghemite and magnetite were still present, but due to the high pH of the medium these iron-oxides are unlikely to sorb Cr(VI) as observed for some iron-oxyhydroxides in relatively acidic lateritic soils.3,33,34 Lesser amounts of calcium-containing minerals, namely calcite (CaCO3), hydrogarnet (Ca3(Al,Fe)2(OH)12), and hydrocalumite (Ca2(Al,Fe)(OH)7 3 3H2O) were identified in the raw sample and in the residue. Common in COPR, hydrogarnet and hydrocalumnite were found to incorporate hexavalent Cr in their structure,31,35 and above pH 11, Cr(VI) solubility would be controlled by Cr(VI)-substituted hydrogarnet and Cr(VI)-hydrocalumite.32 In Cr(VI)-hydrocalumite, Cr(VI) is present in the interstitial spaces of the layered double-hydroxide, and should be easily extractable. In Cr(VI)-bearing hydrogarnet, Cr(VI) is included in the structure, which suggests it is more stable. The incorporation of CrO42‑ in the structure of calcite was also reported by Tang et al.36 and suggested by Hua et al.37 Brownmillerite is also present in SRM 2701 and in the residue, and the incorporation of chromate in its structure in COPR has been reported by Gibbs38 and more recently Chrysochou.30 Furthermore, it is present as resistant nodules where Cr(VI) is not easily extractible.30 Thus, it can be hypothesized that hydrogranet, hydrocalumite, calcite, and/or brownmillerite can trap chromate and make it unavailable for chemical extraction, but further studies are needed to confirm these mechanisms. The INAA and XANES analyses (Table 3) on these fractions showed that the magnetic fraction is highly enriched in both total Cr and Cr(VI) compared to the

nonmagnetic fraction. The XRD analysis clearly shows that the magnetic fraction is enriched in the chromite/maghemite/magnetite phases and the nonmagnetic fraction is enriched in calcite; thus, it can be inferred that the main part of Cr(VI) is not trapped in calcite only, but it is difficult to estimate the enrichment or not in hydrogarnet, hydrocalumnite, and browmillerite. XRD analyses appear to show that the major part of the nonextracted Cr(VI) is associated with phases present in the magnetic fraction of the residue, but their nature remains difficult to determine due to the relative similarity between the XRD spectra of the magnetic and nonmagnetic fractions. Further research is needed to fully understand the mechanism involved in the retention of Cr(VI) and so improve its extraction. The diﬀerence in extraction yield between paint sludge (94.9%) and SRM 2701 (32.2%) can be explained by a diﬀerent repartition of Cr(VI) within mineral phases. Calcite, chromite, and maghemite were the main crystallized phases identiﬁed in the raw paint sludge and were still in the residue while calcite was the major phase removed by the chemical extraction. It can be hypothesized that Cr(VI) was removed with calcite because CrO42‑ is possibly sequestered in its structure.36,37 It can also be mentioned that the XANES spectra of raw paint sludge and its residue are markedly diﬀerent, especially in the postedge region, indicating that the main initial Cr(VI) bearing phase(s), which represents the dominant part of Cr, was removed. For the loamy soil, 57.1% of Cr(VI) was extracted. Quartz, calcite, biotite, feldspars, brucite, and a small amount of chromite were identiﬁed as the main crystalline phases in the raw and residue samples. Calcite was the dominant mineral extracted by 3060A. It can be hypothesized that Cr(VI) is associated with calcite and/or brucite (Mg(OH)2) as hypothesized by Chrysochou.30 Also, the binding of Cr(VI) in Ca(OH)2 at high pH was reported by Ginder-Vogel.39 XANES as a Tool To Improve Measurement of Cr(VI) in Soils Using Chemical Extractions. This study on three different 10498

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology soils showed that 3060A does not completely solubilize all forms of hexavalent chromium, depending on soil matrix. In particular, the soil composed of COPR was found to contain Cr(VI) which was not extracted chemically. This work also showed that the use of spiked soils for extraction studies is not an ideal approach for studying alkaline extraction methods as this does not necessarily reflect the complexities of real soils. XANES represents a powerful tool, but unfortunately synchrotron facilities are difficult to access and generally not conducive to high throughput analytical methods at this time. It is conceivable that XANES spectroscopy can serve as a benchmark for improvements or modifications to alkaline extraction methods, which might consist of sequential extraction approaches or different extraction conditions, tailored to the type of matrix being studied. This future work would be of particular importance in the case of environmental studies dealing with the toxicity of Cr(VI), which may concern easily extractible Cr(VI) but also more insoluble phases.40,41

’ ASSOCIATED CONTENT

bS

Supporting Information. Figures S1S3. This material is available free of charge via the Internet at http://pubs.acs.org.

’ DISCLOSURE Certain commercial equipment, instruments, and materials are identiﬁed in this work to specify adequately the experimental procedure. Such identiﬁcation does not imply recommendation or endorsement by the National Institute of Standards and Technology, nor does it imply that the materials or equipment identiﬁed are necessarily the best available for this purpose. ’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT Use of the National Synchrotron Light Source, Brookhaven National Laboratory, was supported by the U.S. Department of Energy, Oﬃce of Science, Oﬃce of Basic Energy Sciences, under Contract DE-AC02-98CH10886. We are grateful to Bruce Ravel at NIST beamline X23A2 for his assistance in the setup of experiments and for his help with Athena software. We also acknowledge the European Synchrotron Radiation Facility for provision of beam time. Finally, we thank two anonymous reviewers for their comments contributing to improvement of the manuscript. ’ REFERENCES (1) Shaﬀer, R.; Cross, J.; Rose-Pehrsson, S.; Elam, W. Speciation of chromium in simulated soil samples using X-ray absorption spectroscopy and multivariate calibration. Anal. Chim. Acta 2001, 442, 295–304. (2) Pantelouris, A.; Modrow, H.; Pantelouris, M.; Hormes, J.; Reinen, D. The inﬂuence of coordination geometry and valency on the K-edge absorption near edge spectra of selected chromium compounds. Chem. Phys. 2004, 300, 13–22. (3) Fandeur, D.; Juillot, F.; Morin, G.; Olivi, L.; Cognigni, A.; Webb, S. M.; Ambrosi, J. P.; Fritsch, E.; Guyot, F.; Brown, G. E. XANES evidence for oxidation of Cr(III) to Cr(VI) by Mn-oxides in a lateritic regolith developed on serpentinized ultramaﬁc rocks of New Caledonia. Environ. Sci. Technol. 2009, 43, 7384–7390.

ARTICLE

(4) James, B.; Petura, J.; Vitale, R.; Mussoline, G. Hexavalent chromium extraction from soils: A comparison of ﬁve methods. Environ. Sci. Technol. 1995, 29, 2377–2381. (5) Vitale, R.; Mussoline, G.; Petura, J.; James, B. Hexavalent chromium extraction from soils: Evaluation of an alkaline digestion method. J. Environ. Qual. 1994, 23, 1249–1256. (6) Alkaline Digestion for Hexavalent Chromium; United States Environmental Protection Agency Method 3060A; United States EPA: Washington, DC, 1996. (7) Elemental and Speciated Isotope Dilution Mass Spectrometry; United States Environmental Protection Agency Method 6800; United States EPA: Washington, DC, 2007. (8) Huo, D.; Kingston, H. Correction of species transformations in the analysis of Cr(VI) in solid environmental samples using speciated isotope dilution mass spectrometry. Anal. Chem. 2000, 72, 5047–5054. (9) Huo, D.; Yusheng, L.; Kingston, H. Determination and correction of analytical biases and study on chemical mechanisms in the analysis of Cr(VI) in soil samples using EPA protocols. Environ. Sci. Technol. 1998, 32, 3418–3423. (10) Szulczewski, M.; Helmke, P.; Bleam, W. Comparison of XANES analyses and extractions to determine chromium speciation in contaminated soils. Environ. Sci. Technol. 1997, 31, 2954–2959. (11) Dermatas, D.; Chrysochoou, M.; Moon, D.; Grubb, D.; Wazne, M.; Christodoulatos, C. Ettringite-induced heave in chromite ore processing residue (COPR) upon ferrous sulfate treatment. Environ. Sci. Technol. 2006, 40, 5786–5792. (12) Wazne, M.; Jappilla, A.; Moon, D.; Jagupilla, S.; Christodoulatos, C.; Kim, M. Assessment of calcium polysulﬁde for the remediation of hexavalent chromium in chromite ore processing residue (COPR). J. Hazard. Mater. 2007, 143, 620–628. (13) Jagupilla, S. C.; Moon, D. H.; Wazne, M.; Christodoulatos, C.; Kim, M. G. Eﬀects of particle size and acid addition on the remediation of chromite ore processing residue using ferrous sulfate. J. Hazard. Mater. 2009, 168, 121–128. (14) Chrysochoou, M.; Fakra, S.; Marcus, M.; Moon, D.; Dermatas, D. Microstructural analyses of Cr(VI) speciation in chromite ore processing residue (COPR). Environ. Sci. Technol. 2009, 43, 5461–5466. (15) Nagourney, S.; Wilson, S.; Buckley, B.; Kingston, H.; Yang, S.; Long, S. Development of a standard reference material for Cr(VI) in contaminated soil. J. Anal. At. Spectrom. 2008, 23, 1550–1554. (16) Tirez, K.; Scharf, H.; Calzolari, D.; Cleven, R.; Kisser, M.; Luck, D. Validation of a European standard for the determination of hexavalent chromium in solid material. J. Environ. Monit. 2007, 9, 749–759. (17) Kelly, W.; Murphy, K.; Becker, D.; Mann, J. Determination of Cr in certiﬁed reference material HISS-1, marine sediment, by cold plasma isotope dilution ICP-MS and INAA: comparison of microwave versus closed (Carius) tube digestion. J. Anal. At. Spectrom. 2003, 18, 166–169. (18) Rahman, G.; Kingston, H.; Towns, T.; Vitale, R.; Clay, K. Determination of hexavalent chromium by using speciated isotopedilution mass spectrometry after microwave speciated extraction of environmental and other solid materials. Anal. Bioanal. Chem. 2005, 382, 1111–1120. (19) Unceta, N.; Seby, F.; Malherbe, J.; Donard, O. F. X. Chromium speciation in solid matrices and regulation: A review. Anal. Bioanal. Chem. 2010, 397, 1097–1111. (20) Xing, L.; Beauchemin, D. Chromium speciation at trace level in potable water using hyphenated ion exchange chromatography and inductively coupled plasma mass spectrometry with collision/reaction interface. J. Anal. At. Spectrom. 2010, 25, 1046–1055. (21) Coedo, A.; Dorado, T.; Padilla, I.; Alguacil, F. Speciation of chromium in steelmaking solid wastes by selective retention on ionexchange media and determination by isotope dilution inductively coupled plasma mass spectrometry. J. Anal. At. Spectrom. 2000, 15, 1564–1568. (22) Kingston, H.; Huo, D.; Lu, Y.; Chalk, S. Accuracy in species analysis: Speciated isotope dilution mass spectrometry (SIDMS) exempliﬁed by the evaluation of chromium species. Spectrochim. Acta, Part B 1998, 53, 299–309. 10499

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

Environmental Science & Technology

ARTICLE

(23) Kelly, W.; MacDonald, B.; Guthrie, W. Gravimetric approach to the standard addition method in instrumental analysis. Anal. Chem. 2008, 80, 6154–6158. (24) Ravel, B.; Newville, M. Athena, Artemis, Hephaestus: data analysis for X-ray absorption spectroscopy using IFEFFIT. J. Synchrotron Radiat. 2005, 12, 537–541. (25) Farges, F. Chromium speciation in oxide-type compounds: Application to minerals, gems, aqueous solutions and silicate glasses. Phys. Chem. Miner. 2009, 36, 463–481. (26) Parsons, J. G.; Dokken, K.; Peralta-Videa, I. R.; RomeroGonzalez, J.; Gardea-Torresdey, J. L. X-ray absorption near edge structure and extended X-ray absorption ﬁne structure analysis of standards and biological samples containing mixed oxidation states of chromium(III) and chromium(VI). Appl. Spectrosc. 2007, 61, 338–345. (27) Dubrail, J.; Farges, F. Not all chromates show the same pre-edge feature. Implications for the modelling of the speciation of Cr in environmental systems. J. Phys.: Conf. Ser. 2009, 190, 012176. (28) Huggins, F.; Najih, M.; Huﬀman, G. Direct speciation of chromium in coal combustion by-products by X-ray absorption ﬁnestructure spectroscopy. Fuel 1999, 78, 233–242. (29) Goulon, J.; Goulon-Ginet, C.; Cortes, R.; Dubois, J. M. On experimental attenuation factors of the amplitude of the exafs oscillations in absorption, reﬂectivity and luminescence measurements. J. Phys. France 1982, 43, 539–548. (30) Chrysochoou, M.; Dermatas, D. Application of the Rietveld method to assess chromium(VI) speciation in chromite ore processing residue. J. Hazard. Mater. 2007, 141 (2), 370–377. (31) Hillier, S.; Roe, M.; Geelhoed, J.; Fraser, A.; Farmer, J.; Paterson, E. Role of quantitative mineralogical analysis in the investigation of sites contaminated by chromite ore processing residue. Sci. Total Environ. 2003, 308 (13), 195–210. (32) Geelhoed, J.; Meeussen, J.; Hillier, S.; Lumsdon, D.; Thomas, R.; Farmer, J.; Paterson, E. Identiﬁcation and geochemical modeling of processes controlling leaching of Cr(VI) and other major elements from chromite ore processing residue. Geochim. Cosmochim. Acta 2002, 66 (22), 3927–3942. (33) Oze, C.; Fendorf, S.; Bird, D. K.; Coleman, R. G. Chromium geochemistry of serpentine soils. Int. Geol. Rev. 2004, 46, 97–126. (34) Stumm, W. Chemistry of the Solid-Water Interface: Processes at the Mineral-Water and Particle-Water Interface in Natural Systems; John Wiley & Sons: New York, 1992. (35) Palmer, C. D. Precipitates in a Cr(VI)-contaminated concrete. Environ. Sci. Technol. 2000, 34, 4185–4192. (36) Tang, Y.; Elzinga, E. J.; Young, J. L.; Reeder, R. J. Coprecipitation of chromate with calcite: Batch experiments and X-ray absorption spectroscopy. Geochim. Cosmochim. Acta 2007, 71, 1480–1493. (37) Hua, B.; Deng, B.; Thornton, E. C.; Yang, J.; Amonette, J. E. Incorporation of chromate into calcium carbonate structure during coprecipitation. Water, Air, Soil Pollut. 2007, 179, 381–390. (38) Gibb, T. C. Study of calcium chromium iron oxides by extended X-ray absorption ﬁne structure spectroscopy. J. Mater. Chem. 1992, 2, 105–110. (39) Ginder-Vogel, M.; Borch, T.; Mayes, M. A.; Jardine, P. M.; Fendorf, S. Chromate reduction and retention processes within arid subsurface environments. Environ. Sci. Technol. 2005, 39, 7833–7839. (40) Wise, S.; Shaﬄey, F.; LaCerte, C.; Goertz, C.; Dunn, J.; Gulland, F.; Aboueissa, A.; Zheng, T.; Wise, J. Particulate and soluble hexavalent chromium are cytotoxic and genotoxic to Steller sea lion lung cells. Aquat. Toxicol. 2009, 91, 329–335. (41) Beaver, L. M.; Stemmy, E. J.; Constant, S. L.; Schwartz, A.; Little, L. G.; Gigley, J. P.; Chun, G.; Sugden, K. D.; Ceryak, S. M.; Patierno, S. R. Lung injury, inﬂammation and Akt signaling following inhalation of particulate hexavalent chromium. Toxicol. Appl. Pharmacol. 2009, 235, 47–56.

10500

dx.doi.org/10.1021/es201002g |Environ. Sci. Technol. 2011, 45, 10492–10500

ARTICLE pubs.acs.org/est

Application of GCHRMS and GCGCTOFMS To Aid in the Understanding of a Dioxin Assay’s Performance for Soil and Sediment Samples Amy Dindal, Elizabeth Thompson, and Erich Strozier Battelle, 505 King Ave., Columbus, Ohio 43201, United States

Stephen Billets* National Exposure Research Laboratory, U.S. Environmental Protection Agency, 944 E. Harmon Ave., Las Vegas, Nevada 89119, United States

bS Supporting Information ABSTRACT: There have been numerous attempts to correlate results obtained by gas chromatographymass spectrometry (GCMS) to alternative techniques such as immunoassays and bioassays for the analysis of dioxins in environmental samples. In spite of these eﬀorts, uncertainties about the performance of these methods remain. Following a series of performance studies of various dioxin assays, an in-depth evaluation of sample extracts from the Procept Rapid Dioxin Assay was conducted to provide users with a clearer understanding of the diﬀerences in the assay’s results compared to traditional mass spectrometry. Two powerful analytical techniques [high-resolution mass spectrometry (HRMS) and two-dimensional gas chromatography coupled with timeof-ﬂight mass spectrometry (GCGCTOFMS)] were used to provide a unique perspective about the assay’s underlying analytical performance. HRMS analyses demonstrated that the target dioxin and furans were consistently captured in the assay’s extracts. TOFMS analyses revealed that interferents in the sample extracts resulting from inconsistencies in the sample preparation process appear to be the primary factor contributing to the assay’s imprecision. The conclusion of this research was the assay results cannot be expected to correlate directly with HRMS and should only be utilized as a screening technique (e.g., to identify the relative ranking of contamination, to determine if samples are above/ below threshold levels, or to monitor a cleanup) for environmental matrices such as soil and sediment.

’ INTRODUCTION Numerous dioxin-contaminated sites are known to exist in the United States and throughout the world.19 The U.S. Environmental Protection Agency (EPA) has proposed new site remediation guidelines and is reassessing the dioxin exposure risk paradigm.1 Generally, the term “dioxin” also encompasses dioxin-like compounds, such as polychlorinated dibenzofurans and coplanar polychlorinated biphenyls. However, traditional analysis by gas chromatographymass spectrometry (GCMS), such as EPA Method 1613b, which utilizes high-resolution mass spectrometry (HRMS), only involves the quantiﬁcation of 17 dioxin and furans to determine the dioxin toxic equivalence (TEQ) in a sample.10 The cost of traditional analyses ranges from $500 to $1000 or more per sample, and the time delay in getting results can be prohibitive to monitoring dioxin levels during conventional risk assessment or remediation activities. Alternative measurement techniques for dioxin have been proposed to be less costly and easier to perform using biologically based r 2011 American Chemical Society

materials. These technologies are widely accepted in the food and feed industry, including monitoring levels of dioxin in matrices such as grain and ﬁsh tissues.26 However, the application of these technologies to soil and sediment monitoring has been challenging due to the diﬃculty in generating comparable results between the biologically based assay techniques, which are responsive at some level to all dioxins, and the traditional mass spectrometric analysis that focuses on quantifying 17 speciﬁc dioxin compounds as deﬁned above.79,1117 This research was undertaken to present a clearer understanding of the factors that diﬀerentiate the results of these methods from those of traditional GCMS. It provides suggestions for how to improve the

Received: June 23, 2011 Accepted: October 23, 2011 Revised: October 14, 2011 Published: October 23, 2011 10501

dx.doi.org/10.1021/es202149k | Environ. Sci. Technol. 2011, 45, 10501–10508

Environmental Science & Technology correlation and identiﬁes appropriate applications for using dioxin assay technologies. The National Exposure Research Laboratory (NERL) of the U.S. EPA Oﬃce of Research and Development (ORD) has conducted several phases of performance evaluation studies of various measurement technologies for dioxin and furans in soil and sediment samples. 1122 In the ﬁrst phase, ﬁve developers of immunoassay test kits and aryl hydrocarbon receptor (AhR)based assays participated in a ﬁeld demonstration in Saginaw, MI. Over 200 soil, sediment, and extract samples collected from 10 diﬀerent sites around the country with a variety of dioxin congener patterns and other distinguishing attributes, such as high levels of polychlorinated biphenyls (PCBs) and polycyclic aromatic hydrocarbons (PAHs), were analyzed by each developer. 18,19 The results of the Phase 1 study suggested that all of the technologies could be used in some capacity to screen for sample concentrations above and below threshold values (e.g., less than or greater than 1000 pg TEQ/g).1115 However, the tested technologies did not demonstrate high correlation with TEQs generated by conventional GCMS analysis. Phase 11115 generated signiﬁcant interest in whether the correlation could be improved by generating site-speciﬁc correction factors between data generated by the technologies and data generated by GCMS and then by applying the correction factor to other samples from the same site. All developers were invited to participate in the site-speciﬁc (Phase 2) study, and three developers agreed.16,17,21 Phase 2 participants included an immunoassay kit from CAPE Technologies and two aryl hydrocarbon receptor (AhR) based technologies: the Procept Rapid Dioxin Assay from Eichrom Technologies and CALUX by Xenobiotic Detection Systems (XDS). Phase 2 was conducted in the developer’s laboratories rather than a central demonstration site, since Phase 1 suggested that, while these technologies could be mobilized in a ﬁeld environment, they are primarily laboratorybased. In this phase, the developers were given a total of 112 samples that were segregated by ﬁve sites. The results for the AhR-based technologies demonstrated that a site-speciﬁc factor could be used to convert the raw data generated by the CALUX by XDS technology to signiﬁcantly improve the correlation to GCMS.21 This observation had also been reported in other studies.8,9 The data reported by CAPE Technologies’ immunoassay technique also could be site-corrected for some of the sites, but not all, and some uncertainties remained.17 Sitecorrected data from the Procept assay and GCMS results were similar in some cases; however, several large, random quantitative diﬀerences were observed that could not be easily explained.16 Due to these uncertainties, a third and ﬁnal phase of this research was aimed at better understanding the diﬀerences between biologically based dioxin assays and HRMS. Again, all of the Phase 2 developers were invited to participate. Only Eichrom Technologies accepted the oﬀer, expressing an interest in gaining a better understanding of their assay’s performance with a goal of improving the precision and accuracy of the results. Phase 3 ﬁrst focused on analyzing Procept assay extracts for the 17 dioxin and furans determined by GCMS and then attempt to identify other components that may be the source of the reproducibility issues. This goal was accomplished by reanalyzing samples from Phase 2 along with new samples from diﬀerent sites and then employing EPA Method 1613b to determine if the assay’s extracts contained dioxins comparable with the levels found in the conventional GCMS analysis of the same environmental samples. Phase 3 also utilized two-dimensional gas chromatography

ARTICLE

coupled with time-of-ﬂight mass spectrometry (GCGC TOFMS) with the intent of identifying other compounds in the assay’s extracts to gain an understanding of what may be inﬂuencing the assay’s lack of precision and accuracy.

’ EXPERIMENTAL SECTION Procept Rapid Dioxin Assay. Eichrom reported results for 132 soil and sediment samples by the Procept Rapid Dioxin Assay using a procedure detailed elsewhere.16,18 Briefly, the assay is an aryl hydrocarbon (Ah) receptor based polymerase chain reaction (PCR) assay for measuring dioxin in environmental and biological matrices. Manufacturers of dioxin bioassays commonly refer to the determinative data as bioanalytical equivalents (BEQ) to indicate that the assay produces results that are different from EPA Method 1613b toxicity equivalents (TEQ).21 In addition, Eichrom Technologies indicates that their assay will respond to more than the 17 target dioxin and furans with assigned toxic equivalent factors (TEF) as determined by Van den Berg et al. that are used to determine TEQ results.23 The assay’s response to the 17 target dioxin and furans, however, is different than their assigned TEF values.16,18 The difference between BEQ and TEQ indicates that there may be, and likely are, other compounds in the extract that are responding to the Ah receptor and are therefore included in the assay’s results. The Procept sample extraction process is similar to EPA Method 1613b, but the extract cleanup has also been designed to remove PCBs and other classes of compounds (such as PAHs and halogenated pesticides) that are known to respond to the assay. EPA Method 1613b excludes PCBs, PAHs, etc. by selected ion monitoring. In addition, a site-specific correction factor is applied to the assay results to account for process inefficiencies addressed by internal and surrogate spiking with isotopically labeled standards in EPA Method 1613b. Enhanced sample processing and site-specific correction are approaches that are intended to make the assay results comparable to the EPA Method 1613b results; however, as we will report, the precision and accuracy remain problematic even with the use of sophisticated cleanup procedures and site-specific correction factors. Assay Sample Preparation. The soil and sediment samples are extracted using accelerated solvent extraction (ASE) with 30% acetone in toluene, followed by cleanup with a multilayer acidic silica column. The dioxin extracts (in heptane) are transferred to a glass vial and the activation solution containing the Ah receptor, the aryl hydrocarbon nuclear translocator protein (ARNT), and a small DNA response element (DRE) are added. Dioxin-like molecules in the sample form complexes with the Ah receptor, ARNT, and DRE. These complexes are transferred to and immobilized on a capture strip, and excess Ah receptor, ARNT, and DRE are washed away. PCR reagents are added to the capture strips, which are then placed in a real-time PCR instrument. Inside the PCR, DNA fragments are replicated and measured by fluorescence. The output from the PCR instrument for each sample is reported in the threshold cycle (Ct). By comparing the Ct value for an unknown sample to a calibration curve of known standards, the BEQ value for a sample extract is calculated (BEQextract). The value of BEQextract can then be used to determine the assay BEQ of the soil sample using the following equation:

assayBEQ ¼ ðBEQ extract V RF=WÞ MB where the MB is the method blank (the assay BEQ value for blank sample processed through the entire method), V is the 10502

dx.doi.org/10.1021/es202149k |Environ. Sci. Technol. 2011, 45, 10501–10508

Environmental Science & Technology volume in mL of heptane in which the purified sample extract has been dissolved, W is the dry weight of the sample in grams, and RF is the recovery factor (ratio of the known TEQ of the recovery standard/measured BEQ value). The recovery standard is a sample or small group of samples from the same site, which had been analyzed by HRMS and the assay, to allow for a sitespecific calibration. Assay Extracts Preparation for HRMS Analysis. As described fully in an EPA report,10 the assay and EPA Method 1613b sample preparation procedures are similar. Unlike other screening techniques, which may have more simplified cleanup, this assay’s procedures were designed to be more rigorous so that compounds beyond the 17 target dioxin and furans that will respond in the assay are removed. As a result, the assay extract could be analyzed by EPA Method 1613b (which may not have been possible for other screening assays). The assay extracts were spiked with 17 carbon-13 labeled internal and recovery standards used in EPA Method 1613b (Cambridge Isotope Laboratory, EDF-8999 and EDF-5999), evaporated to dryness under a stream of nitrogen, redissolved in nonane, and then concentrated to a final volume of 2050 μL prior to GCHRMS analysis. Extracts were analyzed by GCHRMS following EPA Method 1613b as reported elsewhere.10,18 The concentrations of the 17 dioxin and furans of interest were measured and used to generate TEQ values for the assay extracts. GCGCTOFMS Method. GCGCTOFMS utilizes two chromatography columns coupled sequentially to achieve two chromatographic separation steps for sample components.24,25 GCGCTOFMS was employed with the intent of identifying unknown components of the extracts. Because these analyses were not solely focused on dioxin detection, the GC and MS conditions were selected to facilitate analysis of a wide range of chemical compounds. The method was setup to detect compounds in the boiling point range of C8 to C32 that will chromatograph on a nonpolar capillary column and be detected by an electron-impact MS. GCGCTOFMS operating conditions are listed in the Supporting Information, Table S-1. GCGCTOFMS analyses were conducted using a Leco Pegasus 4D GCGCTOFMS (Leco, St Joseph, MI). The Pegasus 4D utilized an Agilent 6890 GC fitted with a cryogenically cooled four-stage modulator and a secondary temperatureprogrammable oven, both mounted inside the main GC oven. As sample components exit from the first column, they were periodically trapped (every 3 s for this study), refocused, and released into the second column. The second column provided additional separation of many sample components that would coelute or be poorly separated using single column chromatography. The second column was directly connected to a Pegasus TOFMS, which was used to create full scan, electron impact (total ion chromatogram, TIC) mass spectra. Peak finding and spectral deconvolution were performed using Leco’s ChromaTOF Version 3.32 software. Since dioxins were known to be in the samples, a midlevel dioxin standard for EPA Method 1613b (200400 ng/mL) was analyzed to obtain spectral and retention time references for dioxin and furan components.

’ RESULTS AND DISCUSSION Results from the Phase 2 site-speciﬁc study of three dioxin bioassay technologies (Figure 1) led to additional research described in this paper. Figure 1a compares the Phase 2 CAPE, XDS, and Eichrom (Procept) data to the HRMS results for one

ARTICLE

Figure 1. Comparison of bioassay and HRMS data for site-speciﬁc study: (a) CAPE, CALUX by XDS, Procept, and HRMS data for Newark Bay samples; (b) CALUX by XDS, Procept, and HRMS data for samples <100 pg/g; and (c) CALUX by XDS, Procept, and HRMS data for samples >100 pg/g.

site, Newark Bay. As shown in Figure 1a, Procept data agreed well with the others for NB-3 and NB-5 but was signiﬁcantly higher for NB-1, NB-2, and NB-6. Figure 1a shows that the CAPE data agreed well with GCMS data. In Figure 1b,c, data only from the Ah-R based assays are compared for the entire data set, involving samples from seven sampling locations. The CALUX by XDS data tracked consistently well with HRMS data over the wide range of concentrations analyzed (approximately 1010 000 pg TEQ/g). Procept data agreed well with HRMS for some samples, but the inconsistent data often presented large quantitative diﬀerences. The biases were not consistently high or low. Phase 2 demonstrated that the CALUX by XDS data could be calibrated with a site-speciﬁc calibration factor, but the site-speciﬁc calibration was not consistently successful with the Procept method. These observations were the primary reason for this study and for Eichrom’s continued interest in this research. Sampling Site Selections. In Phase 3, 132 environmental samples were analyzed by the assay method from seven sampling sites. The sites were selected from an inventory of samples that had been accumulated to represent a variety of geographic regions and different sources of contamination and to include both soil and sediment matrices. Three sampling sites from Phase 2 (Solutia, Midland, and Newark Bay sites) were included in Phase 3 to allow for direct comparison with previous results. In addition, 10503

dx.doi.org/10.1021/es202149k |Environ. Sci. Technol. 2011, 45, 10501–10508

Environmental Science & Technology

ARTICLE

Table 1. Comparison of Assay BEQ Results with EPA Method 1613b TEQ Results Generated for the Same Soil/Sediment Samples assay bioanalytical equivalent (BEQ)

EPA Method 1613b toxicity equivalent (TEQ)

relative standard site Newark Bay Solutia Midland

ACW Wood Treatment ACW Residential Budd Inlet

Boat Haven

sample ID

relative standard

average concn (pg/g)

eviation (RSD)

average concn (pg/g)

deviation (RSD)

NB-3

76

35

32

6

NB-5

16

68

16

26

S-1

845

23

846

18

S-2

146

25

48

10

M-1

340

40

239

5

M-3

144

39

184

5

M-4 WT-7

99 213

43 15

149 333

7 6

WT-8

104

25

159

1

R-2

115

18

129

3

R-4

5

30

5

7

BI-C4

37

39

30

14

BI-S7

45

27

48

2

BI-C13

40

21

20

22

H-1 H-2

33 18

32 31

46 21

11 5

HS-2

37

38

35

18

H-8

28

19

51

4

H-9

52

33

49

19

four new sampling sites (Boat Haven, Budd Inlet, ACW Residential, and ACW Wood Treatment) were added to expand upon the types of dioxin profiles and matrices included in the Phase 3 study. A complete description of the sites has been presented elsewhere.18 The seven sites contained complex contamination proﬁles that were expected to be challenging to the assay. Table S-2 of the Supporting Information shows the congener distribution for one example sample from each of the seven sites as a percentage of the total TEQ. Three of the seven sites (Midland, Solutia, and Newark Bay) had 2,3,7,8- tetrachlorodibenzo-p-dioxin (TCDD) as the predominant congener, accounting for 4585% of the total TEQ concentration in a representative sample; one site (Budd Inlet) had 1,2,3,7,8-pentachlorodibenzo-p-dioxin (PeCDD) as the predominant congener at 82%; and three sites (Boat Haven, ACW Residential, and ACW Wood Treatment) had 1,2,3,4,6,7, 8-heptachlorodibenzo-p-dioxin (HpCDD) as the predominant congener at 24%0% of the total TEQ concentration. Comparison of Assay BEQ to HRMS TEQ. It is important to begin with the definition of terms for describing the results in Tables 1 and 2. “assay BEQ” represents the original results from the Procept assay. “HRMS TEQ” results are the data generated for EPA Method 1613b for samples prepared for traditional analyses. “Assay TEQ” are results for the same samples that were used to generate the original Procept BEQ results that were also analyzed by HRMS to generate TEQ data for these extracts. Table 1 shows the comparison of the assay BEQ results with EPA Method 1613b TEQ results at the seven sampling sites. Four replicates were analyzed by both the assay and EPA Method 1613b. As shown in Table 1, there were several cases in which the average results were similar. However, relative standard deviations (RSDs) ranged from 15% to 68%, and the assay BEQ measurements were not as reproducible as the EPA Method 1613b TEQ measurements, for which RSDs were all less than

26% (and all but two RSD values were less than 20%). Speciﬁc sample extracts in which the BEQ result did and did not compare well with the corresponding TEQ result, shown in Table 1, were selected for further evaluation by HRMS to determine if inconsistencies in the assay’s results were due to ﬂuctuations in dioxin levels or to other interferents. For example, as shown in Table 1, sample extracts from Newark Bay were selected where the average BEQ concentration agreed with the average TEQ concentration (for sample NB-5, where both concentrations were 16 pg/g), and where the average BEQ concentration was signiﬁcantly diﬀerent than the average TEQ concentration (for sample NB-3, where the BEQ concentration was 76 pg/g and the TEQ concentration was 32 pg/g). It should be noted that a separate evaluation of assay data comparisons to GCMS reported elsewhere18 demonstrated that the BEQ results had <10% false positive and <10% false negatives relative to the interim remediation goal of 72 pg TEQ/g, indicating that BEQ results could be used to screen samples as being above or below threshold levels. However, the current research demonstrated that a direct correlation with GCMS was not achieved, so further analyses were pursued to better understand the discrepancies. Comparison of Assay TEQ to HRMS TEQ. A total of 38 assay extracts were analyzed to determine the “assay TEQ” value. These analyses were aimed at determining if dioxin levels were consistently achieved in the extraction but not reported as such by the Procept assay. Table 2 presents the comparison of assay BEQ results (from Table 1) with assay TEQ results generated for the 38 individual extracts, along with the site-specific assay BEQ correction factor that was applied for each site. As shown in Table 2, the assay TEQ results did not replicate the assay BEQ results in 31 of the 38 extracts. For example, as shown in Table 2, two replicate samples for sample NB-3 produced BEQ results of 113 and 53 pg/g. When these same assay extracts were analyzed by EPA Method 1613b for TEQ, the results were 21 and 19 pg/g. 10504

dx.doi.org/10.1021/es202149k |Environ. Sci. Technol. 2011, 45, 10501–10508

Environmental Science & Technology

ARTICLE

Table 2. Comparison of Selected Assay TEQ Results with Assay BEQ Results concentration (pg/g) site Newark Bay

sample ID NB-3

21

53

19

NB-5

142a 28

9.4 9.0

11

10.5

S-1

1120

430

712

320

145

38

102

47

M-1

508

86

M-3

215 190

115 114

assay BEQ correction factor: 69 S-2 Midland assay BEQ correction factor: 0.44

M-4 ACW Wood Treatment

100

61

82 92

WT-8

252 131

190 105

81

89

R-2

139

95

93

93

4.0

4.0

3.6

3.6

BI-C4

19

16

BI-S7

54 60

20 41

38

52

14

16

BI-C13

a

90

198

assay BEQ correction factor: 0.87

assay BEQ correction factor: 0.47

147

186

R-4

Boat Haven

162

102

WT-7

assay BEQ correction factor: 0.88

Budd Inlet

193

158 assay BEQ correction factor: 0.38

ACW Residential

assay TEQ

113

assay BEQ correction factor: 0.41

Solutia

assay BEQ

52

9

H-1

42

27

H-2

20

15

HS-2

16

20

H-8

46 27

14 21

H-9

42

28

73

21

Bold samples were subsequently analyzed by GCGCTOFMS.

These findings were encouraging, since this data indicated that the 17 target dioxin and furans were being effectively isolated in the Procept method and that the Procept preparation method was capable of producing consistent TEQ results. As shown in Figures S-2 of the Supporting Information, the TEQ value for extracts prepared for the Procept assay measured by GCHRMS correlated very well with the TEQ value measured for samples prepared using the GCHRMS method (R2 = 0.96) as compared to the correlation between the Procept BEQ and GCHRMS results (R2 = 0.84). The Procept TEQ results were not expected to exactly match the EPA Method 1613b TEQ results, since EPA Method 1613b utilizes recovery

Figure 2. Assay average TEQ results (line) compared to 4080% (markers) of EPA Method 1613b average TEQ results. One to four replicates were analyzed for each sample, with a total of 38 analyses in all. Average data are presented for clarity of presentation.

standards to correct the results for ineﬃciencies in the sample preparation process. On the basis of experience with the method, Eichrom indicated dioxin recoveries for the Procept method are expected to be 4080%. Figure 2 is a graph showing that the average Procept TEQ results were within 4080% of the average EPA Method 1613b TEQ results in all but two cases (samples S-2 and BI-S7), where the average Procept TEQ and EPA Method 1613b TEQ results were similar, and therefore the Procept values were above the 80% marker and found to be consistent with method expectations. The average relative standard deviation (RSD) of the EPA Method 1613b TEQ analyses was 10%, and the average precision for the assay BEQ values was 32%. The average RSD for assay extracts measured by HRMS (assay TEQ RSD values) was 19%, which was an improvement over the overall precision for the BEQ measurements (32% RSD). With the exception of two sample sets (BI-C13 and S-1), the reproducibility of the average assay results signiﬁcantly improved when the extracts were analyzed by HRMS. For BI-C13, the assay TEQ values were 9 and 16 pg/g, which resulted in a high %D (52%), but the magnitude of the diﬀerence between these samples (7 pg/g) was small. For S-1, the %D value was 29% for the assay TEQ values and 23% for Procept BEQ values. With these exceptions, the general improvement of precision from 32% RSD for the Procept BEQ values to 19% for assay TEQ values further indicated that there were other contributions to the assay BEQ results that caused variability in the data. GCGCTOFMS Analysis of Assay Extracts for Interfering Compounds. Prior to employing GCGCTOFMS, a cursory study was undertaken to identify other components in the assay extracts. Full-scan low-resolution mass spectrometry [both electron impact (EI) and negative chemical ionization (NCI) modes] were used to compare the components identified in extracts that were prepared for analysis by the Procept method and those that were prepared for EPA Method 1613b analysis.26 The purpose of this study was to provide a preliminary understanding of the types of compounds that were contained in the assay extracts. In general, the assay extracts were found to be much cleaner than the EPA Method 1613b extracts, with overall much lower background levels. That fewer chromatographic peaks were observed was not unexpected, since assay sample extracts underwent a rigorous cleanup. It was also not unexpected that the 10505

dx.doi.org/10.1021/es202149k |Environ. Sci. Technol. 2011, 45, 10501–10508

Environmental Science & Technology EPA Method 1613b extracts would have more background contaminants, because the analytical principles employed to isolate contaminants from the 17 target dioxin and furans are not employed in the assay method (namely, gas chromatographic separation of the desired compounds and m/z filtering by mass spectrometry). The major compounds identified during the EI analysis were similar for the Eichrom and EPA Method 1613b extracts, with alkanes, alkenes, silicones, and phthalates tentatively identified in all but one of the sample extracts. Eichrom indicated that the identified compounds would not produce a response to the Procept assay. The cursory review indicated that a more powerful analytical technique would be needed in order to increase the likelihood of compound identification. GCGC TOFMS was chosen due to its greatly increased separation capability and sensitivity compared to traditional single-column GCMS techniques.24,25 Ten of the 38 assay extracts were analyzed by GCGC TOFMS (identiﬁed in Table 2). Selections represented consistent and variable assay BEQ replicate results. As shown in Table 2, two types of quantitative variability were observed in the BEQ results: (1) diﬀerences between assay BEQ replicate results and (2) diﬀerences between assay BEQ and EPA Method 1613b results. Diﬀerences between the assay replicate results were investigated for selected samples (M-3, NB-5, S-1, and S-2) by analyzing sample replicate pairs representing one higher and one lower assay BEQ response. Samples NB-5, S-1, and S-2 were also selected to examine diﬀerences between the EPA Method 1613b results and the assay results. For evaluation of background levels, samples R-4 and H-2 were chosen to represent samples that had consistent assay BEQ replicate results and/or consistent assay BEQ vs EPA Method 1613b results. Note that, in this context, the “background level” is deﬁned as all other compounds and matrix inﬂuences identiﬁable in the sample beyond the 17 dioxin and furans targeted in EPA Method 1613b. For all samples analyzed by GCGCTOFMS as replicate pairs (M-3, NB-5, S-1, and S-2), each replicate contained many of the same components as its pair, except NB-5. One replicate each of sample pairs M-3, S-1, and S-2 contained higher levels of the observed components compared to the other replicate based on the GCGCTOFMS total ion and extracted ion responses. An example of this diﬀerence in response is depicted for sample M-3 (replicates 2 and 4) in Figure 3, which shows a region for ion channel m/z 306 (monitoring for tetrachlorodibenzofuran and hexachlorobiphenyl ether) overlaid on the same scale. This ion channel was selected for Figure 3 because it best visually demonstrated that replicate sample responses are diﬀerent by approximately a factor of 2. This trend in sample concentrations was consistent in all ion channels and the total ion chromatogram for these sample pairs (although best viewed at the instrument rather than a single dimensional plot; hence, the best ion channel for visualization in this paper was selected). Instrumental variation was eliminated by performing repeat injections, which showed that the results were reproducible. Similar response variations were seen in the comparison of assay TEQ results for M-3 (Ssee Table 2: M-3 replicate values were 162 and 100 pg TEQ/g). Similar trends for the GCGCTOFMS data were observed between the two sample replicates from samples S-1 and S-2; however, these trends were not reﬂected in the assay TEQ results for the S-2 sample pair. In that case, one replicate was slightly lower than the other replicate (See Table 2: S-2 replicate results: 38 and 47 pg TEQ/g). Since the TOFMS diﬀerences in the sample S-2 assay replicates are not reﬂected in

ARTICLE

Figure 3. Overlaid (single dimension) TOFMS chromatograms for ion channel 306 for assay sample M-3 replicates showing nearly twice as much signal when comparing the response for one replicate to another.

the assay TEQ data, something other than concentration diﬀerences is confounding these results. For sample NB-5, assay BEQ results for one replicate (142 pg BEQ/g) were notably higher than assay TEQ results (10 pg TEQ/g); the assay TEQ result agreed with the EPA Method 1613b TEQ (average 16 pg TEQ/g). For the other NB-5 replicate, assay BEQ (11 pg BEQ/g) and assay TEQ (10 pg TEQ/g) results also agreed with the EPA Method 1613b result. GCGCTOFMS analyses of the signiﬁcantly higher NB-5 replicate revealed PAH and thiophene components that were not detected in the other replicate. Procept is known to respond to some PAHs. Relative assay BEQ responses are based on a scale of 01, where 1 represents a response equivalent to that of 2,3,7,8TCDD.16,18,27 Two PAHs known to respond to the Procept assay (benzo[a]anthracene, Procept response = 0.05, and benzo[k]ﬂuoranthene, Procept response = 0.5) were detected in one sample NB-5 replicate. Numerous other PAHs were detected in this sample NB-5 replicate and identiﬁed primarily as alkylated PAHs (e.g., methyl-, dimethyl-, trimethyl-, and propylphenanthrenes and -anthracenes); however, the response of Procept to these alkylated PAH compounds is not documented. PAH concentrations were not measured in these samples so it cannot be ascertained from the existing data whether the diﬀerences in Procept BEQ responses between the two NB-5 replicates are due to the presence of PAHs. On the basis of the presence of benzo[k]ﬂuoranthene and the large numbers of other PAH components in one NB-5 replicate and the lack of these components in another NB-5 replicate, PAHs are a potential cause of the increased assay BEQ in the NB-5 replicate with a signiﬁcantly higher result. The presence of PAHs in this NB-5 replicate is believed to be due to a sample preparation error where the cleanup was accidentally omitted or there was some failure of the silica puriﬁcation procedure. It has been reported that solvents used in the Procept cleanup may contribute to the assay response.27 Two-dimensional contour maps were used to examine overall diﬀerences between samples. Total ion response contour maps for all assay samples in the GCGCTOFMS portion of the study (M-3, NB-5, S-1, S-2, R-4, and H-2) were examined, including the heptane process blank, which was a nonmatrix sample that was carried through the entire sample preparation procedure prior to assay analysis. 10506

dx.doi.org/10.1021/es202149k |Environ. Sci. Technol. 2011, 45, 10501–10508

Environmental Science & Technology The two-dimensional proﬁle for the heptane blank showed responses from numerous high concentration aliphatic hydrocarbon components. This area should be relatively consistent or at least show no diﬀerence for all samples, but in each of the extract samples it was highly variable. As described in the Experimental Section, prior to assay analysis, all samples undergo several cleanup and solvent exchange steps. It is possible that diﬀerences in the execution of these steps may lead to variability due to diﬀerences in sample contents. It should be noted, however, that these were mostly aliphatic hydrocarbons, compounds that would not be expected to aﬀect assay BEQ results; however, the ﬁndings show that the method has signiﬁcant sensitivity to inconsistencies in the sample preparation process. The HRMS analyses of Procept assay extracts demonstrated that the extraction and cleanup procedure used in the assay captures the dioxin congeners and correlates better to the conventional GCMS method than the direct assay results. Understanding that the dioxin levels in the assay extracts were consistent and accurate with HRMS results was an important ﬁnding, but it did not identify the reasons for the inconsistency of the assay BEQ results. Since HRMS analysis utilizes selected ion monitoring to screen out only the dioxin compounds of interest, the presence of other AhR-active compounds in the assay extracts must be contributing to the diﬀerences in the BEQ and EPA Method 1613b values. GCGCTOFMS was used to investigate this. GCGCTOFMS conﬁrmed that the overall levels of contamination in the assay extracts were variable. In some cases, the results were consistent (i.e., relative high/low) with what would be expected on the basis of the assay BEQ analysis. In one case, PAHs that had not been adequately removed during sample cleanup were identiﬁed. Other samples showed high and variable solvent background levels in the GCGCTOFMS analysis. Since the variability and inconsistency of assay BEQ results was more linked to sample preparation than to the method’s ability to isolate the dioxin that is the target of the analysis, inconsistencies could be avoided if the assay sample preparation procedures could be optimized. If these inconsistencies can be overcome, Procept would be a rapid and cost-eﬀective alternative, or at a minimum an important complement, to traditional GCMS analysis. However, in its current format, the Procept Rapid Dioxin Assay should primarily be utilized for screening purposes, for example, to categorize dioxin concentrations as “high” and “low” or above/below regulatory levels, as has been demonstrated in other studies,18 without expectation that a direct correlation with HRMS can be achieved.

’ ASSOCIATED CONTENT

bS Supporting Information. Table S-1, GCGCTOFMS operating conditions and the dioxin congener proﬁles of the sampling sites; Figure S-1, Procept BEQ data versus HRMS TEQ and Procept TEQ data versus HRMS TEQ, demonstrating the improvement in correlation when the Procept extracts were analyzed by HRMS. This material is available free of charge via the Internet at http://pubs.acs.org. ’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]; phone: 702-798-2232; fax: 702-798-2107.

ARTICLE

’ ACKNOWLEDGMENT The U.S. EPA, through its Oﬃce of Research and Development, funded and managed the research described here under Contract No. EP-C-05-057 to Battelle. It has been subjected to Agency review and approved for publication. The authors acknowledge the participation of Eichrom Technologies and in particular the support from Daniel McAlister and Larry Jassin. Mary Schrock of Battelle provided helpful review comments during the preparation of the manuscript. ’ REFERENCES (1) U.S. EPA, 2009. Draft Recommended Interim Preliminary Remediation Goals for Dioxin in Soil at CERCLA and RCRA Sites. Oﬃce of Superfund Remediation and Technology Innovation, Washington, DC, 2009; www.epa.gov/superfund/policy/remedy/pdfs/ Interim_Soil_Dioxin_PRG_Guidance_12-30-09.pdf. (2) Gizzi, G; Hoogenboom, L. A.; Von Holst, C; Rose, M; Anklam, E. Determination of dioxins (PCDDs/PCDFs) and PCBs in food and feed using the DR CALUX bioassay: Results of an international validation study. Food Addit. Contam. 2005, 22 (5), 472–81. (3) Scippo, M.; Soledad Rybertt, M.; Focant, J.; Eppe, G.; Massart, A.-C.; De Pauw, E.; Maghuin-Rogister, G. Evaluation of the DR-CALUX screening of food and feed, according to regulation levels including DLPCB. Organohalogen Compd. 2005, 67, 1397–1402. (4) Hasegawa, J; Guruge, K. S.; Seike, N; Shirai, Y; Yamata, T; Nakamura, M; Handa, H; Yamanaka, N; Miyazaki, S. Determination of PCDD/Fs and dioxin-like PCBs in ﬁsh oils for feed ingredients by congener-speciﬁc chemical analysis and CALUX. Chemosphere 2007, 69 (8), 1188–94. (5) Hoogenboom, L; Traaga, W.; Boveea, T.; Goeyensa, L.; Carbonnellea, S.; van Locoa, J.; Beernaerta, H.; Jacobsa, G.; Schoetersa, G.; Goeyensa, L.; Baeyensa, W. The CALUX bioassay: Current status of its application to screening food and feed. Trends Anal. Chem. 2006, 25 (4), 410–420. (6) Nording, M.; Sporring, S.; Wiberg, K.; Bj€orklund, K.; Haglund, P. Monitoring dioxins in food and feedstuﬀs using accelerated solvent extraction with a novel integrated carbon fractionation cell in combination with a CAFLUX bioassay. Anal. Bioanal. Chem. 2007, 381 (7), 1472–1475. (7) Nording, M.; Nichkova, M.; Spinnel, E.; Persson, Y.; Gee, S. J.; Hammock, B. D.; Haglund, P. Rapid screening of dioxin-contaminated soil by accelerated solvent extraction/puriﬁcation followed by immunochemical detection. Anal. Bioanal. Chem. 2006, 385, 357–366. (8) Nording, M.; Denison, M.; Baston, D.; Persson, Y.; Spinnel, E.; Haglund, P. Analysis of dioxins in contaminated soils with the CALUX and CAFLUS bioassays, an immunoassay, and gas chromatographyhigh-resolution mass spectrometry. Environ. Toxicol. Chem. 2007, 26 (6), 1122–1129. (9) Trindade, M.; Nording, M.; Nichkova, M.; Spinnel, E.; Haglund, P.; Last, M.; Gee, S.; Hammock, B.; Last, J.; Gonzelez-Spienza, G.; Brena, B. Enzyme-linked immunosorbent assay for screening dioxin soil contamination by uncontrolled combustion during informal recycling in slums. Environ. Toxicol. Chem. 2008, 27 (11), 2224–2232. (10) U.S. EPA. EPA Method 1613B. Dioxins, Tetra- thru Octa(CDDs) and Furans (CDFs), EPA/821/B-94-005, 40 Code of Federal Regulations Part 136, Appendix A, October 1994. (11) U.S. EPA. Innovative Technology Veriﬁcation Report, Xenobiotic Detection Systems, CALUX by XDS; EPA/540/R-05/001, July 2005. (12) U.S. EPA. Innovative Technology Veriﬁcation Report, Wako Pure Chemical Industries, Dioxin ELISA Kit; EPA/540/R-05/002, March 2005. (13) U.S. EPA. Innovative Technology Veriﬁcation Report, Abraxis, Coplanar PCB ELISA Kit; EPA/540/R-05/003, March 2005. (14) U.S. EPA. Innovative Technology Veriﬁcation Report, CAPE Technologies, DF1 Dioxin/Furan Immunoassay Kit and PCB TEQ Immunoassay Kit; EPA/540/R-05/004, March 2005. 10507

dx.doi.org/10.1021/es202149k |Environ. Sci. Technol. 2011, 45, 10501–10508

Environmental Science & Technology

ARTICLE

(15) U.S. EPA. Innovative Technology Veriﬁcation Report, Hybrizyme Corporation, AhRC PCR Kit; EPA/540/R-05/005, March 2005. (16) U.S. EPA. Interim Report on the Evolution and Performance of the Eichrom Technologies Procept Rapid Dioxin Assay for Soil and Sediment Samples; EPA/540/R-07/001, January 2007. (17) U.S. EPA. Performance of the CAPE Technologies DF1 Dioxin/ Furan Immunoassay Kit for Soil and Sediment Samples; EPA/540/R-08/ 002, February 2008. (18) U.S. EPA. Final Report on the Performance of the Eichrom Technologies Procept Rapid Dioxin Assay for Soil and Sediment Samples; EPA/540/R-11/002, September 2011. (19) U.S. EPA. Demonstration and Quality Assurance Project Plan; EPA/600/R-04/036, April 2004. (20) Dindal, A.; Billets, S. Experimental design considerations when verifying the performance of monitoring technologies for dioxin and dioxin-like compounds in soils and sediments. Chemosphere 2008, 73, S66–S71. (21) Dindal, A.; Thompson, E.; Aume, L.; Billets, S. Application of site-speciﬁc calibration data using the CALUX by XDS bioassay for dioxin-like chemicals in soil and sediment. Environ. Sci. & Tech. 2007, 41, 8376–8382. (22) Schrock, M.; Dindal, A.; Billets, S. Evaluation of alternative approaches for screening contaminated sediments and soils for PCDD/ PCDF. J Environ. Manage 2009, 90, 1289–1295. (23) van den Berg, M.; Birnbaum, L.; Denison, M.; De Vito, M.; Farland, W.; Feeley, M.; Fiedler, H.; Hakansson, H.; Hanberg, A.; Haws, L.; Rose, M.; Safe, S.; Schrenk, D.; Tohyama, C.; Tritscher, A.; Tuomisto, J.; Tysklind, M.; Walker, N.; Peterson, R. The 2005 Word Health Organization reevaluation of human and mammalian toxic equivalency factors for dioxins and dioxin-like compounds. Toxicol. Sci. 2006, 93 (2), 223–241. (24) Ong, R.; Marriott, P.; Morrison, P.; Haglund, P. Inﬂuence of chromatographic conditions on separation in comprehensive gas chromatography. J. Chromatogr. A 2002, 962 (12), 135–152. (25) Dall€uge, J.; Vreuls, R. J. J.; Beens, J.; Brinkman, U. A. T. Optimization and characterization of comprehensive two-dimensional gas chromatography with time-of-ﬂight mass spectrometric detection (GCGCTOF MS). J. Sep. Science 2002, 25 (4), 201–214. (26) Unpublished EPA report, EPA Contract # EP-C-05-057, Task Order 43. LRMS analysis of Eichrom Extracts. February 2007. (27) Cariou, R.; McAlister, D.; Marchand, P.; Fern, M.; Antignac, J.-P.; Le Bizec, B. Inﬂuence of the solvent quality on the AhR mediated procept assay measurement of dioxin and dioxin-like compounds. Talanta 2010, 80, 2063–2067.

10508

dx.doi.org/10.1021/es202149k |Environ. Sci. Technol. 2011, 45, 10501–10508

ARTICLE pubs.acs.org/est

Sampling Medium Side Resistance to Uptake of Semivolatile Organic Compounds in Passive Air Samplers Xianming Zhang,† Masahiro Tsurukawa,‡,§ Takeshi Nakano,‡,§,|| Ying D. Lei,† and Frank Wania*,† †

)

Department of Chemistry and Department of Physical and Environmental Sciences, University of Toronto Scarborough, Toronto, Ontario M1C 1A4, Canada ‡ Hyogo Prefectural Institute of Environmental Sciences, 3-1-27, Yukihira-cho, Suma-ku, Kobe 654-0037, Japan § Graduate School of Engineering, Osaka University, 2-1, Yamadaoka, Suita 565-0871, Japan Graduate School of Maritime Science, Kobe University, 5-1-1, Fukaeminamimachi, Higashinada-ku, Kobe 658-0022, Japan

bS Supporting Information ABSTRACT: Current theory of the uptake of semivolatile organic compounds in passive air samplers (PAS) assumes uniform chemical distribution and no kinetic resistance within the passive sampling media (PSM) such as polystyrene-divinylbenzene resin (XAD) and polyurethane foam (PUF). However, these assumptions have not been tested experimentally and are challenged by some recently reported observations. To test the assumptions, we performed kinetic uptake experiments indoors using cylindrical PSM that had been concentrically segmented into three layers. Both XAD and PUF were positioned in the same type of sampler housing to eliminate the variation caused by the diﬀerent housing designs, which enabled us to quantify diﬀerences in uptake caused by the properties of the PSM. Duplicated XAD (PUF) samples were retrieved after being deployed for 0, 1 (0.5), 2 (1), 4 (2), 8 (4), 12 (8), and 24 (12) weeks. Upon retrieval, the PSM layers were separated and analyzed individually for PCBs. Passive sampling rates (R) were lower for heavier PCB homologues. Within a homologue group, R for XAD was higher than that for PUF, from which we infer that the design of the “cylindrical can” housing typically used for XAD PAS lowers the R compared to the “double bowl” shelter commonly used for PUF-disk PAS. Outer layers of the PSM sequestered much higher levels of PCBs than inner layers, indicative of a kinetic resistance to chemical transfer within the PSM. The eﬀective diﬀusivities for chemical transfer within PSM were derived and were found negatively correlated with the partition coeﬃcients between the PSM and air. Based on the results, we conclude that the PSM-side kinetic resistance should be considered when investigating factors inﬂuencing R and when deriving R based on the loss of depuration compounds.

’ INTRODUCTION Dynamic-uptake based passive air samplers (PAS) such as those based on polystyrene divinylbenzene (XAD)1 and polyurethane foam (PUF)2 are increasingly used to study persistent semivolatile organic compounds (SVOCs) in the atmosphere. Such PAS are capable of time-integrated sampling with relatively low cost and simple operation, which is independent from power supply and free of noise.13 Because of these advantages over the traditional high-volume air sampler, PAS are widely applied to understand spatial and long-term temporal trends, identify sources, and assess human exposure to SVOCs in various types of environments.46 The mechanism of uptake in PAS is based on the molecular diﬀusion from air to passive sampling medium (PSM). Conceptually, the process of SVOC uptake in PAS has been described using the two-ﬁlm diﬀusion model,2,7 which was originally proposed to describe mass transfer across gasliquid interfaces.7 According to the two-ﬁlm model, “in the main body of either liquid and gas, ...the concentration of solute in the ﬂuid is essentially uniform at all r 2011 American Chemical Society

points”.7 As indicated by the current PAS “theory”,2,3 the kinetic resistance within the PSM is inversely related to a chemical’s PSM/ air partition coeﬃcient and thus negligible for SVOCs due to their large PSM/air partition coeﬃcients. Therefore, the resistance posed by the air boundary layer is regarded as controlling the rate of SVOC uptake in PAS. During the initial uptake stage (operationally deﬁned as the linear uptake range), chemical concentrations on the PSM are so low that surface evaporation is neglible. As such, chemical uptake in PAS can be quantiﬁed with a simple linear equation involving a sampling rate (R, m3/d) that only depends on the surface area of the PSM, the chemical’s molecular diﬀusivity in air (DA), and the boundary layer thickness.3 Because the boundary layer thickness is diﬃcult to quantify directly, in practice, R is usually determined by Received: June 30, 2011 Accepted: November 2, 2011 Revised: October 25, 2011 Published: November 02, 2011 10509

dx.doi.org/10.1021/es2032373 | Environ. Sci. Technol. 2011, 45, 10509–10515

Environmental Science & Technology calibrations against air concentrations determined using active samplers. A number of PAS calibration studies have determined R for both XAD-PAS and PUF-PAS under diﬀerent environmental conditions.1,812 Based on these studies, XAD-PAS have a higher sampling capacity or longer linear uptake range than PUF-PAS.13 The high capacity makes XAD-PAS superior for integrated sampling over long time periods, especially for relatively more volatile compounds such as the ﬂuorotelomer alcohols.14 However, XAD-PAS generally have a 2- to 5-fold lower R than PUFPAS. So far, it is unclear whether the diﬀerent R is caused by diﬀerences in the properties of XAD and PUF or by diﬀerences between the housing conﬁgurations typically employed with the two PAS. R for both of the PAS varies among diﬀerent chemicals or at diﬀerent temperatures.911,15 Such variations are larger than can be explained by the dependence of DA on chemical properties or temperature (FullerSchettlerGiddings equation),16 indicating some other inﬂuential factors may exist. For the PUFPAS, some studies observed higher R for chemicals with lower volatility,11,17 an observation attributed to the binding of such chemicals to particles, which are trapped by the PUF. Conﬂicting results showing lower R for particle-bound chemicals have also been found.15 Previous studies on the temperature dependence of the R for PUF-PAS also yielded inconsistent results. Increased R for some particle-bound PAHs was observed as temperature increases, which was explained with a shifting from particle to gas phase at higher temperatures.11,15 However, a negative correlation was found for BDE-99, which is also likely to undergo gasparticle phase exchange.17 Calibrations for selected pesticides conducted at diﬀerent latitudes for XAD-PAS yielded higher R at higher temperatures.1,9 However, this cannot be due to shifts in the atmospheric phase distribution because the gasparticle exchange behavior of these pesticides is not sensitive to temperature in the environmental temperature range. Here, we hypothesize that SVOCs distribute nonuniformly within the PSM and the PSM-side kinetic resistance could also aﬀect R. This resistance might help explain the variation of R among chemicals and with temperatures. To explain the variation of R with sampling time, Chaemfa et al.8 postulated a two-phase uptake processes: chemicals ﬁrst sorb to the surface of PUF and then penetrate into the PUF at a slower rate. This is essentially similar to our PSM-side kinetic resistance hypothesis. However, no further investigation has sought to conﬁrm this hypothesis that challenges the current PAS uptake theory. In this study, we aim to (1) investigate whether PSM or housing diﬀerences cause the diﬀerent sampling rate between XAD-PAS and PUF-PAS, (2) test our hypothesis on chemical distribution and kinetic resistance within PSM, and (3) quantify the eﬀective diﬀusivity of chemical transfer within the PSM. To achieve these objectives, we performed a kinetic uptake experiment using concentrically segmented XAD and PUF positioned in the same type of sampler housing.

’ MATERIALS AND METHODS Passive Sampling Media. XAD packed in mesh cylinders and PUF were selected for this study because they are the most widely used passive sampling media (PSM) for SVOCs in air. Instead of the PUF disk commonly used in the “double-bowl” type PAS,2 a cylindrical PUF plug (8 cm diameter, 8 cm high) was made from 1-cm-thick PUF sheets (Pacwill Environmental,

ARTICLE

Figure 1. Design of the layered passive air sampling media (XAD and PUF) used to study the distribution of PCBs within the passive sampling medium.

density ∼0.02 g/cm3) and placed in the “cylindrical can” housing commonly used with XAD-PAS (Figure 1) to eliminate the influence of sampler housing design when comparing the uptake characteristics of the two PSM. The XAD-filled mesh cylinder and cylindrical PUF were concentrically segmented into three layers (outer, mid, and inner). The PSM layers can be separated upon sample retrieval. Detailed dimensions of the PSM are given in Figure 1. Before sampling, the segmented PUF components were sequentially cleaned with soap water and deionized water, and Soxhlet-extracted with acetone for 24 h and with petroleum ether for another 24 h. The XAD-2 resin was purchased precleaned (Sigma-Aldrich). Chemicals. Polychlorinated biphenyls (PCBs) were selected as the target chemicals for this study because the PCB congeners cover a wide range of partitioning properties (e.g., PSM/air partition coefficient), which also partially overlaps with other SVOCs of environmental interest such as organochlorine pesticides, polycyclic aromatic compounds, and brominated flame retardants. PSM (XAD and PUF)/air partition coefficients for individual PCB congeners, estimated using poly parameter linear free energy relationships13,18 and recently updated PCB solute descriptors,19 were compiled in Tables S1 and S2 in the Supporting Information (SI) and were used for further data analysis. Sampling Design. Before deployment, the three layers of PSM were spiked with three different groups of depuration compounds (DCs) comprised of 13C-labeled PCB congeners or nonlabeled PCB congeners that are not present in ambient air. Different groups of DCs were applied to different PSM layers. Detailed information on DCs and spiking procedure is provided in the SI. An unoccupied office previously identified as being contaminated with PCBs was selected as the sampling site. Duplicated XAD (PUF) samples were retrieved after been deployed for 0, 1 (0.5), 2 (1), 4 (2), 8 (4), 12 (8), and 24 (12) weeks. Deployment lengths for PUF-PAS were shorter, because we had anticipated faster uptake than for XADPAS. Upon retrieval, the layered PSMs were separated, individually sealed in precleaned aluminum foil and Ziploc bags, and stored at 20 °C before extraction within 2 weeks. Along with the PAS, a low-volume active sampler (BGI Inc., 2.9 ( 0.2 m3/d) with a PUFXAD-PUF sandwich (5 g of XAD between two 2-cm i.d. 3-cm PUF plugs) as the sampling medium was used to measure the PCB air concentrations with monthly resolution. The sampling scheme is illustrated in Figure S1. Sample Extraction and Analysis. Each sample was Soxhlet extracted for 24 h in ∼500 mL of petroleum ether (PUF) or 1:1 10510

dx.doi.org/10.1021/es2032373 |Environ. Sci. Technol. 2011, 45, 10509–10515

Environmental Science & Technology

ARTICLE

acetone/hexane (XAD and PUF-XAD-PUF sandwiches). The extract was rotoevaporated to ∼2 mL and eluted through a disposable pasteur pipet packed with dehydrated sodium sulfate to remove moisture. The eluent was blown down with high purity N2, solvent exchanged to iso-octane, and reduced to ∼0.5 mL in a GC vial, to which 100 ng of mirex was added for volume correction and as internal standard for PCB quantification. PCBs in the samples were analyzed with an Agilent 5890 gas chromatograph (GC) coupled with a JMS-800 double focusing high-resolution mass spectrometer (HRMS, resolution g60 000). The detailed method for instrumental analysis is described by Matsumura et al.20 Brieﬂy, 1.0 μL of the sample was injected in splitless mode with the injector temperature at 280 °C. PCBs in the sample were separated using an HT8-PCB column (0.25 mm i.d. 60 m, SGE Analytical Science) with helium (1 mL/min) as the carrier gas. The GC oven was programmed from 120 to 180 at 20 °C/min, to 260 at 2 °C/min, to 300 at 5 °C/min, and then held isothermal for 4 min. The HRMS was operated under EI and SIM mode with the interface and chamber operated at 260 °C. QA/QC. All samples were duplicated to quantify reproducibility. Data analysis for all samples was based on both duplicates except for the XAD 12-month inner layer, of which one duplicate was lost during sample preparation. The relative difference between the passive sampling rates derived from duplicated samples was generally less than 10% (Figure S2). Duplicated field blanks for both XAD- and PUF-PAS were treated as time zero values in the analysis of chemical uptake kinetics. Prior to extraction, each sample was spiked with 100 μL of 250 pg/μL 13 C12-labeled PCB-77, -101, -141, and -178 (Cambridge Isotope) as surrogate standards. Recoveries of the four surrogate standards ranged between 74 and 131% with an interquartile range <15% (Figure S3). Derivation of Passive Air Sampling Rates. Passive air sampling rates (R, m3/d) and the PSM-side effective diffusivities (DE, m2/h) were obtained by linear least-squares fitting (LLSF) to all duplicated data points. For data below the LOD, random numbers between 0 and LOD were assigned.21,22 The method of using LLSF to derive R has been applied in other studies.2,8 Briefly, R equals the slope of the linear least-squares fitted line of the equivalent sampling volume (Veq) over the PAS deployment time; Veq is calculated as the amount of a chemical sequestered in the PSM (sum of the three layers) divided by the ambient air concentration measured using the active air sampler. Derivation of the Effective Diffusivities on the PSM Side. To derive DE on the PSM side, a two-layered PSM mass balance model was developed (Figure S4, eqs S6S14, and the relevant text in SI). The outer layer in the above-mentioned experiments is referred to as Layer 1; since few PCB congeners were detected in the inner layer, the inner and mid layers in the experiment were combined and are referred to as Layer 2 hereafter. Starting from the chemical mass balance equations (eqs S6 and S7) for the two layers, a relationship between the amounts of chemical sequestered in Layers 1 and 2 was derived m2 ðtÞ ¼

1 A2 kPSM12 ½m1 ð0Þ þ m1 ðtÞt þ m2 ð0Þ 2 V1

ð1Þ

where m1(t) and m2(t) [dimension: M] are the amount of the chemical sequestered in Layer 1 and 2 at time t [T]; kPSM12 [LT1] is the mass transfer coefficient for chemical diffusion between the two layers of the PSM, and kPSM12 = DE, PSM/δ, where DE, PSM [L2T1] is the effective diffusivity of the chemical in the PSM and δ [L] is the diffusion length; A2 [L2] is the area

Figure 2. Comparison of the passive air sampling rates of PCB homologues between the passive sampling media of XAD and PUF positioned in the same type of cylindrical sampling housing.

between Layer 1 and 2; V1 [L3] is the Layer 1 volume of the PSM. Let Xt = [m1(0) + m1(t)] 3 t, Yt = m2(t), and apply LLSF to Xt and Yt, the slope of the fitted line equals to kPSM12 A2/(2V1), from which kPSM12 can be determined. Further, if δ is known, DE, PSM can be determined. Mechanistic Model of Effective Diffusivity in Porous Media. A previously developed modeling approach for the effective diffusivity of chemicals in porous media, such as soil and sediment, which considers sorption and tortuosity16,23 was applied to fit the effective diffusivity in PUF (DE, PUF) derived in this study: 1 f DA f DA ≈ 3 DE, PUF ¼ ΦA 3 f 3 DA ¼ 1 þ rSA 3 KPUF=A 3 3 rSA KPUF=A

ð2Þ where DE, PUF [L2T1] is the effective diffusivity in PUF, DA [L2T1] is the molecular diffusivity in bulk air, ΦA [dimensionless] is the fraction of the chemical in the air-filled PUF pore space, f [dimensionless] is a correction factor related to intra-aggregate porosity and tortuosity,23 rSA [L3(PUF)L3(A)] is the volume ratio between the solid PUF material and the porous air space in PUF, and KPUF/A [L3(A)L3(PUF)] is the chemical partition coefficient between PUF and air. The ratio f/rSA is a property of the porous medium that decreases with increasing density and tortuosity of the PUF.

’ RESULTS AND DISCUSSION Passive Air Sampling Rates. To compare the performance of XAD and PUF, we studied the PCB uptake kinetics on the two PSM placed in housings of the same design (Figure 1). The median of the R for the PCB congeners in each homologue group ranged 0.120.23 and 0.080.16 m3/d for an XAD and PUFbased PAS, respectively (Figure 2). R values derived for the individual PCB congeners are reported in Table S3. Because the configurations of the PSM used in this study were different from those used previously, it is not feasible to directly compare R with those reported in other studies. Therefore, R (m3/d) was normalized to the PSM surface area (dm2) and the normalized sampling rate (SR, m3/d/dm2) was used for comparison (Table S4). XAD-based SR ranged 0.110.32 m3/d/dm2, which is approximately 5- to 10-fold lower than SR from previous outdoor calibrations for XAD-PAS.1,9,12 This is in agreement with 10511

dx.doi.org/10.1021/es2032373 |Environ. Sci. Technol. 2011, 45, 10509–10515

Environmental Science & Technology

ARTICLE

Figure 3. PCB accumulation and distribution in the outer, middle, and inner layers of the passive sampling media (PUF and XAD). Plots are based on duplicated measurements. Mono-PCB (PCB-1) and Penta-PCB (PCB-98/95) are used to illustrate the diﬀerences between PCBs of diﬀerent chlorination or physicochemical properties.

previous studies on the PUF-PAS, which indicate that outdoor SR can be as much as ∼50-fold higher than indoor SR.10,11 The lower SR observed indoors by this and other studies can be attributed to the different extent of air movement indoors and outdoors. Relatively wind-still indoor conditions tend to increase the thickness of the air boundary layer surrounding the PSM and reduce R. The low air movement indoors could also increase the resistance to chemical transfer from ambient air into the PAS housing, which could possibly cause a “starvation” effect24 and make the air concentration within the housing lower than the ambient air. However, such an effect would exist and lower the passive sampling rate only if the resistance for a chemical to diffuse into the housing from ambient air is higher than that for chemical uptake by the PSM. Because it is difficult to measure the actual air concentration of SVOC within the PAS housing without disturbing its normal operational conditions, such a “starvation” effect on PAS for SVOC has so far not been confirmed experimentally. PUF-based SRs of this study ranged 0.020.07 m3/d/dm2, which is ∼5- and ∼30-fold lower than the calibrated indoor SR by Hazrati and Harrad10 and Shoeib and Harner.2 Apart from interstudy variations (∼5 times diﬀerence for the same type of PAS between refs 2 and 10), diﬀerent sampler conﬁgurations could provide a possible explanation for the lower PUF-based SR observed here. In this study, a PUF-cylinder was positioned in a “cylindrical can” rather than the more commonly used arrangement of a disk in a “double bowl” housing.2,8,10,11 This diﬀerent

conﬁguration could increase the thickness of stagnant air around the PUF, increase the kinetic resistance for a chemical to diﬀuse into the housing from ambient air, and thus lower the SR. Evidence of the eﬀect of sampler conﬁguration on passive sampling rates can also be found in studies by Tao et al.:25,26 PUF disks positioned in a more conﬁned housing had ∼10-fold lower SR than PUF disks in a “double bowl” shelter. Such evidence of the eﬀect of sampler conﬁguration on passive sampling rate indicates that the housing design may also contribute to the kinetic resistance to chemical uptake. The homologue-speciﬁc R decreases from the lighter to the heavier PCBs for both PSMs (Figure 2). This is in contrast with previous studies on the PUF-PAS, which found higher R for heavier congeners.8,10,11 A higher fraction of heavier congeners is particle-bound in air. The higher R for heavier congeners was attributed to particles being captured by the PUF-disk.8,10,11 Unlike the PUF-PAS, in which particle-bound chemicals were often detected,27,28 the XAD-PAS is unlikely to trap atmospheric particles since few particle-bound chemicals have ever been detected in XAD-ﬁlled mesh cylinders positioned in a cylindrical housing.5 The semienclosed conﬁguration of the “cylindrical can” shelter greatly limits advective air ﬂow into the housing and thus few particles may enter the housing and get trapped on the PSM. Excluding the eﬀect of particle-bound chemicals, chemical sequestration on the PSM is mainly determined by chemical transport from the ambient air to the PSM via diﬀusion in the gas 10512

dx.doi.org/10.1021/es2032373 |Environ. Sci. Technol. 2011, 45, 10509–10515

Environmental Science & Technology phase. This is supported by the positive correlations between the homologue-speciﬁc passive sampling rate and the gaseous molecular diﬀusivity of the chemicals (Figure S5). Evidence of Kinetic Resistance on Chemical Transfer within PSM. The kinetics of PSM-side mass transfer of the DCs and PCBs from air was investigated by analyzing the amount sequestered in each layer after different deployment times. PCB Uptake from Air. Higher PCB levels were found in the outer layer than in the middle and inner layer over the whole sampling period (Figure 3 and Tables S5 and S6). Within the first month of PAS deployment, the PCBs were either not detected in the middle or inner layers of PUF or detected at levels no different from the blanks. Mono-CBs could be detected in the middle and inner PUF layer after 4 and 8 weeks of deployment, respectively. Nevertheless, even after 12 weeks of deployment, the amount of mono-CBs in the middle and inner PUF layers was only ∼20% and ∼5% of that in the outer layer (Figure 3). Heavier PCBs could hardly be detected in the inner PUF layer, even after 12 weeks. For the mono- to tetra-PCBs detected in the middle PUF layer, the ratio of the amount in middle and outer layer was generally lower for the heavier congeners. No detectable amounts of penta-CBs and higher chlorinated PCBs could be found penetrating to the middle layer even after 12 weeks (Figure 3). Lighter PCBs appeared to diffuse more easily to the inner PUF layer: mono-CBs could diffuse through the 2-cm outer and middle PUF layer into the inner layer. This is because lighter PCBs have lower sorption affinity to PUF (i.e., a lower KPUF/A), allowing for a higher fraction to be in the gas phase of the PUF pores. The nonuniform PCB distribution within the PSM contradicts the assumption in the current passive air sampling theory7 describing chemical uptake in PAS.2,3 Compared to the PUF, less of the mono-CBs were found penetrating into the XAD (Figure 3). Even after 24 weeks, the amount sequestered in the middle layer was only ∼1% of that in the outer layer and no PCBs could be detected in the inner layer. This is in line with KXAD/A being higher than KPUF/A for individual PCB congeners (Tables S1 and S2), which make them less likely to be in the porous air phase and available for diﬀusion through the XAD-PSM. However, despite diﬀerent KXAD/A values, the amount of PCBs sequestered in the middle layer relative to that in the outer layer was very similar for diﬀerent PCB homologues; even for the heavier PCB homologues such as hepta-CBs the middle layer contained approximately ∼1% of the amount in the outer layer. We attribute this to the incomplete shielding of the middle XAD layer from ambient air. The XAD resin may have settled during the deployment period and left the upper part of the XAD in the inner mesh cylinders partially exposed to ambient air. Therefore, we can only infer that less than 1% of the PCBs in the outer XAD layer would penetrate to the middle layer by diﬀusion through the pores. This low diﬀusion rate also indicates that even if only 1% of the middle XAD layer was exposed to ambient air, the amount detected in the middle layer can not reﬂect the diﬀusion across the outer layer. Therefore, we do not further interpret the data for the layered XAD-PSM but focus on the layered PUF-PSM, of which the inside layer was completely covered by the outer one. Depuration Compounds. Transport of the spiked DCs between the PSM layers was observed. The data for the DCs are presented (Figures S6 and S7) and discussed in the SI. Mass Transfer Coefficient for Chemical Diffusion between the Two PUF Layers (kPUF12). kPUF12 was derived by fitting the amount of chemical accumulated in each PUF layer to the

ARTICLE

Figure 4. Relationship between the PUFair partition coeﬃcients (KPUF/A at 20 °C) and the mass transfer coeﬃcients for chemical diﬀusion between the two PUF layers (kPUF12, m/h). The data points represent selected mono-, di-, and tri-CB congeners that penetrated into the inner PUF with detectable amounts. The dashed lines indicate 95% conﬁdence interval of the regression model.

Figure 5. Relationship between the eﬀective diﬀusivity in PUF (DE,PUF, m2/h) and the PUF/air partition coeﬃcient (KPUF/A) for PCBs. The upper- and lower-bound experimentally derived DE,PUF values were based on a diﬀusion length of 1 and 2.5 cm, respectively. The upper- and lower-bound modeled DE,PUF values were based on a f /rSA value of 0.14 and 0.53, respectively.

two-layered mass balance model (eq 1). kPUF12 was calculated only if the coefficient of determination of the LLSF was over 0.7. The kPUF12 could only be derived for mono-, di-, and tri-CBs because heavier PCBs could not be detected in Layer 2. The derived kPUF12 ranged from 4.0 104 m/h for PCB-28 (tri-CB) to 1.1 102 m/h for PCB-1 (mono-CB) (Figure 4). A negative correlation (Spearman’s F = 0.91, p < 104) was found between kPUF12 and the PUF-air partition coefficients (KPUF/A). A simple regression model to predict kPUF12 from KPUF/A (Figure 4) shows that 81% of the variation in the experimentally derived kPUF12 can be accounted for by the variation in KPUF/A. The kPUF12 is related to the diffusion distance within the PUF and thus affected by the dimensions of the PUF. To exclude this factor, we derived the effective diffusivity (DE,PUF). Effective PSM-Side Diffusivities (DE,PUF). As the product of kPUF12 and diffusion length, DE,PUF excludes the effect of PUF 10513

dx.doi.org/10.1021/es2032373 |Environ. Sci. Technol. 2011, 45, 10509–10515

Environmental Science & Technology dimensions and should only depend on the properties of the PUF and chemical. Because kPUF12 was derived from chemical concentrations in two discrete PUF layers of finite thickness, we do not have information on the diffusion length within the PUF. Therefore, a range between 1 cm (thickness of Layer 1) and 2.5 cm (thickness of Layer 1 plus half the thickness of Layer 2) was used to represent the potential distance that chemicals diffusing from Layer 1 to Layer 2 are traversing. The magnitude of DE,PUF ranged from 109 m2/h for tri-CBs to 107 m2/h for mono-CBs (Figure 5). Although chemical diﬀusion in PUF occurs in the air-ﬁlled pore space, the eﬀective diﬀusivity in PUF is lower than the diﬀusivity in air by a factor of 105107. The low diﬀusivity in PUF is mainly attributed to the relatively large KPUF/A and thus a low fraction of the chemical in the porous air phase, where chemical diﬀusion within PUF occurs. Another factor lowering the chemical diﬀusivity in PUF is the tortuous diﬀusion pathway within the PUF, which increases the diﬀusion length and decreases DE,PUF. The inﬂuence of these factors on DE,PUF is also illustrated by the mechanistic model of chemical diﬀusion in porous media (eq 2). Fitting the DE,PUF (upper- and lower-bound value) derived in this study, we estimated f/rSA ranges between 0.18 (95% CI: 0.140.21) and 0.45 (95% CI: 0.350.53). Based on the model, DE,PUF values were calculated for all PCB congeners (Figure 5). DE,PUF decreases by over 5 orders of magnitude from mono- to decaCB. This variation in DE,PUF is mainly due to the variation in KPUF/A, because DA varies by less than 50% among diﬀerent PCB congeners (Figure S8). The upper- and lower-bound DE,PUF from the model diﬀer by ∼0.6 log-unit, which represents the range of f/rSA caused by potential variations of physical PUF properties. Using PUF with densities of 0.021 and 0.035 g/cm3, Chaemfa et al.29 found no signiﬁcant diﬀerence in sampling rates during 12 weeks of uptake. Based on our hypothesis, slightly higher uptake rates would be expected in low density PUF. This ﬁnding suggests that in the currently used PUF, f/rSA varies less than the diﬀerence between our upper- and lower-bound values. Interestingly, although overall uptake rates were not signiﬁcantly diﬀerent, Chaemfa et al. noted a faster uptake of some PCBs in the low-density PUF during early uptake.29 Further Comments on the PSM-Side Kinetic Resistance and Its Implications. Based on our experimental results and evidence from previous studies,8,9,29 we conclude that a kinetic resistance to chemical transfer exists within the PSM (PUF and XAD). The PSM in this study was a cylindrical PUF plug of 8-cm diameter. However, because DE,PUF of a chemical only depends on the properties of the PUF material but not on its shape, it should be possible to extrapolate the results of this study to the widely used 1-cm PUF disk. Because the experiment was conducted indoors and the PSM were positioned in a housing that eﬀectively shields the wind, advective transport of chemicals within the PSM likely did not occur. This agrees with Bohlin et al.,28 who observed only a minor inﬂuence of wind on PUFPAS deployed indoors. In PAS campaigns conducted outdoors, however, wind is likely to pass through the “double bowl”-type housing, resulting in increasing sampling rates with increasing wind speed.30 Such a wind eﬀect on the sampling rate can be caused by a decrease in the thickness of the air boundary layer and/or an increased eﬀective diﬀusivity within the PSM. According to CFD simulations on the PUF-PAS,31 the wind velocity approaches zero at the PUF surface. Therefore, if the wind does not blow directly toward the PUF, wind should have little, if any, eﬀect on DE,PUF. However, the CFD simulations rely on assumed

ARTICLE

scenarios of wind and other conditions. Based on the existing information on PAS under environmental conditions outdoors, we cannot exclude the possibility of advective chemical transport within the PSM. Further studies are needed to understand the potential advective transport within PSM and its eﬀect on the PSM-side kinetic resistance under various wind conditions. The nonuniform chemical distribution within the PSM aﬀects the calculation of the maximum linear uptake capacity of a PAS and the characteristic times of linear uptake or equilibration. Assuming a uniform chemical distribution within the PSM3 will lead to an overestimation of both the uptake capacity and the characteristic times because only the outer layer of the PSM is available for the sampled chemicals. Knowledge of the nonuniform chemical distribution can also help optimization of PAS design. Thinner PSM with a high surface area increase the sampling rate R without a signiﬁcant loss in uptake capacity. The nonuniform chemical distribution within PSM also challenges the current passive air sampling theory.3 Based on the twoﬁlm model,7 it assumes the sampled chemical is uniformly distributed within the PSM and a kinetic resistance to chemical uptake and loss only arises from the air boundary layer. This conceptual approach failed to explain chemical- and temperature-speciﬁc passive sampling rates,9,11 because the experimentally observed variation of R between chemicals and with temperatures was much larger than that can be explained by the compound-speciﬁcity and temperature dependence of DA.1,9 In this study, we found that the PSM-side kinetic resistance correlates with KPUF/A, which varies more among diﬀerent chemicals and at diﬀerent temperatures than DA. Qualitatively, this agrees with the experimental observations. It would be desirable to quantitatively compare the kinetic resistance (i.e., reciprocal of the mass transfer coeﬃcients) introduced by air boundary layer and PSM. However, we currently do not know the thickness of the boundary layer or the average diﬀusion length within the PSM, which are necessary to convert the diﬀusivities to mass transfer coeﬃcients. Because DE,PUF are more than 7 orders of magnitude lower than DA, the PSM side resistance will play a role in the overall uptake as long as the average diﬀusion length within the PSM exceeds 1/107 of the boundary layer thickness. A model that does not rely on the assumption of a uniform chemical distribution within the PSM will be required to quantitatively understand the PSM-side kinetic resistance and its inﬂuence on sampling rates. The current passive air sampling theory has also been used to describe the loss of DCs from the PSM and to derive samplerspeciﬁc sampling rates.3,30 This approach relies on the assumption that the uptake of the sampled chemicals and the loss of the DCs are subjected to the same kinetic resistances.30 This assumption would likely be true if the kinetic resistance of the air boundary layer were rate-limiting. However, because the kinetic resistance on the PSM side is not negligible, the kinetic resistance to uptake and loss would only be identical if the distribution of DCs and sampled chemicals within the PSM were the same. Such rigid conditions are impossible to meet because the distribution of the sampled chemicals within the PSM is unknown beforehand. Therefore, such uncertainty should be considered when interpreting PAS-based air concentrations calculated using R derived from the loss of DCs. Further eﬀorts are necessary to quantify and correct the uncertainty of R derived from the loss of DCs.

’ ASSOCIATED CONTENT

bS

Supporting Information. Further information on experimental methods, QA/QC, model derivation, KXAD/Air and KPUF/Air

10514

dx.doi.org/10.1021/es2032373 |Environ. Sci. Technol. 2011, 45, 10509–10515

Environmental Science & Technology for PCBs, congener-speciﬁc R, and further interpretation on the results. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 416-287-7225; e-mail: [email protected].

’ ACKNOWLEDGMENT We are grateful to James Armitage for sharing the idea for the design of the described experiments and to the Canadian Foundation for Climate and Atmospheric Sciences and the Natural Sciences and Engineering Research Council of Canada for funding. X.Z. acknowledges the Centre for Global Change Science at the University of Toronto for supporting the visit to HIES. ’ REFERENCES (1) Wania, F.; Shen, L.; Lei, Y. D.; Teixeira, C.; Muir, D. C. G. Development and calibration of a resin-based passive sampling system for monitoring persistent organic pollutants in the atmosphere. Environ. Sci. Technol. 2003, 37, 1352–1359. (2) Shoeib, M.; Harner, T. Characterization and comparison of three passive air samplers for persistent organic pollutants. Environ. Sci. Technol. 2002, 36, 4142–4151. (3) Bartkow, M. E.; Booij, K.; Kennedy, K. E.; M€uller, J. F.; Hawker, D. W. Passive air sampling theory for semivolatile organic compounds. Chemosphere 2005, 60, 170–176. (4) Zhang, X. M.; Diamond, M. L.; Robson, M.; Harrad, S. Sources, emissions, and fate of polybrominated diphenyl ethers and polychlorinated biphenyls indoors in Toronto, Canada. Environ. Sci. Technol. 2011, 45, 3268–3274. (5) Shunthirasingham, C.; Oyiliagu, C. E.; Cao, X. S.; Gouin, T.; Wania, F.; Lee, S. C.; Pozo, K.; Harner, T.; Muir, D. C. G. Spatial and temporal pattern of pesticides in the global atmosphere. J. Environ. Monit. 2010, 12, 1650–1657. (6) Bohlin, P.; Jones, K. C.; Levin, J. O.; Lindahl, R.; Strandberg, B. Field evaluation of a passive personal air sampler for screening of PAH exposure in workplaces. J. Environ. Monit. 2010, 12, 1437–1444. (7) Lewis, W. K.; Whitman, W. G. Principles of gas absorption. Ind. Eng. Chem. 1924, 16, 1215–1220. (8) Chaemfa, C.; Barber, J. L.; Gocht, T.; Harner, T.; Holoubek, I.; Klanova, J.; Jones, K. C. Field calibration of polyurethane foam (PUF) disk passive air samplers for PCBs and OC pesticides. Environ. Pollut. 2008, 156, 1290–1297. (9) Gouin, T.; Wania, F.; Ruepert, C.; Castillo, L. E. Field testing passive air samplers for current use pesticides in a tropical environment. Environ. Sci. Technol. 2008, 42, 6625–6630. (10) Hazrati, S.; Harrad, S. Calibration of polyurethane foam (PUF) disk passive air samplers for quantitative measurement of polychlorinated biphenyls (PCBs) and polybrominated diphenyl ethers (PBDEs): Factors inﬂuencing sampling rates. Chemosphere 2007, 67, 448–455. (11) Melymuk, L.; Robson, M.; Helm, P. A.; Diamond, M. L. Evaluation of passive air sampler calibrations: Selection of sampling rates and implications for the measurement of persistent organic pollutants in air. Atmos. Environ. 2011, 45, 1867–1875. (12) Hayward, S. J. Fate of current-use pesticides in the Canadian atmosphere. Ph.D.Thesis. University of Toronto, Toronto, 2010. (13) Hayward, S. J.; Lei, Y. D.; Wania, F. Sorption of a diverse set of organic chemical vapors onto XAD-2 resin: Measurement, prediction and implications for air sampling. Atmos. Environ. 2011, 45, 296–302. (14) Hayward, S. J.; Gouin, T.; Wania, F. Comparison of four active and passive sampling techniques for pesticides in air. Environ. Sci. Technol. 2010, 44, 3410–3416.

ARTICLE

(15) Klanova, J.; Eupr, P.; Kohoutek, J.; Harner, T. Assessing the inﬂuence of meteorological parameters on the performance of polyurethane foam-based passive air samplers. Environ. Sci. Technol. 2008, 42, 550–555. (16) Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry, 2nd ed.; Wiley: Hoboken, NJ, 2003. (17) Chaemfa, C.; Barber, J. L.; Moeckel, C.; Gocht, T.; Harner, T.; Holoubek, I.; Klanova, J.; Jones, K. C. Field calibration of polyurethane foam disk passive air samplers for PBDEs. J. Environ. Monit. 2009, 11, 1859–1865. (18) Kamprad, I.; Goss, K. U. Systematic investigation of the sorption properties of polyurethane foams for organic vapors. Anal. Chem. 2007, 79, 4222–4227. (19) van Noort, P. C. M.; Haftka, J. J. H.; Parsons, J. R. Updated Abraham solvation parameters for polychlorinated biphenyls. Environ. Sci. Technol. 2010, 44, 7037–7042. (20) Matsumura, C.; Tsurukawa, M.; Nakano, T.; Ezaki, T.; Ohashi, M. Eluation orders of all 209 PCBs congeners on capillary column HT8PCB. J. Environ. Chem. 2002, 12, 855–866. (21) Antweiler, R. C.; Taylor, H. E. Evaluation of statistical treatments of left-censored environmental data using coincident uncensored data sets: I. Summary statistics. Environ. Sci. Technol. 2008, 42, 3732–3738. (22) Aruga, R. Treatment of responses below the detection limit: some current techniques compared by factor analysis on environmental data. Anal. Chim. Acta 1997, 354, 255–262. (23) Wu, S. C.; Gschwend, P. M. Sorption kinetics of hydrophobic organic-compounds to natural sediments and soils. Environ. Sci. Technol. 1986, 20, 717–725. (24) Brown, R. H. The use of diﬀusive samplers for monitoring of ambient air. Pure Appl. Chem. 1993, 65, 1859–1874. (25) Tao, S.; Cao, J.; Wang, W. T.; Zhao, J. Y.; Wang, W.; Wang, Z. H.; Cao, H. Y.; Xing, B. S. A passive sampler with improved performance for collecting gaseous and particulate phase polycyclic aromatic hydrocarbons in air. Environ. Sci. Technol. 2009, 43, 4124–4129. (26) Tao, S.; Liu, Y. N.; Xu, W.; Lang, C.; Liu, S. Z.; Dou, H.; Liu, W. X. Calibration of a passive sampler for both gaseous and particulate phase polycyclic aromatic hydrocarbons. Environ. Sci. Technol. 2007, 41, 568–573. (27) Chaemfa, C.; Wild, E.; Davison, B.; Barber, J. L.; Jones, K. C. A study of aerosol entrapment and the inﬂuence of wind speed, chamber design and foam density on polyurethane foam passive air samplers used for persistent organic pollutants. J. Environ. Monit. 2009, 11, 1135–1139. (28) Bohlin, P.; Jones, K. C.; Strandberg, B. Field evaluation of polyurethane foam passive air samplers to assess airborne PAHs in occupational environments. Environ. Sci. Technol. 2010, 44, 749–754. (29) Chaemfa, C.; Barber, J. L.; Kim, K. S.; Harner, T.; Jones, K. C. Further studies on the uptake of persistent organic pollutants (POPs) by polyurethane foam disk passive air samplers. Atmos. Environ. 2009, 43, 3843–3849. (30) Moeckel, C.; Harner, T.; Nizzetto, L.; Strandberg, B.; Lindroth, A.; Jones, K. C. Use of depuration compounds in passive air samplers: Results from active sampling-supported ﬁeld deployment, potential uses, and recommendations. Environ. Sci. Technol. 2009, 43, 3227–3232. (31) Thomas, J.; Holsen, T. M.; Dhaniyala, S. Computational ﬂuid dynamic modeling of two passive samplers. Environ. Pollut. 2006, 144, 384–392.

10515

dx.doi.org/10.1021/es2032373 |Environ. Sci. Technol. 2011, 45, 10509–10515

ARTICLE pubs.acs.org/est

Mapping Flows of Embodied Emissions in the Global Production System Andrew Skelton,† Dabo Guan,*,‡,§ Glen P. Peters,|| and Douglas Crawford-Brown† †

)

Cambridge Centre for Climate Change Mitigation Research, Department of Land Economy, University of Cambridge, Cambridge CB3 9EP, U.K. ‡ School of Earth and Environment, University of Leeds, Leeds LS2 9JT, U.K. § St. Edmund’s College, University of Cambridge, Cambridge CB3 0BN, U.K. Center for International Climate and Environmental Research Oslo (CICERO), N-0318 Oslo, Norway

bS Supporting Information ABSTRACT: Environmentally extended multiregional input-output (MRIO) analysis can be used to investigate ﬁnal production and consumption attributions of emissions. As the distinction between the two attributions has been brought to the attention of policy-makers, there is an ever greater need to understand how and why they diﬀer, by analyzing the connections between production and consumption activities. Seeking to meet this need, we present an approach for mapping ﬂows of embodied emissions through a Leontief production system. The approach, seen as an extension of Structural Path Analysis (SPA), provides an exhaustive map of supply chain linkages between ﬁnal production and consumption attributions of emissions. Whereas SPA is traditionally used to extract and rank individual supply chains according to the emissions occurring at the end of each chain, the mapping approach considers emissions embodied in the ﬂows of intermediate products linking diﬀerent economic sectors along supply chains. Illustrative results are presented from a global MRIO model and CO2 emissions for 2004. The emissions embodied in a sector’s total output of products is also of interest: a method for calculating this is presented and shown to provide further insight into where in the production system a sector’s overall emissions impact is concentrated.

1. INTRODUCTION Assuming that all anthropogenic greenhouse gas (GHG) emissions can ultimately be attributed to consumption activity, environmentally extended multiregional input-output (MRIO) models have been used to investigate annual ﬁnal consumption attributions of emissions caused by national ﬁnal demand of products (i.e., goods and services),16 and, subsequently, to assess the emissions embodied in internationally traded products.711 In contrast, ﬁnal production attributions of emissions detail only the total direct (Scope 1) emissions released by producing entities. The results of the MRIO studies oﬀer a reattribution of emissions from producers to ﬁnal consumers (i.e., households and governments) and investors in capital (that enables further production in later years).12,13 Such an approach supports policies aimed at reducing or shifting consumer demand, at helping consumers understand the composite GHG implications of their choices, or at ensuring that costs of, and responsibilities for, climate change mitigation are allocated to entities and regions based on their roles in driving production processes through consumption. Final consumption attributions quantify the emissions virtually embodied in products purchased by ﬁnal demand the direct r 2011 American Chemical Society

emissions released during the assembly of ﬁnal products and all the indirect emissions released by the producers of intermediate products processed along associated supply chains. A ﬁnal consumption attribution may, for example, reveal that automobiles purchased by ﬁnal demand are high in embodied emissions (perhaps due to emissions occurring in supply chain steel and electricity production); while the automotive sector’s ﬁnal production attribution of emissions might be very low. Discrepancies between ﬁnal production and consumption attributions raise questions such as the following: where have the emissions embodied in ﬁnal products come from; conversely, where have the production emissions from economic sectors gone. Structural Path Analysis (SPA) has helped address these questions.1419 Within the context of a Leontief input-output (IO) model, purchases of intermediate products, instigated by ﬁnal demand purchases of ﬁnal products, can be traced through layers of the production system (e.g., consumers purchase from Received: July 19, 2011 Accepted: November 3, 2011 Revised: October 24, 2011 Published: November 03, 2011 10516

dx.doi.org/10.1021/es202313e | Environ. Sci. Technol. 2011, 45, 10516–10523

Environmental Science & Technology product manufacturers, who purchase from primary manufacturers, who in turn purchase from resource extractors). SPA extracts individual supply chains instigated by ﬁnal demand. When coupled with emissions data, an SPA quantiﬁes the emissions at the end of each supply chain (e.g., emissions occurring in metal production used as input into automobile production, or, for a longer supply chain: emissions occurring in electricity production used as input into metal production used as input into automobile production). The results of an SPA transform the embodied emissions, detailed in a ﬁnal consumption attribution, into a tree-like structure of emissions occurring in diﬀerent economic sectors at diﬀerent points in the production system. However, the interdependence between sectors in an IO framework (e.g., where one sector’s products are used by another sector to produce products used by the initial sector) means that a ﬁnal demand purchase theoretically propagates an inﬁnite number of supply chains through the economy. Still, there is a practical limit to the number of supply chains that can be extracted, both in terms of computational requirements and sensible interpretation.7 Studies using SPA have therefore found it impossible to exhaustively quote SPA results and provide a comprehensive assessment of the connections between ﬁnal production and consumption attributions, focusing instead on the identiﬁcation of emissions hotspots by ranking only the most important supply chains.15,19 The problem with this approach, however, is that while the majority of less important supply chains represent only a small fraction of total emissions, to the point of rendering them individually negligible, they can together represent a considerable share of total emissions.8 As diﬀerences in the apparent roles and responsibilities of actors when viewed through ﬁnal production and consumption attributions of emissions increasingly come to the attention of policy-makers,9 there is an ever greater need to understand how and why such diﬀerences occur. We suggest that part of this need can be met by investigating and illustrating the connections between ﬁnal production and consumption attributions in an exhaustive manner. To achieve this, three problems need to be addressed: (i) how to examine a large number of supply chains linking sectors at diﬀerent stages in the production system; (ii) how to show the relative importance of diﬀerent supply chains in regard to GHG emissions; and, (iii) how to account for the majority of supply chains that each constitute negligibly small emissions sources, but that collectively make up a considerable share of total emissions within an economy. In addressing these problems, the primary contribution of this paper is the introduction of a methodological and diagrammatic approach for mapping ﬂows of embodied emissions through a production system as described by a Leontief input-output model. This mapping approach builds on SPA to enable the exhaustive depiction of the connections between ﬁnal production and consumption attributions of emissions. To illustrate the approach, results are presented from an exploratory application to a global MRIO model and CO2 emissions for 2004. The mapping approach provides intermediate consumption attributions the emissions embodied in a sector’s output of intermediate products purchased at a particular layer within the production system. In addition, we present a method, based on the Pure Backward Linkage measure,20 for calculating a sector’s total consumption attribution the emissions embodied in a sector’s total output of products (i.e., all intermediate and ﬁnal products). The secondary contribution of this paper is to show how knowledge of a sector’s total consumption attribution

ARTICLE

provides further insight into where in the production system a sector’s overall emissions impact is concentrated and to show to what extent ﬁnal production and consumption attributions actually account for a sector’s overall impact.

2. METHODOLOGY MAPPING FLOWS OF EMBODIED EMISSIONS The fundamental equation of the Leontief model (consisting of N economic sectors) links an exogenous N 1 ﬁnal demand vector y with an N 1 total output vector x via x ¼ ðI AÞ1 y ¼ Ly

ð1Þ

where A is an N N matrix of sector intermediate purchases (or direct requirements matrix; an element Apq of A measures direct output from sector p necessary to satisfy unit output from sector q); I is the N N identity matrix; and L is the N N Leontief inverse (or total requirements matrix; an element Lpq of L measures total (i.e., direct and indirect) output from sector p necessary to satisfy unit output from sector q).18 The environmental extension of the model introduces a 1 N intensity vector f of sector direct emissions per unit output, such that total direct emissions (i.e., the ﬁnal production attribution) for sector s = fsxs. For a comprehensive introduction to IO is given by pﬁnal s theory and techniques see Miller and Blair.21 An advantage of the Leontief model is the ability to trace chains of intermediate purchases through layers of a production system instigated by ﬁnal demand. This is achieved by unravelling the Leontief inverse using its power series approximation, as shown in eq 221,23 L ¼ ðI AÞ1 ¼ I þ A þ A 2 þ A 3 þ A 4 þ 3 3 3 provided that limt f ∞ A t ¼ 0

ð2Þ

We deﬁne a production layer (PL) as each term in the power series expansion, PLt = At. Each additional layer, PLt+1 = PLtA, represents the production of intermediate products used as inputs into the preceding layer. For example, the assembly of automobiles (purchased by ﬁnal demand) occurs at PL0, the manufacture of metal products used as inputs into the automotive sector occurs at PL1, which requires inputs from metal production occurring at PL2. In eﬀect, the power series approximation irons-out interdependencies between sectors into linear intersector supply chains. This representation of a production layer is somewhat abstract as each sector is an aggregation of many industries, factories, and processes, and some sectors may be more aggregated than others. In many IO models, for example, agriculture is highly aggregated; thus, internal agricultural supply chains (e.g., seed production as input to vegetable growing) would be represented within a given layer, rather than across several layers were it more disaggregated. Our representation of the production system as a structure of linear intersector supply chains will become more realistic as the number of sectors increases (e.g., toward the resolution of LCA process data) and less realistic for highly aggregated IO models. In addition, an N N emissions multiplier matrix M can be calculated according to M ¼ ^f ðI AÞ1 ¼ ^f L

ð3Þ

where ^f is the diagonal of the intensity vector f.15,22 Similar to an element Lpq, an element Mpq measures direct and indirect emissions released from sector p that have been induced by the 10517

dx.doi.org/10.1021/es202313e |Environ. Sci. Technol. 2011, 45, 10516–10523

Environmental Science & Technology

ARTICLE

Table 1. Direct, Consumption, and Production Attribution Equations for PL0 to PL3 ﬁnal attribution (i.e., at PL ) 0

intermediate attribution at PL

direct

consumption

production

= fiyi

E0i E1j

= mi y i

P0i = Mi:y

= mjAj:y

P1j = Mj:Ay

= fjAj:y

to sector at PL0 from sector from sector at PL

D2k = fkAk:Ay

E2k = mkAk:Ay

P2k = Mk:A2y

at PL D3l = flAl:A2y

E3l = mlAl:A2y

P3l = Ml:A3y

at PL3

production of unit output from sector q (i.e., the emissions from sector p that have become embodied in the unit output from sector q). Manipulation of multiplier matrix M allows two diﬀerent measures of embodied emissions, Type I and Type II, to be calculated: Type I Embodied Emissions. A 1 N multiplier row vector, m, can be defined as the column sum of M, which can also be obtained directly from eq 4 m ¼ f ðI AÞ1 ¼ f L

to sector at PL2

E1f0 = mjAjiyi ji

-

-

E2f0 kji = mkAkjAjiyi

E2f1 = mkAkjAj:y kj

-

E3f0 lkji = mlAlkAkjAjiyi

E3f1 lkj = mlAlkAkjAj:y

E3f2 = mlAlkAk:Ay lk

ð4Þ

An element mq measures emissions from all sectors that have become embodied in unit output from sector q. This measure includes direct emissions (released during the production of sector q’s unit output) and indirect emissions (caused by intersector requirements and feedback eﬀects in the supply chain). Postmultiplication of mq by an elemental ﬁnal or intermediate demand for sector q’s output measures emissions from all sectors that have become embodied in that demand quantity. For the particular case of ﬁnal demand, we deﬁne this measure as sector q’s ﬁnal consumption attribution. For the case of intermediate demand at PLt, we deﬁne this measure as sector q’s intermediate consumption attribution at PLt. Type II Embodied Emissions. Any 1 N row vector, Mp:, extracted from the overall multiplier matrix M measures emissions from sector p that have become embodied in unit output from each sector. Postmultiplication of Mp: by an N 1 column vector of final or intermediate demand for output from each sector will then measure emissions from sector p that have become embodied in that basket of demand quantities. For the particular case of final demand, we define this measure as sector p’s final production attribution (i.e., Mp:y = Pfinal p = fpxp, since sector p’s total direct emissions must have been released by the end of the production system). For the case of intermediate demand at PLt, we define this measure as sector p’s intermediate production attribution at PLt (i.e., the portion of sector p’s final production attribution of emissions that has been released by PLt). By combining eqs 2-4 we arrive at three sets of equations, which for an example of 4 production layers (PL0 to PL3) are given in Table 1. Where i, j, k, and l denote sectors at PL0, PL1, PL2, and PL3, respectively, and ‘:’ denotes all sectors (e.g., Aj: is the 1 N row vector of all intermediate purchases from sector j). An element Dts represents direct emissions released from sector s at PLt (e.g., D1j measures direct emissions released from sector j at PL1 in producing output required to meet intermediate demand given by Aj:y). An element Ets represents emissions that have become embodied in the output of sector s at PLt (e.g., E2k measures emissions from all sectors that have become embodied in the output of sector k at PL2 required to meet the intermediate demand given by Ak:Ay). An element Pts represents emissions

2

from sector

2

intermediate attribution

to sector at PL1

at PL1

1

intermediate attribution at PL

D0i D1j

Table 2. Embodied Emissions Flow Equations down to PL3

3

from sector s that have become embodied in the output of all sectors at PLt (e.g., P3l measures emissions from sector l that have become embodied in the output of all sectors at PL3 required to meet the basket of intermediate demand given by A3y). Further sets of equations can be speciﬁed that measure the emissions from all sectors that have become embodied in the output of a given sector required to meet demand from a chain of ﬁnal and intermediate purchases (i.e., the ﬂows of embodied emissions between sectors at diﬀerent layers) as presented in Table 2. 3f2 , E2f1 measure the emissions that have become E1f0 ji kj , and Elk embodied in the ﬂow of intermediate products between sectors at measures emissions from all adjacent layers. For example, E1f0 ji sectors embodied in the output of sector j at PL1 purchased by sector measures emissions from all sectors i at PL0; similarly, E2f1 kj embodied in the output of sector k at PL2 purchased by sector j at PL1 to meet ﬁnal demand requirements from all sectors at PL0. The remaining equations provide more speciﬁc measuremeasures emissions from all sectors ments. For example, E2f0 kji embodied in the output of sector k at PL2 purchased by sector j at PL1 to meet ﬁnal demand requirements from sector i at PL0. The set of embodied emissions ﬂow equations terminating at PL0 have a similar form to those used in SPA;24 however, the SPA equations are used to calculate direct emissions at the end of a supply chain rather than embodied emissions. For example the = SPA equations for PL1, PL2, and PL3 can be denoted D1f0 ji 3f0 fjAjiyi, D2f0 kji = fkAkjAjiyi, and Dlkji = flAlkAkjAjiyi, respectively. Equation results from Tables 1 and 2, proposed here as an extension of SPA, can be used to map ﬂows of embodied emissions through the production layer expansion of the Leontief production system. Figure 1 illustrates a map for a simple system consisting of just two sectors, s1 and s2. The left-hand side of Figure 1 provides each sector’s ﬁnal production attribution of emissions, Pﬁnal s , while the right-hand side provides each sector’s ﬁnal consumption attribution of emissions, E0i . Between these two ﬁnal attributions, the diagram elaborates: each sector’s intermediate consumption attribution at PL1, PL2, and PL3 , E1j , E2k, and E3l respectively; the release of direct emissions from each sector at PL0 to PL3 D0i , D1j , D2k, and D3l , respectively; and the ﬂows of embodied emissions from sectors at PL1 to PL0, at 3f2 , E2f1 PL2 to PL1, and at PL3 to PL2 E1f0 ji kj , and Elk , respectively. Contributions from higher order layers have been combined to provide a comprehensive depiction of the system. As one would anticipate, the sum of ﬁnal production attributions is equal to the ﬁnal 0 = ∑N sum of ﬁnal consumption attributions, i.e., ∑N s=1Ps i=1Ei . Conceptually we can envisage ﬁnal demand purchases of ﬁnal products triggering cascades of intermediate product purchases that ﬂow upstream through the layers of the production system. Within each sector, at each layer, the processing of products entails the release of emissions into the atmosphere. Similarly, we can envisage cumulative ﬂows of embodied emissions running in the opposite direction, from the depths of the production system to the point of consumption of ﬁnal products (the right-hand side of Figure 1), 10518

dx.doi.org/10.1021/es202313e |Environ. Sci. Technol. 2011, 45, 10516–10523

Environmental Science & Technology

ARTICLE

total output of products from sector s (i.e., component (a) from , can then be above). The total consumption attribution, Etotal s calculated by adding to this the ﬁnal production attribution for sector s, Pﬁnal s Etotal ¼ EPBLs þ Pfinal s s

ð6Þ

Presenting a sector’s ﬁnal and intermediate production and consumption attributions as a percentage of its total consumption attribution allows us to identify where in the production system the sector’s overall emissions impact is concentrated. Furthermore, this allows us to show to what extent the standard ﬁnal consumption and production attributions account for a sector’s overall impact. A graphical means of investigating these relationships is presented in the Application section below. Figure 1. Illustrative map of embodied emissions ﬂows for a simple two sector economy. Diagram ﬂows from left to right and is of a Sankey-type, such that the widths of indicated ﬂows represent their magnitude.25,26 For example, the emissions embodied in the output of ﬁnal products from sector s1 at PL0 to meet ﬁnal demand is given by ﬂow (a), E0s1, the ﬁnal consumption attribution for sector s1. The emissions embodied in the output of intermediate products from sectors s1 and s2 at PL1 purchased 1f0 by sector s1 at PL0 are given by ﬂows (b), E1f0 s1,s1 , and (c), Es2,s1 , respectively. The production processing of these inputs also results in the release of direct emissions from sector s1 at PL0, given by ﬂow (d), D0s1.

picking up additional emissions along the way as intermediate products are processed. While sources of direct emissions are not conceptually understood as ﬂows, they have been represented as such in Figure 1 so as to clearly illustrate how the ﬁnal production attribution can be mapped through to the ﬁnal consumption attribution.

3. METHODOLOGY CALCULATING SECTOR TOTAL CONSUMPTION ATTRIBUTIONS The mapping approach provides a sector’s ﬁnal consumption attribution (i.e., at PL0) and intermediate consumption attributions at earlier layers. To calculate the total consumption attribution of a sector s (i.e., the emissions from all sectors that have become embodied in the total output, xs, of sector s), we need to somehow sum ﬁnal and intermediate consumption attributions across all layers, taking care to avoid double-counting, resulting in a measure that consists of two components: (a) the emissions from all sectors, excluding sector s, that have become embodied in the total output of sector s and (b) the ﬁnal production attribution of emissions for sector s. The IO literature on key sectors and economic linkages provides an algebraic method, introduced by Sonis et al.,20 for calculating economic Pure Backward Linkage (PBL), which gives the upstream impact on the economy from the total output of a given sector that is free from intraindustry demands and feedbacks from the rest of the economy. The PBL calculation can be extended to give the Emissions Pure Backward Linkage (EPBL) of a sector s EPBLs ¼ f ðI A Þ1 A:s xs

ð5Þ

where xs is the total output of sector s; A:s is the N 1 column vector of all intermediate purchases made by s; A* is the N N matrix of intermediate purchases, with purchases by and from s set to zero; and, f* is the 1 N intensity vector of sector direct emissions per unit output, with the emissions intensity for sector s set to zero. The EPBL measure gives the emissions released from all sectors, excluding sector s, induced by the production of the

4. APPLICATION TO GLOBAL SECTOR CO2 EMISSIONS Here we apply the methodological developments introduced above to the case of global economic sector CO2 emissions by using an MRIO model constructed from 2004 global economic data, disaggregated across 113 regions and 57 sectors. Balanced economic data and sector CO2 emissions data were taken from Version 7 of the Global Trade Analysis Project (GTAP)27 and converted into an MRIO table through the proportional allocation of bilateral trade data across interindustry requirements and ﬁnal consumption of imported products.2,13,28 For a discussion of MRIO model uncertainty, see Wiedmann and Lenzen et al.12,29 This particular study has been designed to illustrate the application and interpretation of the methodological developments presented in this paper. 4.1. Mapping Embodied Emissions through the Global Production System. Figure 2 presents a map of embodied CO2

emissions flows for global sectors, constructed in the following manner: • The mapping approach introduced in Section 2 was applied to the MRIO model and sector CO2 emission intensities described above. This provided ﬁnal and intermediate consumption attributions and direct emissions results for each of the 57 sectors in each of the 113 regions at PL0, PL1, PL2, and PL3, and embodied emissions ﬂow results from regional sectors at PL1 to PL0 and at PL2 to PL1. • Results were then aggregated across the 113 regions, reducing the complexity of the system, to give global sector results. • To further simplify the system, results for the 57 global sectors were aggregated to 31 sectors according to the concordance table provided in the Supporting Information (Table SI1). Aggregating to global sectors at a postcalculation stage ensured that regional variation in sector emissions intensities and full implications of international trade (including feedback eﬀects) were accounted for within the study. We can contrast, for example, the components of the construction sector’s ﬁnal consumption attribution with those of electricity production and distribution. For construction, we ﬁnd direct emissions at PL0 account for only 7.7% (0.28 Gt CO2), while the inputs purchased from PL1 have very high embodied emissions associated with them: for example, nonmetallic mineral products accounts for 37.0% (1.35 Gt CO2), metal production and casting 16.3% (0.59 Gt CO2), and fabricated metal products 7.7% (0.28 Gt CO2). For the electricity sector the story is quite diﬀerent: direct emissions at PL0 account for 88.0% (2.73 Gt CO2), with a further 5.6% (0.17 Gt CO2) from intraindustry 10519

dx.doi.org/10.1021/es202313e |Environ. Sci. Technol. 2011, 45, 10516–10523

Environmental Science & Technology

ARTICLE

Figure 2. Map of embodied CO2 emissions ﬂows for global sectors. Diagram ﬂows from left to right and is of a Sankey-type, such that the widths of indicated ﬂows represent their magnitude in Gt CO2. The left-hand side of the map shows the ﬁnal production attributions of CO2 emissions, totalling 22.8 Gt CO2 (although outside the scope of this study, household CO2 emissions are estimated in the model to be 4.5 Gt CO2, giving total global CO2 emissions of 27.3 Gt CO2 for 2004). Contributions from the top 10 sectors are identiﬁed individually, while contributions from the remaining 21 sectors are combined into a single measure. Global electricity production and distribution dominates the ﬁnal production attribution, accounting for 45% (10.1 Gt CO2). Other major sectors include the following: nonmetallic mineral products at 9% (2.0 Gt CO2); road, rail, and other transport at 8% (1.9 Gt CO2); and metal production and casting at 7% (1.6 Gt CO2). The right-hand side of the map presents the ﬁnal consumption attributions, again totalling 22.8 Gt CO2, for each sector. Again contributions from the top 10 sectors are identiﬁed: with the construction sector accounting for 16% (3.7 Gt CO2); electricity production and distribution 14% (3.1 Gt CO2); government services 12% (2.7 Gt CO2); retail and trade 6% (1.4 Gt CO2); and food and beverage products 6% (1.4 Gt CO2). The central part of the diagram reveals the intermediate consumption attributions for each sector at PL1 and PL2. Again the top 10 sectors at each layer are identiﬁed. Direct emissions released by each sector at PL0, PL1, and PL2 are indicated by dark gray ‘ﬂows’ linking back to the ﬁnal production attribution. The top 20 embodied emissions ﬂows between the top 10 sectors at PL1 and PL0 and between those at PL2 and PL1 have been extracted. All remaining ﬂows have been merged and dropped to the background. Finally, contributions from PL3 and all earlier layers have been combined to provide a comprehensive view of the system.

inputs, and the remaining 6.4% (0.20 Gt CO2) from inputs from other sectors at PL1. We should note here that while construction products are heavily purchased by ﬁnal demand, this is not to say that it is household consumers that are making these purchases, but rather, in this case, it is primarily ﬁnal demand from investment in capital. Final construction products can then be seen as enabling further production in later years. 4.2. Normalized Evolution of Production and Consumption Attributions. Using the same MRIO model as the above analysis, here we apply eqs 5 and 6 to calculate each sector’s total consumption attribution as the sum of two components: (a) the emissions from all other sectors that have become embodied in the sector’s total output and (b) the sector’s final production attribution. These results are presented in Figure S1 in the SI and used to normalize results in Figure 3. Figure 3 shows how closely measures of final production and consumption attributions and intermediate

production and consumption attributions at PL1, PL2, and PL3 account for a sector’s total consumption attribution. First, inspecting the data points at the end of each line in Figure 3, we ﬁnd that for the sectors motor vehicles, construction, recreation and other services, food and beverage products, and government services, over 75% of their total consumption attribution can be accounted for by their ﬁnal consumption attribution. For electricity production and distribution and nonmetallic mineral products, the same can be said for their ﬁnal production attribution. However, for all other sectors, neither ﬁnal consumption nor production attributions account for more than 75% of their total consumption attribution. The most extreme case shown is that of fabricated metal products, where its ﬁnal consumption attribution accounts for 19%, and its ﬁnal production attribution only 8%, of its total consumption attribution. This suggests that the analysis of a sector’s emissions impact in terms of only ﬁnal production and consumption attributions runs the risk of underestimating the scale 10520

dx.doi.org/10.1021/es202313e |Environ. Sci. Technol. 2011, 45, 10516–10523

Environmental Science & Technology

ARTICLE

Figure 3. Evolution of sector consumption and production attributions of CO2 emissions from PL3 to PL0 normalized as a percentage of sector total consumption attribution. Solid markers at the end of each line give a sector’s ﬁnal consumption attribution (read oﬀ the y-axis) and ﬁnal production attribution (read oﬀ the x-axis) as a percentage of total consumption attribution. Data points marked with a star indicate intermediate consumption and production attributions at PL1, PL2, and PL3, with PL3 at the start of the line, again as a percentage of total consumption attribution. For example, for the metal production and casting sector: point (a) gives the ﬁnal production attribution (1.56 Gt CO2) as 47% and ﬁnal consumption attribution (0.09 Gt CO2) as 3% of the sector’s total consumption attribution (3.30 Gt CO2); point (b) gives intermediate production attribution at PL1 (1.53 Gt CO2) as 46% and intermediate consumption attribution at PL1 (1.56 Gt CO2) as 47% of total consumption attribution; point (c) gives intermediate production attribution at PL2 (1.02 Gt CO2) as 31% and intermediate consumption attribution at PL2 (1.31 Gt CO2) as 40% of total consumption attribution; and point (d) gives intermediate production attribution at PL3 (0.60 Gt CO2) as 18% and intermediate consumption attribution at PL3 (0.80 Gt CO2) as 24% of total consumption attribution. Solid lines connecting data points are shown to aid the visual inspection of discrete measurements and do not represent a continuous series.

of the sector’s overall impact. This is particularly the case for sectors that produce intermediary products from emissions intensive inputs, such as fabricated metal products and chemical and plastics products. Second, we can inspect the evolution of production and consumption attributions for a particular sector to help us understand where in the production system its overall emissions impact is concentrated and whether this takes the form of direct emissions or inducing upstream emissions from other sectors. Figure 3 shows, for example, that a large share of total direct emissions from the nonmetallic mineral products sector are released at PL1, as indicated by the relatively large diﬀerence between data points at PL2 and PL1 read oﬀ the x-axis these direct emissions alone account for 44% of the total consumption attribution. Furthermore, over two-thirds of the normalized intermediate consumption attribution at PL1 is due to direct emissions released at PL1. We already know (from Figure 2) that the majority of these emissions are in fact passed on to the construction sector at PL0. The sudden drop from intermediate consumption at PL1 to ﬁnal consumption at PL0, coupled with a small increase in the production attribution, can be explained by the fact that few products from the nonmetallic mineral products sector are purchased by ﬁnal demand. We can therefore surmise that the overall emissions impact of the nonmetallic mineral products sector is relatively concentrated as direct emissions released from the sector at PL1. We can contrast this with the metal production and casting sector, which looks to have a similar proﬁle but closer inspection shows otherwise. Again there is a sudden drop in the consumption attribution from PL1 to PL0, coupled with negligible increase in the production attribution, indicating that activity at PL0 does

not signiﬁcantly raise the overall emissions impact. Similar levels of direct emissions are released from PL1 and PL2, and these attributions each account for a third of the normalized consumption attribution at the respective layers. In addition, none of the ﬁnal or intermediate attributions for this sector account for more than 50% of total consumption attribution. We can therefore say that the overall emissions impact of the metal production and casting sector is diﬀused across the production system, with the exception of PL0, and that both direct emissions and upstream emissions from other sectors are important. In general, three broad groups of sectors can be identiﬁed and are indicated in Figure 3 by diﬀerent solid markers at the end of each line. First, the ‘primary producer’ sectors, such as nonmetallic mineral products and metal production and casting, which are characterized by a sudden drop in their consumption attribution and minimal increase in their production attribution at PL0, as they tend to supply few products to ﬁnal demand. Second, the ‘comprehensive producer’ sectors, such as road, rail, and other transport and electricity production and distribution, which are characterized by steadily increasing production and consumption attributions across all layers. These sectors tend to supply products to both ﬁnal demand and the intermediate demand of a wide range of other sectors. And ﬁnally, the ‘consumer facing’ sectors, such as construction and food and beverage products, which are characterized by a low ﬁnal production attribution and a sudden jump in their consumption attribution at PL0. These sectors tend to only supply products to ﬁnal demand that require the processing of typically emissions intensive inputs. 10521

dx.doi.org/10.1021/es202313e |Environ. Sci. Technol. 2011, 45, 10516–10523

Environmental Science & Technology

5. DISCUSSION The methodological developments introduced in this paper are an extension to SPA that reveal how discrepancies between production and consumption attributions of emissions arise. The central concept is that any ﬁnal or intermediate demand for products, within a Leontief production system, can be evaluated in terms of two diﬀerent emissions quantities: (a) direct emissions released from the production of the purchased products and (b) emissions from all sectors that have become embodied in the purchased products. The mapping approach can then be thought of as a toolkit for evaluating ﬂows of emissions embodied in the transactions of intermediate products between economic sectors and ultimately the supply of products to ﬁnal demand. Large input-output models, in particular MRIO models, represent an enormous number of transactions between economic sectors. While it is possible, for a given production layer, to calculate the emissions embodied in each individual transaction (indeed this was the ﬁrst step taken in constructing Figure 2), the visual presentation of all these results is impractical. Furthermore, the production layer expansion form of a Leontief model provides an inﬁnite series of production layers. Hence judgment, on behalf of the analyst, in terms of which ﬂows to illustrate individually and which to aggregate or omit, is required in constructing a map that links ﬁnal production and consumption attributions of emissions. Such judgment would need to reﬂect the issue under investigation. An analyst may, for example, be interested in understanding the linkages between ﬁnal production and consumption in three world regions, China, the US, and the rest of the world: the resulting map may then focus on the ﬂows of embodied emissions exchanged between these world regions without need to elucidate the role played by individual economic sectors. Plots showing the normalized evolution of production and consumption attributions, such as Figure 3, could be used as an eﬀective tool for comparing the emissions impact of sectors across diﬀerent countries or, using time-series IO data, to show how a sector’s emissions impact proﬁle has changed over a certain time period. It is anticipated that future application of the mapping approach and normalized evolution plots could provide supportive evidence for developing regional policies that target ﬁnal and intermediate consumption needs that currently trigger large cascades of emissions though global supply chains. Finally, the approach could be used by ﬁrms to understand the importance of estimating emissions from their supply chains (i.e., upstream Scope 3 emissions) and benchmark against average sector performance, while future developments (e.g., adaption for use with hybrid LCA models and extension to evaluate downstream emissions impacts) could help interpretation of sector and corporate carbon-footprints: an aid that would not only beneﬁt corporate social responsibility initiatives but also assist core business strategy as ﬁrms seek to ameliorate risks from current and future carbon price and regulation. ’ ASSOCIATED CONTENT

bS

Supporting Information. Application concordance table (Table S1) and chart showing sector total consumption attributions (Figure S1). This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

ARTICLE

’ REFERENCES (1) Davis, S. J.; Caldeira, K. Consumption-based accounting of CO2 emissions. Proc. Natl. Acad. Sci. U.S.A. 2010, 107, 5687–5692. (2) Hertwich, E. G.; Peters, G. P. Carbon Footprint of Nations: A Global, Trade-Linked Analysis. Environ. Sci. Technol. 2009, 43, 6414–6420. (3) Wiedmann, T.; Wood, R.; Minx, J. C.; Lenzen, M.; Guan, D.; Harris, R. A carbon footprint time series of the UK Results from a multi-regional input-output model. Econ. Syst. Res. 2010, 22, 19–42. (4) Hubacek, K.; Guan, D.; Barrett, J.; Wiedmann, T. Environmental implications of urbanization and lifestyle change in China: Ecological and Water Footprints. J. Cleaner Prod. 2009, 17, 1241–1248. (5) Guan, D.; Hubacek, K.; Weber, C. L.; Peters, G. P.; Reiner, D. M. The drivers of Chinese CO2 emissions from 1980 to 2030. Global Environ. Change 2008, 18, 626–634. (6) Peters, G. P.; Weber, C. L.; Guan, D.; Hubacek, K. China’s Growing CO2 Emissions - A Race between Increasing Consumption and Eﬃciency Gains. Environ. Sci. Technol. 2007, 41, 5939–5944. (7) Peters, G. P.; Hertwich, E. G. CO2 Embodied in International Trade with Implications for Global Climate Policy. Environ. Sci. Technol. 2008, 42, 1401–1407. (8) Wiedmann, T.; Lenzen, M.; Turner, K.; Barrett, J. Examining the global environmental impact of regional consumption activitiesPart 2: Review of input-output models for the assessment of environmental impacts embodied in trade. Ecol. Econ. 2007, 61, 15–26. (9) Peters, G. P.; Minx, J. C.; Weber, C. L.; Edenhofer, O. Growth in emission transfers via international trade from 1990 to 2008. Proc. Natl. Acad. Sci. U.S.A. 2011, 108, 8903–8908. (10) Guan, D.; Peters, G. P.; Weber, C. L.; Hubacek, K. Journey to world top emitter: An analysis of the driving forces of China’s recent CO2 emissions surge. Geophys. Res. Lett. 2009, 36. (11) Weber, C. L.; Peters, G. P.; Guan, D.; Hubacek, K. The contribution of Chinese exports to climate change. Energy Policy 2008, 36, 3572–3577. (12) Wiedmann, T. A review of recent multi-region input-output models used for consumption-based emission and resource accounting. Ecol. Econ. 2009, 69, 211–222. (13) Peters, G. P. From production-based to consumption-based national emission inventories. Ecol. Econ. 2008, 65, 13–23. (14) Lenzen, M. A guide for compiling inventories in hybrid lifecycle assessments: some Australian results. J. Cleaner Prod. 2002, 10, 545–572. (15) Peters, G. P.; Hertwich, E. G. Structural analysis of international trade: Environmental impacts of Norway. Econ. Syst. Res. 2006, 18, 155–181. (16) Minx, J. C.; Wiedmann, T.; Wood, R.; Peters, G. P.; Lenzen, M.; Owen, A.; Scott, K.; Barrett, J.; Hubacek, K.; Baiocchi, G.; Paul, A.; Dawkins, E.; Briggs, J.; Guan, D.; Suh, S.; Ackerman, F. Input-output analysis and carbon footprinting: an overview of applications. Econ. Syst. Res. 2009, 21, 187. (17) Wood, R.; Lenzen, M. Structural path decomposition. Energy Econ. 2009, 31, 335–341. (18) Lenzen, M. Environmentally important paths, linkages and key sectors in the Australian economy. Struct. Change Econ. Dynam. 2003, 14, 1–34. (19) Acquaye, A. A.; Wiedmann, T.; Feng, K.; Crawford, R. H.; Barrett, J.; Kuylenstierna, J.; Duﬀy, A. P.; Koh, S. C. L.; McQueenMason, S. Identiﬁcation of “Carbon Hot-Spots” and Quantiﬁcation of GHG Intensities in the Biodiesel Supply Chain Using Hybrid LCA and Structural Path Analysis. Environ. Sci. Technol. 2011, 45, 2471–2478. (20) Sonis, M.; Guilhoto, J. J. M.; Hewings, G. J. D.; Martins, E. B. Linkages, key sectors, and structural change: Some new perspectives. Dev. Econ. 2007, 33, 243–246. (21) Miller, R. E.; Blair, P. D. Input-Output Analysis - Foundations and Extensions, 2nd ed.; Cambridge University Press: Cambridge, U.K., 2009. (22) Lenzen, M.; Crawford, R. The Path Exchange Method for Hybrid LCA. Environ. Sci. Technol. 2009, 43, 8251–8256. 10522

dx.doi.org/10.1021/es202313e |Environ. Sci. Technol. 2011, 45, 10516–10523

Environmental Science & Technology

ARTICLE

(23) Suh, S.; Heijungs, R. Power series expansion and structural analysis for life cycle assessment. Int. J. Life Cycle Assess. 2007, 12, 381–390. (24) Treloar, G. J. Extracting embodied energy paths from inputoutput tables: Towards an input-output-based hybrid energy analysis method. Econ. Syst. Res. 1997, 9, 375. (25) Schmidt, M. The Sankey Diagram in Energy and Material Flow Management. Part I: History. J. Ind. Ecol. 2008, 12, 82–94. (26) Schmidt, M. The Sankey Diagram in Energy and Material Flow Management. Part II: Methodology and Current Applications. J. Ind. Ecol. 2008, 12, 173–185. (27) Narayanan, B.; Walmsley, T. Global Trade, Assistance, and Production: The GTAP 7 Data Base, Centre for Global Trade Analysis, Purdue University, Washington, DC, 2008. (28) Peters, G. P.; Andrew, R.; Lennox, J. Constructing an Environmentally-Extended Multi-Regional Input-Output Table using the GTAP database. Econ. Syst. Res. 2011, 23, 131–152. (29) Lenzen, M.; Wood, R.; Wiedmann, T. Uncertainty analysis for multi-region inputoutput models A case study of the UK’s carbon footprint. Econ. Syst. Res. 2010, 22, 43–63.

10523

dx.doi.org/10.1021/es202313e |Environ. Sci. Technol. 2011, 45, 10516–10523

ARTICLE pubs.acs.org/est

Speciation and Degradation of Triphenyltin in Typical Paddy Fields and Its Uptake into Rice Plants Fabiane G. Antes,† Eva Krupp,*,‡,§ Erico M. M. Flores,† Valderi L. Dressler,*,† and Joerg Feldmann‡ †

Departamento de Química, Universidade Federal de Santa Maria, 97105-900 Santa Maria, RS, Brazil Department of Chemistry, University of Aberdeen, Meston Building, Meston Walk, AB24 3TU, Aberdeen, United Kingdom § ACES Aberdeen Centre for Environmental Sustainability, University of Aberdeen, AB23 3UU, Aberdeen, United Kingdom ‡

bS Supporting Information ABSTRACT: Triphenyltin (TPhT) is a biocide used worldwide in agriculture, especially in rice crop farming. The distribution and dissipation of TPhT in rice ﬁelds, as well as uptake of TPhT and other phenyltin compounds (monophenyltin, MPhT, and diphenyltin, DPhT) is still unknown at present. In this study, speciation analysis of phenyltin compounds was carried out in soil and water from a rice ﬁeld where TPhT was applied during rice seeding according to legal application rates in Brazil. The results indicate the degradation of biocide and distribution of tin species into soil and water. To evaluate whether TPhT is taken up by plants, rice plants were exposed to three diﬀerent TPhT application rates in a controlled mesocosm during 7 weeks. After this period, tin speciation was determined in soil, roots, leaves, and grains of rice. Degradation of TPhT was observed in soil, where DPhT and MPhT were detected. MPhT, DPhT, and TPhT were also detected in the roots of plants exposed to all TPhT application rates. Only TPhT was detected in leaves and at relatively low concentration, suggesting selective transport of TPhT in the xylem, in contrast to DPhT and MPhT. Concentration of phenyltin species in rice grains was lower than the limit of detection, suggesting that rice plants do not have the capability to take up TPhT from soil and transport it to the grains.

’ INTRODUCTION Triphenyltin (TPhT) is widely used in agriculture as a biocide.1 In Brazil, this compound is regularly used in rice (Oryza sativa L.) farming. TPhT can be applied directly to the soil or leaves during plant growth, playing the function of molluscicide and fungicide, respectively.2 Brazilian legislation recommends that the maximum residue of TPhT in commercial rice should be below 0.1 mg 3 kg 1.3 However, this limit is not routinely controlled. TPhT is also registered with the U.S. Environmental Protection Agency (EPA), and its use is allowed as a fungicide on pecans, potatoes, and sugar beet crops.4 Among organotin compounds, the trisubstituted compounds are the most toxic, while the nature of the anion group has little or no eﬀect on the biocide activity, except that this anion itself is a toxic component (general formula R3SnX, where R is an alkyl or aryl group and X is an anion such as hydroxide, acetate, etc.1 The toxicity and cardiovascular activity of organotin compounds has been recently reviewed by Nath.5 Experimental studies have shown that TPhT may have endocrine-disrupting capabilities resulting in adverse eﬀects on the reproductive system of mollusks and mammals. 2,6,7 These endocrine-disrupting eﬀects include the development of imposex in gastropods and changes in reproductive organs as well as alterations in sexual hormone levels in rats.8,9 Other studies also showed highly toxic eﬀects of r 2011 American Chemical Society

TPhT on ﬁsh, such as suppression of number of eggs and egg quality, causing a decrease in the fecundity of females, and induced teratogenesis such as eye defects, morphological malformation, and conjoined twins.10 12 Due to food exposed to TPhT, especially ﬁsh and ﬁshery products, humans are indirectly subject to this compound and its toxicological eﬀects. According to Hoch,1 TPhT has a relatively low stability in the environment, and ultraviolet irradiation or biological or chemical cleavage could be responsible for progressive loss of aryl groups from the Sn-organo cation. Consequently, diphenyltin (DPhT), monophenyltin (MPhT), and inorganic tin (Sn) can be formed. Dubascoux et al.13 studied the kinetics of degradation of TPhT in soil spiked with contaminated sewage sludge. According to this study, TPhT degradation occurred very quickly at the beginning of the experiment. Over 85% of TPhT was degraded after 53 days (total experiment time) and the calculated TPhT half-life was 6 ( 1 days. However, DPhT and MPhT showed higher persistence in soil. The degradation of TPhT (as triphenyltin acetate) in selected soil types was studied by Yen et al.14 They found that temperature was the most important factor that contributed to Received: August 13, 2011 Accepted: November 11, 2011 Revised: November 5, 2011 Published: November 11, 2011 10524

dx.doi.org/10.1021/es202832g | Environ. Sci. Technol. 2011, 45, 10524–10530

Environmental Science & Technology the TPhT degradation rate and that soil moisture content and microbial activity did not signiﬁcantly aﬀect the degradation of this compound. The authors observed high degradation rates in soil, with half-lives between 8.3 and 19.4 days, depending on temperature and kind of soil. Little is currently known about the uptake of phenyltin compounds by plants. A study about TPhT uptake from soil by lettuce showed that most of the TPhT that was absorbed remained in roots and only a low amount was transported to the shoots (less than 2%). Additionally, relatively high degradation of TPhT in soil was observed, where only about 20 30% of the initially introduced TPhT was detected after 54 days.15 In some crops, instead of applying the TPhT to the soil, the biocide is applied over the plant leaves. In an experiment performed to evaluate TPhT and its degradation products’ assimilation in leaves and soil from sprayed pecan orchards, it was concluded that TPhT was rapidly degraded to DPhT and mainly to MPhT, probably due to photocatalytic reactions.16 In spite of the use of TPhT in rice crops in many countries, no reports were found in the literature about TPhT or its degradation products’ uptake by rice plants. Here we determined TPhT and its degradation products DPhT and MPhT in soil and water of a paddy ﬁeld that received the routine TPhT application employed in southern Brazil. Additionally, in order to evaluate the extent of TPhT uptake and transportation to diﬀerent parts of rice (roots, grains, and leaves), a study was conducted where rice plants were exposed to diﬀerent concentrations of TPhT added to the irrigation water, in a controlled mesocosm experiment. Speciation analysis of MPhT, DPhT, and TPhT was performed in soil, roots, leaves, and grains via gas chromatography coupled to inductively coupled plasma mass spectrometry (GC-ICP-MS).

’ MATERIALS AND METHODS Sampling of Soil and Water in a Rice Field. Soil samples were collected in a rice farm from Brazil where TPhT hydroxide was applied to the field during rice seeding. Samples (soil volumes of 20 20 20 cm) were collected at five points in a paddy field of 2.5 km2, three days after the application of TPhT hydroxide. Soil samples were dried by lyophilization, ground with mortar and pestle, sieved to obtain particle size lower than 200 μm, and stored at 20 °C. The superficial water layer was also collected in the same rice field. This sample was maintained at 20 °C until speciation analysis. Controlled Mesocosm Experiments. After germination on vermiculite, 30-day-old rice (Oryza sativa L.) plants were transplanted to 1 L plastic pots packed with dry clay-rich soil. The plant pots were placed in a greenhouse with temperature and light simulating tropical conditions (temperature between 22 and 35 °C). Daylight was supplemented with sodium lamps, which were on during 8 h per day. Plants were grown under irrigated conditions by permanent immersion of the pots, in individual trays where water and nutrient solution were added three times a week. Nutrients were supplied through addition of Yoshida solution. A description of this procedure and also the composition of the Yoshida solution is given in Supporting Information. Experiments were performed with 12 plants that were divided into four groups (i.e., three plants/dose): one was the control group and three groups were exposed to three diﬀerent concentrations of TPhT. The treatment consisted of addition of TPhT in 50 mL of Yoshida solution. The concentration of TPhT in

ARTICLE

these solutions was 20, 100, and 1000 μg 3 kg 1 (as Sn), identiﬁed as low, medium, and high TPhT concentration, respectively. The medium concentration was chosen to simulate the amount usually applied in rice ﬁelds in Brazil.3 The addition of TPhT solutions was performed during 7 weeks and ﬁnished 7 days before rice was ready for harvesting. Total mass of TPhT (as Sn) added to the soil during the whole experiment is described in Table S2 (Supporting Information). Harvesting Rice Plants and Sample Preparation for Speciation Analysis. Rice plants were taken out of the pot and the whole plant plus soil mass was weighed. Plants were then carefully separated from soil, and roots were washed with water to remove the soil completely. Roots were separated from shoots by cutting out a piece of 4 cm between roots and shoots, to avoid intercontamination of both parts. Panicles were removed from the leaves, and rice spikelets were removed from panicles manually. The mass of soil, roots, leaves, and spikelets was recorded. Rice grains were separated from their husks by use of a mortar and pestle, and the mass of all grains of each plant was recorded. Rice parts and soil were stored in zip-lock bags at 80 °C before sample preparation for speciation analysis. Evaluation of Sample Preparation Procedures for Speciation Analysis. For the purpose of quality control, the extraction procedure was evaluated using spiked soil and plant material samples due to the lack of certified reference materials for phenyltin compounds. In the case of plant material, the extraction procedure was evaluated by use of rice grains, which were ground in a mortar and pestle under liquid N2, and about 1 g of material was transferred to a glass vial. Soil samples were homogenized with a spatula and also weighed into glass vials. For tin species extraction, two procedures, adapted from Lespes et al.15 and Monperrus et al.,17 were used for soil and plant material. These procedures were evaluated by use of plant material and soil samples spiked with MPhT, DPhT and TPhT. The procedure adapted from Lespes et al.15 showed the best analyte recovery and therefore it was used for MPhT, DPhT, and TPhT speciation analysis. A brief description of sample preparation procedures evaluated is given below. Procedure A: One gram of material (soil or plant) was transferred to a 20 mL glass vial, and 3 mL of methanol + ethyl acetate (1:1) solution was added. The vial was maintained under mechanical shaking for 1 h. Then 5 mL of 0.035 mol 3 L 1 HCl solution prepared in methanol + ethyl acetate (1:1) was added and the vial was shaken for 1 h and then centrifuged at 5000 rpm for 5 min. Procedure B: This was similar to procedure A, except 5 mL of 0.1 mol 3 L 1 HCl plus 1 g of NaCl was used instead of 0.035 mol 3 L 1 HCl. This procedure was evaluated only for soil. Procedure C: One gram of dry material (soil or plant) was transferred to a 20 mL glass vial, and 2 mL of methanol and 5 mL of acetic acid were added. The vial was agitated for 12 h and centrifuged at 5000 rpm for 5 min.Blank samples were prepared in the same way but without the addition of phenyltin compounds. After extractions, 4 mL of supernatant was transferred to a clean glass vial and 50 μL of tripropyltin (TPrT) solution (50 μg 3 L 1 as Sn) was added as internal standard (IS). The pH of this solution was adjusted to 4.9 by use of an acetic acid/sodium acetate buﬀer. This mixture was then submitted to ethylation by the addition of 1 mL of isooctane and 1 mL of 2% (m/v) sodium tetraethylborate solution. The vial was immediately capped and vigorously shaken for 5 min. The solution was centrifuged at 3000 rpm for 10525

dx.doi.org/10.1021/es202832g |Environ. Sci. Technol. 2011, 45, 10524–10530

Environmental Science & Technology 5 min to help phase separation and the organic layer was transferred to a 2 mL amber autosampler vial. Samples from the mesocosm experiments were analyzed in duplicate, and samples collected in the rice ﬁeld were analyzed in triplicate. For analysis of water collected in the rice ﬁeld, 10 mL of sample was derivatized following the same procedure described above. The derivatization was performed in triplicate and extracts were kept at 80 °C until analysis by GC-ICP-MS. Quantiﬁcation was performed via standard addition calibration with IS, and at least two measurements were done per sample. Spiked soil and plant material samples were also prepared by the addition of MPhT, DPhT, and TPhT standard solutions in order to submerge 5 g of soil or plant material (rice grains), which were collected in a place where TPhT had never been applied. After equilibration during 48 h in the dark, the solvent was eliminated with a gentle stream of argon. The concentrations of MPhT, DPhT, and TPhT were 33.4, 30.1, and 71.1 ng 3 g 1 in spiked soil and 65.0, 55.2, and 50.5 ng 3 g 1 in spiked plant material, respectively. Spiked samples were stored at 20 °C before extraction and analysis. Blank samples were prepared in the same way but without the addition of phenyltin compounds. The commercial product Mertin 400, a suspension containing 400 g 3 L 1 TPhT hydroxide used in rice ﬁelds, was also analyzed by GC-ICP-MS in order to evaluate the purity of this product. Therefore, the biocide was weighed and diluted in water and derivatized by the same procedure as applied for the water samples. Soil samples collected in the rice ﬁeld were digested for total tin determination. Digestion was performed with an Ethos II microwave oven (Milestone, Bergamo, Italy) where 0.25 g of dry soil sample was transferred to poly(tetraﬂuoroethylene) (PTFE) vessels and 5, 2, and 2 mL of HNO3, HF, and HCl, respectively, were added. Digestion program consisted of two steps: a ramp of 10 min up to 200 °C, followed by 30 min at this temperature. Total tin was determined by ICP-MS at m/z 120, by use of Perkin-Elmer-SCIEX model Elan DRC II equipment (Thornhill, Canada). Tin Speciation Analysis. An Agilent 6890 gas chromatograph (Palo Alto, CA) was coupled to an Agilent ICP-MS 7500c via a homemade heated transfer line. Dry plasma conditions were used because in this case better species transport was obtained. The GC-ICP-MS operating conditions are described in Table S3 (Supporting Information). Chromatographic signals obtained were processed on the basis of peak area by the WinFAAS 1.0 software. Standards and Reagents. MPhT, DPhT, TPhT, and TPrT stock standard solutions were prepared by dissolving appropriate amounts of the respective compounds (PhSnCl3, Ph2SnCl2, and Ph3SnCl, Sigma Aldrich, Dorset, England; Pr3SnCl, Strem Chemicals, Newburyport, MA) in methanol (Mallinckrodt, Phillipsburg, NJ). Intermediate standard solutions were prepared weekly by dilution of stock solutions in methanol, and work standard solutions were prepared daily in 0.1 mol 3 L 1 HCl. All solutions were maintained in the dark at 20 °C. Methanol, ethyl acetate, NaCl, and HCl (high purity, Merck, Darmstadt, Germany) were used for preparing extraction solutions. A 2% (m/v) NaBEt4 (Sigma Aldrich) solution, prepared in water, was used for ethylation. Adjustment of pH was performed with a 1.0 mol 3 L 1 acetic acid/sodium acetate buffer (pH 4.90). After ethylation, tin species extraction was performed with isooctane (Mallinckrodt). Yoshida’s nutrient solution was prepared from respective analytical-grade salts (see Supporting Information).

ARTICLE

Figure 1. Chromatograms obtained by GC-ICP-MS. (a) Standards (20 μg 3 L 1) of MPhT, DPhT, and TPhT (as Sn); (b) soil sample exposed to high TPhT concentration; (c) roots sample exposed to high TPhT concentration. Peak 1, inorganic Sn; peak 2, TPrT; peak 3, MPhT; peak 4, DPhT; peak 5, TPhT; peak 6, Me3Sn+ (suggested); peak 7, Me2Sn2+ (suggested); peak 8, PhMe2Sn+ (suggested); peak 9, PhMeSn2+ (suggested); peak 10, Ph2MeSn+ (suggested).

’ RESULTS AND DISCUSSION Tin Speciation Analysis and Its Improvement. Due to the relatively low stability of TPhT,18,19 the use of microwave- or ultrasound-assisted extraction procedures is usually not recommended.20 However, considering that extraction by microwave radiation is usually faster than other extraction methods, a procedure using microwave radiation (5 min, 60 °C) was also evaluated (results not presented); however this procedure yielded low recoveries for TPhT and recoveries higher than 100% for DPhT and MPhT. These results indicate species degradation and interconversion, and therefore microwave extraction was not used for further investigations. Therefore, in this study only extraction procedures using mechanical shaking were evaluated. Recoveries obtained for MPhT, DPhT, and TPhT in plant material (using procedures A and C) and in soil (using procedures A, B, and C) are shown in Table S4 (Supporting Information). Recoveries obtained by procedures A and C were lower than 90% and 75%, respectively, for all species in soil samples. Low recoveries were also obtained for procedure C for phenyltin extraction from plant material. On the other hand, recoveries higher than 94% were obtained for plant material and soil by extraction procedures 10526

dx.doi.org/10.1021/es202832g |Environ. Sci. Technol. 2011, 45, 10524–10530

Environmental Science & Technology A and B, respectively. Therefore, procedure A was used for phenyltin extraction from plant material, and procedure B was used for soil samples. Extraction procedures A and B were subsequently applied for phenyltin compound speciation analysis in enriched soil and plant material samples. The recoveries obtained in soil were 79%, 97%, and 95% for MPhT, DPhT, and TPhT, respectively, while the recoveries in plant material were 86%, 95%, and 96% for MPhT, DPhT, and TPhT, respectively. Although relatively lower recovery has been observed for MPhT in enriched soil and plant material samples, the obtained results could be considered suitable when the diﬃculties usually described for phenyltin compound speciation analysis are taken into account.20 Chromatograms obtained for enriched soil and plant material are shown in Figure S1 (Supporting Information). The limits of detection (LOD), calculated following the signal at intercept and 3 times the standard deviation about regression of the calibration curve, for MPhT, DPhT, and TPhT were 0.5, 0.66, and 0.72 μg 3 L 1, respectively (as Sn). The LOD values for 1 g of sample used for extraction were 1.0, 1.3, and 1.4 ng 3 g 1 for MPhT, DPhT, and TPhT, respectively (as Sn). Chromatograms obtained by GC-ICP-MS for standard, soil, and rice (roots) samples are shown in Figure 1a, b, and c, respectively. It is possible to observe a similar peak shape for samples and standard solution and good peak resolution for all species. The peak that corresponds to TPhT is relatively broad due to the high boiling point of ethyl-TPhT (close to 300 °C).21 The transfer line used in this experiment is homemade and has a heating limit of 220 °C. Therefore, the transport of TPhT through the transfer line to the ICP torch is not optimal and causes peak broadening due to insuﬃcient heating. Nonetheless, the triphenyltin compound can be quantiﬁed, even though detection limits are compromised. Other not-identiﬁed peaks can be observed in Figure 1b and c with retention times that do not correspond with retention times of the standards. These peaks were also not observed in chromatograms obtained for enriched soil and plant material (Figure S1, Supporting Information). These peaks are discussed in the next section. Phenyltin Compounds in Soil and Water from a Rice Field. The commercial biocide was analyzed and TPhT was established to be the main organotin compound. Only traces of DPhT had been found, while MPhT was below the detection limit and no other unknown tin species were recorded. Results obtained for phenyltin species for diluted sample are shown in Table S5 (Supporting Information). In general, TPhT is added directly into the irrigation water when the rice is seeded to kill mollusks that could destroy the seeds. However, due to dissipation of TPhT and its degradation in the environment, which leads to formation of the still-toxic MPhT and DPhT compounds, the consequences and damages to other aquatic living organisms could have a serious environmental impact.11 Additionally, the potential of TPhT to contaminate groundwater has to be considered.14 However, very few studies reporting the eﬀects of TPhT use in rice ﬁelds are available, and this compound is still used in indiscriminate ways in many places around the world.22 To evaluate the contamination with phenyltins in a rice ﬁeld, soil and water samples were analyzed by GC-ICP-MS. It is important to mention that the application of TPhT hydroxide in this ﬁeld was done 3 days before sampling. However, this ﬁeld has been used for more then 10 years for rice crops, and TPhT is applied at least once a year. The results obtained are presented in Table 1.

ARTICLE

According to the results shown in Table 1, it can be observed that the distribution of phenyltin compounds is diﬀerent in soil samples from the same ﬁeld collected at diﬀerent points. Phenyltin species were detected in concentrations above LOD in soil samples 3 and 5 (LOD 1.1, 1.3, and 1.4 ng 3 g 1 for MPhT, DPhT, and TPhT, respectively, as Sn). This behavior can be explained by TPhT hydroxide not being homogenously applied in the ﬁeld. Additionally, in this case, rice cultivation is performed in irrigated ﬁelds and the water can be responsible for transportation and dissipation of the biocide. In the water sample that was analyzed, only TPhT was detected, at relatively low concentrations. It is also important to consider the presence of MPhT and DPhT in the soil samples. The analysis of a commercial TPhT hydroxide (Mertin 400) product indicated that the concentration of DPhT was lower than 0.5% of the concentration of TPhT, while MPhT was not detected. This suggests that MPhT and DPhT detected in soil samples were from degradation of TPhT after its application and not from contamination of the commercial biocide used. Triphenyltin Uptake by Rice Plants. The application of TPhT in rice crops in Brazil is usually performed with a commercial suspension of triphenyltin hydroxide (Mertin 400). According to the guidelines on the product, 1 L of a 400 g 3 L 1 suspension is applied to 4 ha of rice field. If it is taken into account that rice is cultivated in irrigated fields, with a water layer of approximately 10 cm, the concentration of triphenyltin hydroxide is expected to be approximately 100 μg 3 L 1. Therefore, the experiment conducted in this work was performed under similar conditions. No phenyltin species were detected in roots, leaves, and grains in control rice samples. In the same way, these species were not detected in control samples of soil. However, MPhT, DPhT, and TPhT species were detected in soil that was exposed to low, medium, and high concentrations of TPhT. Table 2 shows the concentration of phenyltin species in soil and roots from plants exposed to diﬀerent TPhT levels. As can be observed from the results shown in Table 2, TPhT in soil is degraded to DPhT and MPhT species. Apparently, the degradation of TPhT to DPhT and MPhT depends on the amount of TPhT added to soil. The concentration of TPhT in soil exposed to high TPhT concentration is signiﬁcantly lower than those of MPhT and DPhT, indicating that the degradation of TPhT was higher compared to the other two levels. For medium and low concentrations of TPhT added to soils, the concentration is similar for all phenyltin species. The reason for this behavior was not investigated in this study, but it may be related to the diﬀerent stabilities of each species in soil or possible stabilizing eﬀects of substances present in soil that can form complexes preferentially with one of the phenyltin species. The presence of MPhT, DPhT, and TPhT was also detected in the roots. These results are shown in Table 2 for three diﬀerent TPhT levels of concentration. In general, the distribution of phenyltin species in roots for the diﬀerent initial concentrations of TPhT is similar to that observed in soil samples. It is possible to observe an increasing of all phenyltin species concentration in roots with increased TPhT added. Regarding phenyltin species uptake from soil, we postulate some possibilities that could be considered to explain this behavior. The rice plant may have taken up preferentially TPhT, and DPhT and MPhT species could be formed in roots, as a consequence of the plant metabolism. On the other hand, plants could take up the three species from the soil. Another possibility is that both eﬀects could occur simultaneously, that is, the plant could take up the three 10527

dx.doi.org/10.1021/es202832g |Environ. Sci. Technol. 2011, 45, 10524–10530

Environmental Science & Technology

ARTICLE

phenyltin species from soil, and species interconversion and degradation could occur, even in the roots. According to TPhT concentration in roots and soil shown in Table 2, the calculated roots/soil uptake ratio is 6.25, 3.44, and 13.3 for low, medium, and high concentration exposed plants, respectively. In this way, it could be suggested that plant uptake ratio would be higher for plants exposed to higher TPhT levels. In contrast to what was observed for roots, for leaves of plants exposed to low and medium TPhT content, the concentration of MPhT, DPhT, and TPhT was lower than the LOD. Only TPhT was detected (1.33 ( 0.49 ng 3 g 1, as Sn) in leaves of plants treated with high TPhT concentration. DPhT and MPhT concentration was lower than their respective LODs. A chromatogram obtained for leaves exposed to high TPhT level is shown in Figure 2, where a small peak corresponding to TPhT can be observed but no MPhT or DPhT. It is also possible to see an unknown peak with retention time of 200 s, with a relatively high intensity. This peak also appears in chromatograms b and c in Figure 1 and could be related to other tin species, as will be explained afterward. Only a low concentration of TPhT was detected in the leaves, although the concentration of DPhT and MPhT was higher in the roots than TPhT. This would point to preferential TPhT translocation from the roots through xylem sap into the leaves. No phenyltin species were found in the rice grain. According to these results, it seems that rice plants do not easily take up phenyltins and transport them from soil to leaves and grains. Additionally, although relatively high degradation of TPhT has been observed in soil and all phenyltin species were detected in roots and at similar levels, DPhT and MPhT were not transported to the shoots. Probably phenyltin molecules are not easily translocated to the shoots of rice plants, similar to the eﬀect observed for lettuce by Lespes et al.15 These results indicate that rice consumption does not imply risks to health regarding contamination with TPhT and other tin species, although a

thorough ﬁeld study needs to conﬁrm the mesocosm experiments. Although the mesocosm experiment has been performed trying to simulate ﬁeld condition, some factors such as microorganisms, environmental eﬀects like rain, etc., could not be reproduced. Therefore, additional experiments in the ﬁeld would be necessary to conﬁrm the obtained results. As the total amount of TPhT that each plant was exposed to is known (Table S2, Supporting Information) and the mass of each part of plant and soil at the end of the experiment was also known, it was possible to calculate how much was taken up by the plants, by use of the absolute values of phenyltin species found in each plant as listed in Table 3. In this sense, for plants exposed to low TPhT level, the content of TPhT found in the roots after the experiment was only 0.13% ( 0.05% of the amount of TPhT to which the plants were exposed (8 μg). Similarly, for plants exposed to medium and high TPhT levels (72 and 373 μg, respectively), the content of TPhT determined in the roots was 0.09% ( 0.05% and 0.19% ( 0.04% of the amount of TPhT to which plants were exposed. The percentage of TPhT that remained in soil for plants exposed to low, medium, and high levels of TPhT was 1.94% ( 0.07%, 2.41% ( 0.68%, and 1.40% ( 0.38%, respectively. The low recoveries could be attributed to degradation of TPhT to DPhT and MPhT and even to inorganic tin. Although inorganic tin was not quantiﬁed, a peak corresponding to this species with a retention time of about 160 s was observed in all samples. Additionally, losses by volatilization through bioalkylation by formation of volatile tetraalkyltin species may have occurred during the experiment. The formation of tetraalkyltin through Sn methylation is a well-established process that occurs naturally in the environment.23,24 Krupp et al.23 estimated 2 4 μg of Sn 3 (m 3 of landﬁll gas) in two landﬁll sites in

Table 1. Results for Mono-, Di-, and Triphenyltin in Soil and Water Samples from a Rice Fielda sample

MPhT

DPhT

TPhT

(ng 3 g 1, as Sn)

(ng 3 g 1, as Sn)

(ng 3 g 1, as Sn)

soil 1

<1.12

<1.32

<1.44

soil 2

<1.12

<1.32

<1.44

soil 3

3.03 ( 0.21

<1.32

soil 4 soil 5

10.5 ( 0.9

<1.12 1.18 ( 0.13

<1.32 1.72 ( 0.11

<1.44 4.24 ( 0.73

waterb

<1.4b

<1.3b

1.8 ( 0.1b

a Results represent the mean ( standard deviation (n = 3). b Concentrations for water are nanograms of Sn per milliliter.

Figure 2. Tin species detected by GC-ICP-MS. The chromatogram shows detection of TPhT in the extract of rice shoots exposed to high levels of TPhT (373 μg) in the irrigation water. Peak 1, inorganic Sn; peak 2, TPrT; peak 5, TPhT; peak 8, PhMe2Sn+ (suggested).

Table 2. Concentrations of Phenyltin Species in Soil and Roots Samples from Three Diﬀerent Initial TPhT Concentrations of the Mesocosm Experimentsa low (8 μg of TPhT) species 1

MPhT (ng 3 g , as Sn)

a

DPhT (ng 3 g 1, as Sn) TPhT (ng 3 g 1, as Sn)

soil

roots

medium (72 μg of TPhT) soil

roots

high (373 μg of TPhT) soil

roots

0.82 ( 0.26

0.65 ( 0.27

16.2 ( 8.0

5.43 ( 2.70

132 ( 54

87.9 ( 5.1

0.79 ( 0.31 1.04 ( 0.51

2.49 ( 2.08 1.44 ( 1.03

10.6 ( 7.6 11.5 ( 2.1

13.3 ( 7.1 9.25 ( 4.66

139 ( 43 35.4 ( 11.1

88.1 ( 27.4 81.9 ( 40.5

Uncertainties represent the mean and standard deviation (n = 3). Control soil and roots did not contain any phenyltin species. 10528

dx.doi.org/10.1021/es202832g |Environ. Sci. Technol. 2011, 45, 10524–10530

Environmental Science & Technology

ARTICLE

Scotland, coming from tetramethyltin and other ethylated, propylated, and butylated tetraalkyltin compounds, which make up between 35% and 53% of all volatile Sn compounds in the biogas emitted. A close analysis of chromatograms b and c presented in Figure 1 indicates the presence of four unknown peaks, with diﬀerent retention times than those of phenyltin standards. These peaks can be related to methylated tin species that could be formed from degradation products of TPhT or from inorganic tin, by chemical or biological processes.1 To try to identify the possible identity of unknown peaks, a correlation plotting the sum of carbons of Sn ligands from known species (inorganic Sn, TPrT, MPhT, DPhT, and TPhT) against their respective retention time (RT) was performed as suggested Table 3. Total Amount of Phenyltin Species in Each Plant Part and Soil, For Three Diﬀerent Initial TPhT Concentrationsa plant part and soil

MPhT, μg

DPhT, μg

TPhT, μg

Low Initial [TPhT] (8 μg) soil

0.15 ( 0.04

0.11 ( 0.06 (1.35 ( 0.68%) 0.02 ( 0.02

0.16 ( 0.09

roots

(2.68 ( 0.65%) 0.003 ( 0.02

(2.35 ( 1.38%) 0.01 ( 0.01

(0.06 ( 0.02%)

(0.25 ( 0.13%)

(0.13 ( 0.05%)

leaves

<0.009

<0.011

<0.012

grains

<0.001

<0.001

<0.001

Medium Initial [TPhT] (72 μg) soil

2.52 ( 0.054 (5.12 ( 2.77%)

1.46 ( 0.66 (2.46 ( 1.55%)

1.74 ( 0.49 (2.22 ( 0.89%)

roots

0.04 ( 0.03

0.08 ( 0.05

0.06 ( 0.05

leaves

(0.05 ( 0.03%) <0.009

(0.12 ( 0.07%) <0.011

(0.09 ( 0.05%) <0.012

grains

<0.001

<0.001

<0.001

soil

19.4 ( 6.1

21.4 ( 9.1

(5.74 ( 1.89%)

(4.33 ( 0.22%)

High Initial [TPhT] (373 μg)

roots

0.11 ( 0.08

5.20 ( 1.27 (1.57 ( 0.20%)

0.79 ( 0.15

(0.03 ( 0.02%)

(0.21 ( 0.04%)

leaves

<0.009

<0.011

grains

<0.001

<0.001

0.69 ( 0.04 (0.19 ( 0.01%) 0.03 ( 0.02 <0.001

a

Uncertainties represent the mean and standard deviation (n = 3). The value given in parentheses is given as percentage from application rate (as Sn) determined after the controlled mesocosm experiment.

by Krupp et al.23 A detailed description of this procedure is given in Supporting Information. From the regression equation obtained, it was possible to estimate the number of carbons (NC) of unknown species and then suggest the corresponding tin species, always assuming that the original species contained phenyl groups. A good correlation between number of carbons and retention time was obtained, given by the equation RT = 24.1NC 38.01 (R2 = 0.99). The calculated NC and the respective suggested tin species are shown in Table 4, while the correlation graph is shown in Figure S2 (Supporting Information). We assume here that TPhT is partly dephenylated and subsequently or simultaneously methylated, and thus Me replaces a Ph group on the Sn atom. These degradation products would be diphenylmethyltin (Ph2MeSn+) and the methylated MPhT compounds phenylmethyltin (PhMeSn2+) and phenyldimethyltin (PhMe2Sn+), which were detected as their ethylated derivatives. The synthesis of those standards would be necessary to conﬁrm the identity of these species, using the respective standards or a technique that employs molecular mass spectrometry, such as GC coupled with electron impact mass spectrometry (GC-EIMS), which, however, would require much higher concentrations and is not feasible with the small amounts of unknown compounds formed in this experimental setup. A full identiﬁcation of the unknown tin species is, however, beyond the scope of this paper, and dedicated experiments would be needed in environments promoting high methylation rates. Thus, the peak identiﬁcation remains suggestive. Nevertheless, recent work by Krupp et al.23 has conﬁrmed that the assumed correlation between retention time and carbon substitutes is valid, and the earlier observed natural methylation in anaerobic environments1 renders the assumed peak assignments highly probable. Standards and enriched samples (as described above; Figure S1, Supporting Information) did not show any of these peaks; thus we can rule out artiﬁcial methylation occurring during the sample preparation and derivatization steps. Considering that ICP-MS gives the same response for all Sn species, the concentration of the unknown species (as Sn) was calculated, using the standard addition calibration with IS obtained for MPhT, for soil or roots exposed to high concentrations of TPhT. The calculated concentrations are also presented in Table 4. It is important to note that this concentration is very low in comparison with concentrations of MPhT, DPhT, and TPhT determined in plants and soil exposed to high TPhT concentration. The data do not show directly whether MPhT or DPhT were methylated in the soil or in planta; however, only methylated MPhT have been identiﬁed in the roots in signiﬁcant concentration. In conclusion, dimethyland trimethyltin species have tentatively been identiﬁed in

Table 4. Methylated Tin Containing Species in Soil and Roots from the High-Level Mesocosm Experiment (High Application Rate) Corresponding to Unknown Sn Peaks in the Chromatogramsa conc, ng 3 g retention time, s

extrapolated no. of carbonsb

suggested methylated Sn species

1

soil

roots

50

3.7 (5)

Me3Sn as Me3SnEt

<1.0

1.2 ( 0.41

100

5.7 (6)

Me2Sn2+ as Me2SnEt2

1.02 ( 0.25

<1.0

205

10.1 (10)

PhMe2Sn+ as PhSnMe2Et

1.04 ( 0.26

<1.0

239

11.5 (11)

PhMeSn2+ as PhSnMeEt2

ndc

1.52 ( 0.59

320

14.9 (15)

Ph2MeSn+ as Ph2SnMeEt

1.09 ( 0.33

ndc

+

a

Peak assignments are suggested from correlation between retention times versus number of carbons and the approximate concentration of those Sn species, as Sn, calculated from the ICP-MS response for Sn. b According to established relationship between retention time and sum of carbon of all four alkyl or aryl ligands. c nd = not detected. 10529

dx.doi.org/10.1021/es202832g |Environ. Sci. Technol. 2011, 45, 10524–10530

Environmental Science & Technology addition to inorganic Sn. Whether TPhT can directly be methylated or the other phenyltins fully methylated cannot be determined by the experimental setup, since those tetraalkyl/aryltin species are lipophilic and rather volatile and would escape into the atmosphere during the experiment. Hence, the degradation scheme (Figure S3, Supporting Information) contains only species that are not fully alkylated and are amenable with ethylation to analysis by GC-ICP-MS. Suggested methylated tin species were also observed in chromatograms obtained for plants exposed to medium TPhT level. For plants exposed to low TPhT content, the concentration of these species could be considered lower than the limit of detection of MPhT. Toxic eﬀects of TPhT on plants are also not well studied or understood. However, TPhT seems to reduce chloroplast activity in aquatic plants, which further prejudices photosynthesis.25 However, in the present experiments, no visible eﬀects could be observed on plants exposed to diﬀerent TPhT concentrations. Finally, it can be stated that phenyltin degradation is more complex than initially thought, and it is slow enough that plants exposed to this biocide can take up this compound and translocate it inside the plants. Although here no organotin species have been measured in the rice grain, no deﬁnite answer can be given, without a ﬁeld experiment, as to whether or not rice grains from paddies treated with TPhT are organotin-free. On the other hand, it could be observed that TPhT is degraded in the ﬁeld, and this aspect was observed in both mesocosm and real ﬁeld conditions. These results are very important, as no evidence of TPhT distribution and speciation in rice plants and paddy ﬁelds has been previously found in the literature, especially for Brazil.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional text, three ﬁgures, and ﬁve tables describing procedures used. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected], [email protected].

’ ACKNOWLEDGMENT We are grateful to INCT-Bio/CNPq, CAPES, and CNPq for supporting this study. ’ REFERENCES (1) Hoch, M. Organotin compounds in the environment - an overview. Appl. Geochem. 2001, 16, 719–743. (2) Appel, K. E. Organotin compounds: Toxicokinetic aspects. Drug Metab. Rev. 2004, 36, 763–786. (3) Ministerio da Agricultura; http://extranet.agricultura.gov.br. (4) State of New Jersey v. EPA, No. 738-R-99-010 (D.C. Cir. Sep 1999). (5) Nath, M. Toxicity and the cardiovascular activity of organotin compounds: a review. Appl. Organomet. Chem. 2008, 22, 598–612. (6) Nakanishi, T. Endocrine disruption induced by organotin compounds; organotins function as a powerful agonist for nuclear receptors rather than an aromatase inhibitor. J. Toxicol. Sci. 2008, 33, 269–276. (7) Atanasov, A. G.; Nashev, L. G.; Tam, S.; Baker, M. E.; Odermatt, A. Organotins disrupt the 11 beta-hydroxysteroid dehydrogenase type

ARTICLE

2-dependent local inactivation of glucocorticoids. Environ. Health Perspect. 2005, 113, 1600–1606. (8) Grote, K.; Hobler, C.; Andrade, A. J. M.; Grande, S. W.; Gericke, C.; Talsness, C. E.; Appel, K. E.; Chahoud, I. Sex diﬀerences in eﬀects on sexual development in rat oﬀspring after pre- and postnatal exposure to triphenyltin chloride. Toxicology 2009, 260, 53–59. (9) Horiguchi, T.; Shiraishi, H.; Shimizu, M.; Morita, M. Eﬀects of triphenyltin chloride and ﬁve other organotin compounds on the development of imposex in the rock shell, Thais clavigera. Environ. Pollut. 1997, 95, 85–91. (10) O’Halloran, K.; Ahokas, J. T.; Wright, P. F. A. Response of ﬁsh immune cells to in vitro organotin exposures. Aquat. Toxicol. 1998, 40, 141–156. (11) Zhang, Z.; Hu, J.; Zhen, H.; Wu, X.; Huang, C. Reproductive inhibition and transgenerational toxicity of triphenyltin on medaka (Oryzias latipes) at environmentally relevant tip levels. Environ. Sci. Technol. 2008, 42, 8133–8139. (12) Guerin, T.; Sirot, V.; Volatier, J. L.; Leblanc, J. C. Organotin levels in seafood and its implications for health risk in high-seafood consumers. Sci. Total Environ. 2007, 388, 66–77. (13) Dubascoux, S.; Lespes, G.; Denaix, L.; Gautier, M. P. Kinetic monitoring of trisubstituted organotins in soil after sewage sludge application. Appl. Organomet. Chem. 2008, 22, 481–487. (14) Yen, J. H.; Tsai, C. C.; Su, C. C.; Wang, Y. S. Environmental dissipation of fungicide triphenyltin acetate and its potential as a groundwater contaminant. Ecotox. Environ. Saf. 2001, 49, 164–170. (15) Lespes, G.; Marcic, C.; Heroult, J.; Le Hecho, I.; Denaix, L. Tributyltin and triphenyltin uptake by lettuce. J. Environ. Manage. 2009, 90, S60–S68. (16) Kannan, K.; Lee, R. F. Triphenyltin and its degradation products in foliage and soils from sprayed pecan orchards and in ﬁsh from adjacent ponds. Environ. Toxicol. Chem. 1996, 15, 1492–1499. (17) Monperrus, M.; Zuloaga, O.; Krupp, E.; Amouroux, D.; Wahlen, R.; Fairman, B.; Donard, O. F. X. Rapid, accurate and precise determination of tributyltin in sediments and biological samples by species speciﬁc isotope dilution-microwave extraction-gas chromatography-ICP mass spectrometry. J. Anal. Atom. Spectrom. 2003, 18, 247–253. (18) Gomez-Ariza, J. L.; Giraldez, I.; Morales, E.; Ariese, F.; Coﬁno, W.; Quevauviller, P. Stability and storage problems in organotin speciation in environmental samples. J. Environ. Monit. 1999, 1, 197–202. (19) Van, D. N.; Lindberg, R.; Frech, W. Redistribution reactions of butyl- and phenyltin species during storage in methanol. J. Anal. Atom. Spectrom. 2005, 20, 266–272. (20) Van, D. N.; Bui, T. T. X.; Tesfalidet, S. The transformation of phenyltin species during sample preparation of biological tissues using multi-isotope spike SSID-GC-ICPMS. Anal. Bioanal. Chem. 2008, 392, 737–747. (21) Glindemann, D.; Ilgen, G.; Herrmann, R.; Gollan, T. Advanced GC/ICP-MS design for high-boiling analyte speciation and large volume solvent injection. J. Anal. Atom. Spectrom. 2002, 17, 1386–1389. (22) Meng, P.-J.; Lin, J.; Liu, L.-L. Aquatic organotin pollution in Taiwan. J. Environ. Manage. 2009, 90, S8–S15. (23) Krupp, E. M.; Merle, J. K.; Haas, K.; Foote, G.; Maubec, N.; Feldmann, J. Volatilization of organotin species from municipal waste deposits: Novel species identiﬁcation and modeling of atmospheric stability. Environ. Sci. Technol. 2011, 45, 943–950. (24) Mitra, S. K.; Jiang, K. J.; Haas, K.; Feldmann, J. Municipal landﬁlls exhale newly formed organotins. J. Environ. Monit. 2005, 7, 1066–1068. (25) Song, Z. H.; Huang, G. L. Toxic eﬀect of triphenyltin on Lemna polyrhiza. Appl. Organomet. Chem. 2005, 19, 807–810.

10530

dx.doi.org/10.1021/es202832g |Environ. Sci. Technol. 2011, 45, 10524–10530

ARTICLE pubs.acs.org/est

Enhanced Organic Phosphorus Assimilation Promoting Biomass and Shoot P Hyperaccumulations in Lolium multiflorum Grown under Sterile Conditions Nilesh C. Sharma*,† and Shivendra V. Sahi† †

Department of Biology, Western Kentucky University, Bowling Green, 1906 College Heights Boulevard, Kentucky 42101-1080, United States ABSTRACT: Search for plant species prodigious in P use is important for both P-suﬃcient and -deﬁcient conditions. Gulf and Marshall ryegrass (Lolium multiﬂorum), grown in sterile media containing diﬀerent organic P substrates (AMP, ATP, GMP, and IHP), exhibited high rates of growth and shoot P concentrations. Growth increase in Gulf was signiﬁcantly greater on IHP relative to other sources of organic P substrates. Growth was also dependent on an increasing concentration of IHP (020 mM) in this cultivar. P accumulations in Gulf exceeded 1% shoot dry weight from IHP, AMP, and ATP— equivalent to the P accrual from equimolar Pi source. Plants supplied with IHP had phytase activity in root extracts comparable to that in Pi-fed plants or control (no P). The extracellular phytase, however, increased by about 100% relative to that observed in root extracts- for both ryegrass cultivars, but there were no signiﬁcant diﬀerences (P < 0.05) between plant groups grown on diﬀerent substrates (IHP, Pi or control). No signiﬁcant diﬀerences in phosphomonoesterase activities were evident between plant groups supplied with organic P (IHP, G1P) and inorganic source or control. This study establishes the high P-use eﬃciency in ryegrass, irrespective of P source.

’ INTRODUCTION Soil chemistry of phosphorus (P) has attracted more attention in recent times than in previous decades, especially in relation to temperate climate agriculture with over application of organic manures rich in P. Soils typically contain a signiﬁcant quantity of P, but only a small proportion (less than 1%) of the total P is available to plants.1 Organic and inorganic fractions constitute the sum total of soil P. Organic P varies in a wide range (3080%) depending on the type of soils,2 phosphomonoesters constituting the bulk (up to 90%) of the organic P fraction.3 Phytate, the derivative of inositol hexakis-phosphate (IHP), is the most abundant form of phosphomonoesters present in most of the soils, particularly in temperate climates.4 Besides phosphomonoesters, sugar phosphates and phosphate diesters also occur in soils, however, in small quantity, about 5% of total organic P.2 Plants are known to uptake only phosphate anions to meet their P requirement. Thus to be available to plants, organic P must be hydrolyzed by the activity of phosphatase enzymes to release phosphate. Phosphatase activity has also been used as an indicator of organic phosphorus mineralization.5 Phosphatase activity in soil is mainly controlled by soil microorganisms6 although the contribution of plants to the pool of soil phosphatase cannot be undermined. Plants secrete extracellular phosphatase in rhizospheres.7,8 In natural ecosystems, mineralization of soil organic P is considered to provide a major source of plant-available P.9,10 Controlled studies have also demonstrated that mineralization of organic P occurs signiﬁcantly in soils to aﬀect phosphate uptake by plants.11,12 However, the direct r 2011 American Chemical Society

contribution of hydrolyzed organic P to plant P nutrition remains unclear. Phosphatases with diﬀering substrate speciﬁcities for phosphomonoesters and phosphodiesters have been characterized in plant roots.8 6-Phytase (EC 3.1.3.26) is a class of acid phosphatase with high aﬃnity for phytate that may constitute up to 80% of a total organic P2 and may thus be especially important for the hydrolysis of organic P in soils. The presence of extracellular phosphatase in plant roots is supported further by other studies that have directly measured many-fold increases in phosphomonoesterase activity in root zones of plants.6,13 Nevertheless, it is diﬃcult to separate the direct contribution of plants to an observed increase in the activity of phosphatases within the rhizosphere because of the fact that simultaneous increase in microbial population and activity also occurs.14,15 Therefore, the following studies focused on assessing organic P utilization by plants grown in sterile media or under controlled conditions. Wheat seedlings, when grown aseptically in media containing diﬀerent forms of organic P, had diﬀerential ability of P acquisition from the diﬀerent organic P substrates.16,17 These studies also demonstrated a correlation between organic P use and the activities of phytase and acid phosphomonoesterase. Transgenic Arabidopsis plants modiﬁed with phytase gene (phyA) derived from Aspergillus niger demonstrated improved growth and Received: March 21, 2011 Accepted: October 28, 2011 Revised: September 30, 2011 Published: October 28, 2011 10531

dx.doi.org/10.1021/es200942v | Environ. Sci. Technol. 2011, 45, 10531–10537

Environmental Science & Technology P nutrition when supplied with phytate as a source of P.18 Improved utilization of phytate was also reported in transgenic clover and barrel medic that expressed phytase genes, phyA and MtPHY1, respectively, and released extracellular phytase.19,20 Annual ryegrass (Lolium multiﬂorum), closely related to perennial ryegrass (Lolium perenne), is grown all over the world as a key forage grass.21 Marshall and Gulf ryegrass diﬀerent cultivars of Lolium multiﬂorum were recently shown to grow and accumulate P in their shoots when seedlings were grown in hydroponic media and soils containing high concentrations of inorganic P.21,22 Using scanning electron microscopy and electron-dispersive X-ray spectroscopy, it was also shown that the accumulation of P occurs in the cortical and stellar regions of ryegrass shoot.21 These studies formed the basis of evolving a P-phytoremediation strategy using above cultivars for the soils loaded with organic manure and P. Mining of soil P, which includes harvesting P taken up from the soil by a crop grown without external P application, has been proposed as a possible management strategy for P-enriched soils.23 In the above backdrop, particularly keeping in view the unique feature of annual ryegrass, it was interesting to reveal in this species the pattern of P nutrition using organic P sources. In the present study, experiments were thus designed to determine i) plant growth and ii) the quantity of total P stored in shoots and elucidate iii) the relationship between organic P use and root activities of phytase and acid phosphomonoesterase when Marshall- and Gulf ryegrass were grown aseptically in media fortiﬁed with a range of organic P substrates (AMP, ATP, GMP, and phytate).

’ EXPERIMENTAL SECTION Seed Germination. Seeds of Marshall- and Gulf ryegrass (Lolium multiflorum cultivars), provided by USDA Lab, Starkville, MS, were sterilized with sodium hypochlorite (1% v/v) and rinsed several times with sterile deionized water. They were then transferred to water-agar (0.8%) medium in Magenta boxes and maintained at 25 ( 2 °C under 12/12 light/dark regime in a growth chamber. Ten day-old seedlings isolated from agar medium were rinsed with deionized water before the aseptic transfer in different experiments. Growth of Seedlings. Half strength Modified Hoagland’s salts mixture (115 mg/L ammonium nitrate, 2.86 mg/L boric acid, 656.4 mg/L calcium nitrate, 3.0 mg/L ferric chloride, 240.7 mg/L magnesium sulfate, 1.81 mg/L manganese chloride, 0.016 mg/L molybdenum trioxide, 400.6 mg/L potassium nitrate, and 0.22 mg/L zinc sulfate) containing 0.8% (w/v) agar was used as a basal growth medium. To examine the effect of different P substrates, medium was supplemented either with no P (control) or 5 mM each of α-D-glucose 1-phosphate disodium salt (G1P), adenosine 30 :50 cyclic monophosphate sodium salt (AMP), adenosine-50 -triphosphate disodium salt (ATP), myoinositol hexakis-phosphoric acid dodecasodium salt (IHP), or KH2PO4 (Pi) [Sigma Chemical Co., St. Louis, MO]. Another experiment was set up to measure the effect of increasing concentrations (020 mM) of IHP. Three plants were transferred aseptically to each culture tube (15 cm 2.5 cm) containing 15 mL of the medium. Plants were grown in a growth chamber at 25 ( 2 °C in 16/8 light/dark (180200 μmol m2 s1 of cool fluorescent light) for different periods (2 and 5 weeks). Each treatment was replicated three times and experiments were repeated (n = 6).

ARTICLE

Determination of Fresh and Dry Weight. Plants harvested either from 2 weeks of growth (on G1P, AMP, ATP, IHP or Pi) or from 5 weeks of growth (on 020 mM IHP) were measured for the growth of biomass. Harvested shoots pooled from different treatment replicates were blotted with the filter paper before the fresh weight determination and dried in an oven at 70 °C for 48 h for dry weight determination. The average weight of three plants per replicate was determined. Growth was measured by the difference in the final and initial weights of each replicate. Analysis of P in Plant Tissue. Plants from different treatments were harvested, washed thoroughly with deionized water, divided into root and shoot biomass, and dried in an oven at 70 °C for two days. The ground samples were then weighed and placed in 15 mL Teflon beakers. Three mL of concentrated HNO3 was added to the sample, and the beaker was placed on a hot plate set at 100 °C overnight, until evaporated to dryness. The samples were allowed to cool and made up gravimetrically to a volume of 20 mL with 2% HNO3. A VG Elemental Plasma Quad (model PQZ) ICP-AES was used for all data acquisition. Analyses were performed using an external calibration procedure, and internal standards were included to correct for matrix effects and instrumental drift corrections.21,22 Preparation of Root Extracts and Phosphomonoesterase Assay. After 2 weeks of growth on different P treatments, plants were harvested and washed thoroughly with DI water followed by rinse in 2-morpholinoethanesulfonic acid, monohydrate (MES) buffer solution (pH 5.5). Roots were separated, chilled on ice, and homogenized with a mortar and pestle in 15 mM MES buffer (pH 5.5, 0.5 mM CaCl2 3 H2O, 1 mM EDTA). The buffer was added at a ratio of 1:5 (root fresh weight: extraction buffer volume). The extract was centrifuged (13,000 g; 15 min at 4 °C), and the supernatant was used for enzyme assay.16 For the assay of phosphomonoesterase activity, enzyme extract (50 μL) was incubated in a total volume of 500 μL of 15 mM MES buﬀer (pH 5.5, 0.5 mM CaCl2) in the presence of 10 mM p- nitrophenylphosphate, disodium salt (Sigma-Aldrich, St. Louis, MO).16,18 The assay was conducted over 30 min, and reactions were terminated by equal volumes of 0.25 M NaOH. The enzyme activity was calculated from the release of p-nitrophenol (pNP), determined at 412 nm (relative to standard solutions) by a UVvis spectrophotometer (Model Ultrospec 3000, Pharmacia Biotech, USA). Phytase Assay. Harvested plant roots were treated and homogenized as described above. To assay for phytase activity, a 500 μL enzyme extract was incubated in a total volume of 1 mL of 15 mM MES buffer (pH 5.5, 0.5 mM CaCl2) in the presence of 2 mM myo-inositol hexaphosphoric acid (Sigma-Aldrich, St. Louis, MO).16,18 The assay was conducted over 60 min, and reactions were terminated by addition of equal volumes of ice-cold 10% trichloroacetic acid (TCA). Solutions were subsequently centrifuged to remove precipitated material, and their phosphate concentrations were determined by measuring absorbance at 882 nm using the molybdenum-blue reaction.16 Phosphate concentrations were recorded at a fixed time within 1 h following addition of the color reagent to samples, to minimize possible interference. The enzyme assays were conducted at 26 °C using three replicates. Phosphomonoesterase and phytase activities were expressed in mU g1 root fresh weight (FW), where 1 U is defined as the release of 1 μmol of Pi min1 under the assay conditions. Extracellular phytase activity was determined for plants harvested from media containing 5 mM each of IHP, Pi, or control 10532

dx.doi.org/10.1021/es200942v |Environ. Sci. Technol. 2011, 45, 10531–10537

Environmental Science & Technology

ARTICLE

Figure 1. (A-B). A. Fresh weight of Gulf (a-c) and Marshall ryegrass (d-f) B. Dry weight of Gulf (a-d) and Marshall ryegrass (e-f) grown in sterile media containing 5 mM each of G1P, AMP, ATP, IHP, Pi, or no P (control) for a period of 2 weeks. Treatment means (n = 6) labeled with not the same notations are signiﬁcantly diﬀerent (P < 0.05) in each cultivar.

(no P). After the growth period (2 weeks), plants were carefully removed from the agar and incubated in 10 mL of modiﬁed halfstrength Hoagland’s solution (containing no P) for another 16 h under light to collect root exudate. Aliquots of root exudate were analyzed for the enzyme activity as described above. Phytase activity was expressed in mU g1 root fresh weight/day, where 1 U is deﬁned as the release of 1 μmol of Pi min1 under the assay conditions. Statistical Analysis. The data were analyzed by one-way analyses of variance using SYSTAT (Version 9 for Windows, 1999, Systat Software Inc., Richmond, CA). Where variance ratios were significant (p < 0.05), treatment means were compared using LSD (P = 0.05) for each cultivar separately.

’ RESULTS Plant Growth on Media Containing Different P Substrates or No P for 2 Weeks. Gulf ryegrass registered differential fresh

weight growth on different organic substrates (Figure 1A). Growth on G1P (236 mg) and ATP (250 mg) was not significantly different than in control in both cultivars (Figure 1A). However, Gulf displayed growth significantly greater on AMP (312 mg) and IHP (490 mg) than control. IHP-supplemented medium supported even greater increase in fresh weight than Pi containing medium (381 mg). The pattern of growth in Marshall ryegrass was different; when growth on G1P (213 mg) was comparable to control (180 mg), it declined on AMP (156 mg) (Figure 1A). Further, Marshall registered significantly greater growth on ATP (251 mg) and IHP (282 mg) with respect to control but not with respect to Pi (266 mg). A similar trend was noticed in respect of dry weight increase in both cultivars (Figure 1B). Dry weight increase in Gulf was signiﬁcantly greater on organic substrates (25.357.8 mg) than control (18.2). Gulf growth in IHP (57.8 mg) was again signiﬁcantly greater than in other organic substrates (Figure 1B). Marshall registered dry weight growth in a relatively narrow range of 19.731.3 mg; IHP growth not being signiﬁcantly greater than Pi growth (Figure 1B) similar to the fresh weight pattern. Plant Growth on Media Containing 020 mM of IHP or 20 mM of Pi for 5 Weeks. Both cultivars displayed a significant increase (P < 0.05) in fresh weight depending on increasing concentrations of IHP (Figure 2A). Fresh weight growth ranged from 396 to 584 mg on increasing concentrations of IHP (120 mM). Again, growth in Gulf was greater than in Marshall on all concentrations of IHP (Figure 2A). A similar trend was noticeable with respect to dry weight increase in Gulf (Figure 2B). In

Figure 2. (A-B). A. Fresh weight of Gulf (a-d) and Marshall ryegrass (ef) B. Dry weight of Gulf (a-d) and Marshall ryegrass (e-f) grown in sterile media containing (020 mM) of IHP or 20 mM of Pi for a period of 5 weeks. Treatment means (n = 6) labeled with not the same notations are signiﬁcantly diﬀerent (P < 0.05) in each cultivar.

Marshall, dry weight increased with an increase in IHP concentration up to 10 mM and declined thereafter. However, Marshall dry weight gain at 10 mM IHP was comparable to that of 20 mM Pi. Accumulation of P by Plants Grown on Media Containing Different P Substrates or No P for 2 Weeks. Shoot accumulation of P was significantly higher (P < 0.05) on all media than control (no P). Accumulations of P from organic substrates (AMP, ATP, and IHP) ranged from 11,47011,950 mg/kg greater than 1.1% shoot dry weight in Gulf. These accumulations were comparable to the accumulation from inorganic P source, KH2PO4 (11,890 mg/kg) (Figure 3A). Marshall ryegrass exhibited a differential pattern with a maximum P accumulation of 11,930 mg/kg over 1.1% shoot dry weight from ATP-containing medium, followed by accumulation from AMP- medium (Figure 3A). The accumulation from IHP medium was not significantly different (P > 0.05) than from Pi in this cultivar as well. Accumulation of P by Plants Grown on Media Containing 020 mM of IHP or 20 mM of Pi for 5 Weeks. In both cultivars, P accumulation was dependent on increasing concentrations up to 10 mM IHP, after which accumulation leveled off (Figure 3B). However, accumulations in Gulf were significantly greater (P < 0.05) than in Marshall up to 10 mM IHP. There were no significant differences in accumulations from 20 mM IHP and 20 mM Pi in both cultivars (Figure 3B). Phytase and Phosphomonoesterase Activities in Plant Roots Grown on Different P Substrates or No P for 2 Weeks. Figure 4A shows a differential pattern of phytase activities. Both cultivars grown on IHP exhibited a similar activity of phytase 10533

dx.doi.org/10.1021/es200942v |Environ. Sci. Technol. 2011, 45, 10531–10537

Environmental Science & Technology

ARTICLE

Figure 3. (A-B). Shoot P concentration in Gulf (a-c) and Marshall ryegrass (d-f) grown in sterile media containing A. 5 mM each of G1P, AMP, ATP, IHP, Pi, or no P (control). B. 020 mM of IHP or 20 mM of Pi for a period of 5 weeks. Treatment means (n = 6) labeled with not the same notations are signiﬁcantly diﬀerent (P < 0.05) in each cultivar.

Figure 4. (A-B). A. Phytase activity. B. Phosphomonoesterase activity in Gulf (a-b) and Marshall ryegrass (c-e) grown in sterile media containing 5 mM each of G1P, AMP, ATP, IHP, Pi, or no P (control) for a period of 2 weeks. Treatment means (n = 6) labeled with not the same notations are signiﬁcantly diﬀerent (P < 0.05) in each cultivar.

comparable to control (no P) and Pi-plants, ranging from 67.3 mU/g (FW). However, phytase activities were reduced, approximately, 3-fold in media enriched with G1P, AMP, or ATP. Phosphomonoesterase activities of plants grown in G1P, IHP, or Pi were not signiﬁcantly diﬀerent than control (without P), varying in a range of 600740 mU/g FW (Figure 4B). Similar to phytase activities, plants grown in AMP or ATP had lower values of phosphomonoesterase activity (400450 mU/g FW). Extracellular Phytase Activity in Plants Harvested from IHP, Pi, or Control Media. Secreted phytase activity in Gulf and Marshall ryegrass grown on media containing 5 mM each of IHP, Pi, or control (no P) was measured. The secreted activity ranged from 13.018.2 mU/g FW/day in both cultivars (Table 1). Phytase activities among different groups of Gulf were not significantly different, while activity was significantly different in Marshall grown in IHP and Pi.

’ DISCUSSION Both cultivars of annual ryegrass registered signiﬁcant fresh and dry weight growth when grown in the presence of diﬀerent P substrates with respect to control (without P) (Figure 1A-B). However, biomass increase in Gulf ryegrass was generally higher than in Marshall ryegrass on organic and inorganic P (KH2PO4) sources. Growth in these grasses also correlated with increasing concentrations (120 mM) of IHP (Figure 2A-B). An interesting pattern noticed in case of Gulf ryegrass was that its growth on IHP (20 mM) was comparable to that in equimolar Pi medium.

However, it cannot be undermined that equimolar IHP can release three times more Pi on enzyme-catalyzed hydrolysis. Growth proﬁle of these cultivars certainly indicates that these plants can draw P from organic substrates, particularly IHP, as eﬃciently as from an inorganic source to reach an optimum level of growth. Several studies show that growth had signiﬁcantly reduced when Arabidopsis,18 Trifolium,19 and Solanum24 wild-type plants were grown with organic substrates, particularly IHP. Wheat plants grown in IHP, in similar conditions, exhibited more than 40% reduction in mean shoot dry weight compared to Pi-fed controls.16 However, shoot growth in wheat was not signiﬁcantly diﬀerent (P < 0.05) for other sources of P treatments (AMP, G1P, ATP, and PGA) relative to Pi-fed controls. Figure 3(A-B) clearly indicates that both cultivars had signiﬁcantly high shoot P accumulations from organic and inorganic P sources. Gulf plant acquisitions from IHP, AMP, and ATP exceeded 1% (shoot DW) similar to its uptake from Pi. Marshall exhibited a maximum shoot P concentration (>1% DW) when grown on ATP; P acquisition from IHP not being signiﬁcantly (P < 0.05) diﬀerent than from Pi. When ryegrass P use from IHP surpassed many wild-type species (Arabidopsis, Solanum, Trifolium) tested in similar conditions, it compared well with transgenic lines of those species expressing Aspergillus phytase gene, phyA.18,19,24 Transgenic Arabidopsis lines expressing ex: phyA demonstrated several-fold increase in P use and plant growth when grown in IHP-supplemented media.18 Shoot P concentrations in these lines exceeded 1% (shoot DW) as observed in our study. Similarly, P acquisition from phytate in sterile media 10534

dx.doi.org/10.1021/es200942v |Environ. Sci. Technol. 2011, 45, 10531–10537

Environmental Science & Technology

ARTICLE

Table 1. Secreted Phytase Activity (mU/g Root Fresh Weight/Day) in Gulf and Marshall Ryegrass Grown on Media Containing 5 mM Each of IHP, Pi, or no P (Control)a media containing IHP KH2PO4 control (0P)

Gulf ryegrass

Marshall ryegrass

a

14.3a

a

18.2b

a

16.9a

16.0 13.7

13.0

a

Values represent the mean of 6 replicates and, within each column, those not followed by the same letter are signiﬁcantly diﬀerent (P < 0.05).

was increased substantially in a transgenic potato24 and Arabidopsis25 expressing root hair-speciﬁc phytase—similar to our observations in this study. High rates of P accumulation in the present study are also consistent with earlier reports when Gulf and Marshall ryegrass was characterized as a potential P accumulator using diﬀerent regimes of inorganic P.21,22 It is evident from Figure 4A that plants supplied with IHP had phytase activity in soluble extracts comparable to activities in Pifed plants and control (no P). The activity in Gulf and Marshall ryegrass was unlike wheat where plants grown with no P had signiﬁcantly greater activity (P < 0.05) than those grown in IHP and Pi.16 Nevertheless, wild-type Arabidopsis had phytate activity trend comparable to our observations in this study.18 It was also observed that phytase activity in ryegrass was signiﬁcantly reduced on G1P, AMP, and ATP relative to IHP. Similar reduction, however, was not uncommon to wheat and Arabidopsis grown on G1P.16,18 The phytase activity in the external solution (secreted phytase), however, increased by about 100% relative to that observed in soluble extracts in both of the cultivars, but there were no signiﬁcant diﬀerences (P < 0.05) between plant groups (IHP, Pi, or control). Figure 4B shows a diﬀerent pattern of phosphomonoesterase activities in ryegrass. No signiﬁcant differences in phosphomonoesterase activities were evident between plant groups that were supplied with no P or inorganic or organic P sources (IHP, G1P)—similar to the trend reported in wheat and Arabidopsis.16,18 From the above account it is evident that ryegrass, compared to many plant species, exhibits high eﬃciency in IHP utilization as reﬂected in appreciable shoot biomass and P accumulations despite having phytase activity equivalent to 12% of total phosphomonoesterase. Interestingly, the level of P and biomass accumulations in IHP-plants matched to the plants grown in inorganic P. However, some intriguing questions remain with regard to AMP- and ATP-plants where a reduced phosphomonoesterase activity negatively correlated (P < 0.05) to the increased P accumulation. It can be understood that AMP, belonging to the class of phosphodiesters, requires phosphodiesterases to be catalyzed, and the same was not estimated in the present study. Several transgenic lines of Trifolium,19 Solanum,24 and Arabidopsis25 overexpressing (several-fold) extracellular phytase have been reported to acquire P from IHP equal or comparable to the level of contributions that accrue from inorganic source of P. From this standpoint, ryegrass is unique with eﬃcient P nutrition and biomass production utilizing IHP at the expense of a small amount of phytase (12% of total phosphomonoesterases). The observation that enzyme activities are independent of P supply (deﬁciency in case of control- no P- and P suﬃciency in the case of IHP or Pi) makes Gulf ryegrass more interesting with respect to P nutrition. It also points to the probability of

constitutive expression of phosphatase genes, inﬂuencing P metabolism in this species. Only further studies can answer this hypothesis. Many researchers recently attempted to answer the question as why several plant species demonstrate poor ability to utilize phytate from soil and have scanty phytase production despite possessing the phytate gene (phyA). The consensus view that emerges is that the evolutionary pressure for utilization of P from phytate is constrained by factors such as i) low availability of phytate in soil solution or rhizosphere that can be accessed by plants and ii) propensity of phytate to undergo precipitation and sorption reactions in soil environments.18 In this context, what makes plant phytase a critical player in the use of P from phytate is its copious secretion into soil to reach out its substrate. This fact is further substantiated by the demonstrated role of transgenic lines (overexpressing extracellular phytase) belonging to a number of species discussed earlier.19,24,25 Thus the search for wild-type crop species or development of transgenics that are more eﬃcient in P utilization from soil would be particularly beneﬁcial to agriculture in reducing the dependence on P-based fertilizers with diminishing natural resource. Another dimension of P use-eﬃcient species is in their application in P mining from P-loaded soils P phytoremediation, as argued by Novak and Chan.23 Though there are few species reported so far as an eﬀective P remover from soils, annual ryegrass is an interesting candidate for P phytoremediation.22 Studies have shown annual ryegrass eﬃcacy in P removal from P-enriched conditions: hydroponic medium21 to soils acidic, alkaline,22 to soils enriched with poultry litter.26 The instant study while conﬁrming earlier reports of high P use and accumulation in annual ryegrass provides evidence for a biochemical basis of P nutrition from organic P sources, particularly phytate. Hayes et al.,9 when compared P nutrition between some pasture legume and grass species, observed that the legume species (Trifolium spp. and Medicago polymorpha) were more eﬃcient in P use and had higher levels of phytase activity compared to grass species. Higher P acquisition eﬃciency in legumes was also implicated with plant attributes such as follows: I. high storage capacity for inorganic P; II. a favorable ratio of P uptake per unit root length; and III. a high activity of acid phosphatase in the root and capacity to use P from organic P sources when they were grown in soils enriched with organic P.27 This study, in conjunction with previous observations on ryegrass, amply shows that annual ryegrass possesses all these attributes to be eﬃcient in P use. In an interesting comparative study, when Duo grass (Duo festulolium) a hybrid between fescue and ryegrass was grown in similar conditions, sterile solutions containing IHP and other organic substrates, it exhibited comparable phytase and signiﬁcantly greater (<100%) phosphomonoesterase activities relative to annual ryegrass enzyme activities.28 However, its shoot P concentration was observed to be mani-fold less than in Gulf or Marshall ryegrass. This again indicates that a combination of plant attributes, rather than only the enzyme production, is the key determinant to P nutrition in a species. Among the molecular responses, Pi starvation-induced high aﬃnity Pi transporters play a pivotal role in the acquisition and mobilization of Pi in plants, but low aﬃnity Pi transporters may play a major role in the movement of Pi along the concentration gradient.29,30 In one recent study on characterization of Pi responsive genes in Gulf ryegrass, overexpression of LmPAP1 (Lolium multiﬂorum purple acid phosphatase gene) was 10535

dx.doi.org/10.1021/es200942v |Environ. Sci. Technol. 2011, 45, 10531–10537

Environmental Science & Technology implicated in P accumulation.31 Transgenic expression of a purple acid phosphatases gene (derived from M. truncatula) in Arabidopsis also resulted in improved phosphorus acquisition and biomass production.32 Some purple acid phosphatases (AtPAP26 or its orthologs) are constitutively transcribed irrespective of plant’s nutritional Pi status.33 An interesting current study demonstrated the functional characterization of putative organic Pi transporters (G3Pp family) involved in organic P mobilization in Arabidopsis.30 Thus we suggest that Gulf and Marshall ryegrass has evolved mechanisms to hyperaccumulate P under diﬀerent P regimes by activating multiple Pi acquisition and mobilization mechanisms.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 270-745-6593. Fax: 270-745-6856. E-mail: nilesh.sharma@ wku.edu.

’ ACKNOWLEDGMENT Authors acknowledge the Oﬃce of Sponsored Program (ARTP) of Western Kentucky University and the U.S. Department of Agriculture (Grant 58-6406-1-017) for supporting this research. Authors duly acknowledge the contribution of Dr. Ajay Jain (NRCPB, New Delhi-110012, India) in improving the quality of ﬁgures and for his suggestions. ’ REFERENCES (1) Raghothama, K. G. Phosphorus and plant nutrition: an overview. In Phosphorus: Agriculture and the Environment; Sims, J. T., Sharpley, A. N., Eds.; ASA-CSSA-SSSA: Madison, WI, 2005; pp 355378. (2) Dalal, R. C. Soil Organic Phosphorous. Soil organic phosphorus. Adv. Agron. 1977, 29, 85–117. (3) Condron, L. M.; Turner, B. L.; Cade-Menum, B. J. Chemistry and dynamics of soil organic phosphorus. In Phosphorus: Agriculture and the Environment; Sims, J. T., Sharpley, A. N., Eds.; ASA-CSSA-SSSA: Madison, WI, 2005; pp 87121. (4) Turner, B. L.; Paphazy, M. J.; Haygarth, P. M.; Mc-Kelvie, I. D. Inositol phosphates in the environment. Philos. Trans. R. Soc., London 2002, B 357, 449–469. (5) Quiquampoix, H.; Mousain, D. Enzymatic hydrolysis of organic phosphorous. In Turner, B. L., Frossland, E., Baldwin, D. S., Eds.; Organic Phosphorous in the Environment; CABI Publishing: Cambridge, MA, 2005; pp 89112. (6) George, T. S.; Gregory, P. J.; Wood, M.; Read, D.; Buresh, R. J. Phosphatase activity and organic acids in the rhizosphere of potential agroforestry species and maize. Soil Biol. Biochem. 2002, 34, 1487–1494. (7) Hayes, J. E.; Richardson, A. E.; Simpson, R. J. Phytase and acid phosphatase activities in extracts from roots of temperate pasture grass and legume seedlings. Aust. J. Plant Physiol. 1999, 26, 801–809. (8) Raghothama, K. G. Phosphate acquisition. Ann. Rev. Plant Physiol. Plant Mol. Biol. 1999, 50, 665–693. (9) Fox, T. R.; Comerford, N. B. Rhizosphere phosphatase activity and phosphate hydroly- zable organic phosphorous in two forested spodosols. Soil Biol. Biochem. 1992, 24, 579–583. (10) Polglase, P. J.; Attiwill, P. M.; Adams, M. A. Nitrogen and phosphorous cycling in relation to stand age of Eucalyptus reglans. F. Muell. III. Labile inorganic and organic P, phosphatase activity and P availability. Plant Soil 1992, 142, 177–185. (11) Lopez-Hernandez, D.; Brossard, M.; Frossard, E. P-isotopic exchange values in relation to Po mineralization in soils with low P-sorbing capacities. Soil Biol. Biochem. 1998, 30, 1663–1670. (12) Oehl, F.; Oberson, A.; Sinaj, S.; Frossard, E. Organic phosphorous mineralization studies using isotopic dilution techniques. Soil Sci. Soc. Am. J. 2001, 65, 780–787.

ARTICLE

(13) Li, M.; Osaki, M.; Rao, I. M.; Tadano, T. Secretion of phytase from the roots of several plant species under phosphorus-deﬁcient conditions. Plant Soil 1997, 195, 161–169. (14) Tarafdar, J. C.; Jungk, A. Phosphatase activity in the rhizosphere and its relation to the depletion of soil organic phosphorous. Biol. Fertil. Soils 1987, 3, 199–204. (15) Chen, C. R.; Condon, L. M.; Davis, M. R.; Sherlock, R. R. Phosphorous dynamics in the rhizosphere of perennial ryegrass (Lolium perenne) and radiata pine (Pinus radiata D. Don). Soil Biol. Biochem. 2002, 34, 487–499. (16) Richardson, A. E.; Hadobas, P. A.; Hayes, J. E. Acid phosphomonoesterase and phytase activities of wheat (Triticum aestivum L.) roots and utilization of organic phosphorus substrates by seedlings grown in sterile culture. Plant Cell Environ. 2000, 23, 397–405. (17) Tarafdar, J. C.; Claassen, N. Organic phosphorous utilization by wheat plants under sterile conditions. Biol. Fertil. Soils 2003, 39, 25–29. (18) Richardson, A. E.; Hadobas, P. A.; Hayes, J. E. Extracellular secretion of Aspergillus phytase from Arabidopsis roots enables plants to obtain phosphorus from phytate. Plant J. 2001, 25, 641–649. (19) George, T. S.; Richardson, A. E.; Hadobas, P. A.; Simpson, R. J. Characterization of transgenic Trifolium subterraneum L. which expresses phyA and releases extracellular phytase: growth and P nutrition in laboratory media and soil. Plant Cell Environ. 2004, 27, 1351–1361. (20) Xiao, K.; Harrison, M. J.; Wang, Z.-Y. Transgenic expression of a novel M. truncatula phytase gene results in improved acquisition of organic phosphorus by Arabidopsis. Planta 2005, 222, 27–36. (21) Sharma, N. C.; Sahi, S. V.; Jain, J. C.; Raghothama, K. G. Enhanced accumulation of phosphate by Lolium multiﬂorum cultivars grown in phosphate-enriched medium. Environ. Sci. Technol. 2004, 38, 2443–2448. (22) Sharma, N. C.; Sahi, S. V. Characterization of phosphate accumulation in Lolium multiﬂorum for remediation of phosphorusenriched soils. Environ. Sci. Technol. 2005, 39, 5475–5480. (23) Novak, J. M.; Chan, A. S. K. Development of P-hyperaccumulator plant strategies to remediate soils with excess P concentrations. Crit. Rev. Plant Sci. 2002, 21, 493–509. (24) Zimmermann, P.; Zardi, G.; Lehmann, M.; Zelder, C.; Amrhein, N.; Frossard, E.; Bucher, M. Engineering the root-soil interface via targeted expression of a synthetic phytase gene in trichoblasts. Plant Biotechnol. J. 2003, 1, 353–360. (25) Mudge, S. R.; Smith, F. W.; Richardson, A. E. Root speciﬁc and phosphate regulated expression of phytase under the control of a phosphate transporter promoter enables Arabidopsis to grow on phytate as a sole P source. Plant Sci. 2003, 165, 871–878. (26) Starnes, D. L.; Padmanabhan, P.; Sahi, S. Eﬀect of P sources on growth, P accumulation and activities of phytase and acid phosphatases in two cultivars of annual ryegrass (Lolium multiﬂorum L.). Plant Physiol. Biochem. 2008, 46, 580–58. (27) Rao, I. M.; Borrero, V.; Ricaurte, J.; Garcia, R. Adaptive attributes of tropical forage species to acid soils. V. Diﬀerences in phosphorus acquisition from less available inorganic and organic sources of phosphate. J. Plant Nutr. 1999, 22, 1175–1196. (28) Padmanabhan, P.; Sahi, S. V. Inﬂuence of phosphorus nutrition on growth and metabolism of Duo grass (Duo festulolium). Plant Physiol. Biochem. 2009, 47, 31–36. (29) Jain, A.; Vasconcelos, M. J.; Raghothama, K. G.; Sahi, S. Molecular mechanisms of plant adaptation to phosphate deﬁciency. Plant Breed. Rev. 2007, 29, 360–419. (30) Ramaiah, M.; Jain, A.; Baldwin, J. C.; Karthikeyan, A. S.; Raghothama, K. G. Characterization of the phosphate starvation-induced Glycerol-3-phosphate permease ene family in Arabidopsis. Plant Physiol. 2011, 157, 279–291. (31) Venkatachalam, P; Jain, A.; Sahi, S.; Raghothama, K. Molecular cloning and characterization of phosphate (Pi) responsive genes in Gulf ryegrass (Lolium multiﬂorum L.): a Pi hyperaccumulator. Plant Mol. Biol. 2009, 69, 1–21. 10536

dx.doi.org/10.1021/es200942v |Environ. Sci. Technol. 2011, 45, 10531–10537

Environmental Science & Technology

ARTICLE

(32) Xiao, K.; Katagi, H.; Harrison, M.; Wang, Z.-Y. Improved phosphorus acquisition and biomass production in Arabidopsis by transgenic expression of a purple acid phosphatases gene from M. truncatula. Plant Sci. 2006, 170, 191–202. (33) Plaxton, W. C.; Tran, H. T. Metabolic adaptation of phosphatestarved plants. Plant Physiol. 2011, 157, 1006–1015.

10537

dx.doi.org/10.1021/es200942v |Environ. Sci. Technol. 2011, 45, 10531–10537

ARTICLE pubs.acs.org/est

Boron Accumulation and Toxicity in Hybrid Poplar (Populus nigra euramericana) Rainer Rees,†,* Brett H. Robinson,‡ Manoj Menon,†,^ Eberhard Lehmann,§ Madeleine S. G€unthardt-Goerg,|| and Rainer Schulin† †

Institute of Terrestrial Ecosystems, ETH Z€urich, Universit€atstrasse 16, 8092 Z€urich, Switzerland Soil and Physical Sciences, Lincoln University, Burns 222, P.O. Box 84, Lincoln 7647, Christchurch, New Zealand § Spallation Neutron Source Division, Paul-Scherrer-Institut, 5232 Villigen PSI, Switzerland Swiss Federal Institute for Forest, Snow, and Landscape Research WSL, Z€urcherstrasse 111, 8903 Birmensdorf, Switzerland

)

‡

bS Supporting Information ABSTRACT: Poplars accumulate high B concentrations and are thus used for the phytomanagement of B contaminated soils. Here, we performed pot experiments in which Populus nigra euramericana were grown on a substrate with B concentrations ranging from 13 to 280 mg kg1 as H3BO3. Salix viminalis, Brassica juncea, and Lupinus albus were grown under some growing conditions for comparison. Poplar growth was unaﬀected at soil B treatment levels up to 93 mg kg1. Growth was progressively reduced at levels of 168 and 280 mg kg1. None of the other species survived at these substrate B levels. At leaf B concentrations <900 mg kg1 only <10% of the poplar leaf area showed signs of toxicity. Neutron radiography revealed that chlorotic leaf tissues had B concentrations of 10002000 mg kg1, while necrotic tissues had >2000 mg kg1. Average B concentrations of up to 3500 mg kg1 were found in leaves, while spots within leaves had concentrations >7000 mg kg1, showing that B accumulation in leaf tissue continued even after the onset of necrosis. The B accumulation ability of P. nigra euramericana is associated with B hypertolerance in the living tissue and storage of B in dead leaf tissue.

’ INTRODUCTION At low concentrations, boron (B) is an essential plant and animal micronutrient.1 Recent studies suggest that B is also essential for humans.2 Boron deﬁciencies in plants have been reported in over 80 countries for a total of 132 crops.3 Like other trace elements, B becomes toxic for plants at elevated concentrations. The concentration range between B deﬁciency and toxicity is smaller than that for any other nutrient element.4 Boron is transported from soil into roots and thence into stems and leaves primarily by convection, with the stream of transpiration water.5 However, active metabolic-driven uptake has been shown to occur under B deﬁciency conditions.6 High levels of B occur naturally in many soils of arid regions.7 In addition, human activities can lead to high soil B concentrations, such as the irrigation of agricultural ﬁelds with B-laden water, coal mining, or ﬂy ash deposition onto agricultural land.7,8 Poplars (Populus spp.) are used for wood production, supplementary stock fodder during times of drought, and for the phytomanagement of contaminated sites.9,10 Due to their high transpiration rates and B accumulation, poplars have been employed in B phytoremediation to reduce B leaching from contaminated sites into receiving waters.10 Removal of B from contaminated sites can be achieved by harvesting the aboveground biomass.10 Boronenriched poplar twigs and leaves from contaminated sites could be used as livestock forage, providing a supplementary source of this essential trace element.11 r 2011 American Chemical Society

Depending on growth conditions, poplar clone, B application form and salinity, B accumulation in poplar leaves ranges between 500 and 1200 mg kg1, greatly exceeding the B concentrations of the growing substrate and the poplar stems.10,12,13 In comparison to other species, the B accumulation of poplars was much higher in these studies. Apart from ﬁeld surveys where B accumulation in poplars was found,14 there have been no studies following the original report by Ba~nuelos et al.,12 investigating the B accumulation of poplars in more detail, including bioaccumulation factors and B threshold concentrations compared to other species. Various Salix species have been shown to accumulate leaf B concentrations >800 mg kg1, exceeding those of poplars grown on the same ﬂy ash disposal site, rendering also Salix interesting for the purpose of B extraction from contaminated soil.15 The phytoextraction eﬃciency of a plant species for a trace element depends on the respective accumulated concentration of the element and the amount of harvestable biomass.16 Brassica juncea is widely touted for use in phytoremediation and was reported to exhibit a high B tolerance.17 Despite its lower biomass production

Received: May 4, 2011 Accepted: November 3, 2011 Revised: September 28, 2011 Published: November 03, 2011 10538

dx.doi.org/10.1021/es201100b | Environ. Sci. Technol. 2011, 45, 10538–10543

Environmental Science & Technology compared to poplars or willows, the phytoextraction eﬃciency of B. juncea may be similar if its B accumulation were higher. Boron accumulation varies widely among diﬀerent parts of a plant, necessitating the analyses of all plant parts for their B concentration in order to elucidate the total B accumulation.18 The increase of leaf B concentration during the growing period makes it diﬃcult to determine toxicity thresholds for leaf B concentrations by foliar analysis, as B concentrations can vary considerably between old and young leaves. Moreover, when toxicity symptoms become visible in leaves, B concentrations can vary over several orders of magnitude even within single leaves.18,19 Therefore, the distribution of B not only among but also within leaves needs to be analyzed for the determination of B toxicity thresholds in leaf tissue. Various techniques have been applied to measure the spatial B concentration in leaves.1921 However, these methods are either time-consuming, produce an incomplete picture of the B distribution within the leaves, or their suitability for high B concentrations has not been shown. In this study, neutron radiography (NR) was applied for the ﬁrst time to analyze the spatial distribution of 10B in leaves. While the transfer of B from soil into the shoots of poplars is of great interest with respect to potential phytomanagement of contaminated sites, there is little knowledge on B accumulation by poplars. Therefore, the objectives of this study were to determine (1) the aboveground accumulation of B by Populus nigra euramericana in comparison to Salix viminalis, B. juncea, and Lupinus albus and their tolerance to B in soil under controlled growing conditions, (2) the accumulation of B in roots, shoots and leaves of poplars, and (3) the distribution of B within individual poplar leaves in order to identify B threshold concentrations at which the tissue becomes chlorotic or necrotic.

’ MATERIALS AND METHODS Plant Growth. Populus nigra euramericana, (clone “Dorskamp”), S. viminalis (spp.), B. juncea (spp.), and L. albus (L.) plants were grown on a potting mix (PM) under greenhouse conditions with natural lighting at the Swiss Federal Research Institute, WSL (Birmensdorf, 47 210 1600 N, 8 260 1600 O), Switzerland. Populus was chosen because of its known B accumulation and phytoremediation potential of B contaminated sites.10 Salix viminalis and B. juncea were chosen as alternative phytoremediation plants that are often used or proposed for the phytoremediation of contaminated sites,22,23 and L. albus was selected because of the phloem mobility of B in this species.24 Apart from the control treatment with no added B, three soil B treatments were initially established by spiking the PM substrate with different amounts of 10B-enriched H3BO3 (10B > 96%, EaglePicher Technologies, Quapaw, USA). The resulting HNO3- and CaCl2-extractable B concentrations of the substrates, which showed a linear relationship (r2 = 0.88; y = 0.50x 13.1; p < 0.001), are given in Table S1 (Supporting Information (SI)). The chosen B treatments represent the range of soil B concentrations reported in previous studies on B uptake by poplars from contaminated soils.10,13,15 Nitric acid and CaCl2-extractable concentrations of macroand micronutrients in the PM substrate are given in Table S2 of the SI. The pH (CaCl2, substrate: 0.01 mol CaCl2 ratio: 1: 2.5) of the substrate was 5.0, the total carbon concentration was 270.6 g kg1 and the nitrogen concentration was 6.78 g kg1. In April 2005, we prepared three replicate pots (5.5 L) for each treatment and plant species and planted 3 plants in each pot. Planting occurred immediately after the pots were ﬁlled with ca. 4 kg of substrate. P. nigra euramericana and S. viminalis were

ARTICLE

planted as cuttings (20 cm in length and 1 cm diameter), L. albus and B. juncea as seeds. Two weeks after planting, all plants were thinned to one plant per pot. Because S. viminalis, L. albus, and B. juncea did not grow at substrate B concentrations of 168 and 280 mg kg1 two intermediate treatments were set up on the same occasion with B concentrations of 22 and 45 mg kg1. P. nigra euramericana was not planted in these two additional B treatments. The control treatment and the ﬁve B treatments are denoted as T13, T22, T45, T93, T168 and T280 according to the total initial B concentration of the respective substrate. Pots were irrigated with tap water 34 times per week to about ﬁeld capacity, e.g., to the point where water started to drain into the trays. The leachates were collected and reapplied to the pots. All plants were harvested after four months of growth. The aboveground biomass was separated into leaves, stems, and in the case of B. juncea, also into pods. For P. nigra euramericana and S. viminalis, only the new shoot growth and not the originally planted cuttings were used for analysis. Roots were separated from the substrate by washing with tap water, followed by rewashing with deionized water to remove small particles. Fine roots were collected using a 2 mm Nylon sieve. Plant biomass was dried until constant weight was obtained and the biomass was recorded. For P. nigra euramericana we also recorded the position of the leaves in the sequence along the shoot starting with the ﬁrst leaf at the bottom of the plant. Neutron Radiography. We used 10B-enriched B to determine the areal distribution of accumulated B within leaves by means of neutron radiography.25,26 The neutron absorption cross section of 10B as determined at ICON (Instrument for Cold Neutron Radiography) is 8720 E24 cm2. This is several orders of magnitude higher than that of 11B (11.5 E24 cm2), enabling the visualization of 10B within leaf tissue. A preliminary test with NR revealed that only poplar, but none of the other plants accumulated sufficient 10B in their leaves for NR. Neutron radiographs of dried poplar leaves were taken at the ICON facility of the Paul-Scherrer-Institute (Villigen), Switzerland.27 The NR data were calibrated against ICP-OES measurements of leaf B concentrations. After neutron imaging, the leaves were scanned using an office scanner (Agfa, SnapScan 1236) at 150 dpi. Color images were analyzed using WinRhizoPro28 to assess the ratio between healthy and chlorotic or necrotic leaf area (Rh/cn) for each leaf. Chemical Analysis. For chemical analysis, aliquots of dried and ground plant samples were digested in a heating block at 130 C in 15 mL of a 65% HNO3. The digests were analyzed for B and other elements by ICP-OES (Vista MPX, Varian, Australia). Samples of PM substrate were analyzed for B after nitric acid digestion in the same way. Certified plant reference material NCS DC-73350 (poplar leaves, China National Analysis Centre for Iron and Steel, Beijing, China) was used for quality control. The average recovery rate for B was 98.4 ( 2%. To determine extractable concentrations of B and other elements in the PM substrate, 1:10 mixtures of substrate and 0.01 mol CaCl2 were shaken for 16 h, centrifuged at 929.3 g for 10 min, filtered through a 0.25-μm membrane filter and analyzed by ICP-OES. Carbon and nitrogen contents of the PM substrate were measured using an elemental analyzer (CNS-2000, Leco Corp., Saint Joseph, Michigan U.S.). Statistics. Mean whole-plant element concentrations were calculated as mass-weighted average of the respective element concentrations of individual plant parts. KruskalWallis-ANOVA was performed to test for differences in biomass and element 10539

dx.doi.org/10.1021/es201100b |Environ. Sci. Technol. 2011, 45, 10538–10543

Environmental Science & Technology

ARTICLE

contaminated sites. However, both Populus and Salix exhibit considerable inter- and intraspecific genetic and phenotypic variability with respect to B uptake and tolerance.15,29 Therefore, other Populus and Salix species and genotypes may have different B tolerance characteristics. Figure 1 shows that the relative decrease in the biomass of the poplar plants was larger in the roots than in leaves and stems in the T168 and T280 treatments. The shoot:root biomass ratio increased from 6 in the control treatment to 25 and 57 in the T168 and the T280 treatments, respectively. The fact that high soil B concentrations had a stronger negative eﬀect on root than on shoot biomass in P. nigra euramericana indicates a higher B sensitivity of the roots or a mode of biological protection to absorb less B. High concentrations of soil B are known to inhibit root growth relative to shoot growth.30 Reduced growth may be a general response of poplar roots toward contaminants as poplar roots were shown to react in the same way toward elevated soil Zn and Cd concentrations.31 Boron Accumulation and Allocation in the Plants. While in the control treatment shoot B concentrations did not differ among species, significant differences emerged at higher B treatment concentrations (Table 1). The bioconcentration factors (BCF) (plant/soil concentration quotients) ranged between 3.5 and 5 for all species and all treatments, except for B. juncea (BC: 1.52.7) in the B treatments. The highest BCF values were found for poplar in the T168 and T280. Brassica juncea was found to exclude B from entering its shoots. Shoot B concentrations in this species did not differ between T13, T22, and T45 and were still less than half of the surviving L. albus plants in the T93 treatment. The B concentrations found in B. juncea were in the same range as those reported by Ba~nuelos et al..32 If the B tolerance of P. nigra euramericana was due to B exclusion from uptake by the roots, then we would expect nontolerant plants to have higher shoot B concentrations than B-tolerant poplars grown on the same substrate. We did not ﬁnd such a relationship between the plant species used in this study. The ability of the poplars to accumulate higher concentrations of B than the other species was apparently due to a greater B tolerance in their leaf tissues, demonstrating that this characteristic can be a useful strategy to deal with elevated soil B concentrations. The phloem mobility of B in L. albus did not increase its B tolerance in comparison to P. nigra euramericana, L. albus, and S. viminalis. Also, the lower B accumulation in B. juncea did not increase its B tolerance compared to the other species and was less successful under the conditions of our study. These results are consistent with ﬁndings that B can easily penetrate cell membranes, indicating that regulation of B entry

concentrations between B treatments, followed by the Mann Whitney U-Test as posthoc test to compare pairwise differences between treatments. Values given for correlations between variables represent Pearsons’ correlation coefficients. All statistical analyses were carried out using PASW Statistics (Release 17.0.2).

’ RESULTS AND DISCUSSION Biomass. All poplar saplings survived even at the highest B treatment levels, although they showed reduced growth in T168 and severe growth reduction in T280. Our results are consistent with the high B tolerance reported by Robinson et al.13 for poplars growing on B contaminated sites. Figure S1 (SI) shows the aboveground biomass of the harvested plants, excluding the part of the stem axis corresponding to the cutting originally planted in the case of P. nigra euramericana and S. viminalis. L. albus and B. juncea plants survived in the T93 treatment without any reduction in growth, but failed to grow at higher B concentrations. S. viminalis only grew in the T13 and the T22 treatment and its biomass was significantly lower than that of P. nigra euramericana in T13 and that of B. juncea in T13 and T22. Thus, S. viminalis was the least B tolerant of the four species tested, while poplar was the most tolerant. This was a surprising observation given that poplars and willows belong to the same family (Salicaceae). Plants that do not tolerate elevated soil B concentrations are obviously not suited to remediate B

Figure 1. Leaf, stem, and root biomass of 4 month old P. nigra euramericana saplings grown on substrates with diﬀerent B concentrations. The lowest B concentration (13 mg kg1) is the control treatment. The mass of the cutting from which the saplings were grown is not included. Error bars represent standard errors (N = 3).

Table 1. B Accumulation (Mean ( S. E.) in the Aboveground Biomass of L. albus, B. juncea, P. nigra euramericana and S. viminalis Grown on Substrate with Diﬀerent B Concentrationsa,b B concentration L. albus

treatment

P. nigra euramericana

B. juncea

S. viminalis

[mg kg1] T13

a

40.5

(3.44

x yI

T22

114.2

T45

174.6 yz I

T93

304.4 z I

43.5

x

(4.69

43.8 x

(16.6

60.1

x II

(4.37

(27.2

68.1 xy II

(20.7

136.4 y II

(0.29

48.6 x

(4.67

N/A

118.3 y I

(11.3

(17.2

N/A

z

(19.1

392.4 y I

(28.7

z

b

T13 is the control treatment. Statistically signiﬁcant diﬀerences between treatments are indicated by characters and diﬀerences between plant species within the same treatment by roman numerals (Mann-Whitney U-test, p < 0.05, N = 3). N/A: not applicable. c plant died. 10540

dx.doi.org/10.1021/es201100b |Environ. Sci. Technol. 2011, 45, 10538–10543

Environmental Science & Technology

ARTICLE

Figure 2. Concentrations of B in roots, stems and leaves of 4 months old P. nigra euramericana plants. The lowest B concentration (13 mg kg1) is the control treatment. Note that the B concentration is shown on logarithmic scale for better clarity. Error bars represent standard errors (N = 3).

Figure 3. Leaf B concentration as a function of leaf position, counting from bottom to top along the stems of 4 months old poplars grown on substrate with diﬀerent B concentrations. Note that the B concentration is shown on logarithmic scale for better clarity.

into the symplast and further into the root xylem, by means of membrane transporters is ineﬀective.33 Unlike other nutrient elements, B is taken up by plants as the neutral species H3BO3 which is dominant in soil solution at pH <9.24.33 This species has a diameter of only 0.257 nm and thus may easily pass through cell membranes via aquaporins.34 Figure 2 shows that there were no signiﬁcant diﬀerences between root and stem B concentrations, which both increased in the poplar plants with the B concentration of the substrate. In the T168 and T280 treatments, the average leaf B concentration exceeded 1000 mg kg1. This is in agreement with the notion that B is primarily passively transported with the transpiration stream and deposited in the leaves upon evaporation of the water and is consistent with previous reports.10,13 Compared to the other tested species, P. nigra euramericana has good potential for the phytomanagement of B contaminated sites. The total uptake of B into the aboveground biomass of P. nigra euramericana during 4 months was 1 mg per plant in T13 and 8 mg per plant in T93, which represented about 2.1% of the total B initially present in the pots in T93. In the T168 treatment, the total uptake of B was 7.2 mg per plant. In T168, the higher plant B concentration compensated the lower plant biomass in comparison to T93. However, in T168 the 7.2 mg B extracted were only 1% of the total B in the pot. This uptake was higher than found in Gypsophila arrostil and in the same range as reported for Pucinella distans, two species considered as potential B hyperaccumulator plants.35 The highest uptake found for one of the other species tested in this study was 0.7 mg B per plant in B. juncea. With an estimated annual leaf biomass production of 15 t ha1 a1 P. nigra euramericana could extract 6.3 kg B ha1 a1 from contaminated topsoil containing 75 kg B ha1. To prevent the extracted B from returning to the soil via leaf fall, removal of the leaves from the site would be necessary. For that purpose, poplars could be coppiced.13 The B rich leaves could be used as an organic fertilizer on B deﬁcient sites or used as stock fodder.36 Only leaves from T13 and T93 would be suitable as stock fodder, as B concentrations >800 mg kg1 may be toxic to livestock.37 Leaves from the T168 and T280 treatment could still be used as fodder if mixed with fodder produced on unpolluted soil. Partitioning of B in Populus nigra euramericana Leaves. In all treatments, B concentrations decreased exponentially with leaf number from the lower (older) to the upper (younger) leaves of the poplar saplings (Figure 3). There was a more than

10-fold difference in average B concentration between the oldest and the youngest leaves in all B treatments. The B concentration ranges from top to bottom leaves were 22185 (T13), 621725 (T93), 1903241 (T168), and 2983472 (T280) for the respective treatments, with only small differences between the highest treatments T168 and T280. These results have implications for the interpretation of data for B accumulation in poplar trees sampled in the field.18 It is usually only possible to collect and analyze a small number of leaves from a tree. As our results show, B concentration data from leaf samples may vary by an order of magnitude depending on the position of the sampled leaves. Robinson et al.10 found that leaf B concentrations also varied considerably with time over a growing season. Again, these findings are support that B accumulation in the leaves is primarily associated with the transpiration water flow and that there is little or no relocation of B in the phloem of poplars. The leaf B concentrations did not depend on the size of the leaves (data not shown). The leaves emerging in the middle of the growing season were larger than the leaves produced at the beginning and the end of the growing season, while the B concentration of the leaves that emerged in the middle of the growing season steadily increased with age. With increasing leaf B concentrations the fraction of chlorotic and necrotic areas on the sampled leaves increased (Figure 4). At leaf B concentrations <900 mg kg1 Rh/cn was always <10%. The leaf B concentration range 9001199 mg kg1 was a threshold across which Rh/cn jumped to values above 30%. At leaf B concentrations >1200 mg kg1 the value of Rh/cn increased linearly (r2 = 0.98; y = 4.07x + 27.21; p < 0.001), until a second threshold was reached at B concentrations >2100 mg kg1, where Rh/cn increased to >70%. Tripler et al.38 found similar leaf necrosis eﬀects associated with high leaf B concentrations in date palm. Increasing contaminant accumulation and leaf chlorosis/ necrosis with leaf age is also known for Zn and Cd, although these metals were stored in diﬀerent tissues.39,40 Distribution of B within Populus nigra euramericana Leaves. Comparison of the ICP-OES measurements and the NR results showed that local tissue 10B accumulation in leaves was detectable by NR if concentrations in leaves exceeded 300 mg kg1. The detection limit and the spatial resolution of neutron radiographs (130 μm) thus were sufficient for the determination of toxicity thresholds in P. nigra euramericana leaf tissue. Boron concentrations in the leaves of B. juncea, S. viminalis, and L. albus 10541

dx.doi.org/10.1021/es201100b |Environ. Sci. Technol. 2011, 45, 10538–10543

Environmental Science & Technology

ARTICLE

poplars seems not to be active and they do not fulﬁll the criterion that hyperaccumulators should have at least 100-fold higher concentrations of the respective trace element than nonhyperaccumulators when grown in contaminated soil.43 This indicates that B hyperaccumulation in poplars is not hyperaccumulation in the strictest sense, but rather B hypertolerance and thus comparable to the passive arsenic hyperaccumulation in aquatic macrophytes described by Robinson et al.45 Our results indicate that poplar is better suited for phytomanagement of B contaminated soil than S. viminalis or B. juncea, which have been proposed for the phytoextraction of other trace elements.

’ ASSOCIATED CONTENT Figure 4. Chlorotic and necrotic leaf area expressed as percentage of total leaf area (Rh/cn) as a function of leaf B concentration. Note the large increase in chlorotic and necrotic leaf area above 900 mg B kg1. Error bars represent standard errors.

were below the detection limit. Here, laser ablation ICP-MS could be an alternative.20 Within individual leaves, the highest B concentrations occurred at the leaf margins and tips. The margins and tips were also the locations where chlorosis and necrosis occurred ﬁrst and were strongest (Figure S2 & S3 (SI)). At average leaf B concentrations greater than 1000 mg kg1, the leaf margins and tips curled. At higher total leaf B concentrations, necrotic spots occurred throughout the leaf. These spots contained >2000 mg B kg1. Leaf tissue containing between 1000 and 2000 mg B kg1 was chlorotic and tissue containing more than 2000 mg kg1 was necrotic. The ﬁnding of B concentrations >7000 mg kg1 in some spots in necrotic leaf tissue indicates that B accumulation continued in leaf tissue even after the onset of necrosis and that necrotic tissue can still receive B via the transpiration ﬂow. Similar ﬁndings were reported by Reid and Fitzpatrick19 for barley. Deposition of B at high concentrations in discrete patches may be a tolerance mechanism by which a small patch of photosynthetic tissue is sacriﬁced in order to prevent overloading of the surrounding tissues. The ability of P. nigra euramericana to accumulate higher B concentrations in its aerial tissue than the other species tested can be attributed to the high B tolerance of the living leaf tissue and the storage of B in dead leaf tissue. The B accumulation characteristics of P. nigra euramericana are consistent with the criteria compiled by Branquinho et al.41 for hyperaccumulation. The BCF as well as the shoot to root concentration ratio were >1 in P. nigra euramericana and the aboveground B concentration in two (T168 and T280) of three B treatments was more than 10-times higher than in the control (T13). In contrast to many metals,42 there is no established shoot threshold B concentration above which a plant is considered to be a B hyperaccumulator. For example, for Ni the threshold concentration used as criterion for hyperaccumulation is 1000 mg kg1,43 which corresponds to 17.0 mmol kg1. The equivalent mass concentration of B is just 172 mg kg1 because of its 80% lower molar weight compared to Ni. This concentration was exceeded in some of the poplar leaves grown in the control treatment and in more than 85% of the leaves in the treatments with higher B concentrations. In addition, the accumulation of 1000 mg B kg1, a 20-times higher tissue concentration than the 50 mg kg1 that is generally considered to be toxic in tissues of most other plants, is an indicator of B hyperaccumulation in poplar.44 However, as the comparison with other species showed, B accumulation in

bS

Supporting Information. Details on the growing substrate and plant biomass and pictures of poplar leaves showing the pattern of necrosis. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +41 44 633 60 78; E-mail: [email protected]. Present Addresses ^

Kroto Research Institute, University of Sheﬃeld, Sheﬃeld S3 7HQ, United Kingdom.

’ ACKNOWLEDGMENT Funds for this study came from the Swiss National Science Foundation (SNSF). We would also like to thank Rene Saladin from the Soil Protection lab at ETH, Lidija Josic from PSI for help with the NR analysis and providing the 10B cross section data, and Anton Burkart and his team at WSL for the cuttings and tending the plants. ’ REFERENCES (1) Salisbury, F. B.; Ross, C. W. Plant Physiology, 4th ed.; Wadsworth Publishing Co., Inc.: Belmont, California, USA., 1992. (2) Hunt, C. D. Dietary Boron: Evidence for Essentiality and Homeostatic Control in Humans and Animals. In Advances in Plant and Animal Boron Nutrition; Springer: New York, 2007; p 251. (3) Shorrocks, V. M. The occurrence and correction of boron deﬁciency. Plant and Soil 1997, 193 (12), 121–148. (4) Goldberg, S. Reactions of boron with soils. Plant Soil 1997, 193 (12), 35–48. (5) Schulin, R., et al. Trace element deﬁcient soils. In Trace Elements in Soils; Hooda, P., Ed.; Wiley-Blackwell Publishing: Chichester, U.K, 2010; p 175. (6) Takano, J.; Noguchi, K.; Yasumori, M.; Kobayashi, M.; Gajdos, Z.; Miwa, K.; Hayashi, H.; Yoneyama, T.; Fujiwara, T. Arabidopsis boron transporter for xylem loading. Nature 2002, 420 (6913), 337–340. (7) Nable, R. O.; Banuelos, G. S.; Paull, J. G. Boron toxicity. Plant Soil 1997, 193 (12), 181–198. (8) Parks, J. L.; Edwards, M. Boron in the environment. Crit. Rev. Environ. Sci. Technol. 2005, 35 (2), 81–114. (9) Hathaway, R. Short-rotation coppiced willows for sheep fodder in New Zealand. New Zealand Agric. Sci. 1986, 20 (3), 140–142. (10) Robinson, B. H.; Green, S. R.; Chancerel, B.; Mills, T. M.; Clothier, B. E. Poplar for the phytomanagement of boron contaminated sites. Environ. Pollut. 2007, 150 (2), 225–233. (11) Mastromatteo, E.; Sullivan, F. Summary-International Symposium on the health-eﬀects of boron and its compounds. Environ. Health Perspect. 1994, 102, 139–141. 10542

dx.doi.org/10.1021/es201100b |Environ. Sci. Technol. 2011, 45, 10538–10543

Environmental Science & Technology (12) Ba~nuelos, G. S.; Shannon, M. C.; Ajwa, H.; Draper, J. H.; Jordahl, J.; Licht, J. Phytoextraction and accumulation of boron and selenium by Poplar (Populus) hybrid clones. Int. J. Phytoremediation 1999, 1 (1), 8186. DOI 10.1080/15226519908500006 (13) Robinson, B.; Green, S.; Mills, T.; Clothier, B.; Velde, M. v. d.; Laplane, R.; Fung, L.; Deurer, M.; Hurst, S.; Thayalakumaran, T.; Dijssel, C. v. d. Phytoremediation: Using plants as biopumps to improve degraded environments. Aust. J. Soil Res. 2003, 41 (3), 599–511. (14) Dellantonio, A.; Fitz, W. J.; Repmann, F.; Wenzel, W. W. Disposal of Coal Combustion Residues in Terrestrial Systems: Contamination and Risk Management. J. Environ. Qual. 2010, 39 (3), 761–775, DOI: 10.2134/jeq2009.0068. (15) Dellantonio, A.; Fitz, W. J.; Custovic, H.; Repmann, F.; Schneider, B. U.; Grunewald, H.; Gruber, V.; Zgorelec, Z.; Zerem, N.; Carter, C.; Markovic, M.; Puschenreiter, M.; Wenzel, W. W. Environmental risks of farmed and barren alkaline coal ash landﬁlls in Tuzla, Bosnia, and Herzegovina. Environ. Pollut. 2008, 153 (3), 677–686. (16) Pulford, I. D.; Watson, C. Phytoremediation of heavy metalcontaminated land by treesA review. Environ. Int. 2003, 29, 529–540. (17) Ba~nuelos, G. S.; Cardon, G. E.; Phene, C. J.; Wu, L.; Akohoue, S.; Zambrzuski, S. Soil boron and selenium removal by three plant species. Plant Soil 1993, 148 (2), 253–263. (18) Oertli, J. J. Nonhomogeneity of boron distribution in plants and consequences for foliar diagnosis. Commun. Soil Sci. Plant Anal. 1994, 25 (78), 1133–1147. (19) Reid, R.; Fitzpatrick, K. Inﬂuence of leaf tolerance mechanisms and rain on boron toxicity in barley and wheat. Plant Physiol. 2009, 151 (1), 413–420. (20) Wu, B.; Zoriy, M.; Chen, Y.; Becker, J. S. Imaging of nutrient elements in the leaves of Elsholtzia splendens by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS). Talanta 2009, 78 (1), 132–137. (21) Loria, L. G.; Jimenez, R.; Badilla, M.; Lhuissier, F.; Goldbach, H.; Thellier, M. Neutron-capture-radiography study of the distribution of boron in the leaves of coﬀee plants grown in the ﬁeld. J. Trace Microprobe Tech. 1999, 17 (1), 91–99. (22) Dickinson, N. M.; Pulford, I. D. Cadmium phytoextraction using short-rotation coppice Salix: The evidence trail. Environ. Int. 2005, 31 (4), 609–613. (23) Vamerali, T.; Bandiera, M.; Mosca, G. Field crops for phytoremediation of metal-contaminated land. A review. Environ. Chem. Lett. 2010, 8 (1), 1–17. (24) Huang, L.; Bell, R. W.; Dell, B. Evidence of phloem boron transport in response to interrupted boron supply in white lupin (Lupinus albus L. cv. Kiev Mutant) at the reproductive stage. J. Exp. Botany 2008, 59 (3), 575–583, DOI: 10.1093/jxb/erm336. (25) Zawisky, M.; Basturk, M.; Derntl, R.; Dubus, F.; Lehmann, E.; Vontobel, P. Non-destructive B-10 analysis in neutron transmission experiments. Appl. Radiat. Isot. 2004, 61 (4), 517–523. (26) Menon, M.; Robinson, B.; Oswald, S. E.; Kaestner, A.; Abbaspour, K. C.; Lehmann, E.; Schulin, R. Visualization of root growth in heterogeneously contaminated soil using neutron radiography. Eur. J. Soil Sci. 2007, 58 (3), 802–810. (27) Kuhne, G.; Frei, G.; Lehmann, E.; Vontobel, P. CNR—The new beamline for cold neutron imaging at the Swiss spallation neutron source SINQ. Nucl. Instrum. Methods Phys. Res. Sect. A-Accel. Spectrom. Dect. Assoc. Equip. 2005, 542 (13), 264–270. (28) Regent Instruments: WinRhizo Pro, c; 2009. (29) Ba~nuelos, G. S.; LeDuc, D.; Johnson, J. Evaluating the Tolerance of Young Hybrid Poplar Trees to Recycled Waters High in Salinity and Boron. Int. J. Phytoremediation 2010, 12 (5), 419–439. (30) Reid, R. J.; Hayes, J. E.; Post, A.; Stangoulis, J. C. R.; Graham, R. D. A critical analysis of the causes of boron toxicity in plants. Plant, Cell, Environ. 2004, 27 (11), 1405–1414. (31) Dos Santos Utmazian, M. N.; Wieshammer, G.; Vega, R.; Wenzel, W. W. Hydroponic screening for metal resistance and accumulation of cadmium and zinc in twenty clones of willows and poplars. Environ. Pollut. 2007, 148 (1), 155165. DOI: 10.1016/j.envpol.2006.10.045

ARTICLE

(32) Ba~ nuelos, G. S.; Cardon, G. E.; Mackey, B.; Benasher, J.; Wu, L.; Beuselinck, P.; Akohoue, S.; Zambruzski, S. Boron and selenium removal in boron-laden soils by 4 sprinkler irrigated plant-species. J. Environ. Qual. 1993, 22 (4), 786–792. (33) Hu, H.; Brown, P. H. Absorption of boron by plant roots. Plant Soil 1997, 193 (1), 49–58. (34) Tanaka, M.; Fujiwara, T. Physiological roles and transport mechanisms of boron: perspectives from plants. Pﬂugers Arch.Eur. J. Physiol. 2008, 456 (4), 671–677. (35) Stiles, A. R.; Bautista, D.; Atalay, E.; Babaoglu, M.; Terry, N. Mechanisms of boron tolerance and accumulation in plants: A physiological comparison of the extremely boron-tolerant plant species, Puccinellia distans, with the moderately boron-tolerant Gypsophila arrostil. Environ. Sci. Technol. 2010, 44 (18), 7089–7095. (36) Robinson, B.; Mills, T.; Green, S.; Chancerel, B.; Clothier, B.; Fung, L.; Hurst, S.; McIvor, I. Trace element accumulation by poplars and willows used for stock fodder. N. Z. J. Agric. Res. 2005, 48 (4), 489–497. (37) Underwood, E. J.; Suttle, N. F. The Mineral Nutrition of Livestock; CAB International Publishing: Wallingford, UK, 1999. (38) Tripler, E.; Ben-Gal, A.; Shani, U. Consequence of salinity and excess boron on growth, evapotranspiration and ion uptake in date palm (Phoenix dactylifera L., cv. Medjool). Plant Soil 2007, 297 (12), 147–155. (39) Vollenweider, P.; Menard, T.; G€unthardt-Goerg, M. S. Compartmentation of metals in foliage of Populus tremula grown on soils with mixed contamination. I. From the tree crown to leaf cell level. Environ. Pollut. 2011, 159 (1), 324–336. (40) Vollenweider, P.; Cosio, C.; G€unthardt-Goerg, M. S.; Keller, C. Localization and eﬀects of cadmium in leaves of a cadmium-tolerant willow (Salix viminalis L.): Part II Microlocalization and cellular eﬀects of cadmium. Environ. Exp. Botany 2006, 58 (13), 25–40. (41) Branquinho, C.; Serrano, H. C.; Pinto, M. J.; Martins-Louc-~ao, M. A. Revisiting the plant hyperaccumulation criteria to rare plants and earth abundant elements. Environ. Pollut. 2007, 146 (2), 437–443, DOI: 10.1016/j.envpol.2006.06.034. (42) Kramer, U., Metal Hyperaccumulation in Plants. In Annual Review of Plant Biology, Vol 61, Annual Reviews: Palo Alto, 2010; Vol. 61, p 517. (43) Brooks, R. R.; Lee, J.; Reeves, R. D.; Jaﬀre, T. Detection of nickeliferous rocks by analysis of herbarium specimens of indicator plants. J. Geochem. Explor. 1977, 7, 49–57. (44) Jones, J. B. J., Plant tissue analysis in micronutrients. In Micronutrients in Agriculture; 2nd ed.; Mortvedt, J. J., Cox, F. R., Shuman, L. M., Welch, R. M., Eds; SSSA: Madison, WI, U.S., 1991; p 477. (45) Robinson, B.; Kim, N.; Marchetti, M.; Moni, C.; Schroeter, L.; van den Dijssel, C.; Milne, G.; Clothier, B. Arsenic hyperaccumulation by aquatic macrophytes in the Taupo Volcanic Zone, New Zealand. Environ. Exp. Botany 2006, 58 (13), 206–215, DOI: 10.1016/j. envexpbot.2005.08.004.

10543

dx.doi.org/10.1021/es201100b |Environ. Sci. Technol. 2011, 45, 10538–10543

ARTICLE pubs.acs.org/est

Zinc Stabilization Efficiency of Aluminate Spinel Structure and its Leaching Behavior Yuanyuan Tang, Kaimin Shih,* Yanchun Wang, and Tak-Chai Chong Department of Civil Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong SAR, China

bS Supporting Information ABSTRACT: The feasibility of immobilizing zinc in contaminated soil was investigated by observing the role of zinc reacting with aluminum-rich materials under thermal conditions. To observe the process of zinc incorporation, mixtures of ZnO with alumina precursors (γ-Al2O3 and α-Al2O3) were ﬁred at 7501450 °C. Both precursors crystallochemically incorporated zinc into the ZnAl2O4 spinel structure. The incorporation eﬃciencies of a 3 h sintering scheme were ﬁrst quantitatively determined by Rietveld reﬁnement analysis of X-ray diﬀraction data. Diﬀerent zinc incorporation behavior by these two precursors was revealed, although both resulted in nearly 100% transformation at the highest temperature. Diﬀerent product microstructures and thermal densiﬁcation eﬀects were found by observing the sintered products from these two precursors. The leaching performances of ZnO and ZnAl2O4 were compared by a prolonged acid leaching test for 22 d. The leachability analysis pointed to superiority of the ZnAl2O4 structure in stabilizing zinc, suggesting a promising technique for incorporating zinc into the aluminum-rich product. Finally, the sludge collected from water treatment works was calcined and used as an aluminum-rich material to test its ability to stabilize zinc. Successful formation of ZnAl2O4 indicated good potential for employing waterworks sludge to thermally immobilize hazardous metals as a promising waste-to-resource strategy.

’ INTRODUCTION Metal pollution is becoming an increasingly signiﬁcant concern due to rapid industrialization in many regions of the world. The release of large quantities of hazardous metals into the natural environment has resulted in various environmental problems. Hazardous metals, such as zinc, copper, and chromium, are not biodegradable and can accumulate in nature, causing various diseases and disorders when exceeding speciﬁc limits.1,2 The spread of such hazardous metals from mining operations or by the use of agrochemicals, for example,3 can contaminate the soil of surrounding areas.4,5 Traditional environmental remediation techniques (excavation and landﬁlling) are expensive, and only limited resources have been allocated to remediation of contaminated sites; thus, alternative techniques are sought.6 Many investigators have attempted to immobilize hazardous metals in contaminated soil by adding amendments that are able to adsorb, complex, or (co)precipitate elements in the soil.68 Zinc in soil can be immobilized by precipitation with hydroxides, carbonates, phosphates, and sulﬁdes and by forming complexes with organic ligands.6,9,10 However, as a rather mobile element, zinc is easily out-competed by other cations for adsorption.11 Because the mobility of metals is lowest when the soil is near neutral to slightly alkaline,6 the immobilization eﬀect will be reduced when the soil pH becomes more acidic. It was demonstrated in our previous work that after the addition of r 2011 American Chemical Society

aluminum-rich materials into hazardous metal waste, the metals can be stabilized by spinel structure formation through wellcontrolled thermal treatment schemes.1215 By thermally reacting with alumina, hematite, and kaolinite precursors, the nickel and copper in the spinel-type crystalline structure were found to have substantial reduction in their leachability under acidic environments. Formation of zinc aluminate spinel (ZnAl2O4) from alumina (Al2O3) and zinc oxide (ZnO) has been reported in many high-temperature equilibrium experiments,1618 and has also been found in thermally treated sediments.19 However, a reliable and applicable metal stabilization strategy requires quantitative understanding of the metal incorporation eﬃciency at a shorter processing time and more achievable temperatures, the metal leachability and leaching behavior of product, and the inﬂuence of diﬀerent types of precursors. As a waste-to-resource technology, the use of waste sludge resulting from water and wastewater treatment processes has attracted much attention.20,21 In water treatment, for example, all processing systems generate a substantial amount of sludge with the residues of treatment chemicals used as coagulants (commonly aluminum-based), and the use of aluminum-rich Received: May 16, 2011 Accepted: November 9, 2011 Revised: September 10, 2011 Published: November 09, 2011 10544

dx.doi.org/10.1021/es201660t | Environ. Sci. Technol. 2011, 45, 10544–10550

Environmental Science & Technology

ARTICLE

sludge as raw material for ceramic products has shown promise.22 Therefore, in this study, we attempted to quantify the reaction eﬃciency between zinc oxide and two types of alumina precursors within a sintering time of 3 h to assist in the exploration of diﬀerent metal incorporation processes. This was followed by a prolonged leaching experiment to examine the zinc stabilization eﬀect and the leaching behavior of the sintered products. Finally, the water treatment sludge rich in aluminum was reused as a ceramic raw material so that we could evaluate its potential for eﬀectively stabilizing hazardous metals during the sintering process.

’ EXPERIMENTAL METHODS To study the zinc incorporation mechanism, experiments were conducted with samples prepared by ﬁring ZnO separately with two alumina precursors (γ-Al2O3 and α-Al2O3). A sintering scheme with a 3 h dwelling time23 at the targeted temperature was used for temperatures ranging from 750 to 1450 °C. The ZnO powder was purchased from Sigma Aldrich, and the surface area was measured by the BET method, yielding 2.91 ( 0.06 m2/g. Qualitative phase identiﬁcation of HiQ-7223 alumina was conﬁrmed by X-ray diﬀraction (XRD) as boehmite (AlOOH), which converts to the γ-Al2O3 phase upon thermal treatment at 650 °C for 3 h.24 The α-Al2O3 (corundum) used as the other zinc stabilization precursor was obtained by further calcining the as-formed γ-Al2O3 at 1500 °C for 6 h (Figure S1 of the Supporting Information (SI)). To evaluate capacity for incorporating zinc under the sintering environment, a water treatment (waterworks) sludge sample collected in Hong Kong was heated for use as an aluminum-rich ceramic precursor. The sludge was dried and ﬁred at 900 °C for 30 min to remove the organic content and then ground into powder for elemental composition analysis by X-ray ﬂuorescence spectroscopy (XRF) (JEOL JSX3201Z). Samples for all sintering experiments were prepared by mixing each precursor with ZnO powder for a total dry weight of 60 g at a Zn:Al molar ratio of 1:2. The mixing process was carried out by ball milling the powder in water slurry for 18 h. The slurry samples were dried and homogenized by mortar grinding, pressed into 20 mm pellets at 650 MPa, and then ﬁred. After sintering, the pellets were quenched in air and six replicates of pellet diameter measurements were carried out with Vernier calipers. Fired pellets were monitored by scanning electron microscopy (SEM) to observe the development of product microstructures, and then ground into randomly oriented powders for XRD analysis. SEM was performed on a Hitachi S-4800 SEM system equipped with a secondary electron detector for morphologic observation and a backscattered electron detector for energy dispersive spectroscopy (EDS). Prior to SEM, all pellet samples were polished with submicrometer diamond lapping ﬁlms and gold coated to mitigate the electron charging eﬀect. The secondary electron images were used to observe the microstructure of samples, and the backscattered electron images were used to identify compositionally distinct areas. Point- and line-mode EDS and mapping analyses were also carried out. Phase transformation during sintering was monitored using the powder XRD technique. The step-scanned XRD pattern of each powder sample was recorded by a Bruker D8 Advance X-ray powder diﬀractometer equipped with Cu Kα1,2 X-ray radiation source (40 kV 40 mA) and a LynxEye detector. The 2θ scanning range was 10130°, and the step size was 0.02° with a scan speed

Figure 1. XRD patterns of sintering (a) ZnO + γ-Al2O3, at 750 to 1450 °C for 3 h and (b) ZnO + corundum at 7501450 °C for 3 h. The standard patterns retrieved from the ICDD database include ZnO (PDF#36-1451), ZnAl2O4 (PDF#05-0669), and corundum (α-Al2O3; PDF#10-0173).

of 0.3 s/step. Qualitative phase identiﬁcation was executed by matching powder XRD patterns with those retrieved from the standard powder diﬀraction database of the International Centre for Diﬀraction Data (ICDD PDF-2, Release 2008). Zinccontaining crystalline phases found in products were ZnO (PDF#361451) and ZnAl2O4 (PDF#050669). Within γ-Al2O3 and corundum precursor systems, the identiﬁed crystalline phases were subjected to quantitative analysis to evaluate the eﬃciency of zinc incorporation into the ZnAl2O4 structure. XRD data were analyzed by means of the FullProf Suite (February 2007 version) for Rietveld reﬁnement. With reﬁnement of the diﬀraction patterns, the weights of all crystalline phases in samples were determined and expressed as percentages. The reﬁnement was evaluated by the derived factors as the criteria of ﬁt provided in SI Table S1. To assess the validity of the reﬁnement procedure in quantifying the crystalline phases used in the study, powder mixtures containing authentic ZnO and ZnAl2O4 phases in known weight fractions of 20:80, 40:60, 60:40, and 80:20 were tested, with satisfactory results (SI Table S2). 10545

dx.doi.org/10.1021/es201660t |Environ. Sci. Technol. 2011, 45, 10544–10550

Environmental Science & Technology

ARTICLE

Figure 2. The transformation ratio (TR, %) of zinc into the ZnAl2O4 spinel structure by γ-Al2O3 and corundum precursors. The TR values derived from the γ-Al2O3 precursor are much higher than those derived from the corundum precursor at low temperatures. The incorporation of zinc by both precursors reached 100% by the end of sintering processes.

The leachability of single-phase ZnO and ZnAl2O4 samples was tested by means of a leaching experiment modiﬁed from the U.S. EPA Toxicity Characteristic Leaching Procedure (TCLP), with a pH 2.9 acetic acid solution (extraction ﬂuid #2) as the leaching ﬂuid. Each leaching vial was ﬁlled with 10 mL of TCLP extraction ﬂuid and 0.5 g of powder. The leaching vials were rotated end-over-end at 60 rpm for agitation periods of 0.75 to 22 d. At the end of each agitation period, the leachates were ﬁltered with 0.2-μm syringe ﬁlters, the pH was measured, and the concentrations of all metals were derived from an Optima 3300DV inductively coupled plasma atomic emission spectrometer (PerkinElmer). The sample surface areas were determined from the nitrogen adsorptiondesorption isotherms at liquid nitrogen temperature (77K) on a Beckman Coulter SA3100 surface area and pore size analyzer by BrunauerEmmett Teller (BET) method.

’ RESULTS AND DISCUSSION ZnAl2O4 Formation from Alumina Precursors. Figure 1 collates the XRD patterns of the sintered samples within the temperature range of 7501450 °C, and ZnAl2O4 is indicated as the predominant product phase after sintering. Therefore, the potential reactions between ZnO and the two alumina precursors (γ-Al2O3 and α-Al2O3) are expressed as follows:

ZnO þ γ-Al2 O3 f ZnAl2 O4

ð1Þ

ZnO þ α-Al2 O3 f ZnAl2 O4

ð2Þ

When γ-Al2O3 was used as the precursor, the ZnAl2O4 spinel phase was clearly initiated with sintered at 750 °C for 3 h. However, when corundum was used as the precursor, the ZnAl2O4 spinel was first detected in the sintering product when the temperature reached 950 °C. Once the sintering temperatures were raised above those corresponding to the first appearance of the spinel phase, continuing growth of ZnAl2O4 was observed with increased peak intensity. The signals of reactants eventually diminished after 3 h of sintering at 1450 and 1250 °C for the ZnO + γ-Al2O3 and ZnO + α-Al2O3 systems, respectively.

Figure 3. Even after extensive polishing, the secondary electron micrographs show (a) a highly dense microstructure sintered from the ZnO + γ-Al2O3 mixture at 1350 °C for 3 h, and (b) a very porous texture of the product of the ZnO + corundum mixture sintered under the same conditions.

To further quantify the variation of zinc incorporation efficiency in both precursor systems at different sintering temperatures, the distribution of crystalline phases in the products can be expressed by a “transformation ratio (TR, %)” index defined as follows: wt% of ZnAl2 O4 MW of ZnAl2 O4 TRð%Þ ¼ wt% of ZnAl2 O4 wt% of ZnO þ MW of ZnO MW of ZnAl2 O4 where MW = molecular weight. For TR = 100%, complete transformation of zinc into ZnAl2O4 spinel structure occurred. Figure 2 summarizes the TR values of zinc incorporated into ZnAl2O4 phase when ZnO + γ-Al2O3 and ZnO + α-Al2O3 samples were sintered over the temperature range of 750 1450 °C. The reaction of zinc with γ-Al2O3 precursor was strong and even reached over 60% transformation at the lowest sintering temperature (750 °C). The zinc transformation continued to increase with the elevation in sintering temperature and reached nearly full incorporation after sintering at 1450 °C. In contrast, no ZnAl2O4 was observed when sintering ZnO with corundum precursor at temperatures below 850 °C. Nevertheless, when the sintering temperature reached 1250 °C, complete transformation of zinc into ZnAl2O4 structure (TR = 100%) occurred. 10546

dx.doi.org/10.1021/es201660t |Environ. Sci. Technol. 2011, 45, 10544–10550

Environmental Science & Technology

ARTICLE

Figure 4. Variation in pellet diameters produced by sintering ZnO with γ-Al2O3 and corundum precursors at temperature ranging from 750 to 1450 °C for 3 h. Diameters of the sintered ZnO + γ-Al2O3 pellets decreased when sintering temperatures increased, whereas the sintered ZnO + corundum pellets increased in diameter with increases in temperature.

The crossover of the two curves at about 1100 °C suggests that the ZnO and γ-Al2O3 interaction was dominant at lower temperatures, whereas the corundum precursor may greatly facilitate zinc incorporation at higher temperatures. The driving forces for mass transfer during sintering are the diﬀerences in chemical potential (free enthalpy or molar Gibbs energy) between ﬁred materials.25 The value of ΔHfo (standard enthalpy of formation) for γ-Al2O3 is +18.8 kJ/mol higher than that of corundum,26 and thus the reaction between ZnO and γ-Al2O3 is more energetically favored in comparison to that between ZnO and α-Al2O3. This may explain the greater zinc incorporation eﬃciency by γ-Al2O3 at lower temperatures. Influence of Precursors on Products Microstructure. The product microstructure plays a vital role in affecting the mechanical properties and determining the potential product applications. The different types of alumina precursors were found to generate spinel-containing products with distinct microstructures. Part a of Figure 3 shows the product microstructure resulting from the mixture of ZnO and γ-Al2O3 sintered at 1350 °C for 3 h. On the electron scanning micrograph, the ZnAl2O4 product grains appear tightly associated with each other. The product sintered under the same conditions but with the corundum precursor appears to have finer spinel grains and contains numerous voids (part b of Figure 3). Because greater zinc incorporation efficiency was observed with the use of corundum precursor at higher temperatures (see Figure 2), the product microstructure suggests enhanced incorporation owing to the additional interfacial diffusion in ZnO + α-Al2O3 samples in comparison to the lattice diffusion dominant in sintering ZnO + γ-Al2O3 samples. Although the change in product pellet size can be inﬂuenced by factors such as impurities, powder particle size, and sintering conditions, understanding the basic eﬀect on pellet size caused by the zinc incorporation reaction is important. The overall densiﬁcation or thermal expansion will reﬂect an increase or decrease in pellet sizes. Figure 4 demonstrates the pellet size variation after 3 h sintering, showing the diﬀerent thermal eﬀects when γ-Al2O3 and corundum were used as precursors. Interaction

Figure 5. Leaching performance of ZnO and ZnAl2O4 phases demonstrated by (a) pH values of both leachates, (b) the normalized zinc concentrations in both leachates, and (c) the [Al]/[Zn] molar ratios in ZnAl2O4 leachates.

between ZnO and γ-Al2O3 led to a decrease in pellet diameters after sintering. Shrinkage due to the thermal densiﬁcation of samples continued until the end of the sintering process, consistent with the increased zinc transformation into ZnAl2O4 phase shown in Figure 2. However, when corundum was used, the pellet diameters increased with increases in temperature through the same sintering process as with γ-Al2O3 as the precursor. The most signiﬁcant expansion appeared at the temperature range of 9501150 °C, and the pellet reached its largest size after complete zinc transformation was achieved, as 10547

dx.doi.org/10.1021/es201660t |Environ. Sci. Technol. 2011, 45, 10544–10550

Environmental Science & Technology shown in Figure 2. In porous compacts, such as powder-pressed pellets, mass transport from grain boundaries to pores leads to densiﬁcation.27 As a more eﬃcient transport mechanism during sintering, grain boundary diﬀusion is determined by its intrinsic mobility and drag forces from pores.28 The pore drag is substantial at high porosity, and it decreases during densiﬁcation. If the drag force falls below the required minimum, grain boundaries and pores become separated, and intragranular pore entrapment occurs. Because the surface area of γ-Al2O3 (204.8 m2/g) is substantially greater than that of corundum (1.645 m2/g), the larger pore drag forces in the γ-Al2O3 precursor system may lead to the stronger densiﬁcation eﬀect after sintering. Leaching Performance of ZnO and ZnAl2O4. To investigate the effect of zinc immobilization after incorporation by the spinel structure, the inherent leachability of the two zinc-containing phases, i.e. ZnO and ZnAl2O4, was evaluated through the prolonged TCLP leaching procedure. The as-received ZnO powder was used for the leaching experiment. To produce a powder sample with ZnAl2O4 as the single phase, a mixture of ZnO and γ-Al2O3 powders was prepared at a Zn/Al molar ratio equal to 1:2, pelletized, and sintered at 1350 °C for 48 h before being ground into powder form. The extended sintering time was to further ensure a complete reaction and homogeneity of the spinel formation. The XRD pattern of this synthesized ZnAl2O4 powder sample is shown in SI Figure S2. It illustrates the successful achievement of ZnAl2O4 phase without any peaks of reactant phases (ZnO or Al2O3) observed in the sample. The BET surface area of the ZnAl2O4 powder was 0.830 ( 0.022 m2/g. Throughout the 22 day leaching experiment, the pH values of ZnO and ZnAl2O4 leachates were monitored, as shown in part a of Figure 5, and the greater pH increase of ZnO leachate was revealed. Within the ﬁrst 18 h, the pH of ZnO leachate increased substantially from 2.9 to 6.3, and then it remained at approximately 6.5 throughout the rest of the leaching period. In contrast, the pH of the ZnAl2O4 leachate remained at its initial value throughout the entire leaching period. The increase in the leachate pH may be due to the dissolution of cations through ion exchange with protons in the solution, accompanied by the destruction of crystals at the solid surface by the acidic leaching ﬂuid. The results indicate that ZnAl2O4 is of higher inherent resistance to such acidic attack and that ZnO may be much more vulnerable to proton-mediated dissolution. Because the leaching of a solid is likely dominated by surface reactions, such reactions are expected to be proportional to the sample surface area. In addition, because samples of the same weight (0.5 g) were always used, the total zinc content in each sample, subject to the diﬀerent zinc phases, can be normalized for comparison. Part b of Figure 5 summarizes the amounts of leached zinc normalized with respect to the surface areas and the total zinc content of tested samples. After such normalization, the amount of leached zinc in the ZnO leachate was about 3 orders of magnitude greater than that in the ZnAl2O4. In comparison to ZnO, the ZnAl2O4 spinel demonstrated much higher inherent resistance to acidic attack, and thus the spinel incorporation strategy proved to be beneﬁcial in stabilizing zinc. The inset of part b of Figure 5 further provides the details of the normalized zinc content in the ZnAl2O4 leachate. When the pH of ZnO leachate reached about 6.5, the zinc concentration in the leachate stabilized at about 3000 mg/L (101.3 M). As a general assumption of the cation-proton exchange mechanism, the destruction of zinc oxide by the acidic

ARTICLE

attack of the solution can be expressed as follows: 2þ ZnOðsÞ þ 2Hþ ðeqÞ f ZnðeqÞ þ H2 O

ð3Þ

However, the concentration of zinc ions in the solution [Zn2+(aq)] is limited by the potential precipitation/dissolution reactions, such as in respect to Zn(OH)2(s): ZnðOHÞ2ðsÞ T Zn2þ ðeqÞ þ 2OHðeqÞ

ð4Þ

where the solubility constant (Ksp) of eq 4 is 1016.35.29 At pH 6.5, the product of [Zn2+(aq)] [OH(aq)]2 was found to be 1016.3, which is very close to the Ksp of Zn(OH)2(s). This indicates that the system was in equilibrium with Zn(OH)2(s), and explains why the zinc concentration stabilized at about 3000 mg/L after the substantial increase in the leachate pH during the ﬁrst 18 h of leaching. When leaching the ZnAl2O4 phase, a “congruent dissolution” through the cation-proton exchange reaction can be written as follows: 2þ 3þ ZnAl2 O4ðsÞ þ 8Hþ ðeqÞ f ZnðeqÞ þ 2AlðeqÞ þ 4H2 O

ð5Þ

Such congruent dissolution would usually result in a theoretical [Al3+(aq)]/[Zn2+(aq)] molar ratio of 2.0 in the leachates. However, this ratio was observed at 0.600.95 in the leachate of ZnAl2O4, as shown in part c of Figure 5. Because the system was maintained in a more acidic environment (about pH 2.9) and the zinc concentration was much lower than that in the ZnO leachate, the zinc concentrations in the leachates of ZnAl2O4 should be all considerably under-saturated with respect to the Zn(OH)2(s). Furthermore, the maximum aluminum concentrations measured in the ZnAl2O4 leachates were about 0.91 mg/L (104.47 M), and the reaction of amorphous aluminum hydroxide (am 3 Al(OH)3(s)) precipitation/dissolution is expressed as follows: am 3 AlðOHÞ3ðsÞ T Al3þ ðeqÞ þ 3OHðeqÞ

ð6Þ

where the solubility constant (Ksp) of eq 6 is 1032.7.29 The product of [Al3+(aq)] [OH(aq)]3 was found to be 1037.8 in the leaching system, and it did not reach the saturation ([Al3+(aq)] [OH(aq)]3 = 1032.7) of amorphous Al(OH)3(s). Therefore, neither zinc nor aluminum ions were subject to reprecipitation from the leachates, and the observed [Al3+(aq)]/ [Zn 2+ (aq)] ratio may indicate incongruent dissolution of ZnAl2O4 in the leaching experiment, where the majority of the Al—O bonds still remained on the ZnAl2O4 spinel surface. Although some previous studies 30,31 also suggested the possibility of surface reorganization of remaining molecules during the incongruent dissolution process, the overall result suggests the accumulation of aluminum-rich substance(s) on the surface of leached ZnAl2O4, which may also be beneﬁcial for preventing further leaching of Zn from the spinel product. Potential of Using Waterworks Sludge for Zinc Stabilization. Water treatment works in Hong Kong use aluminum-based coagulants (aluminum sulfate, aluminum chloride, poly aluminum sulfate, etc.) to facilitate particulate sedimentation. Because aluminum-rich ceramic raw materials (γ-Al2O3 and corundum) have shown high efficiency in the incorporation of zinc into spinel structure, the use of waterworks sludge as a precursor material to facilitate stabilization of zinc under thermal conditions may be a promising waste-to-resource strategy. 10548

dx.doi.org/10.1021/es201660t |Environ. Sci. Technol. 2011, 45, 10544–10550

Environmental Science & Technology

ARTICLE

ZnO are not observable in the XRD patterns, whereas the ZnAl2O4 spinel is the predominant Zn-containing phase observed in the sample. Virtually all known drinking water processing systems generate enormous amounts of residual sludge, and what to do with this rapidly increasing waste stream in a beneﬁcial and environmentally sustainable manner remains an important issue.21 In this study, the successful incorporation of zinc into the ZnAl2O4 structure by waterworks sludge at an attainable sintering temperature indicates that such sludge may be economically reused as an aluminum-rich material to immobilize hazardous metals in soils under thermal conditions. The wasteto-resource strategy of reusing waterworks sludge material, together with its role in facilitating natural resource remediation, may provide a strategy to further protect public health and enhance environmental sustainability.

’ ASSOCIATED CONTENT

bS

Supporting Information. Two tables and six ﬁgures demonstrating the powder XRD patterns of precursor materials, selected sintered products, and Rietveld reﬁnement. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +852-28591973; fax: +852-25595337; e-mail: kshih@ hku.hk.

’ ACKNOWLEDGMENT This work was supported ﬁnancially by the General Research Fund Scheme of the Research Grants Council of Hong Kong (HKU 716310E). Contribution of HiQ-7223 alumina by the Alcoa Corporation is gratefully acknowledged. Figure 6. XRD patterns of waterworks sludge sintered at (a) 900 °C for 30 min and ZnO + sludge mixture sintered at (b) 1150 to 1350 °C for 3 h. The standard patterns retrieved from the ICDD database include quartz (SiO2, PDF#47-1144), mullite (3Al2O3 3 2SiO2; PDF#79-1455), ZnAl2O4 (PDF#05-0669), and cristobalite (SiO2; PDF#76-0938).

The weight losses of dry sludge heated at diﬀerent temperatures are shown in SI Figure S3. The weight loss continued until the temperature reached 600 °C, when the maximum weight loss was around 47%. The sludge was further heated at 900 °C for 30 min so that the elemental compositions could be obtained via XRF. Normalization into metal oxides (SI Figure S4) shows aluminum to be the predominant constituent, and the XRD pattern (part a of Figure 6) further indicates that the aluminum component may exist as mullite and poor-crystalline phase(s) in the calcined sludge. A strong quartz signal was also found, which may reﬂect the second-largest component (silicon) detected by XRF. At temperatures above 1150 °C, signiﬁcant zinc incorporation was observed by alumina precursors, and therefore a sintering temperature range of 11501350 °C was used to investigate the incorporation of zinc into the sludge precursor. A mixture of ZnO and 900 °C calcined waterworks sludge at a Zn:Al molar ratio of 1:2 was prepared for the 3-h sintering scheme, and the XRD patterns show the success of incorporating zinc into the ZnAl2O4 spinel structure (part b of Figure 6). The diﬀraction peaks of

’ REFERENCES (1) Gode, F.; Pehlivan, E. A comparative study of two chelating ion exchange resins for the removal of chromium(III) from aqueous solution. J. Hazard. Mater. 2003, B100, 231–243. (2) Veli, S.; Aly€uz, B. Adsorption of copper and zinc from aqueous solutions by using natural clay. J. Hazard. Mater. 2007, 149, 226–233. (3) Lado, L. R.; Hengl, T.; Reuter, H. I. Heavy metals in European soils: A geostatistical analysis of the FOREGS Geochemical database. Geoderma 2008, 148, 189–199. (4) Friesl, W.; Friedl, J.; Platzer, K.; Horak, O.; Gerzabek, M. H. Remediation of contaminated agricultural soils near a former Pb/Zn smelter in Austria: Batch, pot and ﬁeld experiments. Environ. Pollut. 2006, 144, 40–50. (5) Pusz, A. Inﬂuence of brown coal on limit of phytotoxicity of soils contaminated with heavy metals. J. Hazard. Mater. 2007, 149, 590–597. (6) Kumpiene, J.; Lagerkvist, A.; Maurice, C. Stabilization of As, Cr, Cu, Pb, and Zn in soil using amendments—A review. Waste Manag. 2008, 28, 215–225. (7) Janos, P.; Vavrova, J.; Herzogova, L.; Pilarova, V. Eﬀects of inorganic and organic amendments on the mobility (leachability) of heavy metals in contaminated soil: A sequential extraction study. Geoderma 2010, 159, 335–341. (8) Mulligan, C. N.; Yong, R. N.; Gibbs, B. F. Remediation technologies for metal-contaminated soils and groundwater: An evaluation. Eng. Geol. 2001, 60, 193–207. (9) Brown, S.; Chaney, R.; Hallfrisch, J.; Ryan, J. A.; Berti, W. R. In situ soil treatments to reduce the phyto- and bioavailability of lead, zinc and cadmium. J. Environ. Qual. 2004, 33, 522–531. 10549

dx.doi.org/10.1021/es201660t |Environ. Sci. Technol. 2011, 45, 10544–10550

Environmental Science & Technology

ARTICLE

(10) Hamon, R. E.; McLaughlin, M. J.; Cozens, G. Mechanisms of attenuation of metal availability in in situ remediation treatments. Environ. Sci. Technol. 2002, 36, 3991–3996. (11) Cao, X.; Ma, L. Q.; Rhue, D. R.; Appel, C. S. Mechanisms of lead, copper, and zinc retention by phosphate rock. Environ. Pollut. 2004, 131, 435–444. (12) Shih, K.; White, T.; Leckie, J. O. Spinel formation for stabilizing simulated nickel-laden sludge with aluminum-rich ceramic precursors. Environ. Sci. Technol. 2006, 40, 5077–5083. (13) Shih, K.; White, T.; Leckie, J. O. Nickel stabilization eﬃciency of aluminate and ferrite spinels and their leaching behavior. Environ. Sci. Technol. 2006, 40, 5520–5526. (14) Tang, Y.; Shih, K.; Chan, K. Copper aluminate spinel in the stabilization and detoxiﬁcation of simulated copper-laden sludge. Chemosphere 2010, 80, 375–380. (15) Shih, K.; Tang, Y. Prolonged toxicity characteristic leaching procedure for nickel and copper aluminates. J. Environ. Monitor. 2011, 13, 829–835. (16) Ballarini, A. D.; Bocanegra, S. A.; Castro, A. A.; de Miguel, S. R.; Scelza, O. A. Characterization of ZnAl2O4 obtained by diﬀerent methods and used as catalytic support of Pt. Catal. Lett. 2009, 129, 293–302. (17) Bunting, E. N. Phase equilibria in the system SiO2-ZnO-Al2O3. Bur. Stand. J. Res. 1932, 8, 279–287. (18) Hansson, R.; Hayes, P. C.; Jak, E. Experimental study of phase equilibria in the Al-Fe-Zn-O system in air. Metall. Mater. Trans. B 2004, 35B, 633–642. (19) Ndiba, P. K.; Axe, L. Sequential extraction of phosphate- and thermal-treated New York/New Jersey Harbor dredged sediments. Environ. Eng. Sci. 2009, 26, 1755–1764. (20) Merrington, G.; Oliver, I.; Smernik, R. J.; McLaughlin, M. J. The inﬂuence of sewage sludge properties on sludge-borne metal availability. Adv. Environ. Res. 2003, 8, 21–36. (21) Babatunde, A. O.; Zhao., Y. Q. Constructive approaches toward water treatment works sludge management: An international review of beneﬁcial reuses. Crit. Rev. Env. Sci. Technol. 2007, 37, 129–164. (22) Vicenzi, J.; Moura Bernardes, A.; Perez Bergmann, C. Evaluation of alum sludge as raw material for ceramic products. J. Ind. Ceram. 2005, 25, 171–180. (23) Sun, D. D.; Tay, J. H.; Cheong, H. K.; Leung, D. L. K.; Qian, G. R. Recovery of heavy metals and stabilization of spent hydrotreating catalyst using a glass-ceramic matrix. J. Hazard. Mater. 2001, B87, 213–223. (24) Zhou, R. S.; Snyder, R. L. Structures and transformation mechanisms of the η, γ, and θ transition aluminas. Acta Crystallogr. 1991, B47, 617–630. € (25) Sarıkaya, Y.; Ada, K.; Onal, M. A model for initial-stage sintering thermodynamics of an alumina powder. Powder Technol. 2008, 188, 9–12. (26) McHale, J. M.; Navrotsky, A. Eﬀects of increased surface area and chemisorbed H2O on the relative stability of nanocrystalline γ-Al2O3 and α-Al2O3. J. Phys. Chem. B 1997, 101, 603–613. (27) Yan, M. F. Eﬀect of grain size distribution on sintered density. Mater. Sci. Eng. 1983, 60, 275–281. (28) Yan, M. F.; Cannon, R. M.; Bowen, H. K. Grain boundary migration in ceramics. In Ceramic Microstructures; Fulrath, R. M., Pask, J. A., Eds.; Westview Press: Boulder, CO, 1977; pp 276307. (29) Stumm, W., Morgan, J. J. Aquatic Chemistry, 3rd, ed.; Wiley Interscience: New York, 1996. (30) Cailleteau, C.; Angeli, F.; Devreux, F.; Gin, S.; Jestin, J.; Jollivet, P.; Spalla, O. Insight into silicate-glass corrosion mechanisms. Nat. Mater. 2008, 7, 978–983. (31) Ohlin, C. A.; Villa, E. M.; Rustad, J. R.; Casey, W. H. Dissolution of insulating oxide materials at the molecular scale. Nat. Mater. 2010, 9, 11–19.

10550

dx.doi.org/10.1021/es201660t |Environ. Sci. Technol. 2011, 45, 10544–10550

ARTICLE pubs.acs.org/est

Recovery of Pure ZnO Nanoparticles from Spent Zn-MnO2 Alkaline Batteries Akash Deep,*,† Kamal Kumar,‡ Parveen Kumar,† Pawan Kumar,† Amit L Sharma,† Bina Gupta,§ and Lalit M Bharadwaj† †

Biomolecular Electronics and Nanotechnology Division, Central Scientiﬁc Instruments Organization (CSIR-CSIO), Sector 30C, Chandigarh 160030, India ‡ Department of Physics, Gurukula Kangri University, Haridwar 249404, India § Department of Chemistry, Indian Institute of Technology Roorkee, Roorkee 247667, India

bS Supporting Information ABSTRACT: The recovery of pure ZnO (zinc oxide) nanoparticles from spent ZnMn dry alkaline batteries is reported. Spent batteries were dismantled to separate the contained valuable metals of the cell electrodes in the form of black powder. Treatment of this black powder with 5 mol L1 HCl produced leach liquor, primarily containing 2.90 g L1 Zn and 2.02 g L1 Mn. Selective and quantitative liquidliquid extraction of Zn(II) was then carried out in three counter current steps by using Cyanex 923 (0.10 mol L1 in n-hexane). Zn(II) distributed in the organic phase as complex ZnCl2 3 2R (R = Cyanex 923 molecule). The metal loaded organic phase was subjected to combust at 600 °C to yield pure ZnO nanoparticles (4050 nm). Important characteristics of the synthesized nanoparticles were investigated by ﬁeld emission scanning electron microscopy (FESEM), energy dispersive X-ray spectroscopy (EDX), X-ray diﬀraction spectroscopy (XRD), and atomic force microscopy (AFM).

’ INTRODUCTION Zincmanganese oxide dry alkaline batteries are frequently used to operate many electronic and electrical appliances. Recycling of spent Zn-MnO2 batteries is very important to minimize the risk of environmental pollution. However, recycling with an unadorned purpose of waste treatment is not an attractive business, particularly in developing countries where economic interests supersede environmental obligations. In this scenario, the idea of recovering a valuable product (e.g., pure ZnO nanoparticles) from spent batteries may be useful to promote their recycling due to the projected economic beneﬁts. Several pyrometallurgical and hydrometallurgical processes have been suggested to recover valuable metal components from diﬀerent kinds of spent batteries, including NiCd, 13 Ni metal hydride, 24 and lithium ion. 3,58 The recycling of ZnMn- and ZnC-based batteries has also been a topic of research interest. 9 The recovery of contained valuable metals from spent batteries is generally carried out by ammoniacal and acidic leaching1012 processes. Precipitation1214 and thermal15,16 treatments have been cited to yield reusable oxides or ferrites. Most of the abovereported processes do not necessarily catch the fancy due to limited applicability and low commercial value of the end product. r 2011 American Chemical Society

The liquidliquid extraction of metals from spent batteries58 is a simple technique to recover high purity products. Using this technique, some researchers have proposed the separation of valuable metals from spent ZnMn17,18 batteries with the application of phosphonic and phosphinic acid extractants, namely Cyanex 272 and Cyanex 301. In these processes, the metal loaded organic phase needs to be treated with HCl solution in order to back-extract the desired metals. In an earlier published study,19 we identiﬁed Cyanex 923 (a mixture of four tri alkyl phosphine oxides) as a very useful reagent for the extraction of 3d transition metals. Cyanex 923 proved to be a better choice over Cyanex 272 and Cyanex 301 as the studied extractant oﬀered faster extraction kinetics, better loading capacity and improved phase separations. Moreover, Cyanex 923 does not solidify at low ambient temperatures and is free from foul smell. The present work demonstrates the use of Cyanex 923 for the extraction of zinc from the waste black powder of exhausted ZnMnO2 dry alkaline batteries. After extraction and a subsequent washing step, the pure zinc loaded organic phase has been Received: May 23, 2011 Accepted: November 3, 2011 Revised: October 10, 2011 Published: November 03, 2011 10551

dx.doi.org/10.1021/es201744t | Environ. Sci. Technol. 2011, 45, 10551–10556

Environmental Science & Technology

ARTICLE

Table 1. Composition of Digest and Leach Liquors Processed from the Black Powder “S1” of Spent Batteriesa concentration in digested

concentration in leaching

%

metal

sample g L1

sample g L1

leaching

Zn Mn

3.05 ( 0.05 7.05 ( 0.10

2.93 ( 0.04 2.03 ( 0.04

96.1 28.8

Ni

(3.1 ( 0.05) 103

(3.07 ( 0.05) 103

g99

(50.5 ( 1) 103

g99

(2.47 ( 0.02) 103

g99

Fe Cd

(51 ( 1) 10

3

(2.5 ( 0.02) 103

Sample concentration =20 g L1, dissolving medium = aqua regia for digestion, 5 mol L1 HCl for leaching. a

Figure 2. Extraction of metals as a function of initial aqueous phase acidity. [Cyanex 923] = 0.10 mol L1; [HCl] = 0.50, 0.60, 0.70, 0.80, 0.90, 1.0, 1.1, 1.2 mol L1; A/O ratio =1; t = 5 min; T = 25 °C.

Figure 1. Extraction of metals from the leach liquor with varying concentrations of Cyanex 923 in n-hexane. [Cyanex 923] = 0.01, 0.05, 0.075, 0.10, 0.125, 0.15 mol L1; A/O ratio =1; t = 5 min; T = 25 °C.

subjected to combust at 600 °C for the synthesis of high purity ZnO nanoparticles. Some important structural characteristics of the synthesized nanoparticles have been ascertained by ﬁeld emission scanning electron microscopy (FESEM), energy dispersive X-ray spectroscopy (EDX), X-ray diﬀraction spectroscopy (XRD), atomic force microscopy (AFM), and ultraviolet visible (UVvis) spectrometry. The recovered pure ZnO nanoparticles may ﬁnd applications in piezoelectric transducers, gas sensors, photonic crystals, light-emitting devices, photodetectors, photodiodes, optical waveguides, transparent conductive ﬁlms, varistors, and solar cells.

’ EXPERIMENTAL PROCEDURES Materials and Equipment. Metal standards for the atomic absorption spectroscopy were purchased from Merck Chemicals, India. Other reagents, solvents, and titration indicator were A.R. grade materials from Fisher Scientific/Sigma Aldrich. Cyanex 923 was received from Cytec Canada Inc., Canada. This extractant was a 93% pure mixture of four trialkylphosphine oxides: R3PdO, R0 R2PdO, R2R0 PdO, and R0 3PdO, where R and R0 represent n-octyl and n-hexyl hydrocarbon chains, respectively. The concentrations of different metals in aqueous solutions were determined by Atomic Absorption Spectrometer (AAS, Perkin-Elmer, AAnalyst 200). Spectral and topographical features of the synthesized nanoparticles were studied by UVvis spectrometer (Varian, Cary 5000), field emission scanning electron microscope energy dispersive

X-ray spectroscope (FESEM-EDX, Hitachi, 4800 SE), X-ray diffractometer (XRD, Shimadzu, 6000), and atomic force microscope (AFM, Park Systems, XE-NSOM). Dismantling of Spent Batteries and Study of Elemental Composition. Spent Zn-MnO2 dry alkaline cells (AA size, 1.5 V, electrolyte-KOH) of a particular brand were collected from the local sources. A set of 20 spent batteries was manually dismantled. The desired electrode material (black powder) was carefully separated from the scrap paper, plastic film, outer metallic body, and membrane. The black powder thus collected was washed with water to remove entrained electrolyte. The moist sample was dried at 120 °C (24 h), followed by manual milling to obtain fine particles (S1). One g of this sample “S1” was mixed with 20 mL of aqua-regia, heated to boil for 1 h, cooled, and then appropriately diluted with double-distilled water. The above digested sample solution was assayed by AAS for the determination of metal contents. Leaching of Zinc from Black Powder. Ten g sample (S1) of the black powder was treated with 100 mL of 5 mol L1 HCl for 2 h at 70 °C. The contents were allowed to cool and then filtered. Residual mass was washed in four steps with a total of 100 mL of double distilled water. Final volume of the leach liquor was made up to 500 mL (L1). LiquidLiquid Extraction Studies. Equal volumes of the leach liquor (L1) and the extractant solution (Cyanex 923 in n-hexane) were equilibrated in separatory funnels for 5 min at 25 °C to attain optimum mass transfer. The emulsion was allowed to settle (2 min) for a clear phase separation followed by the collection of aqueous and organic fractions in separate vials. The concentration of metals in aqueous phase was determined by AAS, whereas the concentration of metals in organic phase was estimated by mass balance. The zinc loaded organic fraction was equilibrated with 2.0 mol L1 HCl in order to cleanse it from other coextracted metal impurities. Synthesis of ZnO Nanoparticles. Ten mL of the zinc loaded organic phase was taken in a platinum crucible and then thermally treated for 1 h at elevated temperature (300 °C, 400 °C, 500 °C, 600 °C). Thus synthesized nanoparticles were characterized by FESEM-EDX, XRD, AFM, and UVvis spectrometry, 10552

dx.doi.org/10.1021/es201744t |Environ. Sci. Technol. 2011, 45, 10551–10556

Environmental Science & Technology

ARTICLE

’ RESULTS AND DISCUSSION Composition of Spent Zn-MnO2 Dry Alkaline Batteries and Leaching Tests. AAS analysis (average of 3 measurements)

of the digested black powder solution (1 g “S1” in 50 mL of final volume) revealed the metallic contents in following concentrations: Zn: 3.05 ( 0.05 g L1, Mn: 7.05 ( 0.10 g L1, Ni: (3.1 ( 0.05) 103 g L1, Fe: (51 ( 1.0) 103 g L1, Cd: (2.5 ( 0.02) 103 g L1. Black powder samples from the different sets of spent batteries were studied in the above manner for establishing their average metallic composition as follows: 15 ( 0.3% Zn, 35 ( 0.5% Mn, 0.015 ( 0.005% Ni, 0.25 ( 0.03% Fe, 0.012 ( 0.005% Cd. Data on the leaching of metals from the black powder “S1” with 5 mol L1 HCl solution (solution ‘L1’) are given in Table 1. It can be estimated that the implemented leaching step has transferred almost 96% of the zinc content into the aqueous phase. Nickel, iron, and cadmium were almost completely dissolved, whereas the recovery of manganese was around 30%. The concentration of free HCl in the ﬁnal leach liquor was estimated by acidbase titration using bromophenol blue indicator; and found to be 0.82 ( 0.05 mol L1. Extraction of Metals from Leach Liquor of Spent Batteries by Cyanex 923. The leach liquor “L1” was equilibrated with varying concentrations (0.01, 0.05, 0.075, 0.10, 0.125, 0.15 mol L1) of Cyanex 923. As shown in Figure 1, the extraction of Zn(II) in Cyanex 923 was higher than that of Fe(III) and Cd(II). Around 87% extraction of Zn(II) was achieved in a single extraction step with 0.10 mol L1 Cyanex 923. Around 22% of Fe(III) and 25% of Cd(II) were coextracted with the above extractant solution. The extraction of Mn(II) and Ni(II) remained negligible (<1%) in all the investigated extractant concentrations. 0.10 mol L1

concentration of Cyanex 923 was preferred for all further liquidliquid extraction studies. This selected extractant concentration provided a reasonably high extraction of the desired metal ion and limited the coextraction of Fe(III) and Cd(II) to a tolerable level. Since the HCl content in the leach liquor could inﬂuence the metal distribution in Cyanex 923, the extraction patterns of Zn(II), Mn(II), Ni(II), Fe(III), and Cd(II) were further investigated with respect to the initial aqueous phase acidity (Figure 2). Synthetic solutions were used for this particular study, whose elemental compositions matched with the leach liquor but the acid content varied (0.501.2 mol L1). Better extractions of Zn(II), Fe(III) and Cd(II) were observed vis-a-vis the aqueous phase acidity. This pattern may be attributed to the fact that the mechanism of the metal distribution in Cyanex 923 is governed through the solvation of neutral Zn chloro species. At low HCl acidity (e.g., 0.600.90 mol L1), the predominating presence of anionic ZnCl species over the neutral ZnCl2 species accounted the realization of lower metal extractions. Higher HCl acidity (1.0 mol L1 or more) of the aqueous phase facilitated the predominance of neutral ZnCl2 species; and consequently, the Zn extraction improved to 90% and more. Similar extraction mechanism also governs the distribution of Fe(III) and Cd(II) in Cyanex 923, and therefore, the coextraction of Fe(III) and Cd(II) also increases with the increasing aqueous phase acidity. Extraction Equilibrium. The extraction of Zn(II) with Cyanex 923 (R) proceeds according to the following reaction: Zn2þ þ 2Cl þ nR w ZnCl2 3 nR

ð1Þ

The stoichiometric extraction constant Kex of above reaction can be represented as follows: Kex ¼

ZnCl2 3 nR ½Zn2þ ½Cl 2 ½Rn

ð2Þ

Deﬁning the term (ZnCl2 3 nR)/[Zn2+] as the distribution ratio D, the eq 2 can be rearranged as under the following: log D ¼ log Kext þ 2log Cl þ nlogR

Figure 3. Extraction isotherm of Zn from the leach liquor. [Cyanex 923] = 0.10 mol L1; A/O ratio =4, 3, 2, 1, 0.5, 0.33, 0.25; t = 5 min; T = 25 °C.

ð3Þ

A graphical illustration (log [R] vs log [D], Figure 1) of eq 3 has been used to estimate the value of n. On the basis of this obtained value (n = 2), the distribution of Zn in Cyanex 923 is proposed in the form of complex ZnCl2 3 2R. Further, solving eq 3 for diﬀerent extractant concentrations have estimated the value of conditional average extraction constant Kex as (1.92 ( 0.04) 104. The above obtained values of n and Kex have been deduced by taking the [Cl] as 0.82 mol L1 (the acidity of the herein used leaching solution). Since the selection of acid concentration for

Table 2. Extraction of Zn from the Leach Liquor “L1” by Using Cyanex 923

metal

estimated concentration in

concentration in organic

concentration

concentration in aqueous phase

organic phase after

phase after scrubbing with

% recovery of Zn in organic

in leach liquor (g L1)

after extraction (g L1)

extraction (g L1)

2.0 mol L1 HCl (g L1)

phase with purity g98% (recovery)

Zn

2.93 ( 0.04

0.04 ( 0.01

2.89

2.88 ( 0.02

Mn

2.03 ( 0.04

2.03 ( 0.05

<0.01

not detected

Ni

(3.07 ( 0.05) 103

(3.07 ( 0.05) 103

<0.01

not detected

(20.0 ( 0.50) 103

30.5 103

<0.01

(1.25 ( 0.02) 103

1.22 103

<0.01

Fe Cd

3

(50.5 ( 1) 10

(2.47 ( 0.02) 103

g99 (relative purity)

a Aqueous phase =200 mL of leach liquor, organic phase =200 mL of 0.10 mol L1 Cyanex 923, no. of extraction stages =3, impurity scrubbing a

solution =2.0 mol L1 HCl (A/O = 0.25).

10553

dx.doi.org/10.1021/es201744t |Environ. Sci. Technol. 2011, 45, 10551–10556

Environmental Science & Technology the leaching step is a conditional parameter, the signiﬁcance of the above calculated values of n and Kex depends upon their constancy for other [Cl]. Therefore, the values of n and Kex have also been computed for some other [Cl], e.g., 0.6, 0.8, 1.0, and 1.2 mol L1. Graphical representations of log [R] vs log D (Figure S-1, Supporting Information) and necessary computations have revealed the values of n (= 2 in each case) and Kex {Kex,0.6 M HCl = (1.93 ( 0.02) 104; Kex,0.8 M HCl = (1.97 ( 0.02) 104; Kex,1.0 M HCl = (1.97 ( 0.03) 104; Kex,1.2 M HCl =

Figure 4. Surface morphology (FESEM) and elemental composition (EDX) of the black powder from spent batteries.

ARTICLE

(1.96 ( 0.03) 104}. It is clear that the values of n and Kex remain more or less constant irrespective of the aqueous phase acidity. This observation proves the eﬀectiveness of Cyanex 923 for the recovery of Zn from the leach liquors of varying acidity. Extraction Isotherm. The loading capacity of Cyanex 923 was investigated by recording the distribution of Zn(II) at varying (4, 3, 2, 1, 0.5, 0.33, and 0.25) aqueous (leach liquor) to organic (0.10 mol L1 Cyanex 923) phase ratios. Graphical illustration (Figure 3) of the collected data is useful to estimate the required number of extraction stages for the complete extraction of Zn(II) from the leach liquor at any desired A/O ratio. The study has indicated that the complete extraction of Zn(II) at an A/O ratio of 1 requires three counter-current extraction steps. Recovery of Pure Zn(II) from Leaching Solution. 200 mL aliquot of the leach liquor “L1”, containing 2.93 g L1 Zn(II), 2.03 g L1 Mn(II), 3.07 103 g L1 Ni(II), 50.5 103 g L1 Fe(III), and 2.47 103 g L1 Cd(II) was equilibrated with 0.10 mol L1 Cyanex 923 in three stages of counter-current extraction (A/O = 1). The composition of the leach liquor before and after the extraction is shown in Table 2. The estimated composition (by mass balance) of the organic phase is also given. Zn(II) was almost completely transferred into the organic phase with some coextraction of Fe(III) and Cd(II). Mn(II) and Ni(II) remained in the aqueous raffinate. After phase separation, the zinc loaded organic phase was washed with 50 mL of 2.0 mol L1 HCl (four times) in order to scrub out the impurities and recover a pure Zn(II) loaded organic phase (O1). The relative recovery and purity of the extracted Zn(II) were confirmed by taking out a small (1.0 mL) sample from the “O1” and stripping the loaded metal content into an aqueous phase (1.0 mL of 2.0 mol L1

Figure 5. Surface morphology of ZnO nanoparticles synthesized by the combustion of Zn loaded Cyanex 923 phase at (a) 300 °C, (b) 400 °C, (c) 500 °C, and (d) 600 °C. 10554

dx.doi.org/10.1021/es201744t |Environ. Sci. Technol. 2011, 45, 10551–10556

Environmental Science & Technology

ARTICLE

Figure 6. Noncontact AFM topography and 3-dimensional NCM phase of ZnO6000C nanoparticles.

elemental (Figure S-2 of the Supporting Information) studies of the recovered products have indicated that the combustion at 300 and 400 °C delivers ZnO particles of more than 100 nm diameter. These products were also contaminated with phosphorus, which might have surfaced due to incomplete vaporization of Cyanex 923. Solution combustion at a higher temperature (500 and 600 °C) resulted into the formation of the ZnO nanoparticles. These nanoparticles were also free from any metallic or nonmetallic impurity. In overall, the formation of ZnO nanoparticles from the leach liquor can be represented by the following reactions. Zn2þ þ 2Cl þ nR w ZnCl2 3 nR Δ, in presence of oxygen

ZnCl2 3 nR s f ZnO þ Cl v þ R v Figure 7. XRD spectrum of ZnO6000C nanoparticles with interplanar spacing (experimental and JCPDS data card) and corresponding (hkl) values.

HNO3). The analyses have indicated that the above developed process for the extraction of Zn(II) from the leaching solution “L1” provides more than 98% recovery of the desired metal ion with a purity of more than 99%. The whole process is simple, fast and can be accomplished in commonly existing industrial mixersettlers or column extractors. Remaining metals in raffinate and scrubbing solutions can be recovered by standard cementation/ precipitation (e.g., Cd, Fe) or electrolytic deposition (e.g., Mn). The left-over solutions after the leaching, extraction and other involved processes can be reused in closed process loop to maintain water balance. All of the suggested leaching, extraction, and washing steps construct a standard hydrometallurgical engineering setup. Modern day engineering advancements in such systems are well capable of minimizing the waste and byproduct generation. Treatment of Loaded Organic Phase to Yield ZnO Nanoparticles. FESEM and EDX (Figures 4) investigations of the starting material (black powder of spent batteries) revealed macroscopic structures of Zn, Mn, Fe, Cd and Ni with no particular orientation or shape. Thermal treatment (300 °C, 400 °C, 500 °C, 600 °C) of the Zn(II) loaded Cyanex 923 phase “O1” has yielded ZnO particles. Morphological (Figure 5) and

ð4Þ ð5Þ

Combustion of “O1” at 600 °C has produced good quality ZnO nanoparticles (3550 nm). It may be pertinent to mention here that the treatment of vapors evolved during the combustion process may be required for the safety of the environment. The boiling points of the involved chemicals, i.e., n-hexane (69 °C) and Cyanex 923 (310 °C), are fairly distinct from the ﬁnal annealing temperature (600 °C). During practical operations, the whole setup can easily be designed to recycle the used reagents by introducing two fractional distillation steps before the ﬁnal annealing. A rough estimation of the process cost (Sheet 1 of Supporting Information) indicates that the described recovery of the ZnO nanoparticles is economically beneﬁcial. The use of recycled chemicals may further improve the process economics. AFM data of the synthesized ZnO nanoparticles are shown in Figure 6. True Non-Contact AFM surface topographic scanning conﬁrmed the particle size to be in the range of 4050 nm. Three dimensional view of the scan area also supports our claim of nanoparticle formation. Figure S-3 of Supporting Information shows the energy absorption characteristics of the ZnO6000C nanoparticles. Maximum absorption at 351 nm with a sharp onset represents the crystalline semiconducting ZnO nanostructure. XRD spectrum (Figure 7) of the synthesized nanoparticles is also characterized with well-deﬁned diﬀraction peaks for a hexagonal crystalline structure. Clearly, the product was free from impurities. 10555

dx.doi.org/10.1021/es201744t |Environ. Sci. Technol. 2011, 45, 10551–10556

Environmental Science & Technology

’ ASSOCIATED CONTENT

bS

Supporting Information. Further information on the Zn(II) extraction characteristics with respect to the varying aqueous phase acidity, EDX elemental data, UVvis absorption data of the synthesized ZnO nanoparticles, and the process cost estimation is presented in Figures S-1 to S-3 and Sheet 1. This material is available free of charge via the Internet at http://pubs. acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 91-172-2637311; Fax: 91-172-2657082; E-mail: [email protected].

’ ACKNOWLEDGMENT Authors are thankful to Dr. Pawan Kapur, Director, CSIRCSIO, Chandigarh, India for providing necessary infrastructure facilities. We thank Ms. Manpreet for helping in part of the experimental work.

ARTICLE

(13) Peng, C.-h.; Bai, B.-s.; Chen, Y.-f. Study on the preparation of Mn-Zn soft magnetic ferrite powders from waste Zn-Mn dry batteries. Waste Manage. 2008, 28 (2), 326–332. (14) Freitas, M. B. J. G.; Pegoretti, V. C.; Pietre, M. K. Recycling manganese from spent Zn-MnO2 primary batteries. J. Power Sources 2007, 164 (2), 947–952. (15) Xiao, L.; Zhou, T.; Meng, J. Hydrothermal synthesis of Mn-Zn ferrites from spent alkaline Zn-Mn batteries. Particuology 2009, 7 (6), 491–495. (16) Ferella, F.; De Michelis, I.; Veglio, F. Process for the recycling of alkaline and zinc-carbon spent batteries. J. Power Sources 2008, 183 (2), 805–811. (17) Salgado, A. L.; Veloso, A. M. O.; Pereira, D. D.; Gontijo, G. S.; Salum, A.; Mansur, M. B. Recovery of zinc and manganese from spent alkaline batteries by liquid-liquid extraction with Cyanex 272. J. Power Sources 2003, 115 (2), 367–373. (18) El-Nadi, Y. A.; Daoud, J. A.; Aly, H. F. Leaching and separation of zinc from the black paste of spent MnO2-Zn dry cell batteries. J. Hazard. Mater. 2007, 143 (12), 328–334. (19) Gupta, B.; Deep, A.; Malik, P.; Tandon, S. N. Extraction and separation of some 3d transition metal ions using Cyanex 923. Solvent Extr. Ion Exch. 2002, 20 (1), 81–96.

’ REFERENCES (1) Huang, K.; Li, J.; Xu, Z. Characterization and recycling of cadmium from waste nickel cadmium batteries. Waste Manage. 2010, 30 (11), 2292–2298. (2) Huang, K.; Li, J.; Xu, Z. Enhancement of the recycling of waste Ni-Cd and Ni-MH batteries by mechanical treatment. Waste Manage. 2011, 31 (6), 1292–1299. (3) Mantuano, D. P.; Dorella, G.; Elias, R. C. A.; Mansur, M. B. Analysis of a hydrometallurgical route to recover base metals from spent rechargeable batteries by liquid-liquid extraction with Cyanex 272. J. Power Sources 2006, 159 (2), 1510–1518. (4) Rodrigues, L. E. O. C.; Mansur, M. B. Hydrometallurgical separation of rare earth elements, cobalt and nickel from spent nickelmetal-hydride batteries. J. Power Sources 2010, 195 (11), 3735–3741. (5) Chen, L.; Tang, X.; Zhang, Y.; Li, L.; Zeng, Z.; Zhang, Y. Process for the recovery of cobalt oxalate from spent lithium-ion batteries. Hydrometallurgy 2011, 108 (12), 80–86. (6) Pranolo, Y.; Zhang, W.; Cheng, C. Y. Recovery of metals from spent lithium-ion battery leach solutions with a mixed solvent extractant system. Hydrometallurgy 2010, 102 (14), 37–42. (7) Kang, J.; Senanayake, G.; Sohn, J.; Shin, S. M. Recovery of cobalt sulfate from spent lithium ion batteries by reductive leaching and solvent extraction with Cyanex 272. Hydrometallurgy 2010, 100 (34), 168–171. (8) Zhao, J. M.; Shen, X. Y.; Deng, F. L.; Wang, F. C.; Wu, Y.; Liu, H. Z. Synergistic extraction and separation of valuable metals from waste cathodic material of lithium ion batteries using Cyanex272 and PC-88A. Sep. Purif. Technol. 2011, 78 (3), 345–351. (9) Sayilgan, E.; Kukrer, T.; Civelekoglu, G.; Ferella, F.; Akcil, A.; Veglio, F.; Kitis, M. A review of technologies for the recovery of metals from spent alkaline and zinc-carbon batteries. Hydrometallurgy 2009, 97 (34), 158–166. (10) Senanayake, G.; Shin, S. M.; Senaputra, A.; Winn, A.; Pugaev, D.; Avraamides, J.; Sohn, J. S.; Kim, D. J. Comparative leaching of spent zinc-manganese-carbon batteries using sulfur dioxide in ammoniacal and sulfuric acid solutions. Hydrometallurgy 2010, 105 (12), 36–41. (11) De Michelis, I.; Ferella, F.; Karakaya, E.; Beolchini, F.; Veglio, F. Recovery of zinc and manganese from alkaline and zinc-carbon spent batteries. J. Power Sources 2007, 172 (2), 975–983. (12) Sayilgan, E.; Kukrer, T.; Yigit, N. O.; Civelekoglu, G.; Kitis, M. Acidic leaching and precipitation of zinc and manganese from spent battery powders using various reductants. J. Hazard. Mater. 2010, 173 (13), 137–143. 10556

dx.doi.org/10.1021/es201744t |Environ. Sci. Technol. 2011, 45, 10551–10556

ARTICLE pubs.acs.org/est

Biocathodic Nitrous Oxide Removal in Bioelectrochemical Systems Joachim Desloover,† Sebastia Puig,‡ Bernardino Virdis,§ Peter Clauwaert,† Pascal Boeckx,|| Willy Verstraete,† and Nico Boon*,† †

Laboratory of Microbial Ecology and Technology (LabMET), Ghent University, Coupure Links 653, B-9000 Gent, Belgium Laboratory of Chemical and Environmental Engineering (LEQUIA-UdG), Institute of the Environment, University of Girona, Campus Montilivi s/n, Facultat de Ciencies, E-17071 Girona, Spain § The University of Queensland, Advanced Water Management Centre, Level 4, Gehrmann Building (60), Brisbane, QLD 4072, Australia Laboratory of Applied Physical Chemistry (ISOFYS), Ghent University, Coupure Links 653, 9000 Gent, Belgium

)

‡

bS Supporting Information ABSTRACT: Anthropogenic nitrous oxide (N2O) emissions represent up to 40% of the global N2O emission and are constantly increasing. Mitigation of these emissions is warranted since N2O is a strong greenhouse gas and important ozone-depleting compound. Until now, only physicochemical technologies have been applied to mitigate point sources of N2O, and no biological treatment technology has been developed so far. In this study, a bioelectrochemical system (BES) with an autotrophic denitrifying biocathode was considered for the removal of N2O. The high N2O removal rates obtained ranged between 0.76 and 1.83 kg N m3 net cathodic compartment (NCC) d1 and were proportional to the current production, resulting in cathodic coulombic eﬃciencies near 100%. Furthermore, our experiments suggested the active involvement of microorganisms as the catalyst for the reduction of N2O to N2, and the optimal cathode potential ranged from 200 to 0 mV vs standard hydrogen electrode (SHE) in order to obtain high conversion rates. Successful operation of the system for more than 115 days with N2O as the sole cathodic electron acceptor strongly indicated that N2O respiration yielded enough energy to maintain the biological process. To our knowledge, this study provides for the ﬁrst time proof of concept of biocathodic N2O removal at long-term without the need for high temperatures and expensive catalysts.

’ INTRODUCTION Mitigation of nitrous oxide (N2O) emissions is a necessity since it is an important greenhouse gas with a global warming potential of about 300 times CO21 and represents about 7.9% of the global greenhouse gas budget when expressed in CO2 equivalents (calculated from Denman et al.2). Next to its strong global warming potential, N2O release is currently also the single most important ozone-depleting substance and is expected to remain the largest throughout the 21st century.3 Anthropogenic N2O emissions are responsible for almost 40% of the total emission (calculated from Denman et al.2), rising at a pace of 0.20.3% year1.1 Besides the well-known diﬀusive anthropogenic sources like agriculture, point sources contribute signiﬁcantly to the global N2O emission,2 and thus need to be mitigated as well. Examples of such sources include production plants of chemicals such as nitric acid, adipic acid and caprolactam, stationary and mobile combustion of fossil fuels and biomass, and wastewater treatment plants performing biological nitrogen removal.49 For the treatment of these N2O emissions only physicochemical removal technologies are applied such as thermal decomposition and selective catalytic reduction, requiring high temperatures (5001000 °C) and expensive catalysts.10,11 r 2011 American Chemical Society

Bioelectrochemical systems (BES) with a denitrifying biocathode have been described previously and can provide a more sustainable alternative to physicochemical approaches, since N2O is an intermediate of the denitriﬁcation pathway.12,13 The latter comprises the stepwise reduction of oxidized nitrogen compounds nitrate and nitrite (NO3 and NO2) to nitric oxide (NO), nitrous oxide (N2O), and dinitrogen gas (N2), performed by denitrifying microorganisms.14 BES are electrochemical devices where the oxidation of an electron donor at the anode is coupled with the reduction of an electron acceptor at the cathode, using bacteria to catalyze one or both reactions.15 In a standard conﬁguration, anode and cathode compartments are separated by an ion exchange membrane. The concept of using a solid-state electrode to supply electrons for biological nitrate reduction to nitrite was reported for the ﬁrst time by Gregory and co-workers.16 However, it has been only in recent years that the cathodic bioelectrochemical denitriﬁcation has been coupled with a bioanode to supply electrons.12,13,1720 In particular, the reduction of N2O to N2 represents a respiratory process in its Received: June 16, 2011 Accepted: November 9, 2011 Revised: October 20, 2011 Published: November 09, 2011 10557

dx.doi.org/10.1021/es202047x | Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

own (reaction 1).14 Yet, this ﬁnal denitriﬁcation step has not been investigated in detail at a biocathode. N2 O þ 2e þ 2Hþ f N2 þ H2 O E°0 ¼ þ 1:36V vs SHEðG°0 ¼ 262kJ 3 mol1 Þ

ð1Þ

From a thermodynamic point of view, N2O should be a more favorable electron acceptor compared to the other oxidized nitrogen species of the denitriﬁcation pathway.21 However, not often the denitrifying microbial community is prone to take advantage of this since signiﬁcant amounts of N2O can be released in microbiological active environments.22 At present, a strong debate exists in regards to the biochemical energy conservation of this reaction.14,2224 Given the current lack of knowledge existing in regards to biocathodic N2O reduction, this study aimed at investigating whether N2O can be eﬀectively removed by a denitrifying biocathode. Therefore, the performance and eﬃciency of a N2O reducing biocathode was characterized in terms of its volumetric removal rate, current production and cathodic coulombic eﬃciency. Furthermore, the possible role of microorganisms present in the cathode was indicated by experiments at open circuit, ﬁxed cathodic potential and abiotic controls. Finally, the advantage of biocathodic N2O removal and the main challenges for further development are discussed.

’ MATERIAL AND METHODS BES construction. A 2-chambered BES was made of two polycarbonate frames (8.0 8.0 1.9 cm3) placed side by side. Anodic and cathodic compartments (0.121 L each) were filled with granular graphite (type 00514, diameter 1.55 mm, Mersen, Wemmel, Belgium). As a result, the net anodic (NAC) and net cathodic (NCC) compartment, was 0.060 L for each compartment. Contact to the external electrical circuit was guaranteed by placing two graphite rods (5 mm diameter, Morgan, Belgium) in intimate contact with the granular matrix. A cation exchange membrane (CEM; Ultrex CMI7000, Membranes International Inc., USA) was used to separate the anodic and cathodic compartments. The same BES was used during the entire experimental period except for the abiotic cathode controls for which an identical cell was used with similar anode operating conditions. Inoculum and Synthetic Medium. The anodic compartment was inoculated with anode effluent from an already active BES treating acetate present in our laboratories, whereas the cathodic compartment was inoculated with activated nitrifying denitrifying sludge (Ossemeersen municipal WWTP, Ghent, Belgium). The anodic and cathodic liquid streams consisted of an autoclaved and nitrogen-purged modified M9 medium with addition of trace elements as previously described.12 No nitrogen source was added in the M9 medium. Operational Conditions. The experimental study was divided into three main periods, comprising a 82-days continuous feeding period with nitrate as the electron acceptor at the cathode (day 081), followed by a batch feeding period of 70 days with N2O as the electron acceptor (day 82151) and finally a continuous N2O feeding period of 46 days (day 152197). The anode was always fed continuously with sodium acetate and M9 medium. A summary of the operational conditions is given in Supporting Information (SI) Table S.1 and a scheme of the reactor setup is presented in SI Figure S.5.

Continuous Nitrate Feeding Period. A continuous nitrate feeding period was maintained in order to develop a denitrifying biocathode. During the nitrate-feeding period (day 082), both anode and cathode compartments were operated continuously. In order to obtain the desired loading rate, concentrated solutions of sodium acetate (1.11 g L1) and sodium nitrate (2.07 g L1) were injected with a syringe pump (8.3 mL d1) in the influent stream (M9 medium; 0.9 L d1) of the anode and cathode compartment, respectively. The hydraulic retention time (HRT) for each compartment was 1.6 h (SI Table S.1). To minimize concentration polarization in the anode, the loading rate of the anode was always twice as much as the loading rate of the cathode on a coulombic basis, giving 1.19 kg COD m3 NAC d1 and 0.21 kg NO3-N m3 NCC d1 for the anode and cathode compartment, respectively. By doing so, the anode was expected to be nonlimiting in the performance of the BES, and was verified by performing polarization curves on a weekly basis. Finally, in order to guarantee well-mixed conditions within the compartments and thus avoid concentration gradients, both anolyte and catholyte were recirculated at a rate of 2.4 L h1. In order to collect the gases that would have accumulated during the operation of the reactor, the effluent tubes of both compartments were connected to a gas trap. Batch Operation Tests with Nitrate and N2O. Batch operation tests with nitrate or N2O were performed in the cathode in order to assess the production and removal of denitrification intermediates (nitrate, nitrite and N2O; nitrate batch test), and to characterize the performance and efficiency of the biocathode under conditions where N2O was the sole electron acceptor present (batch operation tests with N2O). Therefore, from day 82nd onward, the cathode compartment was switched to a batch operation mode, while the anode operation remained continuous at an acetate loading rate of 2.39 kg COD m3 NAC d1. Prior to each batch test, residual oxidized nitrogen species present in the catholyte and recirculation vessel were removed by maintaining the system at closed circuit. The latter was veriﬁed by measuring the concentration of nitrate, nitrite and N2O, and also by allowing the cell voltage to reach almost zero (510 mV). Meanwhile, autoclaved M9 medium was prepared in a new recirculation vessel and ﬂushed with nitrogen gas for 15 min prior to connection with the cathode compartment. The latter rendered a total cathodic liquid and headspace volume of 0.860 and 0.350 L, respectively. Approximately 30 min following connection, the desired amount of nitrogen was injected in the recirculation vessel by dosing a certain volume of a concentrated solution of sodium nitrate (nitrate batch test), or by adding a volume of 100% N2O gas in the headspace of the recirculation vessel with a syringe (N2O batch tests). All batch tests were run over a 10 Ω external resistor, except when a potentiostat was used (batch test E). Control Experiments: Open Circuit Tests and Tests with an Abiotic Cathode. Control experiments were performed to determine the electron donor for the cathodic removal of N2O (batch test B and C), and to elucidate the role of the microorganisms present in the biocathode (batch test D). Open circuit tests were identical to closed circuit tests, except that for these cases the external electrical circuit was interrupted at the moment of N2O addition, preventing migration of electrons from anode to cathode. Tests with an abiotic cathode were done by placing autoclaved graphite granules in the cathode compartment. By autoclaving, microbiological activity was assumed to be absent. The amount 10558

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

of granules was equal to the amount present during the tests with a denitrifying biocathode. Batch Tests with N2O under Fixed Cathodic Potential. Batch tests were performed at fixed cathode potentials of 200, 0, and +100 mV vs standard hydrogen electrode (SHE) (batch test E). At the start of each batch test, a certain volume of 100% N2O gas was injected in the headspace of the recirculation vessel. Electrodes were connected with a potentiostat, which kept the system at the desired fixed cathode potential (PAR Bi-Stat Potentiostat, Princeton Applied Research, France; three electrode setup). The current and cathodic potential were measured and recorded every 5 s. Tests were performed in triplicate in the same BES. N2O Feeding during Batch Operation. N2O was supplied each time the cell voltage reached almost zero (510 mV) during the time periods between the different performed batch experiments presented in this paper. Furthermore, the recirculation buffer was replenished at least three times a week in order to prevent possible interference of microbial products (e.g. redox mediators) and to keep the pH constant (7.2 ( 0.1). Continuous N2O Feeding Period. The operation of the cathode was switched to a continuous mode on day 152 after a batch operation period of 70 days. A 100% N2O saturated solution (0.64 g L1 N2ON) was fed to the cathode at a rate of 200 mL d1, giving a loading rate of 2.1 kg N2ON m3 NCC d1, or 10 mA (167 A m3 NCC) when recalculated to current production (see eq. 1). Electrochemical Monitoring. The cell voltage over a fixed resistor and the cathode potential were recorded every minute with a Data Acquisition Unit (HP 34970A, Agilent, Santa Clara, CA). The cathodic potential was monitored with an Ag/AgCl reference electrode (3.5 M KCl; + 0.206 V vs SHE). Polarization curves were obtained according to Clauwaert and co-workers.12 Calculations. Current and power production were calculated according to Ohm’s law. The volumetric current density could be expressed as a theoretical nitrogen removal rate (D; kg N m3 NCC d1) according to the equation (eq eq. 1) reported in Clauwaert et al.12 D¼

IM 86400 s d1 ¼ 2:507 103 I Fn 1000 g kg 1

ðeq:1Þ

I = the volumetric current density (A m3 NCC), M = the molar mass of nitrogen (14 g N mol1), F = Faraday’s number (96485 C mol1), and n = the moles of electrons exchanged per mole nitrate (5 moles of electrons) or mole N2ON (1 mol of electrons) reduced. A gas trap was used to assess the gas production during the continuous nitrate feeding period. The cathodic coulombic eﬃciency (εcathode) was evaluated as the ratio of the coulombs produced and the theoretical amount of coulombs needed basing on the oxidized nitrogen compounds removed at the cathode. Nitrogen removal rates were expressed relative to the net cathodic volume (0.060 L). Removal rates reported with R2 were obtained after ﬁtting the data points enclosed in the given time interval to a linear regression curve. Chemical Analysis. The concentration of NO3 and NO2 were determined using an ion chromatograph (Compact IC 761 with conductivity detector, Metrohm, Switzerland). Ammonium (NH4+) concentration was determined according to the colorimetric Nessler procedure.25

Gas-phase N2O concentration was measured with a gas chromatograph (14B, Shimadzu, Japan) ﬁtted with a 63Ni electron capture detector (detection limit 300 ppbv) according to Roobroeck and co-workers.26 Measurements of the concentration N2O in the liquid-phase (solubility: 0.029 M atm1 at 20 °C27) were performed according to Desloover et al.5 and brieﬂy described in SI.

’ RESULTS The Development of a Denitrifying Biocathode. Continuous feeding with nitrate as cathodic electron acceptor was maintained during an 82 days period in order to selectively enrich for denitrifying microorganisms at the cathode. After inoculation (day 0), a start-up period of 46 days was observed. From then on, an average nitrogen removal rate of 0.21 ( 0.01 kg NO3-N m3 NCC d1 was obtained, resulting in 100% removal efficiency (Table 1). No residual nitrite was detected in the effluent and the concentration of N2O in the gas trap of the cathode compartment was always below 10 μL L1. Considering the average gas production together with the nitrogen load during the stable nitrate removal period, the N2O emission represented only 0.0010.002% of the nitrogen load. An additional nitrate batch test was performed at the cathode in order to quantify the production and removal of denitriﬁcation intermediates during the process (Figure 2). An average nitrogen removal rate and cathodic coulombic eﬃciency of 0.51 ( 0.02 kg N m3 NCC d1 and 95% was obtained, respectively (Table 1). Nitrite accumulation was observed during the ﬁrst eight hours of the experiment, representing up to 55% of the nitrate initially added. In contrast, no such eﬀect could be observed for N2O as the average concentration measured throughout the nitrate batch test amounted only for 0.025 ( 0.022% of the nitrogen initially injected. Furthermore, NH4+ was not detected in the cathodic liquid during both the continuous operation period and the nitrate batch test. A Denitrifying Biocathode with Nitrous Oxide As the Sole Electron Acceptor. After successful enrichment with nitrate, the latter was replaced by N2O as the sole electron acceptor in the cathode compartment. The ability of the biocathode to reduce N2O was investigated in a batch experiment (batch test A, Figure 1 and Table 1). In order to assess accurately the fate of N2O at the cathode, both gas- and liquid-phase concentrations of N2O were measured. The total N2O removal rate clearly followed the same trend as did the current production and the cathodic potential. After reaching maximum performance at 2.5 h, a gradual decrease of the N2O removal rate, current production and cathode potential could be seen until complete depletion of N2O. The cathodic coulombic efficiency of the batch test amounted to 99%. No NH4+, NO3 or NO2 were detected in the catholyte. Changing from nitrate to N2O as the sole electron acceptor present also had an inﬂuence on the diversity of the microbial community (data not shown). Overall, a decrease in bacterial diversity was observed during N2O reduction. This is in line with the fact that a signiﬁcant fraction of the denitrifying population does not have the genetic capacity to reduce N2O.28 Therefore, bacteria lacking the NosZ gene might have been outcompeted, resulting in a decreased bacterial diversity. Most likely, the dominant bacteria can be found in the groups of Proteobacteria, Firmicutes, and Chloroﬂexi as shown by an extensive microbiological study on denitrifying cathodes.29 Furthermore, Virdis and co-workers showed that Paracoccus and Pseudomonas ssp. 10559

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

Table 1. Summary of Results Obtained during Continuous Nitrate Feeding, Batch Operation Tests with Nitrate and N2O, and Continuous N2O Feedinga nitrogen removal rateb experiment

3

(kg N m

1

NCC d )

current production

removal eﬃciency (%)

(mA)

(A m3 NCC)

εcathode (%)

cathode potential (V vs SHE)

continuous nitrate feeding 0.21 ( 0.01c

100 ( 0

5.05 ( 0.57

84 ( 9

100 ( 11

0.206 ( 0.027 (Rext =5Ω)

nitrate batch test N2O batch test A

0.51 ( 0.02 (R2 = 0.9869) 1.83 ( 0.08 (R2 = 0.9931)e

100 100

10.64 ( 3.10 7.32d

177 ( 52 122d

95 99

0.053 ( 0.060 0.138d

closed circuit (01 h)

3.26 ( 0.02 (R2 = 0.9985)

70

10.28d

167d

77

0.079d

27

N/A

N/A

N/A

+0.110d

1.24f 0.07 ( 0.01 (R2 = 0.9954)

50

N/A

N/A

N/A

+0.087d

closed circuit (01 h)

1.02 ( 0.20 (R2 = 0.9271)

35

0.33 ( 0.20

5 ( 3

N/A

0.427 ( 0.021

closed circuit (124 h)

0.04 ( 0.01 (R2 = 0.9607) 41

N/A

N/A

N/A

0.462 ( 0.012

60 ( 3 35 ( 2

7.01 ( 0.39 3.82 ( 0.05

117 ( 7 64 ( 1

95 ( 6 94 ( 14

ﬁxed ﬁxed

0.167 ( 0.015 (Rext =10 Ω)

N2O batch test B closed circuit (12 h)

1.63 ( 0.02 (R = 0.9995)

open circuit (01 h)

1.74 ( 0.04 (R2 = 0.9899)

open circuit (12 h)

0.32 ( 0.10 (R = 0.8384)

2

2

N2O batch test C open circuit (01 h) open circuit (124 h) N2O batch Test D

open circuit (01 h)

0.73 ( 0.28 (R2 = 0.7749)

open circuit (124 h)

0.07 ( 0.01 (R2 = 0.9954)

N2O batch Test E 200 mV vs SHE (n = 3) 0 mV vs SHE (n = 3)

1.45 ( 0.12 (R2 = 0.9730) 0.85 ( 0.04 (R2 = 0.9884)

+100 mV vs SHE (n = 3)

0.65 ( 0.08 (R2 = 0.9374)

26 ( 1

0.11 ( 0.01

2(0

3(0

ﬁxed

continuous N2O feeding

0.76 ( 0.26g

36 ( 12h

3.63 ( 1.25

61 ( 21

N/A

0.204 ( 0.055

N indicates the number of replicates, average ( standard error. N/A: Not applicable. b Nitrogen removal rates were obtained from a linear regression curve including the data points of the addressed time period of the experiment, and were presented as mean value ( standard error unless stated otherwise. c Nitrogen removal rate calculated as mean ( standard error of ﬁve weekly samples. d Maximum current production or cathode potential obtained during test. e Nitrogen removal rate obtained during ﬁrst 2.5 h of the test. f Nitrogen removal rate calculated from two data points (at 0 and 1 h). g Nitrogen removal rate recalculated from the current production. h Assumed that the loading rate was constant (inﬂuent 100% saturated with N2O, giving a loading rate of 2.10 kg N m3 NCC d1). a

were among the most abundant denitrifying organisms at a cathode.30 Open versus Closed Circuit Removal of N2O. Comparing closed and open circuit experiments allowed the investigation of the removal of N2O with and without the supply of electrons derived from the anode (batch test B, Figure 3 and Table 1). N2O was monitored in both gas- and liquid-phase for 2 h in order to assess the nitrogen removal during open and closed circuit operation. During the closed circuit experiment (Figure 3A), the N2O removal rate during the second hour decreased with approximately 50% compared to the first hour, and overall a cathodic coulombic efficiency of 77% was obtained. During open circuit operation (Figure 3B), 23% of the N2O initially injected was removed during the ﬁrst hour. Overall, 27% of the N2O initially added was removed after 2 h. An additional open circuit experiment was performed over 24 h (batch test C) in order to assess the N2O removal over a longer period (Table 1 and SI Figure S.1.). At the end of the experiment, 50% of the added N2O was removed of which 19% during the ﬁrst 2 h. No NH4+, NO3 or NO2 were detected in the cathodic liquid both during the open and closed circuit experiment. N2O Removal in an Abiotic Cathode. In order to address the catalyzing role of the denitrifying microorganisms in the N2O removal process, a 24 h experiment under closed and open circuit conditions (SI Figure S.2) was performed with an abiotic cathode

filled with autoclaved granules (batch test D; Table 1). In total, 35% of the added N2O was removed during the closed circuit experiment, from which 18% during the first hour. For the open circuit experiment, 41% of the N2O initially added was removed, of which 14% during the first hour. N2O Removal at Different Cathodic Polarizations. The removal of N2O was assessed at three different poised cathode potentials (200, 0 and +100 mV vs SHE; Batch test E; Figure 4 and Table 1). By doing so, the theoretical energy gain, determined by the voltage difference between the cathode and the redox potential of the final electron acceptor (N2O; E°0 = +1.36 V vs SHE), could be controlled. In general, lower N2O removal rates were observed at higher cathodic potentials. The removal rates obtained at 200 and 0 mV vs SHE correlated well with the average current production expressed as a nitrogen removal rate, equal to 1.46 ( 0.08 kg N m3 NCC d1 and 0.80 ( 0.01 kg N m3 NCC d1, respectively. In contrast, almost no current was produced at +100 mV vs SHE (Figure 4A), although N2O removal was also observed. The mass balance was veriﬁed by relating the amount of N2O removed expressed as coulombs with the amount coulombs produced through current generation, giving 95 ( 6%, 94 ( 14% and 2.9 ( 0.3% cathodic coulombic eﬃciency for the tests performed at a ﬁxed cathode potential of 200, 0 mV and +100 mV vs SHE, 10560

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

Figure 1. Nitrate batch test. Amount of nitrate (circles; black), nitrite (squares; black), gas-phase N2O (triangles; white), total amount of nitrogen (diamonds; white), current production (solid line; black) and cathode potential (solid line; gray) in function of time.

respectively (Figure 4B). No NH4+, NO3 and NO2 were detected in the catholyte. Long-Term Performance. The cathode of the BES was fed with N2O in a batch operation mode over a period of 70 days (days 82151). From day 152 onward, the BES was operated in continuous mode for 45 days in order to assess the long-term performance of the system. During this period, cell voltage, current production and cathode potential were monitored continuously. From day 156 onward, an average current production of 3.62 ( 1.25 mA (61 ( 21 A m3 NCC) was obtained, which corresponded to a calculated N2O removal rate of 0.76 ( 0.26 kg N m3 NCC d1 (Table 1 and SI Figure S.3). The average cathode potential during this period was 0.204 ( 0.055 V vs SHE (Table 1). Polarization curves were obtained on a weekly basis during the batch and continuous N2O feeding periods as an additional means to monitor the performance of the reactor (SI Figure S.4). From day 95 until day 150 (batch operation period), maximum current and power production decreased from 217 to 73 A m3 NCC and from 25 to 7.6 W m3 NCC, respectively. However, after 20 days of continuous N2O feeding (day 172), maximum current and power production increased again to 104 A m3 NCC and 11 W m3 NCC, respectively.

’ DISCUSSION Autotrophic Nitrate Removal: Performance and Evolution of Nitrogen Intermediates. A denitrifying biocathode was

enriched with nitrate for 82 days as a step to the development of an N2O reducing biocathode. The obtained removal rate and cathode potential during the 35-day (days 4681) stable period during continuous nitrate-feeding (Table 1) were in the same range as the values reported in other studies concerning autotrophic nitrate removal in BES.12,13,17,19 During the nitrate batch experiment (Figure 2), nitrite accumulation was observed. This was in accordance with the results obtained by Puig and coworkers,31 and can be explained from both a thermodynamical and kinetic perspective. The lower redox potential of nitrite (E°0 = +0.35 V vs SHE) compared to nitrate (E°0 = +0.43 V vs SHE), implies that theoretically, less energy is available for the

micro-organisms when nitrite is the electron acceptor,21 and the nitrate reduction rate of the present microbial community appeared to be higher compared to that of nitrite. The N2O emission observed during both the continuous and batch operation with nitrate was low compared to the values reported in previous studies on bioelectrochemical nitrogen removal.13,18,19 The production and emission of N2O are strongly inﬂuenced by many parameters, among which the concentration of ammonium and nitrite, as well as the dissolved oxygen levels are the most important.7 Most likely, the absence of oxygen and ammonium was beneﬁcial for the low N2O emission in this case. Furthermore, the nitrate batch test revealed a high cathodic coulombic eﬃciency of 95%, and indicated that the cathode was the principle electron donor for nitrate reduction. N2O As the Sole Electron Acceptor in a Denitrifying Biocathode: Establishment, Performance and Efficiency. After 35 days of stable nitrate removal, the biocathode was investigated for its ability to treat N2O as the sole electron acceptor present (batch test A, Figure 2 and Table 1). The N2O removal rate was up to 2.6 times higher than the nitrate removal rate observed during the nitrate batch test. This is not surprising since fewer electrons are needed to achieve reduction to N2.21 Therefore, the microbial community might have compensated the shorter electron transport chain by a higher activity. The obtained N2O removal rates in the order of 1 kg N2O N m3 NCC d1 can be considered as relatively high. For comparison, the obtained removal rates in biological nitrate removal systems range from 0.03 to about 1.5 kg N m3 d1.3234 The latter could also explain the absence of N2O emission or accumulation during both the continuous nitrate feeding period and the nitrate batch test. Furthermore, a high cathodic coulombic eﬃciency near 100% was observed and indicated that the electrons originating from the anode were the main electron donors for N2O reduction, and that complete removal of N2O could be established. Removal Mechanisms and Biocatalysis. At open circuit, anode and cathode are not electrically connected. Consequently, no cathodic reduction reaction should be observed and the concentration of N2O should remain constant. Surprisingly, 10561

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

Figure 2. N2O batch test A. Amount of N2O total: squares; gas-phase: triangles; liquid-phase: circles), current production (solid line; black) and cathode potential (solid line; gray) in function of time.

the N2O level also decreased in open circuit. However, this seemed to be mainly a temporary effect since the removal of N2O decreased significantly after 1 h during the open circuit experiments (batch test B, Figure 3B and Table 1; batch test C, SI Figure S.1 and Table 1). Indeed, The total N2O removal rate observed from 1 to 24 hrs during the 24 h open circuit experiment was ca. 20 times lower compared to closed circuit experiments and can therefore most likely be accounted as gas diffusion loss from the reactor. Interestingly, the abovementioned temporary N2O removal mechanism was also observed during the closed circuit experiments since the 2 h closed circuit experiment (batch test B, Figure 3A and Table 1) showed a 2 times higher N2O removal during the first hour compared to the second hour, and a relatively low cathodic coulombic efficiency of 77% was observed. The latter observations thus suggest a temporary alternative removal mechanism during the first hour of both the open and closed batch experiments. Virdis and co-workers explained a similar phenomenon with the capacitance of the graphite granular matrix constituting the electrodes.19 A steep rise of the cathode potential, observed at the beginning of each experiment, results from the discharge of the electrode during which electrons are liberated. This local flow of electrons can subsequently be used by bacteria to endure N2O reduction. However, this does not result in net current production since these electrons were already present in the charged graphite granules in the cathode compartment. Calculations corroborating the latter hypothesis are presented in SI Table S.2. A closed and open circuit experiment with an abiotic cathode (Batch test D; SI Figure S.2 and Table 1) suggested that microorganisms catalyze the reduction reaction since the obtained removal rates obtained between 1 and 24 h were similar to the rates obtained in open circuit with a biocathode. However, when looking to the N2O proﬁle during the ﬁrst hour of both the closed an open circuit abiotic experiment, a 1020 times higher N2O removal was observed. Since no steep rise of the cathode potential took place, a discharge eﬀect of the electrode was likely to be absent and N2O was removed through another mechanism. The absence of ammonium, nitrate, and nitrite

throughout all the batch tests with N2O as an electron acceptor excludes any other reductive or oxidative removal pathway. A plausible explanation would be an initial adsorption eﬀect of N2O to the graphite granules. The latter could also have played a role during the experiments with a biocathode. Most likely, both processes occurred simultaneously. Autotrophic N2O Removal: Energetic Constraints and Energy Conservation. A batch test performed under poised cathode potentials (Batch test E, Figure 4 and Table 1) revealed that the N2O removal rate increased with a factor 2.2 when the cathode potential was decreased from +100 to 200 mV vs SHE. The latter is corroborated by the thermodynamics of the process, since a lower cathode potential increases the theoretical energy gain for the microorganisms. Virdis and co-workers performed similar experiments with nitrate, and observed N2O accumulation, indicating lower specific N2O removal rates as compared with nitrate removal rates.19 In this study, N2O was the sole electron acceptor present, and the observed N2O removal rates were 1.92.5 times higher compared to the values reported by Virdis et al.19 This is a strong indication that a specialized N2O reducing community was present in this study. The current production was proportional to the N2O removal rate for the applied cathode potentials of 200 and 0 mV vs SHE, giving cathodic coulombic efficiencies near 100% (SI Table S.2). However, no current production was observed at a poised cathode potential of +100 mV vs SHE, and indicated that the microorganisms were not able to take up electrons from the cathode at that potential. Nevertheless, N2O removal was observed without concomitant current production, explaining the low cathodic coulombic efficiency observed. The latter suggests again the involvement of an alternative N2O removal mechanism. The removal rate observed here was 10 times higher than what was previously accounted for as gas diffusion loss from the reactor, and leads to a capacitance or adsorption effect as possible explanations. Interestingly, much debate exists regarding the energy conservation associated with the reduction of N2O to N2. Wasser and co-workers stated that the energy derived from N2O reduction is generally dissipated as heat.24 In contrast, Zumft and 10562

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

Figure 3. N2O batch test B. Amount of N2O (A and B, total: squares; gas-phase: triangles; liquid-phase: circles), current production (A, solid line; black), cell voltage (B, dashed line; black) and cathode potential (A and B, solid line; gray) in function of time during closed (Figure 3A) and open (Figure 3B) circuit operation mode.

Richardson reported that each reaction in the denitriﬁcation pathway is catalyzed by an enzyme that is coupled to the production of proton motive force and thus energy conservation.14,23 Yet, the lack of N2O reduction makes only 20% diﬀerence to the bioenergetics of denitrifying bacteria.22 This aspect was investigated on long-term tests by continuing batch feeding of the cathode for 70 days, followed by a continuous feeding period of 45 days. The current proﬁle of the continuous feeding period (SI Figure S.3) revealed that current production of about 61 ( 21 A m3 NCC was sustainable (Table 1). In addition, the cathodic potential of around 200 mV vs SHE that was observed during continuous operations was similar to those reported in similar studies treating nitrate or nitrite.12,17,31 Analysis of the obtained polarization curves (SI Figure S.4) revealed a decreasing performance of the BES over the batch operation period, and was almost entirely due to a decreasing activity of the biocathode (SI Figure S.4B).

However, operating the cathode continuously resulted in an increased performance of the biocathode, and suggest that the microbial community prefers a continuous feeding over a feast and famine regime. The latter was, together with the fact that the biocathode was fed for 115 days with N2O as the sole electron acceptor, a strong indication that the microorganisms could conserve energy from the reduction of N2O without the need for higher oxidized nitrogen species. Bioelectrochemical N2O Removal: Advantages and Challenges. The results clearly show that an N2O reducing biocathode can be sustained at long-term and high activity. The main advantages are that BES can remove N2O at ambient temperatures, and no additional expensive catalysts are needed to perform the reduction reaction. Furthermore, these systems are able to decouple oxidation and reduction processes and can be operated at lower carbon to nitrogen ratios compared to conventional denitrification systems,13,35 leading to a lower demand 10563

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

Figure 4. N2O batch test E. A: Total amount of N2O (gas + liquid phase) and current production in function of time at 200 (circles; black solid line), 0 (triangles; gray solid line) and +100 mV vs SHE (squares; dark gray solid line). B: Coulombs produced (black bars) and amount of N2O removed recalculated to coulombs removed (gray bars). Error bars indicate standard deviation of triplicate experiments.

for organic carbon. Nevertheless, it is worth noting that this technology faces a number of scale-up challenges that need to be resolved prior to the development of practical applications. Furthermore, the O2 sensitivity of the enzyme catalyzing N2O reduction36 can be considered as a major constraint. However, biocathodic denitrification at high dissolved oxygen cocentration is feasible30 and species like P. stutzeri TR2 have been shown to denitrify efficiently at high oxygen partial pressures.37 Once these limitations can be solved, niches could be found in the biological nitrogen removal sector where N2O can be released in significant amounts.4,5,7,38 Hereby aiming at minimizing N2O emissions by

the treatment of recycle streams or effluents rich in dissolved N2O or even N2O contaminated off-gases.

’ ASSOCIATED CONTENT

bS

Supporting Information. Calculations corroborating the capacitance eﬀect are provided. Furthermore, the supplemental ﬁgures of N2O batch tests C and D, the continuous N2O feeding period, the performed polarization curves and the reactor schemes are given. This material is available free of charge via the Internet at http://pubs.acs.org.

10564

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

’ AUTHOR INFORMATION Corresponding Author

*Phone: +32 (0)9 264 59 76; fax: +32 (0)9 264 62 48; E-mail: [email protected].

’ ACKNOWLEDGMENT This work was supported by the project grant for J.D. from the Institute for the Promotion and Innovation through Science and Technology in Flanders (IWT-Vlaanderen, SB-091144). S.P. acknowledges the University of Girona (mobility grant 2010), Befesa Water (CENIT-E TEcoAgua) and the Spanish Government (MCYT-CTQ2008-06865-C02-01/PPQ and CONSOLIDER-CSD2007-00055) for the ﬁnancial support. We thank Tom Hennebel, Jan Arends, and Siegfried E. Vlaeminck for the useful suggestions and critically reading the manuscript. BV wishes to thank the Australian Research Council (Grant DP0985000). ’ REFERENCES (1) Solomon, S.; Qin, D.; Manning, M.; Chen, Z.; Marquis, M.; Averyt, K. B.; Tignor, M.; Miller, H. L. In Climate Change 2007: The Physical Science Basis; Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, A. B., Tignor, M., Miller, H. L., Eds.; Cambridge University Press: Cambridge, 2007; pp 2192. (2) Denman, K. L.; Brasseur, G.; Chidthaisong, A.; Ciais, P.; Cox, P. M.; Dickinson, R. E.; Hauglustaine, D.; Heinze, C.; Holland, E.; Jacob, D.; Lohmann, U.; Ramachandran, S.; da Silva Dias, P. L.; Wofsy, S. C.; Zhang, X. In Climate Change 2007: The Physical Science Basis; Solomon, S., Qin, D.; Manning, M.; Chen, Z.; Marquis, M., Averyt, K. B., Tignor, M., H.L., M., Eds.; Cambrigde University Press: Cambridge, United Kingdom, 2007; pp 501588. (3) Ravishankara, A. R.; Daniel, J. S.; Portmann, R. W. Nitrous oxide (N2O): The dominant ozone-depleting substance emitted in the 21st century. Science 2009, 326 (5949), 123–125. (4) Ahn, J. H.; Kim, S.; Park, H.; Rahm, B.; Pagilla, K.; Chandran, K. N2O Emissions from activated sludge processes, 20082009: Results of a national monitoring survey in the United States. Environ. Sci. Technol. 2010, 44 (12), 4505–4511. (5) Desloover, J.; De Clippeleir, H.; Boeckx, P.; Du Laing, G.; Colsen, J.; Verstraete, W.; Vlaeminck, S. E. Floc-based sequential partial nitritation and anammox at full scale with contrasting N2O emissions. Water Res. 2011, 45, 2811–2821. (6) Hayhurst, A. N.; Lawrence, A. D. Emissions of nitrous-oxide from combustion sources. Prog. Energy Combust. Sci. 1992, 18 (6), 529–552. (7) Kampschreur, M. J.; Temmink, H.; Kleerebezem, R.; Jetten, M. S. M.; van Loosdrecht, M. C. M. Nitrous oxide emission during wastewater treatment. Water Res. 2009, 43 (17), 4093–4103. (8) Lee, S.-J.; Ryu, I.-S.; Kim, B.-M.; Moon, S.-H. A review of the current application of N2O emission reduction in CDM projects. Int. J. Greenhouse Gas Control 2011, 5 (1), 167–176. (9) Wojtowicz, M. A.; Pels, J. R.; Moulijn, J. A. N2O emission control in coal combustion. Fuel 1994, 73 (9), 1416–1422. (10) Centi, G.; Perathoner, S.; Vazzana, F.; Marella, M.; Tomaselli, M.; Mantegazza, M. Novel catalysts and catalytic technologies for N2O removal from industrial emissions containing O2, H2O and SO2. Adv. Environ. Res. 2000, 4 (4), 325–338. (11) Kapteijn, F.; RodriguezMirasol, J.; Moulijn, J. A. Heterogeneous catalytic decomposition of nitrous oxide. Appl. Catal., B 1996, 9 (14), 25–64. (12) Clauwaert, P.; Rabaey, K.; Aelterman, P.; De Schamphelaire, L.; Ham, T. H.; Boeckx, P.; Boon, N.; Verstraete, W. Biological denitriﬁcation in microbial fuel cells. Environ. Sci. Technol. 2007, 41 (9), 3354–3360. (13) Virdis, B.; Rabaey, K.; Yuan, Z.; Keller, J. Microbial fuel cells for simultaneous carbon and nitrogen removal. Water Res. 2008, 42 (12), 3013–3024.

ARTICLE

(14) Zumft, W. G. Cell biology and molecular basis of denitriﬁcation. Microbiol Mol Biol Rev 1997, 61 (4), 533–616. (15) Rabaey, K.; Rozendal, R. A. Microbial electrosynthesis— Revisiting the electrical route for microbial production. Nature Reviews Microbiol. 2010, 8 (10), 706–716. (16) Gregory, K. B.; Bond, D. R.; Lovley, D. R. Graphite electrodes as electron donors for anaerobic respiration. Environ. Microbiol. 2004, 6 (6), 596–604. (17) Clauwaert, P.; Desloover, J.; Shea, C.; Nerenberg, R.; Boon, N.; Verstraete, W. Enhanced nitrogen removal in bio-electrochemical systems by pH control. Biotechnol. Lett. 2009, 31 (10), 1537–1543. (18) Virdis, B.; Rabaey, K.; Rozendal, R. A.; Yuan, Z.; Keller, J. Simultaneous nitriﬁcation, denitriﬁcation and carbon removal in microbial fuel cells. Water Res. 2010, 44, 2979–2980. (19) Virdis, B.; Rabaey, K.; Yuan, Z. G.; Rozendal, R. A.; Keller, J. Electron ﬂuxes in a microbial fuel cell performing carbon and nitrogen removal. Environ. Sci. Technol. 2009, 43 (13), 5144–5149. (20) Xie, S.; Liang, P.; Chen, Y.; Xia, X.; Huang, X. Simultaneous carbon and nitrogen removal using an oxic/anoxic-biocathode microbial fuel cells coupled system. Bioresour. Technol. 2011, 102 (1), 348–354. (21) Thauer, R. K.; Jungermann, K.; Decker, K. Energy conservation in chemotropic anaerobic bacteria. Bacteriol. Rev. 1977, 41 (1), 100–180. (22) Richardson, D.; Felgate, H.; Watmough, N.; Thomson, A.; Baggs, E. Mitigating release of the potent greenhouse gas N2O from the nitrogen cycle - could enzymic regulation hold the key? Trends Biotechnol 2009, 27 (7), 388–397. (23) Richardson, D. J. Bacterial respiration: A ﬂexible process for a changing environment. Microbiology-Sgm 2000, 146, 551–571. (24) Wasser, I. M.; de Vries, S.; Moenne-Loccoz, P.; Schroder, I.; Karlin, K. D. Nitric oxide in biological denitriﬁcation: Fe/Cu metalloenzyme and metal complex NOx redox chemistry. Chem. Rev. 2002, 102 (4), 1201–1234. (25) Greenberg, A. E., Standard Methods for the Examination of Water and Wastewater; American Public Health Association: Washington DC, 1992. (26) Roobroeck, D.; Butterbach-Bahl, K.; Brueggemann, N.; Boeckx, P. Dinitrogen and nitrous oxide exchanges from an undrained monolith fen: Short-term responses following nitrate addition. Eur. J. Soil Sci. 2010, 61 (5), 662–670. (27) Sander, R. Compilation of Henry’S Law Constants for Inorganic and Organic Species of Potential Importance in Environmental Chemistry; Max-Planck Institute of Chemistry: Mainz, 1999. (28) Hallin, S.; Philippot, L.; Andert, J.; Jones, C. M.; Bru, D. Importance of denitriﬁers lacking the genes encoding the nitrous oxide reductase for N2O emissions from soil. Global Change Biol. 2011, 17 (3), 1497–1504. (29) Wrighton, K. C.; Virdis, B.; Clauwaert, P.; Read, S. T.; Daly, R. A.; Boon, N.; Piceno, Y.; Andersen, G. L.; Coates, J. D.; Rabaey, K. Bacterial community structure corresponds to performance during cathodic nitrate reduction. ISME J. 2010, 4 (11), 1443–1455. (30) Virdis, B.; Read, S. T.; Rabaey, K.; Rozendal, R. A.; Yuan, Z. G.; Keller, J. Bioﬁlm stratiﬁcation during simultaneous nitriﬁcation and denitriﬁcation (SND) at a biocathode. Bioresour. Technol. 2011, 102 (1), 334–341. (31) Puig, S.; Serra, M.; Vilar-Sanz, A.; Cabre, M.; Ba~ neras, L.; Colprim, J.; Balaguer, M. D. Autotrophic nitrite removal in the cathode of microbial fuel cells. Bioresour. Technol. 2011, 102 (6), 4462–4467. (32) Kim, D.; Kim, K. Y.; Ryu, H. D.; Min, K. K.; Lee, S. I. Long term operation of pilot-scale biological nutrient removal process in treating municipal wastewater. Bioresour. Technol. 2009, 100 (13), 3180–3184. (33) Pedros, P. B.; Onnis-Hayden, A.; Tyler, C. Investigation of nitriﬁcation and nitrogen removal from centrate in a submerged attached-growth bioreactor. Water Environ Res 2008, 80 (3), 222–228. (34) Vlaeminck, S. E.; Terada, A.; Smets, B. F.; De Clippeleir, H.; Schaubroeck, T.; Bolca, S.; Demeestere, L.; Mast, J.; Boon, N.; Carballa, M.; Verstraete, W. Aggregate size and architecture determine microbial activity balance for one-stage partial nitritation and anammox. Appl. Environ. Microbiol. 2010, 76 (3), 900–909. 10565

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

Environmental Science & Technology

ARTICLE

(35) Ahn, Y. H. Sustainable nitrogen elimination biotechnologies: A review. Process Biochem 2006, 41 (8), 1709–1721. (36) Korner, H.; Zumft, W. G. Expression of denitriﬁcation enzymes in response to the dissolved oxygen level and respiratory substrate in continuous culture of Pseudomonas stutzeri. Appl. Environ. Microbiol. 1989, 55 (7), 1670–1676. (37) Miyahara, M.; Kim, S. W.; Fushinobu, S.; Takaki, K.; Yamada, T.; Watanabe, A.; Miyauchi, K.; Endo, G.; Wakagi, T.; Shoun, H. Potential of aerobic denitriﬁcation by Pseudomonas stutzeri TR2 to reduce nitrous oxide emissions fromwastewater treatment plants. Appl. Environ. Microbiol. 2010, 76 (14), 4619–4625. (38) Foley, J.; de Haas, D.; Yuan, Z. G.; Lant, P. Nitrous oxide generation in full-scale biological nutrient removal wastewater treatment plants. Water Res. 2010, 44 (3), 831–844.

10566

dx.doi.org/10.1021/es202047x |Environ. Sci. Technol. 2011, 45, 10557–10566

ARTICLE pubs.acs.org/est

Field-Scale Reduction of PCB Bioavailability with Activated Carbon Amendment to River Sediments Barbara Beckingham and Upal Ghosh* Department of Chemical, Biochemical, and Environmental Engineering, University of Maryland Baltimore County, 1000 Hilltop Circle, Baltimore, Maryland 21250, United States

bS Supporting Information ABSTRACT: Remediation of contaminated sediments remains a technological challenge because traditional approaches do not always achieve risk reduction goals for human health and ecosystem protection and can even be destructive for natural resources. Recent work has shown that uptake of persistent organic pollutants such as polychlorinated biphenyls (PCBs) in the food web is strongly inﬂuenced by the nature of contaminant binding, especially to black carbon surfaces in sediments. We demonstrate for the ﬁrst time in a contaminated river that application of activated carbon to sediments in the ﬁeld reduces biouptake of PCBs in benthic organisms. After treatment with activated carbon applied at a dose similar to the native organic carbon of sediment, bioaccumulation in freshwater oligochaete worms was reduced compared to preamendment conditions by 69 to 99%, and concentrations of PCBs in water at equilibrium with the sediment were reduced by greater than 93% at all treatment sites for up to three years of monitoring. By comparing measured reductions in bioaccumulation of tetra- and penta-chlorinated PCB congeners resulting from ﬁeld application of activated carbon to a laboratory study where PCBs were preloaded onto activated carbon, it is evident that equilibrium sorption had not been achieved in the ﬁeld. Although other remedies may be appropriate for some highly contaminated sites, we show through this pilot study that PCB exposure from moderately contaminated river sediments may be managed eﬀectively through activated carbon amendment in sediments.

’ INTRODUCTION Fish consumption is known to have several health beneﬁts for humans, but accumulation of legacy toxic chemicals, such as polychlorinated biphenyls (PCBs), in both farmed and wild caught ﬁsh can pose a hazard.1,2 It has been estimated that ﬁsh consumption advisories now cover 43% of the area of lakes and 39% of all river miles in the United States.3 In many of these deteriorated inland and estuarine water bodies some remedial action will be needed to address contaminated sediments, which often serve as the long-term source of legacy contaminants to the aquatic food web. A recent study by the National Research Council found that of the 26 sediment megasites (remediation expense > $50M) that have undergone dredging operations, about half did not achieve the contaminant cleanup levels in sediment immediately following remediation and very few sites documented long-term success .4 Sediment risk assessments are often based on bulk total concentrations and the presumption that all of the chemicals in sediments are available for exposure, which can overestimate risk and lead to bias against any remedy other than sediment removal by dredging or isolation by capping.57 Biological availability of hydrophobic contaminants in sediments, such as PCBs and polycyclic aromatic hydrocarbons, is aﬀected by the nature of binding to native sediment organic matter types, especially to strongly sorbing black carbons, such as soot, coke, and charcoal.8,9 The two key pathways of exposure to ﬁsh as illustrated in Figure S1 r 2011 American Chemical Society

are (1) dietary uptake following accumulation in benthic invertebrates (related to contaminant bioavailability in sediments) and (2) ﬂux of contaminants to the water column and uptake in the pelagic food web (related to diﬀusive and resuspension driven ﬂux into the water column, impacted by partitioning and hydrodynamics). Exposure through both of these pathways can be reduced by altering sediment geochemistry through amendment of stable sorbents such as activated carbon. Activated carbon has a long history of use in air and water cleanup applications such as in gas masks, water treatment systems, and mercury capture in coal-ﬁred power plants. Activated carbon can be produced from coal or from renewable biomass sources such as coconut shells and other agricultural residue, has a very high speciﬁc surface area for sorption (∼1000 m2/g), and is the best low-cost adsorbent for a wide range of toxic compounds including PCBs, dioxins/ furans, polyaromatic hydrocarbons, and chlorinated pesticides. When made from a biomass source and ultimately buried in sediments, activated carbon also has the potential for carbon storage.10,11 Several laboratory studies have demonstrated that amendment of activated carbon (AC) to sediments can suppress biouptake in Received: June 28, 2011 Accepted: November 10, 2011 Revised: November 6, 2011 Published: November 10, 2011 10567

dx.doi.org/10.1021/es202218p | Environ. Sci. Technol. 2011, 45, 10567–10574

Environmental Science & Technology a variety of invertebrate species (Table S1). For example, PCB bioaccumulation in oligochaetes decreased by 7090%, depending on mixing time and particle size, after the addition of 2.6% activated carbon to freshwater sediments, a dose which amounted to half the sediment native organic carbon.12 Although recent laboratory studies have demonstrated contaminant bioavailability reductions in sediment and a small ﬁeld study was conducted in a marine mudﬂat, 13,14 more experience is needed to better understand the feasibility of ﬁeld-scale application of activated carbon in river sediments and eﬀectiveness in PCB bioavailability reduction in the ﬁeld. We report here on a unique study conducted in a PCB-impacted river to evaluate the novel approach of sediment remediation using sorbent amendments. The objectives of the research were to demonstrate that activated carbon application to contaminated river sediments is feasible using large-scale equipment, the applied carbon is stable in the ﬂowing river environment, and is eﬀective in reducing contaminant release into water and uptake at the base of the food chain. Also, we compare bioavailability reductions with ﬁeld amendment to the levels that may be achieved under optimal conditions, such as when most PCBs are associated with the activated carbon in sediment.

’ MATERIALS AND METHODS Field Location and Pilot Study Design. The field site for this study was the lower Grasse River, a tributary to the St. Lawrence River which has been impacted by historic releases of PCBs from an industrial facility and is currently under a fish-consumption advisory. The river is relatively slow-flowing and approximately 4.6 m deep. A full description of the pilot study location is given by Beckingham and Ghosh.15 The lower Grasse River, near Massena, NY, USA is currently undergoing deliberation of remedial alternatives to address legacy sediment contamination by PCBs. The pilot study site is located approximately 5.6 km downstream from a former industrial source of PCBs to the river. Sediments in the study area are composed primarily of sand and silt and in 2006 measured in the range of 2.03.9 μg/g dry wt. for total PCB concentration and averaged 5.8 ( 0.7% by dry wt. (N = 13) for total organic carbon content. In 2006, granular activated carbon (particle size: 75300 μm) was added to sediments at a target dose of 3.75% by dry weight to the top ∼15 cm of surficial sediment as a slurry by three modes of amendment: (1) mixed (using an enclosed tilling device), (2) layered (without mixing), and (3) injected (injection into surficial sediments using two rows of hollow tines) (Figure 1A and B). Bituminous coal-based AC (Carbsorb, Calgon Carbon Corp.) was amended in the mixed and injected applications, and a coconut shell-based AC (055C-CNS-V000, Calgon Carbon Corp.) was used in the layered application. Monitoring sites were established within each of three treatment areas: six in the mixed treatment area (M1M6), three in the injected treatment area (UTA 3, 5, 9), and three in the layered treatment area (UTA 14, 15, 17) (Figure 1C). An upstream background site was established approximately 150 m upstream of the treatment areas for monitoring changes naturally occurring in the river. Two additional background sites, located 15 m up- and downstream of the original background site, were also monitored in 2008 and 2009. Monitoring conducted in 4 years (before amendment and up to 3 years postapplication) at several sites within the upstream background and treatment areas included distribution of activated carbon in bulk sediments and single-point or 5-point composite cores, bioaccumulation from

ARTICLE

sediments in a freshwater oligochaete worm, Lumbriculus variegatus, and measures of aqueous equilibrium, a surrogate for direct pore water measurements. For mixed treatment area and background sites, bioaccumulation tests were performed both in the ﬁeld and in the laboratory. Also, at sampling locations approximately 150 m down-river the potential redistribution of AC to the sediment bed downstream was monitored. Sediment Collection. Bulk sediments for use in bioaccumulation tests, PCB analysis, and aqueous equilibrium measurements were collected using a petit Ponar dredge. Several dredges were combined to obtain enough sediment for all tests. Although effort was made to minimize sediment mixing during collection, some mixing occurred during allocation of sediment between in situ exposure chambers and glass jars which were shipped for laboratory tests. Sediment cores were collected to analyze the depth distribution of activated carbon application within each treatment area (Figure 1C) using manual push cores with Lexan tubing such that the top 30.5 cm of sediment was obtained. Cores were either analyzed separately as single points, or as a composite of 5 cores collected within a ∼1 m2 area to estimate an areaaverage activated carbon dose which is less impacted by smallscale variability. Single-point cores were sectioned into 6 increments (03.8, 3.87.6, 7.611.4, 11.415.2, 15.222.9, and 22.930.5 cm) and 5-point composite cores were sectioned into 3 increments (07.6, 7.615.2, and 15.230.5 cm) below the sedimentwater interface. Laboratory and Field Bioaccumulation Tests. Detailed methods for conducting bioaccumulation tests in the laboratory and in the field are described in Beckingham and Ghosh.15 Bioaccumulation tests were performed with L. variegatus, a freshwater oligochaete worm, by standard exposure tests conducted in the laboratory for 14 days with bulk sediments collected from each background and treatment area monitoring site in beakers with daily water renewal.16 There were 5 replicate exposure beakers for each site each containing ∼0.5 g wet wt. of organisms, ∼150 mL of sediment, and ∼100 mL of Grasse River water. A 16:8 light/dark photoperiod was provided. Water quality parameters were monitored in pooled replicates on a daily (temperature, dissolved oxygen) and weekly (conductivity, alkalinity, ammonia-nitrogen) basis, and met criteria established in the guidelines with the exception of a few minor deviations from the temperature criteria. Organisms were exposed in the ﬁeld according to a method adapted from Burton et al.17 and described in detail elsewhere.15 Site sediments and L. variegatus were introduced to small plastic cylindrical enclosures with end-caps and ﬁne polypropylene mesh screens on opposite sides to allow water ﬂow-through. Chambers were attached in replicates of 6 to a wire basket and anchored to the sediment bottom at 6 mixed treatment sites and 13 upstream background sites. Following the 14-d exposures to sediments in the laboratory or the ﬁeld, worms were collected and cleaned of debris with gentle streams of water, allowed to depurate gut contents for 6 h, and then tissues were weighed and frozen until analysis. Bioaccumulation Tests with Spiked PCBs. An additional laboratory bioaccumulation study exposed L. variegates, according to the same guidelines noted above,16 to either clean sediment spiked with PCBs or activated carbon (coal-based, 75300 μm) spiked with PCBs and then added to clean sediment. The purpose of this experiment was to test the maximum achievable bioavailability alteration of PCBs with activated carbon amendment in a scenario where most of the PCBs are on the AC and 10568

dx.doi.org/10.1021/es202218p |Environ. Sci. Technol. 2011, 45, 10567–10574

Environmental Science & Technology

ARTICLE

Figure 1. (A) Pilot-scale application of activated carbon was carried out in a PCB-impacted river within a 2000 m2 area. The site was enclosed with a silt curtain and application was performed in 2006 using a barge-mounted crane. (B) Placement of the AC was achieved using two custom-made devices: (1) a 2.5 m 4 m rototiller-type enclosed mixing unit with slurry injectors (left) that was used to place and mix activated carbon into surﬁcial sediment or used to apply carbon as a layer on sediment without the mixing action; and (2) a 2 m 3 m tine sled device (right) with nozzles that injected a slurry of activated carbon into surﬁcial sediments (< 10 cm). (C) Diagram of the treatment area locations with respect to one another and the sampling sites. Bulk sediment sampling sites (0) coincided with deployment of ﬁeld bioaccumulation chambers. Sediment cores were collected as single cores (Δ) and as 5-point composite cores (b). The arrow points in the direction of river ﬂow. The colored boxes delineate the target area for activated carbon application: 23 46 m for mixed and 15 30 m for each of the injected and layered treatment areas. (Photo in A courtesy of R.G. Luthy, Stanford University).

fouling is minimal. The clean sediment used in this experiment was obtained from the Rhode River in MD, USA. Total organic carbon content was measured as 3% by dry wt. and PCB concentrations were determined to be below the level of detection (0.01 mg/kg dry wt.). The two treatments were prepared in replicates of four in 500-mL glass wide-mouth jars, by spiking either a slurry of activated carbon (∼2 g) or sediments (∼210 g wet wt.) in 200 mL of filtered streamwater with PCB Aroclor 1260, a commercial mixture of PCB congeners in methanol, then placing the jars on a roller for a 10-d contact period. Following the 10-d contact, the same mass of sediments as in the sediment-only treatment was added to the activated carbon replicates, and both treatments were allowed to settle for 48 h. In this spiking study, the total concentration of PCBs in both treatment groups was 2 μg/g dry wt., and the dose of activated carbon in sediment was 5% by dry weight.

Aqueous Equilibrium Tests. Aqueous equilibrium concentrations were measured at yearly monitoring events according to previously established methods.15,18 In brief, bulk sediments collected from each site and Grasse River water with sodium azide as biocide were gently placed as distinct phases in glass jars and contacted on a very slow orbital shaker for 30 d. The gentle mixing allowed the overlying water column to be mixed without resuspending sediments. At the end of the contact period, the water phase was flocculated with alum to remove colloids, and transferred to a separatory funnel for liquidliquid extraction with hexane to transfer PCBs to the solvent phase for congener-level PCB analysis. Analytical Methods. Tissue and sediment samples were extracted by ultrasonication in hexane/acetone mixture (1:1 by volume) and samples were processed by standard U.S. EPA 10569

dx.doi.org/10.1021/es202218p |Environ. Sci. Technol. 2011, 45, 10567–10574

Environmental Science & Technology guidelines (SW846 methods 3550B, 3660B, 3665A, and 3630C). Prior to extraction, surrogate recovery standards (PCB BZ#14 and 65) were added to assess processing efficiency. Percent surrogate recovery in all analyzed worm, bulk sediment, and aqueous equilibrium samples was within the criteria of 100 ( 30% with only few exceptions for worm samples in 2006 and 2007 where recovery was 100 ( 40%. Congener-level quantification of polychlorinated biphenyls in tissue, sediment, and aqueous equilibrium extracts was performed by gas chromatography with electron capture detection using PCB BZ#30 and 204 as internal standards. A quality control plan was implemented to ensure that the chemical analyses performed were accurate. Matrix blank and matrix spike samples analyzed for cultured L. variegatus found PCB concentrations in organisms to be below the limit of quantiﬁcation and indicated no matrix eﬀects (percent recovery of matrix spike was 100 ( 30% at spike levels of 610 μg and 183 μg total PCBs). Sediment matrix spike recoveries were within 100 ( 30% for background unamended sediments. However, a matrix eﬀect was observed for activated carbon treated sediments. PCBs were found to be not as extractable from AC-treated sediments and therefore, data from the background sites only are used to assess trends in surﬁcial sediment PCB concentration over time (see Figure S2 for details). Activated carbon content of sediments was measured by wet chemical oxidation.19 This method entails oxidation of natural organic matter with sulfuric acid and potassium dichromate, followed by thermal oxidation of the black carbon remaining in the sample by a Shimadzu TOC analyzer. The value of black carbon content measured by the instrument is corrected for carbon content of the AC to determine AC dose in the sediment sample.

’ RESULTS AND DISCUSSION Distribution of AC in Sediments. Measurements of activated carbon content in bulk sediment and sediment cores demonstrated that the amendment was largely applied in surficial sediments (015 cm depth), was present at or above the target dose at most sites in each year, and showed significant small-scale variability (Figures 2 and S3). Activated carbon was not lost from the application area after 3 years of field exposure and could not be detected in sediments downstream of the treatment area. For instance, the average black carbon content of bulk surficial sediments in the study area before activated carbon amendment was 0.20 ( 0.03% (N = 13), which compares well to the downstream measurements in sediment cores in Figure 3D. Also, the average recovery of AC from the top 15 cm of sediment based on composite core analysis (Figure 3AC) ranged 97156% for the mixed sites, 121192% for the layered sites, and 133189% for the injected sites over the 3 years of monitoring after AC application. The generally higher than expected recovery is partly influenced by the high spatial variability of the AC and dry bulk density of the sediment. Over time, the AC content of treated sediments in the composite core sections closest to the sediment water interface (08 cm depth) changed little, while the content of the deeper section (815 cm depth) increased. Single-point cores show similar results but the topmost (04 cm) section decreases in AC content over time and deeper cores indicate an increase in AC content. For example, in the mixed treatment area, the highest level of AC in 2007 was found in the 04 cm depth, and in 2009 was found in the 811 cm depth. However, single cores are also impacted by small-scale hetereogeneity of activated

ARTICLE

Figure 2. Average activated carbon (or black carbon for downstream site) content in sectioned composite-cores taken in 2007 (stripe), 2008 (white), and 2009 (gray) in the (A) mixed (N = 10), (B) layered (N = 8), and (C) injected (N = 8) treatment areas, and (D) downstream (N = 2) of treatment areas. Error bars show (1 standard deviation among core sections.

carbon placement (Figure S4). This trend of increasing AC in deeper cores with time indicates that there was burial of the amendment by sediment deposition onto the treatment plots and some mixing of carbon into these deposited sediments, perhaps by bioturbation. Observation of burial of AC with time is consistent with previous reports of historical long-term average deposition rate of 23 cm/yr using 137Cs dating of sediment cores.20 Reduction in PCB Bioaccumulation. Compared to baseline bioaccumulation measurements at each site, total PCB concentrations in worm tissues from sediment exposures in 2009 were reduced by 8597% (field exposures) and 8998% (laboratory exposures) in sites where the AC was mixed with the sediment (Figure 3). Similar bioaccumulation reductions were observed at the layered application sites (9296%) and injected application sites (9095%). These reductions in bioaccumulation are in a range similar to or higher than those observed in laboratory studies with sediments from other sites (Table S1). The exceptional reductions reported here may be attributed to the relatively low intrinsic binding capacity of Grasse River sediments and dominance of lower chlorinated PCBs. For instance, the fraction of fast desorbing PCBs based on a Tenax desorption study was 6683% for Grasse River sediment,12 compared to 2535% for Hunters Point sediment.13 Total PCB concentrations in worms from each monitoring site are presented in Figure S5 to show variability among exposure replicates. A large reduction in PCB bioaccumulation occurred in the first year after amendment (6995% at all sites above the target AC dose), and in the subsequent 2 years of monitoring this lowered bioaccumulation was either maintained or improved. The background untreated site also demonstrated reductions in PCB bioaccumulation over the monitoring period, 46% for field exposures and 81% for laboratory exposures. Part of this reduced bioaccumulation at the control sites can be explained by the overall reduction of 40% in 10570

dx.doi.org/10.1021/es202218p |Environ. Sci. Technol. 2011, 45, 10567–10574

Environmental Science & Technology

ARTICLE

Figure 4. Average percent reduction in bioaccumulation of the 60 most abundant PCB congeners for ﬁeld exposures at mixed treatment sites compared to (A) baseline and (B) background in each year show increased eﬀectiveness over the monitoring period, especially in 2009 for the more chlorinated, hydrophobic PCBs with higher octanolwater partitioning coeﬃcients (Log KOW).

Figure 3. Relationship between (A) laboratory bioaccumulation, (B) ﬁeld bioaccumulation, and (C) laboratory aqueous equilibrium concentration and activated carbon dose at diﬀerent treatment area sites in each year show increasing eﬀectiveness with activated carbon dose.

sediment PCB concentration over time (Figure S6), mainly caused by deposition of fresh (cleaner) sediment in the study area. Therefore, ongoing natural recovery processes in the river, including deposition of cleaner sediment, biodegradation, and other losses of PCBs from sediment, influence interpretation of results from the pilot-study. Whereas deposition of less contaminated sediment can result in PCB concentration reductions in the control untreated area, the same process has a reverse effect in the treated areas by depositing untreated sediment on top of carbon-amended sediment. However, despite this reduction over time in background bioaccumulation, biouptake in L. variegatus from sites receiving the target AC dose was reduced by 6293% in field tests and 6399% in laboratory tests when compared to the background site in the same year as illustrated in

Figure 3. At mixed treatment sites, the reductions in the higher chlorinated PCBs for field bioaccumulation tests improved with time in comparison to baseline and background sediments, suggesting that longer contact time is needed for the slowly diffusing, more hydrophobic congeners to be sequestered into the AC. As shown in Figure 4, the percent reduction in bioaccumulation 1 year after AC amendment fell sharply with PCB congener hydrophobicity (log KOW). However, with time this trend was lost, and in 2009, 3 years after AC amendment, the percent reductions in bioaccumulation were mostly greater than 80%. This is consistent with descriptions of mass transfer modeling of PCBs in activated carbon-amended sediments which indicate that the higher chlorinated PCBs are more slow to transfer from sediment to AC.21 Reduction in PCB Aqueous Equilibrium. Aqueous equilibrium concentrations were reduced by 95% to greater than 99% compared to background sites (and >93% compared to baseline) for all treatment sites with AC at the target dose or higher in each year (Figure 3). These reductions were comparable to observations at Hunters Point, where mixed addition of 3.2% by dry wt. of coal-based activated carbon reduced aqueous equilibrium concentrations by more than 95% compared to untreated sediments up to 18 months postapplication.13 Aqueous concentration reductions were highest for the monochloro biphenyls 10571

dx.doi.org/10.1021/es202218p |Environ. Sci. Technol. 2011, 45, 10567–10574

Environmental Science & Technology approaching 100%, and decreased with increasing level of chlorination. While removal of DOC by adsorption to AC and resulting loss of DOC-associated PCBs could partly contribute to total aqueous PCB reduction, we would have seen a greater reduction of more chlorinated PCBs compared to the less chlorinated PCBs if this was a dominant mechanism. Reduction in equilibrium aqueous PCBs demonstrates the effectiveness of activated carbon amendment to reduce pore water concentrations, and subsequently the driving force for diffusive flux of PCBs from sediments to the water column. Based on the 3-year post-treatment monitoring data, we show that PCB bioaccumulation and aqueous concentration decreased with increasing dose of AC up to an AC dose close to the native organic carbon content of the sediment (Figure 3). Above an AC dose of 35%, reductions in total PCB aqueous equilibrium concentration approached 100%, but reduction in total PCB concentration in worm tissue in 2009 compared to background are less (7299%) which may be due to the dominance of lower chlorinated congeners in the dissolved phase (80% mono and di) which tend to transfer more easily into AC, and additional route of worm exposure to sediment PCBs through sediment ingestion22 where the ingested sediments have not yet achieved equilibrium distribution of PCBs with AC and water. Comparison of Predicted and Observed Reductions in Bioaccumulation and Aqueous PCBs. To provide perspective on how the level of reductions achieved in the field related to the reductions that may be expected under optimum conditions, bioaccumulation measurements from field-amended sediments were compared to the exposures where AC was first preloaded with PCBs. As shown in Figure S7, when PCB Aroclor 1260 (more chlorinated than Grasse River sediment PCBs) is preloaded on AC and amended to clean sediment, the bioaccumulation is 93% less for total PCBs, with bioaccumulation of tetra and penta chlorobiphenyls less by 98 and 96%, respectively. Bioaccumulation in laboratory exposures with mixed treatment area sediments was plotted against activated carbon dose for PCB congeners that were at similar concentrations in the Grasse River field sediments and in the direct-AC spiked exposures (one coeluting pair of tetra-chlorinated PCBs BZ#66 + 95 and a pentachlorinated PCB BZ#101). As shown in Figure 5, bioaccumulation is greatly reduced with field-aged AC amendment, but not to the level achieved when the majority of PCBs have been transferred to the AC, as in the case of the spiked AC. The difference indicates that optimum equilibrium sorption has not been achieved in the field, likely as a result of the combination of relatively slow mass transfer kinetics and sorption attenuation of PCBs in the field by natural organic matter. To the same purpose, aqueous equilibrium measurements were compared to modeled pore water concentrations assuming equilibrium conditions with a range of activated carbon doses according to a two-carbon model: CS ¼ f OC K OC CPW þ f AC K AC ðCPW Þn where sediment concentration (CS; mg/kg dry wt.) and sediment organic carbon content (fOC) are known values measured for background untreated sediments in 2007 and 2006, KOC values are site-speciﬁc determined from batch equilibrium studies with background untreated and mixed treated sites (N = 7) in 2006, and Freundlich partitioning coeﬃcients (KAC) and linearity parameters (n) were measured in isotherm tests for Carbsorb AC (see Table S2). The Freundlich parameter, n, was found to be

ARTICLE

Figure 5. Bioaccumulation measured in laboratory exposures with mixed treatment area sediments over time ()) for representative (A) tetra- and (B) penta-chlorinated PCB congeners compared to bioaccumulation under optimum conditions when the majority of PCBs are sorbed to AC.

close to 1 over a relatively narrow, low concentration range in the isotherm near the concentrations found in Grasse River water and therefore partitioning was modeled as linear. Werner et al.23 also recently supported the use of a linear partitioning model for black carbon at low, environmentally relevant concentrations (picogram to microgram per liter). The aqueous pore water concentrations of several dominant PCB congeners modeled at equilibrium are 12 orders of magnitude lower than the aqueous equilibrium batch test results for mixed treatment sites over the 3-year monitoring period (Figure 6). This implies that more time is needed to reach equilibrium status in the ﬁeld, or that sorption capacity of the activated carbon for PCBs is attenuated in the ﬁeld due to competition with other sorbates, such as natural organic matter. However, it is important to note that this large apparent diﬀerence is partly accentuated by the fact that several of the aqueous equilibrium concentrations (20 out of 54) for ACtreated sediments were not detected and these data points may actually lie closer to the predictions which are below detection limits. We demonstrate for the ﬁrst time that primary exposure pathways to the aquatic food web can be restricted through pollutant binding in activated carbon amended into contaminated river sediments. Furthermore, AC was successfully applied using both the enclosed tiller (mixed and unmixed) and injection tine devices, and the amendment was stable over time. Although reductions in bioavailability measurements were similar among 10572

dx.doi.org/10.1021/es202218p |Environ. Sci. Technol. 2011, 45, 10567–10574

Environmental Science & Technology

ARTICLE

could be a way to reduce variability and incorporate AC into freshly depositing cleaner sediments more evenly both spatially and also within the surﬁcial sediment depth. Amendment with AC will be most eﬀective at sites that are depositional in nature, less prone to sediment erosion, where native bioavailability of contaminants is high, and ongoing contributions from upstream and terrestrial sources have been controlled. AC amendment provides several advantages over traditional remediation methods, including less disruption to benthic habitats in sensitive rivers and wetlands, amenability to shallow or constricted locations, and potential for lower cost. In situ amendments can also be used in combination with other remedies. This pilot study shows the promise of AC amendment as a new strategy to help address the widespread need to reduce contamination of the aquatic food web from exposure to sediment-bound legacy hydrophobic contaminants.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional ﬁgures and tables. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]; phone: 410-455-8665.

Figure 6. Aqueous equilibrium concentrations measured in batch tests with mixed treatment area sediments over time ()) for representative (A) di-, (B) tri-, and (C) tetra-chlorinated PCB congeners compared to pore water concentrations modeled assuming equilibrium conditions. Measured values below detection limit have not been plotted in the ﬁgure.

application sites, it is diﬃcult to use this result as a basis for supporting a general recommendation for a particular application mode. There may have been some impact of unmeasured natural mixing in the ﬁeld, upon collection of sediment, or by bioturbation of worms within exposure chambers that masks the inﬂuence of application method on bioavailability reduction. Previous laboratory work12 also showed that application of AC without mixing is nearly as eﬀective as application with initial brief mixing, especially when benthic organisms are present that induce mixing through bioturbation. Sediment core analysis showed that there was signiﬁcant small-scale variability in AC dose achieved in sediments. Further improvements in engineering application are needed to reduce AC variability and to improve eﬃciency. For example, multiple applications in small doses over a few years

’ ACKNOWLEDGMENT We thank Alcoa, the U.S. Environmental Protection Agency, Anchor-QEA, and Arcadis-BBL for their support and inputs to the implementation of this project. L. McShea, B. Cook, R.G. Luthy, C. Patmont, P. LaRosa, H. Vanderwalker, D. Buys, and Y. Chang are thanked for initial discussions, planning, and ﬁeld implementation for this study. We also thank R.G. Luthy for the picture in Figure 1A. Adam Grossman is thanked for sediment carbon measurements. The monitoring and laboratory component of this research was supported in part by an unrestricted research gift to UMBC from Alcoa. B.B. was also partly supported by the National Science Foundation Integrative Graduate Education and Research Traineeship “Water in the Urban Environment” program (Grant 0549469). U.G. is a coinventor of two patents related to the technology described in this paper for which he is entitled to receive royalties. One invention was issued to Stanford University (U.S. Patent 7,101,115 B2), and the other to the University of Maryland Baltimore County (UMBC) (U.S. Patent 7,824,129). In addition, U.G. is a partner in a startup company (Sediment Solutions) that has licensed the technology from Stanford and UMBC and is transitioning the technology in the ﬁeld. ’ REFERENCES (1) Hites, R. A.; Foran, J. A.; Carpenter, D. O.; Hamilton, M. C.; Knuth, B. A.; Schwager, S. J. Global assessment of organic contaminants in farmed salmon. Science 2004, 303, 226–229. (2) Kelly, B. C.; Ikonomou, M. G.; Blair, J. D.; Morin, A. E.; Gobas, F. A. P. C. Food webspeciﬁc biomagniﬁcation of persistent organic pollutants. Science 2007, 317, 236–239. (3) U.S. Environmental Protection Agency. 2008 Biennial National Listing of Fish Advisories: Technical Fact Sheet; EPA-823-F-09-007; September 2009; www.epa.gov/ﬁshadvisories. (accessed April 2011). 10573

dx.doi.org/10.1021/es202218p |Environ. Sci. Technol. 2011, 45, 10567–10574

Environmental Science & Technology

ARTICLE

(4) National Research Council. Sediment Dredging at Supefund Megasites: Assessing the Eﬀectiveness; The National Academies Press: Washington, DC, 2007. (5) Ehlers, L. J.; Luthy, R. G. Contaminant bioavailability in soil and sediment. Environ. Sci. Technol. 2003, 37, 295A–302A. (6) Alexander, M. Aging, bioavailability, and overestimation of risk from environmental pollutants. Environ. Sci. Technol. 2000, 34, 4259–4265. (7) Traina, S. J.; Laperche, V. Contaminant bioavailability in soils, sediments, and aquatic environments. Proc. Natl. Acad. Sci. 1999, 96, 3365–3371. (8) Ghosh, U.; Gillette, J. S.; Luthy, R. G.; Zare, R. N. Microscale location, characterization, and association of polycyclic aromatic hydrocarbons on harbor sediment particles. Environ. Sci. Technol. 2000, 34, 1729–1736. (9) Luthy, R. G.; Aiken, G. R.; Brusseau, M. L.; Cunningham, S. D.; Gschwend, P. M.; Pignatello, J. J.; Reinhard, M.; Traina, S. J.; Weber, W. J. Sequestration of hydrophobic organic contaminants by geosorbents. Environ. Sci. Technol. 1997, 31, 3341–3347. (10) Marris, E. Black is the new green. Nature 2006, 442, 624–626. (11) Lehmann, J. A handful of carbon. Nature 2007, 447, 143–144. (12) Sun, X.; Ghosh, U. PCB bioavailability control in Lumbriculus variegatus through diﬀerent modes of activated carbon addition to sediments. Environ. Sci. Technol. 2007, 41, 4774–4780. (13) Cho, Y.-M.; Ghosh, U.; Kennedy, A. J.; Grossman, A.; Ray, G.; Tomaszewski, J. E.; Smithenry, D. W.; Bridges, T. S.; Luthy, R. G. Field application of activated carbon amendment for in-situ stabilization of polychlorinated biphenyls in marine sediment. Environ. Sci. Technol. 2009, 43, 3815–3823. (14) Ghosh, U.; Luthy, R. G.; Cornelissen, G.; Werner, D.; Menzie, C. A. In-situ sorbent amendments: A new direction in contaminated sediment management. Environ. Sci. Technol. 2001, 45, 1163–1168. (15) Beckingham, B. A.; Ghosh, U. Comparison of ﬁeld and laboratory exposures of Lumbriculus variegatus to polychlorinated biphenylimpacted river sediments. Environ. Toxicol. Chem. 2010, 29, 2851–2858. (16) U.S. Environmental Protection Agency. Methods for Measuring the Toxicity and Bioaccumulation of Sediment-Associated Contaminants with Freshwater Invertebrates; EPA 600/R-99/064; Washington, DC, 2000. (17) Burton, G. A.; Greenberg, M. S.; Rowland, C. D.; Irvine, C. A.; Lavoie, D. R.; Brooker, J. A.; Moore, L.; Raymer, D. F. N.; McWilliam, R. A. In situ exposures using caged organisms: A multi-compartment approach to detect aquatic toxicity and bioaccumulation. Environ. Pollut. 2005, 134, 133–144. (18) Ghosh, U.; Weber, A. S.; Jensen, J. N.; Smith, J. R. Relationship between PCB desorption equilibrium, kinetics, and availability during land biotreatment. Environ. Sci. Technol. 2000, 34, 2542–2548. (19) Grossman, A.; Ghosh, U. Measurement of activated carbon and other black carbons in sediments. Chemosphere 2009, 75, 469–475. (20) Comprehensive Characterization of the Lower Grasse River. Volume I - Main Report (pg ES-56). Alcoa, Massena, NY. Amended April 2001. http://www.thegrasseriver.com/pdf/CCLGR_Rept.pdf. (21) Werner, D.; Ghosh, U.; Luthy, R. G. Modeling polychlorinated biphenyl mass transfer after amendment of contaminated sediment with activated carbon. Environ. Sci. Technol. 2006, 40, 4211–4218. (22) Lepp€anen, M. T.; Kukkonen, J. V. K. Relative importance of ingested sediment and pore water as bioaccumulation routes for pyrene to oligochaete (Lumbriculus variegatus, M€uller). Environ. Sci. Technol. 1998, 32, 1503–1508. (23) Werner, D.; Hale, S. E.; Ghosh, U.; Luthy, R. G. Polychlorinated biphenyl sorption and availability in ﬁeld-contaminated sediments. Environ. Sci. Technol. 2010, 44, 2809–2815.

10574

dx.doi.org/10.1021/es202218p |Environ. Sci. Technol. 2011, 45, 10567–10574

ARTICLE pubs.acs.org/est

Oxidation of Trimethoprim by Ferrate(VI): Kinetics, Products, and Antibacterial Activity George A. K. Anquandah,† Virender K. Sharma,*,† D. Andrew Knight,† Sudha Rani Batchu,‡ and Piero R. Gardinali‡ † ‡

Chemistry Department, Florida Institute of Technology, 150 West University Boulevard, Melbourne, Florida 32901, United States Chemistry Department, Florida International University, 3000 N.E. 151st St, North Miami, Florida 33181, United States

bS Supporting Information ABSTRACT:

Kinetics, stoichiometry, and products of the oxidation of trimethoprim (TMP), one of the most commonly detected antibacterial agents in surface waters and municipal wastewaters, by ferrate(VI) (Fe(VI)) were determined. The pH dependent second-order rate constants of the reactions of Fe(VI) with TMP were examined using acidbase properties of Fe(VI) and TMP. The kinetics of reactions of diaminopyrimidine (DAP) and trimethoxytoluene (TMT) with Fe(VI) were also determined to understand the reactivity of Fe(VI) with TMP. Oxidation products of the reactions of Fe(VI) with TMP and DAP were identiﬁed by liquid chromatography-tandem mass spectrometry (LCMS/MS). Reaction pathways of oxidation of TMP by Fe(VI) are proposed to demonstrate the cleavage of the TMP molecule to ultimately result in 3,4,5,-trimethoxybenzaldehyde and 2,4-dinitropyrimidine as among the ﬁnal identiﬁed products. The oxidized products mixture exhibited no antibacterial activity against E. coli after complete consumption of TMP. Removal of TMP in the secondary eﬄuent by Fe(VI) was achieved.

’ INTRODUCTION The presence of pharmaceuticals in streamwater, wastewater eﬄuent, and drinking water has led to widespread research in their monitoring, degradation, and possible adverse eﬀects to aquatic environments.1,2 Active pharmaceuticals and their residues have been detected in the range of trace levels to μg L1 in aqueous environment.3,4 Among the diﬀerent classes of pharmaceuticals detected in the environment, antibiotics are among the most prevalent.5 Antibiotics are frequently used in human and veterinary medicine as well as in aquaculture and farming for both prevention and treating of microbial infections.6 Antibiotics have been detected in freshwater resources and wastewater eﬄuents.4,7 The presence of antibiotics in the aqueous environment can potentially change ecosystems and lead to antibioticresistant bacteria.810 Trimethoprim (5-(3, 4, 5-trimethoxybenzyl)pyrimidine-2,4-diamine, TMP) (Figure S1 of the Supporting Information, SI) has shown little evidence of reversibility of its resistance once established.11 TMP acts as an inhibitor to dihydropteroate synthesase by blocking a step in folate production, which has been attributed to the 2,4-diaminopyrimidine moiety (DAP, Figure S1).12,13 TMP has been frequently detected around the world in surface water and wastewater eﬄuent in the concentration range 30150 ng L1 and 2037 000 ng L1, respectively.14 r 2011 American Chemical Society

TMP is generally not removed by existing conventional water and sewage treatment techniques.15 When TMP is exposed to the commonly used water treatment method of chlorination, mono and dichlorinated products are produced while the 2,4diaminopyrimidine substructure remains intact.16 Ozone oxidation studies of TMP in river water and wastewater have shown g90% removal with ozone,16,17 however, the removal may not necessarily obviate the antibacterial properties completely since the 2,4-diaminopyrimidine substructure was present in some of the degradates.18 The antibacterial study has however demonstrated that oxidation of TMP by ozone or OH radicals resulted in much lower antibacterial activity than that of the TMP.17 The use of ozone raises the concern of production of carcinogenic bromate ion in the treated water.19 Other techniques such as photolysis and TiO2 photocatalysis have been shown to degrade TMP, but some toxicity of oxidation products on V. ﬁscheri was observed.20 This present work describes for the ﬁrst time that the oxidant, ferrate(VI) (Fe(VI)) degrades TMP completely with cleavage of the original molecule, oxidation of Received: June 29, 2011 Accepted: October 27, 2011 Revised: October 23, 2011 Published: October 27, 2011 10575

dx.doi.org/10.1021/es202237g | Environ. Sci. Technol. 2011, 45, 10575–10581

Environmental Science & Technology amino groups of the pyrimidine moiety, and ﬁnally the elimination of antibacterial activity of TMP. In recent years, studies on the oxidation of pharmaceuticals by Fe(VI) have been forthcoming.2126 Kinetics of the reactions have been studied, but the knowledge of stoichiometry and products of the oxidations are very limited. Signiﬁcantly, antibacterial activity of the reaction products, after elimination of the parent drug molecule is missing in the literature. The objectives of the present paper are to: (i) determine species-speciﬁc rate constants for the reactions of Fe(VI) and TMP, and DAP (Figure S1 of the SI) by studying the kinetics of oxidation as a function of pH (ii) determine stoichiometry and identify oxidized products of the reaction between Fe(VI) and TMP in order to propose reaction pathways, and (iii) determine the antibacterial activity of the reaction mixtures against Escherichia coli.

’ EXPERIMENTAL SECTION Reagents. Trimethoprim (TMP), diaminopyrimidine (DAP), 3,4,5-trimethoxytoluene (TMT), LuriaBertani broth, sodium acetate, sodium borate, and sodium hydrogen phosphate were all obtained from Sigma-Aldrich or Fisher with purity higher than 97%. Potassium ferrate solid of ∼98% purity used in the experiments was synthesized by the wet method.27 Fe(VI) solutions were prepared by addition of solid Fe(VI) to 1 103 M Na2B4O7.10H2O/5 103 M Na2HPO4 at pH 9.0. Concentrations of Fe(VI) in the solution were determined spectroscopically at a wavelength of 510 nm using an Agilent 8453 UVvisible spectrophotometer. A molar absorption coefficient, ε510 nm = 1150 M1 cm1 was used to determine Fe(VI) concentration at pH 9.0.27 All solutions were prepared using doubly distilled water that had been passed through an 18 MΩ Milli-Q (Millipore) water purification system. Stock solutions of TMP were prepared by dissolving the solid compound in an appropriate buffer solution and lowering the pH to ∼2.0 in order to facilitate the dissolution. The pH of the dissolved TMP was then adjusted to the desired pH. Diaminopyrimidine stock solutions were prepared in 0.01 M Na2HPO4 buffer solution. Solution of trimethoxytoluene was prepared in 12.5 M methanol0.01 M Na2HPO4 buffer mixture. Methanol aided in dissolution of trimethoxytoluene to achieve the concentration of 2.50 102 M. Kinetic Studies. A stopped-flow spectrophotometer (SX. Eighteen MV, Applied Photophysics, and U.K.) with a photomultiplier (PM) detector was used for the kinetic studies in which the substrate was in excess. Time spectra were collected in the wavelength range from 350 to 750 nm. Kinetic traces were collected at a wavelength of 510 nm to determine the pseudofirst-order rate constants.28 Data collected from the stopped flow were analyzed using the nonlinear least-squares algorithm of SX-18MV global software. Rate constants obtained represent the average of six replicate runs. The UVvisible spectrophotometer was used to study kinetic reactions at pH above 8 where reactions were found to have slow rates. Stoichiometry and Product Studies. The stoichiometric molar ratio of TMP and Fe(VI) was determined by mixing equal solution volumes of 10 mL and the reaction mixtures were maintained at pH 9.0. The TMP was prepared in 1.0 104 M Na2HPO4, while the Fe(VI) in 1 103 M Na2B4O7.10H2O/ 5 103 M Na2HPO4. The concentration of the TMP was kept at 1.0 104 M and the concentrations of Fe(VI) varied from 1.0 104 - 6.0 104 M. This resulted in a [Fe(VI):[TMP] ratio in the reaction mixture of 1: 1 to 6: 1. The Fe(VI)

ARTICLE

concentration was monitored until no residual Fe(VI) was detectable. After completion of the reaction, solutions were filtered using 0.45 μm nylon filters into HPLC vials. The concentrations of the TMP in the resulting reaction mixtures were quantified by the use of a Waters Alliance 2695 HPLC with an Alltima C18 column (250 10 mm, 5 μm) at a wavelength of 271 nm. A binary mobile phase consisting of 70% Solvent A (0.1% HCOOH and acetonitrile at 1:9 (v/v)) and 30% B (0.1% formic acid in water) at a flow rate of 0.4 mL min1 were used in an isocratic elution mode. The mixed solutions, after completion of the reaction between Fe(VI) and TMP, were also subjected to product analysis. The possible inorganic products from the reaction of TMP and Fe(VI) analyzed were NO2, NO3, and NH3. NO3 or NO2 was analyzed by using the Waters Alliance 2695 HPLC with a Waters ion chromatography (IC)-PAC Anion column (4.6 75 mm). Borate gluconate eluant at a ﬂow rate of 1.0 mL min1 was used and the injected volume of the sample was 100 μL. Possible ammonia evolution was tested using an Orion model 9512 ammonia electrode in combination with an Orion benchtop pH/ISE meter model 720A. The organic products formed from the reactions of Fe(VI) with TMP, and Fe(VI) with DAP were studied by LC-MS/MS. The LCQ advantage max ion trap mass spectrometer (IT-MS) was operated under electrospray ionization (ESI) attached to quaternary surveyor LC system (ThermoFinnigan, San Jose, CA). The separations were achieved on a Luna C-18 column (150 cm x 4.6 mm x 5 μm - Phenomenex, Torrance, CA). A binary gradient mobile phase at a ﬂow rate of 0.5 mL/min was used. The mobile phase consisted of methanol (A) and 0.1% formic acid with water (B). Phase A was maintained at 10% for the ﬁrst 2 min, then the percentage was increased to 80% during the next 3 min, then back to 10% in the next 3 min and was maintained at the same level for the last 3 min. The ESI-IT-MS/ MS was operated in the positive ion mode with the capillary temperature at 315 °C and a spray voltage of 4500 V. Data were acquired in the full scan mode (m/z 50400) for identifying intermediates and MS/MS information on the identiﬁed products was obtained in the data dependent scan mode with a collision induced dissociation energy (CID) of 35 eV. Antibacterial Activity Study. Experiments were performed with E. coli 01K1H7 wild type strain at 1 106 CFU/mL. The initial concentration of TMP in the reaction mixtures was TMP 1.0 104 M while the concentration of Fe(VI) concentrations varied from 1.08.0 104 M. All reactions were performed at pH 9.0. After completion of reaction, an aliquot of the reaction mixture was taken and subjected to HPLC analysis. The completion of the reaction was determined when no Fe(VI) remained in the reaction mixture. The residual TMP in each of the reaction samples was determined by the HPLC techniques. The broth microdilution method29 was used to test the biological activity of the mixed solution without filtering (Text S1 of the SI). Control experiments were also performed whereby Fe(VI) which has been reduced to Fe(III) were mixed without TMP. A detailed description of this study is given in Text S1 of the SI. All experiments were conducted in triplicate. The corresponding absorbance readings from the plate reader were converted to growth inhibition percentage by using eq 1. Ið%Þ ¼ fðAmax AÞ=Amax g 100%

ð1Þ

The Amax represents the maximum absorbance reading; implying there is 0% growth inhibition and the inhibition growth I, thus 10576

dx.doi.org/10.1021/es202237g |Environ. Sci. Technol. 2011, 45, 10575–10581

Environmental Science & Technology

ARTICLE

varies from 0 to 100%. Nonlinear regression analysis was used to quantitatively evaluate the antibacterial activity. The relationship between sample dilution and the corresponding growth inhibition I, (%) was fitted using the four parameters sigmodal regression given as eq 2. ð2Þ þ ðY Y Þ=1 þ ðeðX -XoÞ=b Þ I% ¼ Y min

max

min

where Ymax and Ymin represent the maximum and minimum values of the growth inhibition respectively, b is a slope (dimensionless), and Xo is the TMP concentration that provoked a response halfway between the baseline and maximum (EC50). SigmaPlot 2001 software was used to fit the experimental data using eq 2. Removal study. Secondary effluent samples were collected from the Melbourne Reclamation Facility (Florida, USA). The secondary effluent sample had water characteristics as: pH = 7.25, TSS = 0.9 mg/L, nitrogen as N = 11.30 mg/L, phosphorus as P = 1.63 mg/L, BOD = 1.7 mg/L and DOC= 13.9 mg/L. The sample was spiked with 1.7 106 M TMP and various concentrations of Fe(VI) were added to determine the removal of TMP from the effluent. The concentrations of TMP remaining in the mixed solutions, after completion of the reactions, were determined using the LC/MS as described above.

’ RESULTS AND DISCUSSION Kinetics. Initially, the oxidation of TMP by Fe(VI) was studied at pH 7.3 and 25 °C by performing spectral measurements during the reaction (Figure S2 of the SI). Fe(VI) decayed without the apparent formation of Fe(VI)-TMP intermediate in the time scale of the studied reaction. The kinetics of the reaction was followed at 510 nm at different concentrations of TMP under pseudo-first-order conditions. The inset of Figure S3 shows the decrease in absorbance of Fe(VI) with time, which could be fitted nicely to a single exponential decay. This suggests that the reaction is first-order with respect to the concentration of Fe(VI). The pseudo-firstorder rate constants (k0 ) were obtained at different concentrations of TMP and a plot of k0 versus [TMP] showed a linear relationship (Figure S3). A loglog plot of the data from Figure S3 gave a slope of 1.00 ( 0.03; indicating the reaction is also first-order with respect to the concentration of TMP. The second-order rate constants (k), for the reaction of Fe(VI) with TMP were then determined at different pH (Figure 1). Generally, the values of k increased with decrease in pH, similar to results from most studies with Fe(VI).22,24,30,31 Similar kinetic studies at different pH were also performed for the reaction of Fe(VI) with DAP and the results are shown in Figure 1. The trend was similar to those observed for reaction of Fe(VI) with TMP. The variation in the values of k with pH in Figure 1 can be explained by considering reactions between acidbase species of Fe(VI) (H3FeO4+ h H+ + H2FeO4, pKa1 = 1.9; H2FeO4 h H+ + HFeO4, pKa2 = 3.5; HFeO4 h H+ + FeO42‑, pKa2 = 7.2332) and trimethoprim (H2TMP2+ h H+ + HTMP+, pK0 a1 = 3.2;33 HTMP+ h H+ + TMP, pK0 a2 = 7.234). The pH dependence of k for the reaction of Fe(VI) and trimethoprim could be modeled by eq 3. k½FeðVIÞtot ½TMPtot ¼

∑

i ¼ 1, 2, 3, 4, j ¼ 1, 2, 3

Figure 1. Second-order rate constants for the oxidation of TMP, DAP and TMP as a function of pH at 25 °C. (Experimental conditions: [TMP]initial and [DAP]initial = 1 10 3 - 5 10 3 M and [Fe(VI)]initial ≈ 104 M; Solid lines were drawn using kinetic model given in eq 3).

where [Fe(VI)]tot = [H3FeO4+] + [H2FeO4] + [HFeO4]+ [FeO42‑]; [TMP]tot = [H2TMP2+] + [HTMP+] + [TMP]; αi and βj represent the respective species distribution coeﬃcients for Fe(VI) and TMP, i and j represent each of the species of Fe(VI) and TMP respectively and kij is the species-speciﬁc second-order rate constant for the reaction between the Fe(VI) species i and the TMP species j. While there are twelve possible reactions from the mathematical expression in eq 3, only three reactions (eqs 4-6) were needed to ﬁt the experimental values of k (solid line in Figure 1). H2 FeO4 þ HTMPþ f FeðOHÞ3 þ ProductðsÞk22 ¼ ð1:6 ( 0:2Þ 103 M1 s1

ð4Þ

HFeO4 þ HTMPþ f FeðOHÞ3 þ ProductsðsÞk32 ¼ ð8:0 ( 0:5Þ 101 M1 s1

ð5Þ

FeO4 2- þ HTMP f FeðOHÞ3 þ ProductsðsÞk42 ¼ ð2:5 ( 2:1Þ 101 M1 s1

ð6Þ

The rate constants for reactions 46 demonstrate that the diprotonated (or monoprotonated) species of Fe(VI) react faster with TMP species than the monoprotonated (or unprotonated) species of Fe(VI). This is consistent with the general trend predicted using theoretical calculations.35 The species-speciﬁc rate constants for the oxidation of 2,4-diamino pyrimidine (HDAP+ h H+ + DAP, pK0 a2 = 7.434) were also obtained using eq 3 and are shown in eqs 7-9. H2 FeO4 þ HDAPþ f FeðOHÞ3 þ ProductðsÞ k22 ¼ ð2:1 ( 0:2Þ 102 M1 s1

ð7Þ

HFeO4 þ HDAPþ f FeðOHÞ3 þ ProductsðsÞ k32 ¼ ð2:4 ( 0:1Þ 101 M1 s1

kij αiβj ½FeðVIÞtot ½TMPtot

ð8Þ

FeO4 2 þ HDAP f FeðOHÞ3 þ ProductsðsÞ k42 ¼ ð1:4 ( 0:4Þ 101 M1 s1

ð3Þ 10577

ð9Þ

dx.doi.org/10.1021/es202237g |Environ. Sci. Technol. 2011, 45, 10575–10581

Environmental Science & Technology

ARTICLE

Figure 2. Stoichiometry of the reaction between Fe(VI) and TMP at pH 9.0. ([Fe(VI)]R = [Fe(VI)] reacted with TMP).

The higher reactivity of the protonated species than the deprotonated species has been attributed to the larger spin density on the oxo ligands of protonated species.35 Relatively, the k32 and k42 diﬀer by ∼3 fold in the case of TMP, but are similar for DAP. This indicates the involvement of the methylene group of TMP in the oxidation by Fe(VI). However, no such possibility exists in the oxidation of DAP by Fe(VI). The reactivity of Fe(VI) with TMT was also performed in the pH range from 5.0 9.0. The rates were similar with and without 1.25 102 M TMT present in the reaction mixture. This suggests no signiﬁcant reaction between the Fe(VI) and TMT occurred. Stoichiometry and Products. Two sets of experiments were conducted to determine the stoichiometry of the reaction between Fe(VI) and TMP at pH 9.0. In the first set, solutions of different concentrations of Fe(VI) were mixed with solutions that had a fixed initial concentration of TMP. An increase in the concentration of the Fe(VI) resulted in a decrease in the concentration of the TMP, however, Fe(VI) could also react simultaneously with buffered water (Table S1 of the SI). Hence, Fe(VI) and buffered solutions were mixed and the decay of Fe(VI) was monitored in the second set of experiments. The amount of Fe(VI) reacted with only TMP was obtained from this set of experiments. The plot [TMP] vs the [Fe(VI)]reacted showed a linear relationship with the slope (Δ[TMP]/Δ[Fe(VI)]) being 0.20 ( 0.02 (Figure 2). This indicates that the stoichiometric ratio is ∼5:1 ([Fe(VI)]:[TMP]) and the stoichiometry may be written as follows: 5FeðVIÞ þ TMP f 5FeðIIIÞ þ productðsÞ

ð10Þ

Next, a study on the products of reaction 10 was performed. The presence of amine groups in the TMP (see Figure S1 of the SI) suggests the possibility of the formation of inorganic products such as ammonia, nitrite, and nitrate. A ﬁxed concentration of TMP of 2.0 104 M was reacted with diﬀerent concentrations of Fe(VI). Ammonia analyses were conducted using an ammonia electrode in all the reaction mixtures. There was no signiﬁcant evolution of NH3 as the concentration of Fe(VI) was increased in the reaction mixture, suggesting that there was no NH3 released in the oxidation process. The same reaction mixture samples were also subjected to analysis for NO2 and NO3. No detectable levels of either of the ions were observed. This suggests that the amine groups of TMP were not converted to inorganic oxy-nitrogen compounds or ions.

Figure 3. Integrated peak areas in the reaction of Fe(VI) reaction with TMP and DAP at pH 9.0 and 25 °C. ([TMP] = 1.0 104 M and [DAP] = 1.0 104 M, ﬁlled symbols reaction between Fe(VI) and TMP; unﬁlled symbols - reaction between Fe(VI) and DAP; DNP* DAP found in oxidation of DAP by Fe(VI)).

Organic products of the oxidation of TMP by Fe(VI) were studied at diﬀerent molar ratios of added Fe(VI) to TMP (ﬁlled symbols, Figure 3A). Mass spectra of the identiﬁed organic products of the oxidation of TMP by Fe(VI) after separation by chromatography are shown in Figures S4AF of the SI. Among the identiﬁed intermediates and ﬁnal products had m/z = 307 (TMPOH), m/z = 305 (TMP=O), m/z = 197 (TMBA), and m/z = 171 (DNP) (Table S2 of the SI; Figure 3A). TMPOH was observed at a 0.5:1 molar ratio of Fe(VI): TMP, which decreased with the increase in concentration of Fe(VI) until it degraded completely at a molar ratio of 2.7: 1. The formation of nitropyrimidine DNP, was observed when excess amount of Fe(VI) was reacted with TMP. DAP as a possible intermediate before yielding DNP was not seen. This could be attributed to competing reaction rates of Fe(VI) with TMP and DAP which have similar rate constants at pH 9.0. Formation of DNP as the ﬁnal product from DAP most likely occurred through several intermediates, which also reacted with Fe(VI). In a separate experiment, the reaction between Fe(VI) and DAP also formed DNP as the product (unﬁlled symbols, Figure 3). This supports the DAP as the possible intermediate in the reaction of Fe(VI) with TMP. Products and their levels in the reaction mixture indicate that the oxidation of TMP by Fe(VI) involved oxidation and cleavage steps. Proposed reaction pathways for the reaction of Fe(VI) with TMP are presented in Figure 4, which are based on identiﬁed organic products. A number of initial reaction sites on TMP are possible including the exocyclic amino groups of the diaminopyridine ring and an activated bridging methylene group. In pathway (1), Fe(VI) preferentially attacks the reactive bridging methylene group of TMP to undergo oxidation to form intermediate, TMPOH with m/z of 307. The methylene bridge is activated due to the presence of both the electron withdrawing trimethoxybenzene and diaminopyimide rings which can stabilize the resulting intermediate through resonance. Activated methylene oxidation by iron complexes has previously been reported for a number of substrates including cinnamyl alcohol, 10578

dx.doi.org/10.1021/es202237g |Environ. Sci. Technol. 2011, 45, 10575–10581

Environmental Science & Technology

ARTICLE

Figure 4. Reaction pathways for oxidation of TMP by Fe(VI).

tetrahydronaphthalene, ﬂuorene, diphenylmethane, and dihydroanthracene.36 The decrease in TMP and the formation of TMPOH were seen at a Fe(VI) to TMP molar ratio of 0.5:1 (Figure 3). At higher molar ratios of Fe(VI) to TMP, a new product, TMPdO, with m/z 305, was observed (Figure 3). This indicates that TMPOH underwent further reaction to yield the ketone product, TMPdO (pathway (2)). Both reaction pathways have also been suggested in the reaction of TMP with permanganate.37 Additionally, Fe(VI) has been shown to selectively oxidize secondary alcohols to ketones.38 Signiﬁcantly, TMPOH showed a continuous decrease with the increase in the amount of Fe(VI) while TMPdO remained in the solution as a stable product (Figure 3). This is not surprising considering no proton is available on TMP=O for abstraction by Fe(VI). Moreover, diaryl ketons has resonance stabilization, which may not be allowing further oxidation of TMP=O. Therefore, pathway 3 is proposed in which cleavage of TMPOH takes place, to produce DAP and TMBA of m/z 111 and m/z 197, respectively. The amino product, DAP, showed further reactivity with Fe(VI) (Figure 1), hence its concentration decreased and a new product, DNP with m/z of 171 (Figure S4E) was formed (pathway 4). The results of the reaction between Fe(VI) and DAP, shown in Figure 3 (unﬁlled symbols), further indicate the decrease of DAP and concomitant formation of DNP (Figure 3). There was no further decrease in DNP, analogous to the reaction of Fe(VI) with TMP (Figure 3). Formation of nitro compounds was also observed in the reaction of Fe(VI) with arylamines.39,40 Antibacterial Activity after Oxidation. The results of E. coli growth experiments in the absence and presence of Fe(VI) treatment of TMP are presented in Figure 5A. Increases in the concentration of Fe(VI) eventually resulted in no growth

inhibition of E. coli at a concentration of Fe(VI) of 8.2 104 M, which corresponded to a molar ratio of 8:1 of Fe(VI) added to TMP in the reaction mixture. This implies that oxidation of the TMP by Fe(VI) resulted in elimination of the antibacterial properties of TMP. This is illustrated by the shift of the doseresponse curve from the left to the right, where the leftmost curve is the untreated TMP. The antibacterial properties of TMP have been attributed to the 2,4-diaminopyrimidine moiety,12,13 and since this moiety was altered to 2,4- dinitropyrimidine, antibacterial properties were no longer evident. Another possible explanation can be attributed to the lower concentration of the TMP as a result of the reaction, or to any oxidized product(s) that might still have antibacterial properties but less than the minimum inhibition concentration (MIC). A MIC of ∼0.2 x106 M and EC50 of ∼8.0 106 M were estimated from the doseresponse plot. In other studies where E. coli (ATCC 25922) were used with temperatures of 22 °C, 28 and 35 °C and corresponding incubation times of 24 - 28 h, 44 - 48 h, 16 - 20 h respectively, all resulted in a MIC median of 0.21 106 M .41 A control experiment in which E. coli growth was monitored in the presence of Fe(III) showed no growth inhibition of the bacteria, similar to when excess of Fe(VI) was used to consume all TMP from the reaction. Correlating the residual concentration of the TMP in respective samples as the Fe(VI) dosage was increased in reaction, a potency equivalent quotient (PEQ) value can be calculated for each sample using eq 11. PEQ ¼ EC50, 0 =EC50, X

ð11Þ

EC50,0 in eq 11 represents the EC50 value calculated from the measured doseresponse relationship for the TMP without any 10579

dx.doi.org/10.1021/es202237g |Environ. Sci. Technol. 2011, 45, 10575–10581

Environmental Science & Technology

ARTICLE

Figure 6. Residual TMP in secondary eﬄuents of Water Reclamation facility after adding Fe(VI) at pH 9.0 (In mixed solution: [TMP] = 1.7 106 M).

Figure 5. (A) Doseresponse relationships for initial [TMP] = 1.0 104 M with increasing concentration of Fe(VI) at 37 °C and 18 h inoculation, and (B) Potency equivalent quotient plots with the corresponding TMP oxidation by Fe(VI).

oxidizing Fe(VI) added and EC50,x represents the EC50 for the corresponding plots with M Fe(VI) dosed to the TMP to cause oxidation (Figure 5A). The individual calculated PEQ’s are plotted against the [TMP]/[TMP]o in each sample (Figure 5B). This plot was used to evaluate the quantitative relationship between the decrease of the TMP compound and corresponding changes in its biological activity. The straight line represents in which a loss of one mole of TMP would result in the loss of one PEQ of the TMP antibacterial properties. The experimental data represented in Figure 5B showed a negative deviation. This indicates that not only the oxidation products had no antibacterial properties, but also interfered with the inhibition properties of TMP.17 Furthermore, the unequal negative deviation from the ideal at diﬀerent dosages of Fe(VI) suggest that the oxidation products formed at various dosages of Fe(VI) caused diﬀerent interference(s) to the inhibition properties of TMP. A possible role of Fe(III) in the negative deviation (Figure 5B) may not be signiﬁcant because no proportional deviation with increase in the concentration of Fe(III), produced from Fe(VI), was seen. The mixture of products resulting from the Fe(VI) reactions with TMP had insigniﬁcant antibacterial potency in comparison to the TMP. Removal Study. Results of the removal experiments, using secondary effluent water demonstrate that Fe(VI) is effective in oxidizing TMP (Figure 6). The secondary effluent water was relatively clean since it had undergone treatment and did not supposedly have many reactive compounds. At a concentration of 1.7 106 M TMP, Fe(VI) removed TMP completely

(Figure 6). However, demand for Fe(VI) was more than predicted by stoichiometry of the reaction. This indicates that the other components present in the effluent competed to cause an increase in the demand of Fe(VI). Additionally, if excess pollutants present in the secondary effluent have higher rate constants for oxidation with Fe(VI) than TMP, the demand for Fe(VI) would increase for complete removal of TMP from the secondary effluent water. The water matrices such as dissolved organic matter (DOM) would also increase the demand for Fe(VI) since they would be in higher concentration compared to the TMP. Nevertheless, Fe(VI) showed a potential to remove TMP and demand of Fe(VI) dose would vary with the concentration of TMP and constituents of water matrices.

’ ASSOCIATED CONTENT

bS

Supporting Information. Supporting Information (Tables S1S2, Figures S1S4, and Text S1). This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 321-674-7310; fax: 321-674-8951; e-mail: vsharma@ ﬁt.edu.

’ ACKNOWLEDGMENT The authors wish to thank the Center for Ferrate Excellence. Authors also thank Professors Clayton Baum, Nasri Nesnas, and Mary Sohn for their comments and Christa Simmers for guidance with E. coli experiments. The authors also wish to thank the anonymous reviewers for their comments which improved the manuscript greatly. ’ REFERENCES (1) Miege, C.; Choubert, J. M.; Ribeiro, L.; Eusebe, M.; Coquery, M. Fate of pharmaceuticals and personal care products in wastewater treatment plants—Conception of a database and ﬁrst results. Environ. Pollut. 2009, 157, 1721–1726. (2) Mohring, S. A. I.; Strzysch, I.; Fernandes, M. R.; Kiﬀmeyer, T. K.; Tuerk, J.; Hamscher, G. Degradation and elimination of various sulfonamides during anaerobic fermentation: A promising step on the 10580

dx.doi.org/10.1021/es202237g |Environ. Sci. Technol. 2011, 45, 10575–10581

Environmental Science & Technology way to sustainable pharmacy? Environ. Sci. Technol. 2009, 43, 2569– 2574. (3) Benotti, M. J.; Trenholm, R. A.; Vanderford, B. J.; Holady, J. C.; Stanford, B. D.; Snyder, S. A. Pharmaceuticals and Endocrine Disrupting Compounds in U.S. Drinking Water. Environ. Sci. Technol. 2009, 43, 597–603. (4) Pal, A.; Gin, K. Y. H.; Lin, A. Y. C.; Reinhard, M. Impacts of emerging organic contaminants on freshwater resources: Review of recent occurrences, sources, fate and eﬀects. Sci. Total Environ. 2010, 408, 6062–6069. (5) Santos, L. H. M. L. M.; Araujo, A. N.; Fachini, A.; Pena, A.; Delerue-Matos, C.; Montenegro, M. C. B. S. M. Ecotoxicological aspects related to the presence of pharmaceuticals in the aquatic environment. J. Hazard. Mater. 2010, 175, 45–95. (6) K€ummerer, K. Antibiotics in the aquatic environment—A review —Part I. Chemosphere 2009, 75, 417–434. (7) Nagulapally, S. R.; Ahmad, A.; Henry, A.; Marchin, G. L.; Zurek, L.; Bhandari, A. Occurrence of ciproﬂoxacin-, trimethoprim-sulfamethoxazole-, and vancomycin-resistant bacteria in a municipal wastewater treatment plant. Water Environ. Res. 2009, 81, 82–90. (8) K€ummerer, K. Antibiotics in the aquatic environment—A review —Part II. Chemosphere 2009, 75, 435–441. (9) Akiyama, T.; Savin, M. C. Populations of antibiotic-resistant coliform bacteria change rapidly in a wastewater eﬄuent dominated stream. Sci. Total Environ. 2010, 408, 6192–6201. (10) Knapp, C. W.; Dolﬁng, J.; Ehlert, P. A. I.; Graham, D. W. Evidence of increasing antibiotic resistance gene abundances in archived soils since 1940. Environ. Sci. Technol. 2010, 44, 580–587. (11) Sundqvist, M.; Geli, P.; Andersson, D. I.; Sjoelund-Karlsson, M.; Runehagen, A.; Cars, H.; Abelson-Storby, K.; Cars, O.; Kahlmeter, G. Little evidence for reversibility of trimethoprim resistance after a drastic reduction in trimethoprim use. J. Antimicrob. Chemother. 2010, 65, 350–360. (12) Simo, B.; Perello, L.; Ortiz, R.; Castineiras, A.; Latorre, J.; Canton, E. Interactions of metal ions with a 2,4-diaminopyrimidine derivative (trimethoprim). Antibacterial studies. J. Inorg. Biochem. 2000, 81, 275–283. (13) Vandanyan, R. Hruby, V. Synthesis of Essential Drugs; Elsevier: Amsterdam, The Netherlands, 2006; p 617. (14) Fatta-Kassinos, D.; Meric, S.; Nikolaou, A. Pharmaceutical residues in environmental waters and wastewater: current state of knowledge and future research. Anal. Bioanal. Chem. 2011, 399, 251–275. (15) Le-Minh, N.; Khan, S. J.; Drewes, J. E.; Stuetz, R. M. Fate of antibiotics during municipal water recycling treatment processes. Water Res. 2010, 44, 4295–4323. (16) Dodd, M. C.; Huang, C. Aqueous chlorination of the antibacterial agent trimethoprim: Reaction kinetics and pathways. Water Res. 2007, 41, 647–655. (17) Dodd, M. C.; Kohler, H. P. E.; von Gunten, U. Oxidation of antibacterial compounds by ozone and hydroxyl radical: Elimination of biological activity during aqueous ozonation processes. Environ. Sci. Technol. 2009, 43, 2498–2504. (18) Radjenovic, J.; Godehardt, M.; Petrovic, M.; Hein, A.; Farre, M.; Jekel, M.; Barcelo, D. Evidencing Generation of Persistent Ozonation Products of Antibiotics Roxithromycin and Trimethoprim. Environ. Sci. Technol. 2009, 43, 6808–6815. (19) Bonacquisti, T. P. A drinking water utility’s perspective on bromide, bromate, and ozonation. Toxicology 2006, 221, 145–148. (20) Sirtori, C.; Ag€uera, A.; Gernjak, W.; Malato, S. Eﬀect of watermatrix composition on Trimethoprim solar photodegradation kinetics and pathways. Water Res. 2010, 44, 2735–2744. (21) Lee, Y.; von Gunten, U. Oxidative transformation of micropollutants during municipal wastewater treatment: Comparison of kinetic aspects of selective (chlorine, chlorine dioxide, ferrateVI, and ozone) and non-selective oxidants (hydroxyl radical). Water Res. 2010, 44, 555–566. (22) Lee, Y.; Zimmermann, S. G.; Kieu, A. T.; von Gunten, U. Ferrate (Fe(VI)) application for municipal wastewater treatment: A novel process for simultaneous micropollutant oxidation and phosphate removal. Environ. Sci. Technol. 2009, 43, 3831–3838.

ARTICLE

(23) Hu, L.; Page, M.; Marinas, B.; Shisler, J. L.; Strathmann, T. J. Treatment of emerging pathogens and micropollutants with potassium ferrate (VI). Proceedings - Water Quality Technology Conference and Exposition 2010, hul1/1–hul1/8. (24) Hu, L.; Martin, H. M.; Arce-bulted, O.; Sugihara, M. N.; Keating, K. A.; Strathmann, T. J. Oxidation of carbamazepine by Mn(VII) and Fe(VI): reaction kinetics and mechanism. Environ. Sci. Technol. 2009, 43, 509–515. (25) Sharma, V. K.; Li, X. Z.; Graham, N.; Doong, R. A. Ferrate(VI) oxidation of endocrine disruptors and antimicrobials in water. J. Water Supply: Res. Technol.-AQUA. 2008, 57, 419–426. (26) Sharma, V. K.; Mishra, S. K.; Nesnas, N. Oxidation of sulfonamide antimicrobials by ferrate(VI) [FeVIO42‑]. Environ. Sci. Technol. 2006, 40, 7222–7227. (27) Luo, Z.; Strouse, M.; Jiang, J. Q.; Sharma, V. K. Methodologies for the analytical determination of ferrate(VI): A Review. J. Environ. Sci. Health - Part A: Toxic/Hazard. Subs. Environ. Eng. 2011, 46, 453–460. (28) Yngard, R. A.; Sharma, V. K.; Filip, J.; Zboril, R. Ferrate(VI) oxidation of weak-acid dissociable cyanides. Environ. Sci. Technol. 2008, 42, 3005–3010. (29) National Committee for Clinical Laboratory Standards Reference method for broth dilution antifugal susceptibility testing of ﬁlamentous fungi. Approved standards 38-A, 2002. (30) Li, C.; Li, X. Z.; Graham, N.; Gao, N. Y. The aqueous degradation of bisphenol A and steroid estrogens by ferrate. Water Res. 2008, 42, 109–120. (31) Sharma, V. K. Oxidation of nitrogen containing pollutants by novel ferrate(VI) technology: A review. J. Environ. Sci. Health, Part A: Toxic/Hazard. Subst. Environ. Eng. 2010, 45, 645–667. (32) Sharma, V. K.; Burnett, C. R.; Millero, F. J. Dissociation constants of monoprotic ferrate(VI) ions in NaCl media. Phys. Chem. Chem. Phys. 2001, 3, 2059–2062. (33) Qiang, Z.; Adams, C. Potentiometric determination of acid dissociation constants (pKa) for human and veterinary antibiotics. Water Res. 2004, 38, 2874–2890. (34) Roth, B.; Strelitz, J. Z. Protonation of 2,4-diaminopyrimidines. I. Dissociation constants and substituent eﬀects. J. Org. Chem. 1969, 34, 821–836. (35) Kamachi, T.; Nakayama, T.; Yoshizawa, K. Mechanism and kinetics of cyanide decomposition by ferrate. Bull. Chem. Soc. Jpn. 2008, 81, 1212–1218. (36) Shejwalkar, P.; Rath, N. P.; Bauer, E. B. New chiral phosphoramidite complexes of iron as catalytic precursors in the oxidation of activated methylene groups. Molecules 2010, 15, 2631–2650. (37) Hu, L.; Stemig, A. M.; Wammer, K. H.; Strathmann, T. J. Oxidation of Antibiotics during Water Treatment with Potassium Permanganate: Reaction Pathways and Deactivation. Environ. Sci. Technol. 2011, 45, 3635–3642. (38) Lee, D. G.; Gai, H. Kinetics and mechanism of the oxidation of alcohols by ferrate ion. Can. J. Chem. 1993, 71, 1394–1400. (39) Johnson, M. D.; Hornstein, B. J. Unexpected selectivity in the oxidation of arylamine with ferrate-preliminary mechanistic considerations. Chem. Commun. 1996, 965–966. (40) Huang, H.; Sommerfeld, D.; Dunn, B. C.; Lloyd, C. R.; Eyring, E. M. Ferrate(VI) oxidation of aniline. J. Chem. Soc., Dalton Trans. 2001, 1301–1305. (41) Miller, R. A.; Walker, R. D.; Carson, J.; Coles, M.; Coyne, R.; Dalsgaard, I.; Gieseker, C.; Hsu, H. M.; Mathers, J. J.; Papapetropoulou, M.; Petty, B.; Teitzel, C.; Reimschuessel, R. Standardization of a broth microdilution susceptibility testing method to determine minimum inhibitory concentrations of aquatic bacteria. Dis. Aquat. Org. 2005, 64, 211–222.

10581

dx.doi.org/10.1021/es202237g |Environ. Sci. Technol. 2011, 45, 10575–10581

ARTICLE pubs.acs.org/est

Mechanism of Perchlorate Formation on Boron-Doped Diamond Film Anodes Orchideh Azizi,† David Hubler,‡ Glenn Schrader,‡ James Farrell,‡ and Brian P. Chaplin*,† †

Department of Civil and Environmental Engineering and Villanova Center for the Advancement of Sustainable Engineering, Villanova University, Villanova, Pennsylvania 19085, United States ‡ Department of Chemical and Environmental Engineering, University of Arizona, Tucson, Arizona 85721, United States

bS Supporting Information ABSTRACT: This research investigated the mechanism of perchlorate (ClO4) formation from chlorate (ClO3) on boron-doped diamond (BDD) ﬁlm anodes by use of a rotating disk electrode reactor. Rates of ClO4 formation were determined as functions of the electrode potential (2.29—2.70 V/ standard hydrogen electrode, SHE) and temperature (10—40 °C). At all applied potentials and a ClO3 concentration of 1 mM, ClO4 production rates were zeroth-order with respect to ClO4 concentration. Experimental and density functional theory (DFT) results indicate that ClO3 oxidation proceeds via a combination of direct electron transfer and hydroxyl radical oxidation with a measured apparent activation energy of 6.9 ( 1.8 kJ 3 mol1 at a potential of 2.60 V/SHE. DFT simulations indicate that the ClO4 formation mechanism involves direct oxidation of ClO3 at the BDD surface to form ClO3•, which becomes activationless at potentials > 0.76 V/SHE. Perchloric acid is then formed via the activationless homogeneous reaction between ClO3• and OH• in the diﬀuse layer next to the BDD surface. DFT simulations also indicate that the reduction of ClO3• can occur at radical sites on the BDD surface to form ClO3 and ClO2, which limits the overall rate of ClO4 formation.

’ INTRODUCTION Boron-doped diamond (BDD) ﬁlm electrodes have gained increasing interest for their ability to oxidize recalcitrant and complex aqueous waste streams.15 The high oxidizing power of BDD electrodes originates from their ability to oxidize compounds by a combination of direct electron transfer reactions at the electrode surface and indirect oxidation via hydroxyl radicals (OH•) produced from water oxidation.14,6 Various emerging water treatment applications are being developed utilizing BDD electrodes, including: treatment of landﬁll leachate, industrial wastewater treatment, and electrochemical disinfection of cooling tower waters, drinking water, wastewater, swimming pools, and spas.7,8 However, the extreme promise of these electrodes for water treatment is tempered by recent studies showing the production of ClO4 during the electrolysis of chloride-containing waters.914 The production of ClO4 during electrolysis is problematic due to the known health risks, which include disruption of the normal function of the thyroid gland and carcinogenic potential.1517 These risks have prompted the U.S. Environmental Protection Agency (EPA) to issue a health advisory target of 15 parts per billion (ppb) for drinking water sources,18 and two states, California and Massachusetts, have mandated even lower limits of 6 and 2 ppb, respectively.15,19,20 Recent research has shown very high concentrations of both ClO3 and ClO4 formed during extended electrolysis of Cl r 2011 American Chemical Society

and ClOx solutions by use of BDD and Pt anodes.814,21 These previous studies have focused primarily on the relationship between operating conditions (e.g., temperature, ﬂow rate, current density, and Cl concentration) and ClO4 formation.9,10 It has been found that the most important parameters aﬀecting ClO4 formation are the mass-transfer rate to the electrode surface9,10 and the concentration of competitive ions.9,10,12 Low mass-transfer rates enhance ClO 4 formation due to the multistep pathway leading to its formation from Cl as shown: Cl f OCl f ClO2 f ClO3 f ClO4

ð1Þ

where the rate-determining step in this pathway is the oxidation of ClO3 to ClO4.10,21 High concentrations of competitive ions (e.g., Cl) have been shown to inhibit ClO4 formation, due to adsorption at the electrode surface, which blocks the oxidation of ClO3 to ClO4.9 Several studies have investigated the mechanisms and kinetics of ClO4 formation on Pt, Pt/Ti, and PbO2 anodes.2123 However, the proposed mechanisms for the formation of ClO4 on these electrode materials are still Received: July 21, 2011 Accepted: October 27, 2011 Revised: October 14, 2011 Published: October 27, 2011 10582

dx.doi.org/10.1021/es202534w | Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology speculative, and mechanistic studies involving ClO4 formation on BDD electrodes have not been conducted. Past work has shown that the functional groups on the BDD surface have a major eﬀect on the physical, chemical, and electronic characteristics of the BDD surface and the mechanisms of both anodic and cathodic reactions.2426 Freshly prepared BDD surfaces are terminated with hydrogen atoms.25,2733 However, anodic polarization results in oxidation of some surface hydrogen atoms and produces various oxygenated functional groups,2935 such as carboxyl, carbonyl, and hydroxyl groups.25,27,3033,36,37 Evidence suggests that these oxygenated groups mediate electron transfer at BDD electrodes2426 and remain on the surface even after cathodic polarization.25,38 The aim of this work was to develop a mechanistic understanding of ClO4 formation on BDD electrodes. Speciﬁcally, the oxidation of ClO3 to ClO4 was investigated as a function of electrode potential and temperature. The mechanisms of ClO4 formation via various pathways and BDD functional groups were investigated by density functional theory (DFT) modeling.

’ MATERIALS AND METHODS Reagents. All chemicals were reagent-grade and were obtained from Fisher Scientific. All chemicals were used as received without additional purification. All solutions were made from Milli-Q ultrapure water (18.2 MΩ 3 cm at 21 °C). Rotating Disk Electrode Experiments. Reaction rates for ClO3 removal and ClO4 formation were measured at constant potential conditions by use of a rotating disk electrode (RDE) experimental setup. Currents and electrode potentials were controlled and measured with a Gamry series 6000 potentiostat/galvanostat. Experiments were performed over a temperature range of 10—40 °C by use of a circulating water bath (Thermo Electron Corp., Neslab RTE7). Ultrananocrystalline BDD films on 1.0 cm2 surface area p-silicon substrates were used as the working electrode (Advanced Diamond Technologies, Romeoville, IL). Chemical vapor deposition of the BDD films was performed with a concentration of trimethylborane of 75012 000 ppm in flowing CH4, and at a temperature between 700 and 800 °C. The BDD film thickness was approximately 2 μm with a resistivity of 0.05—0.1 Ω 3 cm. The electrochemical cell used for RDE experiments is shown in the Supporting Information (Figure S-1). The working electrode was mounted in a custom-made poly(ether ether ketone) (PEEK) holder attached to a Pine Research Instruments rotator assembly (model AFMSRCE) and rotated at 3000 rotations per minute (rpm) to eliminate both mass-transfer limitations on the reaction rate of ClO3 and current gradients on the BDD surface. The electrode holder exposed a 0.35 cm2 electrode surface area to the electrolyte. The calculated Reynolds number was 34 800. The counterelectrode was a 12 cm long, 0.3 mm diameter Pt wire, and the reference electrode was a single-junction Hg/Hg2SO4/ K2SO4 (mercury sulfate electrode, MSE) (Pine Research Instruments), whose internal filling solution was changed before each experiment. Anode and cathode chambers were separated by a Nafion N115 membrane (Ion Power, Inc., New Castle, DE) in order to isolate anodic and cathodic reactions. All potentials were adjusted for uncompensated solution resistance and are reported versus the standard hydrogen electrode (SHE). Experiments were conducted in 50 mL of either 10 mM or 1 M KH2PO4 buffer, pH 4.5, as a background electrolyte. Before each experiment

ARTICLE

the BDD electrode was preconditioned in a blank electrolyte solution at a current density of 20 mA 3 cm2 for 20 min to remove adsorbed species. All experiments were conducted in duplicate. Linear sweep voltammetry experiments were conducted by use of the same experimental setup as described above, except the electrode was stationary. The potential was swept from the open circuit potential to 2.74 V/SHE at a scan rate of 2 mV 3 s1, in 1.0 M KH2PO4 electrolyte, pH 4.5. Reaction Rate. Two methods were used to calculate reaction rates as a function of electrode potential. In one method (current analysis), the electrode potential was stepped anodically from its open circuit potential in the blank electrolyte solution (10 mM KH2PO4, pH 4.5) to the desired potential (2.29—2.70 V), which generated a constant current for water oxidation. After 3 min, 1 mM ClO3 was added to the electrolyte and the current increase (Δi) was recorded. The current increase was converted to a reaction rate r (moles per hour) by use of Faraday’s law: r ¼

Δi nF

ð2Þ

where n is the number of electrons transferred and F is the Faraday constant. Control experiments were conducted where 1 mM ClO4 was added to the blank electrolyte in place of ClO3. These experiments did not show a measurable current increase, indicating that the solution resistance was not significantly changed by compound addition. The choice of ClO4 for control experiments was made because it has previously been shown to be nonreactive at BDD anodes.39 Analytically determined reaction rates were also calculated by measuring both the disappearance of ClO3 and formation of ClO4 with time. Linear regression of the concentration versus time profiles was used to obtain the reaction rates. All errors reported represent 95% confidence intervals obtained by regression analysis. Analytical Methods. Concentrations of ClO3 and ClO4 were determined by ion chromatography (Dionex ICS-3000; Dionex IonPac AS16 column; KOH eluent; 1 mL/min eluent flow rate). Free available chlorine was measured by Hach method 8021. An Accumet model 25 pH probe was used to measure the solution pH. Electron Transfer Coefficient. The dimensionless electron transfer coefficient (α) can be used to determine the ratedetermining step in an electrochemical reaction mechanism. The dependence of electrochemical reaction rates on potential is described by the ButlerVolmer equation: i ¼ i0 ½eα~FðE Eeq Þ=RT eAαFðE Eeq Þ=RT

ð3Þ

where i is reaction current density, i0 is exchange current density, R is the universal gas constant, T is temperature, E is electrode potential, Eeq is equilibrium potential for the redox reaction, and α B and α A are dimensionless forward (oxidation) and reverse (reduction) electron transfer coefficients, respectively. Combining eq 3 with the Nernst equation and assuming that the reverse reaction is negligible at high overpotentials40 yields the following relationship: 2:3RT d log r ~ α¼ ð4Þ F dE where r is the measured reaction rate. Plots of log r versus electrode potential are used to calculate values of α B from the measured data. A detailed description of how αB relates to a multistep electrontransfer reaction has been provided by Bockris et al.41 The value of α B is a function of the number of electrons transferred before 10583

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology

ARTICLE

Figure 1. Polarization curves recorded on BDD at scan rate of 2 mV 3 s1 in the absence and presence of diﬀerent concentrations of ClO3: 0, 1, 5, 10, 50, and 100 mM. Electrolyte = 1 M KH2PO4, pH 4.5.

(γ B) and after (γ A) the rate-determining step, the number of times the rate-determining step occurs (υ), and the symmetry factor (β) of the reaction.41 Therefore, αB can be expressed as ~ α¼

~ γ þ rβ υ

ð5Þ

where r = 1 for a rate-determining step that involves direct electron transfer and r = 0 for a rate-determining step that is dependent on chemical factors. The parameter β is dependent on the symmetry of the potential energy surface between the reactant and the transition state for the rate-determining step and is close to 0.5 for a direct electron transfer reaction occurring on a metal electrode.41 Quantum Mechanical Simulations. Density functional theory (DFT) simulations were performed to investigate activation barriers for possible reactions involving ClO3. All DFT calculations were performed with the DMol342,43 package in the Accelrys Materials Studio44 modeling suite on a personal computer. All simulations used double-numeric with polarization (DNP) basis sets45 and the gradient-corrected BeckeLee YangParr (BLYP)46,47 functionals for exchange and correlation. The nuclei and core electrons were described by DFT optimized semilocal pseudopotentials.48 Implicit solvation was incorporated into all simulations by use of the COSMO-ibs model.49 The activation energies (Ea) for direct electron transfer as a function of the electrode potential were calculated by the method of Anderson and Kang.50 Reactions with the BDD electrode surface were modeled by use of a previously described 10-carbon atom cluster containing hydrogen and oxygen surface terminations.26 Activation energies for reaction with the surface and with OH• were calculated by minimizing the energy of the system for fixed distances between reacting atoms.

’ RESULTS AND DISCUSSION Experimental Results. Figure 1 shows linear sweep voltammetry profiles recorded on the stationary BDD electrode at a scan rate of 2 mV 3 s1 in the presence of different concentrations of ClO3 (0—100 mM) in the 1.0 M KH2PO4 supporting electrolyte. At potentials lower than those necessary for significant water oxidation, an increase in ClO3 concentration above

5 mM leads to the appearance of an oxidation peak (∼2.4—2.6 V). This peak becomes higher and shifts toward higher potentials with increasing ClO3 concentration, providing evidence that ClO3 reacts on the BDD surface via a direct electron-transfer reaction.1,51 In the region of significant water oxidation (∼ 2.7 V), a drop in the current is observed at all ClO3 concentrations relative to the blank electrolyte. At a potential of 2.7 V, the current progressively decreases up to ClO3 concentrations of 10 mM, and at ClO3 concentrations > 10 mM the current then begins to increase. At low ClO3 concentrations (<10 mM) the blockage of water oxidation sites by adsorbed ClO3 (or reaction products) induces this initial current decrease. However, at higher ClO3 concentration (>10 mM), suppression of the water oxidation reaction is compensated by increased ClO3 oxidation, which leads to a net increase in the observed current. Chronoamperometry experiments were performed at 21 °C to determine the rate of ClO4 formation as a function of electrode potential. These experiments were conducted in a 10 mM KH2PO4 electrolyte, with the RDE rotated at 3000 rpm. Proﬁles for chronoamperometry experiments are shown in Figure S-2 in the Supporting Information. At potentials of 2.60 and 2.70 V, the injection of 1 mM ClO3 to the electrolyte solution resulted in an increase in current compared to the blank electrolyte, providing further evidence of direct electron-transfer reactions.1 Unlike the linear polarization experiments that showed a decrease in current in the presence of 1 mM ClO3 relative to the blank electrolyte at these potentials, chronoamperometric experiments were conducted with a rotating electrode, which prevented mass-transfer control of ClO3 concentrations at the BDD surface, and thus higher currents were observed. At potentials of 2.29 and 2.44 V, the total currents were similar in the presence of 1 mM ClO3 compared to those in the blank electrolyte (Figure S-2, Supporting Information), which support the linear sweep polarization experiments shown in Figure 1. Concentration versus electrolysis time proﬁles for ClO3 removal and ClO4 formation at potentials ranging from 2.29 to 2.70 V at a temperature of 21 °C are shown in Figure 2. Duplicate control experiments, which were conducted without an applied potential, showed that ClO3 was removed from the anode chamber during the ﬁrst 30 min of the experiment. After this time, ClO3 concentrations were approximately constant. Analysis of the reference electrode ﬁlling solution at the completion of the experiments detected an average of 930 μM ClO3, indicating it transferred into the reference electrode through the porous ceramic frit. The ﬁnal mass balance for the control experiments was 103% with respect to the initial ClO3 concentration. Results from the control experiments are plotted in Figure 2a, along with the ClO3 data measured at applied potentials between 2.29 and 2.70 V. All ClO3 data shows a similar trend for the ﬁrst 30 min of reaction, due to transport into the reference electrode. Therefore, rates of ClO3 oxidation were calculated by regressing the data at times >30 min. ClO3 oxidation rates increased with increasing potential (0.006—1.17 μmol 3 h1) and were not statistically diﬀerent than ClO4 formation rates (0.008—1.60 μmol 3 h1) at the 95% conﬁdence level (Figure 2b), indicating that ClO3 was primarily transformed to ClO4. A mass balance for Cl is shown in Supporting Information for all potentials investigated (Figure S-3). The mass balance is based on summation of the measured ClO3 and ClO4 concentrations in each experiment and normalized to the ClO3 concentrations measured in the control experiments. Final mass balances between 98% and 100% were found during oxidation 10584

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology

ARTICLE

Figure 2. (a) ClO3 and (b) ClO4 concentrations as a function of time, at diﬀerent anodic potentials on BDD electrode. (b) Average values of duplicate experiments; (—) linear regressions. Rate constants are presented as micromoles per hour. Experiments were conducted in 50 mL of 10 mM KH2PO4 buﬀer background electrolyte, pH 4.5; at 21 °C. Reported errors represent 95% conﬁdence intervals. (]) Control experiment conducted in the reactor without the BDD electrode. (O) ClO4 concentrations were below the detection limit (0.2 μM) during the ﬁrst 30 min of reaction at 2.29 V/ SHE, and thus these data were not used in the regression.

experiments, and analysis of the liquid samples did not detect other Cl species (e.g., Cl, Cl2, or ClO2). However, the shaft of the RDE prevents a sealed reactor, and the escape of trace volatile compounds was possible under the vigorous mixing conditions employed in the experiments (3000 rpm). In previous nonelectrochemical studies, Cl2O6 (volatile species) was proposed as an intermediate in ClO4 formation,52 through dimerization of ClO3•.53 Evidence was not found to support the formation of Cl2O6, as ClO4 formation was insensitive to the initial ClO3 concentration (data not shown). However, its formation and subsequent volatilization from solution was possible at trace levels. A comparison between the two methods used to calculate the reaction rates of ClO4 formation shows that, at high oxidation potentials (2.60 and 2.70 V; Figure S-2, Supporting Information), the rates calculated by current analysis using eq 2 for a one-electron transfer reaction are 3.73 and 9.70 μmol 3 h1 at 2.60 and 2.70 V, respectively, which are approximately 4 and 6 times higher than the analytically measured ClO4 formation

rates of 0.98 and 1.60 μmol 3 h1 at these same potentials. These results indicate that additional direct electron transfer reactions involving either ClO3 or reaction products are occurring at the BDD surface that do not directly lead to ClO4 formation. The lack of a measurable electrode response to 1 mM ClO3 injection at potentials of 2.29 and 2.44 V is likely due to the fact that the calculated Δi values were 1.23 and 24.7 μA 3 cm2, respectively, which were determined by plugging the analytically measured rates into eq 2. These values are comparable to the variation in the measured currents, which were (7 and (20 μA 3 cm2 at 2.29 and 2.44 V, respectively. Total current eﬃciencies for ClO4 formation from ClO3 ranged from 2.2% to 4.0% for a twoelectron transfer, indicating that the reaction is not highly favorable on the electrode surface compared to water or organic compound oxidation. For example, oxidation of N-nitrosodimethylamine at BDD electrodes achieved current eﬃciencies of 77—99% over a similar electrode potential range (i.e., 2.39— 2.64 V) and substrate concentration (i.e., 1.35 mM).3 10585

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology

ARTICLE

Figure 4. Activation barrier calculation as a function of electrode potential for direct oxidation of ClO3. Atom key: Cl, green; O, red.

Figure 3. Calculation of electron-transfer coeﬃcients by regression of (a) log rate of ClO3 removal versus potential and (b) log rate of ClO4 formation versus potential.

Insight into the mechanism of ClO4 formation may be gained by examining values of α B obtained for ClO3 oxidation and ClO4 formation by use of eq 4. Values of 0.40 ( 0.08 and 0.45 ( 0.12 were obtained for α BClO4, respectively (Figure 3a). The BClO3 and α regressions shown in Figure 3 do not include the rates measured at 2.70 V, which were omitted because rates began to plateau at potentials higher than 2.60 V. This observation has been made in other studies and has been attributed to oxygen bubble formation on the electrode surface that physically blocks reaction sites3 or to leveling oﬀ of the OH• concentration at potentials approaching 3.0 V.54 Therefore, in order to avoid speculation on data interpretation, only potentials e 2.60 V are considered. A value for αBClO3 = 0.40 is close to the theoretical value of α B = 0.5 determined by eq 5, suggesting a one-electron direct transfer reaction is the ratedetermining step for ClO3 oxidation. This ﬁnding is consistent with results from linear sweep voltammetry and chronoamperometry experiments that suggest a direct electron transfer pathway for ClO3 oxidation (Figure 1 and Figure S-2, Supporting Information). The fact that a value of <0.5 was found for αBClO3 is likely related to an asymmetrical potential energy surface with β < 0.5. Typical values reported for α B for direct electron transfer reactions on BDD electrodes range from 0.3 to 0.4.55 The conversion of ClO3 to ClO4 involves an overall twoelectron transfer reaction that also involves the rearrangement of chemical bonds (i.e., the addition of oxygen). The half-reaction can be written as ClO3 þ H2 O f ClO4 þ 2Hþ þ 2e

ð6Þ

A measured value of α BClO4 = 0.45 ( 0.12 was found from the rates of ClO4 formation (Figure 3b). The similar values found for α BClO4 indicate that oxidation of ClO3 via direct BClO3 and α electron transfer is the rate-determining step for ClO4 formation. If transfer of the second electron was the rate-determining step, α BClO4 close to 1.5 should be observed, according to eq 5. If the overall reaction rate were limited by a chemical reaction (i.e, r = 0), α BClO4 close to 0 or 1.0 should be observed.

The temperature dependence of the ClO4 formation rate was used to calculate an apparent activation energy for the oxidation of ClO3 to ClO4. The ClO4 formation rate was measured at 2.60 V and temperatures from 10 to 40 °C, which yielded an apparent Ea of 6.9 ( 1.8 kJ 3 mol1 (Figure S-4, Supporting Information). Values for Ea this small are normally indicative of activationless processes,56 such as temperature eﬀects on the composition and thickness of the electrical double layer or the relative adsorption strengths of water and ClO3 on the electrode surface. Density Functional Theory Modeling. In order to elucidate the reaction mechanisms of ClO4 formation, DFT simulations were used to calculate Ea values for direct electron transfer from ClO3 and for the homogeneous reaction between OH• and ClO3. DFT results for the direct electron transfer reaction ClO3 f ClO3 • þ e

ð7Þ

are shown in Figure 4. Ea decreases as a function of the applied potential and becomes activationless at potentials > 0.76 V (Figure 4). This result is consistent with results from linear sweep voltammetry and chronoamperometry experiments and calculated values for α B that support a direct electron transfer pathway. DFT simulations indicate that removing an additional electron from ClO3• did not produce a stable structure, and therefore a direct two-electron transfer reaction is not the likely pathway for ClO4 formation. Additionally, DFT simulations indicate that the direct homogeneous reaction between OH• and ClO3 (as shown in eq 8) did not occur, which supports experimental evidence in the literature that this reaction rate is below the quantification limit of the spin trap method (<1 106 M1 3 s1).57 ClO3 þ OH• f ClO4 þ Hþ þ e

ð8Þ

Additional DFT simulations were conducted to gain information on surface complexes that may form with ClO3•. These simulations investigated the interaction of ClO3• with diﬀerent functional groups on the 10-carbon atom diamond cluster (Supporting Information, Figure S-5). X-ray photoelectron spectroscopy has identiﬁed hydrogen (tCH), hydroxyl (tCOH), aldehyde (=CHO), and carbonyl (=C=O) surface terminations on BDD surfaces, where the tCOH group is the most prevalent oxygenated site and has been determined to represent ∼10—20% of surface groups.3033 Under anodic polarization, these surface functional groups may become oxidized and undergo loss of either an electron or H atom, producing surface radical sites. DFT results indicate that ClO3• did not chemisorb to either tCH or =C=O sites. The calculated Ea associated with ClO3• adsorption at a =CHO• site was 158 kJ 3 mol1. The high Ea 10586

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology

ARTICLE

Scheme 1. Proposed Reaction Mechanism for Formation of ClO4 from ClO3 on BDD Anodesa

a

Atom key: C, gray; Cl, green; H, white; and O, red. Full structure of 10-carbon diamond with diﬀerent functional groups is presented in Supporting Information, Figure S-5.

indicates that it is not an important reaction site at room temperature. However, ClO3• formed activationless chemisorbed complexes with tC• and tCO• sites, indicating that at least two distinct sites on the BDD surface participate in chemisorption of ClO3•. The tCO• site may be an important site for oxygen evolution through an electrochemical desorption mechanism (eq 9). Therefore, the fact that DFT simulations indicate that ClO3• adsorbs at this site supports the experimental results showing that low concentrations of ClO3 (<10 mM) block water oxidation. CO• þ OH• f C• þ Hþ þ O2 þ e

ð9Þ

•

Other DFT simulations indicate that ClO3 chemisorbs to the tC• site by both its oxygen atom (H15C10OClO2) and by its chlorine atom (H15C10ClO3), as shown in Figures S-6 and S-7 in Supporting Information, respectively. Adsorption via both Cl and O atoms at the tC• site is activationless, with overall reaction energies of 32 and 279 kJ 3 mol1, respectively. DFT simulations involving OH• attack on chemically adsorbed ClO3• with either bonding conﬁguration yields HClO3 as a product, which is released into solution while the oxygen remains on the BDD surface, forming a tCO site, as shown in eqs 10 and 11: COClO2 þ OH• f CO þ HClO3 CClO3 þ OH• f CO þ HClO3

ð10Þ ð11Þ

The reactants and products of the reaction between physisorbed OH• and chemisorbed ClO3• by its Cl atom at the tCO site

are shown in the Supporting Information (Figure S-8). The overall reaction energy for OH• attack of ClO3• bonding via its Cl atom was 560 kJ 3 mol1 with an Ea of 27.6 kJ 3 mol1, and the overall reaction energy for OH• attack of ClO3• bonding via its O atom was 64.6 kJ 3 mol1 with an Ea of 12.7 kJ 3 mol1. The relatively low Ea values indicate that these reactions can occur at the temperatures in our experiments. The chemically bonded ClO3• intermediate that subsequently reacts back to HClO3 explains why the measured rate of ClO4 production was much less than the calculated rate using eq 2. These results indicate that monitoring only the faradic current and correlating it to reaction rates is problematic and should be used with extreme caution. From a practical standpoint, this pathway also may act to limit ClO4 formation. Two other reactions were also found to take place at the tCO• and =CHO• sites. The ﬁrst reaction involves ClO3• reacting at the BDD surface to form ClO2 and O2, through coordination of ClO3• with its oxygen atom at the tCO• and = CHO• sites. For the =CHO• site, the overall reaction energy was 79.9 kJ 3 mol1 with an Ea of 56.9 kJ 3 mol1. The relatively high Ea indicates that it is not an important reaction in our experiments. However, for the tCO• site the reaction was activationless and may provide an additional pathway that could limit ClO4 formation, as shown below: CO• þ ClO3 • f C• þ ClO2 þ O2

ð12Þ

Another possible reaction involves OH attack on ClO3• chemisorbed to the tCO• site. This reaction produces HClO4 with an overall reaction energy of 112 kJ 3 mol1 and a calculated Ea of 66.0 kJ 3 mol1. The relatively high Ea compared •

10587

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology

ARTICLE

to the measured value indicates that this mechanism does not likely contribute signiﬁcantly to ClO4 formation in our experiments. In addition to bonding to the BDD surface, ClO3• may also react with OH• in the solution adjacent to the electrode surface. The existence of OH• in the bulk solution is unlikely since its lifetime in aqueous solution is on the order of <1 μs,58 and thus it would not diﬀuse out of the boundary layer in this amount of time. The energy proﬁles of this reaction as a function of the ClOH bond length, along with the reactants and products, are shown in Figure S-9 in the Supporting Information. The DFT simulations indicate that the homogeneous reaction between ClO3• and OH• is activationless and leads to HClO4 formation, and thus this pathway is likely the primary contributor to ClO4 formation in our experiments: ClO3 • þ OH• f HClO4

ð13Þ

Speculation that eq 13 may be involved in electrochemical perchlorate formation has been previously reported.59 Perchlorate Formation Mechanism and Environmental Significance. The proposed reaction mechanism for the formation of ClO4 from ClO3 on BDD anodes is summarized in Scheme 1. Over the potential range investigated in this study, the experimental and DFT results indicate that formation of ClO4 from the oxidation of ClO3 proceeds through a two-step mechanism. The first step involves the direct transfer of one electron from ClO3 to the BDD anode, which becomes activationless at potentials >0.76 V. Subsequent solution-phase reaction of ClO3• with OH• produced from water oxidation at potentials greater than 1.5 V54 produces ClO4 via an activationless pathway. Calculations estimate that 16—26% of the ClO3• formed via direct electron transfer goes on to produce ClO4. ClO3• forms chemisorption complexes with the BDD surface at tC• sites via an activationless step and subsequently reacts with physisorbed OH• to produce HClO3 and an oxidized surface site (tCO) via pathways with low to moderate Ea values (12.7— 27.6 kJ 3 mol1) (Scheme 1a). ClO3• also either chemisorbs or reacts at tCO• sites on the BDD surface (Scheme 1b). The reaction between ClO3• and tCO• produces ClO2 and O2 via an activationless step. Subsequent reaction of chemisorbed ClO3 at the tCO• site with OH• produces HClO4 via a high activation barrier step (66 kJ 3 mol1) (Scheme 1b). The mechanistic insights provided in this study help explain the 2 orders of magnitude higher ClO4 formation rate with BDD electrodes as compared to other electrodes (i.e., Pt, IrO2, IrO2RuO2) (Table S-1, Supporting Information),9 which is primarily due to the ability of the BDD electrode to produce both ClO3• and OH• at high concentrations. The resulting transformation products of ClO3• were found to be sensitive to speciﬁc functional groups on the BDD surface, resulting in the production of ClO4, ClO3, or ClO2. Due to the great potential of BDD electrodes to oxidize aqueous waste streams, either electrode modiﬁcations or operational strategies will be needed to limit or eliminate ClO4 formation. For example, the preparation of BDD electrodes with a high density of tCO sites may lead to suﬃcient side reactions that would signiﬁcantly limit ClO4 formation and thus will be investigated in future work. Results from this study indicate that extreme caution must be taken when BDD electrodes are used for the oxidation of chloride-containing waters. Perchlorate concentrations at the end of the 3.5-h oxidation experiments ranged from 0.7 to 120 μM (70—12 000 ppb) between 2.3 and 2.7 V, respectively. These

values are well over the U.S. EPA’s health advisory target of 15 ppb and drinking water limits set by California and Massachusetts, 6 and 2 ppb, respectively. Studies using BDD electrodes to oxidize reverse osmosis brines and landﬁll leachates have documented levels of ClO3 as high as 630 and 900 mg/L, respectively.7,60 The waters used in these studies initially contained high levels of both dissolved organic carbon (20—300 mg/L as C) and NH4+ (200—800 mg/L), which both should scavenge the Cl2 produced and thus limit ﬁnal ClO3 concentrations. While ClO4 was not measured in these studies, the observed ClO3 concentrations were an order of magnitude higher than the initial ClO3 concentration used in our study (83 mg/L), indicating that ClO4 formation likely occurred. However, research is needed to investigate the mechanisms of ClO4 formation in complex waste streams, as prior work has indicated that ClO3• can react with organic compounds, which could substantially lower ﬁnal ClO4 concentrations.4

’ ASSOCIATED CONTENT

bS

Supporting Information. Nine ﬁgures and one table, showing electrochemical cell setup, chronoamperometry experiments, mass balance of RDE experiments, Arrhenius plot for ClO4 formation, molecular structures of all compounds investigated, DFT simulations of ClO3 oxidation on BDD, and comparison of ClO4 formation on diﬀerent electrodes. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 610-519-4967; fax: 610-519-6754; e-mail: brian.chaplin@ villanova.edu.

’ ACKNOWLEDGMENT Funding for this work was provided by the National Science Foundation (CBET-0931749) and Villanova University. ’ REFERENCES (1) Zhi, J. F.; Wang, H. B.; Nakashima, T.; Rao, T. N.; Fujishima, A. Electrochemical incineration of organic pollutants on boron-doped diamond electrode. Evidence for direct electrochemical oxidation pathway. J. Phys. Chem. B 2003, 107 (48), 13389–13395. (2) Carter, K. E.; Farrell, J. Oxidative destruction of perﬂuorooctane sulfonate using boron-doped diamond ﬁlm electrodes. Environ. Sci. Technol. 2008, 42 (16), 6111–6115. (3) Chaplin, B. P.; Schrader, G.; Farrell, J. Electrochemical oxidation of N-nitrosodimethylamine with boron-doped diamond ﬁlm electrodes. Environ. Sci. Technol. 2009, 43 (21), 8302–8307. (4) Chaplin, B. P.; Schrader, G.; Farrell, J. Electrochemical destruction of N-nitrosodimethylamine in reverse osmosis concentrates using boron-doped diamond ﬁlm electrodes. Environ. Sci. Technol. 2010, 44 (11), 4264–4269. (5) Pacheco, M. J.; Santos, V.; Ciriaco, L.; Lopes, A. Electrochemical degradation of aromatic amines on bdd electrodes. J. Hazard. Mater. 2011, 186 (23), 1033–1041. (6) Kapalka, A.; Foti, G.; Comninellis, C. Investigation of the anodic oxidation of acetic acid on boron-doped diamond electrodes. J. Electrochem. Soc. 2008, 155 (3), E27–E32. (7) Anglada, A.; Urtiaga, A.; Ortiz, I. Pilot scale performance of the electro-oxidation of landﬁll leachate at boron-doped diamond anodes. Environ. Sci. Technol. 2009, 43 (6), 2035–2040. 10588

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology (8) Bergmann, M. E. H., Drinking water disinfection by in-line electrolysis-product and inorganic by-product formation. In Electrochemistry for the Environment; Comninellis, C., Chen, G., Eds.; Springer: New York, 2009; pp 163205. (9) Bergmann, M. E. H.; Rollin, J.; Iourtchouk, T. The occurrence of perchlorate during drinking water electrolysis using BDD anodes. Electrochim. Acta 2009, 54, 2102–2107. (10) Bergmann, M. E. H.; Rollin, J. Product and by-product formation in laboratory studies on disinfection electrolysis of water using boron-doped diamond anodes. Catal. Today 2007, 124, 198–203. (11) Murata, M.; Ivandini, T. A.; Shibata, M.; Nomura, S.; Fujishima, A.; Einaga, Y. Electrochemical detection of free chlorine at highly borondoped diamond electrodes. J. Electroanal. Chem. 2008, 612 (1), 29–36. (12) Sanchez-Carretero, A.; Saez, C.; Canizares, P.; Rodrigo, M. A. Electrochemical production of perchlorates using conductive diamond electrolyses. Chem. Eng. J. 2011, 166 (2), 710–714. (13) Polcaro, A. M.; Vacca, A.; Mascia, M.; Ferrara, F. Product and by-product formation in electrolysis of dilute chloride solutions. J. Appl. Electrochem. 2008, 38, 979–984. (14) Bergmann, M. E. H.; Iourtchouk, T.; Schmidt, W.; N€uske, G.; Fischer, M., Perchlorate formation in electrochemical water disinfection. In Perchlorates: Production, Uses and Health Eﬀects; Matthews, L. E., Ed.; Nova Science Publishers: Hauppage, NY, 2011; pp 111141. (15) Urbansky, E. T.; Schock, M. R. Issues in managing the risks associated with perchlorate in drinking water. J. Environ. Manage. 1999, 56 (2), 79–95. (16) Urbansky, E. Perchlorate as an environmental contaminant. Environ. Sci. Pollut. Res. 2002, 9 (3), 187–192. (17) Lang, G.; Horanyi, G. Some interesting aspects of the catalytic and electrocatalytic reduction of perchlorate ions. J. Electroanal. Chem. 2003, 552, 197–211. (18) Interim drinking water health advisory for perchlorate; EPA 822R-08-025; U.S. Environmental Protection Agency, Washington, DC, December 2008. (19) Perchlorate in drinking water; R-16-04; California Department of Health Services, Sacramento, CA, 2006. (20) Perchlorate fact sheet for public water suppliers; Massachusetts Department of Environmental Protection, Boston, MA, 2006. (21) Jung, Y. J.; Baek, K. W.; Oh, B. S.; Kang, J. W. An investigation of the formation of chlorate and perchlorate during electrolysis using Pt/Ti electrodes: The eﬀects of pH and reactive oxygen species and the results of kinetic studies. Water Res. 2010, 44 (18), 5345–5355. (22) Munichandraiah, N.; Sathyanarayana, S. Kinetics and mechanism of anodic oxidation of chlorate ion to perchlorate ion on lead dioxide electrodes. J. Appl. Electrochem. 1987, 17 (1), 33–48. (23) Janssen, L. J. J.; Vanderheyden, P. D. L. Mechanism of anodic oxidation of chlorate to perchlorate on platinum-electrode. J. Appl. Electrochem. 1995, 25 (2), 126–136. (24) Granger, M. C.; Swain, G. M. The inﬂuence of surface interactions on the reversibility of ferri/ferrocyanide at boron-doped diamond thin-ﬁlm electrodes. J. Electrochem. Soc. 1999, 146 (12), 4551–4558. (25) Girard, H.; Simon, N.; Ballutaud, D.; Herlem, M.; Etcheberry, A. Eﬀect of anodic and cathodic treatments on the charge transfer of boron doped diamond electrodes. Diamond Relat. Mater. 2007, 16 (2), 316–325. (26) Mishra, D.; Liao, Z. H.; Farrell, J. Understanding reductive dechlorination of trichloroethene on boron-doped diamond ﬁlm electrodes. Environ. Sci. Technol. 2008, 42 (24), 9344–9349. (27) Xu, J.; Granger, M. C.; Chen, Q.; Strojek, J. W.; Lister, T. E.; Swain, G. M. Boron-doped diamond thin-ﬁlm electrodes. Anal. Chem. 1997, 69, 591A–597A. (28) Holt, K. B.; Bard, A. J.; Show, Y.; Swain, G. M. Scanning electrochemical microscopy and conductive probe atomic force microscopy studies of hydrogen-terminated boron-doped diamond electrodes with diﬀerent doping levels. J. Phys. Chem. B 2004, 108 (39), 15117–15127. (29) Martin, H.; Argoitia, A.; Landau, U.; Anderson, A. B.; Angus, J. C. Hydrogen and oxygen evolution on boron-doped diamond electrodes. J. Electrochem. Soc. 1996, 143 (6), L133–L-136.

ARTICLE

(30) Goeting, C. H.; Marken, F.; Gutierrez-Sosa, A.; Compton, R. G.; Foord, J. S. Electrochemically induced surface modiﬁcations of boron-doped diamond electrodes: An X-ray photoelectron spectroscopy study. Diamond Relat. Mater. 2000, 9 (36), 390–396. (31) Simon, N.; Girard, H.; Ballutaud, D.; Ghodbane, S.; Deneuville, A.; Herlem, M.; Etcheberry, A. Eﬀect of H and O termination on the charge transfer of moderately boron doped diamond electrodes. Diamond Relat. Mater. 2005, 14 (37), 1179–1182. (32) Wang, M.; Simon, N.; Decorse-Pascanut, C.; Bouttemy, M.; Etcheberry, A.; Li, M. S.; Boukherroub, R.; Szunerits, S. Comparison of the chemical composition of boron-doped diamond surfaces upon diﬀerent oxidation processes. Electrochim. Acta 2009, 54 (24), 5818–5824. (33) Wang, M.; Simon, N.; Charrier, G.; Bouttemy, M.; Etcheberry, A.; Li, M. S.; Boukherroub, R.; Szunerits, S. Distinction between surface hydroxyl and ether groups on boron-doped diamond electrodes using a chemical approach. Electrochem. Commun. 2010, 12 (3), 351–354. (34) Duo, I.; Fujishima, A.; Comninellis, C. H. Electron transfer kinetics on composite diamond (sp3)-graphite (sp2) electrodes. Electrochem. Commun. 2003, 5 (8), 695–700. (35) Santana, M. H. P.; De Faria, L. A.; Boodts, J. F. C. Electrochemical characterisation and oxygen evolution at a heavily boron doped diamond electrode. Electrochim. Acta 2005, 50 (10), 2017–2027. (36) Martin, H. B.; Argoitia, A.; Angus, J. C.; Landau, U. Voltammetry studies of single-crystal and polycrystalline diamond electrodes. J. Electrochem. Soc. 1999, 146 (8), 2959–2964. (37) Latto, M. N.; Riley, D. J.; May, P. W. Impedance studies of boron-doped CVD diamond electrodes. Diamond Relat. Mater. 2000, 9 (36), 1181–1183. (38) Notsu, H.; Yagi, I.; Tatsuma, T.; Tryk, D. A.; Fujishima, A. Introduction of oxygen-containing functional groups onto diamond electrode surfaces by oxygen plasma and anodic polarization. Electrochem. Solid State 1999, 2 (10), 522–524. (39) Iniesta, J.; Michaud, P. A.; Panizza, M.; Cerisola, G.; Aldaz, A.; Comninellis, C. Electrochemical oxidation of phenol at boron-doped diamond ﬁlm electrode. Electrochim. Acta 2001, 46, 3573–3578. (40) Bard, A. J.; Faulkner, L. R., Electrochemical Methods: Fundamentals and Applications; John Wiley and Sons: New York , 1980; p 718. (41) Bockris, J. O. M.; Reddy, A. K.; Gambra-Aldeco, M. Modern Electrochemistry, 2nd ed.; Plenum Press: New York, 2000; Vol. 2A. (42) Delley, B. An all-electron numerical-method for solving the local density functional for polyatomic-molecules. J. Chem. Phys. 1990, 92, 508–517. (43) Delley, B. From molecules to solids with the DMOL3 approach. J. Chem. Phys. 2000, 113, 7756–7764. (44) Materials Studio, v.4.2; Accelrys Corporation, San Diego, CA, 2007. (45) Delley, B. Fast calculation of electrostatics in crystals and large molecules. J. Phys. Chem. 1996, 100, 6107–6110. (46) Becke, A. Density-functional exchange-energy approximation with correct asymptotic-behavior. Phys. Rev. A 1988, 38 (6), 3098–3100. (47) Lee, C.; Yang, W.; Parr, R. Development of the Colle-Salvetti correlation-energy formula into a functional of the electron-density. Phys. Rev. B 1988, 37 (2), 785–789. (48) Delley, B. Hardness conserving semilocal pseudopotentials. Phys. Rev. B 2002, 66 (15), No. 155125. (49) Delley, B. The conductor-like screening model for polymers and surfaces. Mol. Simul. 2006, 32, 117–123. (50) Anderson, A. B.; Kang, D. B. Quantum chemical approach to redox reactions including potential dependence: Application to a model for hydrogen evolution from diamond. J. Phys. Chem. A 1998, 102 (29), 5993–5996. (51) Wang, J. Analytical Electrochemistry, 3rd ed.; John Wiley and Sons, Inc.: Hoboken, NJ, 2006; p 250. (52) Rao, B.; Anderson, T. A.; Redder, A.; Jackson, W. A. Perchlorate formation by ozone oxidation of aqueous chlorine/oxy-chlorine species: Role of ClxOy radicals. Environ. Sci. Technol. 2010, 44 (8), 2961–2967. (53) Zuo, Z. H.; Katsumura, Y.; Ueda, K.; Ishigure, K. Laser photolysis study on reactions of sulfate radical and nitrate radical with 10589

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

Environmental Science & Technology

ARTICLE

chlorate ion in aqueous solutions - formation and reduction potential of ClO3• radical. J. Chem. Soc., Faraday Trans. 1997, 93 (4), 533–536. (54) Farrell, J.; Martin, F. J.; Martin, H. B.; O’Grady, W. E.; Natishan, P. Anodically generated short-lived species on boron-doped diamond ﬁlm electrodes. J. Electrochem. Soc. 2005, 152 (1), E14–E17. (55) Duo, I.; Levy-Clement, C.; Fujishima, A.; Comninellis, C. Electron transfer kinetics on boron-doped diamond part I: Inﬂuence of anodic treatment. J. Appl. Electrochem. 2004, 34 (9), 935–943. (56) Smith, J. M. Chemical Engineering Kinetics; McGraw-Hill: New York, 1980. (57) Buxton, G. V.; Greenstock, C. L.; Helman, W. P.; Ross, A. B. Critical review of rate constants for reactions of hydrated electrons, hydrogen atoms and hydroxyl radicals in aqueous solution. J. Phys. Chem. Ref. Data 1988, 17, 513–886. (58) George, V. Critical review of rate constants for reactions of hydrated electrons, hydrogen atoms and hydroxyl radicals (HO•/•O) in aqueous solution. J. Phys. Chem. Ref. Data 1988, 17, 513–886. (59) Bergmann, M. E. H. New results of applying diamond electrodes for water hygienization. In Yearbook of Surface Technology; Leuze Electronic: Bad Saulgau, Germany, 2009; pp 317329 (60) Perez, G.; Fernandez-Alba, A. R.; Urtiaga, A. M.; Ortiz, I. Electro-oxidation of reverse osmosis concentrates generated in tertiary water treatment. Water Res. 2010, 44 (9), 2763–2772.

10590

dx.doi.org/10.1021/es202534w |Environ. Sci. Technol. 2011, 45, 10582–10590

ARTICLE pubs.acs.org/est

Effect of Diesel Oxidation Catalysts on the Diesel Particulate Filter Regeneration Process Leonardo Lizarraga,* Stamatios Souentie, Antoinette Boreave, Christian George, Barbara D’Anna, and Philippe Vernoux Universite de Lyon, Institut de Recherches sur la Catalyse et l’Environnement de Lyon, UMR 5256, CNRS, Universite Claude Bernard Lyon 1, 2 Avenue A. Einstein, 69626 Villeurbanne, France ABSTRACT: A Diesel Particulate Filter (DPF) regeneration process was investigated during aftertreatment exhaust of a simulated diesel engine under the inﬂuence of a Diesel Oxidation Catalyst (DOC). Aerosol mass spectrometry analysis showed that the presence of the DOC decreases the Organic Carbon (OC) fraction adsorbed to soot particles. The activation energy values determined for soot nanoparticles oxidation were 97 ( 5 and 101 ( 8 kJ mol1 with and without the DOC, respectively; suggesting that the DOC does not facilitate elementary carbon oxidation. The minimum temperature necessary for DPF regeneration was strongly aﬀected by the presence of the DOC in the aftertreatment. The conversion of NO to NO2 inside the DOC induced the DPF regeneration process at a lower temperature than O2 (ΔT = 30 K). Also, it was veriﬁed that the OC fraction, which decreases in the presence of the DOC, plays an important role to ignite soot combustion.

’ INTRODUCTION Diesel particulate matter (PM) signiﬁcantly contributes to urban air pollution and has often been associated with adverse health eﬀects.1,2 Furthermore, particles from combustion processes may also inﬂuence the radiation budget of the atmosphere and therefore the climate 3 and the hydrogeological cycle.4 In particular, diesel engine emissions constitute a major source of ultraﬁne particles in urban environments.57 Diesel exhaust PM is a complex mixture of carbonaceous material and hundreds of combustion products. This complex composition highly depends on the engine operation, fuel composition, lubriﬁcation oil aftertreatment technology, and exhaust sampling procedure. Exhaust pipes of modern diesel engines present particle size distributions centered at ∼80 nm with a typical number concentration close to 2.0 107 # cm3.8 Total carbon (TC) of diesel PM consists of elemental and organic carbon (EC and OC). EC is formed during fuel pyrolysis, and it is principally graphitic carbon. OC is originated from incomplete fuel combustion and slip of lubrication oil past engine seals; and it primarily consists of polycyclic aromatic hydrocarbons (PAHs) and aliphatic hydrocarbons.6,9 PAHs are harmful substances for human health due to their high carcinogenic, mutagenic, and allergenic potential;2 more than 100 diﬀerent PAHs have been found in PM. Indeed, soot particles can transport genotoxic compounds across the cellular membrane acting as a Trojan horse.10,11 Diesel particulate ﬁlters (DPFs) are considered as the key technology to detoxify the diesel exhaust, reducing more than 95% the number of particles emitted. These ﬁlters consist of ceramic monoliths with alternating ﬂow channels, which are closed at the end to force the exhaust ﬂow to go through the porous wall of the honeycomb ﬁlter.12,13 Soot accumulation r 2011 American Chemical Society

inside the DPF increases the pressure drop in the aftertreatment. Thus, a periodical regeneration process is required. The optimal way to regenerate DPF is to oxidize the carbonaceous compounds to CO2 and H2O. However, the temperature needed for PM oxidation is elevated, commonly exceeding 773 K.14,15 Diesel oxidation catalysts (DOC) are also used during diesel emission aftertreatment; they consist of honeycomb monoliths impregnated with an active noble metal such as Pt and/or Pd. They are used to optimize the maximum conversion of hydrocarbons and carbon monoxide.16,17 It is well reported that NO2 is more eﬃcient than O2 for soot oxidation, and NO2 can oxidize soot at temperatures as low as 523 K.1823 For the latter, a continuous regenerating trap (CRT) is one of the technologies used for promoting the DPF regeneration,13,23,24 which is composed of two parts, a DOC followed by a DPF. In the DOC, the exhaust temperature is increased by hydrocarbon oxidation, and NO2 is produced by oxidation of NO generated in the engine. The NO2 is then able to oxidize the soot deposited on a DPF to CO2 but with elevated concomitant emission of NO. Recently, secondary eﬀects of including a DOC and/or DPF in the diesel engine afterteatment have been extensively studied.10,14,2533 These works have principally focused on possible reactions occurring in the DPF and/or DOC due to reactant accumulation in the DPF, elevated temperature, and elongated residence time; e.g., formation of nitro-PAHs, which Received: July 27, 2011 Accepted: November 3, 2011 Revised: October 26, 2011 Published: November 03, 2011 10591

dx.doi.org/10.1021/es2026054 | Environ. Sci. Technol. 2011, 45, 10591–10597

Environmental Science & Technology

ARTICLE

Figure 1. Experimental design to simulate the diesel engine aftertreatment and the analysis system.

have much higher mutagenicity and carcinogenic properties than PAHs.31,32 Furthermore, recent works have reported the emission of nanoparticles with size distribution below 20 nm during DPF regeneration.14,33 The main objective of this work was to investigate the eﬀect of a DOC on the DPF regeneration temperature procedure, using a simulated diesel engine aftertreatment. The novel aspect of this study is the correlation of the chemical composition analysis of the nonrefractory fraction of soot particles, with the oxidizing capacity of NO2 produced in the DOC on the DPF regeneration process.

’ EXPERIMENTAL SECTION Experimental Setup. Figure 1 shows the experimental design used to characterize the exhaust of a simulated diesel engine. The setup consisted of a soot generator; a proxy for diesel vehicles aftertreatment; and a series of instruments to analyze PM and gas phase in the stream. A mini Combustion Aerosol Standard (miniCAST, Jing Ltd. Switzerland) was used as soot generator. Additional gases were injected into the soot generated flow to mimic the diesel engine exhaust. The total flow rate of the stream was 20 L min1 using N2 as carrier gas, and the initial concentrations of the gas phase compounds were 500 ppmv NO, 1000 ppmv propene (C3H6), 1000 ppmv propane (C3H8), 9.0% v/v CO2, and 10% v/v O2 for all experiments unless differently indicated. The hydrocarbon concentration was elevated compared to a diesel exhaust to highlight the impact of DOC performance on soot emission. The total flow was preheated using a heating pipeline, which allowed temperature increase up to 873 K. A stainless steel reactor was used to fix a conventional DOC and a commercial mini DPF, and it was placed in a tubular furnace. The temperatures of the inlet and outlet gas stream (Tinlet and Toutlet) were measured using two thermocouples chromel alumel (type K). These thermocouples were located inside the reactor just before the DOC (Tinlet) and just after the DPF (Toutlet), and both were in contact with the gas stream. The pressure drop values between inlet and the outlet of the reactor were measured using a pressure sensor (Keller). The DOC was composed of a cylindrical cordierite monolith with 2.5 cm length and 2.5 cm diameter. The catalytically active metal phase in the DOC was a mixture of Pt and Pd with a metal ratio equal to 1 and a total metal loading of 3.2 g L1. This type of

DOC is highly eﬀective in hydrocarbons and carbon monoxide oxidation at low temperatures. The mini DPFs used (IBIDEN, Japan) were composed of silicon carbide, and a cell density of 400 cpsi, with 5 cm length and 2.5 cm diameter. To study the eﬀect of a DOC in the aftertreatment on particle characterization, the DPF was replaced by a silicon carbide monolith with open channels to allow PM circulation in the system, which was analyzed by scanning mobility particle sizer (SMPS) and aerosol mass spectrometry (AMS). The monolith was the same dimensions as the DPF to equal its dead volume inside the reactor; avoiding homogeneous or noncatalyzed and/ or nucleation processes. The concentration of the diﬀerent compounds in the gas stream were analyzed online using a micro-GC (R 3000 SRA) for C3H8, C3H6, O2, CO, and CO2; and a chemiluminescence’s detector for NO, NO2 and NOx (ECO Physics CLD 62). As shown in Figure 1, a DPF was placed before the micro-CG and the chemiluminescence’s detector to ﬁltrate PM. c-TOF-AMS. A compact time-of-flight (c-TOF, Tofwerk) Aerodyne Aerosol Mass Spectrometer (AMS) was used to analyze the non-refractory fraction of combustion PM. The methodology used within the C-TOF AMS is fully described in Drewnick et al.34 The soot particles were sampled through a critical orifice of 100 μm ID at 80 mL min1. An aerodynamic lens focuses particles as a narrow beam into the vacuum chamber, where they are accelerated to a terminal velocity which depends on their aerodynamic size. A mechanical wheel with two radial slits is used for chopping the particle beam. After traveling through the particle-sizing chamber, the particles enter the evaporation and ionization chamber. The nonrefractory aerosol components are flash-vaporized on a hot surface (∼873 K) and ionized by 70 eV electrons emitted from a tungsten filament. Positive charged ions are transferred into the time-of-flight mass spectrometer, collecting the complete mass spectra as a function of particle time-offlight. Mass spectra scans were obtained from m/z 4 to 400. The data were acquired in MS (mass spectrum) and P-TOF (Particle Time-of-Flight) modes. The MS and P-TOF modes were run for cycles of 20 s. Signals from soot particles and from gas phase background were acquired for 10 s in each mode. The MS is the result of the difference between the MS obtained with open and close chopper. Spectra were averaged and saved with a frequency of 3 cycles. The following controls were performed at the beginning of the experiments: baseline check; single ion calibration; 10592

dx.doi.org/10.1021/es2026054 |Environ. Sci. Technol. 2011, 45, 10591–10597

Environmental Science & Technology

ARTICLE

Table 1. Reactor Temperature, Hydrocarbons Conversions and NO2 Formation with and without the DOC Tinlet (K) Toutlet (K) XC3H8 (%) XC3H6 (%) NO2 (ppmv) without DOC

553

573

0

0

0

with DOC

553

653

13

100

60

m/z calibration for nominal masses 32, 28, 40, and 184; ion efficiency calibration for NH4NO3; and size calibration using polystyrene latex spheres. SQUIRREL data analysis package 34 was used to process mass spectra. Scanning Mobility Particle Sizer. Soot particle size distribution was measured using a SMPS (Model 3080, TSI Inc., St.Paul, MN, USA) with a differential mobility analyzer (DMA) (TSI, Model 3081) column and condensation particle counter (CPC) (TSI, Model 3772). It was operated with a sheath flow of 4 L min1; and an aerosol flow of 0.3 L min1. The particle diameter scanning was from 12.4 to 562 nm, and each scan required 3 min. Soot Generator. The miniCAST soot generator produces soot particles with physical and chemical properties similar to those of diesel engines. It allows size control and chemical composition in a wide range.3537 The generator uses a laminar diffusion flame, and consists of an inner burner gas flow, and an outer sheath air flow. Oxygen is transported by diffusion from the sheath flow, oxidizing the fuel gas. The flame is “cut open” at halfway to its tip, using N2 (Linde Gaz, 99.9995%) for quenching the reaction and stabilizing the formed soot particles. Propane (Air Liquide, 99.9995%) was used as fuel, and filtered compressed air was used as sheath gas flow. The miniCAST was operated at a propane flow of 0.06 L min1, and an air sheath flow of 1.55 L min1. The N2 flow for quenching the flame was 7.5 L min1. After soot particle formation, an air dilution flow of 6.2 L min1 was fixed inside the miniCAST to obtain a total flow of 15 L min1 at the outlet. Under these operating conditions, the miniCAST produces 400 mg of soot per m3 and 2.5 1014 particles number per cm3, and presents a c-TOF-AMS spectrum and a size particle distribution similar to those given by diesel engines.8,38 Thus, soot generated by the miniCAST is a good proxy for diesel PM. By means of AMS, Ferge et al. 35 reported a similar EC/TC ratio (∼0.95) for PM emitted from a diesel engine and from a mini CAST under similar conditions to those used in the present work. In contrast, classical thermal or thermaloptical techniques 39 do not allow estimating low OC concentrations adsorbed on soot particles.

’ RESULTS AND DISCUSSION Chemical Composition of Soot Particles. Mass spectra of soot particles were obtained at two different aftertreatment configurations; i.e., with and without a DOC, to analyze the effect of the DOC on the chemical composition of the PM. As described in the Experimental Section, a nonplugged monolith was fixed in the reactor downstream the DOC to allow soot particles analysis at the SMPS and AMS. The preheating pipeline and the furnace temperatures were set to Tinlet = 553 K in both aftertreatment configurations. Table 1 displays Toutlet, C3H8 and C3H6 conversions (XC3H8 and XC3H6), and NO2 concentration values with and without the DOC. XC3H6 expressed in % is defined as ([C3H6]initial [C3H6]outlet)/ [C3H6]initial 100; an analogue expression was used to calculate XC3H8. As shown in Table 1, in presence of the DOC the hydrocarbons were partially or completely oxidized while NO2 was possibly formed via NO

Figure 2. Mass spectra acquired with the c- TOF-AMS. Blue line mass spectrum with the DOC; red line mass spectrum without the DOC.

oxidation. As well, the Toutlet increase was due to the exothermicity of the hydrocarbons combustion reactions. Indeed, a thermodynamic calculation considering XC3H8 and XC3H6 values and the total flow of 20 L min1 predicts ∼80 K temperature increase with the DOC, in agreement with the temperature increase of Toutlet. The mass spectra obtained with the AMS for various organic species may show slight greater fragmentation than standard electron impact spectra due to the higher internal energy acquired during vaporization of the aerosol.40 However, the PAHs molecular ion signals are observed with signiﬁcant intensity because aromatic rings are very resistant to fragmentation.40 Thus, PAHs can be used as indicative species of the condensed OC content on soot. Figure 2 shows two c-TOF-AMS mass spectra obtained for soot particles with and without the DOC. Only m/z values between 200 and 400 are shown, which is the range relevant to PAHs molecular ions. Signals with m/z 207, 221, 281, and 295 were due to polydimethylsiloxane contamination by the silicone tubing used in pipeline connections.41 The signals of PAHs molecular ion nominal masses with even carbon number in their structure are signaled with arrows in Figure 2. These ions are more abundant and stable than those with odd number of carbons.40 The signals indicated with the arrows were more intense in the mass spectrum obtained without the DOC than those obtained using the DOC in the reactor. Besides, the estimation of total organic mass loading values were 18 ( 1 and 45 ( 1 μg m3 with and without the DOC, respectively. These results clearly indicate a decrease in the amount of organic compounds in PM due to the DOC inclusion, which is in agreement with previous studies.10,27,29,30 In those studies, however, PM was collected onto ﬁlters and then OC and/or PAHs were extracted with diﬀerent methods and analyzed by chromatographic techniques. Recently, Chirico et al. 42 have reported a decrease of OC concentration on PM from a real diesel engine in the presence of the DOC using AMS. It is worth noting that only the ion signal at m/z 295 (inset Figure 2) was more intense without the DOC; such an ion may be assigned to NO2PAH (e.g., NO2benzo(k)ﬂuranthene or NO2benzopyrene) formation, via reaction between PAHs and NO2. Heeb et al. 10 have found NO2PAHs production in presence of oxidative catalysts in the aftertreatment. Effect of the DOC on Elemental Carbon Oxidation. Experiments on the effect of the DOC on soot oxidation were also performed without the DPF. Soot combustion was followed via variation of particle size evolution with temperature. The particle 10593

dx.doi.org/10.1021/es2026054 |Environ. Sci. Technol. 2011, 45, 10591–10597

Environmental Science & Technology

ARTICLE

Figure 4. Dependence of ΔP on Toutlet during the DPF regeneration (a) without the DOC and regeneration with air; (b) with the DOC and regeneration with air; and (c) with the DOC and regeneration with reactive mixture.

Figure 3. Kinetic experiments to calculate Ea for soot combustion using the SMPS. (a) with the DOC; (b) without the DOC.

size distribution was measured at different temperatures inside the reactor using a SMPS with and without the DOC upstream the nonplugged honeycomb monolith. The rate of size decrease was modeled using a modified Arrhenius expression, which was introduced by Higgins et al. for the study of soot particles:43,44 T0 Ea ΔDp ¼ Aθ0 1=2 exp ð1Þ RT T where ΔDp is the change of the mode diameter value in the particle size distribution, A is the pre-exponential factor, T0 is 298 K, θ0 is the residence time in the reactor (6.7 s) at T0, Ea is the activation energy, and T is the Tout of the reactor. Figure 3 shows the size distribution of soot particles with (a) and without (b) the DOC for diﬀerent values of Tout. Shrivastava et al. 22 showed that at 373 K, neither soot oxidation nor evaporation of strongly adsorbed nonvolatile OC occurs. Thus, the initial particle diameter used in this work to calculate ΔDp was the mode diameter of the particles size distribution at 373 K, which was equal to 88 nm. No change of the soot particles size distribution was detected at temperatures below 783 K. However, at higher temperatures, the particle size distribution shifted to lower values, reaching a mode value of 70 nm at 858 K. As the EC/TC ratio in the PM generated from the CAST was near 0.95;

changes in the size distribution due to evaporation and oxidation of OC, occurring between 473 and 673 K,4345 can be assumed negligible. Consequently, the decrease of particle size above 773 K was attributed to EC oxidation. The insets of Figure 3 show the ΔDp values as a function of T, from which Ea can be calculated using eq 1. Moreover, NO2 formation was detected in presence of the DOC and its concentration was practically constant at 60 ppmv in the Tout range 773853 K. The ﬁtting was performed using the nonlinear LevenbergMarquard algorithm. The Ea values were 97 ( 5 and 101 ( 8 kJ mol1 for the soot oxidation reaction with and without the DOC, respectively. These Ea values are in concordance with the reported values for soot oxidation with air,44,47 while they are twice the values reported by Shrivasrava et al. 22 for soot oxidation with NO2. Thus, the use of the DOC did not facilitate the EC oxidation on the soot particles in the gas stream at low temperatures. This can be probably explained by the short residence time of particles in the reactor. Effect of the DOC on the DPF Regeneration. The DPF regeneration process was investigated with and without a DOC. As mentioned in the Experimental Section, the DPF was placed downstream the DOC (Figure 1). Initially, a step of soot charging in the DPF was performed for 1 h under a total flow rate of 20 L min1 (with initial concentrations as described in detail in the Experimental setup section) to accumulate 250 mg of soot, equivalent to 10 g of soot per L of DPF. The latter is in agreement with the soot concentration inside a DPF before the regeneration process is performed in a real diesel aftertreatment.48,49 During this step, no particles were detected with the SMPS at the outlet gas stream, indicating the excellent filtering efficiency of the DPF. The DPF regeneration process was performed after the soot charging step under a total gas flow rate of 10 L min1 using either (i) compressed air or (ii) a reactive mixture of the same concentration as the DPF charging flow (500 ppmv NO, 1000 ppmv C3H6, 1000 ppmv C3H8, 9.0% v/v CO2, 10.0% v/v O2 and N2 as carrier). In the regeneration process, a constant heating rate of 10 °C min1 was used from 573 to 973 K. Figure 4 shows the dependence of the pressure drop (ΔP) on Toutlet during the DPF regeneration process for three diﬀerent conﬁgurations, using compressed air with (a) and without (b) DOC, and using the reactive gas mixture with the DOC (c). 10594

dx.doi.org/10.1021/es2026054 |Environ. Sci. Technol. 2011, 45, 10591–10597

Environmental Science & Technology

ARTICLE

the kinetic experiments (without the soot accumulation step) shown in Figure 3. Summarizing, the presence of the DOC in the aftertreatment induces two opposite eﬀects on the DPF regeneration: (I) depletion of OC fraction on the soot accumulated inside the DPF, and (II) NO2 formation. According to our knowledge, this is the ﬁrst study that combines diﬀerent approaches to investigate the inﬂuence of a DOC in the DPF regeneration process.

’ AUTHOR INFORMATION Corresponding Author

*Tel: +33 472 431 054; Fax: +33 472 431 695; E-mail: [email protected]. Figure 5. Dependence of ΔP and the NO2 formation yield (YNO2) on Toutlet during the DPF regeneration with the DOC under reactive mixture feed.

The temperature associated to the maximum of ΔP indicated that the DPF regeneration process was already initiated. This temperature was used to compare the diﬀerent conﬁgurations and it was addressed as critical temperature (Tc). As shown in Figure 4, in the case where only air was fed, a signiﬁcant increase in Tc was observed from 813 to 873 K in the presence of DOC. This increase may be rationalized by a diﬀerence in the composition of the soot particles stored in the DPF during the charging step due to the DOC. This is partly conﬁrmed from a decrease of the organic mass loading and PAHs (Figure 2) observed when the DOC is used upstream the monolith. It is worth noting that the temperature at the reactor outlet, Toutlet, in presence of the DOC during charging was ∼80 K higher than without the DOC. This is possibly due to the exothermicity of the catalytic combustion of the cofeeded hydrocarbons. Furthermore, NO2 formation was detected (see Table 1). Thus, the higher operation temperature and NO2 presence could result in pronounced evaporation or oxidation of the PAHs, or more generally OC, from the soot particles. Hence, the OC content of the soot particles stored in the DPF would be signiﬁcantly lower, as shown by the AMS data. The OC can facilitate the ignition of the combustion of the black carbon nucleus,46 providing the necessary heat by the exothermicity of their combustion reaction. When the gas mixture was used for the regeneration, the measured Tc (783 K) was lower than without the DOC and feeding only air (813 K). The remarkable Tc diﬀerence (783 K vs 883 K) between the two cases where the DOC was utilized (a,c) can suggest a diﬀerent soot oxidation mechanism. Figure 5 presents the eﬀect of temperature on ΔP in the DPF and the NO2 formation yield during DPF regeneration under feed conditions similar to those during the soot charging step in presence of the DOC. As shown in the ﬁgure, at temperatures below 623 K a signiﬁcant NO2 production yield (∼30%) was observed, which decreased to lower than 5% at higher temperatures. The latter is more likely related to the reduction of NO2 to NO by soot oxidation 18,20,23,50 than to thermodynamic equilibrium limitations. As mentioned above, NO2 is known as a stronger oxidizing agent than O2 for soot oxidation. In this case, the obtained Tc for soot combustion was 783 K, which overbalances the observed negative eﬀect of the DOC in Figure 4, when the regeneration process was performed with air. Soot accumulation inside the DPF allowed EC oxidation by NO2 generated in the DOC, in contrast to the results obtained from

’ ACKNOWLEDGMENT The authors thank Gilles D’Orazio and Frederic Bourgain from the technical service of IRCELYON. The authors acknowledge PSA Peugeot-Citro€en to provide the DOC used in the study. The authors are also grateful to the “ADEME PIREP” project (Contract No. 06-66-C0138 from PREDIT 3 VPE 2006 program) for the funding of this study. ’ REFERENCES (1) WHO:Guidelines for Air Quality. World Heath Organization; 2005. (2) International Agency for Research on Cancer (IARC) Air pollution, part 1, some non-heterocyclic polycyclic aromatic hydrocarbons and some related industrial exposures. In: Monographs on the Evaluation of Carcinogenic Risks to Humans; Lyon, 2008. (3) Jacobson, M. Z. Strong radiative heating due to the mixing state of black carbon in atmospheric aerosols. Nature 2001, 409 (6821), 695–697. (4) Lohmann, U.; Feichter, J. Global indirect aerosol eﬀects: a review. Atmos. Chem. Phys. 2005, 5, 715–737. (5) Morawska, L.; Ristovski, Z.; Jayaratne, E. R.; Keogh, D. U.; Ling, X. Ambient nano and ultraﬁne particles from motor vehicle emissions: Characteristics, ambient processing and implications on human exposure. Atmos. Environ. 2008, 42 (35), 8113–8138. (6) Kittelson, D. B. Engines and nanoparticles: A review. J. Aerosol. Sci. 1998, 29 (56), 575–588. (7) Querol, X.; Alastuey, A.; Ruiz, C. R.; Artinano, B.; Hansson, H. C.; Harrison, R. M.; Buringh, E.; ten Brink, H. M.; Lutz, M.; Bruckmann, P.; Straehl, P.; Schneider, J. Speciation and origin of PM10 and PM2.5 in selected European cities. Atmos. Environ. 2004, 38 (38), 6547–6555. (8) Barrios, C. C.; Dominguez-Saez, A.; Rubio, J. R.; Pujadas, M. Development and evaluation of on-board measurement system for nanoparticle emissions from diesel engine. Aerosol Sci. Technol. 2011, 45 (5), 570–580. (9) Maricq, M. M. Chemical characterization of particulate emissions from diesel engines: A review. J. Aerosol. Sci. 2007, 38 (11), 1079–1118. (10) Heeb, N. V.; Schmid, P.; Kohler, M.; Gujer, E.; Zennegg, M.; Wenger, D.; Wichser, A.; Ulrich, A.; Gfeller, U.; Honegger, P.; Zeyer, K.; Emmenegger, L.; Petermann, J. L.; Czerwinski, J.; Mosimann, T.; Kasper, M.; Mayer, A. Impact of low- and high-oxidation diesel particulate ﬁlters on genotoxic exhaust constituents. Environ. Sci. Technol. 2010, 44 (3), 1078–1084. (11) Rothen-Rutishauser, B. M.; Schurch, S.; Haenni, B.; Kapp, N.; Gehr, P. Interaction of ﬁne particles and nanoparticles with red blood cells visualized with advanced microscopic techniques. Environ. Sci. Technol. 2006, 40 (14), 4353–4359. (12) Liati, A.; Dimopoulos Eggenschwiler, P. Characterization of particulate matter deposited in diesel particulate ﬁlters: Visual and analytical approach in macro-, micro-, and nano-scales. Combust. Flame 2010, 157 (9), 1658–1670. 10595

dx.doi.org/10.1021/es2026054 |Environ. Sci. Technol. 2011, 45, 10591–10597

Environmental Science & Technology (13) van Setten, B.; Makkee, M.; Moulijn, J. A. Science and technology of catalytic diesel particulate ﬁlters. Catal. Rev.-Sci. Eng. 2001, 43 (4), 489–564. (14) Dwyer, H.; Ayala, A.; Zhang, S.; Collins, J.; Huai, T.; Herner, J.; Chau, W. Emissions from a diesel car during regeneration of an active diesel particulate ﬁlter. J. Aerosol. Sci. 2010, 41 (6), 541–552. (15) Caroca, J. C.; Millo, F.; Vezza, D.; Vlachos, T.; De Filippo, A.; Bensaid, S.; Russo, N.; Fino, D. Detailed investigation on soot particle size distribution during DPF regeneration, using standard and bio-diesel fuels. Ind. Eng. Chem. Res. 2011, 50 (5), 2650–2658. (16) Farrauto, R. J.; Voss, K. E. Monolithic diesel oxidation catalysts. Appl. Catal. B-Environ. 1996, 10 (13), 29–51. (17) Stein, H. J. Diesel oxidation catalysts for commercial vehicle engines: Strategies on their application for controlling particulate emissions. Appl. Catal. B-Environ. 1996, 10 (13), 69–82. (18) Jeguirim, M.; Tschamber, V.; Ehrburger, P. Catalytic eﬀect of platinum on the kinetics of carbon oxidation by NO2 and O2. Appl. Catal. B: Environ. 2007, 76 (34), 235–240. (19) Kandylas, I. P.; Haralampous, O. A.; Koltsakis, G. C. Diesel soot oxidation with NO2: Engine experiments and simulations. Ind. Eng. Chem. Res. 2002, 41 (22), 5372–5384. (20) Schejbal, M.; Stepanek, J.; Marek, M.; Koci, P.; Kubicek, M. Modelling of soot oxidation by NO2 in various types of diesel particulate ﬁlters. Fuel 2010, 89 (9), 2365–2375. (21) Stanmore, B. R.; Tschamber, V.; Brilhac, J. F. Oxidation of carbon by NOx, with particular reference to NO2 and N2O. Fuel 2008, 87 (2), 131–146. (22) Shrivastava, M.; Nguyen, A.; Zheng, Z. Q.; Wu, H. W.; Jung, H. S. Kinetics of soot oxidation by NO2. Environ. Sci. Technol. 2010, 44 (12), 4796–4801. (23) Setiabudi, A.; Makkee, M.; Moulijn, J. A. The role of NO2 and O2 in the accelerated combustion of soot in diesel exhaust gases. Appl. Catal. B: Environ. 2004, 50 (3), 185–194. (24) Mohr, M.; Forss, A. M.; Lehmann, U. Particle emissions from diesel passenger cars equipped with a particle trap in comparison to other technologies. Environ. Sci. Technol. 2006, 40 (7), 2375–2383. (25) Carrara, M.; Wolf, J. C.; Niessner, R. Nitro-PAH formation studied by interacting artiﬁcially PAH-coated soot aerosol with NO2 in the temperature range of 295523 K. Atmos. Environ. 2010, 44 (32), 3878–3885. (26) Heeb, N. V.; Schmid, P.; Kohler, M.; Gujer, E.; Zennegg, M.; Wenger, D.; Wichser, A.; Ulrich, A.; Gfeller, U.; Honegger, P.; Zeyer, K.; Emmenegger, L.; Petermann, J.-L.; Czerwinski, J.; Mosimann, T.; Kasper, M.; Mayer, A. Secondary eﬀects of catalytic diesel particulate ﬁlters: Conversion of PAHs versus formation of Nitro-PAHs. Environ. Sci. Technol. 2008, 42 (10), 3773–3779. (27) Liu, Z. G.; Berg, D. R.; Swor, T. A.; Schauer, J. J. Comparative analysis on the eﬀects of diesel particulate ﬁlter and selective catalytic reduction systems on a wide spectrum of chemical species emissions. Environ. Sci. Technol. 2008, 42 (16), 6080–6085. (28) Ratcliﬀ, M. A.; Dane, A. J.; Williams, A.; Ireland, J.; Luecke, J.; McCormick, R. L.; Voorhees, K. J. Diesel particle ﬁlter and fuel eﬀects on heavy-duty diesel engine emissions. Environ. Sci. Technol. 2010, 44 (21), 8343–8349. (29) Vaaraslahti, K.; Ristim€aki, J.; Virtanen, A.; Keskinen, J.; Giechaskiel, B.; Solla, A. Eﬀect of oxidation catalysts on diesel soot particles. Environ. Sci. Technol. 2006, 40 (15), 4776–4781. (30) Wenger, D.; Gerecke, A. C.; Heeb, N. V.; Zennegg, M.; Kohler, M.; Naegeli, H.; Zenobi, R. Secondary eﬀects of catalytic diesel particulate ﬁlters: Reduced aryl hydrocarbon receptor-mediated activity of the exhaust. Environ. Sci. Technol. 2008, 42 (8), 2992–2998. (31) Bamford, H. A.; Bezabeh, D. Z.; Schantz, M. M.; Wise, S. A.; Baker, J. E. Determination and comparison of nitrated-polycyclic aromatic hydrocarbons measured in air and diesel particulate reference materials. Chemosphere 2003, 50 (5), 575–587. (32) Murakami, M.; Yamada, J.; Kumata, H.; Takada, H. Sorptive behavior of nitro-PAHs in street runoﬀ and their potential as indicators of diesel vehicle exhaust particles. Environ. Sci. Technol. 2008, 42 (4), 1144–1150.

ARTICLE

(33) Cauda, E.; Hernandez, S.; Fino, D.; Saracco, G.; Specchia, V. PM0.1 emissions during diesel trap regeneration. Environ. Sci. Technol. 2006, 40 (17), 5532–5537. (34) Drewnick, F.; Hings, S. S.; DeCarlo, P.; Jayne, J. T.; Gonin, M.; Fuhrer, K.; Weimer, S.; Jimenez, J. L.; Demerjian, K. L.; Borrmann, S.; Worsnop, D. R. A new time-of-ﬂight aerosol mass spectrometer (TOF-AMS)—Instrument description and ﬁrst ﬁeld deployment. Aerosol Sci. Technol. 2005, 39 (7), 637–658. (35) Ferge, T.; Karg, E.; Schroppel, A.; Coﬀee, K. R.; Tobias, H. J.; Frank, M.; Gard, E. E.; Zimmermann, R. Fast determination of the relative elemental and organic carbon content of aerosol samples by online single-particle aerosol time-of-ﬂight mass spectrometry. Environ. Sci. Technol. 2006, 40 (10), 3327–3335. (36) Giechaskiel, B.; Wang, X.; Horn, H. G.; Spielvogel, J.; Gerhart, C.; Southgate, J.; Jing, L.; Kasper, M.; Drossinos, Y.; Krasenbrink, A. Calibration of condensation particle counters for legislated vehicle number emission measurements. Aerosol Sci. Technol. 2009, 43 (12), 1164–1173. (37) Slowik, J. G.; Stainken, K.; Davidovits, P.; Williams, L. R.; Jayne, J. T.; Kolb, C. E.; Worsnop, D. R.; Rudich, Y.; DeCarlo, P. F.; Jimenez, J. L. Particle morphology and density characterization by combined mobility and aerodynamic diameter measurements. Part 2: Application to combustion-generated soot aerosols as a function of fuel equivalence ratio. Aerosol Sci. Technol. 2004, 38 (12), 1206–1222. (38) Canagaratna, M. R.; Jayne, J. T.; Ghertner, D. A.; Herndon, S.; Shi, Q.; Jimenez, J. L.; Silva, P. J.; Williams, P.; Lanni, T.; Drewnick, F.; Demerjian, K. L.; Kolb, C. E.; Worsnop, D. R. Chase studies of particulate emissions from in-use New York City vehicles. Aerosol Sci. Technol. 2004, 38 (6), 555–573. (39) Kirchstetter, T. H. Novakov, Controlled generation of black carbon particles from diﬀusion ﬂame and applicants in evaluating balck carbon measurement methods. Atmos. Environ. 2007, 41 (9), 1874– 1888. (40) Dzepina, K.; Arey, J.; Marr, L. C.; Worsnop, D. R.; Salcedo, D.; Zhang, Q.; Onasch, T. B.; Molina, L. T.; Molina, M. J.; Jimenez, J. L. Detection of particle-phase polycyclic aromatic hydrocarbons in Mexico City using an aerosol mass spectrometer. Int. J. Mass Spectrom. 2007, 263 (23), 152–170. (41) Timko, M. T.; Yu, Z.; Kroll, J.; Jayne, J. T.; Worsnop, D. R.; Miake-Lye, R. C.; Onasch, T. B.; Liscinsky, D.; Kirchstetter, T. W.; Destaillats, H.; Holder, A. L.; Smith, J. D.; Wilson, K. R. Sampling artifacts from conductive silicone tubing. Aerosol Sci. Technol. 2009, 43 (9), 855–865. (42) Chirico, R.; DeCarlo, P. F.; Heringa, M. F.; Tritscher, T.; Richter, R.; Prevot, A. S. H.; Dommen, J.; Weingartner, E.; Wehrle, G.; Gysel, M.; Laborde, M.; Baltensperger, U. Impact of aftertreatment devices on primary emissions and secondary organic aerosol formation potential from in-use diesel vehicles: results from smog chamber experiments. Atmos. Chem. Phys. 2010, 10 (23), 11545– 11563. (43) Higgins, K. J.; Jung, H.; Kittelson, D. B.; Roberts, J. T.; Zachariah, M. R. Size-selected nanoparticle chemistry: Kinetics of soot oxidation. J. Phys. Chem. A 2002, 106 (1), 96–103. (44) Higgins, K. J.; Jung, H.; Kittelson, D. B.; Roberts, J. T.; Zachariah, M. R. Kinetics of diesel nanoparticle oxidation. Environ. Sci. Technol. 2003, 37 (9), 1949–1954. (45) Schneider, J.; Hock, N.; Weimer, S.; Borrmann, S.; Kirchner, U.; Vogt, R.; Scheer, V. Nucleation particles in diesel exhaust: Composition inferred from in situ mass spectrometric analysis. Environ. Sci. Technol. 2005, 39 (16), 6153–6161. (46) Giechaskiel, B.; Chirico, R.; DeCarlo, P. F.; Clairotte, M.; Adam, T.; Martini, G.; Heringa, M. F.; Richter, R.; Prevot, A. S. H.; Baltensperger, U.; Astorga, C. Evaluation of the particle measurement programme (PMP) protocol to remove the vehicles’ exhaust aerosol volatile phase. Sci. Total Environ. 2010, 408 (21), 5106–5116. (47) Stanmore, B. R.; Brilhac, J. F.; Gilot, P. The oxidation of soot: a review of experiments, mechanisms and models. Carbon 2001, 39 (15), 2247–2268. 10596

dx.doi.org/10.1021/es2026054 |Environ. Sci. Technol. 2011, 45, 10591–10597

Environmental Science & Technology

ARTICLE

(48) Chen, K.; Martirosyan, K. S.; Luss, D. Temperature gradients within a soot layer during DPF regeneration. Chem. Eng. Sci. 2011, 66 (13), 2968–2973. (49) Liati, A.; Eggenschwiler, P. D. Characterization of particulate matter deposited in diesel particulate ﬁlters: Visual and analytical approach in macro-, micro- and nano-scales. Combust. Flame 2010, 157 (9), 1658–1670. (50) Atribak, I.; Bueno-Lopez, A.; García-García, A. Uncatalysed and catalysed soot combustion under NOx + O2: Real diesel versus model soots. Combust. Flame 2010, 157 (11), 2086–2094.

10597

dx.doi.org/10.1021/es2026054 |Environ. Sci. Technol. 2011, 45, 10591–10597

ARTICLE pubs.acs.org/est

Photosensitized Oxidation of Emerging Organic Pollutants by Tetrakis C60 Aminofullerene-Derivatized Silica under Visible Light Irradiation Jaesang Lee,† Seokwon Hong,† Yuri Mackeyev,‡ Changha Lee,§ Eunhyea Chung,† Lon J. Wilson,‡ Jae-Hong Kim,|| and Pedro J. J. Alvarez*,^ †

Water Research Center, Korea Institute of Science and Technology, Seoul 136-791, Korea Department of Chemistry, Rice University, Houston, Texas 77005, United States § Department of Urban and Environmental Engineering, Ulsan National Institute of Science and Technology, Ulsan 689-798, Korea Department of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332, United States ^ Department of Civil and Environmental Engineering, Rice University, Houston, Texas 77005, United States

)

‡

bS Supporting Information ABSTRACT: We recently reported that C60 aminofullerenes immobilized on silica support (aminoC60/silica) eﬃciently produce singlet oxygen (1O2) and inactivate virus and bacteria under visible light irradiation.1 We herein evaluate this new photocatalyst for oxidative degradation of 11 emerging organic contaminants, including pharmaceuticals such as acetaminophen, carbamazepine, cimetidine, propranolol, ranitidine, sulﬁsoxazole, and trimethoprim, and endocrine disruptors such as bisphenol A and pentachlorophenol. Tetrakis aminoC60/silica degraded pharmaceuticals under visible light irradiation faster than common semiconductor photocatalysts such as platinized WO3 and carbon-doped TiO2. Furthermore, aminoC60/silica exhibited high target-speciﬁcity without signiﬁcant interference by natural organic matter. AminoC60/silica was more eﬃcient than unsupported (water-suspended) C60 aminofullerene. This was attributed to kinetically enhanced 1O2 production after immobilization, which reduces agglomeration of the photocatalyst, and to adsorption of pharmaceuticals onto the silica support, which increases exposure to 1O2 near photocatalytic sites. Removal eﬃciency increased with pH for contaminants with a phenolic moiety, such as bisphenol A and acetaminophen, because the electron-rich phenolates that form at alkaline pH are more vulnerable to singlet oxygenation.

’ INTRODUCTION Many previous studies have attempted the degradation of various organic contaminants using photosensitizers that produce reactive oxygen species (ROS) upon irradiation by light.2,3 Recently, increasing attention has been given to the development of photosensitizers that utilize visible light4 6 to overcome limitations of commonly used metal oxide photocatalysts such as TiO2 and ZnO that require UV irradiation for activation. Such eﬀorts include enhancing the susceptibility of metal oxide photocatalysts to visible light by introducing various dopants,4,5,7 anchoring of organic photosensitizers,8,9 and hybridization with other semiconductors. New semiconductor photocatalysts (e.g., BiVO4, PbBi2Nb2O9)10 12 and macrocyclic functional dyes (e.g., porphyrins and phthalocyanines)13 that exhibit inherent visible light activity have also been considered as photocatalysts for water or wastewater treatment. C60 fullerene and C60 derivatives with various surface functional groups undergo facile photoexcitation under irradiation by UV and visible light (λ < 550 nm). The resulting photoexcited r 2011 American Chemical Society

triplet state of fullerenes eﬃciently mediates the transfer of energy to molecular oxygen to produce singlet oxygen (1O2), as well as the transfer of electrons to produce superoxide radical anions (O2• ).14,15 Although C60 is extremely hydrophobic, C60 derivatives with hydrophilic surface functional groups enable the application of the above phenomena in the aqueous phase.1,16 19 For water and wastewater treatment applications, photochemically produced 1O2 is a primary oxidant for electron-rich moieties including polycyclic aromatic rings,20 benzene rings activated with electron-donating substituents,21 and conjugated double bonds22 that are commonly found in organic contaminants and cellular components (e.g., proteins, lipids, and DNA). Accordingly, multiple hydroxylated C60, commonly known as fullerenol, Received: August 25, 2011 Accepted: November 5, 2011 Revised: November 4, 2011 Published: November 05, 2011 10598

dx.doi.org/10.1021/es2029944 | Environ. Sci. Technol. 2011, 45, 10598–10604

Environmental Science & Technology has been shown to photochemically oxidize 2-chlorophenol17 and inactivated MS-2 bacteriophage16 in water. We previously developed C60 adducts with amine-, carboxylic-, and hydroxyl functional groups as a photocatalyst to inactivate Escherichia coli and MS-2 phage via photochemically produced 1 O2.19 We also demonstrated that cationic C60 aminofullerene, immobilized onto silica beads for eﬀective catalyst recycling, eﬃciently inactivated MS-2 phage under natural sunlight irradiation through rapid singlet oxygenation of viral capsid proteins.18 Whereas these studies veriﬁed the high eﬃcacy of functionalized C60 for photodynamic microbial inactivation, there is a need to delineate the applicability and limitations of these novel nanoparticles as visible-light-activated 1O2 photosensitizers for broader water treatment applications. In particular, there is a need for a multiactivity assessment with multiple emerging pollutants to discern susceptible chemical structures, and to compare functionalized fullerene performance versus other photocatalysts under various irradiation conditions. Pharmaceuticals and endocrine disruptors are emerging environmental contaminants with signiﬁcant concerns of interfering with hormones and reproductive systems, and thus causing severe adverse biological eﬀects.23 The present study explores the potential use of tetrakis C60 aminofullerene immobilized onto silica beads (referred to herein as aminoC60/silica) for oxidative degradation of various pharmaceuticals and endocrine disruptors under ﬂuorescent and visible light irradiation. Photochemical reactivity for 1O2 yield is measured as a function of fullerene content in the presence and absence of natural organic matter (NOM) or L-histidine (a 1O2 quencher). The eﬃcacy of aminoC60/silica is also compared to that of other widely studied photocatalysts, namely, TiO2, carbon-doped TiO2 (C-TiO2), and platinized WO3 (Pt/WO3).

’ MATERIALS AND METHODS Preparation of Tetrakis C60 Aminofullerenes and Immobilization onto Silica Support. Tetrakis C60 aminofullerene was

synthesized and purified as previously described.19 A brief description of the methods and chemicals is given in the Supporting Information (Text S1, Figure S1). The C60 aminofullerene was further immobilized onto 3-(2-succinic anhydride)propyl functionalized silica gel through amide bond. Detailed procedures for synthesis and purification are available in our previous publication1 and a brief summary is included in the Supporting Information (Text S2, Figure S2). Characterization by FIB/SEM and AFM. Cross-sectioning and electron beam imaging were performed using a dual beam focused ion beam/scanning electron microscopy (FIB/SEM) system coupled with an energy dispersive X-ray spectrometer (EDS) (Quanta 3D FEG, FEI Co., USA). Thin platinum and carbon layers were deposited on the surface of the materials (silica gel and aminoC60/silica) before ion milling to protect the area of interest and enhance charge and heat transfer. The FIB milling was performed by a focused gallium ion (Ga+) beam at 30 kV and 50 pA. SEM images of the FIB-prepared section were taken at an accelerating voltage of 5 kV. Topographic surface images of the silica support and aminoC60/silica were produced by atomic force microscopy (AFM, Dimension edge, Bruker AXS) in tapping mode. Preparation of Platinized WO3 and Carbon-Doped TiO2. Platinization of WO3 was performed using a photodeposition method.6 Briefly, an aqueous suspension containing 0.5 g/L

ARTICLE

Figure 1. (a) Photosensitized 1O2 production (i.e., degradation of furfuryl alcohol, FFA) by 50 μM tetrakis C60 aminofullerene versus tetrakis aminoC60/silica with a fullerene content of 0.05 mmol/g under ﬂuorescent light irradiation, and eﬀects of L-histidine and Suwannee River natural organic matter; and (b) FFA degradation by tetrakis aminoC60/silica as a function of fullerene content. Dashed curves represent nonlinear regression ﬁts to pseudo-ﬁrst-order decay ([aminoC60/silica]0 = 1 g/L; [FFA]0 = 100 μM; [L-histidine]0 = 0.1 M; [SRNOM]0 = 10 mg/L; [phosphate]0 = 10 mM; pHi = 7).

WO3 (nanopowder (<100 nm), Aldrich), 12.2 μM chloroplatinic acid, and 1 M methanol (electron donor) was irradiated by a 200-W medium pressure mercury lamp for 30 min. Pt/WO3 was then collected by filtration and washed with distilled water. A typical Pt loading on WO3 was estimated to be approximately 0.5 wt %. Carbon doping of TiO2 was achieved without addition of an external carbon source according to Park et al.7 Briefly, 1 N HCl (7.36 mL) was added dropwise to a mixture of titanium tetrabutoxide (97%, 11.22 mL) and 2-propanol (50 mL) under vigorous mixing in an ice bath. The gel products were aged under ambient condition for 1 day, dried at 80 C, and subsequently heat-treated at 250 C for 3 h. Photochemical Experiments. Photochemical experiments were carried out using a magnetically stirred 60-mL cylindrical quartz reactor surrounded by six 4-W commercial fluorescent lamps (emission wavelength 350 650 nm, Philips Co.) at an ambient temperature (22 C). The light intensity of the fluorescent lamp was measured with a pyranometer (Apogee, PYR-P) and determined to be 1.105 mW/cm2. The UV portion was relatively small (Figure S3) and the difference in light intensity between fluorescent and visible light was not detectable. Photochemical reactions under visible light irradiation were performed with a UV cutoff filter which blocked light below 400 nm. Note that the light source was polychromatic, which precludes determination of the quantum yield. 24 Typical reaction suspensions, buﬀered using 10 mM phosphate for pH 7 and 10 mM carbonate for pH 10, contained 1 g/L 10599

dx.doi.org/10.1021/es2029944 |Environ. Sci. Technol. 2011, 45, 10598–10604

Environmental Science & Technology

ARTICLE

Figure 2. (a) Cross-sectional and (b) topographic surface images of tetrakis C60 aminofullerene deposited on the functionalized silica. The small (20 60 nm) clusters bracketed in panel (b) appear to be C60 aminofullerene, based on comparison with the surface topology of the bare silica support (see Figure S5).

photosensitizers including aminoC60/silica, TiO2, C-TiO2, or Pt/WO3, and 0.1 mM target substrates. Sample aliquots of 1 mL were withdrawn from the illuminated reactor using a 1-mL syringe, ﬁltered through a 0.22-μm PTFE ﬁlter (Millipore), and injected into a 2-mL amber glass vial for further analysis. Photolytic experiments were performed in duplicate or more. The residual concentrations of target substrates at a constant time interval were quantiﬁed using a HPLC (Waters 2695) equipped with a C-18 column (ZORBAX Eclipse XDB-C18) and a photodiode-array detector (Waters 996). The eluent consisted of a binary mixture of 0.1% (v/v) phosphoric acid aqueous solution and acetonitrile (typically 70:30 by volume), while cimetidine, furfuryl alcohol, and ranitidine were quantiﬁed using the mobile phase comprising a mixture of water and methanol (60:40 v/v). The analysis of trimethoprim was carried out using an eluent of water containing 25 mM ammonium acetate/acetonitrile (70:30 v/v). The degradation of FFA and the target organic compounds generally followed an exponential decay pattern, and pseudoﬁrst-order rate constants (k values) were determined by nonlinear regression of concentration versus time data (Figures 1 and 3). This approach inherently assumes that the concentration of the primary oxidant (1O2) was at steady state.25,26

’ RESULTS AND DISCUSSION Properties of Aminofullerene Immobilized onto Silica. The 1O2 probe, furfuryl alcohol (FFA),25 was rapidly degraded in the presence of aminoC60/silica under fluorescent light irradiation (Figure 1a). FFA decay was completely inhibited by 0.1 M L-histidine as a 1O2 quencher. Negligible FFA degradation was observed with aminoC60/silica in the dark or by light irradiation alone without C60 (data not shown). These results corroborate that aminoC60/silica photochemically produced 1 O2, as previously shown for water-suspended C60 aminofullerene using electron paramagnetic resonance (EPR) analysis.19 The rate of photochemical 1O2 generation increased with the

amount of C60 aminofullerene loaded onto the silica support, approaching a maximum when the loaded C60 aminofullerene reached 0.02 0.05 mmol/g (Figure 1b). It is noteworthy that Suwannee River natural organic matter (SRNOM), which was added at an environmentally relevant concentration (10 mg/L), did not affect 1O2 production rate, suggesting that the presence of NOM in water at this level would not significantly inhibit the photosensitized degradation of the tested organic contaminants by aminoC60/silica. This result appears to contrast earlier findings that NOM quenches ROS such as •OH and 1O2.27,28 Immobilizing C60 aminofullerene onto silica support enhanced the kinetics for 1O2 production by up to about 2.2-fold compared to the same concentration of unsupported watersuspended C60 aminofullerene (for aminoC60/silica, k = 1.378 ( 0.015 h 1 (R2 = 0.9981); for C60 aminofullerene, k = 0.611 ( 0.021 h 1 (R2 = 0.9943)) (Figure 1a). Apparently, bonding of C60 aminofullerene onto the silica surface precluded its aggregation in the aqueous phase19 and mitigated against the subsequent loss of photochemical activity due to excited state quenching.29 This explanation is supported by cross-sectional observations using FIB/SEM (Figure 2a) and the EDS spectra (Figure S4) measured at several points on the cross-sectional area, which indicate the presence of C60 aminofullerene on the silica surface (in the dark layer under the bright platinum layer) (Figure 2a). Surface morphology analyses by tapping-mode AFM also indicate that small-sized clusters of C60 aminofullerenes are present on the silica surface (Figure 2b), while such morphological features do not exist on bare silica support (Figure S5). Such marked diﬀerences in surface morphology between aminoC60/ silica and silica support were consistently observed (Figure S6). Immobilization onto large (∼ 37 74 μm) silica support also facilitates the separation of the catalyst from the water matrix for recycling. Photographs of suspensions (Figure S7) and light absorption spectra (Figure S8) before and after ﬁltration with 11-μm ﬁlter paper suggest that aminoC60/silica can be readily recovered by ﬁltration process, whereas the unsupported homogenized suspensions cannot. 10600

dx.doi.org/10.1021/es2029944 |Environ. Sci. Technol. 2011, 45, 10598–10604

Environmental Science & Technology

ARTICLE

Figure 4. Dark adsorption of pharmaceutical compounds on tetrakis aminoC60/silica with a fullerene content of 0.05 mmol/g ([aminoC60/ silica]0 = 1 g/L; [pharmaceutical compound]0 = 100 μM; [phosphate]0 = 10 mM; pHi = 7).

Figure 3. Faster degradation of pharmaceutical compounds by (a) tetrakis aminoC60/silica with a fullerene content of 0.05 mmol/g than (b) homogenized tetrakis C60 aminofullerene in water under ﬂuorescent light irradiation. Curves represent nonlinear regression ﬁts to pseudoﬁrst-order decay ([tetrakis C60 aminofullerene]0 = 50 μM; [aminoC60/ silica]0 = 1 g/L; [pharmaceutical compound]0 = 100 μM; [phosphate]0 = 10 mM; pHi = 7).

Photochemical Degradation of Pharmaceutical Compounds. The photochemical reactivity of aminoC60/silica could

be used for oxidative degradation of a variety of emerging contaminants such as pharmaceuticals associated with toxicological impacts to aquatic environments.30,31 Figure 3a shows that aminoC60/silica enabled near complete degradation of 100 μM cimetidine, ranitidine, and propranolol with 30 min of fluorescent light irradiation. Degradation of trimethoprim and sulfisoxazole was slower, while carbamazepine was not degraded at all. The differences in the kinetics for photochemical pharmaceutical degradation could be attributed to the selective reactivity of 1O2, which was the primary oxidant in this system. Cimetidine, ranitidine, and propranolol contain chemical moieties such as furan, imidazole, and naphthalene, that are very vulnerable to singlet oxygenation (e.g., k(furan + 1O2 in CH2Cl2) = 1.4 107 M 1 s 1;32 k(imidazole + 1O2 in H2O) = 3.4 107 M 1 s 1;32 k(naphthalene + 1O2 in 1-butanol) = 5.2 108 M 1 s 1 32). Immobilizing C60 aminofullerene onto silica signiﬁcantly enhanced the kinetics of pharmaceutical degradation (Figure 3). The degradation of ranitidine also followed pseudo-ﬁrst-order kinetics, and was 31-fold faster for aminoC60/silica (k = 13.987 ( 0.016 h 1; R2 = 0.998) than for C60 aminofullerene (k = 0.445 ( 0.019 h 1; R2 = 0.992). The enhancement due to immobilization was higher (75-fold) for propranolol, with k = 10.77 ( 0.019 h 1 for

Figure 5. Degradation of phenolic compounds with pKa values (a) above 9 and (b) below 9 by tetrakis aminoC60/silica with a fullerene content of 0.05 mmol/g in water under ﬂuorescent light irradiation ([aminoC60/silica]0 = 1 g/L; [acetaminophen]0 = [bisphenol A]0 = [4chlorophenol]0 = [trichlorophenol]0 = 100 μM; [pentachlorophenol]0 = 25 μM; [phosphate]0 = 10 mM (pHi = 7); [carbonate]0 = 10 mM (pHi = 10)).

aminoC60/silica (R2 = 0.997) and k = 0.145 ( 0.015 h 1 for aminofullerene (R2 = 0.975). Trimethoprim was degraded in the presence of immobilized aminofullerene, but no degradation occurred with suspensions of unsupported aminofullerene. This kinetic enhancement cannot be solely attributed to the increase in 1 O2 production by aminoC60/silica (Figure 1). Experiments performed in the dark (Figure 4) suggest that the silica support readily adsorbed pharmaceuticals that experienced enhanced oxidation 10601

dx.doi.org/10.1021/es2029944 |Environ. Sci. Technol. 2011, 45, 10598–10604

Environmental Science & Technology

ARTICLE

Figure 6. Degradation of pharmaceuticals by tetrakis aminoC60/silica with a fullerene content of 0.05 mmol/g in water under visible light irradiation ([aminoC60/silica]0 = 1 g/L; [pharmaceutical compound]0 = 100 μM; [phosphate]0 = 10 mM; pHi = 7).

kinetics with immobilized aminofullerene; i.e., propranolol, ranitidine, and trimethoprim. On the other hand, negligible removal (by sorption) of pharmaceuticals occurred in the presence of suspensions of unsupported homogenized aminofullerene in the dark (Figure S9). Considering the short lifetime of 1O2 in the aqueous phase (∼ 2 μs in distilled water33), contaminant adsorption would contribute to enhancing degradation of pharmaceutical by increasing exposure to 1O2 near photocatalytic sites. On the other hand, repulsive interaction between catalyst surface and sulﬁsoxazole appears to occur, which may hinder singlet oxygenation of sulﬁsoxazole on aminoC60/silica (Figure 3). pH-Dependent Kinetics for Oxidation of Phenolics. Figure 5 shows the kinetics of photosensitized oxidation of selected phenolic compounds including acetaminophen, bisphenol A, 4-chlorophenol, 2,4,6-trichlorophenol, and pentachlorophenol by aminoC60/silica under neutral and alkaline conditions. Adsorption and direct photolysis did not cause any detectable loss of these compounds under both conditions (data not shown). Even though alkaline conditions (pH 10) slightly decreased photochemical 1O2 production (inset, Figure 5), they significantly favored oxidative degradation of acetaminophen, bisphenol A, and 4-chlorophenol, which have pKa values near 9.5 (pKa(acetaminophen) = 9.5;34 pKa(bisphenol A) = 9.6 and 10.2;35 pKa(4-chlorophenol) = 9.4136) (Figure 5a). This effect is caused by the deprotonation of phenol to phenolate anion with the increase in pH. For example, the ratio of phenolate form of acetaminophen to phenol form, which was calculated based on the corresponding acid-dissociation equilibrium constant, pKa, increases from 0.3% to 76% as pH increases from 7 to 10. Phenolate anion is more electrophilic than neutral phenol and exhibits nearly 2 orders of magnitude higher reactivity toward 1 O221 (k(phenolate anion + 1O2 in H2O) = 1.8 108 M 1 s 1;32 k(phenol + 1O2 in H2O) = 3 106 M 1 s 1;32 k(4-chlorophenolate anion + 1O2 in H2O) = 1.9 108 M 1 s 1;32 k(4-chlorophenol + 1 O2 in H2O) = 6 106 M 1 s 1).32 Similarly, the oxidation of anionic phenolate by ozone is 105 106 times faster than for neutral phenol (k(phenolate anion + O3 in H2O) = 1.4 ( 0.4 109 M 1 s 1;32,37 k(phenol + O3 in H2O) = 1.3 ( 0.2 103 M 1 s 1 32,37). At neutral pH, highly chlorinated phenols such as 2,4,6-trichlorophenol and pentachlorophenol are predominantly present in

Figure 7. Degradation of (a) furfuryl alcohol and (b) ranitidine (RA) and cimetidine (CM) by TiO2, carbon-doped TiO2 (C-TiO2), platinized WO3 (Pt@WO3), and tetrakis aminoC60/silica with a fullerene content of 0.05 mmol/g in water under ﬂuorescent and visible light irradiation ([photocatalysts (or photosensitizers)]0 = 1 g/L; [furfuryl alcohol]0 = [pharmaceutical compound]0 = 100 μM; [phosphate]0 = 10 mM; pHi = 7).

dissociated (phenolate) forms (pKa(2,4,6-trichlorophenol) = 6.23;36 pKa (pentachlorophenol) = 4.7038) and were eﬀectively degraded by aminoC60/silica (Figure 5b). Faster oxidation of 2,4,6-trichlorophenol was similarly observed as the pH increased from 7 to 10 (Figure 5b), which would have increased the fraction of phenolate from 85% to nearly 100%. On the other hand, pentachlorophenol degradation was retarded at alkaline pH. This reﬂects that further deprotonation of pentachlorophenol from pH 7 to 10 was negligible, while the eﬃcacy of 1O2 production decreased under alkaline conditions. Comparison to Other Photocatalysts. Efficient oxidation of pharmaceuticals by aminoC60/silica was observed using visible light irradiation (when light below 400 nm was removed by a UV cutoff filter) (Figure 6). Figures S3 and S10 compare UV vis reflectance spectra of aminoC60/silica with varying fullerene content to the emission profile of a fluorescent lamp. A significant absorption in the spectral range of 400 to 600 nm implies utilization of visible light by aminoC60/silica for photochemical 1 O2 generation. Consistently, FFA degradation rates by visible light (λ > 400 nm) and fluorescent lamp light were very similar (i.e., k = 1.378 ( 0.015 h 1 for fluorescent light irradiation, versus 1.248 ( 0.024 h 1 for visible light-irradiation) (Figure 7a). Figure 7a and b compare the photocatalytic activity of aminoC60/ silica to those of the widely researched semiconducting photocatalysts TiO2, C-TiO2, and Pt/WO3 in terms of FFA, ranitidine, and cimetidine degradation under ﬂuorescent and visible light irradiation. Whereas TiO2 catalyzed signiﬁcant FFA oxidation under ﬂuorescent light, FFA was not degraded when a UV cutoﬀ ﬁlter was placed, conﬁrming the lack of visible light activity by TiO2 (Figure 7a). Similarly, unmodiﬁed TiO2 degraded ranitidine and 10602

dx.doi.org/10.1021/es2029944 |Environ. Sci. Technol. 2011, 45, 10598–10604

Environmental Science & Technology cimetidine under ﬂuorescent lamp light, but at a much slower rate compared to aminoC60/silica under visible light irradiation (Figure 7b). C-TiO2 did not catalyze FFA degradation under either ﬂuorescent or visible light (Figures 7a), while it exhibited moderate eﬃcacy in ranitidine and cimetidine degradation under visible light irradiation (Figures 7b). These results are likely due to reduced oxidizing power of the holes generated in midgap states of C-TiO240,41 and accelerated charge recombination in dopinginduced intermediate states.42,43 Pt/WO3 degraded FFA faster than the other photocatalysts tested under both ﬂuorescent and visible light conditions, as Pt/WO3 eﬀectively produces •OH via multiple electron transfer to oxygen.39 However, ranitidine and cimetidine degradation was faster with aminoC60/silica than with Pt/WO3 under visible light irradiation (Figure 7b). Although such comparisons should recognize that these photoactive materials generate diﬀerent oxidizing species (e.g., mainly 1O2 for aminoC60, •OH for Pt/WO3 6,39 and TiO2, and valence band hole for C-TiO2), results suggest that aminoC60/silica has a potential for application as an alternative environmental photocatalyst. Potential Applications. This study demonstrates that the photochemical activity of aminoC60/silica results in efficient 1O2 production, which could be used to degrade emerging organic micropollutants such as pharmaceuticals and endocrine disruptors. This newly developed photocatalyst has several advantages over existing technologies. First, aminofullerene immobilized onto relatively large support media facilitates separation for recycling. Second, it functions efficiently under visible light irradiation, overcoming a major drawback for many semiconductor-based photocatalysts that require UV irradiation and its associated infrastructure. Third, 1O2 is highly selective toward organic contaminants containing electron-rich moieties (e.g., furan and imidazole),32 which would enhance its efficacy when such pollutants are present in complex water matrices with relatively high concentrations of background organics (e.g., wastewater). In contrast, •OH (which is the major oxidant in semiconductor photocatalyst) has poor selectivity that offsets its higher oxidation capacity, and is wastefully consumed by nontarget substrates (including natural organic matter) and byproducts.44,45 It is noteworthy that even though complete oxidation of target micropollutants by 1O2 is unlikely to occur, slight oxidative modifications often cause drastic reduction of biological/estrogenic activities of pharmaceuticals,46 elimination of hazardous effects of algal toxins,47 and effective decolorization of dyes.48 Whereas these results suggest several potential advantages of aminoC60/silica for water pollution control applications, more studies are needed to identify potential critical limitations associated with scale-up issues and long-term operation on a larger scale under real world conditions. Further photocatalyst development should also consider diﬀerent types of functionalized C60, preferably with improved visible light quantum yield, modifying the physical and chemical properties of catalyst support to increase surface area available for the photocatalytic reaction, and introducing high-aﬃnity adsorption sites to enhance photocatalytic degradation.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional ﬁgures and text. This information is available free of charge via the Internet at http://pubs.acs.org/.

ARTICLE

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]; phone: (713) 348-5903; fax: (713) 348-5203.

’ ACKNOWLEDGMENT This study was supported by the U.S. National Science Foundation (Award CBET-0932872). Partial funding was also provided by the Korea Ministry of Environment as “Eco-Innovation program (Environmental Research Laboratory)” (414-111011) and as “Converging Technology Project” (191-101-001), by the Robert A. Welch Foundation (Grant C-0627), and by Basic Science Research Program through the National Research Foundation of Korea funded by the Ministry of Education, Science, and Technology (200110005647). We thank Dr. Jungwon Kim and Dr. Yiseul Park at POSTECH for their assistant on synthesis of platinized WO3 and carbon-doped TiO2. ’ REFERENCES (1) Lee, J.; Mackeyev, Y.; Cho, M.; Wilson, L. J.; Kim, J.-H.; Alvarez, P. J. J. C60 aminofullerene immobilized on silica as a visible-lightactivated photocatalyst. Environ. Sci. Technol. 2010, 44, 9488–9495. (2) Hoﬀmann, M. R.; Martin, S. T.; Choi, W. Y.; Bahnemann, D. W. Environmental applications of semiconductor photocatalysis. Chem. Rev. 1995, 95, 69–96. (3) Mills, A.; LeHunte, S. An overview of semiconductor photocatalysis. J. Photochem. Photobiol., A 1997, 108, 1–35. (4) Asahi, R.; Morikawa, T.; Ohwaki, T.; Aoki, K.; Taga, Y. Visible light photocatalysis in nitrogen doped titanium oxides. Science 2001, 293, 269–271. (5) Choi, J.; Park, H.; Hoﬀmann, M. R. Eﬀects of single metal ion doping on the visible light photoreactivity of TiO2. J. Phys. Chem. C 2010, 114, 783–792. (6) Kim, J.; Lee, C. W.; Choi, W. Platinized WO3 as an environmental photocatalyst that generates OH radicals under visible light. Environ. Sci. Technol. 2010, 44, 6849–6854. (7) Park, Y.; Kim, W.; Park, H.; Tachikawa, T.; Majima, T.; Choi, W. Carbon-doped TiO2 photocatalyst synthesized without using an external carbon precursor and the visible light activity. Appl. Catal. B: Environ. 2009, 91, 355–361. (8) Park, Y.; Lee, S. H.; Kang, S. O.; Choi, W. Organic dye-sensitized TiO2 for the redox conversion of water pollutants under visible light. Chem. Commun. 2010, 46, 2477–2479. (9) Li, G. S.; Zhang, D. Q.; Yu, J. C. A new visible light photocatalyst: CdS quantum dots embedded mesoporous TiO2. Environ. Sci. Technol. 2009, 43, 7079–7085. (10) Gopidas, K. R.; Bohorquez, M.; Kamat, P. V. Photophysical and photochemical aspects of coupled semiconductors. Charge transfer processes in colloidal CdS-TiO2 and CdS-AgI systems. J. Phys. Chem. 1990, 94, 6435–6440. (11) Kim, H. G.; Hwang, D. W.; Lee, J. S. An undoped, single phase oxide photocatalyst working under visible light. J. Am. Chem. Soc. 2004, 126, 8912–8913. (12) Li, G.; Zhang, D.; Yu, J. C. Ordered mesoporous BiVO4 through nanocasting: A superior visible light driven photocatalyst. Chem. Mater. 2008, 20, 3983–3992. (13) Kim, W.; Park, J.; Jo, H. J.; Kim, H. J.; Choi, W. Visible light photocatalysts based on homogeneous and heterogenized tin porphyrins. J. Phys. Chem. C 2008, 112, 491–499. (14) Arbogast, J. W.; Darmanyan, A. P.; Foote, C. S.; Rubin, Y.; Diederich, F. N.; Alvarez, M. M.; Anz, S. J.; Whetten, R. L. Photophysical properties of C60. J. Phys. Chem. 1991, 95, 11–12. (15) Yamakoshi, Y.; Umezawa, N.; Ryu, A.; Arakane, K.; Miyata, N.; Goda, Y.; Masumizu, T.; Nagano, T. Active oxygen species generated 10603

dx.doi.org/10.1021/es2029944 |Environ. Sci. Technol. 2011, 45, 10598–10604

Environmental Science & Technology from photoexcited fullerene (C60) as potential medicines: O2 versus 1 O2. J. Am. Chem. Soc. 2003, 125, 12803–12809. (16) Badireddy, A. R.; Hotze, E. M.; Chellam, S.; Alvarez, P.; Wiesner, M. R. Inactivation of Bacteriophages via photosensitization of fullerol nanoparticles. Environ. Sci. Technol. 2007, 41, 6627–6632. (17) Chae, S. R.; Hotze, E. M.; Wiesner, M. R. Evaluation of the oxidation of organic compounds by aqueous suspensions of photosensitized hydroxylated-C60 fullerene aggregates. Environ. Sci. Technol. 2009, 43, 6208–6213. (18) Cho, M.; Lee, J.; Mackeyev, Y.; Wilson, L. J.; Alvarez, P. J. J.; Hughes, J. B.; Kim, J.-H. Visible light sensitized inactivation of MS-2 bacteriophage by a cationic amine-functionalized C60 derivative. Environ. Sci. Technol. 2010, 44, 6685–6691. (19) Lee, J.; Mackeyev, Y.; Cho, M.; Li, D.; Kim, J.-H.; Wilson, L. J.; Alvarez, P. J. J. Photochemical and antimicrobial properties of novel C60 derivatives in aqueous systems. Environ. Sci. Technol. 2009, 43, 6604–6610. (20) Stevens, B.; Perez, S. R.; Ors, J. A. Photoperoxidation of unsaturated organic molecules. XIV. 1O2 acceptor properties and reactivity. J. Am. Chem. Soc. 1974, 96, 6846–6850. (21) Tratnyek, P. G.; Holgne, J. Oxidation of substituted phenols in the environment - A QSAR analysis of rate constants for reaction with singlet oxygen. Environ. Sci. Technol. 1991, 25, 1596–1604. (22) Jensen, A. W.; Daniels, C. Fullerene-coated beads as reusable catalysts. J. Org. Chem. 2003, 68, 207–210. (23) Schwarzenbach, R. P.; Escher, B. I.; Fenner, K.; Hofstetter, T. B.; Johnson, C. A.; von Gunten, U.; Wehrli, B. The challenge of micropollutants in aquatic systems. Science 2006, 313, 1072–1077. (24) Stumn, W.; Morgan, J. J. Aquatic Chemistry; John Wiley & Sons: New York, 1996. (25) Haag, W. R.; Hoigne, J. Singlet oxygen in surface waters. 3. Photochemical formation and steady-state concentrations in various types of waters. Environ. Sci. Technol. 1986, 20, 341–348. (26) Hoigne, J.; Faust, B. C.; Haar, W. R.; Scully, F. E.; Zepp, R. G., Aquatic humic substances as sources and sinks of photochemically produced transient reactants. In In Aquatic Humic Substances: Inﬂuence on Fate and Treatment of Pollutants; American Chemical Society: Washington, DC, 1989. (27) Zepp, R. G.; Hoigne, J.; Bader, H. Nitrate-induced photooxidation of trace organic chemicals in water. Environ. Sci. Technol. 1987, 21, 443–450. (28) Sandvik, S. L. H.; Bilski, P.; Pakulski, J. D.; Chignell, C. F.; Coﬃn, R. B. Photogeneration of singlet oxygen and free radicals in dissolved organic matter isolated from the Mississippi and Atchafalaya River plumes. Mar. Chem. 2000, 69, 139–152. (29) Lee, J.; Yamakoshi, Y.; Hughes, J. B.; Kim, J.-H. Mechanism of C60 photoreactivity in water: Fate of triplet state and radical anion and production of reactive oxygen species. Environ. Sci. Technol. 2008, 42, 3459–3464. (30) Halling-Sorensen, B.; Nielsen, S. N.; Lanzky, P. F.; Ingerslev, F.; Lutzhoft, H. C. H.; Jorgensen, S. E. Occurrence, fate and eﬀects of pharmaceutical substances in the environment - A review. Chemosphere 1998, 36, 357–394. (31) Richardson, M. L.; Bowron, J. M. The fate of pharmaceutical chemicals in the aquatic environment. J. Pharm. Pharmacol. 1985, 37, 1–12. (32) Wilkinson, F.; Helman, W. P.; Ross, A. B. Rate constants for the decay and reactions of the lowest electronically excited singlet state of molecular oxygen in solution - An expanded and revised compilation. J. Phys. Chem. Ref. Data 1995, 24, 663–1021. (33) Merkel, P. B.; Kearns, D. R. Radiationless decay of singlet molecular oxygen in solution. An experimental and theoretical study of electronic-to-vibrational energy transfer. J. Am. Chem. Soc. 1972, 94, 7244–7253. (34) Bailey, D. N.; Briggs, J. R. The binding of acetaminophen, lidocaine, and valproic acid to human milk. Amer. J. Clin. Pathol. 2004, 121, 754–757.

ARTICLE

(35) Kosky, P. G.; Silva, J. M.; Guggenheim, E. A. The aqueous phase in the interfacial synthesis of polycarbonates 0.1. Ionic equilibria and experimental solubilities in the BPA-NaOH-H2O System. Ind. Eng. Chem. Res. 1991, 30, 462–467. (36) Serjeant, E. P.; Dempsey, B. Ionisation Constants of Organic Acids in Aqueous Solution; Pergamon: Oxford, UK, 1979. (37) Hoigne, J.; Bader, H. Rate constants of reactions of ozone with organic and inorganic compounds in water. 2. Dissociating organic compounds. Wat. Res. 1983, 17, 185–194. (38) Cessna, A. J.; Grover, R. Spectrophotometric determination of dissociation constants of selected acidic herbicides. J. Agric. Food Chem. 1978, 26, 289–292. (39) Abe, R.; Takami, H.; Murakami, N.; Ohtani, B. Pristine simple oxides as visible light driven photocatalysts: Highly eﬃcient decomposition of organic compounds over platinum-loaded tungsten oxide. J. Am. Chem. Soc. 2008, 130, 7780–7781. (40) Mrowetz, M.; Balcerski, W.; Colussi, A. J.; Hoﬀmann, M. R. Oxidative power of nitrogen doped TiO2 photocatalysts under visible illumination. J. Phys. Chem. B 2004, 108, 17269–17273. (41) Rockafellow, E. M.; Stewart, L. K.; Jenks, W. S. Is sulfur-doped TiO2 an eﬀective visible light photocatalyst for remediation? Appl. Catal. B: Environ. 2009, 91, 554–562. (42) Choi, W. Y.; Termin, A.; Hoﬀmann, M. R. The role of metal ion dopants in quantum sized TiO2 - Correlation between photoreactivity and charge carrier recombination dynamics. J. Phys. Chem. 1994, 98, 13669–13679. (43) Martin, S. T.; Morrison, C. L.; Hoﬀmann, M. R. Photochemical mechanism of size quantized vanadium doped TiO2 particles. J. Phys. Chem. 1994, 98, 13695–13704. (44) Buxton, G. V.; Greenstock, C. L.; Helman, W. P.; Ross, A. B. Critical review of rate constants for reactions of hydrated electrons, hydrogen atoms and hydroxyl radicals (OH•/O• ) in aqueous solution. J. Phys. Chem. Ref. Data 1988, 17, 513–886. (45) Lee, Y.; von Gunten, U. Oxidative transformation of micropollutants during municipal wastewater treatment: Comparison of kinetic aspects of selective (chlorine, chlorine dioxide, ferrate(VI), and ozone) and non-selective oxidants (hydroxyl radical). Wat. Res. 2010, 44, 555–566. (46) Lee, Y.; Escher, B. I.; Von Gunten, U. Eﬃcient removal of estrogenic activity during oxidative treatment of waters containing steroid estrogens. Environ. Sci. Technol. 2008, 42, 6333–6339. (47) Lawton, L. A.; Robertson, P. K. J. Physico-chemical treatment methods for the removal of microcystins (cyanobacterial hepatotoxins) from potable waters. Chem. Soc. Rev. 1999, 28, 217–224. (48) Szpyrkowicz, L.; Juzzolino, C.; Kaul, S. N. A comparative study on oxidation of disperse dyes by electrochemical process, ozone, hypochlorite and Fenton reagent. Wat. Res. 2001, 35, 2129–2136.

10604

dx.doi.org/10.1021/es2029944 |Environ. Sci. Technol. 2011, 45, 10598–10604

ARTICLE pubs.acs.org/est

Ammonia Removal Using Activated Carbons: Effect of the Surface Chemistry in Dry and Moist Conditions Maraisa Gonc-alves, Laura Sanchez-García, Erika de Oliveira Jardim, Joaquín Silvestre-Albero,* and Francisco Rodríguez-Reinoso Laboratorio de Materiales Avanzados, Departamento de Química Inorganica-Instituto Universitario de Materiales, Universidad de Alicante, Ap. 99, E-03080 Alicante, Spain ABSTRACT:

The eﬀect of surface chemistry (nature and amount of oxygen groups) in the removal of ammonia was studied using a modiﬁed resin-based activated carbon. NH3 breakthrough column experiments show that the modiﬁcation of the original activated carbon with nitric acid, that is, the incorporation of oxygen surface groups, highly improves the adsorption behavior at room temperature. Apparently, there is a linear relationship between the total adsorption capacity and the amount of the more acidic and less stable oxygen surface groups. Similar experiments using moist air clearly show that the eﬀect of humidity highly depends on the surface chemistry of the carbon used. Moisture highly improves the adsorption behavior for samples with a low concentration of oxygen functionalities, probably due to the preferential adsorption of ammonia via dissolution into water. On the contrary, moisture exhibits a small eﬀect on samples with a rich surface chemistry due to the preferential adsorption pathway via Brønsted and Lewis acid centers from the carbon surface. FTIR analyses of the exhausted oxidized samples conﬁrm both the formation of NH4+ species interacting with the Brønsted acid sites, together with the presence of NH3 species coordinated, through the lone pair electron, to Lewis acid sites on the graphene layers.

1. INTRODUCTION New air quality emission standards are forcing the diﬀerent governments to adopt new regulations to control pollutant emissions. Among them, ammonia is considered an important health hazard because it is poisonous if inhaled in great quantities (breathing levels above 50 100 ppm), while it can cause eye, throat, and nose irritation in lesser concentrations.1 Actual ammonia emissions are generated mainly from the fertilizer manufacture industry, coke manufacture, fossil fuel combustion, livestock and poultry management, and refrigeration methods. Among all of these sources, livestock waste management and fertilizers production account for about 90% of total ammonia emissions.1,2 Many techniques have been proposed in the literature for the removal of NH3 in industrial eﬄuents. These include absorption by solution, reaction with other gases, ion exchange using polymeric resins, separation using membranes, thermal treatment, catalytic decomposition and adsorption by porous solids.3 5 Some of these techniques (e.g., thermal combustion) r 2011 American Chemical Society

are economically convenient for high concentration pollutants, whereas they become economically unviable for diluted waste streams. For these special cases, adsorption on porous solids (e.g., activated carbons, zeolites, and so on) can be an excellent approach.6 12 Among the diﬀerent porous solids described in the literature, activated carbons exhibit certain advantages such as a high “apparent” surface area, a highly developed porous structure, and, most importantly, the possibility to tailor the porous structure and surface chemistry to adapt it for a special application.13 These modiﬁcations using pre- and postsynthesis treatments are very convenient when trying to adsorb polar gases (e.g., NH3), because activated carbons usually exhibit nonpolar surfaces and surface modiﬁcations are mandatory. Received: September 3, 2011 Accepted: November 3, 2011 Revised: November 2, 2011 Published: November 03, 2011 10605

dx.doi.org/10.1021/es203093v | Environ. Sci. Technol. 2011, 45, 10605–10610

Environmental Science & Technology Previous studies described in the literature have anticipated that the surface chemistry is probably the most critical parameter deﬁning the total adsorption capacity of activated carbons for a basic molecule such as ammonia.6,7,9 Kim et al. showed that not only the amount but also the nature of the oxygen surface groups present on the carbon surface, that is, acidic groups, are responsible for the total amount adsorbed.7 Furthermore, Huang et al. reported a linear correlation between NH3 breakthrough capacity and the total amount of acidic groups.6 A similar beneﬁcial eﬀect for the ammonia removal has been described in the literature for metal-modiﬁed (Fe, Co, Cr, Mo, and W) activated carbons.8,10,14 Diﬀerent reaction mechanisms have been proposed for the adsorption process involving mainly Brønsted acid sites, but the true pathway is still under debate. Unfortunately, many of these studies described in the literature deal with a wide variety of carbon samples, that is, diﬀerent textural and chemical properties, together with additional eﬀects due to the presence of metal species, which makes diﬃcult a clear correlation between the inherent parameters of the carbon support and the adsorption mechanism.8,10,14 Furthermore, only few studies have paid attention to the role of moisture usually present in industrial streams. The objective of this Article is to evaluate the real role of the oxygen surface groups present on the surface of activated carbons in the removal of ammonia, either in the presence or in the absence of moisture. For this purpose, an activated carbon prepared from a resin precursor was modiﬁed by an oxidation treatment with HNO3, followed by a subsequent thermal treatment at diﬀerent temperatures to selectively remove some of the oxygen surface groups. As compared to a previous analysis described in the literature, the sole modiﬁcation of the surface chemistry in a controlled way will allow one to analyze the eﬀect of the surface chemistry in the ammonia adsorption process, under dry or moist conditions, avoiding the inﬂuence of additional parameters (e.g., porous structure or inorganic matter).

2. EXPERIMENTAL SECTION 2.1. Materials. A spherical activated carbon (MA2, supplied by MAST Carbon International, UK) with particle size around 0.32 mm was prepared by carbonization and subsequent activation using CO2 (43% burnoff) of a porous resin obtained by cross-linking of phenol-formaldehyde Novolac precursor with hexamethylenetetramine and using ethylene glycol as solventpore former. For the oxidation treatment, 25 g of carbon was treated with 250 mL of HNO3 solution (6 M) at 90 °C for 1 h. The oxidized sample (MA2ox) was washed until neutral pH and dried overnight at 85 °C. To selectively remove the surface functional groups, sample MA2ox was submitted to a subsequent thermal treatment at 300, 500, and 700 °C under a helium flow (50 mL/min) for 1 h. Samples were labeled MA2ox300, MA2ox500, and MA2ox700, respectively. 2.2. Characterization. N2 adsorption/desorption isotherms were performed at 195 °C in a homemade high precision volumetric equipment. Before any experiment, samples were degassed under vacuum (10 3 Pa) at 150 °C for 4 h. “Apparent” surface area was calculated from the nitrogen adsorption data after application of the BET equation, while the micropore volume (V0) was obtained after application of the Dubinin Radushkevich (D R) equation. CO2 adsorption isotherms were performed in the same apparatus at 0 °C following the same protocol. Application of the D R equation to the CO2 adsorption data was

ARTICLE

used to calculate the volume of narrow micropores (Vn), that is, those below 0.7 nm.15 Temperature-programmed decomposition experiments (TPD) were performed to evaluate the amount and nature of oxygen surface functionalities for the diﬀerent samples. 100 mg of carbon was placed in a quartz reactor under a helium ﬂow (50 mL/min) and submitted a heat treatment up to 1000 °C (10 °C/min). The diﬀerent gas species evolved as a result of surface groups decomposition (mainly CO and CO2) were analyzed using an online mass spectrometer (Omnistar TM, Balzers). Quantitative analyses were performed after calibration using CaC2O4 3 H2O as a reference material. FTIR spectra of the exhausted samples were obtained in the range 4000 700 cm 1, by adding 100 scans at a resolution of 5 cm 1, using a Mattson Inﬁnity Gold FTIR spectrometer in the diﬀuse reﬂectance method. Before any experiment, all samples were diluted with KBr. 2.3. Breakthrough Column Experiments. The adsorption capacity under dynamic conditions for ammonia removal at room temperature (23 °C) was measured using a 1 cm i.d. column containing 2 cm carbon bed height (∼0.6 g of sorbent). A total flow of 300 mL/min of air containing 1000 ppm of ammonia was passed through the column containing the fresh sample (without any pretreatment). The concentration of ammonia at the exit was followed using a Polytron 3000 (Drager) detector. Adsorption experiments were performed using either dry or moist air (70% humidity). For moist conditions, humidity was introduced into the NH3/air flow using a calibrated syringe. The breakthrough experiment was arbitrarily stopped when the ammonia concentration at the outlet reached a concentration of 100 ppm. The adsorption capacity was determined by integration of the area above the breakthrough curve and considering the inlet concentration, the total flow rate, and the amount of sorbent used. To further clarify the role of humidity, additional experiments were performed using prehumidified samples (20%, 40%, and 70% relative humidity). For this purpose, the carbon samples were first submitted to a thermal treatment at 100 °C overnight to remove the residual humidity, and, in a subsequent step, samples were exposed to a given H2O/H2SO4 mixture for 24 h at room temperature under controlled humidity conditions. Breakthrough column experiments using the prehumidified samples were always performed under dry conditions (NH3/dry air).

3. RESULTS AND DISCUSSION 3.1. Sample Characterization. Figure 1 shows the nitrogen adsorption/desorption isotherms for the original and the modified activated carbon. As can be observed, the resin-based activated carbon (MA2) exhibits a highly developed microporous structure, that is, closed knee in the nitrogen adsorption isotherm at low relative pressure, together with capillary condensation processes at p/p0 ≈ 0.8, corresponding to the filling of large mesopores. The oxidation treatment with HNO3 produces slight changes in the porous structure, the main effect being the partial blockage of the microporosity due to the creation of polar oxygen surface groups at the entrance of narrow pores; this slightly decreases the adsorption of a molecule such as nitrogen, due to steric effects at the pore mouth.16,17 A subsequent thermal treatment at high temperature (700 °C) under a helium flow produces the widening of the existing micro-/mesopores due to the partial gasification of the carbon structure.13,18 The textural properties of the modified activated carbon obtained from the N2 and CO2 adsorption data, 10606

dx.doi.org/10.1021/es203093v |Environ. Sci. Technol. 2011, 45, 10605–10610

Environmental Science & Technology

ARTICLE

Table 2. Total Amount of CO2 and CO Groups Evolved in the TPD Experiment carbons

Figure 1. Nitrogen adsorption/desorption isotherms at the diﬀerent activated carbons.

195 °C for

Table 1. Textural Parameters for the Diﬀerent Activated Carbons Obtained from the N2 and CO2 Adsorption Data at 195 and 0 °C, Respectively SBET

V0

Vmeso

Vt(0.97)

Vna

2

(m /g)

3

(cm /g)

3

(cm /g)

3

(cm /g)

(cm3/g)

MA2

1550

0.61

0.68

1.29

0.55

MA2ox

1470

0.61

0.63

1.24

0.53

MA2ox300

1570

0.63

0.67

1.30

0.54

MA2ox500

1580

0.64

0.67

1.31

0.54

MA2ox700

1700

0.69

0.74

1.43

0.61

samples

a

Obtained after application of the D R equation to the CO2 adsorption data at 0 °C.

after application of the corresponding equations, are reported in Table 1. The surface chemistry of the original and modiﬁed activated carbons was evaluated using temperature-programmed decomposition (TPD) experiments. The amount of CO2 and CO evolved in the temperature range from room temperature to 1000 °C was analyzed as described in the Experimental Section. Quantitative data for the diﬀerent samples are compiled in Table 2. CO2 groups usually evolve at low temperatures, and they correspond to the decomposition of the more acidic and less stable oxygen surface groups (mainly carboxylic and lactone groups), while CO evolution is due to the decomposition of the more stable and less acidic oxygen groups (mainly phenolic, carbonyl, and quinone groups) and takes place at a higher temperature.19 The as-received activated carbon (MA2) shows a relatively poor surface chemistry; that is, it has a small amount of oxygen surface groups, with a larger content of the less acidic groups as compared to those evolved as CO2 (CO/CO2 ratio of 4.6). As expected, the oxidation treatment with nitric acid produces a large increase in the amount of both types of oxygen surface groups, this enhancement being larger for the CO2 groups (CO/CO2 ratio of 1.6), in accordance with previous observations.17 A subsequent thermal treatment at low temperature (300 °C) produces an important decrease in the amount of the less stable and more acidic oxygen groups, that is, those evolved as CO2, while the more stable groups remain mainly unaﬀected. Temperatures higher than 300 °C are required to selectively remove not only the more acidic but also the majority of the more stable oxygen surface groups, that is, those evolved as CO in the TPD experiment.

CO (mmol/g)

CO2 (mmol/g)

MA2

0.837

0.180

MA2ox MA2ox300

2.800 2.820

1.770 0.862

MA2ox500

1.822

0.245

MA2ox700

1.048

0.093

3.2. Breakthrough Column Experiments. Breakthrough column experiments were performed in a glass tubular reactor using a total flow of 300 mL/min of either dry or moist air (70% humidity) containing 1000 ppm NH3. Figure 2 shows the corresponding curves obtained using (a) dry air and (b) moist conditions. The total adsorption capacity (mg/g) for the different samples is reported in Table 3. Dynamic adsorption experiments show important diﬀerences among the diﬀerent samples either in the presence or in the absence of moisture. In general, all samples exhibit a sharp breakthrough saturation proﬁle that proves the absence of kinetic restrictions in the adsorption process. This observation must be attributed to fast adsorption kinetics of NH3 through the mesoporous network on these carbon materials; the mesopores constitute the channels to access the inner microporosity. Under dry conditions (see Figure 2a), the total adsorption capacity largely improves (close to 5-fold increase) after an oxidation treatment of the original carbon with HNO3. The total adsorption capacity for the oxidized sample (MA2ox) achieves a value as high as 17.5 mg/g, in close agreement with previous measurements reported in the literature for oxidized activated carbons.6,8,20 A subsequent thermal treatment at low and high temperature produces a large decrease in the uptake down to a residual value even lower than that of the original sample. Interestingly, the decline in the adsorption capacity is very drastic after the thermal treatment at low temperature (sample MA2ox300), whereas a small decrease is observed thereafter (up to 700 °C). This observation anticipates the crucial role of the oxygen surface groups in the gas phase removal of ammonia and, more speciﬁcally, the important eﬀect of the more acidic and less stable (those decomposing at low temperature) oxygen surface groups (mainly carboxylic groups) in the adsorption process under dry conditions.6,7 Incorporation of 70% relative humidity in the inlet stream has a diﬀerent eﬀect depending on the surface chemistry (Figure 2b). The oxidized sample (MA2ox) exhibits a moderate improvement in the adsorption behavior after moisture incorporation, this clearly suggesting the predominant role of the oxygen surface groups in the adsorption process; that is, water must exhibit a marginal eﬀect in the adsorption mechanism. Concerning the thermally treated samples, Figure 2b shows that the improvement in the adsorption capacity due to moisture incorporation is larger as compared to the oxidized sample and it remains mainly constant for all samples, independent of the temperature of the thermal treatment applied (see Table 3). Apparently on these thermally treated samples, that is, these samples were the more acidic and less stable oxygen surface groups have been removed, moisture exhibits a crucial role in the adsorption mechanism, independent of the amount and nature of the remaining surface functionalities. On the contrary, on carbons with a rich surface chemistry, the preferential interaction of ammonia with the 10607

dx.doi.org/10.1021/es203093v |Environ. Sci. Technol. 2011, 45, 10605–10610

Environmental Science & Technology

ARTICLE

Figure 2. Breakthrough curves of ammonia adsorbed on the diﬀerent activated carbons (1000 ppm, 300 mL/min, 23 °C) using (a) dry and (b) moist (70% relative humidity) air.

Table 3. Total Adsorption Capacity for Ammonia under Dry and Moist (70% Relative Humidity) Air Conditions carbons MA2

NH3 dried

NH3 70% RH

Δm (mg/g)

(mg/g)

(mg/g)

(0 70% RH) 0.6

4.7

5.3

17.5

20.1

2.6

MA2ox300 MA2ox500

7.6 3.4

12.3 7.6

4.7 4.2

MA2ox700

1.9

6.1

4.2

MA2ox

carbon surface, assisted by the oxygen surface groups and, more speciﬁcally, by the more acidic groups, could be the key factor deﬁning the total adsorption capacity, independent of the presence or absence of moisture. The large adsorption capacity achieved with the oxidized sample (MAox) under dry conditions, together with the slight improvement observed after moisture incorporation is somehow in contradiction with previous studies described in the literature, which proposed a crucial role of water in the formation of NH4+ ions, necessary for the adsorption process via interaction with the Brønsted acid groups of the carbon surface.8 To further clarify the real role of the oxygen surface groups in the adsorption process, Figure 3a compares the total adsorption capacity, either under dry or under moist conditions, and the total amount of acidic oxygen surface groups, that is, those decomposing as CO2 in a typical TPD run. As described above, the total adsorption capacity is larger in the presence of moist air for all samples, independent of the surface chemistry. However, contrary to previous analysis described in the literature, the presence of moisture does not seem to be so critical for the ammonia adsorption process when oxygen surface groups are present (e.g., sample MAox); in other words, the adsorption capacity of the oxidized sample under dry conditions is rather high, and the improvement after moisture incorporation is rather small.8 Interestingly, there is a linear correlation between the amount of ammonia adsorbed and the total amount of acidic groups on the carbon surface, either in the presence or in the absence of humidity. Furthermore, a closer look to Figure 3a shows that the beneﬁcial eﬀect of moisture slightly decreases when the amount of oxygen surface groups increases, thus suggesting a diﬀerent adsorption mechanism in the presence or absence of surface functionalities (mainly acidic oxygen surface groups). While moisture highly improves the adsorption behavior of the

carbon materials with a poor surface chemistry (69% improvement on sample MA2ox700), moisture seems to exhibit a minor eﬀect (13% improvement on sample MA2ox) in the presence of oxygen functionalities. Incorporation of oxygen functionalities to the carbon surface can modify the adsorption behavior through two diﬀerent mechanisms: (i) the creation of active sites, preferentially acidic oxygen groups, at the periphery of the basal planes, which will promote NH3 adsorption via a Brønsted acid base process through ammonium ion formation (NH4+), or (ii) the creation of Lewis acid centers on the graphene layers due to the withdrawal of electron density by the more electronegative oxygen surface groups on the periphery, thus promoting the interaction of ammonia (Lewis base) via the lone pair electron. Taking into account that the incorporation of the oxygen functionalities to the carbon surface promotes both processes at the same time, the presence of a linear correlation as observed in Figure 3a does not provide enough information to exclude one mechanism against the other. Previous studies described in the literature using a large variety of activated carbons have shown that when the adsorption process is governed by a pure Brønsted acid base process, the adsorption behavior increases exponentially below a certain pH (∼4.5), which corresponds to the pKa for dissociation of carboxylic acid groups on the carbon surface.8 In this sense, Figure 3b shows the relationship between the total amount adsorbed and the surface pH of the diﬀerent samples used. At this point, it is important to highlight that at the pH of all activated carbons used in this Article (pH < 6.5), NH3 will be protonated to NH4+ in a proportion close to 100% (pka for ammonia is 9.3). As it can be observed, the absence of an exponential increase in the amount adsorbed below a certain pH value, either in the presence or in the absence of moisture, clearly suggests that the NH3 adsorption mechanism on activated carbons is not exclusively governed by the Brønsted acid sites on the surface, as proposed in the literature.8 Most probably, both adsorption mechanisms, that is, Lewis and Brønsted mechanisms, take place on activated carbons even in the presence or absence of moisture. To further clarify the role of humidity, a new series of experiments were performed using prehumidiﬁed samples. In a ﬁrst step, carbon materials were heat treated at 100 °C overnight to remove the residual humidity; in a second step, activated carbons were submitted to a prehumidiﬁcation treatment at diﬀerent percentages (20%, 40%, and 70% relative humidity) using a H2O/H2SO4 mixture, and, in a ﬁnal step, carbon samples were 10608

dx.doi.org/10.1021/es203093v |Environ. Sci. Technol. 2011, 45, 10605–10610

Environmental Science & Technology

ARTICLE

Figure 3. Dependence of the total breakthrough column capacity (a) as a function of the amount of acidic oxygen surface groups and (b) as a function of surface pH, in the presence or absence of moisture.

Figure 4. Dependence of ammonia adsorption capacity on the amount of water adsorbed under diﬀerent prehumidiﬁcation conditions (RH 0 70%) for samples MA2ox and MA2ox700.

analyzed in ammonia breakthrough column experiments using dry air. Figure 4 reports the amount of NH3 adsorbed for samples MA2ox and MA2ox700, that is, samples with a completely different surface chemistry, under diﬀerent prehumidiﬁcation conditions and as a function of the weight gain after the prehumidiﬁcation step, that is, the amount of humidity retained by the sample. As expected, the oxidized sample (MA2ox) is more hydrophilic, and, consequently, it is able to absorb a larger amount of water under the prehumidiﬁcation conditions used. Figure 4 shows that for the sample with a poor surface chemistry (sample MA2ox700), there is a linear relationship between the amount of water adsorbed by the carbon sample and the subsequent adsorption capacity for a basic molecule such as ammonia. This linear relationship conﬁrms that moisture plays a crucial role in the adsorption behavior on this sample. Most probably in samples with a poor surface chemistry (absence of acidic groups), adsorption of ammonia takes place via dissolution into water, in such a way that the amount of ammonia dissolved increases with the amount of solvent. The total adsorption capacity ranges from 1.1 mg/g for the dried sample to 5.3 mg/g for the sample prehumidiﬁed with 70% RH, this meaning a 4.8-fold increase. On the contrary, the oxidized sample exhibits a completely diﬀerent behavior. As described above, the adsorption capacity of the untreated sample, that is, in the absence of humidity, is as large as 15.6 mg/g, thus conﬁrming the crucial role of the oxygen surface groups in the adsorption of a basic molecule such as ammonia. At this point, it

Figure 5. FTIR spectra of activated carbons (a) MA2ox700, (b)MA2ox700-70% RH, (c) MA2ox, and (d) MA2ox-70% RH, after NH3 breakthrough column experiments.

must be highlighted that this value is slightly smaller than the one reported in Table 3 (17.5 vs 15.6 mg/g) for the same sample, thus conﬁrming the beneﬁcial eﬀect of the inherent humidity on the fresh sample. Prehumidiﬁcation with 20% RH produces a slight increase in the adsorption capacity up to a value of 17.5 mg/g, remaining constant thereafter (up to 70% RH). Taking into account that the amount of water adsorbed highly increases with the prehumidiﬁcation step (from 0 to 70% RH) on sample MA2ox, the absence of signiﬁcant changes in the adsorption capacity for this sample above 20% RH clearly suggests that the adsorption mechanism in carbons with a rich surface chemistry follows a diﬀerent pathway. Under these conditions, moisture exhibits a scarce eﬀect, the main contribution to the adsorption process coming from the oxygen groups, mainly acidic groups, present on the carbon surface as suggested in Figure 3a. As described above, the adsorption mechanism on oxidized samples can proceed following two approaches: (i) by interaction of NH3 through the lone pair electron with the Lewis acidic sites on the graphene planes; or (ii) by interaction of NH3 with acidic oxygen surface groups via NH4+ ions formation. Apparently, both adsorption mechanisms take place in the oxidized sample (MA2ox) either in the presence and in the absence of moisture, although with a certain prevalence for the Lewis acid base mechanism, according to Figure 3b. In a ﬁnal step, the nature of the active species after the adsorption process has been analyzed on the exhausted samples using FTIR. As can be observed in Figure 5, the heat-treated 10609

dx.doi.org/10.1021/es203093v |Environ. Sci. Technol. 2011, 45, 10605–10610

Environmental Science & Technology sample MA2ox700 exhibits a poor FTIR proﬁle; that is, the amount of surface species after the adsorption process is very small, in accordance with the above breakthrough column tests. FTIR spectra of the oxidized sample, MA2ox, after the adsorption process show intense bands, mainly when the experiment is performed in the presence of moisture. Mutually overlapping bands of OH and NH stretching vibration appear as a broad contribution at 3271 cm 1 as a result of NH3 adsorption. Furthermore, bands at 1637 and 1082 cm 1, δas NH3 and δsym NH3, respectively, denote the presence of NH3 coordinated to Lewis acid sites on the carbon surface (graphene layers).11 Formation of surface ammonium salts of carboxylic acid appears as a band at 1439 cm 1 (δsym NH4+ ion), as well as bands at 1521 cm 1 (νas COO ) and 1393 1236 cm 1 (νsym COO ). Consequently, FTIR analysis of the exhausted samples conﬁrms the presence of two diﬀerent adsorption mechanisms for ammonia adsorption on oxidized activated carbons: (i) adsorption on Brønsted acid sites (carboxylic acid sites), via ammonium ion (NH4+) formation, and (ii) adsorption on Lewis acid sites on the graphene layers, via the lone pair electron of the ammonia molecule. Interestingly, both mechanisms dominate in oxidized carbon samples either in the presence or in the absence of moisture.

’ AUTHOR INFORMATION Corresponding Author

*Phone/fax: +34965909350/3454; e-mail: [email protected].

’ ACKNOWLEDGMENT We acknowledge ﬁnancial support from MEC (project MAT200761734 FEDER) and Generalitat Valenciana (PROMETEO/ 2009/002). The European Commission is also acknowledged (project FRESP CA, contract 218138). J.S.-A. acknowledges support from MEC, GV, and UA (RyC2137/06).

ARTICLE

(11) Zawadzki, J.; Wisniewski, M. In situ characterization of interaction of ammonia with carbon surface in oxygen atmosphere. Carbon 2003, 41, 2257–2267. (12) Bernal, M. P.; Lopez-Real, J. M. Natural zeolites and sepiolite as ammonium and ammonia adsorbent materials. Bioresour. Technol. 1993, 43, 27–33. (13) Marsh, H.; Rodriguez-Reinoso, F. Activated Carbon; Elsevier: London, 2006. (14) Bandosz, T. J.; Petit, C. On the reactive adsorption of ammonia on activated carbons modiﬁed by impregnation with inorganic compounds. J. Colloid Interface Sci. 2009, 338, 329–345. (15) Garrido, J.; Linares-Solano, A.; Martín-Martínez, J. M.; MolinaSabio, M.; Rodríguez-Reinoso, F.; Torregrosa, R. Use of nitrogen vs carbon dioxide in the characterization of activated carbons. Langmuir 1987, 3, 76–81. (16) Rodríguez-Reinoso, F.; Molina-Sabio, M.; Mu~ necas-Vidal, M. A. Eﬀect of microporosity and oxygen surface groups of activated carbon in the adsorption of molecules of diﬀerent polarity. J. Phys. Chem. 1992, 96, 2707–2713. (17) Silvestre-Albero, A.; Silvestre-Albero, J.; Sepulveda-Escribano, A.; Rodríguez-Reinoso, F. Ethanol removal using activated carbon: Eﬀect of porous structure and surface chemistry. Microporous Mesoporous Mater. 2009, 120, 62–68. (18) Molina-Sabio, M.; Mu~ necas, M. A.; Rodríguez-Reinoso, F. Modiﬁcation in porous texture and oxygen surface groups of activated carbons by oxidation. Stud. Surf. Sci. Catal. 1991, 62, 329–339. (19) Figueiredo, J. L.; Pereira, M. F. R.; Freitas, M. M. A.; Orfao, J. J. M. Modiﬁcation of the surface chemistry of activated carbons. Carbon 1999, 37, 1379–1389. (20) Rodrigues, C. C.; Moraes, D. d., Jr.; da Nobrega, S. W.; Barboza, M. G. Ammonia adsorption in a ﬁxed bed activated carbon. Bioresour. Technol. 2007, 98, 886–891.

’ REFERENCES (1) Phillips, J. Control and pollution prevention options for ammonia emissions. EPA-456/R-95-002, 1995. (2) Calvert, S.; Englund, H. M. Handbook of Air Pollution Technology; Wiley: New York, 1984. (3) Pez, G. P.; Laciak, D. V. Ammonia separation using semipermeable membranes. U.S. patent 4,762,535, 1988. (4) Blonigen, S. J.; Fassbender, A. G.; Litt, R. D.; Monzyk, B. F.; Neﬀ, R. Method for ammonia removal from waste streams. U.S. patent 6,558,643, 2003. (5) Blonigen, S. J.; Fassbender, A. G.; Litt, R. D.; Monzyk, B. F.; Neﬀ, R. Apparatus and method for ammonia removal from waste streams. U.S. patent 6,838,069, 2005. (6) Huang, C.-C.; Li, H.-S.; Chen, C.-H. Eﬀect of surface acidic oxides of activated carbon on adsorption of ammonia. J. Hazard. Mater. 2008, 159, 523–527. (7) Kim, B.-J.; Park, S.-J. Eﬀect of carbonyl group formation on ammonia adsorption of porous carbon surfaces. J. Colloid Interface Sci. 2007, 311, 311–314. (8) Le Leuch, L. M.; Bandosz, T. J. The role of water and surface acidity on the reactive adsorption of ammonia on modiﬁed activated carbons. Carbon 2007, 45, 568–578. (9) Park, S.-J.; Jin, S.-Y. Eﬀect of ozone treatment on ammonia removal of activated carbons. J. Colloid Interface Sci. 2005, 286, 417–419. (10) Petit, C.; Bandosz, T. J. Role of surface heterogeneity in the removal of ammonia from air on micro/mesoporous activated carbons modiﬁed with molybdenum and tungsten oxides. Microporous Mesoporous Mater. 2009, 118, 61–67. 10610

dx.doi.org/10.1021/es203093v |Environ. Sci. Technol. 2011, 45, 10605–10610

ARTICLE pubs.acs.org/est

Laccase-Carrying Electrospun Fibrous Membranes for Adsorption and Degradation of PAHs in Shoal Soils Yunrong Dai,† Lifeng Yin,† and Junfeng Niu* State Key Laboratory of Water Environment Simulation, School of Environment, Beijing Normal University, Beijing 100875, P.R. China

bS Supporting Information ABSTRACT: The removal of polycyclic aromatic hydrocarbons (PAHs) from soil is costly and time-consuming. The high hydrophobicity of PAHs makes PAH diﬀusion from soil particles by hydraulic ﬂow diﬃcult. The phase transfer of PAHs from soil to another available mediator is crucial for PAH removal. This study focuses on the remediation of PAH-contaminated shoal soil, located in Yangtze, China, using three types of laccase-carrying electrospun ﬁbrous membranes (LCEFMs) fabricated via emulsion electrospinning. These LCEFMs were composed of coreshell structural nanoﬁbers (for PAH adsorption), with laccase in the core (for PAH degradation) and pores on the shell (for mass transfer). The LCEFMs with strong adsorptivity extracted the PAHs from the soil particles, resulting in an obvious enhancement of PAH degradation. The removal eﬃciencies in 6 h for phenanthrene, ﬂuoranthene, benz[a]anthracene and benzo[a]pyrene were greater than 95.1%, 93.2%, 79.1%, and 72.5%, respectively. The removal halflives were 0.0031.52 h, much shorter than those by free laccase (17.967.9 h) or membrane adsorption (1.2512.50 h). The third-order reaction kinetics suggested that the superﬁcial adsorption and internal diﬀusion were the rate-limiting steps of the overall reaction. A synergistic eﬀect between adsorption and degradation was also proposed on the basis of the triple phase distribution and kinetics analyses.

’ INTRODUCTION Polycyclic aromatic hydrocarbons (PAHs) are a class of potentially carcinogenic and ubiquitous global pollutants consisting of two or more fused benzene rings. PAHs can be emitted to the environment as a result of natural combustion processes and human activities.1,2 Due to their hydrophobic nature, PAHs tend to bind to particles or dust, and they eventually enter soils through direct deposition or rainwash. Soils act as a natural repository for PAHs, resulting in increasingly serious soil environmental pollution.35 These PAHs exert adverse eﬀects on both the edaphon and crops. In some instances, PAH may also cause serious health problems in humans and/or cause genetic alterations, through uptake and accumulation in food chains.6,7 PAHs are strongly hydrophobic and are diﬃcult to dissolve in water. This property makes it easy for PAHs to adsorb onto soil particles but diﬃcult for them to dissociate from these particles.8 Therefore, PAHs in soils are of low bioavailability, have poor degradability, and are diﬃcult to be removed from the soils.9,10 For this reason, the phase transfer of PAHs (from soils to another available mediator) is crucial for the puriﬁcation of soils.11,12 Much eﬀort has been made to extract PAHs from soil particles. PAHs could be washed oﬀ from the soil particles by surfactant solubilization.13 However, it is diﬃcult for researchers to ﬁnd an environmentally friendly surfactant with low cost and high eﬃciency. Other approaches, including extraction methods by supercritical liquid,14 organic solvent,15 hot water,16 and steam,17 could also transfer PAHs from soil particles to a liquid or gaseous phase. r 2011 American Chemical Society

Nevertheless, the feasibility and practicality of these techniques need to be improved economically. In addition, the extracted PAH requires further treatment. Therefore, a feasible method for PAH to interphase transfer is urgently needed for the removal of PAHs from soils. Moreover, a subsequent degradation procedure of minimized environmental interference and high removal eﬃciency is indispensable for rapid puriﬁcation. In our previous study, the strong adsorption of PAHs on electrospun ﬁbrous membranes (EFMs) was evidenced based on the micro and macro characterizations.18 Emulsion electrospinning, a feasible method to obtain the coreshell nanoﬁbers, has also been developed for enzyme immobilization in EFMs in the degradation of crystal violet (triphenylmethane dye).19 These ﬁndings inspired us to combine the two approaches for the adsorption and degradation of PAHs in soils via a phase transfer procedure. Herein, laccase was employed as the biocatalyst for PAH degradation. Laccase (p-diphenol:dioxygen oxidoreductases, EC 1.10.3.2) is a copper-containing oxidase. Most of the 16 priority PAHs listed by the United States Environmental Protection Agency (USEPA) can be transformed by laccase, and produce some intermediate products, including dihydroxyphenanthrene, dihydrodiol, diphenic acid, anthraquinone, anthrahydroquinone, benzo[a]pyrene Received: September 19, 2011 Accepted: November 2, 2011 Revised: October 28, 2011 Published: November 02, 2011 10611

dx.doi.org/10.1021/es203286e | Environ. Sci. Technol. 2011, 45, 10611–10618

Environmental Science & Technology quinones, hydroxybenzo[a]pyrene, anthraquinone, etc.2023 Wu et al. revealed the remediation of PAH contaminated soils by free laccase and found signiﬁcant dissimilation of benzo[a]pyrene and anthracene after 14 days of treatment.24 Nevertheless, free laccase is expendable and easily loses its activity when exposed to the actual environment. For practical application, the immobilization of laccase has been considered to enhance stabilities, adjust reaction, and improve recyclability of laccase. The improved performance of laccase-carrying electrospun ﬁbrous membranes (LCEFMs) has been evidenced in wastewater treatment.19 On the other hand, considering the strong hydrophobicity of PAHs, we speculated that the polymer ﬁbers of LCEFMs might extract PAHs from soil particles in the aqueous solution. However, there is still some doubt about whether the LCEFMs can remove PAHs from soil eﬃciently. In this study, three types of polymers, including poly(D,L-lactide) (PDLLA), poly(D,L-lactide-co-glycolide) (PDLGA), and methoxypolyethylene glycolpoly(lactide-coglycolide) (MPEGPLGA), were utilized as the support materials for the immobilization of laccase. All the selected polymers have excellent biocompatibility, mechanical properties, spinnability, and biodegradability. Emulsion electrospinning was introduced to fabricate LCEFMs for the immobilization of laccase. The LCEFMs were then used for the removal of phenanthrene, ﬂuoranthene, benz[a]anthracene, and benzo[a]pyrene from soil in the aqueous solution. The cooperation of membrane-adsorption and laccase-degradation was investigated on the basis of the phase distribution and reaction kinetic analyses. The removal mechanisms and the possible interactions between the LCEFMs and the PAHs in the soil are also discussed.

’ EXPERIMENTAL SECTION Materials. PDLLA, PDLGA, and MPEGPLGA (MW 100 000, for each) were provided by Daigang Biomaterials (Jinan, China). A triblock copolymer PEOPPOPEO (F108) was supplied by BASF. Laccase (p-diphenol:dioxygen oxidoreductases, EC 1.10.3.2) from Trametes versicolor with a 23 U mg1 solid activity and a substrate of 2,2-azinobis-3-ethylbenzothiazoline-6-sulfonate (ABTS, 99%) were obtained from Sigma-Aldrich. Fluorescein isothiocyanate (FITC), phenanthrene (99.7%), fluoranthene (99.5%), benz[a]anthracene (99.7%), and benzo[a]pyrene (99.0%) were obtained from J&K, and some of their properties are summarized in Table S1 (see the Supporting Information). The soil samples were collected from Baisha Shoal, which is located near to the Yangtze River, China. The soil samples were cleared oﬀ large pieces of debris (e.g., stones, bricks, and plant residue) and then they were freeze-dried and passed through a 250 μm sieve. The concentrations of the four typical PAHs in the soil were 312.7 ( 0.5, 228.5 ( 1.3, 292.6 ( 2.1, and 186.3 ( 0.9 μg kg1, respectively. The soil was stored at 18 °C prior to analysis and use. The physicochemical properties of the shoal soil are listed in Table S2 (Supporting Information). Preparation of Emulsions. The polymer was first dissolved in methylene dichloride with stirring for 3 h to form a homogeneous solution. The polymer concentrations of PDLLA, PDLGA, and MPEGPLGA were 15, 10, and 12 wt % in methylene dichloride, respectively. Each polymer solution concentration was determined upon formation of stable jets and bead-free fibers under stable electrospinning. To obtain the stable W/O emulsions, 10 wt % (relative to polymers) of F108 was used as an emulsifier in the polymer/ methylene dichloride solution. Then, 0.5 mL of 20 mg mL1 laccase solution was mixed with the polymer/F108/methylene

ARTICLE

dichloride solution to obtain the uniform emulsions. To verify that the laccase was or was not encapsulated in the electrospun fibers, the same amount of FITC-labeled laccase was used in the emulsions for laser confocal scanning microscopy (LCSM, LSM-510, Zeiss, Germany) observation. The laccase was labeled by FITC, as described in detail elsewhere.19 Moreover, three types of pure polymer solutions were prepared without F108 and laccase. Electrospinning. Electrospinning was conducted on a selfmade multiend electrospinning apparatus (see Figure S1, Supporting Information). In a typical procedure, the emulsion was first loaded into a 10 mL spinning solution cartridge with 12 30-gauge needles (0.5 mm inner diameter) attached. A syringe pump (RWD) was set to inject the emulsion at a flow rate of 0.5 mL min1. Electrospinning was conducted at a voltage of 12 kV, and the distance between the tip of the needle and the collector was 15 cm. The electrospun fibers were collected on a barrel covered with aluminum foil. It usually took a few minutes to obtain acceptably thick and integrated LCEFMs. These LCEFMs were then kept in glutaraldehyde vapor for 30 min. The vapor was obtained from a vacuum vessel containing 10 mL of glutaraldehyde aqueous solution (25 wt %) under 0.5 bar at 30 °C. These LCEFMs were then stored at 4 °C before application. The deactivated LCEFMs and pure electrospun fibrous membranes (PEFMs) were prepared with inactivated laccase (boiling for 10 min) and pure polymer solutions, respectively. Using the multiend electrospinning, the LCEFMs could be obtained immediately, minimizing the adverse effects of solvent, electric field, and dehydration on the activity of laccase. Characterization. The morphology of the LCEFMs was characterized using a field emission scanning electron microscope (FESEM S-4800, Hitachi). The fibers were collected on microscope glass slides during electrospinning and observed under LCSM. The excitation and emission wavelengths were 488 and 535 nm, respectively. The specific surface area and pore volume of the LCEFMs were measured using an ASAP 2020 analyzer (Micromeritics). The contact angle was measured on a Dataphysics OCA20 angle measuring instrument. Activity and Stability Assays. The activity of the laccase was determined by measuring the change of absorbance during the catalytic oxidation of ABTS by laccase, with a UVvis spectrophotometer (Cary 50, Varian) at 420 nm. The detailed measurement and calculation methods of laccase activity have been described elsewhere.19 The residual activities were measured for one month to evaluate the storage stability of the free laccase and the LCEFMs. Between each measurement, the samples were stored in a phosphate buﬀer (pH 3.5) at 4 °C. To assess the operational stability, the LCEFMs were separated from the reaction system after each run in an assay. These LCEFMs were washed three times with the phosphate buﬀer and then transferred to the fresh ABTS solution. This operation was repeated 10 times. Purification of PAH Contaminated Soil. All PAH removal experiments were carried out at 25 ( 1 °C and pH 6.5. First, 5 g (dry weight) of PAH-contaminated soil was added into 25 mL of deionized water, and the suspension was shaken at 150 rpm in the dark for 24 h (on KS4000i, IKA) to reach the adsorption desorption equilibrium. Then, five LCEFMs (2 cm 2 cm, total weight 195200 mg) loaded with 5 mg of laccase were added. The reaction mixtures were then stirred for 24 h and sampled periodically. For each sampling, 50 μL of sodium azide (20 mmol L1) was added to terminate the laccase catalysis. Control experiments 10612

dx.doi.org/10.1021/es203286e |Environ. Sci. Technol. 2011, 45, 10611–10618

Environmental Science & Technology

ARTICLE

with the equivalent amount of free laccase, deactivated LCEFMs, PEFMs, and PEFMs/free laccase were carried out in the same reactor. Moreover, blank experiments without membranes and laccase were also performed. All samples were produced in triplicate, including the controls. Analytical Methods. Water samples were extracted with 25 mL of dichloromethane/methanol mixture (1:1, v/v), and then the aqueous extracts were reduced to 2 mL in a rotary evaporator (RV 05 basic, IKA, Germany), and re-extracted with 5 mL of methanol again. PAHs were extracted from soil samples through accelerated solvent extraction apparatus (ASE300, Dionex) with 40 mL of hexane/acetone mixture (1:1, v/v). The extracts were concentrated to 2 mL and purified by using open glass columns containing 2 g of 5% deactivated silica gel (100200 mesh, from Qingdao Ocean Chemical Plant, China) and 1 g of anhydrous sodium sulfate. The columns were eluted with 10 mL of hexane/methylene dichloride mixture (1:1, v/v). The concentration and solvent replacement procedures of the eluents were similar to water samples. The extractions of PAHs from the LCEFMs were carried out in an ultrasonic bath (KQ-502B, Kunshan Ultrasonic Instruments, China) for 30 min by using acetonitrile as the solvent. Quantification analysis of the PAHs was performed on a high-performance liquid chromatograph (HPLC, Dionex U3000) equipped with a ChromSep C18 column (250 4.6 mm, 5 μm, Varian) and a fluorescence detector. For three types of LCEFMs, both degradation (degraded by laccase) and removal (adsorbed by electrospun ﬁbrous membrane and degraded by laccase) eﬃciencies were introduced to evaluate the PAH removal from soils. The adsorption eﬃciency was deﬁned as the PAHs adsorbed by the deactivated LCEFMs. The degradation eﬃciency and removal eﬃciency were expressed as follows degradation efficiency ð%Þ ¼ 1 ðA1 þ A2 þ A3 þ BÞ ð1Þ removal efficiency ð%Þ ¼ 1 ðA3 þ BÞ

ð2Þ

where A1, A2, and A3 are the mass ratio of PAHs in the aqueous solution, on the membrane, and in the soil particle to initial total PAHs, respectively, and B is the mass ratio of the experimental loss to initial total PAHs.

’ RESULTS AND DISCUSSION Morphology and Structure of LCEFMs. The SEM images and diameter distribution of the three types of LCEFMs are shown in Figure S2 (Supporting Information). All three membranes consist of continuous porous fibers with an average diameter of 100500 nm. The three kinds of fibers were covered with nanoscale pores. The formation of pores on the surface of the fibers was considered as a dominant phase separation mechanism between the polymer and the air. Under the stress of strong electric fields, the polymer in the solution may be distributed randomly, resulting in the formation of polymer-rich and polymer-poor regions on the surface of the fibers. Probably the pores are formed from bubbles in the polymer-poor regions during the evaporation of the methylene dichloride.25 The formation of pores was also affected by the ambient humidity, the organic solvent, and the polymer.26,27 The porous structure might act not only as an arena for the adsorption but also as the access for mass transfer during the reaction.19,28

Figure 1. Laser scanning confocal microscope (LSCM) images of PDLLA (a), PDLGA (b), and MPEGPLGA (c) laccase-carrying electrospun ﬁbrous membranes (laccase was labeled by ﬂuorescein isothiocyanate).

Emulsion electrospinning has been used to prepare ultraﬁne ﬁbers with coreshell structures, in which the water-soluble enzyme or drug could be encapsulated into a hydrophobic or amphiphilic polymer.29,30 In our case, all three ﬁbers emitted green ﬂuorescence, which indicated that the laccase had been successfully encapsulated into them and also veriﬁed their core shell structures (Figure 1). The formation mechanisms of the coreshell structured ﬁbers during emulsion electrospinning were illustrated in our previous work.19 Activity and Stability Assays. The retained activities of the three types of LCEFMs relative to free laccase were all >75% (see Table S3, Supporting Information). Here, the biocompatible polymer shell both maintained the activity of the enzyme and protected the laccase from the influence of the external surroundings. Meanwhile, the porous structure on the surface of the fibers may facilitate the diffusion of substrate to/off the laccase, resulting in increased laccase activity. Furthermore, these special structures improved the storage and operational stabilities of the LCEFMs. As shown in Figure 2a, the free laccase lost more than 50% of its initial activity in 2 weeks. After a month of storage, the relative activity of the free laccase was less than 30%, whereas the LCEFMs still retained more than 70% of their initial activities. Figure 2b illustrates that the PDLLA LCEFMs retained 85% of its initial activity after oxidizing 10 batches of 0.5 mM ABTS, and the PDLGA and MPEGPLGA LCEFMs also retained more than 70% of their initial activities. In addition to the special structures of LCEFMs, the glutaraldehyde vapor treatment can also improve their stability. Glutaraldehyde may react with the primary amine groups of the enzymes, forming intermolecular cross-links and larger enzyme aggregates, which would be less likely to leak out of the fibers.31 However, there are also some diﬀerences among the LCEFMs prepared from the diﬀerent polymers. The storage and operational stabilities of the three LCEFMs follow the order of PDLLA > PDLGA > MPEGPLGA, which might be due to the differences in the hydrophilicityhydrophobicity of the polymers 10613

dx.doi.org/10.1021/es203286e |Environ. Sci. Technol. 2011, 45, 10611–10618

Environmental Science & Technology

ARTICLE

Figure 2. Storage stability (in phosphate buﬀer at 4 °C) and reusability of laccase-carrying electrospun ﬁbrous membranes.

Figure 4. The schema of the PAH phase transfer and degradation process by laccase-carrying electrospun ﬁbrous membrane. Figure 3. The degradation and removal eﬃciencies of phenanthrene, ﬂuoranthene, benz[a]anthracene, and benzo[a]pyrene by free laccase, PDLLA, PDLGA, and MPEGPLGA laccase-carrying electrospun ﬁbrous membranes. The solid columns represent degradation eﬃciency, and the addition of solid columns and hollow columns represent removal eﬃciency.

(see Table S3, Supporting Information). The most hydrophobic PDLLA might undermine the conformation stability of the enzyme, resulting in the lowest activity retention. Both the highest retained activity and the worst stability of the MPEGPLGA LCEFMs were due to the hydrophilicity. If the MPEGPLGA LCEFMs are immersed in an aqueous medium for a long time, their hydrophility may lead to the swelling and disintegration of the ﬁbers, causing enzyme leakage and activity loss. In comparison, the most hydrophobic PDLLA LCEFMs maintained the best storage and operational stability. PAH Removal by Free Laccase and LCEFMs. In our previous study, it was demonstrated that EFMs adsorbed a large amount of PAHs.18 Figure 3 shows the degradation and removal efficiencies of PAHs by free laccase and the three types of LCEFMs. The degradation efficiencies for the four PAHs were ranked in the order of phenanthrene > fluoranthene > benz[a]anthracene > benzo[a]pyrene, which was inversely related to the variation of log Kow (shown in Table S1, Supporting Information). However, the degradation efficiencies of the PAHs by all three LCEFMs were much higher than those by free laccase, especially for benzo[a]pyrene, whose degradation efficiency by free laccase was less than 30%, whereas those by the three LCEFMs exceeded 70%. Generally, it is accepted that the activity of immobilized laccase should be lower than that of free laccase. The unexpected higher degradation efficiencies might be due to the adsorption of PAHs on the membranes. In the soil suspension, some PAHs in

the soils were released into the aqueous solution, but the concentrations of PAHs in the aqueous solution were low. The degradation rate by free laccase for low concentrations of substrate is usually low because of the low probability of collision and the diffusion limitation. Moreover, free laccase can easily lose its activity after exposure to an aqueous solution for a long time, which results in the low degradation efficiency. However, for LCEFMs, PAHs in the aqueous solution can be adsorbed on the surface of the LCEFMs and concentrated around the active sites of laccase; thus, the degradation rates of PAHs were significantly enhanced. Therefore, the higher degradation efficiencies of PAHs by the LCEFMs than those by free laccase may be mainly due to the preconcentration of PAHs on the LCEFMs. For the three types of LCEFMs, their removal efficiencies for PAHs were higher than the degradation efficiencies because some PAHs adsorbed on the membranes had not been completely degraded by the laccase at 24 h. The difference in the removal efficiencies of the three LCEFMs may be due to the different adsorptivities and laccase activities of the LCEFMs for the four PAHs. The preconcentration process of PAHs on the LCEFMs can be explained as shown in Figure 4. To achieve equilibrium distribution between soil surface and aqueous phase, the PAHs were released into the water from the soil particles. The process is very slow due to the hydrophobicity of the PAHs. However, in the presence of the LCEFMs, the mass transfer of PAHs was accelerated since the PAHs in the water could be adsorbed onto the membrane. Beneﬁtting from the strong adsorptivity of the membrane, little PAH was left in the aqueous phase and more PAH was released from the soil particles under the stress of equilibrium. The adsorbed PAHs were strong enough to maintain a stable PAH “atmosphere” around the ﬁbers, meanwhile, they would migrate into the core through the porous shell and react with 10614

dx.doi.org/10.1021/es203286e |Environ. Sci. Technol. 2011, 45, 10611–10618

Environmental Science & Technology

ARTICLE

Figure 5. The triphase (soil, membrane, and aqueous phase) distribution diagram of phenanthrene, ﬂuoranthene, benz[a]anthracene, and benzo[a]pyrene in the presence/absence of laccase degradation. The data points for each laccase-carrying electrospun ﬁbrous membrane were drawn from the samples at 6, 12, and 24 h.

laccase immobilized in the core. As shown in Figure S2 (Supporting Information), the diameter of the pores on the ﬁbers was tens of nanometers, which was suﬃciently wide for the mass transfer of the PAHs and their degradation products. Therefore, the porous shell of the ﬁbers may not only shelter the laccase from environmental interference but may also provide access for the exchange of substance. With the migration of PAHs from the surface into the interior of the ﬁbers, there would be more sites for PAH adsorption, facilitating the phase transfer of PAHs from the soil to the membrane. These functional structures of the LCEFMs were advantageous for enhancing the degradation eﬃciencies of the PAHs in the soils. The triphase distribution diagram (Figure 5) shows the relationship between adsorption and degradation. Obviously, little PAH could be removed by free laccase under the limitation of external diﬀusion, and the degradation eﬃciencies for most of the PAHs at 24 h were lower than 50%. However, the four PAHs were rapidly adsorbed on the deactivated LCEFMs. More than 90% of the low ring PAHs, including phenanthrene and ﬂuoranthene, were transferred from the soil onto the deactivated LCEFMs. Benz[a]anthracene and benzo[a]pyrene were so hydrophobic that only 6080% of them could be transferred. As time proceeded, the distribution ratios of the PAHs on the membranes increased, while those on the soils decreased. The adsorptivities of the three deactivated LCEFMs for the diﬀerent PAHs mainly depended on the hydrophilichydrophobic properties of the polymers and the log Kow of the PAHs, which have been discussed in the previous study.18 As a result of the preconcentration of PAHs, the degradation eﬃciencies of the active LCEFM were improved to more than 90% (for phenanthrene and ﬂuoranthene) and 80% (for benz[a]anthracene and benzo[a]pyrene). By comparison, the distribution ratios of PAHs on the soils in the active LCEFM systems were higher than those in the deactivated LCEFM systems, whereas the distribution ratios of the PAHs on membranes were much less, as a result of the reduction of total PAH amount after

the laccase degradation. In addition, the distribution ratios of the PAHs in the aqueous solution were low because they could be rapidly adsorbed onto the active LCEFMs and degraded by laccase. The PAHs were transferred from the soils to the aqueous solution and then distributed onto the active LCEFMs. Theoretically, this process continued until all PAHs were are removed from the entire system. The synergistic eﬀect between membrane adsorption and laccase degradation need to be further evidenced by kinetic analysis. The degradation/adsorption kinetic order can be determined by the corresponding parameters of the regression equation. The kinetic rate constant (k) and the reaction half-life (t1/2) can be calculated by eqs 35 (IUPAC, Compendium of Chemical Terminology, 2nd ed., 1997). The ﬁrst-order reaction for PAH degradation by free laccase (eq 3), second-order reaction for PAH adsorption by the deactivated LCEFMs (eq 4), and thirdorder reaction (eq 5) for PAH degradation and removal by the active LCEFMs are expressed as listed lnðC0 =CA Þ ¼ kt;

t1=2 ¼ ln 2=k

CA =½C0 ð1 CA Þ ¼ kt; ð1=CA 2 1=C0 2 Þ ¼ 2kt;

t1=2 ¼ 1=ðkC0 Þ t1=2 ¼ 3=ð2kC0 2 Þ

ð3Þ ð4Þ ð5Þ

where C0 and CA are the initial and instantaneous concentration, respectively. The detailed kinetic parameters are listed in Table 1. The kinetic parameters for the catalytic degradation of PAHs by free and immobilized laccase can be well-regressed. For example, the degradation rate constant (kdeg) of the free laccase for phenanthrene, ﬂuoranthene, benz[a]anthracene, and benzo[a]pyrene were 0.0387, 0.0258, 0.0172, and 0.0102 h 1 , respectively, whereas their degradation rate constants (kdeg) by the active PDLLA LCEFM were 3.12, 0.51, 0.12, and 0.09 dm6 mmol1 h1, respectively. Correspondingly, the degradation half-lives of the PAHs were dramatically shortened from 17.967.9 to 0.1151.875 h. 10615

dx.doi.org/10.1021/es203286e |Environ. Sci. Technol. 2011, 45, 10611–10618

Environmental Science & Technology

ARTICLE

Table 1. Degradation/Adsorption/Removal Half-Life (t1/2) and Reaction Rate Constant (k) by Free Laccase and Three LaccaseCarrying Electrospun Fibrous Membranes Phea

kinetics parameter free laccase PDLLA

t1/2 (h)

adsorption

0.0387 2.82

0.0258 4.12

0.0172 3.76

0.0102 6.71

kads (dm3 mmol1 h1)

3.50

2.52

2.71

1.57

t1/2 (h)

0.115

0.307

1.525

1.875

kdeg (dm6 mmol1 h1)

1.32

0.57

0.12

0.08

17.9

26.8

40.3

67.9

t1/2 (h)

0.029

0.052

0.214

0.301

krem (dm6 mmol1 h1)

5.11

2.96

0.72

0.52

adsorption

t1/2 (h)

2.07

3.73

5.56

9.09

degradation

kads (dm3 mmol1 h1) t1/2 (h)

5.07 0.048

2.74 0.317

1.83 1.525

1.13 1.666

kdeg (dm6 mmol1 h1)

3.12

0.51

0.12

0.09

t1/2 (h)

0.007

0.094

0.375

0.751

21.60

1.60

0.40

0.20

t1/2 (h)

1.25

2.78

7.69

12.50

kads (dm3 mmol1 h1)

8.07

3.62

1.33

0.86

degradation

t1/2 (h)

0.428

0.306

1.875

2.509

removal

kdeg (dm6 mmol1 h1) t1/2 (h)

3.52 0.003

0.52 0.0125

0.08 1.52

0.06 1.45

1.20

0.13

0.09

removal

krem (dm6 mmol1 h1) adsorption

krem (dm6 mmol1 h1) a

BaPa

kdeg (h1) t1/2 (h)

removal

MPEGPLGA

BaAa

degradation

degradation

PDLGA

Flua

45.40

Phe, phenanthrene; Flu, ﬂuoranthene; BaA, benz[a]anthracene; BaP, benzo[a]pyrene.

Generally, the activity of the immobilized enzyme is lower than that of the free enzyme under the same conditions.32,33 To the best of our knowledge, the signiﬁcantly enhanced activity of immobilized laccase for PAHs has been rarely reported. More interestingly, the PAH degradation by the active LCEFMs was a third-order reaction. It is generally accepted that an enzyme reaction should be a ﬁrst-order reaction under the low concentration of the substrate or a zero-order reaction under the high concentration of the substrate in terms of the LineweaverBurk equation.34 The change of the reaction order might be attributed to the multiple control steps. During the PAH degradation by the LCEFMs, the reaction might be aﬀected by the nature of soil particles, the surface of ﬁbers, the porous structure, and the immobilized laccase. In addition, the PAH diﬀusions from the soil particle, in the solution, on the aqueous ﬁlm, on the surface of ﬁbers, in the pores, and near the immobilized laccase also inﬂuenced the kinetics of the overall reaction. However, it is diﬃcult to determine which step was the rate-limiting step. As shown in Table 1, the reaction rates of PAH degradation by the LCEFMs were much faster than those of the membrane adsorption. Therefore, the adsorption and diﬀusion might be the rate-limiting steps of the entire process. PAH adsorption onto the membrane (second-order kinetics) can be viewed as a link of the overall reaction. Other than the adsorption, the mass transfer from the ﬁbrous surface to the active sites of laccase was another possible rate-limiting step related to the PAH concentration. Hence, the third-order reaction might be attributed to the synergy of superﬁcial adsorption and internal diﬀusion. However, the speciﬁc kinetics for this complex system requires further investigation. The results showed that the PAH degradation reactions by LCEFMs were much faster than those by free laccase. The exact laccase-catalysis reaction rate was diﬃcult to measure. However, the ideal reaction rates on laccase in the LCEFMs,

at least, were faster than the apparent removal rates by LCEFMs, because the adsorption and diﬀusion were the rate-limiting steps. The absolute enhancement could be explained by the strong interactions between the concentrated PAHs and the active sites of the laccase. Therefore, the additivity of the reaction orders conveyed the synergistic eﬀect of the membrane adsorption and laccase degradation. However, diﬀerences in the performance of the diﬀerent LCEFMs were also observed on the basis of the kinetic analysis. As shown in Table 1, the removal rate of phenanthrene by the MPEG PLGA and the PDLGA LCEFMs was much faster than that by the PDLLA LCEFMs, whereas the removal rate of benzo[a]pyrene by the MPEGPLGA and PDLGA LCEFMs was lower than that by PDLLA LCEFMs. The log Kow of the PAHs determined the diﬀerence in the removal rates. Moreover, the membrane adsorption enhanced the diﬀerences in the laccase degradation. The strong adsorption of the membrane, the mass transfer via the pores, and the high activity of the laccase enabled the LCEFMs to eﬃciently remove the PAHs from soils. In the PAH removal process, the PAHs were ﬁrst released into the water under the adsorptiondesorption equilibrium of the soilwater. The distribution partition shifted immediately when the LCEFMs were introduced into the PAH solution. The PAHs were rapidly extracted by the LCEFMs, transferred into the ﬁbers via the pores, and subsequently degraded by the laccase. Obviously, the preconcentration of PAHs enhanced the degradation eﬃciency of the laccase. The functional activity and stability of the LCEFMs were much higher than those of free laccase during the removal of PAHs from contaminated soil. This ﬁnding might be helpful for the new environmental application of enzymes and raise a new approach to the use of nanomaterials for the removal of pollutants from soils. 10616

dx.doi.org/10.1021/es203286e |Environ. Sci. Technol. 2011, 45, 10611–10618

Environmental Science & Technology

’ ASSOCIATED CONTENT

bS

Supporting Information. Physicochemical properties of chemicals and samples and additional experimental results. This material is available free of charge via the Internet at http:// pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]; tel.: +86-10-5880 7612; fax: +86-10-5880 7612. Author Contributions †

These authors contributed equally to this work.

’ ACKNOWLEDGMENT This work was supported by the National Basic Research Program of China (973 Program, 2010CB429003), the National Natural Science Foundation of China (21077010), and the Key Project of Chinese Ministry of Education (No. 109026). ’ REFERENCES (1) Ahrens, M. J.; Depree, C. V. A source mixing model to apportion PAHs from coal tar and asphalt binders in street pavements and urban aquatic sediments. Chemosphere 2010, 81 (11), 1526–1535. (2) Wan, X. L.; Chen, J. W.; Tian, F. L.; Sun, W. J.; Yang, F. L.; Saiki, K. Source apportionment of PAHs in atmospheric particulates of Dalian: Factor analysis with nonnegative constraints and emission inventory analysis. Atmos. Environ. 2006, 40 (34), 6666–6675. (3) Timoney, K. P.; Lee, P. Polycyclic aromatic hydrocarbons increase in Aathabasca River delta sediment: Temporal trends and environmental correlates. Environ. Sci. Technol. 2011, 45 (10), 4278–4284. (4) Wang, Z.; Chen, J. W.; Tian, F. L.; Yang, P.; Qiao, X. L.; Yao, Z. Application of factor analysis with nonnegative constraints for source apportionment of soil polycyclic aromatic hydrocarbons (PAHs) in Liaoning, China. Environ. Forensics 2010, 11 (12), 161–167. (5) Wang, Z.; Chen, J. W.; Qiao, X. L.; Yang, P.; Tian, F. L.; Huang, L. P. Distribution and sources of polycyclic aromatic hydrocarbons from urban to rural soils: A case study in Dalian, China. Chemosphere 2007, 68 (5), 965–971. (6) Guzzella, L.; Poma, G.; De Paolis, A.; Roscioli, C.; Viviano, G. Organic persistent toxic substances in soils, waters and sediments along an altitudinal gradient at Mt. Sagarmatha, Himalayas, Nepal. Environ. Pollut. 2011, 159 (10), 2552–2564. (7) Wang, Z.; Chen, J. W.; Yang, P.; Qiao, X. L.; Tian, F. L. Polycyclic aromatic hydrocarbons in Dalian soils: Distribution and toxicity assessment. J. Environ. Monitor 2007, 9 (2), 199–204. (8) Jung, J. E.; Lee, D. S.; Kim, S. J.; Kim, D. W.; Kim, S. K.; Kim, J. G. Proximity of ﬁeld distribution of polycyclic aromatic hydrocarbons to chemical equilibria among air, water, soil, and sediment and its implications to the coherence criteria of environmental quality objectives. Environ. Sci. Technol. 2010, 44 (21), 8056–8061. (9) Couling, N. R.; Towell, M. G.; Semple, K. T. Biodegradation of PAHs in soil: Inﬂuence of chemical structure, concentration and multiple amendment. Environ. Pollut. 2010, 158 (11), 3411–3420. (10) Cave, M. R.; Wragg, J.; Harrison, I.; Vane, C. H.; Van de Wiele, T.; De Groeve, E.; Nathanail, C. P.; Ashmore, M.; Thomas, R.; Robinson, J.; Daly, P. Comparison of batch mode and dynamic physiologically based bioaccessibility tests for PAHs in soil samples. Environ. Sci. Technol. 2010, 44 (7), 2654–2660. (11) Rehmann, L.; Prpich, G. P.; Daugulis, A. J. Remediation of PAH contaminated soils: Application of a solid-liquid two-phase partitioning bioreactor. Chemosphere 2008, 73 (5), 798–804.

ARTICLE

(12) Yeom, I.; Ghosh, M. M. Mass transfer limitation in PAHcontaminated soil remediation. Water Sci. Technol. 1998, 37 (8), 111–118. (13) Peng, S.; Wu, W.; Chen, J. Removal of PAHs with surfactantenhanced soil washing: Inﬂuencing factors and removal eﬀectiveness. Chemosphere 2011, 82 (8), 1173–1177. (14) Dankers, J.; Groenenboom, M.; Scholtis, L. H. A.; van der Heiden, C. High-speed supercritical ﬂuid extraction method for routine measurement of polycyclic aromatic hydrocarbons in environmental soils with dichloromethane as a static modiﬁer. J. Chromatogr. A 1993, 641 (2), 357–362. (15) Rulkens, W. H.; Bruning, H.; van Hasselt, H. J.; Rienks, J.; van Veen, H. J.; Terlingen, J. P. M. Design of a solvent extraction process for PAH-contaminated sediments: The WAUacetone process. Water Sci. Technol. 1998, 37 (67), 411–418. (16) Dadkhah, A. A.; Akgerman, A. Hot water extraction with in situ wet oxidation: PAHs removal from soil. J. Hazard. Mater. 2002, 93 (3), 307–320. (17) George, C. E.; Azwell, D. E.; Adams, P. A.; Rao, G. V. N.; Averett, D. E. Evaluation of steam as a sweep gas in low temperature thermal desorption processes used for contaminated soil clean up. Waste Manage. 1995, 15 (56), 407–416. (18) Dai, Y. R.; Niu, J. F.; Yin, L. F.; Xu, J. J.; Xi, Y. H. Sorption of polycyclic aromatic hydrocarbons on electrospun nanoﬁbrous membranes: Sorption kinetics and mechanism. J. Hazard. Mater. 2011, 192 (3), 1409–1417. (19) Dai, Y. R.; Niu, J. F.; Liu, J.; Yin, L. F.; Xu, J. J. In situ encapsulation of laccase in microﬁbers by emulsion electrospinning: Preparation, characterization, and application. Bioresour. Technol. 2010, 101 (23), 8942–8947. (20) Zumarraga, M.; Plou, F. J.; Garcia-Arellano, H.; Ballesteros, A.; Alcalde, M. Bioremediation of polycyclic aromatic hydrocarbons by fungal laccases engineered by directed evolution. Biocatal. Biotransform. 2007, 25 (24), 219–228. (21) Johannes, C.; Majcherczyk, A. Natural mediators in the oxidation of polycyclic aromatic hydrocarbons by laccase mediator systems. Appl. Environ. Microbiol. 2000, 66 (2), 524–528. (22) Cho, S. J.; Park, S. J.; Lim, J. S.; Rhee, Y. H.; Shin, K. S. Oxidation of polycyclic aromatic hydrocarbons by laccase of Coriolus hirsutus. Biotechnol. Lett. 2002, 24 (16), 1337–1340. (23) Majcherczyk, A.; Johannes, C.; H€uttermann, A. Oxidation of polycyclic aromatic hydrocarbons (PAH) by laccase of trametes versicolor. Enzyme Microb. Technol. 1998, 22 (5), 335–341. (24) Wu, Y. C.; Teng, Y.; Li, Z. G.; Liao, X. W.; Luo, Y. M. Potential role of polycyclic aromatic hydrocarbons (PAHs) oxidation by fungal laccase in the remediation of an aged contaminated soil. Soil Biol. Biochem. 2008, 40 (3), 789–796. (25) Greiner, A.; Wendorﬀ, J. Electrospinning: A fascinating method for the preparation of ultrathin ﬁbers. Angew. Chem. Int. Ed. 2007, 46 (30), 5670–5703. (26) Li, D.; Xia, Y. Electrospinning of nanoﬁbers: Reinventing the wheel? Adv. Mater. 2004, 16 (14), 1151–1170. (27) Wang, P. Nanoscale biocatalyst systems. Curr. Opin. Biotechnol. 2006, 17 (6), 574–579. (28) Dai, Y. R.; Niu, J. F.; Yin, L. F.; Liu, J.; Jiang, G. X. Electrospun nanoﬁber membranes as supports for enzyme immobilization and its application. Prog. Chem. 2010, 22 (9), 1808–1818. (29) Yang, Y.; Li, X. H.; Qi, M. B.; Zhou, S. B.; Weng, J. Release pattern and structural integrity of lysozyme encapsulated in core-sheath structured poly(DL-lactide) ultraﬁne ﬁbers prepared by emulsion electrospinning. Eur. J. Pharm. Biopharm. 2008, 69 (1), 106–116. (30) Xu, X. L.; Zhuang, X. L.; Chen, X. S.; Wang, X. R.; Yang, L. X.; Jing, X. B. Preparation of core-sheath composite nanoﬁbers by emulsion electrospinning. Macromol. Rapid Commun. 2006, 27 (19), 1637–1642. (31) Thurston, E.; Herricks, A. S. K. B. Direct fabrication of enzymecarrying polymer nanoﬁbers by electrospinning. J. Mater. Chem. 2005, 15, 3241–3245. 10617

dx.doi.org/10.1021/es203286e |Environ. Sci. Technol. 2011, 45, 10611–10618

Environmental Science & Technology

ARTICLE

(32) Wu, L. L.; Yuan, X. Y.; Sheng, J. Immobilization of cellulase in nanoﬁbrous PVA membranes by electrospinning. J. Membr. Sci. 2005, 250 (12), 167–173. (33) Stoilova, O.; Manolova, N.; Gabrovska, K.; Marinov, I.; Godjevargova, T.; Mita, D. G.; Rashkov, I. Electrospun polyacrylonitrile nanoﬁbrous membranes tailored for acetylcholinesterase immobilization. J. Bioact. Compat. Polym. 2010, 25, 40–57. (34) Lineweaver, H.; Burk, D. The determination of enzyme dissociation constants. J. Am. Chem. Soc. 1934, 56 (3), 658–666.

10618

dx.doi.org/10.1021/es203286e |Environ. Sci. Technol. 2011, 45, 10611–10618

ARTICLE pubs.acs.org/est

Exploring a Water/Energy Trade-off in Regional Sourcing of Livestock Feed Crops Martin C. Heller* and Gregory A. Keoleian Center for Sustainable Systems, School of Natural Resources and Environment, University of Michigan, 3012 Dana Building, 440 Church Street, Ann Arbor, Michigan 48109-1041, United States

bS Supporting Information ABSTRACT: Feed production constitutes a major portion of the energy and water resource inputs in modern livestock production. Schemes to reduce these inputs may include local sourcing of animal feed. However, in water stressed regions where irrigation of feed crops is necessary, a trade-oﬀ between local sourcing (with high water stress) and transport from less water stressed regions can occur. We demonstrate this trade-oﬀ in the U.S. by combining state-level irrigation water use and pumping energy demand from USDA surveys with fertilizer and transportation energy demands for producing major feed crops (corn grain, soybean, alfalfa hay, corn silage) in each state and delivering them to two hypothetical dairy farms located in Kersey, CO and Rosendale, WI. A back-up technology approach is employed to express freshwater resource depletion in units of energy, allowing direct comparison with other energy resource demands. Corn grain, soybean, and alfalfa hay delivered to CO demonstrate a clear trade-oﬀ between transportation energy (proportional to the distance between CO and the production state) and water stress. On the other hand, transportation burdens dominate for corn silage, making local production most attractive, even in water stressed regions. All crops delivered to WI (a region of low water stress and minimal irrigation) are dominated by transportation burdens, making local production preferable, but this is clearly not a universal principal, as other cases show. This paper quantitatively elucidates the waterenergy trade-oﬀ in sourcing feed for livestock and the method is expected to be applicable in managing supply chain logistics of other farm commodities.

’ INTRODUCTION Recent reports by the FAO have drawn attention to the global environmental impact of animal agriculture.1 Ruminant animals contribute directly to greenhouse gas emissions through enteric fermentation, and manure management plays a critical role in nutrient related impacts (acidiﬁcation, eutrophication); however, most other environmental impacts of animal agriculture (land use, water use, energy use) are dominated by the production of feed for livestock. There is a rising interest in local foods, supported by a host of purported beneﬁts, including preserving taste, nutrient content, and freshness; maintaining money in the local economy; fostering closer relationships between producers and consumers; and reducing “food miles,” thus leading to a more energy eﬃcient and environmentally sustainable diet.2 There has been considerable recent debate on whether localizing food production oﬀers signiﬁcant energy eﬃciency advantages,24 but few holistic, quantitative studies exist that explore the potential trade-oﬀs across multiple environmental impacts. Eﬀorts by major retailers such as Wal-Mart to “green” their supply chain have driven food manufacturers to closely examine their environmental impact.5 Not surprisingly, analysis has demonstrated the importance of feed production in the overall life cycle impact of animal-based food products.6,7 It may be r 2011 American Chemical Society

desirable to source feed crop production physically near livestock in order to minimize transportation, and in turn, reduce the environmental burden. However, some environmental impacts, such as water use, are highly dependent on the location of crop production. The impacts of water use in crop production are especially acute in water stressed regions; for animal production located in these regions, a potential trade-oﬀ arises between local production of feed crops and transporting feed from less water stressed regions. This paper explores a method for quantitatively evaluating this trade-oﬀ. Fresh water is widely recognized as a natural resource in limited supply.8,9 As the human population grows and places greater demand on freshwater resources, greater shortages, and potential conﬂicts, are expected. Water for crop irrigation represents over two-thirds of global human withdrawals, but other demands such as energy generation and growing urban populations are increasing.10 Using the water footprint methods of Hoekstra et al.,11 Mekonnen and Hoekstra have reported on the global water use in major crop12 and animal products.13 Received: May 14, 2011 Accepted: November 1, 2011 Revised: October 31, 2011 Published: November 01, 2011 10619

dx.doi.org/10.1021/es2016473 | Environ. Sci. Technol. 2011, 45, 10619–10626

Environmental Science & Technology Analysts point to the close interdependence of energy and water resources and the need to integrate supply management in policy and design.14 Despite its importance, freshwater use has received limited attention in life cycle assessment (LCA) and similar life cycle-based sustainability assessment concepts,15 and there is ongoing debate as to the most appropriate way to quantify the impact of freshwater use.1618 One emerging methodological approach, presented by Pﬁster et al.19 provides a framework for evaluating the water and energy use trade-oﬀ in agricultural systems. In this paper, we apply integrated resource management principles to the livestock feed supply chain. We combine USDA irrigation survey data with the freshwater resource impact assessment methods of Pﬁster and estimates of the life cycle energy use of transportation and fertilizer production to explore the trade-oﬀ between energy and water resource impacts in supplying feed crops in the U.S. This study does not attempt to incorporate a full life cycle assessment of the feed supply chain but serves to quantify and compare the relative contributions of energy and water resource use. We consider feed crops typical of dairy operations: corn grain, corn silage, soybeans, and alfalfa hay; delivered to two hypothetical feeding locations in Kersey, CO and Rosendale, WI, chosen for demonstrative purposes. However, our approach is not speciﬁc to dairy, or livestock in general for that matter, and may ﬁnd application to any sector that sources commodity grain crops.

’ METHODS The goal of this study is to quantify the water stress due to irrigation of feed crops across the U.S., and compare that resource demand with the energy resources needed for irrigation pumping, fertilizer production, and transport of those feed crops to a given destination. To make the incommensurable indicators of water stress (in units of liter equivalents) and energy use (in units of MJ) directly comparable, we employ a “backup technology” concept to translate water resource use into energy equivalents. Detailed methods are described below. Irrigation Intensity. Irrigation application rates for various crops are reported at the state level in the USDA 2008 Farm and Ranch Irrigation Survey.20 Applied irrigation rates were combined with 2007 state total irrigated acres from the 2007 Census of Agriculture21 and divided by the state total harvested quantity to give an average state irrigation intensity (L/kg harvested). Note that this “irrigation intensity” is distributed across the total state harvest, from both irrigated and nonirrigated acreage. The analysis focuses on the continental U.S.: AK and HI are omitted. Irrigation Pumping Energy Use. The USDA Farm and Ranch Irrigation Survey also reports total irrigation pumping expenditures for each state, by fuel type. Using state-specific (when available) 2008 fuel prices from U.S. EIA,2224 fuel expenditures were converted to total pumping energy use, then divided by the total irrigated acres in each state. Survey irrigation expenses are not differentiated by crop type; thus, this method assumes the same pumping energy consumed per acre for all crop types. Pumping energy per acre is then multiplied by the irrigated acres per crop and state and divided by total harvested quantity as above to give pumping energy per kg harvested. Water Stress and Impact to Resources. Water stress assessment was based on the methods of Pfister et al.19 and utilized the WaterGAP2 model to estimate annual freshwater withdrawals relative to hydrological availability.25 Watershed polygons for the

ARTICLE

continental U.S. were clipped at state borders to provide 502 separate polygons. The fraction of freshwater consumption that contributes to depletion (Fdepletion) was calculated for each polygon based on eq 11 from Pfister et al.19 State-level aggregated Fdepletion values were calculated in analogy to country-level values derived by Pfister et al., i.e., using total annual withdrawal within the watershed as a weighting factor, and distributing crossboundary watersheds according to the area share within the specific state (eq 12 from Pfister et al.19). A midpoint water stress indicator was calculated by multiplying the irrigation intensity by a state-level water stress index (WSI).19 The resulting score is a weighted “water scarcity potential”, expressed in volume-equivalent of water use (i.e., L eq./kg). To make water use directly comparable to other types of resource uses (speciﬁcally, energy resources), the backup-technology concept,26 expressed as surplus energy to make the resource available in the future, was used according to Pﬁster et al.19 Desalination of seawater (with an energy demand set to Edesal. = 11 MJ/m3) was used as the backup technology. Impact from depletion of freshwater resources is then calculated as ΔR ¼ Edesal: 3 Fdepletion, state 3 WUirrigation;crop;state

ðeq:1Þ

where WUirrigation,crop,state is the irrigation intensity in L/kg for a given crop and state, as described above. Use of desalination as a backup technology for water resource depletion does not imply that all depleted water will be desalinated; it is merely used as a theoretical indicator to compare otherwise incommensurable resource types. It is important to note that this method assumes overuse of water resources occurs when freshwater withdrawals exceed hydrological availability. This may not always be true. Further, fossil groundwater depletion (and water stress) may occur even when the ratio of freshwater withdrawals to hydrological availability as calculated by WaterGAP2 is less than unity. Transportation Energy. Energy required to transport feed crops from the state of origin to the consuming livestock farm was based on life cycle data from Franklin Associates27 for transport via diesel truck (1.02 MJ/t km) and via rail (0.26 MJ/t km). Primary energy for both upstream activities (resource extraction, fuel processing, and distribution) and combustion (fuel use) were evaluated. Distances were estimated using GoogleMaps from the geographic center of the feed crop producing state to the consuming livestock farm located, in this exercise, in Kersey, CO and Rosendale, WI. Fertilizer. Fertilizer production constitutes a large portion of the embodied energy in many crop production systems. Further, application rates may vary regionally based on local soil conditions, yield expectations, or other agronomic factors. To demonstrate the effect of fertilizer application on the overall energy resource balance, state-level primary nutrient (N, PO43, K, S) application rates from the USDA NASS Agricultural Chemical Usage survey were used. Data for corn and soybeans were available for the 200528 and 200129 season, respectively. Data were reported for a limited number of states (representing 93% and 71% of the U.S. planted acreage of corn and soybeans, respectively), and an average of reported values was used as a proxy for unreported states. The life cycle energy demands of producing fertilizer were derived from Ecoinvent databases,30 aggregating fertilizer forms (urea, NH3, etc.) to represent a U.S. national average. The resulting energy intensities used were 61.3 MJ/kg N, 56.7 MJ/kg PO43, 13.7 MJ/kg K, and 89.4 MJ/kg S. 10620

dx.doi.org/10.1021/es2016473 |Environ. Sci. Technol. 2011, 45, 10619–10626

Environmental Science & Technology

ARTICLE

Table 1. Irrigation Intensities and Water Stress Assessment for Producing Major Feed Crops, by Statea irrigation intensity (L/kg) state

alfalfa hay

water stress indicator (L eq./ kg)

corn grain

soybean

corn silage

corn grain

soybean

alfalfa hay

corn silage

AL

40.2

15.8

3.7

16.0

0.8

0.3

0.1

0.3

AZ

730.4

NP

1102.0

238.0

728.9

NP

1099.7

237.5

AR

246.8

760.4

(NI)

(D)

41.7

128.5

(NI)

(D)

CA

533.2

NP

742.2

151.1

519.5

NP

723.2

147.2

CO

351.4

689.6

566.1

80.2

251.6

493.7

405.3

57.4

CT

(D)

(D)

(NI)

(NI)

(D)

(D)

(NI)

(NI)

DE

111.6

167.1

33.8

10.1

5.6

8.3

1.7

0.5

FL GA

86.7 207.8

(NI) 120.4

(NI) 41.2

58.5 58.9

78.8 5.9

(NI) 3.4

(NI) 1.2

53.1 1.7

ID

(D)

NP

522.9

111.2

(D)

NP

44.1

9.4

IL

3.6

4.5

0.3

0.1

0.1

0.1

0.0

0.0

IN

8.1

8.9

2.2

2.3

0.2

0.2

0.1

0.1

IA

1.3

2.1

0.4

0.2

0.1

0.1

0.0

0.0

KS

190.4

143.3

141.7

44.0

29.1

21.9

21.7

6.7

KY

3.3

6.2

(D)

(D)

0.1

0.1

(D)

(D)

LA ME

89.2 (NI)

122.5 (NI)

(D) (NI)

3.9 (NI)

6.5 (NI)

8.9 (NI)

(D) (NI)

0.3 (NI)

MD

33.2

54.1

3.2

1.1

28.8

46.9

2.8

1.0

MA

(D)

(NI)

(D)

(D)

(D)

(NI)

(D)

(D)

MI

24.4

22.3

2.6

1.7

0.8

0.8

0.1

0.1

MN

6.4

9.7

4.8

1.8

0.2

0.3

0.2

0.1

MS

118.5

322.4

(NI)

(NI)

2.1

5.7

(NI)

(NI)

MO

34.3

69.8

0.9

1.7

3.7

7.5

0.1

0.2

MT NE

490.8 159.0

(D) 225.3

316.1 120.5

91.8 34.7

57.5 143.0

(D) 202.6

37.1 108.4

10.8 31.2

NV

(NI)

NP

948.5

176.3

(NI)

NP

743.9

138.3

NH

(D)

NP

(D)

(NI)

(D)

NP

(D)

(NI)

NJ

19.4

43.9

19.0

0.1

16.8

38.1

16.5

0.1

NM

497.6

(D)

692.9

115.5

391.6

(D)

545.3

90.9

NY

0.3

(D)

0.1

0.1

0.2

(D)

0.1

0.1

NC

21.1

53.8

8.8

1.0

1.5

3.9

0.6

0.1

ND OH

14.0 1.0

3.7 0.8

10.6 1.2

5.1 0.1

2.0 0.0

0.5 0.0

1.5 0.0

0.7 0.0

OK

186.3

99.4

51.5

52.6

32.1

17.1

8.9

9.1

OR

727.7

(D)

611.4

92.0

135.5

(D)

113.8

17.1

PA

0.3

(D)

0.2

0.1

0.1

(D)

0.0

0.0

RI

(NI)

NP

(D)

(D)

(NI)

NP

(D)

(D)

SC

35.4

31.0

33.6

14.5

3.4

3.0

3.3

1.4

SD

14.4

9.2

23.4

3.9

2.5

1.6

4.1

0.7

TN TX

7.1 244.1

6.0 214.4

(D) 298.1

0.8 69.2

0.2 207.5

0.1 182.2

(D) 253.4

0.0 58.9

UT

894.0

NP

794.6

152.7

873.5

NP

776.4

149.2

VT

(NI)

(NI)

(D)

(NI)

(NI)

(NI)

(D)

(NI)

VA

10.7

14.4

1.2

0.9

2.6

3.5

0.3

0.2

WA

(D)

1193.4

497.2

91.6

(D)

71.0

29.6

5.5

WV

(D)

(D)

(D)

(D)

(D)

(D)

(D)

(D)

WI

8.8

12.4

2.3

0.6

0.3

0.4

0.1

0.0

WY

614.5

NP

793.0

128.2

472.7

NP

610.0

98.6

a

(D) = One or more necessary data points suppressed by USDA for conﬁdentiality purposes. (NI) = irrigation data not available through FRIS.20 NP = no reported production in state.

10621

dx.doi.org/10.1021/es2016473 |Environ. Sci. Technol. 2011, 45, 10619–10626

Environmental Science & Technology

ARTICLE

Figure 1. Water stress due to irrigation of corn grain in a given state plotted against the energy demand for transporting the corn grain (by road) to Kersey, CO.

’ RESULTS Calculated irrigation intensities by state for corn grain, soybean, alfalfa hay, and corn silage are reported in Table 1. This irrigation intensity is derived from survey reported rates, and should therefore reﬂect actual usage rather than predicted irrigation needs based on evapotranspiration models. Also shown in Table 1 is the characterized water stress assessment corresponding to the crop irrigation. The water stress characterization serves to emphasize water use in regions where overall usage is large relative to hydrological availability (water stressed regions). Figure 1 plots the water stress for corn grain production in a given state against the energy required to transport that corn to a farm in Kersey, CO by road via truck. CO and neighboring states show signiﬁcant water stress, whereas transport distance is lowest; other states (UT, AZ) that are relatively close in proximity have extremely high water stress. Sourcing corn grain from FL, on the other hand, represents the greatest transport distance but very low water stress. Herein lies the potential trade-oﬀ between water and energy resource use: optimizing either indicator independently could lead to a suboptimal systems outcome. A similar plot of water stress vs the cumulative energy demands (transportation, irrigation pumping, and fertilizer production) is included in the Supporting Information. When the water stress indicator is expressed in terms of the energy required to replace damaged resources, and then compiled with other energy resource ﬂows involved in crop production and delivery as in Figure 2, a number of interesting trends begin

to emerge. Figure 2 shows the compiled energy resource indicators for producing corn grain in a given state and delivering it to Kersey, CO. Note that states are ranked based on the sum of transportation, irrigation pumping, and desalination (water stress) energies. Fertilizer embodied energy is included in Figure 2 to provide perspective, but does not show signiﬁcant variation among states in this assessment. Whereas transportation energy is lower for corn grown in CO and neighboring states, this is counterbalanced by the impact of irrigation. Mountain states, such as CO and WY, which show high irrigation intensity in Table 1, have lower irrigation pumping energy than Great Plains states such as KS, OK, NE, and TX. This is presumably due to the fact that much of the irrigation in mountain states is gravity fed from reservoir-collected precipitation, whereas much of the irrigation in the Great Plains is from deep well aquifers, thus requiring signiﬁcantly greater pumping energy. Although the transportation depicted in Figures 1 and 2 is by road via truck and trailer, much of the corn shipped long distances in the U.S. is transported by rail or barge, thus imposing a reduced energy demand. Results using rail transport of corn grain are included in the Supporting Information. The reduced energy of rail transport further emphasizes the impacts of irrigation, making states throughout the Midwest more attractive sources of corn grain delivered to CO than neighboring Mountain or Great Plains states. Compiled energy resource indicators for the production of both soybean and alfalfa hay follow trends similar to those for 10622

dx.doi.org/10.1021/es2016473 |Environ. Sci. Technol. 2011, 45, 10619–10626

Environmental Science & Technology

ARTICLE

Figure 2. Cumulative energy demand (transportation, irrigation pumping, water stress, fertilizer) to produce corn grain in a given state and deliver it by road to Kersey, CO. Note that state ranking does not include fertilizer.

Figure 3. Cumulative energy demand (transportation, irrigation pumping, water stress, fertilizer) to produce corn silage in a given state and deliver it by road to Kersey, CO. Note that state ranking does not include fertilizer. 10623

dx.doi.org/10.1021/es2016473 |Environ. Sci. Technol. 2011, 45, 10619–10626

Environmental Science & Technology

ARTICLE

Figure 4. Cumulative energy demand (transportation, irrigation pumping, water stress, fertilizer) to produce corn grain in a given state and deliver it by road to Rosendale, WI. Note that state ranking does not include fertilizer.

corn grain (Figure 2), and are included in Supporting Information. One noted exception is that virtually no soybean is grown in the Western U.S., further skewing results toward Great Plains and Midwestern states. Figure 3 shows the irrigation and transportation impact of producing corn silage and delivering it to CO. Corn silage is typically produced near the consuming animals because long distance transportation is economically and logistically prohibitive, and this empirical conclusion is reﬂected in our analysis. Whereas irrigation and transportation burdens were balanced in corn grain production, transportation burdens dominate the corn silage analysis, thus favoring local production (CO and surrounding states). The irrigation demands per acre are roughly equivalent for growing corn grain and corn silage, but the harvested mass per acre is much greater for corn silage. When expressed as irrigation per kg harvested, irrigation impacts for corn grain are therefore more pronounced than corn silage. Figure 4 shows the same analysis, but for corn grain delivered to central WI. Irrigation of corn is not prevalent in WI and neighboring Midwest states, so the irrigation burdens for these states are low in our analysis, and thus transportation dominates, making local production favorable. This same trend (low irrigation burdens, ranking determined by transportation distance) is seen for all crops analyzed for delivery to central WI (shown in Supporting Information).

’ DISCUSSION The results shown here demonstrate a trade-oﬀ between water and energy resource use, and begin to oﬀer a glimpse into the nuances of local sourcing of food and feed crops. From an

environmental impact perspective, a “local is best” philosophy may not always hold true when sourcing irrigated crops. Eﬀectively balancing multiple environmental impact categories can be challenging; the methodology presented in this paper is an attempt to quantitatively demonstrate the trade-oﬀ between fresh water and energy resource consumption. Of course, sourcing decisions must be made in consideration of a collection of factors, including price, availability and timing, and social considerations: it is ill advised to base decisions on environmental factors alone. Additional environmental factors, including nutrient cycling indicators such as eutrophication and acidiﬁcation, may be of particular local import, requiring further consideration. Still, there is a known balance between energy and water resource use, and this balance has not been quantitatively described previously in the literature for feed crops. The physical distance between livestock and feed crops also impacts the feasibility of utilizing animal manures as a nutrient source. Although manure is a valuable source of N, P, and organic matter for crop production, transport of manure over long distances can be logistically and economically prohibitive. Trends toward increased concentration in the livestock sector have led to the separation of animal and crop production; the amount of land per animal unit across all animal types declined nearly 40% between 1982 and 1997.31 Although beyond the scope of this paper, one could imagine incorporating into the analysis presented here the energy cost of manure transport as well as the reduced need for chemical fertilizer when manures are utilized in crop production. Energy demands in addition to the ones considered in this study exist in the overall life cycle of feed crop production. These include in-ﬁeld fuel use, pesticide production, and other ancillary 10624

dx.doi.org/10.1021/es2016473 |Environ. Sci. Technol. 2011, 45, 10619–10626

Environmental Science & Technology inputs. Their exclusion in this assessment is not to suggest that they are insigniﬁcant, but instead that they are assumed to be relatively constant across diﬀerent geographic production regions. Although this assumption seems reasonable, state-level data to justify the assumption are not readily available. An additional energy demand that may have signiﬁcant regional variation is grain drying. Typical corn drying (from 25% moisture to 15% moisture) requires an estimated 0.40.8 MJ/kg of corn.32 While dependent on seasonal conditions, it is expected that the need for grain drying will be higher in wet, humid regions, perhaps adding a signiﬁcant energy demand to production in these regions. Again, data to validate this speculation is not available. Water stress indicators were compiled at the state level in order to make comparisons with other data available at this level. This creates complications because watersheds with signiﬁcantly diﬀerent water stress values may exist within the same state border. Such is the case, e.g., for CO: with the aggregation method used here, the state-level Fdepletion for CO is 0.34, whereas watersheds within the state have Fdepletion values ranging from 0 to 0.93. Evaluation at the watershed level would be the preferred method, but little agronomic data is available at this resolution. Furthermore, water availability as computed by the WaterGAP2 model does not account for exploitation of fossil groundwater reserves, which are particularly relevant in Great Plains states that utilize the Ogallala aquifer.33 In addition, as pointed out by Hoekstra et al.,17 the water stress characterization factor as developed by Pﬁster et al.19 does not account for local environmental ﬂow requirements; in other words, the “water use” needed for a local functioning ecosystem is not accounted for in the withdrawal-to-availability ratio. While ecosystem impact is certainly of great import, it is not the focus of this study. The use of theoretical backup technologies to assess the impacts of resource use may be considered controversial, and, according to Pﬁster, “...should be interpreted as a conceptual approach of quantifying environmental impacts for future generations in terms of additional energy consumption eventually needed to compensate for water scarcity.”19 Our approach may not provide an accurate energy budget for supplying fresh water, but this method does provide a relative valuation of the energy costs of water resource depletion. The energy of desalination value used here represents a stateof-the-art technology,34 and while of the same order of magnitude, it is on the lower end of reported values.35,36 A larger Edesal. value would further emphasize water stress impacts relative to the energy demands of transport. Whereas this study focuses on crop production within the U.S., there is great interest in the environmental impact of international trade in agricultural crops, including livestock feed crops (e.g., Pﬁster et al.,37 Siebert and Doll38). To the extent that necessary data (irrigation intensity, irrigation pumping, agronomic diﬀerences, etc.) is available, the methodology presented here could ﬁnd application in comparing domestically produced and imported feed crops. In comparing regions with grossly diﬀerent production practices, however, it may be necessary to include agronomic parameters in addition to those in the present analysis. In addition, international transport of feed crops is likely to involve composite modes, such as barge plus rail, adding complexity to the analysis. The assessment presented here brings light to the spatially explicit nature of resource demand in producing and sourcing feed crops for livestock production. Diverse climatic zones within

ARTICLE

the U.S. create a situation where signiﬁcantly diﬀerent resources are needed to produce equivalent commodity feedstuﬀs. When combined with the energy demand for transport of these feed crops, a trade-oﬀ in resource use emerges. As competition for both water and energy resources escalates, such spatially explicit sourcing schemes may be warranted in order to optimize resource use eﬃciency.

’ ASSOCIATED CONTENT

bS

Supporting Information. Results ﬁgures for combinations of additional crops, transport by rail, and transport to WI. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT We are grateful to the Aurora Organic Dairy Foundation for ﬁnancial support. The views, opinions, conclusions, and recommendations presented here do not necessarily reﬂect those of Aurora Organic Dairy. ’ REFERENCES (1) Steinfeld, H.; Gerber, P.; Wassenaar, T.; Castel, V.; Rosales, M.; de Haan, C. Livestock’s Long Shadow: Environmental Issues and Options; Food and Agriculture Organization of the United Nations: Rome, 2006. (2) Mariola, M. J. The local industrial complex? Questioning the link between local foods and energy use. Agric. Human Values 2008, 25 (2), 193–196. (3) Weber, C. L.; Matthews, H. S. Food-miles and the relative climate impacts of food choices in the United States. Environ. Sci. Technol. 2008, 42 (10), 3508–3513. (4) Edwards-Jones, G. Does eating local food reduce the environmental impact of food production and enhance consumer health? Proc Nutr. Soc. 2010, 69 (4), 582–591. (5) Rosenbloom, S. Wal-Mart Unveils Plan to Make Supply Chain Greener. The New York Times February 25, 2010. (6) Heller, M. C.; Keoleian, G. A. Life Cycle Energy and Greenhouse Gas Analysis of a Large-Scale Vertically Integrated Organic Dairy in the United States. Environ. Sci. Technol. 2011, 45 (5), 1903–1910. (7) Bastianoni, S.; Boggia, A.; Castellini, C.; Stefano, C. D.; Niccolucci, V.; Novelli, E.; Palotti, L.; Pizzigallo, A. Measuring Environmental Sustainability of Intensive Poultry-Rearing System. In Sustainable Agriculture Reviews 4; Lichtfouse, E., Ed.; Springer: 2010; pp 277309 (http:// www.springer.com/life+sciences/agriculture/book/978-90-481-8740-9? cm_mmc=Google-_-Book%20Search-_-Springer-_-0). (8) Gleick, P. H. Water and Conﬂict: Fresh Water Resources and International Security. Int. Security 1993, 18 (1), 79–112. (9) Vitousek, P. M.; Mooney, H. A.; Lubchenco, J.; Melillo, J. M. Human domination of Earth’s ecosystems. Science 1997, 277 (5325), 494–499. (10) Strzepek, K.; Boehlert, B. Competition for water for the food system. Philos. Trans. R. Soc., B 2010, 365 (1554), 2927–2940. (11) Hoekstra, A. Y.; Chapagain, A. K.; Aldaya, M. M.; Mekonnen, M. M. Water Footprint Manual: State of the Art 2009; Water Footprint Network: Enschede, The Netherlands, 2009. (12) Mekonnen, M. M.; Hoekstra, A. Y. The Green, Blue and Grey Water Footprint of Crops and Derived Crop Products; Value of Water Research Report Series No. 47; UNESCO-IHE: Delft, The Netherlands, 2010. 10625

dx.doi.org/10.1021/es2016473 |Environ. Sci. Technol. 2011, 45, 10619–10626

Environmental Science & Technology (13) Mekonnen, M. M.; Hoekstra, A. Y. The Green, Blue and Grey Water Footprint of Farm Animals and Animal Products; Value of Water Research Report Series No. 48; UNESCO-IHE: Delft, The Netherlands, 2010. (14) Hightower, M.; Pierce, S. A. The energy challenge. Nature 2008, 452 (7185), 285–286. (15) Koehler, A. Water use in LCA: managing the planet’s freshwater resources. Int. J. Life Cycle Assess. 2008, 13 (6), 451–455. (16) Berger, M.; Finkbeiner, M. Water Footprinting: How to Address Water Use in Life Cycle Assessment? Sustainability 2010, 2, 919–944 (17) Hoekstra, A. Y.; Gerbens-Leenes, W.; van der Meer, T. H. Reply to Pﬁster and Hellweg: Water footprint accounting, impact assessment and life-cycle assessment. Proc. Natl. Acad. Sci. U. S. A. 2009, 106 (40), E114. (18) Pﬁster, S.; Hellweg, S. The water 00 shoesize00 vs. footprint of bioenergy. Proc. Natl. Acad. Sci. U. S. A. 2009, 106 (35), E93–E94. (19) Pﬁster, S.; Koehler, A.; Hellweg, S. Assessing the Environmental Impacts of Freshwater Consumption in LCA. Environ. Sci. Technol. 2009, 43 (11), 4098–4104. (20) National Agricultural Statistics Service. Census of Agriculture Farm and Ranch Irrigation Survey; AC-07-SS-1; USDA, 2008; www. agcensus.usda.gov/Publications/2007/Online_Highlights/Farm_and_ Ranch_Irrigation_Survey/index.asp. (21) National Agricultural Statistics Service. Census of Agriculture; USDA, 2007; http://quickstats.nass.usda.gov. (22) Energy Information Administration. Weekly Retail Gasoline and Diesel Prices; US Department of Energy, 2008; www.eia.gov/dnav/ pet/pet_pri_gnd_a_epmr_pte_dpgal_a.htm. (23) Energy Information Administration. Electric Sales, Revenue, and Average Price 2008; U.S. Department of Energy: Washington, DC, 2009. (24) Energy Information Administration. Natural Gas Annual 2008; DOE/EIA-0131(08); U.S. Department of Energy: Washington, DC, 2010. (25) Alcamo, J.; Doll, P.; Henrichs, T.; Kaspar, F.; Lehner, B.; Rosch, T.; Siebert, S. Development and testing of the WaterGAP 2 global model of water use and availability. Hydrol. Sci. J. 2003, 48 (3), 317–337. (26) Stewart, M.; Weidema, B. A Consistent Framework for Assessing the Impacts from Resource Use: A focus on resource functionality. Int. J. Life Cycle Assess. 2005, 10 (4), 240–247. (27) Franklin Associates. USA LCI Database Documentation; Prairie Village, KS, 1998. (28) National Agricultural Statistics Service. Agricultural Chemical Usage 2005 Field Crops Summary; Ag Ch 1 (06); USDA: Washington, DC, 2006. (29) National Agricultural Statistics Service. Agricultural Chemical Usage 2001 Field Crops Summary; Ag Ch 1 (02)a; USDA: Washington, DC, 2002. (30) Ecoinvent Database version 2.0. Swiss Center for Life Cycle Inventories, 2007. (31) Ribaudo, M.; Gollehon, N.; Aillery, M.; Kaplan, J.; Johansson, R.; Agapoﬀ, J.; Christensen, L.; Breneman, V.; Peters, M. Manure Management for Water Quality: Costs to Animal Feeding Operations of Applying Manure Nutrients to Land; Agricultural Economic Report 824; USDA Economic Research Service: Washington, DC, 2003. (32) USDA-NRCS. Energy Self Assessment Conservation Tools; grain drying; http://www.ruralenergy.wisc.edu/conservation/grain_drying/ prequalify_graindrying.aspx. (33) Little, J. B., The Ogallala Aquifer: Saving a Vital U.S. Water Source. Sci. Am., Special Editions March 30, 2009. (34) Fritzmann, C.; Lowenberg, J.; Wintgens, T.; Melin, T. State-of-theart of reverse osmosis desalination. Desalination 2007, 216 (13), 1–76. (35) Stokes, J.; Horvath, A. Life cycle energy assessment of alternative water supply systems. Int. J. Life Cycle Assess. 2006, 11 (5), 335–343. (36) Vince, F.; Aoustin, E.; Breant, P.; Marechal, F. LCA tool for the environmental evaluation of potable water production. Desalination 2008, 220 (13), 37–56.

ARTICLE

(37) Pﬁster, S.; Bayer, P.; Koehler, A.; Hellweg, S. Environmental Impacts of Water Use in Global Crop Production: Hotspots and TradeOﬀs with Land Use. Environ. Sci. Technol. 2011, 45 (13), 5761–5768. (38) Siebert, S.; Doll, P. Quantifying blue and green virtual water contents in global crop production as well as potential production losses without irrigation. J. Hydrol. 2010, 384 (34), 198–217.

10626

dx.doi.org/10.1021/es2016473 |Environ. Sci. Technol. 2011, 45, 10619–10626

ARTICLE pubs.acs.org/est

Ditetraalkylammonium Amino Acid Ionic Liquids as CO2 Absorbents of High Capacity Jing-wen Ma,† Zheng Zhou,† Feng Zhang,*,†,‡ Cheng-gang Fang,† You-ting Wu,† Zhi-bing Zhang,† and Ai-min Li*,† †

State Key Laboratory of Pollution Control & Resource Reuse, School of Chemistry and Chemical Engineering, School of the Environment, Nanjing University, Nanjing 210093, China ‡ School of Pharmacy, China Pharmaceutical University, Nanjing 210009, China

bS Supporting Information ABSTRACT: By grafting butyl or ethyl onto tetramethylethylenediamine, quaternary ammonium salts with two positive charge centers were formed at the ﬁrst step. Metathesis with Ag2O followed. Through neutralization with glycine, L-alanine, or valine, a series of new ditetraalkylammonium amino acid ionic liquids (DILs) for CO2 capture were generated. The structures of DILs, as shown in Figure 1, were veriﬁed by using 1H NMR and EA. These DILs were found to be of quite high viscosity which militated against their industrial application in CO2 removal. Drawing on the experience of mixed amines’ aqueous solutions, these DILs were blended with water or N-methyldiethanolamine (MDEA) aqueous solutions to act as special absorbents of CO2. Using a Double-Tank Absorption System, the absorption performance of these DIL solutions was investigated in detail. The experimental results indicated that among the three aqueous solutions of DILs (20%, 40%, and 80 wt %), the solution of 40% DIL had a higher absorption rate of CO2 than the other two, demonstrating the diﬀerent eﬀects of concentration and viscosity on the absorption. The solution of 40% DIL or the 15% DIL + 15% MDEA had much higher capacity for CO2 than the corresponding monocation tetraalkylammonium AAILs, due to the special structure of the dication which could inﬂuence the solubility of CO2 in the aqueous solution.

1. INTRODUCTION As a worldwide concern, the treatment of CO2 has attracted more and more attention.1 Many techniques for CO2 removal have been developed. However, CO2 is not just a greenhouse gas but also an important chemical resource widely used in many ﬁelds. Carbon dioxide can be transformed into an environmentalfriendly extraction solvent in green chemical industries. Moreover, as a C1 resource, CO2 can be transformed into some basic chemical materials, such as CO, syngas, and methanol.27 Therefore, the capture of CO2 is good for both the environment and the recycling of carbon resource. At present, one of the most successfully and commercially utilized technologies for CO2 recovery is the absorption of CO2 by aqueous solutions of amines, including monoethanolamine (MEA), diethanolamine (DEA), N-methyldiethanolamine (MDEA), etc.8 However, there are several shortcomings in the use of alkanolamine,911 such as equipment corrosion, easy heat decomposition, oxidation of amines, and secondary pollution due to its high volatility. In view of the above-mentioned drawbacks of amines aqueous solutions, ionic liquids (ILs) are considered to be promising alternatives for the uptake of CO2. Room temperature ionic liquids (RTILs) are novel type of green solvents with special properties12,13 such as low vapor pressure, wide liquid process temperature, high stability, and easy assembly. RTILs have aroused considerable r 2011 American Chemical Society

interest due to their potential application as new green solvents.12 Compared with alkanolamines, RTILs have negligible volatility and remarkable thermal stability which can avoid loss of absorbents for CO2 uptake. Many research groups, especially that of Brennecke,14 have carried out signiﬁcant research on the solubility of CO2 in ILs. They15,16 ﬁrst reported that CO2 was highly soluble in [bmim][PF6](1-butyl-3-methylimidazolium hexaﬂuorophosphate) with a mole fraction of 0.6 at 8 MPa. However, when the pressure of CO2 was below ambient pressure, the solubility of CO2 in conventional ILs was only up to 0.035 mol fraction. In general, the absorption of CO2 by these traditional ILs was physical and had to be performed under extreme conditions such as very high pressure (90 bar or even higher) and very long time for the equilibrium to be reached (up to 24 h). Inspiringly, the unique “assembly” character of ionic liquids makes rapid and eﬃcient absorption of CO2 in ILs possible. Introducing special groups which can greatly enhance CO2 capture into the anion or the cation of ILs, functionalized ionic liquids by designing the structures of cation or anion according to Received: May 31, 2011 Accepted: November 8, 2011 Revised: October 29, 2011 Published: November 08, 2011 10627

dx.doi.org/10.1021/es201808e | Environ. Sci. Technol. 2011, 45, 10627–10633

Environmental Science & Technology the practical needs and special purposes are expected to have excellent performance in the uptake of CO2. Bates et al.17 reported that the molar uptake of CO2 per mole of [pNH2bim][PF6] (ionic liquid functionalized the cation with amine) during 3 h approached 0.5 under normal pressure and temperature. The absorption mechanism of CO2 in the amine functionalized ionic liquid was also proposed as the primary amines. Since then, lots of functional ILs19 with the amino group in cation have been synthesized for CO2 uptake. In 2005, amino-acid-based ILs [emim][amino acid] were prepared from 20 natural amino acids by Ohno’s group.20 Then, tetraalkylphosphonium amino acid ILs was also reported.19,20 Due to their high viscosities (245 745 mPa•s), tetraalkylphosphonium amino acid ILs had to be loaded on porous silica gel to absorb CO2.20 In 2010, Brennecke’s group21 successfully used a computational molecular design approach to identify a new class of ILs based on the AHAs (aprotic heterocyclic anions), which were shown to react stoichiometrically and reversibly with CO2 and to not suﬀer the large viscosity increases that have plagued previous attempts to create CO2-complexing ILs. Our research group have successfully synthesized nine tetraalkylammonium-based amino acid ionic liquids which could rapidly absorb CO2 (the absorption equilibrium could all be reached within 60 min).22 In 2010, lysine, which had two amino groups in one molecule, was also used as anion to obtain tetramethylammonium lysine ([N1111][Lys]) and tetraethylammonium lysine ([N2222][Lys]).19 It was found that [N1111][Lys] and [N2222][Lys] had signiﬁcant mole absorptivity of CO2. The absorption mechanism of CO2 in pure amino-acid-based ILs is similar to that of [pNH2bim][PF6], i.e., one CO2 molecule is combined with two amino groups23

ARTICLE

Practically, in the aqueous solution, the amino-acid-based ILs completely dissociate into hydrated cations and [H2N-CHRCOO] anion. Since the amino acid anions react with CO2 like the primary alkanolamine, [H2N-CHR-COO] can be represented as RNH2. In the aqueous solution of amino-acid-based ILs, the carbon dioxide absorption is usually described by the zwitterion mechanism.25 First, zwitterion is obtained through the reaction of CO2 with amino acid RNH2 þ CO2 h RNH2 þ COO

Then, the zwitterion is deprotonated by the bases (denoted by B) in the solution RNH2 þ COO þ B h RNHCOO þ BHþ

ð3Þ

where B includes RNH2, H2O, OH, MDEA, etc. In the aqueous solutions of IL and MDEA, the amino acid ﬁrst reacts quickly with CO2 to form zwitterions which will transfer protons to MDEA. Therefore, the new CO2 absorbent AAILs + MDEA + water combines the rapid absorption rate of AAILs and the high CO2 capacity of MDEA. The equilibrium reactions in the liquid phase are suggested as follows19 RNH2 þ COO þ RNH2 h RNHCOO þ RNH3 þ

ð4Þ

RNHCOO þ H2 O h RNH2 þ HCO3

ð5Þ

RNH3 þ h RNH2 þ Hþ

ð6Þ

CO2 þ H2 O h Hþ þ HCO3

ð7Þ

HCO3 h Hþ þ CO3 2

ð8Þ

H2 O h Hþ þ OH

ð9Þ

MDEAHþ h MDEA þ Hþ

The viscosity of amino acid ILs (AAILs) is still quite high.2224 Generally, the diﬀusivity of CO2 in the liquid is determined by the solvent’s viscosity. The lower the viscosity of solvent is, the higher is diﬀusivity in the solvent and thus the faster absorption rate. The biggest ﬂaw with the functionalized ILs is their high viscosity, which greatly hinders their application in CO2 absorption. Drawing on the successful experience of the mixed amines in CO2 uptake, combining functionalized ILs with water or MDEA solution is a good way to apply the functionalized ILs,18 since the amino acid ILs have high solubility in water. According to a US patent,25 [bmim][BF4] (1-butyl-3-methylimidazolium tetraﬂuoroborate) and [bmim][acetate] aqueous solutions and the hybrid of ionic liquid and MDEA (MEA) were used for the removal of CO2. However, the absorption of CO2 in these solvents was quite limited, since [bmim][BF4] and [bmim][acetate] were not very good for the absorption of CO2. In 2010, four functionalized amino acid ILs, tetramethylammonium glycinate ([N1111][Gly]), tetraethylammonium glycinate([N2222][Gly]), tetramethylammonium lysinate([N1111][Lys]), and tetraethylammonium lysinate([N2222][Lys]), were mixed with aqueous solution of MDEA to form a new type of absorbents (AAILs + MDEA + water) for CO2 capture. It was found that adding amino acid IL greatly enhanced CO2 absorption in the MDEA aqueous solution.18,19

ð2Þ

ð10Þ

Through eqs 4 10, the concentration of species can be calculated. In recent years, dicationic ionic liquids have attracted increasing attention for their excellent properties. Some imidazoliumbased dicationic ILs have been synthesized by Ohno and his coworkers,27 who have also synthesized pyridinium- and ammonium-based dicationic ILs with a polyether linker.28 Armstrong29 synthesized 39 kinds of dication ionic liquids, combining cations which contain two imidazolium or pyrrolidinium cations joined via hydrocarbon linkage chains of diﬀerent lengths (from 3 to 12 carbons long) and the anions including Br, NTf2, BF4, and PF6. The conductivity, glass transition temperature, melting point, surface tension, and shear viscosity in some dicationic ILs have been reported.30 Some dicationic ILs have been applied as stationary phases for gas chromatography,3133 as solvents for high-temperature organic reactions,34 as high-temperature lubricants,3538 and as electrolytes in secondary batteries39 and dye sensitized solar cells.40,41There are usually two ways to obtain functionalized ILs: cation or anion functionalized. As a medium for CO2 capture, higher absorption capacity of the functionalized ILs is always preferred. In the present work, for the ﬁrst time, dual-cations amino acid ILs were synthesized, in which the cation was equipped with two charges and the anions were functionalized with two appended amine groups. So far, most of the investigations on DILs29,37,4244 have focused on their synthesis or their basic physicochemical properties, while little work has been done 10628

dx.doi.org/10.1021/es201808e |Environ. Sci. Technol. 2011, 45, 10627–10633

Environmental Science & Technology

ARTICLE

Figure 1. Cation and anion of the target ionic liquids.

concerning the speciﬁc uses of DILs, especially the amino-acidbased DILs. In our work, DILs and their solutions were ﬁrst applied in the absorption of CO2 gas. Then, absorption of CO2 in the DIL aqueous solutions and the DIL + MDEA aqueous solutions was investigated respectively to explore the eﬀect of carbon structure and two positive charge centers on CO2 uptake.

2. EXPERIMENTAL SECTION 2.1. Materials. Tetramethylethylenediamine, 1-ethyl bromide, 1-butyl bromide, silver oxide, glycine, lysine, and valine of analytical grade were provided by Sinopharm Chemical Reagent. CO2 (purity above 99.99%) was purchased from Jiangsu Topgrand Petrochemical industrial gas Co., Ltd. 2.2. Synthesis of Dicationic Ionic Liquids. The synthesis process of amino-acid-based DILs was as follows: Step 1. Tetramethylethylenediamine and bromine alkane (methyl bromide or butyl bromide) reacted in ethanol at 80 °C for 48 h. The crude dication salt was then obtained through removing the solvent and unreacted reagents by rotary evaporator. After anhydration with vacuum drying over phosphorus pentoxide, pure dibromide salt was obtained eventually.

Step 2. Dication salt from Step 1 was mixed with potassium hydroxide in ethanol and stirred at room temperature for 24 h to obtain ditetraalkylammonium hydroxides. Since the solubility of potassium bromide in ethanol was quite low (0.034 g KBr/100 g ethanol at 25 °C), the precipitated potassium bromide could be removed by ﬁltration.

Step 3. The small amount of bromide ion in the solution generated from Step 2 was measured with the Mohr titration. One molar equivalent of the bromide salt dissolved in water would react with 1 mol equiv of solid silver oxide which was added intermittently. After being stirred in darkness at room temperature for 8 h, the upper supernatant was separated and treated by centrifugalization at 4500 rpm for 15 min. Silver bromide which was nearly insoluble in ditetraalkylammonium hydroxides solution could be removed.

Step 4. As the concentration of OH in the solution generated from Step 3 was calibrated by 1 mol/L standard hydrochloric acid, OH reacted with amino acid at mole ratio of 1:1 for 6 h at ambient temperature. Then water was removed through rotaevaporation and vacuum drying over phosphorus pentoxide to obtain the crude products.

Step 5. The crude products were extracted by using the mixture of acetonitrile and chloroform (the volume fraction was 50%). By removing solvent through rota-evaporation and vacuum drying over phosphorus pentoxide, respectively, six pure ionic liquids would be obtained. The structures of the anion and the cation for the DILs are shown in Figure 1. 2.3. Characterization of Dicationic Ionic Liquids. The structures of synthesized DILs were identified by 1 H NMR spectroscopy (Varian XL-300) and EA (Elementar vario EL II). The thermal stability was measured with TGA 10629

dx.doi.org/10.1021/es201808e |Environ. Sci. Technol. 2011, 45, 10627–10633

Environmental Science & Technology

ARTICLE

Figure 3. Absorption of CO2 in aqueous solution of C2(N114)2Gly2.

Figure 2. Schematic diagrams of gas absorption apparatus.

(Perkin-Elmer, 7 series thermal analysis system), and the samples were placed in aluminum pans and heated from room temperature to 500 at 10 °C min1 under nitrogen. The amount of water was measured by using Karl Fisher coulometric titration (Brinkmann Metrohm 756 KF Coulometer).The Br concentration, which was found to be (240 to 285) ppm, was measured by the Mohr titration. 2.4. Experiments of CO2 Absorption. Since all pure DILs are of high viscosity and low liquidity, they will be prepared as aqueous solutions for CO2 uptake like the previous treatment for the amino-acid-based ILs.18,19 Considering the economic benefit and the reactivity, glycine is an excellent anion provider and CO2 stabilizing agent. Therefore, C2(N114)2Gly2 and C2(N112)2Gly2 were chosen as functional absorbents for CO2 capture. Two kinds of solutions were prepared as absorbents: (1) aqueous solutions of DILs with weight fraction of 20%, 40%, and 80% and (2) 15% DIL + 15% MDEA aqueous solutions. The densities of all the solutions were measured by density meter Anton Paar DMA 5000, with a precision of 0.000001 g 3 cm3. The viscosities of the solutions were measured by HAAKE Rheostress 600. The absorption reactor is made of stainless steel and can afford pressures up to 120 bar. The whole test device consists of an isothermal water bath, an absorption equilibrium system, and the data receiving sections. This device has two vessels, and the storage vessel (176.3 mL) isolates CO2 before it contacts the samples in the absorption vessel (57.3 mL) which is equipped with a magnetic stirrer. The temperature of both vessels is controlled by the water bath. The pressure in the two vessels is monitored by using the pressure gauge (WIDEPLUS-8). Before the experiment, a known mass of the sample (about 4 g) was placed into the absorption vessel, and the whole device was degassed by a vacuum pump for at least 2 h. As shown in Figure 2, the two vessels were first separated by using a needle valve. When the storage vessel received the known amount of CO2, its initial pressure P0 was measured by a pressure gauge. Then, the needle valve between the two vessels was turned on to let the CO2 be introduced to the sample in the absorption vessel, and the CO2 gas would be absorbed by the liquid in the bottom of the vessel. After the equilibration was reached, the CO2 uptake could be calculated. A detailed description of the experiments has been presented in ref 19. The amount of absorbed CO2 can be calculated by the following equation nCO2 ¼

P0 VS PðVS þ VA VL Þ þ Pv VL RT

ð15Þ

Figure 4. Absorption of CO2 in aqueous solution of C2(N112)2Gly2.

where VS and VA denote the volume of the storage vessel and the absorption vessel respectively. VL represents the volume of liquid, and Pv represents the saturated vapor pressure of liquid. According to the Damping-Film Theory, for the isothermal absorption of gas, the relationship between the partial pressure of the gas versus the time can be described as follows lnðP Pe Þ=ðPo Pe Þ ¼ ut

ð16Þ

where P represents the partial pressure of CO2 at time t. Po and Pe denote the partial pressure at the beginning and at the absorption equilibrium, respectively. u stands for the apparent absorption rate constant based on pressure. Taking part in the Peng Robinson cubic equation of state, eq 16 will be written in the form of n-- the amount of CO2 lnðn ne Þ=ðno ne Þ ¼ Kt

ð17Þ

where, like u, K is also the apparent absorption rate constant. Noticeably, K strictly equals to u only for the ideal gas.

3. RESULTS AND DISCUSSION 3.1. The Characterization Results. The results of 1H NMR,

EA, TG, the water content, and the halide concentration of the sample are presented in the Supporting Information. The characteristic data are in good agreement with the expected structures and compositions. 3.2. Absorption of CO2 in Solutions of Various Concentrations. Absorption of CO2 by aqueous solutions of DILs with P0 = 97 KPa and T = 298 K is illustrated in Figures 3 and 4. It was found that the two types of DILs, C2(N114)2Gly2 and C2(N112)2Gly2 had the same absorption trend in aqueous solutions. Among the three aqueous solutions of DIL (20%, 40%, and 80%), the 40% DIL aqueous solution had a higher absorption rate and larger absorption capacity for CO2 than the 20% and 80% DIL aqueous solutions. A possible explanation is that compared with the 40% DIL aqueous solution, the 20% DIL 10630

dx.doi.org/10.1021/es201808e |Environ. Sci. Technol. 2011, 45, 10627–10633

Environmental Science & Technology

ARTICLE

Table 1. Physical Property of the Absorbents F (g 3 cm3)

μ (mP 3 s)

nCO2 (mol)

α (molCO2mol1Gly)

3.883

1.02243

4.58

0.0031909

0.6617

3.8995

1.04073

19.85

0.004296

0.4434

80%C2(N112)2Gly2

3.9923

1.07320

618

0.004254

0.2825

20% C2(N114)2Gly2

3.9968

1.02259

5.32

0.002950

0.6593

40% C2(N114)2Gly2

4.0012

1.04103

20.3

0.004107

0.4032

80% C2(N114)2Gly2

4.0037

1.074193

630

0.004082

0.2794

absorbents

m (g)

20% C2(N112)2Gly2 40% C2(N112)2Gly2

Figure 5. Apparent absorption rates in aqueous solution of C2(N114)2Gly2.

solution had less effective absorbent (RNH2COO), while the 80% DIL solution had higher viscosity which greatly hindered the absorption of CO2. Generally, the viscosity of aqueous solutions increased with the uptake CO2, further weakening the absorption performance. The absorption capacity of the DIL aqueous solutions was compared in Table 1, which showed that the 20% aqueous solution had the highest CO2 load of 0.66 molCO2/molgly and the 40% DIL aqueous solution had the largest absorbing amount, while the viscosity of the 80% DIL aqueous solution was 30 times higher than that of the 40% DIL aqueous solution. In general, the absorption rate of CO2 in IL aqueous solutions is mainly inﬂuenced by the concentration and viscosity of the solution. High concentration of IL or low viscosity of solution both accelerates the absorption speed. For the IL aqueous solution, the viscosity is dominated by the IL’s concentration, and a higher concentration of the IL leads to a higher viscosity of the solution. Therefore, the eﬀect of the IL’s concentration on the absorption rate is conﬂicting. As is displayed in Figure 5, for C2(N114)2Gly2, among the three diﬀerent concentrations, the 40% DIL solution had the highest apparent absorption rate. The apparent absorption rate of the 80% DIL solution was close to that of the 20% solution during the early 6 min. As time went on, the apparent absorption rate of the 80% solution got gradually higher than that of the 20% DIL solution. A possible explanation is that the values of the concentration and viscosity of the 40% DIL aqueous solution were ﬁtter for the absorption. As for the 20% and 80% DIL solutions, concentration played the leading role in the absorption, especially when the absorption time reached over 6 min. Noticeably, the case of C2(N112)2Gly2 (see Figure 6) was the same with C2(N114)2Gly2. 3.3. Effect of Different Lengths of Alkyl Chain. The absorption of CO2 in 40% aqueous solutions of C2(N114)2Gly2 and C2(N112)2Gly2 were compared in Figure 7 and Figure 8, showing that the amount of absorbed CO2 was a little larger and the uptake rate a little higher for C2(N112)2Gly2 aqueous solution than those of C2(N114)2Gly2. This could be explained that there were more molecules of C2(N112)2Gly2 under the same weight,

Figure 6. Apparent absorption rates in aqueous solution of C2(N112)2Gly2.

Figure 7. Absorption of CO2 in 40% DIL aqueous solution.

Figure 8. Apparent absorption rates of the absorbents.

rising from its lighter molecular weight. As shown in Figure 10, the molar absorption load of CO2 in the C2(N114)2Gly2 aqueous solution was similar to that of the C2(N112)2Gly2 aqueous solution. Muldoon et al.28 have reported the influence of alkyl chain length on absorption load. They compared the solubility of CO2 in [C8H4F13mim][Tf2N] and [C6H4F9mim][Tf2N] and found that the solubility of CO2 increased with the rise in the alkyl chain length, but the changes were not obvious. 10631

dx.doi.org/10.1021/es201808e |Environ. Sci. Technol. 2011, 45, 10627–10633

Environmental Science & Technology

Figure 9. Absorption capacity of CO2 in 40% ILs aqueous solution.

ARTICLE

pressure in the same way. When the pressure was less than 25 kPa, the CO2 load increased most rapidly with the rising pressure; when the pressure was between 25100 kPa, the increase of CO2 load with the pressure was moderate; when the pressure was over 100 kPa, there was only slight increase in CO2 load with the still rising pressure. Obviously, the absorption performance of CO2 in these solutions was similar to each other when the pressure was less than 20 kPa. When the pressure reaches over 20 kPa, the CO2 load of DILs mixtures were higher than that of MIL mixtures and the difference became more noticeable with the further increasing pressure, indicating the potential use of DIL in CO2 uptake under high pressure. This proved that the charge of ionic liquids had significant influence on the absorption of CO2. In certain concentration range, CO2 solubility increases with the rise in the number of electric charges. Practically, according to Schumpe’s model,45 the physical solubility of gas in solution can be affected by the concentration and the specific parameters of ions.

’ ASSOCIATED CONTENT

bS Figure 10. Solubility of CO2 in 15% ILs + 15% MDEA aqueous solutions.

3.4. Effect of Different Cations on the Absorption Capacity of Aqueous Solutions. Due to the excellent absorption proper-

ties of the 40% DIL aqueous solution, its concentration was chosen to investigate the solubility of CO2 under different pressure. The absorption load of dicationic and monocationic ILs in the 40% aqueous solution were compared in Figure 9. Since the DIL has two anions of glycine, the absorption loads of CO2 were calculated by per mole anion. The solubility of CO2 in aqueous solution could approach one mole CO2 per mole DILs when the pressure is close to 250 kpa. It should be noted that the mole load of [N1111] [Gly] aqueous solution was much lower than those of C2(N114)2Gly2 and C2(N112)2Gly2, indicating the significant influence of cation characteristic on CO2 absorption in the aqueous solutions. That is, the dication of two positive charge centers could more effectively attract anions which contained the absorbed CO2, such as RNH2COO. The absorption load of CO2 was almost the same in C2(N114)2Gly2 and C2(N112)2Gly2 aqueous solutions. 3.5. Effect of Different Cations on the Absorption Capacity of Mixed Absorbent. At present, an effective chemical method for CO2 uptake usually employs aqueous solutions of alkanolamine, including monoethanolamine (MEA), diethanolamine (DEA), N-methyldiethanolamine (MDEA), etc. Among them, MDEA of high load performance is mainly used, but the slow absorption rate hindered its application in the industrial process. It has been proved that adding functionalized amino acid ionic liquids into MDEA aqueous solution could significantly improve the absorption performance of MDEA.20 The absorption characteristics of 15% [N1111][Gly] + 15% MDEA aqueous solution have been investigated in detail.18,19 In the present work, the absorption characteristics of the DILs + MDEA aqueous solutions were also examined and compared with those of MIL ([N1111][Gly]). As shown in Figure 10, the CO2 load of all the IL + MDEA solutions increased with the rise in the equilibrium

Supporting Information.

The results of 1H NMR, EA, TG, the water content, and the halide concentration of the sample. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +86-25-83596665. Fax: +86-25-83593772. E-mail: [email protected] (F.Z.). Phone: +86-25-835952222. Fax: +8625-83595222. E-mail: [email protected] (A.M.L.).

’ ACKNOWLEDGMENT The authors are grateful for the ﬁnancial support from National Natural Science Foundation of China (No. 200906046), National Nature Science Fund for Distinguished Young Scientists (Grant No. 50825802), Granted Special Grade of the Financial Support from the China Postdoctoral Science Foundation (No. 201003567), and National Nature Science Foundation of China (21076101), and Fundamental Research Funds for the Central Universities (No.1114020504). ’ REFERENCES (1) George, M.; Weiss, R. G. Chemically reversible organogels: Aliphatic amines as 00 latent00 gelators with carbon dioxide. J. Am. Chem. Soc. 2001, 123 (42), 10393–10394. (2) Arakawa, H.; Aresta, M.; Armor, J. N.; Barteau, M. A.; Beckman, E. J.; Bell, A. T.; Bercaw, J. E.; Creutz, C.; Dinjus, E.; Dixon, D. A; et al. Catalysis research of relevance to carbon management: Progress, challenges, and opportunities. Chem. Rev. 2001, 101 (4), 953–996. (3) Song, C. S. Global challenges and strategies for control, conversion and utilization of CO2 for sustainable development involving energy, catalysis, adsorption and chemical processing. Catal. Today 2006, 115 (14), 2–32. (4) Amenomiya, Y. Methanol synthesis from CO2+ H2 II. Copperbased binary and ternary catalysts. Appl. Catal. 1987, 30 (1), 57–68. (5) Aresta, M.; Dibenedetto, A. Utilisation of CO2 as a chemical feedstock: opportunities and challenges. Dalton Trans. 2007, 28, 2975– 2992. (6) Sakakura, T.; Choi, J. C.; Yasuda, H. Transformation of carbon dioxide. Chem. Rev. 2007, 107 (6), 2365–2387. 10632

dx.doi.org/10.1021/es201808e |Environ. Sci. Technol. 2011, 45, 10627–10633

Environmental Science & Technology (7) Yu, K. M. K.; Curcic, I.; Gabriel, J.; Tsang, S. C. E. Recent Advances in CO2 Capture and Utilization. ChemSusChem. 2008, 1 (11), 893–899. (8) Navaza, J. M.; Gomez-Diaz, D.; La Rubia, M. D. Removal process of CO2 using MDEA aqueous solutions in a bubble column reactor. Chem. Eng. J. 2009, 146 (2), 184–188. (9) Bishnoi, S.; Rochelle, G. T. Absorption of carbon dioxide in aqueous piperazine/methyldiethanolamine. AIChE J. 2002, 48 (12), 2788–2799. (10) Bonenfant, D.; Mimeault, M.; Hausler, R. Determination of the structural features of distinct amines important for the absorption of CO2 and regeneration in aqueous solution. Ind. Eng. Chem. Res. 2003, 42 (14), 3179–3184. (11) Park, S. H.; Lee, K. B.; Hyun, J. C.; Kim, S. H. Correlation and prediction of the solubility of carbon dioxide in aqueous alkanolamine and mixed alkanolamine solutions. Ind. Eng. Chem. Res. 2002, 41 (6), 1658–1665. (12) Huang, J. H.; Ruther, T. Why are Ionic Liquids Attractive for CO2 Absorption? An Overview. Aust. J. Chem. 2009, 62 (4), 298–308. (13) Shiﬂett, M. B.; Yokozeki, A. Solubilities and diﬀusivities of carbon dioxide in ionic liquids: bmim PF6 and bmim BF4. Ind. Eng. Chem. Res. 2005, 44 (12), 4453–4464. (14) Blanchard, L. A.; Hancu, D.; Beckman, E. J.; Brennecke, J. F. Green processing using ionic liquids and CO2. Nature 1999, 399 (6731), 28–29. (15) Aki, S. N. V. K.; Mellein, B. R.; Saurer, E. M.; Brennecke, J. F. High-pressure phase behavior of carbon dioxide with imidazolium-based ionic liquids. J. Phys. Chem. B 2004, 108 (52), 20355–20365. (16) Blanchard, L. A.; Gu, Z.; Brennecke, J. F. High-pressure phase behavior of ionic liquid/CO2 systems. J. Phys. Chem. B 2001, 105 (12), 2437–2444. (17) Bates, E. D.; Mayton, R. D.; Ntai, I.; Davis, J. H. CO2 capture by a task-speciﬁc ionic liquid. J. Am. Chem. Soc. 2002, 124 (6), 926–927. (18) Wang, Y. T.; Fang, C. G.; Zhang, F.; Geng, J.; Zhang, Z. B. The performances of CO2 absorption in mixed aqueous solution of MDEA and amino acid ionic liquids. CIESC J. 2009, 60 (11), 2781–2786. (19) Zhang, F.; Fang, C. G.; Wu, Y. T.; Wang, Y. T.; Li, A. M.; Zhang, Z. B. Absorption of CO2 in the aqueous solutions of functionalized ionic liquids and MDEA. Chem. Eng. J. 2010, 160 (2), 691–697. (20) Fukumoto, K.; Yoshizawa, M.; Ohno, H. Room temperature ionic liquids from 20 natural amino acids. J. Am. Chem. Soc. 2005, 127 (8), 2398–2399. (21) Gurkan, B.; Goodrich, B.; Mindrup, E.; Ficke, L.; Massel, M.; Seo, S.; Senftle, T.; Wu, H.; Glaser, M.; Shah, J. Molecular Design of High Capacity, Low Viscosity, Chemically Tunable Ionic Liquids for CO2 Capture. J. Phys. Chem. Lett. 2010, 1, 3494–3499. (22) Jiang, Y. Y.; Wang, G. N.; Zhou, Z.; Wu, Y. T.; Geng, J.; Zhang, Z. B. Tetraalkylammonium amino acids as functionalized ionic liquids of low viscosity. Chem. Commun. 2008, 4, 505–507. (23) Zhang, J. M.; Zhang, S. J.; Dong, K.; Zhang, Y. Q.; Shen, Y. Q.; Lv, X. M. Supported absorption of CO2 by tetrabutylphosphonium amino acid ionic liquids. Chem.—Eur. J. 2006, 12 (15), 4021–4026. (24) Zhang, Y. Q.; Zhang, S. J.; Lu, X. M.; Zhou, Q.; Fan, W.; Zhang, X. P. Dual Amino-Functionalised Phosphonium Ionic Liquids for CO2 Capture. Chem.—Eur. J. 2009, 15 (12), 3003–3011. (25) Chinn, D.; Vu, D.; Driver, M. S.; Boudreau, L. C. CO2 removal from gas using ionic liquid absorbents, US 2005/0129598 A1. In. United States; 2005. (26) Portugal, A. F.; Derks, P. W. J.; Versteeg, G. E.; Magalhaes, F. D.; Mendes, A. Characterization of potassium glycinate for carbon dioxide absorption purposes. Chem. Eng. Sci. 2007, 62 (23), 6534–6547. (27) Ito, K.; Nishina, N.; Ohno, H. Enhanced ion conduction in imidazolium-type molten salts. Electrochim. Acta 2000, 45 (89), 1295–1298. (28) Muldoon, M. J.; Anderson, J. L.; Dixon, J. K. Improving carbon dioxide solubility in ionic liquids. J. Phys. Chem. B. 2007, 111, 9001– 9009. (29) Anderson, J. L.; Ding, R. F.; Ellern, A.; Armstrong, D. W. Structure and properties of high stability geminal dicationic ionic liquids. J. Am. Chem. Soc. 2005, 127 (2), 593–604.

ARTICLE

(30) Shirota, H.; Mandai, T.; Fukazawa, H.; Kato, T. Comparison between Dicationic and Monocationic Ionic Liquids: Liquid Density, Thermal Properties, Surface Tension, and Shear Viscosity. J. Chem. Eng. Data 2011. (31) Anderson, J. L.; Armstrong, D. W. Immobilized ionic liquids as high-selectivity/high-temperature/high-stability gas chromatography stationary phases. Anal. Chem. 2005, 77 (19), 6453–6462. (32) Qi, M.; Armstrong, D. W. Dicationic ionic liquid stationary phase for GC-MS analysis of volatile compounds in herbal plants. Anal. Bioanal. Chem. 2007, 388 (4), 889–899. (33) Huang, K.; Han, X.; Zhang, X.; Armstrong, D. W. PEG-linked geminal dicationic ionic liquids as selective, high-stability gas chromatographic stationary phases. Anal. Bioanal. Chem. 2007, 389 (7), 2265– 2275. (34) Han, X.; Armstrong, D. W. Using geminal dicationic ionic liquids as solvents for high-temperature organic reactions. Org. Lett. 2005, 7 (19), 4205–4208. (35) Jin, C. M.; Ye, C.; Phillips, B. S.; Zabinski, J. S.; Liu, X.; Liu, W.; Jean’ne, M. S. Polyethylene glycol functionalized dicationic ionic liquids with alkyl or polyﬂuoroalkyl substituents as high temperature lubricants. J. Mater. Chem. 2006, 16 (16), 1529–1535. (36) Zeng, Z.; Phillips, B. S.; Xiao, J. C.; Shreeve, J. M. Polyﬂuoroalkyl, polyethylene glycol, 1, 4-bismethylenebenzene, or 1, 4-bismethylene-2, 3, 5, 6-tetraﬂuorobenzene bridged functionalized dicationic ionic liquids: synthesis and properties as high temperature lubricants. Chem. Mater. 2008, 20 (8), 2719–2726. (37) Yu, G. Q.; Yan, S. Q.; Zhou, F.; Liu, X. Q.; Liu, W. M.; Liang, Y. M. Synthesis of dicationic symmetrical and asymmetrical ionic liquids and their tribological properties as ultrathin ﬁlms. Tribol. Lett. 2007, 25 (3), 197–205. (38) Palacio, M.; Bhushan, B. Molecularly thick dicationic ionic liquid ﬁlms for nanolubrication. J. Vac. Sci. Technol., A 2009, 27, 986. (39) Zhang, Z. X.; Zhou, H. Y.; Yang, L.; Tachibana, K.; Kamijima, K.; Xu, J. Asymmetrical dicationic ionic liquids based on both imidazolium and aliphatic ammonium as potential electrolyte additives applied to lithium secondary batteries. Electrochim. Acta 2008, 53 (14), 4833–4838. (40) Kim, J. Y.; Kim, T. H.; Kim, D. Y.; Park, N. G.; Ahn, K. D. Novel thixotropic gel electrolytes based on dicationic bis-imidazolium salts for quasi-solid-state dye-sensitized solar cells. J. Power Sources 2008, 175 (1), 692–697. (41) Zafer, C.; Ocakoglu, K.; Ozsoy, C.; Icli, S. Dicationic bisimidazolium molten salts for eﬃcient dye sensitized solar cells: Synthesis and photovoltaic properties. Electrochim. Acta 2009, 54 (24), 5709–5714. (42) Han, X. X.; Armstrong, D. W. Using geminal dicationic ionic liquids as solvents for high-temperature organic reactions. Org. Lett. 2005, 7 (19), 4205–4208. (43) Liu, Q. B.; van Rantwijk, F.; Sheldon, R. A. Synthesis and application of dicationic ionic liquids. J. Chem. Technol. Biotechnol. 2006, 81 (3), 401–405. (44) Zhang, Z.; Yang, L.; Luo, S.; Tian, M.; Tachibana, K.; Kamijima, K. Ionic liquids based on aliphatic tetraalkylammonium dications and TFSI anion as potential electrolytes. J. Power Sources 2007, 167 (1), 217–222. (45) Schumpe, A. The estimation of gas solubilities in salt solutions. Chem. Eng. Sci. 1993, 48 (1), 153–158.

10633

dx.doi.org/10.1021/es201808e |Environ. Sci. Technol. 2011, 45, 10627–10633

ARTICLE pubs.acs.org/est

Exergy Efficiency in Industry: Where Do We Stand? Robert U. Ayres,*,† Laura Talens Peiro,† and Gara Villalba Mendez‡ † ‡

INSEAD- Campus Europe, Boulevard de Constance, 77305 Fontainebleau, France Department of Chemical Engineering, Ediﬁci Q, Universitat Autonoma de Barcelona (UAB), ES-08193 Bellaterra, Barcelona, Spain

bS Supporting Information ABSTRACT: Eﬃciency is a term generally used to determine how well a system performs. However, eﬃciency can have diﬀerent meanings and, unaccompanied by a formal deﬁnition or taken out of context, can lead to serious misconceptions. In many oﬃcial publications, eﬃciency is calculated as the ratio of useful output to energy input. This measure reﬂects the ﬁrst law of thermodynamics (conservation of energy) but does not reﬂect the potential for improvement. A better measure, that also reﬂects the second law of thermodynamics, is the ratio of the potential useful (exergy) output to the potential useful (exergy) input. We estimate second law eﬃciencies for the inorganic and organic chemical industries to be 29% and 35% respectively. We also estimate the eﬃciency of the U.S. industry sector as a whole to be 37.6%, as compared to only 7.7% for the overall U.S. economy. These ﬁgures are far lower than the “ﬁrst law” ﬁgures published by the U.S. Department of Energy (80% for industry and 42.5% overall) and they imply a signiﬁcant potential for improvement.

1. INTRODUCTION The main subject of this paper is “eﬃciency”, a term often used without a formal deﬁnition, as if the term really needed none. Unfortunately, that is not the case. In fact, the term is widely used in three diﬀerent ways, the more common of which is seriously misleading. For reasons not relevant to this paper, the U.S. Energy Information Agency (USEIA) has published a diagram in its annual energy review since 1950, as shown in Figure 1 and Supporting Information (SI 1). The 2008 version is almost identical to the one for 1970, except that all the ﬂows are proportionally bigger.1 According to this chart, total U.S. energy consumption in 2008 was 99.2 quads (quadrillion BTU) of which 42.15 (42.5%) are classed as “useful”. The remaining 57.5% is classed as “rejected” energy, of which 27.39 quads are from electricity generation and 20.90 quads are from transportation. Remarkably, this chart only shows 4.78 quads of “rejected” energy from industry, as compared to 19.15 quads classed as “useful”; 1.71 quads rejected from commerce as compared to 6.86 “useful” and 2.29 quads rejected from residences (households) as compared to 9.18 quads that were supposed to be “useful”. The fact that the term “eﬃciency” is not used in the USEIA ﬂowchart does not justify the misleading underestimation of potential for future energy (exergy) savings. The ratio of “rejected energy” to “total energy” is, of course, the loss fraction. Subtracting the loss fraction from unity yields a number that is easily interpreted as a measure of eﬃciency. We show these eﬃciency measures by sector for the USEIA energy ﬂowcharts since 1950 in Table 1 below. The deﬁnition applied by the USEIA is a ratio between “useful” (in some sense) output and energy input. This is now r 2011 American Chemical Society

called “ﬁrst law” eﬃciency because it reﬂects the fact that massenergy is conserved, so the mass and energy of the inputs and outputs must be the same. However, this deﬁnition is misleading because it only distinguishes between “useful” and “rejected” energy but does not tell us how much of the theoretically “useful” output is actually available to do work (is exergy) and how much of that output is unavailable (is anergy) due to entropic irreversibility losses. This point was clariﬁed by a summer study sponsored by the American Physical Society (APS) in 1975.2 “Second-law eﬃciency”, as deﬁned in the report of the APS summer study is now termed “exergy eﬃciency” in most textbooks.3 Exergy eﬃciency is the ratio of potentially useful (exergy) output to potentially useful (exergy) input. The reference to “second law” reﬂects the fact that losses of potential work result from irreversibilities due to the second law of thermodynamics (e2). Second-law eﬃciency is usually less than ﬁrst law eﬃciency. For instance, a manufacturer may describe a gas-ﬁred hot water boiler as “80% eﬃcient” if for each unit of heat from the ﬂame 0.8 units go into the water and 0.2 units are “rejected” into the exhaust pipe. The ratio of minimum energy needed to raise the temperature of water from 25 to 100 °C, to the heat produced by the ﬂame is 9.92%. Hence, the hot water boiler is, in fact, very ineﬃcient (see SI 2 for examples). The numbers in Table 1 are derived from the Sankey diagrams by USEIA, but most of the “useful” ﬁgures in those diagrams Received: June 27, 2011 Accepted: November 1, 2011 Revised: October 31, 2011 Published: November 01, 2011 10634

dx.doi.org/10.1021/es202193u | Environ. Sci. Technol. 2011, 45, 10634–10641

Environmental Science & Technology

ARTICLE

Figure 1. U.S. Energy Flow diagram for 2008. Source: ref 4.

Table 1. U.S. Estimated Energy “Eﬃciencies”. Source: Authors Based on USEIAa sector

1950

1970

1990

2000

2008

electricity generation

25

36

33

31

0.32

residential and commercial

73

75

75

75

0.80

industrial

72

75

75

80

0.80

transport

26

26

25

20

0.24

aggregate

50

50

44

38

0.42

a

Source: LLNL= Lawrence Livermore National Laboratories; DOE = United States Department of Energy.

must have been based on nothing more than guesswork, at best. Only the eﬃciency of electric power generation (top line) is roughly accurate. None of the other eﬃciency numbers can be derived from published data. As regards transport, it seems likely that the “useful” numbers are based on the textbook eﬃciency of internal combustion engines operating in ideal conditions, viz. no part-load penalty, no stop-start penalty, no allowance for power train losses or parasitic loads and no adjustment for the fact that the purpose of transportation is to move people and goods, not vehicles. Calculations made by Dewulf and Van Langenhove of exergy service eﬃciency for several transport modes, taking all of these limitations into account, yield shockingly diﬀerent results.5 For trucks, cars, and electric trains, the eﬃciency of the service is 0.37%, 0.44%, and 1.99% respectively (for a distance greater than 135 km at a speed of 80 km/h). Authors did not do a calculation for aircraft because the transport service provided is diﬀerent (higher speed). However for obvious reasons, it is unlikely that

aircraft payload eﬃciency is as high as that for electric trains with a high load-factor. Given the predominance of highway vehicles, the overall service eﬃciency of the transport sector (excluding the production of fuels) is less than 1%, showing that the potential for improvement is very large. As regards the residential-commercial (buildings) sector, where most energy consumption is for space-heating and lighting, analysis in the APS summer study, indicates that the real (second-law) eﬃciency of most space-heating systems (by steam or hot water) is around 5% (electric heating is less), whereas water heating and lighting rarely achieve 10%.2 The same eﬃciency value for lighting and 6% for space heating are given for Norway in 1995.6 A recent paper suggests that the exergy eﬃciencies of a condensing boiler and a conventional boiler using LNG, and an external airair heat pump are estimated to be between 7% and 9%.7 In Japan, the exergy eﬃciency in the residential and commercial sectors are 6.33% and 5.74%, respectively.8 For our purposes, it is safe to say that the overall eﬃciency of the residentialcommercial sector is less than 10%, rather than 80% as implied by Table 1. It is worth noting that Reistad’s eﬃciency estimates (1975), based on second-law analysis were far more realistic. They were as follows: residentialcommercial (13.7%), industrial (36%) and transportation (20%), for an overall eﬃciency of 21% for the U.S. economy.9 Reistad’s estimate for transportation was too high, because he considered only the eﬃciency of the carriers (vehicles), without allowing for the payload. Several authors have undertaken studies comparing the exergy eﬃciency of diﬀerent countries, notably Wall and Ertesvag. Wall estimated the exergy eﬃciency of Sweden (1980) as 22%, Japan (1985) as 19%, and Italy (1990) as 17%.1012 Using Wall’s approach, 10635

dx.doi.org/10.1021/es202193u |Environ. Sci. Technol. 2011, 45, 10634–10641

Environmental Science & Technology Ertesvag estimated the eﬃciencies for Sweden as 22% (1980), Brazil as 23% (1987), Italy as 17% (1990), Turkey as 13% (1995), and Norway as 24% (1995).13 The relatively high eﬃciencies for Norway and Sweden are partly due to high contribution from hydroelectricity. We focus hereafter on industrial energy (exergy) eﬃciency, with special emphasis on the chemical sector. The Sections 2 and 3 deal with technicalities and deﬁnitions. Section 4 presents numerical results for the production of inorganic and organic chemicals. Section 5 shows exergy eﬃciency values for other industries and estimates the eﬃciency of the U.S. economy.

2. EXERGY With the convergence of policies for resource optimization and energy conservation, the idea of analyzing the eﬃciency of any process or system in terms of inputs and outputs becomes attractive. A brief word about economics: It is traditional to think of physical capital in mass or value terms, but on deeper reﬂection, capital goods can be divided into two categories: active and passive. Passive capital consists of material goods that provide a service merely by existing in a certain place or situation. The obvious examples of passive capital goods are roads, bridges, tunnels, structures, wires, pipes, and containers. Active capital consists of engines and machines. Active capital requires available energy (exergy) to function. All production (and consumption) in the economic system depend on active capital and available energy. It is important to understand from the outset that not all energy is available. Technically, available energy, now called exergy, is deﬁned as the maximum amount of work that can be done by a subsystem as it approaches thermodynamic equilibrium with its surroundings by a sequence of reversible processes (for example, Szargut et al 198817). Equilibrium is a homogeneous unchanging state in which there are no gradients. This implies uniformity of temperature, pressure, density, chemical composition as well as uniform gravitational and electro-magnetic ﬁelds. In practice, the equilibrium end-states for all chemical (and other) processes may be the atmosphere, the oceans or the earth’s crust, depending on the chemical elements involved. Exergy, as potential work, is therefore deﬁnable for any subsystem that is not in thermodynamic equilibrium with its surroundings (air, ocean water, and soil). It is also deﬁnable as that component of energy potentially capable of doing work, in contrast with energy not capable of doing work (anergy). Fuels can be thought of as exergy carriers (along with process steam, ﬂywheels, storage batteries, and electricity). Energy is conserved in every action or reaction (ﬁrst law of thermodynamics) whereas exergy is not conserved. Exergy is destroyed when work is done, whereas anergy (like entropy) increases during every process or activity. There are four types of exergy: kinetic, potential (e.g., in the gravitational ﬁeld of the earth), physical (based on temperature and pressure, with respect to the local environment) and chemical exergy (e.g., the heat output of an exothermic reaction). Electricity is almost pure exergy, as it can be totally converted to high temperature heat in an electric arc furnace or to mechanical work by an electric motor. On the other hand, low temperature heat has very little exergy content, because only a small fraction can be recovered as useful work. From this perspective, all natural and industrial materials can be characterized in terms of exergy “content”. In short all

ARTICLE

materials are exergy carriers, and this fact is particularly relevant to eﬃciency calculations in industry. Chemically inert materials such as water, carbon dioxide, silicon dioxide, iron oxide (in fact, most metal oxides) have low exergy content. Fuels and biomass, as well as reactive chemicals like industrial acids and caustic soda have high exergy content.

3. EFFICIENCY AND EFFECTIVENESS It is logical, at ﬁrst glance, to suppose that the eﬃciency of any activity or process (including chemical or metallurgical processes) can be calculated by comparing the energy content of the useful output with the energy content of the inputs, including materials. This is known as “ﬁrst law” eﬃciency, using the terminology of the APS study, because it reﬂects the fact that the total energy input is equal to the total energy of the output, partly useful and partly “rejected”.2 energy output ð1Þ e1 ¼ energy inputs An obvious extension of this deﬁnition is “second-law” eﬃciency, deﬁned as the ratio of exergy embodied in “useful” products to the total exergy of all inputs. As noted already, the term “second law” reﬂects the fact that the exergy embodied in most energy ﬂows (whether inputs or outputs) is less than the energy because some of the energy is not “useful”. For example, the exergy of process steam is much less than the energy (heat content or enthalpy) of the steam; because the steam is at a ﬁnite temperature (the two measures only coincide exactly at inﬁnite temperature). More generally, second law eﬃciency reﬂects the fact that there are losses, due to irreversibilities in every process. e2 ¼

usef ul exergy output total exergy inputs

ð2Þ

As emphasized by the APS group “second law” eﬃciency is generally a much better measure of potential for improvement than ﬁrst law eﬃciency (there is no diﬀerence in the case of electric power generation because the output is pure exergy). However, it is easy to demonstrate that “ﬁrst law” eﬃciency (as applied to a piece of equipment such as a boiler) can be much greater than “second law” eﬃciency. The APS study considered a wide range of examples of energy eﬃciency, including heating systems, refrigeration, heat pumps, engines, generators, and motor vehicles. The study did not consider chemical or metallurgical processes. This paper, by contrast, considers process eﬃciency, almost exclusively. In this context, it is important to bear in mind that chemical reactions are of two kinds. Endothermic reactions, such as ammonia synthesis, consume (destroy) exergy in a process, so the exergy content of the useful product becomes the numerator of the eﬃciency fraction. Exothermic reactions, like combustion, generate heat which may be useful in driving a second reaction, or may be wasted. In exothermic reactions there are usually energy-rich byproducts in the output stream. Blast furnace gas from iron smelting is an example. For purposes of calculating eﬃciency, it depends on whether the energy-rich byproduct (say B.F. gas) is subsequently utilized or not. In addition, there may be unreacted inputs (such as nitrogen from the air) that reappear in the output. In this case, it may be convenient to subtract the exergy content of these unreacted compounds from both the numerator and the denominator, or it may be more practical to count the unreacted inputs as byproducts. 10636

dx.doi.org/10.1021/es202193u |Environ. Sci. Technol. 2011, 45, 10634–10641

Environmental Science & Technology

ARTICLE

In exothermic reactions the diﬀerence between input exergy and output exergy is divided between byproduct and waste products. But what exactly is a waste? After all, some of it is usually heat, and some heat can be recovered, depending on temperature and equipment. Some of it may be other chemical compounds that have little economic value (chlorination processes are examples). In cases where combustion (or oxidation) is not involved, the deﬁnition of a waste is case dependent, and may also be equipment dependent. In some cases, where the inputs can be divided explicitly between “feedstocks” and “utilities”, but where the process is otherwise undeﬁned, one can calculate second law eﬃciency as the exergy increase in the product (as compared to the exergy of the feedstock, excluding utilities), divided by the exergy of feedstocks consumed to drive the process.14 This applies, for instance, to petroleum reﬁning taken as a whole. e2 ¼

increase exergy between f eedstocks and usef ul products exergy f uel consumed ð3Þ

Many chemical reactions consist of both exothermic and endothermic steps that cannot be physically separated. The exothermic stage produces heat that can drive a later reaction. Examples include ammonia synthesis, methanol synthesis, and carbothermic smelting. For this reason, it is often meaningless to calculate the exergy output of an exothermic reaction in isolation, because the exergy (as heat) may be diﬃcult to recover and utilize, whereas it may be utilized quite eﬃciently to drive a subsequent endothermic reaction. In fact, this situation is the norm in the chemical industry. In this paper, we apply the second law eﬃciency calculation to the chemical industry as a whole. As noted above, the second law eﬃciency is deﬁned as the ratio of the exergy content of the products to the exergy content of the inputs to the process. The exergy inputs are the sum of raw materials or feed-stocks and utilities (mainly electricity) required by the industry. Accounting for all those inputs together helps us to compare processes and product substitutes in terms of resource use, waste and emissions generation.15,16 For calculating second law eﬃciency, at the product level, it is necessary to have details of the material balances, stream composition and conditions, and ﬂow-sheets of the process under study. The exergy of a product and materials is calculated using the reference state deﬁned by Szargut et al. namely a reference temperature (298 K), pressure (1 atm) and an average conventional composition of the Earth’s litho-, hydro-, and atmosphere.17 The exergy of utilities is based on the amount of electricity, natural gas, diesel, water, and steam supplied to the processes. For electricity, 1 MJ of electrical energy equals 1 MJ of exergy.13 Calculating the eﬃciency of a process can help quantify how changes in production parameters may change yields and/or reduce emissions and exergy losses.

4. THE EFFICIENCY OF THE U.S. CHEMICAL INDUSTRY Quantitative data about industrial chemicals can be estimated with reasonable accuracy from industry production statistics. In the U.S., production statistics were published annually by the U.S. International Trade Commission (USITC) prior to the mid 1990s. However, in recent years the availability of production data to the public has become progressively worse as the number

of producers of high volume chemicals has declined. Hence, this paper used production statistics from USITC in 19911993 (there are other sources of information such Chemical Engineering News, but unfortunately they do not include statistical production data of hydrocarbons). USITC included production data for virtually all industrial chemicals, including intermediates. While production data for individual chemicals has ﬂuctuated signiﬁcantly since then, we think that the overall eﬃciencies have not changed by more than a few percent at most. To compare inputs and outputs for the whole sector, and avoid double counting and ensure the consistency of results, the list can be divided into two groups: (1) basic chemicals, which are made directly from raw materials, and (2) all others, including intermediates. For such classiﬁcation, some knowledge of the industry is required. For instance, sulfuric acid is mainly made by burning sulfur, but is now also produced as a byproduct of copper smelting. Hydrochloric acid, is no longer made from salt but as byproduct of many downstream chlorination processes and thus is not included as a “basic” inorganic chemical. By combining production and process information of the basic inorganic and organic chemicals, raw material inputs are quantiﬁed in mass terms, whence a material balance by elements (C, H, O, N, Cl, S, Na, Ca, etc) can be performed. The diﬀerence between mass inputs and useful outputs are wastes and emissions. For the industry as a whole, wastes are characterized by elemental composition but they can be estimated approximately as a mix of compounds (CO2, CO, H2O, NaCl, CaSO4 etc.) based on knowledge of process reactions. For the production of sulfuric acid from sulfur, the inputs to account for are sulfur, oxygen, and hydrogen. For chlorine and sodium hydroxide, the inputs are sodium chloride and water, and for ammonia, the inputs are methane and air. The amount of the input materials is calculated based on their content in the ﬁnal product. SI 3 shows the elemental mass balance for the production of basic inorganic and organic chemicals in U.S. in 1991. The energy requirement for the production of the inorganic chemicals sector as a whole is estimated based on the standard unit process descriptions for sulfuric acid, ammonia, chlorine and sodium hydroxide, multiplied by the tonnages reported. These calculations have been checked against reported energy consumption by subsectors for sulfuric acid (SIC 2819), ammonia (SIC 2873) and chlorine (SIC 2812) published in the Manufacturing Energy Consumption Survey (MECS) by the USEIA.18 When the material balance is closed and energy inputs are estimated, the exergy of inputs and outputs can be calculated using the procedure explained in Section 3. Once the exergy of inputs and outputs is known, the exergy balance follows directly and the second law eﬃciency can be calculated. The diﬀerence between the exergy of inputs and outputs are wastes and losses from the system. The basic inorganic chemicals made from raw materials are sulfuric acid, ammonia, chlorine and caustic soda. The total production of these four chemicals represents 75% of the total mass of inorganic chemical production.19 Sulfuric acid, the world’s largest-volume industrial chemical, is principally used in the manufacture of phosphoric acid and phosphate fertilizers. Sulfuric acid is also used in making a number of other chemical products, such as hydro-ﬂuoric acid, synthetic detergents, dyes and pigments, drugs, explosives, plasticizers, adhesives, rubbers, edible oils, lubricants and the manufacture of food acids such as citric acid and lactic acid. It is used in petroleum reﬁning (to wash impurities out of reﬁnery products), water treatment, 10637

dx.doi.org/10.1021/es202193u |Environ. Sci. Technol. 2011, 45, 10634–10641

Environmental Science & Technology

ARTICLE

Figure 2. Mass (in MMT) and exergy (PJ) balance of the production of inorganic and organic chemicals in U.S. in 1991.

for cleaning iron and steel surfaces before plating, and in mining copper, uranium, and vanadium ores. Ammonia is a starting material used in a wide variety of nitrogen fertilizer materials and industrial products. About 90% of all ammonia production is consumed in fertilizers: urea, ammonium nitrate, ammonium sulfate, and ammonium phosphates.20 The rest is currently used in the manufacture of other inorganic products including nitric acid, nitrates and nitriles, hydrocyanic acid, sodium cyanide, aniline, plastics, ﬁbers (nylon), explosives, hydrazine, amines, amides, and other organic nitrogen compounds that serve as intermediates in dyes and pharmaceuticals manufacturing.21 Most of the world’s ammonia is produced from natural gas by steam reforming, except in China, where ammonia is produced from synthesis gas from coal.22 Chlorine and sodium hydroxide are produced as coproducts by electrolytic decomposition of sodium chloride solutions obtained from brines.19 They are two of the most important inorganic chemical commodities. Most of the chlorine and its organic compounds are converted to other products (as hydrochloric acid) and recycled within industry several times before being embodied in the ﬁnal product or discarded as waste. Approximately 60% of the value added in the chemical industry involves chlorinated chemicals at some stage, even though chlorine is not contained in the ﬁnal product. In 1993, 61% was used to make polyvinyl chloride (PVC), 18% went to industrial and commercial solvents (such as dry cleaning ﬂuids), 15% went to other organics, and 6% was used in inorganics such as bleaches.2326 The percentages have not changed very much. Sodium hydroxide was once used mainly in the manufacture of soap (from animal fat), and it is still used in the manufacture of household detergents and cleaners. But caustic soda is now used in many other industrial processes, notably bauxite reﬁning, in the paper pulping process, and in the petroleum and natural gas industry to neutralize acidic contaminants in gas and oil processing. It is also used for pH control, acid neutralization, and oﬀ-gas scrubbing. Based on the process descriptions of those basic inorganic chemicals, we assumed that sodium chloride, sulfur, methane, oxygen, nitrogen, and hydrogen are the raw material inputs. Figure 2 shows the material and energy balance for the production of the listed basic inorganic chemicals. The numbers in italics are the mass in million metric tons (MMT) of each input and output to the process. The mass balance shows that about 9%

of the total mass inputs are wasted in compounds of carbon, chlorine, sodium, and sulfur. Exergy inputs and outputs are represented in bold numbers and in PJ. The second law eﬃciency for the production of inorganic industrial chemicals works out to 29%, corresponding to a loss of more than 1,100 PJ. Based on USTIC statistics for 19911993, the major endproducts from organic chemicals are plastics (as polyethylene, polypropylene, polystyrene and polyvinyl chloride), nylon 6, ethylene glycol (antifreeze), and methyl tert-butyl-ether (a fuel additive that has been largely phased out since 1991). The production of these chemicals, in mass terms, represented about 80% of total U.S. production of organics in 1991.27 From standard unit process descriptions, we have estimated as basic inputs: hydrocarbons, and also inorganic raw materials as chlorine, sulfuric acid, ammonia, and caustic soda. Most organic chemicals are produced from feed-stocks from natural gas or petroleum reﬁneries, with a very small share from coal. There are three categories of feedstock: paraﬃn, oleﬁns, and cyclic/ aromatics. Paraﬃns are saturated straight or branched-chain hydrocarbons. Examples include methane, ethane, propane, isobutene, and n-butane. Oleﬁns are unsaturated aliphatic compounds with one or more double bonds. Examples include ethylene, propylene, butylenes, and butadiene. Cyclic aromatics are benzene, toluene, xylene, cyclopentene, cyclohexane, and naphthalene. A detailed breakdown of hydrocarbons included in the assessment is included in the SI 4. Figure 2 illustrates the material and energy balance for the production of the listed basic organic chemicals. Again, mass values represented in italics are in MMT and exergy values in bold numbers are given in PJ. The mass balance shows that about 60% of the total mass inputs are wasted as compounds of carbon, hydrogen, oxygen, nitrogen, chlorine, sodium, and sulfur. The second law eﬃciency for the production of organic industrial chemicals turns out to be 35%, corresponding to a loss of 2,100 PJ of exergy.

5. THE EFFICIENCY OF U.S. INDUSTRY AND ECONOMY The second law eﬃciency values obtained for inorganic and organic chemicals together with published second law eﬃciencies for other industries serve to estimate the overall eﬃciency of U.S. industry. SI 5 shows “the second law eﬃciency” for several industrial processes given by diﬀerent authors. The most signiﬁcant 10638

dx.doi.org/10.1021/es202193u |Environ. Sci. Technol. 2011, 45, 10634–10641

Environmental Science & Technology

ARTICLE

diﬀerences are found in exergy eﬃciency values given for the glass, paper and plastics industries. As regards the glass industry, the diﬀerence between the Hall (22.1%) and Ayres (2.9%) eﬃciency estimates are due to an assumption, in the Hall case, that high temperature waste heat is recovered—to preheat input materials, for instance—whereas in the Ayres calculation no recovery was assumed. For present purposes, we accept the Hall et al. calculation as more realistic. For the paper industry, the exergy eﬃciency results of Gyftopoulos et al. were calculated based on the exergy change of the feedstock and useful product, independently from the process.14 The eﬃciency result is misleading for two reasons: the change in the available useful work (exergy) for the sulfate alkaline process is small (only 0.12 GJ/ton) whereas the amount of fuel input is high (41.14 GJ/ton), and the eﬃciency calculation does not include the fuel or equivalent materials in the form of wood bark and spent pulp liquors generated by the process and used in the boiler. In this paper, we use the exergy eﬃciency value of 48% published by the U.S. Oﬃce of Technology Assessment (USOTA). For plastics, Hall et al. calculate the exergy eﬃciency based on the production of polymers from monomers whereas Gaines Table 2. Annual Production of Selected End Products (MMT) and Their Exergy Eﬃciency in 1993 production 1993 (MMT)

exergy eﬃciency

43.6

23%

ferrous

steel

non-ferrous

aluminum

3.7

24%

copper

1.8

10%

lead

0.3

11%a

zinc

0.2

5%a

industrial chemicals

inorganic

83.9

29%

organic

44.4

35%

other primary

glass

10.6

22%

materials

paper

82.2

48%

cement

67.0

14%b

337.8

30%

totals a

Calculated from Ayres and Ayres 2001.19 b Average exergy eﬃciency published by U.S. Congress Oﬃce of Technology Assessment (USOTA).

and Shen and Dewulf et al. included the production of the monomers.2830 The exergy eﬃciency of all plastics falls from 87 to 90% to 29 to 54% when including the production of monomers as ethylene, propylene, styrene, and vinyl chloride. In our assessment, we have included plastics as outputs from the organic chemical industry for assessing the U.S. industry eﬃciency, thus we do not include them as a separate industry. Table 2 shows the second law eﬃciency values for ferrous, nonferrous, and other primary materials are taken from published sources and based in previous comments. For the nonferrous metals lead and zinc, exergy eﬃciency values where calculated using data from published exergy ﬂow diagrams.19 The overall industry exergy eﬃciency (e2) is calculated by adjusting the contribution of each useful product, based on production values, to the production of the whole industry. This methodology gives more importance to useful products generated in larger amounts. The useful products considered for the assessment are ferrous metals, nonferrous metals (aluminum, copper, lead, and zinc), industrial inorganic and organic chemicals, and other primary materials as glass, paper and cement. Table 2 shows the annual production of those products and their second law eﬃciencies. The annual production data for each industry is directly taken from U.S. annual statistics and technical reports.3136 The second law eﬃciency of U.S. industry works out to about 30%, an eﬃciency value well below 7580% estimated by USDOE. The results provides critical information to policy makers as they suggest where there is still a potential for improving present industrial production processes. Further studies about the eﬃciency of industrial processes are in order to identify potential saving at process level. Potential savings may come from better technologies for heat recovery, and also the recycling of waste. A possible way to increase potential savings at higher level is to promote the exchange of useful ﬂows of materials and energy between industries concentrated within short distances. Potential saving will also depend on the type of industries located nearby, thus data and information about processes is crucial to ensure their success. The results of the exergy eﬃciency of U.S. industry can also be used to assess the eﬃciency of the U.S. economy, as a whole. Table 3 includes in the ﬁrst column the total energy input used by the residential and commercial, industrial and transportation sectors of the U.S. economy in 1991.37 The second and third columns include the eﬃciency values deduced from USEIA Sankey diagrams (available in SI 4) and the useful energy output for each sector.38 The fourth and ﬁfth columns have the exergy eﬃciency values for each sector and the useful energy output obtained from them.2,5

Table 3. Eﬃciency of U.S. Economy in 1991 USEIA

New estimation based on exergy

energy input by sector (quads) eﬃciency (%) useful energy output (quads) exergy eﬃciency (%) useful energy output (quads)

a b

residential and commercial

16.20

75

12.20

industrial

16.30

80

13.00

transportation

22.20

25

5.60

7.5a 30 1.0b

1.22 4.89 0.22

useful energy (exergy)

30.80

total energy consumption

82

6.33 82

eﬃciency

37.6%

7.7%

Average exergy eﬃciency value based on Carnahan et al. 1975.2 Similar exergy eﬃciency values are given by Yildiz and G€ung€or 20097 and Kondo 2009.8 Exergy eﬃciency value given by Dewulf and Van Langenhove 2003.5 10639

dx.doi.org/10.1021/es202193u |Environ. Sci. Technol. 2011, 45, 10634–10641

Environmental Science & Technology We note that Reistad estimated that the residential and commercial sector eﬃciency was 13.7%.9 His estimate was based on two subsidiary estimates with which that we disagree, namely that the direct fuel use (for space and water heating) was 12.5% eﬃcient and that the electrical use was 21.9% eﬃcient, allowing for the eﬃciency of generation. However, the ﬁrst ﬁgure is too high because the temperature of the water going into the roomheating radiator or the bathtub—as opposed to the water in the tank—is less than 100 °C resulting in a Carnot eﬃciency of no more than 5%.2 The second ﬁgure (electrical use) is also too high because a signiﬁcant amount of the electrical energy was (and is) used for lighting, which was less than 5% eﬃcient in 1975, and is less than 10% eﬃcient today. For this reason, we think it is more realistic to use exergy eﬃciency values published by Kondo, which yields an overall eﬃciency for the sector of less than 10%.8 The eﬃciency of U.S. economy can be calculated by dividing the useful energy output by the total energy consumption. The eﬃciency given by USEIA equals 37.6%, compared to 7.7% given using the second law eﬃciency for the year 1991. The results show that more than 90% of high quality energy extracted from the earth is wasted. The transportation sector is by far the poorer in performance compared to industrial and even commercial and residential. These second law eﬃciencies indicate where eﬀorts to improve the use of quality energy should be made and show that it is still possible to use resources much more eﬃciently than they are used, and that there is still a lot of potential for doing so. A recent study estimated that the current eﬃciency of U.S. economy is about 13% in 2009, but using an estimate of transportation eﬃciency based on the movement of vehicles, not payload.39,40 USEIA reports suggest that Russia, China, and India remain less energy eﬃcient than the U.S. (at least in the industrial sectors) whereas Japan, the UK, and Austria reach 20% eﬃciency.41

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional material as noted in the text. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +33 160 724 189; fax: +33 160 745 566; e-mail: robert. [email protected].

’ ACKNOWLEDGMENT We thank the INSEAD and Marie Curie fellowship (FP7PLEOPLE-2010-IEF 272206). ’ REFERENCES (1) U.S. Energy Information Agency (US EIA). Annual energy review 2008, DOE/EIA-0384 (2008); U.S Department of Energy: Washington, DC, June 2009. (2) Carnahan, W.; Ford, K. W.; Prosperetti, A.; Rochlin, G. I.; Rosenfeld, A. H.; Ross, M. H.; Rothberg, J. E.; Seidel, G. M.; Socolow, R. H. Eﬃcient Use of Energy: A Physics Perspective, 399; American Physical Society: New York, January, 1975. (3) Dewulf, J.; Van Langenhove, H.; Muys, B.; Bruers, S.; Bakshi, B.; Grubb, G.; D.M., P.; E., S. Exergy: Its potential and limitations in Environmental Science and Technology. Environ. Sci. Technol. 2008, 42 (7), 2221–2232.

ARTICLE

(4) Lawrence Livermore National Laboratory (LLNL). Estimated US Energy use in 2008, LLNL-MI-410527; U.S. Department of Energy, 2009. (5) Dewulf, J.; Van Langenhove, H. Exergetic material input per unit of service (EMIPS) for the assessment of resource productivity of transport commodities. Resour., Conserv. Recycl. 2003, 38, 161–174. (6) Ertesvag, I. S.; Mielnik, M. Exergy analysis of the Norwegian society. Energy 2000, 25 (10), 957–973. (7) Yildiz, A.; G€ung€ or, A. Energy and exergy analyses of space heating in buildings. Appl. Energy 2009, 86 (10), 1939–1948. (8) Kondo, K. Energy and exergy utilization eﬃciencies in the Japanese residential/commercial sectors. Energy Policy 2009, 37 (9), 3475–3483. (9) Reistad, G. M. Available energy conversion and utilization in the United States (USA). Trans. ASME, J. Eng. Power 1975, 97 (3), 429–434. (10) Wall, G. Exergy conversion in the Swedish society. Resour. Energy 1987, 9, 55–73. (11) Wall, G. Exergy conversion in the Japanese society. Energy 1990, 15. (12) Wall, G.; Sciubba, E.; Naso, V. Exergy use in the Italian society. Energy 1994, 19 (12), 1267–1274. (13) Ertesvag, I. Society exergy analysis: A comparison of diﬀerent societies. Energy 2001, 26, 253–270. (14) Gyftopoulos, E. P.; Lazaridis, L. J.; Widmer, T. F. Potential Fuel Eﬀectiveness in Industry; Ballinger Publishing Company: Cambridge MA, 1974. (15) Ayres, R. U.; Martinas, K.; Ayres, L. W. Exergy, waste accounting and life cycle analysis. Energy 1998, 23 (5), 355–363. (16) Talens Peiro, L.; Villalba Mendez, G.; Sciubba, E.; Gabarrell i Durany, X. Extended exergy accounting applied to biodiesel production. Energy 2010, 35 (7), 2861–2869. (17) Szargut, J.; Morris, D. R.; Steward, F. R. Exergy Analysis of Thermal, Chemical, And Metallurgical Processes; Hemisphere Publishing Corporation: New York, 1988. (18) U.S. Energy Information Agency (US EIA). Manufacturers Energy Consumption Survey, 1991. (19) Ayres, R. U.; Ayres, L. W., Accounting for Resources 2: The Life Cycle of Materials; Edward Elgar: Cheltenham, UK and Lyme MA, 1999. (20) Suresh, B.; Fujita, K. Ammonia; SRI consulting: 2007. (21) Febre-Domene, L. A.; Ayres, R. U. Nitrogen’s role in industrial systems. J. Ind. Ecol. 2001, 5 (1), 77–103. (22) Glauser, J.; Kumamoto, T. Chemical Economics Handbook: Ammonia; Stanford Research Institute Consulting: Englewood, CO, 2010. (23) Ayres, R. U. The life cycle of chlorine: Part III: Accounting for ﬁnal use. J. Ind. Ecol. 1998, II (2), 65–89. (24) Ayres, R. U. The life cycle of chlorine: Part I; Chlorine production and the chlorine-mercury connection. J. Ind. Ecol. 1997, I (1), 81–94. (25) Ayres, R. U.; Ayres, L. W. The life cycle of chlorine: Part II; Conversion processes and use in the European chemical industry. J. Ind. Ecol. 1997, I (2), 93–115. (26) Ayres, R. U.; Ayres, L. W. The life cycle of chlorine: Part IV: Accounting for persistent cyclic organo-chlorines. J. Ind. Ecol. 2000, IV (1), 123–132. (27) Ayres, R. U.; Ayres, L. W., Accounting for Resources 1: EconomyWide Applications of Mass-Balance Principles to Materials and Waste; Edward Elgar: Cheltenham, UK and Lyme MA, 1998. (28) Hall, E. H.; Hanna, W. H.; Reed, L. D.; Varga, J. J.; Williams, D. N.; Wilkes, K. E.; Johnson, B. E.; Mueller, W. J.; Bradbury, E. J.; Frederick, W. J. Evaluation of the Theoretical Potential for Energy Conservation in Seven Basic Industries, PB-244,772; Battelle Columbus Laboratories: Columbus OH, July 11, 1975. (29) Gaines, L. L.; Shen, S. Y. Energy and Materials Flows in the Production of Oleﬁns and Their Derivatives, ANL/CNSV-9; Argonne National Laboratory: Argonne IL, August, 1980. (30) Dewulf, J.; Van Langenhove, H. Thermodynamic optimization of the life cycle of plastics by exergy analysis. Int. J. Energy Res. 2004, 28 (11), 969–976. 10640

dx.doi.org/10.1021/es202193u |Environ. Sci. Technol. 2011, 45, 10634–10641

Environmental Science & Technology

ARTICLE

(31) PEDCo-Environmental. Industrial Process Proﬁles for Environmental Use: Chapter 29; Primary Copper Industry, PB81-164915; EPA600/2-80-170; PEDCo-Environmental: Cincinnati OH, July, 1980. (32) PEDCo-Environmental. Industrial Process Proﬁles for Environmental Use: Chapter 27; Primary Lead Industry; PB81-110926; EPA-600/ 2-80-168; PEDCo-Environmental: Cincinnati OH, July, 1980. (33) PEDCo-Environmental. Industrial Process Proﬁles for Environmental Use: Chapter 28; Primary Zinc Industry, PB80-225717; EPA-600/ 2-80-169; PEDCo-Environmental: Cincinnati OH, July, 1980. (34) United States International Trade Commission. Synthetic Organic Chemicals 1992; United States Government Printing Oﬃce: Washington DC, 1992. (35) United States International Trade Commission, Synthetic organic chemicals 1991. United States Government Printing Oﬃce: Washington DC, 1991. (36) Gaines, L. L. Energy and Material Flows in the Copper Industry; Argonne National Laboratory: Argonne IL, 1980. (37) Borg, I. Y.; Briggs, C. K. US Energy ﬂow-1991, UCID-19227-91; Lawrence Livermore National Laboratory, Department of Energy: Livermore, CA, June 1992, 1992; p 26. (38) Borg, I. Y.; Briggs, C. K. US Energy ﬂow-1992; Lawrence Livermore National Laboratory, Department of Energy: Livermore, CA, October 1993, 1993; p 24. (39) Ayres, R. U.; Warr, B. S., Energy eﬃciency and economic growth: The “rebound eﬀect” as a driver. In Energy Eﬃciency and Sustainable Consumption, Chapter 6; Herring, H.; Sorrell, S., Eds. Palgrave Macmillan: London, 2009; pp 121137. (40) Ayres, R. U.; Ayres, E. H., Crossing the Energy Divide: Moving from Fossil Fuel Dependence to a Clean-Energy Future; Wharton School publishing: Upper Saddle River NJ, 2010. (41) Warr, B. S.; Eisenmenger, N.; Krausmann, F.; Schandl, H.; Ayres, R. U. Energy use and economic development; a comparative analysis of useful work supply in Austria, Japan, the United Kingdom and the US during 100 years of economic growth. Ecol. Econ. 2010in press.

10641

dx.doi.org/10.1021/es202193u |Environ. Sci. Technol. 2011, 45, 10634–10641

ARTICLE pubs.acs.org/est

Bidirectional Permeation of Electrolytes in Osmotically Driven Membrane Processes Nathan T. Hancock,† William A. Phillip,*,‡ Menachem Elimelech,§ and Tzahi Y. Cath*,† †

Division of Environmental Science and Engineering, Colorado School of Mines, 1500 Illinois Street, Golden, Colorado 80401, United States ‡ Department of Chemical and Biomolecular Engineering, University of Notre Dame, 182 Fitzpatrick Hall, Notre Dame, Indiana 46556-5637, United States § Department of Chemical and Environmental Engineering, Yale University, P.O. Box 208286, New Haven, Connecticut 06520-8286, United States

bS Supporting Information ABSTRACT: Osmotically driven membrane processes (ODMP) are emerging water treatment and energy conversion technologies. In this work, we investigated the simultaneous forward and reverse (i.e., bidirectional) solute ﬂuxes that occur in ODMP. Numerous experiments were conducted using ternary systems (i.e., systems containing three distinct ions) and quaternary systems (i.e., systems containing four distinct ions) in conjunction with a membrane in a forward osmosis orientation. Ten diﬀerent combinations of strong electrolyte salts constitute the ternary systems; common anion systems studied included KClNaCl, KBr-NaBr, KNO3-NaNO3, KCl-CaCl2, and KCl-SrCl2; and common cation systems explored were KCl-KH2PO4, NaCl-NaClO4, NaCl-Na2SO4, NaCl-NaNO3, and CaCl2-Ca(NO3)2. For each combination, two experiments were conducted with each salt being used once in the draw solution and once in the feed solution. Quaternary systems studied were NaCl-KNO3, NaCl-MgSO4, MgSO4-KNO3, and NaCl-K2SO4. Experimental ﬂuxes of the individual ions were quantiﬁed and compared to a set of equations developed to predict bidirectional electrolyte permeation for ODMP in a forward osmosis orientation. Results demonstrate that ion ﬂuxes from the draw solution to the feed solution are well predicted; however, ion ﬂuxes from the feed solution to the draw solution show slight deviations from the model that can be rationalized in terms of the electrostatic interactions between charged ions. The model poorly predicts the ﬂux of nitrate containing solutions; however, several unique mass transfer mechanisms are observed with implications for ODMP process design.

’ INTRODUCTION Osmotically driven membrane processes (ODMP), such as forward osmosis (FO), pressure retarded osmosis (PRO), and osmotic dilution, are an emerging class of technologies that may be used for water treatment1,2 and for sustainable energy generation.3,4 FO technology has demonstrated unique abilities to treat highly impaired water sources59 as well as to desalinate seawater and brackish water.10,11 The versatility of ODMP is partly due to their lower irreversible fouling propensity1214 and to the spontaneity of mass transport that eliminates the need for operation with high hydraulic pressure.1 Recent research to improve ODMP led to the development of new membranes with unique support structures that reduce the mass transfer limitations imposed by internal concentration polarization1518 and the exploration of novel draw solutions to enable treatment of highly saline feed streams.10 However, fundamental problems related to the bidirectional transport of solutes through ODMP membranes require additional investigation to further improve the processes. Reverse permeation of solutes from the draw solution through the membrane into the r 2011 American Chemical Society

feed stream, in the opposite direction of the water ﬂux, decreases the driving force for water permeation and may necessitate replenishment of the draw solution. Likewise, permeation of feed solutes through the membrane into the draw solution may negatively impact the performance of the downstream process. This may be especially acute when sparingly soluble salts accumulate and precipitate in systems that employ a closed-loop draw solution reconcentration process. Previous studies5,1924 examined the transport of solutes in FO. These studies identiﬁed the reverse permeation of draw solution solutes as a potential obstacle for future process design and implementation. Two recent studies19,25 found that the speciﬁc reverse ﬂux (the inverse of reverse ﬂux selectivity20) through a cellulose triacetate (CTA) membrane into deionized water feed can vary by an order of magnitude between diﬀerent Received: July 27, 2011 Accepted: October 28, 2011 Revised: October 20, 2011 Published: October 28, 2011 10642

dx.doi.org/10.1021/es202608y | Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

ARTICLE

between charged ions from the feed and draw solutions. A limited number of quaternary ion systems were explored to investigate what aspects of the model could be extended to higher order systems. Implications of the results for the design and operation of ODMPs are discussed.

’ THEORETICAL BACKGROUND Forward and Reverse Fluxes of Noninteracting Electrolytes. In this study we are exploring how electrostatic interac-

Figure 1. Schematics of concentration proﬁles for (a) the draw solution electrolyte and (b) the feed solution electrolyte. Local coordinates x and z are established with respect to the forward water ﬂux (Jw). Both electrolytes are transported down their concentration gradient resulting in a net reverse ﬂux of draw solution electrolyte, J1, and a net forward ﬂux of feed solution electrolyte, J2. Electrolytes must be transported across the external concentration polarization layer, active layer, and porous support layer with thickness δ, ta, and ts, respectively. Subscripts are used to denote either the draw solution electrolyte (1) or feed solution electrolyte (2). Superscripts represent the concentration of these ions at speciﬁc locations along their concentration proﬁle, where F and D correspond to the feed and draw solution sides of the active layer, and b, i, and a correspond to the bulk solution, interface, or active layer, respectively.

draw solutions. A recent study20 derived a model to predict the rate of reverse permeation of strong electrolytes (e.g., NaCl), and related the reverse ﬂux selectivity of the membrane to the membrane electrolyte and solvent (water) permeabilities. In another recent study19 we explored the bidirectional transport of solutes through membranes in FO. Solutes were observed to diﬀuse at diﬀerent rates based primarily on the size of the hydrated molecule. Ions that form larger hydration shells (e.g., barium, magnesium, calcium) exhibited lower ﬂuxes compared to ions with smaller hydration shells (e.g., potassium, sodium, ammonium).26 During all experiments, ions were observed to diﬀuse at rates that maintained electroneutrality, and Donnan eﬀects27 were also observed. For example, in one set of experiments, a feed stream consisting of BaNO3 was used. The highly mobile nitrate ion was able to permeate through the membrane at a substantially faster rate than barium. In this case, chloride from the NaCl draw solution was observed to permeate at a greater rate relative to sodium to maintain the electroneutrality of the solutions. The objective of this study was to develop a comprehensive understanding of bidirecitonal solute permeation through the membrane during FO. Speciﬁcally, the potential impact of electrostatic interactions on the bidirectional permeation of ions through the membrane was explored using a combination of bench-scale experiments and phenomenological modeling. A series of ternary ion experiments with an inorganic electrolyte dissolved in the feed solution and another distinct electrolyte, but one which shares a common anion or cation, dissolved in the draw solution were conducted, and the ﬂux of each ion was measured. The measured ion ﬂuxes were compared to a model developed for electrolyte permeation in the absence of interactions between the draw solution and feed solution ions. Deviations between the model and experimental results were reconciled in terms of the electrostatic interactions that occur

tions between ions can influence their forward and reverse permeation in FO. To begin, we derive expressions for the electrolyte fluxes, ignoring the electrostatic interactions between the constituent ions. Please note that we use the term electrolyte to refer to the parent salt that when dissolved in solution forms ions. A schematic of an asymmetric membrane operating in FO mode is shown in Figure 1. The permeation of an electrolyte from the draw solution into the feed solution (Figure 1a) and an electrolyte from the feed solution into the draw solution (Figure 1b) requires their transport across three distinct regions: the external boundary layer, the dense active layer, and the porous support layer. Transport of electrolytes in the support layer and boundary layer occurs by both diﬀusion and convection, but only diﬀusion controls electrolyte transport through the active layer — consistent with the solution-diﬀusion transport mechanism.28,29 By writing mass balances on each of the layers and equating the electrolyte ﬂuxes at the interfaces between the layers, expressions for the electrolyte ﬂuxes in the absence of electrostatic interactions can be derived (eqs 1 and 2). The details of these derivations were previously reported30 and are provided in the SI. Here we present the ﬁnal results for the draw solution electrolyte and feed solution electrolyte ﬂuxes, J1 and J2, respectively J1 ¼

Jw B1 ðc1F, b expðPeF1 þ PeD1 Þ c1D, b Þ ðB1 expðPeF1 Þ þ Jw Þ expðPeD1 Þ B1

ð1Þ

J2 ¼

Jw B2 ðc2F, b expðPeF2 þ PeD2 Þ c2D, b Þ ðB2 expðPeF2 Þ þ Jw Þ expðPeD2 Þ B2

ð2Þ

where Jw is the superﬁcial ﬂuid velocity (which is equivalent to the water ﬂux), B1 is the electrolyte permeability coeﬃcient, cF,b 1 is the concentration of draw solution electrolyte in the bulk feed solution, and cD,b 1 is the concentration of draw solution electrolyte in the bulk draw solution. PeF1 and PeD 1 are the Peclet numbers of the draw solution electrolyte in the external boundary layer and the support layer, respectively. These Peclet numbers quantify the relative importance of convective transport to diﬀusive transport in each layer and are deﬁned as PeF1 Jw/k for the external boundary layer, where k is the feed side mass transfer coeﬃcient, and PeD 1 (Jwtsτ)/(DDSε) for the support layer, where DDS is the bulk binary diﬀusion coeﬃcient of the draw solution electrolyte and water, ts is the support layer thickness, τ its tortuosity, and ε its porosity. All the variables have the same meaning in eq 2 but are now deﬁned for the feed solution electrolyte instead of the draw solution electrolyte. Equations 1 and 2 can be used to predict the reverse and forward electrolyte ﬂuxes, respectively, in the absence of interactions between feed and draw solution electrolytes and will be used as a reference for comparison. Note that the electrolyte permeability and diﬀusion coeﬃcients in these equations refer to 10643

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

ARTICLE

those measured for a single electrolyte where the fact that the electrolyte forms multiple ions when dissolved in solution has been ignored.20 By doing so, the electrostatic interactions between the ions are ignored and electroneutrality is implicitly assumed. Below we consider these interactions more explicitly. Effects of Electrostatic Interactions on Ion Permeation. In FO, ions from the feed and draw electrolytes will be present at appreciable concentrations within the dense active layer of the membrane. Therefore, it is important to understand how these ions are transported across the active layer and how they may interact with each other. Explicitly, considering that electrolytes form ions within the active layer of the membrane is particularly interesting. Our prior work with an NaCl draw solution and a deionized water feed solution demonstrated that the NaCl flux is linearly proportional to the NaCl concentration difference across the active layer.20 This result implies that the sodium and chloride permeate the active layer as charged ions and not an ion pair.31 Despite the evidence that ions permeate the active layer as two separate entities, their transport could be described by a single transport coefficient.20,28 A Single Transport Coefficient Describes the Permeation of a Binary System. Dissolved solutes are transported through the dense active layer of the FO membrane by the solutiondiffusion mechanism.28 In this mechanism, the diffusing solute, which in the case of a dissolved electrolyte is the constituent cations and anions, must first partition into the active layer phase before diffusing across it. The partitioning of the cation and anion can be quantified using a Henry’s-law-like constant, by assuming equilibrium at the active layer-aqueous solution interface μai ¼ μsi

ð3Þ

where μi is the chemical potential of species i, and the superscripts represent the phase: (a) active layer and (s) solution. The index of species i represents the specific ion in solution: 1 is an ion in the draw solution, 2 is an ion in the feed solution, and 3 is the counterion, which for ternary systems will be the common ion found in both the feed and draw solutions. For the binary case of a single 1:1 electrolyte, eq 3 can be written for each ion as follows a a s0 s s μa0 1 þ RT ln c1 þ z1 Fϕ ¼ μ1 þ RT ln c1 þ z1 Fϕ

ð4aÞ a a s0 s s μa0 3 þ RT ln c3 þ z3 Fϕ ¼ μ3 þ RT ln c3 þ z3 Fϕ

ð4bÞ where μ0i is a reference chemical potential, ci is the concentration of species i, zi is the valence, F is the Faraday constant, R is the ideal gas constant, T is the absolute temperature, and ϕ is the electrostatic potential. Adding eqs 4a and 4b, rearranging, and using the fact that for a 1:1 electrolyte c1 = c3 = c yields pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ca ¼ H1 H3 ¼ H1=3 cs

The extended NernstPlanck equation can be used to demonstrate that the electrostatic coupling between the oppositely charged ions allows their motion to be described by a single diﬀusion coeﬃcient. The ﬂux of a charged species i deﬁned by the extended NernstPlanck equation is29 F ð7Þ Ji ¼ Di ∇ci þ zi ci ∇ϕ RT where Ji is the ﬂux of species i, and Di is the diﬀusion coeﬃcient in the active layer. Additionally, the system must maintain electroneutrality, thus ∑icizi = 0, and because there is no applied potential, the current, I, is equal to 0 (i.e., I = ∑iziJi = 0). This system of equations can be used to show that the diﬀusion coeﬃcient of the 1:1 electrolyte in the active layer, D1/3, is equal to the harmonic average of the ionic diﬀusion coeﬃcients, D1 and D3 2D1 D3 Js ¼ J1 ¼ J3 ¼ ð8Þ ∇c ¼ D1=3 ∇c D1 þ D3 Equation 8 can be integrated over the active layer thickness, ta. Then, using eq 5, the concentration diﬀerence can be written in terms of the electrolyte concentration in solution to yield the following expression for the ﬂux across the active layer Jsa ¼

Therefore, the partition coefficient for the electrolyte (i.e., the coupled anion and cation), H1/3, is the geometric average of the partition coefficient for each ion.29

ð9Þ

The group of terms D1/3H1/3/ta is more commonly measured and reported as the electrolyte permeability coeﬃcient, B1/3 (i.e., B1 used in eq 1). Thus, a single coeﬃcient can be used to quantify the ﬂux of a binary (i.e., anion and cation) system.29 Electrostatic Interactions Have a Small Effect on the Permeation of Ternary Systems. The electrostatic coupling between anion and cation of the binary system, which allows that system to be described by a single transport coefficient, may affect ion permeation in FO processes when ions from the feed solution interact with ions from the draw solution. We now consider a ternary system consisting of three ions with a common anion or cation in the feed and draw solution (e.g., an NaCl draw solution and a KCl feed or an NaCl draw solution and an NaClO4 feed). A ternary system is used because analytical solutions do not exist for higher order systems. Multicomponent partitioning effects in ternary systems have been studied for cellulose acetate membranes designed for RO operations.32 In general, the observed effects were small. One important conclusion from these studies was that at operational pHs cellulose acetate membranes contain negligible fixed charged groups. Therefore, charges contained within the membrane will not have an observable effect on ion partitioning.33 The NernstPlanck equation, the electroneutrality constraint, and I = 0 are once again the starting points for deriving the transport coeﬃcients. This system leads to29 Ji ¼ Dij ∇cj

ð5Þ

where c is the electrolyte concentration, and Hi, the partition coefficient for the individual species i, is equal to ! Δμ0i ð6Þ Hi ¼ exp RT

D1=3 H1=3 Δc ¼ B1=3 Δc ta

ð10aÞ

2

D1 z21 c1 ðD3 D1 Þ 6 D1 þ 3 6 Dk z2k ck 6 6 k ¼ 1 Dij ¼ 6 6 D2 z1 z2 c2 ðD3 D1 Þ 6 6 3 4 Dk z2k ck

∑

∑

k¼1

D1 z1 z2 c1 ðD3 D2 Þ

3

7 7 7 7 k¼1 7 D2 z22 c2 ðD3 D2 Þ 7 7 D2 þ 7 3 5 Dk z2k ck 3

∑

Dk z2k ck

∑

k¼1

ð10bÞ 10644

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

ARTICLE

where the diﬀusion coeﬃcients Dij quantify the proportionality between the ﬂux of species i and the concentration gradient of species j. Equation 10b is written as a 2 2 matrix because the ﬂux of the common ion (species 3) can be determined from the ﬂux of the ion in the draw solution, J1, the ﬂux of the ion in the feed solution, J2, and the fact that I = 0. By itself, eq 10b does not provide any clear physical insight into the bidirectional permeation of ions in FO. Therefore, it is helpful to consider what occurs when c1 . c2, a pertinent condition in FO where the draw solution concentration is much greater than the feed concentration. Under this condition, eq 10b reduces to 2 3 ðjz1 j þ jz3 jÞD1 D3 D1 z2 ðD3 D2 Þ ¼ D 1=3 6 jz1 jD1 þ jz3 jD3 7 6 jz1 jD1 þ jz3 jD3 7 7 Dij ¼ 6 4 5 0 D2 ð11Þ This provides much more physical insight into the transport of ions in FO. The cross-term D21, which quantiﬁes how much the ﬂux of species 2 (the feed solution ion) is inﬂuenced by the gradient of species 1 (the draw solution ion), goes to 0. Therefore, the concentration gradient of the draw ion should not aﬀect the ﬂux of the feed ion, although the draw solution concentration must be high enough that the limit c1 . c2 is satisﬁed. The term D22 reduces to D2 (i.e., the ionic diﬀusion coeﬃcient of species 2), which implies that because there is excess of the common counterion (species 3) present, species 2 is able to permeate through the active layer at a rate independent of species 1 and 3. The ability of species 2 to permeate independently may produce some noticeable variations in its ﬂux that can be observed during experiments; however, the diﬀerence between the ionic diﬀusion coeﬃcients, D2 and D3, is often negligible and any variations are likely to be small. The diﬀusion coeﬃcients for species 1 also yield interesting values under this condition (c1 . c2). The cross-term D12 is nonzero because species 1 adjusts its permeation rate to allow species 2 to permeate at its independent rate, while still maintaining electroneutrality. The main term coeﬃcient, D11, reduces to D1/3, which is equal to the weighted average of the ionic diﬀusion coeﬃcients of ions 1 and 3 (i.e., eq 8). Examining the relative magnitudes of D12rc2 and D11rc1, it is easily shown that D11rc1 will dominate the ﬂux of species 1. Therefore, the electrostatic coupling between ions from the feed and draw solutions will have little eﬀect on the ﬂux of ions from the draw solution.

’ MATERIAL AND METHODS Model Inorganic Salts. Thirteen ACS grade salts were used in this study: NaCl (Sigma-Aldrich/Fluka, Buchs, Switzerland); SrCl2 and NaClO4 (Sigma-Aldrich, St. Louis, MO); KH2PO4, KNO3, Ca(NO3)2, and NaNO3 (Fisher Scientific, El Monte, CA); Na2SO4, MgSO4, NaBr, KCl, and CaCl2 (Mallinckrodt Chemicals, Phillipsburg, NJ); and KBr (Spectrum Chemical Products, Gardena, CA). Ternary and quaternary ion experiments were conducted to explore the bidirectional fluxes of dissolved salts. Both ternary and quaternary experiments were conducted with a 1 M draw solution concentration and 0.05 M feed solution concentration, with the exception of a few experiments that are explicitly defined in the Results and Discussion section in which the concentrations were different.

Bench-Scale System Design. A bench-scale testing apparatus similar to that described in a previous publication19 was employed for this study. The testing cell was constructed with symmetric flow chambers, each containing two 575 mm 48 mm 1.6 mm (L W H) channels separated by a gasket. Draw and feed solutions were circulated cocurrently through the chambers at a flow rate of 2 L/min. A programmable logic control system was developed to maintain constant experimental conditions (i.e., feed volume of 3 L, draw solution concentration, and system temperature of 20 °C) and to collect experimental data. Details about the design and operation of the system are provided elsewhere.30 Sampling and Analytical Methods. Samples of the feed and draw solutions were collected at the beginning of each experiment and after 1 and 2 L of water permeated through the FO membrane. Diluted samples were analyzed with ion chromatography (IC) (ICS-90, Dionex, Sunnyvale, CA) and inductively coupled plasma atomic emission spectroscopy (ICP-AES) (Optima 5300, Perkin-Elmer, Fremont, CA) to determine the concentration of anions and cations, respectively. Anion concentrations analyzed with IC were measured in duplicate with a typical relative standard deviation of 2.2%. Cation concentrations analyzed with ICP-AES were measured in triplicate with a typical relative standard deviation of 1.8%. The mass of draw solution ions in the feed stream was recorded as a function of time and normalized by the membrane area to determine the reverse ﬂux of draw solution ions. An identical process was followed to measure the forward ﬂux of the feed solution ion. Forward Osmosis Membrane Characterization. The membrane used for this study (Hydration Technology Innovations, Albany, OR) is similar to the CTA membranes used in previous studies.19,20 The membrane is believed to be composed of a cellulose triacetate layer with an embedded woven support mesh.1,34 Membrane integrity tests, using a 1 M NaCl draw solution and a Milli-Q deionized water feed, were routinely performed to ensure that membrane performance (i.e., water flux and reverse NaCl salt flux) were consistent between experiments. The membrane was replaced if either water or reverse NaCl salt flux deviated more than 5% from their established baseline values. The average water flux and reverse NaCl salt flux over 26 integrity tests were 10.2 ( 0.35 L/m2 3 hr and 127 ( 9 mmol/ m2 3 hr, respectively. Pure water permeability (A) and the membrane structural parameter (S tsτ/ε) were measured in FO mode (i.e., draw solution facing the support layer and no applied hydraulic pressure) using the method reported in a previous publication.8 Water ﬂux was measured at three diﬀerent draw solution concentrations (0.5, 1.0, and 1.5 M NaCl) with deionized water feed. Thermophysical modeling software (OLI Analyzer v3.0, OLI Systems Incorporated, Morris Plains, NJ) was employed to calculate the osmotic pressure of the draw solutions to account for the nonideality of concentrated solutions. Two samples of the same commercial membrane were used during these experiments. The diﬀerent samples had slightly diﬀerent transport characteristics. The pure water permeability and structural parameter of Membrane I were determined to be 0.44 ( 0.04 L/m2 3 hr 3 bar (1.36 1012 ( 9.80 1014 m/ Pa 3 s) and 439 ( 26 μm, respectively. Membrane II was more permeable but had a higher structural parameter (A = 0.52 ( 0.02 L/m2 3 hr 3 bar (1.45 1012 ( 4.90 1014 m/Pa 3 s) and 10645

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

ARTICLE

Table 1. Water and Cation Fluxes Measured during PRO Mode Experiments and Characteristic Transport Coeﬃcientsa DDS45 (10‑9 m2/s)

Jw (10‑6 m/s)

cation ﬂux (mmol/m2 3 hr)

B (10‑8 m/s)

NaClO4

1.53

3.14

NaNO3

1.56

3.52

1.17

1297.8 ( 21.7

49.92

1.19

420.5 ( 10.1

Ca(NO3)2

1.29

16.28

5.18

1.05

74.5 ( 1.5

5.04

KBr NaBr

2.02

4.43

1.41

400.0 ( 4.3

16.19

1.62

4.50

1.22

288.0 ( 2.8

12.56

KCl

1.99

4.35

1.40

190.4 ( 0.2

7.64

NaCl

1.61

3.79

1.25

156.7 ( 7.7

6.02

CaCl2 SrCl2

1.33 1.34

5.71 5.21

1.07 1.07

33.7 ( 0.7 27.4 ( 1.8

2.38 1.99

KH2PO4

1.18

2.18

0.99

18.3 ( 4.4

1.26

Na2SO4

0.41

2.68

1.01

30.5 ( 2.9

1.72

KCl

1.99

4.89

1.40

521.6 ( 2.0

19.44

NaCl

1.61

4.75

1.25

280.4 ( 0.6

12.22

CaCl2

1.33

5.99

1.07

85.9 ( 1.7

4.17

salt

k (10‑5 m/s) Membrane I

Membrane II

a

The electrolyte diﬀusion coeﬃcients in water were calculated using eq 8, the mass transfer coeﬃcients were obtained from the correlation developed in ref 35, and the electrolyte permeation coeﬃcients were determined using the method in ref 20. Salt pairs are listed in order from most chaotropic to most kosmotropic based on their anion.37 The two membrane samples used in this work are listed as Membrane I (A = 1.36 1012 ( 9.80 1014 m/Pa 3 s; S = 439 ( 26 μm) and Membrane II (A = 1.45 1012 ( 4.90 1014 m/Pa 3 s; S = 559 ( 42 μm).

S = 559 ( 42 μm). These variations are similar to those observed in the published literature for this membrane.20,34 Additional experiments were conducted in PRO mode (i.e., draw solution facing the active layer and no applied hydraulic pressure) to determine the permeability coeﬃcient (B) for various electrolytes using the method reported in a previous publication.20 In each experiment, the draw solution concentration of electrolyte was 1 M and the feed was deionized water. Samples were drawn at predetermined times throughout each experiment and analyzed with IC and ICP-AES. The ﬂux of ions into the feed was calculated using the method reported above.

’ RESULTS AND DISCUSSION Membrane Performance Parameters. Permeability coefficients, B, for the various electrolytes used in this study were determined from PRO mode experiments and are reported in Table 1. The electrolyte diffusion coefficient in water,29 observed water flux, mass transfer coefficient,35 and molar flux of the cation used in this calculation are also summarized in Table 1 (molar flux of the anions are provided in the SI). The rate that salts with similar cations permeate through the membrane correlates with their anion. Permeabilities were highest for perchlorate containing salts, and incrementally decreased in the order of nitrate, bromine, chloride, dihydrogen phosphate, and sulfate containing salts. Empirically, this follows a similar trend as the Hofmeister series.3638 A trend in the permeability coeﬃcients for the electrolytes with similar anions can also be seen in Table 1. Salts containing divalent cations diﬀuse more slowly than salts with similar anions that contain monovalent cations. Among monovalent cations, diﬀusion rates are consistently greater for salts containing potassium instead of sodium. This behavior is in accordance with previous observations19,25,29,39 and indicates that a smaller hydration radius correlates to faster permeation through the membrane.

Model Validation of Experimental Water Flux. Experiments were conducted in FO mode to measure the water flux and bidirectional flux of ions. Ten combinations of salts were used in 18 different experiments. Measured water fluxes from these experiments were compared to predicted values from existing models34,40,41 (Figure 2). In many cases, the water flux was under predicted but was within 10% of perfect agreement. Predicted water flux values were used in eqs 1 and 2 to calculate the forward and reverse flux of electrolytes a priori. Ion Permeation in Common Anion Systems. Experiments were performed using ternary systems that share a common anion in the feed and draw solutions. The experimentally determined ion fluxes are compared to the predicted fluxes calculated from eqs 1 and 2. One of the goals of this work is to explore the impact of electrostatic interactions on the bidirectional flux of ions in ODMP; thus, the predicted cation fluxes were calculated using the electrolyte permeability coefficients in Table 1. The model used has already been validated for single salt systems,20 meaning any significant deviations between the predicted and experimental ion fluxes are indicative of differences that arise from the presence of another ion. The predicted anion fluxes can be found by taking the sum of the predicted cation fluxes. The predicted and measured ﬂuxes are compared on loglog plots in Figure 3a and b, where the systems have been separated by the relative mobility of the draw electrolyte based on the permeability coeﬃcients, B. Results from experiments with the more mobile cation in the draw solution are shown in Figure 3a, and results from common anion experiments with the more mobile cation in the feed solution are shown in Figure 3b. Water ﬂux, cation and anion ﬂuxes, and experimental error data for all the systems studied are provided in the SI. The molar ﬂuxes of ions from the draw solution into the feed solution are clustered in the top right region of both panels (Figure 3a and b). As expected, these ﬂuxes are higher than the 10646

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

Figure 2. Comparison of the observed water ﬂux to the predicted water ﬂux for ternary electrolyte experiments. The predicted water ﬂuxes are calculated using equations derived in previous studies.34,40 The aqueous diﬀusion coeﬃcient, D, and feed side mass transfer coeﬃcient, k, for each salt are reported in Table 1, S = 439 μm, and A = 0.44 L m2 h1 bar1. The dashed line (slope = 1 represents perfect agreement between predicted water ﬂux values and those observed from experiments. Two water ﬂuxes are reported for each system of electrolytes because each salt was used once as in the draw solution and once in the feed solution. Only one data point is reported for the KH2PO4 and Na2SO4 containing systems because neither salt was soluble enough to prepare a 1.0 M draw solution. Error bars represent one standard deviation.

feed ion ﬂux for a given experiment.19,20,25,42 In the ternary systems, the draw solution ion ﬂuxes are predicted by eqs 1 and 2 with relative accuracy as demonstrated by the close proximity of data points to the line with slope = 1 (i.e., perfect agreement) in Figure 3a and b. One critique of the model (eq 2) is that it may slightly overpredict the ﬂux of the cation in the feed solution (shown in the lower left quadrant and central region of Figure 3a and b, respectively). Analytical error associated with sample preparation and analysis may partially account for this deviation. Yet, this observation may also indicate that the feed cation is permeating through the active layer at a diﬀerent rate due to the presence of a high concentration of the draw solution ions, as discussed in the interpretation of eq 11. Additional experiments were performed to further explore this observation. The CaCl2 draw solution and KCl feed solution system was selected for further experiments because this system showed the greatest deviations from the model predictions. Experiments were conducted using a CaCl2 draw solution concentration between 0.5 and 2.0 M and a 0.05 M KCl feed solution. Membrane II, which was observed to have slightly diﬀerent permeabilities and structural parameter (Table 1), was used during this portion of the investigation. The molar ﬂux of potassium as a function of CaCl2 concentration is presented in Figure S1.

ARTICLE

Results from experiments show that the ﬂux of potassium ions is independent of the CaCl2 concentration gradient, which is in agreement with the limit presented in eq 11. Furthermore, the measured ﬂux of potassium is consistently lower than the ﬂux predicted by eq 2, which seems to indicate that the presence of the concentrated draw solution enables the feed ion to permeate through the membrane active layer at a rate set by the ionic permeability of potassium, not the potassium chloride permeability. Again, this is consistent with eq 11. An experiment was also performed with a 1 M CaCl2 draw solution and a 0.15 M KCl feed solution to demonstrate that the ﬂux of potassium does increase when the KCl feed concentration increases. Reassuringly, a 3-fold increase in KCl concentration lead to a corresponding increase in potassium ﬂux (11.3 mmol m2 hr1 to 34.2 mmol m2 hr1). The ﬂux of calcium ions from the draw solution into the feed remained constant when the KCl concentration changed (26.4 mmol m2 hr1 at 0.05 M KCl and 27.0 mmol m2 hr1 at 0.15 M KCl). Taken together, these results suggest that if only electrostatic interactions exist between ions from the feed solution and draw solution (i) the ﬂux of ions from the draw solution into feed solution will not be aﬀected by the presence of ions from the feed solution and (ii) the ﬂux of ions from the feed solution into the draw solution will be slightly aﬀected due to the high concentration of draw solution ions. Ion Permeation in Common Cation Systems. Ternary ion experiments were also performed with solutions that contain common cations and dissimilar anions. The predicted and measured ion fluxes are compared on loglog plots in Figure 4a and b. Experiments are divided into two groups: those performed with salts that do not contain nitrate (Figure 4a) and those that do contain nitrate (Figure 4b). Similar to results in Figure 3a and b, the molar fluxes of ions from the draw solution into the feed solution are clustered in the top right region of both figures, while the feed anion data points occupy the central and bottom left regions. For systems that did not contain nitrate, the experimental fluxes of ions from the draw solution were well predicted by the application of eqs 1 and 2 and the permeability coefficients in Table 1. Forward feed ion fluxes again deviated from those predicted by eq 2, although in this case the fluxes were slightly underpredicted. The observation that the feed anions in common cation systems permeate more quickly than predicted while the feed cations in common anion systems permeate more slowly is consistent with the knowledge that ionic diﬀusion coeﬃcients of anions tend to be higher than those of cations.29 Therefore, because the high draw solution concentrations enable the feed solution ions to permeate independently, the cations “slow down” and the anions “speed up” relative to the electrolyte permeability coeﬃcient. The ion ﬂuxes in common cation experiments with nitrate containing salts (Figure 4b) are poorly predicted by eqs 1 and 2. It is possible that this is because interactions between the polar structure of nitrate and the membrane increase the nitrate partitioning into the membrane.43 We are currently pursuing further understanding of the mechanism of nitrate permeation through the membrane. Ion Permeation in Quaternary Systems. Feed streams will typically contain a large number of dissolved ions; however, analytical solutions to the NernstPlanck equation only exist for systems containing up to three ions. In order to explore what aspects of the understanding developed from the analytical solution (eq 11) might be extended to higher order systems, a 10647

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

ARTICLE

Figure 3. Comparison of the observed ion ﬂuxes to the predicted ion ﬂuxes for common anion systems. The systems are divided into two categories: (a) those with the more mobile cation in the draw solution and (b) those with the more mobile cation in the feed solution. Predicted ion ﬂuxes are calculated using eqs 1 and 2 along with the predicted water ﬂuxes, Jw, reported in Figure 2. The transport parameters D, B, and k correspond to those reported in Table 1 for Membrane I, and S = 439 μm. In the legend, the ﬁrst salt is for the draw solution (1.0 M), and the second salt is for the feed solution (0.05 M) (e.g., NaCl-KCl refers to an experiment using a 1.0 M NaCl draw solution and a 0.05 M KCl feed solution). Three ion ﬂuxes are reported for each experiment: the ﬂuxes of the two distinct cations and the ﬂux of the shared common anion. The dashed line (slope = 1) represents perfect agreement between predicted ionic ﬂuxes and those measured from experiments. Error bars represent one standard deviation.

Figure 4. Comparison of the observed ion ﬂuxes to the predicted ion ﬂuxes for common cation ternary electrolyte systems. The systems are separated into two categories: (a) systems that do not use nitrate containing salts and (b) systems that do use nitrate containing salts. Predicted ion ﬂuxes are calculated using eqs 1 and 2 and the predicted water ﬂuxes, Jw, are reported in Figure 2. The transport parameters D, B, and k correspond to those reported in Table 1 for Membrane I, and S = 439 μm. In the legend, the ﬁrst salt is for the draw solution (1.0 M) and the second salt is for the feed solution (0.05 M). Three ion ﬂuxes are reported for each experiment: the ﬂuxes of the two distinct anions and the ﬂux of the shared common cation. The dashed line (slope = 1) represents perfect agreement between predicted ionic ﬂuxes and those observed during experiments. Error bars represent one standard deviation.

limited number of experiments were conducted to investigate bidirectional ion permeation in quaternary systems (i.e., two distinct ions in the feed and another two distinct ions in the draw

solution). Three experiments were conducted, and a fourth was adopted from our previous study.19 Detailed quantitative results are provided in the SI. 10648

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

ARTICLE

of these eﬀorts can aid the future development of ODMPs. Speciﬁcally, the experiments here demonstrate that solute solute (e.g., electrostatic) interactions do exist in the application of ODMPs and do aﬀect the ionic ﬂuxes through the membrane. These eﬀects may be signiﬁcant as is the case with nitrate containing systems or they may be much less important as is the case with electrostatic interactions. If only electrostatic interactions between charged species exists, the model developed here along with the electrolyte permeability coeﬃcients measured from binary systems should adequately predict the reverse and forward ion ﬂuxes in ODMPs. The model predictions will be very accurate for the reverse ion ﬂuxes but less accurate for the forward ion ﬂuxes because of the electrostatic interaction between ions. However, we suspect these diﬀerences will be negligible compared to some issues faced in the operation of ﬁeld systems such as feed stream variability.44

’ ASSOCIATED CONTENT Figure 5. Comparison of the observed ion ﬂuxes to the predicted ion ﬂuxes for quaternary electrolyte systems. Predicted ion ﬂuxes are calculated using eqs 1 and 2, and the predicted water ﬂuxes, Jw, are reported in the SI. The transport parameters D, B, and k correspond to those reported in Table 1, and S = 439 μm. In the legend, the experiments are identiﬁed using the same naming convention as that in Figures 3 and 4. Four ion ﬂuxes are reported for each experiment because each dissolved salt produces two distinct ions. Please note that two symbols for the NaCl-KNO3 system are obscured by symbols from other systems. Both are located in the upper right-hand corner: one is located behind upward facing triangles of the MgSO4KNO3 system and the other behind the downward facing triangles of the NaCl-K2SO4 system. The solid line (slope = 1) represents perfect agreement between predicted ionic ﬂuxes and those observed from experiments. Error bars represent one standard deviation.

Similar to the ternary experiments, water ﬂux in each experiment was predicted using established equations to within 10% of the observed values for these experiments (data not shown). Measured ion ﬂuxes were compared to the ﬂuxes predicted by eqs 1 and 2, assuming that the individual ions permeate at a rate set by the electrolyte permeability coeﬃcient of their parent salts (Table 1). The predicted and measured ion ﬂuxes are compared on a loglog plot shown in Figure 5. The observed results from these experiments are consistent with behaviors identiﬁed in the ternary experiments. For systems that do not contain nitrate, the predicted reverse ﬂux of draw solution ions strongly agrees with the observed values. This indicates that electrostatic interactions between ions have a negligible eﬀect on the reverse permeation of the draw electrolyte. The model predictions of the feed ion ﬂuxes (eq 2) slightly deviate from experiments for these two cases, as we also observed in the ternary systems. For systems that contain nitrate, the model is unable to accurately predict the ion ﬂuxes for the NaCl draw solution and KNO3 feed experiment. Deviations between the model and experiment are more noticeable when an NaCl draw solution is used compared to an MgSO4 draw solution. The development of osmotically driven membrane processes requires that a variety of complex phenomena be understood. One of those phenomena, the bidirectional permeation of ions, was thoroughly explored here through modeling and experimental eﬀorts. Therefore, it is useful to consider how the results

bS

Supporting Information. Details on the derivation of the bidirectional solute permeation model; tabulated experimental results including water ﬂux, solute ﬂux, and experimental error from binary, ternary, and quaternary systems. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected] (W.A.P.), [email protected] (T.Y.C.).

’ ACKNOWLEDGMENT We acknowledge the support of California Department of Water Resources (Grant 46-7446-R-08), the AWWA Abel Wolman Fellowship, and Oasys Water Inc. Special thanks to Hydration Technology Innovations for providing the membrane utilized in this study. ’ NOMENCLATURE A water permeability coeﬃcient B electrolyte permeability coeﬃcient c molar concentration of electrolyte molar concentration of species i ci bulk diﬀusion coeﬃcient of electrolyte in water DDS diﬀusion coeﬃcient of species i when inﬂuenced by Dij gradient of species j average of diﬀusion coeﬃcient of ions i and j Di/j F Faraday constant partition coeﬃcient of species i Hi average partition coeﬃcient of ions i and j Hi/j I current density molar ﬂux of species i Ji volumetric water ﬂux Jw k external mass transfer coeﬃcient R ideal gas constant S membrane structural parameter T absolute temperature thickness of active layer ta thickness of support layer ts x, z local coordinate systems valence of species i zi 10649

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology δ ε ϕ μi π τ

thickness of external concentration polarization layer porosity of support layer electrostatic potential chemical potential of species i osmotic pressure tortuosity of support layer

’ SUBSCRIPTS 1 refers to species originating from the draw solution (electrolyte or ion) 2 refers to species originating from the feed solution (electrolyte or ion) 3 refers to the counterion to ions 1 and 2; in ternary experiments, it is the common ion ’ REFERENCES (1) Cath, T. Y.; Childress, A. E.; Elimelech, M. Forward osmosis: principles, applications, and recent developments. J. Membr. Sci. 2006, 281 (12), 70–87. (2) McGinnis, R. L.; Elimelech, M. Global Challenges in Energy and Water Supply: The Promise of Engineered Osmosis. Environ. Sci. Technol. 2008, 42 (23), 8625–8629. (3) Loeb, S. Osmotic power plants. Science 1974, 189, 350. (4) Yip, N. Y.; Tiraferri, A.; Phillip, W. A.; Schiﬀman, J. D.; Hoover, L. A.; Kim, Y. C.; Elimelech, M. Thin-Film Composite Pressure Retarded Osmosis Membranes for Sustainable Power Generation from Salinity Gradients. Environ. Sci. Technol. 2011, 45 (10), 4360–4369. (5) Achilli, A.; Cath, T. Y.; Marchand, E. A.; Childress, A. E. The forward osmosis membrane bioreactor: A low fouling alternative to MBR processes. Desalination 2009, 239, 10–21. (6) Cartinella, J. L.; Cath, T. Y.; Flynn, M. T.; Miller, G. C.; Hunter, K. W.; Childress, A. E. Removal of natural steroid hormones from wastewater using membrane contactor processes. Environ. Sci. Technol. 2006, 40 (23), 7381–7386. (7) Cath, T. Y.; Gormly, S.; Beaudry, E. G.; Michael, T. F.; Adams, V. D.; Childress, A. E. Membrane contactor processes for wastewater reclamation in space: Part I. Direct osmosis concentration as pretreatment for reverse osmosis. J. Membr. Sci. 2005, 257 (12), 85–98. (8) Cath, T. Y.; Hancock, N. T.; Lundin, C. D.; Hoppe-Jones, C.; Drewes, J. r. E. A multi-barrier osmotic dilution process for simultaneous desalination and puriﬁcation of impaired water. J. Membr. Sci. 2010, 362 (12), 417–426. (9) Holloway, R. W.; Childress, A. E.; Dennett, K. E.; Cath, T. Y. Forward osmosis for concentration of centrate from anaerobic digester. Water Res. 2007, 41, 4005–4014. (10) McCutcheon, J. R.; McGinnis, R. L.; Elimelech, M. A novel ammonia-carbon dioxide forward (direct) osmosis desalination process. Desalination 2005, 174, 1–11. (11) Martinetti, C. R.; Cath, T. Y.; Childress, A. E. High recovery of concentrated RO brines using forward osmosis and membrane distillation. J. Membr. Sci. 2009, 331, 31–39. (12) Mi, B.; Elimelech, M. Organic fouling of forward osmosis membranes: Fouling reversibility and cleaning without chemical reagents. J. Membr. Sci. 2010, 348 (12), 337–345. (13) Lee, S.; Boo, C.; Elimelech, M.; Hong, S. Comparison of fouling behavior in forward osmosis (FO) and reverse osmosis (RO). J. Membr. Sci. 2010, 365 (12), 34–39. (14) Lay, W. C. L.; Chong, T. H.; Tang, C. Y. Y.; Fane, A. G.; Zhang, J. S.; Liu, Y. Fouling propensity of forward osmosis: Investigation of the slower ﬂux decline phenomenon. Water Sci. Technol. 2010, 61, 927–936. (15) Yip, N. Y.; Tiraferri, A.; Phillip, W. A.; Schiﬀman, J. D.; Elimelech, M. High Performance Thin-Film Composite Forward Osmosis Membrane. Environ. Sci. Technol. 2010, 44 (10), 3812–3818. (16) Zhang, S.; Wang, K. Y.; Chung, T.-S.; Chen, H.; Jean, Y. C.; Amy, G. Well-constructed cellulose acetate membranes for forward

ARTICLE

osmosis: Minimized internal concentration polarization with an ultrathin selective layer. J. Membr. Sci. 2010, 360 (12), 522–535. (17) Tiraferri, A.; Yip, N. Y.; Phillip, W. A.; Schiﬀman, J. D.; Elimelech, M. Relating performance of thin-ﬁlm composite forward osmosis membranes to support layer formation and structure. J. Membr. Sci. 2011, 367 (12), 340–352. (18) Bui, N.; Lind, M. L.; Hoek, E. M. V.; McCutcheon, J. R. Electrospun Nanoﬁber Supported Thin Film Composite Membranes for Engineered Osmosis. J. Membr. Sci. 2011, DOI: (doi:10.1016/ j.memsci.2011.08.002). (19) Hancock, N. T.; Cath, T. Y. Solute coupled diﬀusion in osmotically driven membrane processes. Environ. Sci. Technol. 2009, 43, 6769–6775. (20) Phillip, W. A.; Yong, J. S.; Elimelech, M. Reverse Draw Solute Permeation in Forward Osmosis: Modeling and Experiments. Environ. Sci. Technol. 2010, 44 (13), 5170–5176. (21) Xiao, D.; Tang, C. Y.; Zhang, J.; Lay, W. C. L.; Wang, R.; Fane, A. G. Modeling salt accumulation in osmotic membrane bioreactors: Implications for FO membrane selection and system operation. J. Membr. Sci. 2011, 366 (12), 314–324. (22) Cornelissen, E. R.; Harmsen, D.; de Korte, K. F.; Ruiken, C. J.; Qin, J.-J.; Oo, H.; Wessels, L. P. Membrane fouling and process performance of forward osmosis membranes on activated sludge. J. Membr. Sci. 2008, 319 (12), 158–168. (23) Jin, X.; Tang, C. Y.; Gu, Y.; She, Q.; Qi, S. Boric Acid Permeation in Forward Osmosis Membrane Processes: Modeling, Experiments, and Implications. Environ. Sci. Technol. 2011, 45 (6), 2323–2330. (24) Yaroshchuk, A. Inﬂuence of osmosis on the diﬀusion from concentrated solutions through composite/asymmetric membranes: Theoretical analysis. J. Membr. Sci. 2010, 355 (12), 98–103. (25) Achilli, A.; Cath, T. Y.; Childress, A. E. Selection of inorganicbased draw solutions for forward osmosis applications. J. Membr. Sci. 2010, 364 (12), 233–241. (26) Nightingale, E. R. Phenomenological theory of ion solvation. Eﬀective radii of hydrated ions. J. Phys. Chem. 1959, 63 (9), 1381–1387. (27) Donnan, F. G. The Theory of Membrane Equilibria. Chem. Rev. 1924, 1 (1), 73–90. (28) Paul, D. R. Reformulation of the solution-diﬀusion theory of reverse osmosis. J. Membr. Sci. 2004, 241 (2), 371–386. (29) Cussler, E. L. Diﬀusion mass transfer in ﬂuid systems, 3rd ed.; University Press: Cambridge, 2009. (30) Hancock, N. T. Engineered Osmosis: Assessment of Mass Transport and Sustainable Hybrid System Conﬁgurations for Desalination and Water Reclamation. Colorado School of Mines, Golden, 2011. (31) Reusch, C. F.; Cussler, E. L. Selective Membrane Transport. AIChE J. 1973, 19 (4), 736–741. (32) Heyde, M. E.; Anderson, J. E. Ion sorption by cellulose acetate membranes from binary salt solutions. J. Phys. Chem. 1975, 79 (16), 1659–64. (33) Paul, D. R.; Geise, G. M.; Lee, H. S.; Miller, D. J.; Freeman, B. D.; Mcgrath, J. E. Water Puriﬁcation by Membranes: The Role of Polymer Science. J. Polym. Sci., Part B: Polym. Phys. 2010, 48 (15), 1685–1718. (34) McCutcheon, J. R.; Elimelech, M. Inﬂuence of concentrative and dilutive internal concentration polarization on ﬂux behavior in forward osmosis. J. Membr. Sci. 2006, 284 (12), 237–247. (35) Hoek, E. M. V.; Kim, A. S.; Elimelech, M. Inﬂuence of crossﬂow membrane ﬁlter geometry and shear rate on colloidal fouling in reverse osmosis and nanoﬁltration separations. Environ. Eng. Sci. 2002, 19 (6), 357–372. (36) Hofmeister, F. Zur Lehre von der Wirkung der Salze [Title translation: About the science of the eﬀect of salts]. Arch. Exp. Pathol. Pharmakol. 1888, 24, 247–360. (37) Zhang, Y.; Cremer, P. S. Interactions between macromolecules and ions: the Hofmeister series. Curr. Opin. Chem. Biol. 2006, 10, 658–663. 10650

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

Environmental Science & Technology

ARTICLE

(38) Kunz, W.; Lo Nostro, P.; Ninham, B. W. The present state of aﬀairs with Hofmeister eﬀects. Curr. Opin. Colloid Interface Sci. 2004, 9 (12), 1–18. (39) Mallevialle, J.; Odendaal, P. E.; Wiesner, M. R. Water treatment membrane processes; McGraw-Hill: USA, 1996. (40) Mehta, G. D.; Loeb, S. Internal polarization in the porous substructure of a semi-permeable membrane under pressure-retarded osmosis. J. Membr. Sci. 1978, 4, 261. (41) Loeb, S.; Titelman, L.; Korngold, E.; Freiman, J. Eﬀect of porous support fabric on osmosis through a Loeb-Sourirajan type asymmetric membrane. J. Membr. Sci. 1997, 129, 243–249. (42) Robinson, R. A.; Stokes, R. H. Electrolyte Solutions: the measurement and interpretation of conductance, chemical potential, and diﬀusion in solutions of simple electrolytes, 2nd ed. (revised) ed.; Buttersworths Scientiﬁc Publications: London, 1959. (43) Paugam, L.; Taha, S.; Dorange, G.; Joauen, P.; Quemeneur, F. Mechanism of nitrate ions transfer in nanoﬁltration depending on pressure, pH, concentration and medium composition. J. Membr. Sci. 2004, 231, 37–46. (44) Christoﬁdes, P. D.; Zhu, A. H.; Cohen, Y. Energy Consumption Optimization of Reverse Osmosis Membrane Water Desalination Subject to Feed Salinity Fluctuation. Ind. Eng. Chem. Res. 2009, 48 (21), 9581–9589. (45) Li, Y.; Gregory, S. Diﬀusion of ions in seawater and in deep-sea sediments. Geochim. Cosmochim. Acta 1974, 38, 703–714.

10651

dx.doi.org/10.1021/es202608y |Environ. Sci. Technol. 2011, 45, 10642–10651

ARTICLE pubs.acs.org/est

Parental Transfer of Polybrominated Diphenyl Ethers (PBDEs) and Thyroid Endocrine Disruption in Zebrafish Liqin Yu,†,|| James C. W. Lam,§,|| Yongyong Guo,† Rudolf S. S. Wu,‡ Paul K. S. Lam,*,§ and Bingsheng Zhou*,† †

State Key Laboratory of Freshwater Ecology and Biotechnology, Institute of Hydrobiology, Chinese Academy of Sciences, Wuhan 430072, China ‡ School of Biological Sciences, The University of Hong Kong, Hong Kong SAR, China § State Key Laboratory in Marine Pollution; Department of Biology and Chemistry, City University of Hong Kong, Hong Kong SAR, China

bS Supporting Information ABSTRACT: Polybrominated diphenyl ethers (PBDEs) have the potential to disrupt the thyroid endocrine system. The objective of the present study was to characterize the disrupting eﬀects of longterm exposure on the thyroid endocrine system in adult ﬁsh and their progeny following parental exposure to PBDEs. Zebraﬁsh (Danio rerio) embryos were exposed to environmentally relevant concentrations (1, 3, and 10 μg/L) of the PBDE mixture DE-71 for 5 months until sexual maturation. In the F0 generation, exposure to DE-71 signiﬁcantly increased plasma thyroxine (T4) but not 3,5,30 -triiodothyronine (T3) in females. This increased T4 was accompanied by decreased mRNA levels of corticotropinreleasing hormone (CRH) and thyrotropin β-subunit (TSHβ) in the brain. The F1 generation was further examined with or without continued DE-71 treatment conditions. Exposure to DE-71 in the F0 ﬁsh caused signiﬁcant increases in T4 and T3 levels in the F1 larvae and modiﬁed gene expressions in the hypothalamic pituitarythyroid axis (HPT axis) under both conditions. Decreased hatching and inhibition of growth in the F1 oﬀspring were observed in the condition without DE-71 treatment. Continued DE-71 treatment in the F1 embryos/larvae resulted in further decreased hatching, and increased malformation rates compared with those without DE-71 exposure. Analysis of F1 eggs indicated that parental exposure to DE-71 could result in a transfer of PBDEs and thyroid hormones (THs) to their oﬀspring. For the ﬁrst time, we demonstrated that parental exposure to low concentrations of PBDEs could aﬀect THs in the oﬀspring and the transgenerational PBDE-induced toxicity in subsequent nonexposed generations.

’ INTRODUCTION Polybrominated diphenyl ethers (PBDEs) are additives in ﬂame retardants. They are hydrophobic and lipophilic, and some congeners of PBDEs share many characteristics with classical persistent organic pollutants (POPs). Tetra-, penta-, hexa-, and hepta-BDE have recently been added to the POPs list under the Stockholm Convention1 due to great concerns that have arisen concerning the potential environmental and human health risks associated with PBDE exposure. Because of their structural similarity to thyroid hormones (THs), PBDEs are thyroid endocrine disruptors. Decreased concentrations of circulating total thyroxine (T4) have been observed in rats and mice following short- to long-term exposure to various PBDEs.27 In ﬁsh, studies have also shown that PBDEs perturb TH homeostasis. Plasma levels of T4 are lower in PBDEexposed juvenile lake trout (Salvelinus namaycush) compared with control ﬁsh after dietary PBDE exposure.8 Reduced plasma T4 levels have been observed in fathead minnows (Pimephales promelas) r 2011 American Chemical Society

administered BDE-47.9 Exposure of the PBDE mixture DE-71 to zebraﬁsh embryos/larvae has also been shown to result in a reduction of T4 in the larvae.10 Like other lipophilic organic compounds, maternal transfer of several PBDEs has previously been observed in various species, such as frogs,11 birds,12,13 rats,14 and humans.15 Recently, maternal transfer of PBDEs to the oﬀspring following parental exposure via feed was observed in zebraﬁsh in laboratory experiments16 and in the ﬁeld.17 This vertical transfer provides an opportunity for PBDEs to interfere with developmental eﬀects on the progeny. Many ﬁsh species are known to be most sensitive to organic contaminants in their early life stages. In ﬁsh, PBDEs may be transferred maternally in the lipid stores of oocytes, and the Received: August 28, 2010 Accepted: October 31, 2011 Revised: September 28, 2011 Published: October 31, 2011 10652

dx.doi.org/10.1021/es2026592 | Environ. Sci. Technol. 2011, 45, 10652–10659

Environmental Science & Technology

ARTICLE

Table 1. Eﬀects of DE-71 on Zebraﬁsh Developmental Parametersa hatching rates (%) ﬁsh F0

F1(DE-71)

F1(+DE-71)

DE-71 (μg/L)

malformation rates (%)

3d

5d

survival (%) 5d

10d

body weight (mg) 5d

10d

0

78.1 ( 2.1

1.67 ( 0.7

80.3 ( 1.8

78.3 ( 1.3

0.34 ( 0.01

0.38 ( 0.02

1

79.8 ( 2.7

2.33 ( 0.3

81.6 ( 1.6

70.5 ( 2.5

0.35 ( 0.02

0.38 ( 0.02

3

72.6 ( 1.5

2.00 ( 0.6

76.2 ( 1.7

73.7 ( 1.2

0.31 ( 0.01

0.35 ( 0.01

10

71.2 ( 2.7

2.67 ( 0.3

75.7 ( 0.6

69.3 ( 5.8

0.31 ( 0.01

0.33 ( 0.03

0

82.4 ( 2.7

1.17 ( 0.2

80.0 ( 2.0

74.8 ( 3.6

0.33 ( 0.01

0.38 ( 0.02

1

86.0 ( 0.8

1.25 ( 0.3

81.4 ( 1.4

71.1 ( 1.1

0.31 ( 0.03

0.36 ( 0.01

3

74.8 ( 1.5*

1.64 ( 0.5

76.0 ( 2.7

72.0 ( 4.4

0.26 ( 0.02*

0.28 ( 0.00*

10 0

73.2 ( 1.5* 82.4 ( 2.7

1.31 ( 0.3 1.17 ( 0.2

79.8 ( 5.0 80.0 ( 2.0

76.1 ( 4.3 74.8 ( 3.6

0.28 ( 0.01* 0.33 ( 0.01

0.31 ( 0.01* 0.38 ( 0.02

1

69.8 ( 2.5*a

1.36 ( 0.3

76.8 ( 4.2

74.9 ( 4.2

0.26 ( 0.01*

0.28 ( 0.01*

3

62.2 ( 2.6*b

2.89 ( 0.8*b

74.8 ( 2.8

68.9 ( 2.8

0.28 ( 0.01*

0.29 ( 0.03*

10

67.3 ( 4.9*c

2.59 ( 0.3*c

76.5 ( 2.7

64.5 ( 4.2

0.28 ( 0.02*

0.31 ( 0.03*

a

Hatching, malformation, survival and growth in the F0 embryos/larvae exposed to DE-71 and F1 embryos/larvae with (+DE-71) or without (DE-71) continued DE-71 treatments were evaluated. Asterisk indicates signiﬁcantly diﬀerent between exposure groups and their corresponding control (P < 0.05) (One way ANOVA, followed by Tukey’s test). In the F1 generation, the diﬀerences between groups treated with or without continued DE-71 treatment were also compared (Student’s t-test). a,b,cRepresents signiﬁcant diﬀerences between continued (+DE-71) treatment at 1, 3, and 10 μg/L and without (DE-71) treatment, respectively. Results are given as mean values of three replicates of 50 embryos or larvae for each exposure condition at 5 and 10 dpf. All data are expressed as means ( SEM.

oﬀspring can be exposed to these compounds during the earliest stages of embryogenesis. Due to limited reports regarding the thyroid endocrine eﬀects of PBDEs in ﬁsh, and because little information is known about transgenerational toxicity in particular, the purpose of our study was to evaluate the endocrine disrupting activities of PBDE exposure in the thyroid system and the impact on oﬀspring. We selected the DE-71 commercial mixture since it contains tetraand penta-brominated congeners that are prevalent in environmental and biological samples (BDE-47, BDE-99, BDE-100, BDE-153, and BDE-154).18,19 This study focused on parental exposure to DE-71 and the transfer of these compounds to the oﬀspring, and investigated the eﬀects on the thyroid endocrine system in both generations. Speciﬁcally, thyroid hormone levels, gene expression patterns in the hypothalamicpituitarythyroid axis, and developmental toxicity were examined.

’ MATERIALS AND METHODS Chemicals. The commercial PBDE mixture DE-71 (purity >99.9%) was obtained from Wellington Laboratory, Inc. (Ontario, Canada). The TRIzol reagent and SYBR Green PCR kit were purchased from Invitrogen (New Jersey) and Toyobo (Osaka, Japan), respectively. All other chemicals used in the present study were of analytical grade. Zebrafish Maintenance and Experimental Design. Adult zebrafish (AB strain) were maintained and the embryos were exposed to DE-71 as described previously.10 Briefly, the embryos that had developed normally and reached the blastula stage (2 h postfertilization, hpf) were selected for the experiments. The embryos were randomly distributed into glass beakers containing 500 mL of DE-71 exposure solution (0, 1, 3, and 10 μg/L). There were three replicates for each exposure concentration, and each beaker contained 100 embryos. The selected exposure concentrations were based on our previous study.10 At 10 days post fertilization (dpf), the larvae were transferred into 20-L tanks. At 40 dpf, the fry were transferred into 30-L tanks. After 150 days of

exposure, the survival and growth (length and weight) were determined, from which a condition factor was obtained. The fish were paired (18 males and 18 females), and eggs were immediately collected for DE-71 and TH assays. The embryos were divided into two groups: one group received continued treatment with the same DE-71 concentrations as did their parents, and the other group received no further DE-71 treatment. In the no treatment group, the embryos were washed with freshwater five times and placed in glass dishes in freshwater without DE-71 to evaluate the parental transfer of PBDEs and transgenerational toxicity. During all the experimental period, 50% of the exposure solution was renewed daily, and the appropriate concentration of PBDEs was maintained. The control and treated groups received 0.003% (v/v) dimethyl sulfoxide (DMSO). The hatching, malformation, survival, and growth were determined in both generations. The F1 larvae were randomly sampled at 5 and 10 dpf, immediately frozen in liquid nitrogen, and stored at 80 °C for subsequent gene expression analysis and TH assays. TH Assays. After 150 days of exposure, the adult fish were anesthetized in 0.03% MS-222, and blood was collected from the caudal vein of each fish. The blood samples from four fish of the same sex were pooled as one replicate (about 40 μL). The plasma was stored at 80 °C until analysis. The methods for extraction of whole body THs content in eggs and larvae were from a previous method in fathead minnow with a small modiﬁcation (Text S1, Supporting Information). For THs levels in plasma of adult zebraﬁsh and in the F1 eggs and larvae, the total T4 and T3 levels were measured using the commercial enzyme-linked immunosorbent assay (ELISA) test kits purchased from Wuhan EIAab Science Co. Ltd. (Wuhan, China) following the manufacturer’s instructions. Quantitative Real-Time PCR Assay. The liver and brain (including hypothalamus and pituitary) were collected and preserved in TRIzol reagent at 80 °C. Extraction, purification, and quantification of total RNA and first-strand cDNA synthesis were performed as described previously10 (Text S2, Supporting Information). The primer sequences of the selected genes were 10653

dx.doi.org/10.1021/es2026592 |Environ. Sci. Technol. 2011, 45, 10652–10659

Environmental Science & Technology

ARTICLE

Table 2. Total T4 and T3 Levels in F0 Adult Zebraﬁsh after Exposure to DE-71 and in the F1 Eggs and Larvaea DE-71 (μg/L) ﬁsh F0

TH levels T4 T3

eggs F1(-DE-71)

F1(+DE-71)

sex

0

1

3

10

female

20.50 ( 1.08

22.18 ( 1.40

21.06 ( 0.17

27.52 ( 1.52*

male

22.59 ( 1.85

22.50 ( 1.81

20.80 ( 1.60

26.11 ( 4.61

female

2.58 ( 0.30

3.36 ( 0.61

3.63 ( 0.46

4.03 ( 1.30

male

2.67 ( 0.23

3.13 ( 0.59

3.32 ( 0.44

3.41 ( 0.44 11.00 ( 1.27*

T4

7.52 ( 0.18

6.54 ( 1.17

7.80 ( 0.38

T3

0.75 ( 0.02

0.80 ( 0.11

1.11 ( 0.10*

1.28 ( 0.08**

T4

5d

15.30 ( 0.37

14.58 ( 1.05

20.22 ( 2.36

23.97 ( 3.49*

T3

10d 5d

16.53 ( 0.46 0.94 ( 0.11

17.55 ( 3.76 1.12 ( 0.09

21.06 ( 1.88 1.54 ( 0.04*

23.13 ( 1.39*a 1.61 ( 0.16*

10d

1.80 ( 0.24

1.98 ( 0.13

2.26 ( 0.08

2.17 ( 0.16

T4

5d

15.30 ( 0.37

17.68 ( 3.54

19.38 ( 2.65

26.39 ( 2.05**

10d

16.53 ( 0.46

18.48 ( 2.74

19.39 ( 1.13

29.23 ( 1.84**b

T3

5d

0.94 ( 0.11

1.42 ( 0.36

1.21 ( 0.24

1.75 ( 0.14*

10d

1.80 ( 0.24

2.09 ( 0.25

2.09 ( 0.28

2.15 ( 0.26

a For the adult zebraﬁsh, plasma samples from four individual ﬁsh were pooled and tested with three replicates. For the F1 eggs and larvae, 200 eggs or 300 larvae were measured also with three replicates. The eggs were collected immediately after spawning, while the TH contents were measured at 5 and 10 dpf with or without continued DE-71 treatment. The TH levels in the F0 zebraﬁsh are expressed as ng/mL and ng/g wet weight in the F1 eggs and larvae, respectively. All data are expressed as means ( SEM.*P < 0.05 and **P < 0.01 indicate signiﬁcant diﬀerences between exposure groups and the corresponding control group. a,bRepresents signiﬁcant diﬀerences between continued (+DE-71) treatment at 10 μg/L and without (DE-71) treatment (Student’s t-test, P < 0.05).

obtained by using the online Primer 3 program (http://frodo.wi. mit.edu/) (Table S1, Supporting Information). Quantification of PBDEs in F0 Zebrafish and F1 Eggs. Concentrations of PBDEs were determined in the adult fish (F0) and the eggs (F1). The detailed protocols for extraction, clean up, analysis, and quality assurance and quality control (QA/QC) are provided in the Supporting Information (Text S3). Statistical Analysis. All data are expressed as means ( standard error (SEM). The normality of the data was verified using the KolmogorovSmirnov test. The homogeneity of variances was analyzed by Levene’s test. The differences between the control and each exposure group were evaluated by one-way analysis of variance (ANOVA) followed by Tukey’s test by using SPSS 13.0 software (SPSS, Chicago, IL). A P value <0.05 was considered statistically significant.

’ RESULTS Developmental Toxicity in the F0 and F1 Generations. Exposure of the F0 embryos to DE-71 did not significantly affect hatching, malformation, survival rates, or growth. In the F1 generation without DE-71 treatment, however, the hatching rates were significantly decreased at 3 dpf, and inhibition of growth was observed at 5 and 10 dpf (P < 0.05) (Table 1). In the F1 generation with continued DE-71 treatment, decreased hatching at 3 dpf and increased malformation rates were observed at 5 dpf (P < 0.05) (Table 1). There was no significant difference in the survival rates in the F0 and F1 larvae recorded at 5 and 10 dpf (Table 1). Among the groups exposed to different concentrations of DE-71 until sexual maturity, there were no significant differences in the survival rates (60.2 ( 3.7, 60.0 ( 3.3, 61.4 ( 3.5, and 58.6 ( 3.8 in the control, 1, 3, and 10 μg/L groups, respectively). There were also no significant differences in the effects of DE-71 on growth and condition factor in F0 adult fish (Table S2, Supporting Information).

Assessments of THs. In the F0 generation, DE-71 exposure significantly increased the plasma total T4 level by 34.2% in the 10 μg/L treatment group in the females (P < 0.05) (Table 2). There was a slight but nonsignificant increased trend of plasma T4 of males in the exposure groups compared with those in the control group (P > 0.05) (Table 2). The plasma total T3 level also showed a trend toward an increase in exposure groups in both the females and males, but it was again not statistically significant relative to the control (P > 0.05) (Table 2). Whole body levels of THs were measured in the F1 eggs and F1 larvae at 5 and 10 dpf with or without continued DE-71 treatment. In the F1 eggs, the T4 levels were signiﬁcantly increased in the 10 μg/L group (P < 0.05), and T3 levels were increased in the 3 and 10 μg/L groups with parental exposure of 3 and 10 μg/L groups (P < 0.05) (Table 2). At 5 dpf, in the F1 larvae without DE-71 treatment, the total T4 levels were signiﬁcantly increased by 32.2% and 56.5% with parental groups exposed to 3 and 10 μg/L, respectively, compared with those in the control (P < 0.05) (Table 2). For the respective F1 larvae groups given continued treatment with 1, 3, and 10 μg/L of DE-71, concentration-dependent increases of T4 (15.6%, 26.7%, and 72.5%) were detected at 5dpf (P < 0.05) (Table 2); at 10 dpf, increases in T4 were also found in the F1 larvae with (11.8%, 17.3%, 76.8%; P < 0.05) or without (6.2%, 27.4%, 39.9%; P < 0.05) continued DE-71 treatment (Table 2). A signiﬁcant increase in the total T4 levels was observed in the F1 generation with continued DE-71 treatment (+DE-71, 10 μg/L) compared with those without treatment (-DE-71) at 10 dpf (Student’s t-test, P < 0.05) (Table 2). At 5 dpf, in the group without DE-71 treatment, the T3 levels were signiﬁcantly increased (63.8%, 71.3%) with parental groups exposed to 3 and 10 μg/L, respectively (P < 0.05). In the group where DE-71 treatment was continued, a signiﬁcantly increased T3 level (86.2%) was observed with parental exposure to 10 μg/L (P < 0.01) (Table 2). At 10 dpf, a trend of increased T3 levels was 10654

dx.doi.org/10.1021/es2026592 |Environ. Sci. Technol. 2011, 45, 10652–10659

Environmental Science & Technology detected in the groups with (16.1%, 16.1%, 19.4%) and without (10.0%, 25.6%, 20.6%) continued DE-71 treatment, but the diﬀerences were not signiﬁcantly signiﬁcant (P > 0.05) (Table 2). There was no signiﬁcant diﬀerence in the total T3 levels between the F1 generation with continued DE-71 treatment (+DE-71) and the F1 generation without treatment (DE-71) at both 5 and 10 dpf (Student’s t-test, P > 0.05) (Table 2). Gene Expression in F0 and F1 Generations. In the F0 fish, several genes involved in regulation, transport, binding, and metabolism of THs were examined. In the brain of the females, both corticotropin-releasing hormone (CRH) and thyroid-stimulating hormone (TSHβ) gene expressions were significantly down-regulated in the 10 μg/L exposure group compared with those in the control (P < 0.01). Likewise, in the males, a small but significant down-regulation of CRH gene expression was observed in the 10 μg/L exposure group (P < 0.05), and a concentration-dependent down-regulation of TSHβ gene expression was observed in the 1, 3, and 10 μg/L exposure groups (P < 0.05) (Table S3, Supporting Information). In the liver, transthyretin (TTR) gene expression was signiﬁcantly down-regulated in the 10 μg/L exposure group in the females (P < 0.05), while hepatic uridine diphosphoglucuronosyl transferase (UGT1) was increased in the 10 μg/L exposure group (P < 0.01). Expressions of the deiodinases (Dio1 and Dio2) were signiﬁcantly down-regulated in the 10 μg/L treated group compared with those in the control group (P < 0.05). In the liver of the males, down-regulation of TTR, and up-regulation of UGT1 were observed upon exposure to 10 μg/L (P < 0.01). The gene expressions of Dio1 and Dio2 were also down-regulated in the 10 μg/L exposure group (P < 0.05) (Table S3, Supporting Information). In zebraﬁsh larvae (F1), CRH and TSHβ, marker genes involved in thyroid gland development and growth (e.g., hhex and nkx2.1), THs synthesis (e.g., thyroglobulin, TG) and binding (TTR) were examined at 5 and 10 dpf with or without continued DE-71 treatment. The gene expression levels were similar in the control group of both treatments. At 5 dpf, TSHβ gene expression was down-regulated in the groups without DE-71 exposure (P < 0.01). Hhex, nkx2.1, and TG (P < 0.05) were up-regulated in the groups under both conditions. Signiﬁcant up-regulation of Dio1 (P < 0.01) and down-regulation of Dio2 (P < 0.05) were observed in the groups without DE-71 exposure. The down regulation of Dio1 was observed in the larvae subjected to continued treatment with DE-71 in the 3 and 10 μg/L (P < 0.05) and a small but signiﬁcant upregulation of Dio2 was observed in the 10 μg/L group (P < 0.05). However, UGT1 gene expression was downregulated in the group without DE-71 treatment (P < 0.01). The TTR gene expression was not changed in the group without DE-71 exposure, but its up-regulation was found in the group with continued DE-71 treatment (P < 0.01) (Table S4, Supporting Information). At 10 dpf, the CRH gene expressions were signiﬁcantly downregulated in the groups subjected to continued treatment with DE-71 (P < 0.01). The gene expressions of TSHβ were signiﬁcantly down-regulated (P < 0.05) under both conditions, while the hhex and nkx2.1 gene expressions were up-regulated (P < 0.05). Decreases and increases in the TG gene expression were observed in the groups with and without continued DE-71 treatment (P < 0.05), respectively. Signiﬁcant down- and upregulations of Dio1 were measured in the groups with (P < 0.01) and without continued DE-71 exposure (P < 0.01), respectively. Meanwhile, Dio2 gene expression was signiﬁcantly inhibited

ARTICLE

(P < 0.01) under both treatment conditions. The gene expressions of UGT1 (P < 0.05) and TTR (P < 0.01) were signiﬁcantly increased in the group without DE-71 exposure, while those in the DE-71 continued treatment group were down-regulated (P < 0.01) (Table S4, Supporting Information). PBDE Content in F0 Adult Fish and F1 Eggs. Seven congeners were detected in exposed F0 zebrafish, where BDE47 contributed to most of the total PBDE body burden, followed by BDE-100 and 99, in both females and males (Figure 1A). The contents of individual congeners and total PBDEs showed clear dose-dependent relationships between body burdens and their water exposure concentrations. The total body burdens of PBDEs were higher in males than in females in all the exposure groups. The detected total contents of PBDEs were 2.03 ( 0.88 ng/ g wet weight in the control females and 7706 ( 578, 19 954 ( 224, 55 029 ( 4444 in the 1, 3, and 10 μg/L exposure groups, respectively. In the males, the detected total PBDE contents were 2.58 ( 0.62 ng/g in the control and 11 816 ( 1314, 40 821 ( 2646, 118 419 ( 9087 ng/g in the 1, 3, and 10 μg/L exposure groups, respectively. The estimated bioaccumulation factors (detected concentrations in fish/nominated concentrations in water) are 7700, 6650, and 5530 with treatment of 1, 3, and 10 μg/L in the females and 11 820, 13 600, and 11 400 in the males, respectively. In the F1 eggs, seven congeners were detected; BDE-47 was the predominant congener, followed by BDE-100 and BDE-99 (Figure 1B). The total body burden also showed a dose-dependent relationship between parental exposure concentrations of 1, 3, and 10 μg/L with the values of 1689 ( 289, 2569 ( 528, and 13 701 ( 420 ng/g wet weight, respectively. The detected total PBDE content in the control eggs was 0.85 ( 0.34 ng/g.

’ DISCUSSION The aim of the study reported here was to investigate a possible transgenerational eﬀect of DE-71 following parental exposure as this is potentially relevant for environmental risk assessments. The present study demonstrated that parental exposure to low concentrations of DE-71 signiﬁcantly disturbed THs in both generations. Parental exposure to DE-71 did not cause any changes in hatching and maternal growth. However, oﬀspring following DE71 parental exposure demonstrated reduced hatching and growth retardation compared to controls. This observation is consistent with recent reports on parental exposure of DE-71 to ranch minks (Mustela vison)20 and rats.14 Our results indicated that this alteration was transferred to the F1 generation, and developmental eﬀects in oﬀspring were more sensitive than those in their respective parents. A signiﬁcant increase in the T4 level was observed in the adult females. In contrast, most studies have found reduced T4 levels with treatment at higher doses of PBDEs in lake trout (Salvelinus namaycush),8 fathead minnows (Pimephales promelas),9 ﬂounder (Platichthys ﬂesus),21 and mammalian models.27 A previous study also reported that total T4 plasma levels were elevated at 3 and 6 days in adult rats following a single exposure to PBDE99 (8.2 mg/kg), but they returned to normal levels after 12 days.22 In zebraﬁsh, waterborne exposure to DE-71 (0, 5, 16, 50, 160, and 500 μg/L) for 30 days revealed a trend of concentrationdependent increase of plasma T4.21 In our previous study, zebraﬁsh embryos were exposed to DE-71 (1, 3, and 10 μg/L) for 14 days, but T4 was signiﬁcantly reduced at 10 μg/L in the 10655

dx.doi.org/10.1021/es2026592 |Environ. Sci. Technol. 2011, 45, 10652–10659

Environmental Science & Technology

ARTICLE

Figure 1. PBDE content in (A) F0 adult zebraﬁsh after exposure to 0, 1, 3, or 10 μg/L DE-71 for 150 days; (B) PBDEs in the F1 eggs. For adult ﬁsh, the values represent means ( standard error (SEM) of three individual replicate ﬁsh. For eggs, PBDEs were measured in 100 eggs, with three replicate samples. a,b,c,dRepresent the eggs derived from parental exposure of 0, 1, 3, and 10 μg/L DE-71, respectively.

larvae.10 Therefore, the levels of THs may vary due to diﬀerent exposure regimes, and given the evidence that in the environment where mostly low concentrations of PBDEs are detected, testing of long-term lower dose exposures would be better for assessing environmental risks. Decreased serum T3 upon exposure to PBDEs has not generally been observed in previous rodent studies, but a recent study did show a decrease in T3 concentrations from relatively low doses of DE-71 (0.5 mg/L) in ranch mink.20 In our study, an increasing trend of T3 was also observed in the adult zebraﬁsh, and this result is consistent with a previous report, which demonstrated a concentration-dependent increase in T3 levels in zebraﬁsh exposed to DE-71.21 In ﬁsh, THs are present in high quantities in eggs and are presumably of maternal origin.23 Thus, the TH content in eggs probably reﬂects that of the maternal plasma. In the F1 eggs,

signiﬁcant increases in both T4 and T3 contents were measured, indicating maternal transfer of increased THs to the oﬀspring. Thus, the signiﬁcant increases in THs in oﬀspring could also be due to maternal hyperthyroidism caused by DE-71 exposure. This may have a direct eﬀect on the development of thyroid function in the young larvae. However, previous analysis of THs during embryo development in fathead minnow also showed a signiﬁcant rise in both T4 and T3 during the prehatch period, indicating embryonic production of both thyroid hormones and signiﬁcantly increased synthesis of THs in an early developmental stage after hatching in ﬁsh.24,25 In the present study, TH levels were increased in the larvae and the expressions of genes related to thyroid development and growth (e.g., hhex, nkx2.1) and TH synthesis (e.g., TG) were all signiﬁcantly up-regulated as well, indicating that the increased THs in the larvae could also be 10656

dx.doi.org/10.1021/es2026592 |Environ. Sci. Technol. 2011, 45, 10652–10659

Environmental Science & Technology from enhanced synthesis of THs in the larvae. Increased T4 levels have also been observed in ranch mink juveniles after the adult had been treated with lower doses of DE-71 (0.5 mg/L)20 and in female rat oﬀspring at postnatal day 60 following DE-71 exposure (1.7 and 10.2 mg/kg/day).14 Consistent with these previous reports, our study showed that parental exposure to PBDEs could result in increased TH levels and thus thyroid endocrine disruption in the developing oﬀspring. In ﬁsh, CRH and TSH secretions function as common regulators of the thyroidal axis as feedback mechanisms are triggered by changes in the concentrations of circulating THs.26,27 Therefore, the down-regulation of both CRH and TSHβ genes observed in the present study could be explained as a negative feedback response of increased levels of T4. On the other hand, a low dietary dose of BDE-47 was shown to elevate expression of TSHβ in the pituitary gland, associated with reduced plasma levels of T4 in adult fathead minnows.9 In this regard, evaluation of CRH and TSHβ gene expression can be used to determine whether environmental chemicals cause thyroid dysfunction. Although the mechanisms of disturbing THs by PBDE exposure are not well-understood, three general explanations have been proposed, including direct interference at the thyroid gland by changes in thyroid gland histology and morphology,3 interference with TH metabolic enzymes, and interference with the plasma transport of THs.6,28 In the present study, TTR gene expression in adult exposed zebraﬁsh was down-regulated, which may have reduced the amount of TTR proteins available to bind and transport free T4 to target organs. UGTs play a role in decreasing circulating THs, and up-regulation of UGT gene expression or enzyme activities have generally been observed in PBDE-treated animal models.3,29 In our study, the increased UGT1 gene expression could possibly be explained as an autoregulatory response to increased T4 levels, by increased biliary elimination of the conjugated hormone within the thyroid axis. Hepatic deiodinases are important regulators of circulating and peripheral TH levels in vertebrates. In ﬁsh, it has been demonstrated that Dio2 plays a pivotal role in producing active T3, allowing an adequate availability of local and systemic T3.30 Dio1 is mainly expressed in the kidney and is thought to play a minimal role in plasma TH homeostasis, while it has a considerable inﬂuence on iodine recovery and TH degradation.30,31 In our study, both Dio1 and Dio2 were signiﬁcantly down-regulated in the adult ﬁsh liver. Our results are consistent with previous reports showing that hyperthyroidism suppresses Dio1 and Dio2 activities and expression of their mRNA, while hypothyroidism increases them in ﬁsh.30,31 The down-regulation of gene expression of the deiodinases may indicate a regulatory role in response to increased THs, indicating that the observed eﬀects have functional consequences and indicate a true hyperthyroid state. In the F1 larvae, the gene expressions were examined with or without continued DE-71 treatment. In the F1 larvae without continued DE-71 treatment, Dio2 gene expression was downregulated, while Dio1 was up-regulated. The eﬀects of THs on deiodinases in developing zebraﬁsh embryos/larvae have been studied previously. Treatment with 5 nM T3 was shown to downregulate Dio2 mRNA expression in zebraﬁsh larvae, without aﬀecting Dio1 expression.25 Knockdown of Dio1 alone does not aﬀect developmental progression, while Dio2 knockdown results in a clear developmental delay,32 suggesting that Dio2 is the major contributor to TH activation in developing zebraﬁsh embryos/larvae. These results indicate that zebraﬁsh larvae are dependent on T4 to T3 conversion for normal development, and

ARTICLE

down-regulation of Dio2 transcription may aﬀect zebraﬁsh development. The increased T4 levels were measured in F1 larvae without DE-71 treatment, while the gene expressions of CRH and TSHβ were signiﬁcantly down-regulated in the F1 larvae. This response thus can indicate a negative response to the hyperthyroid levels. In addition, it is also well-known that expression levels of hepatic deiodinases respond to both hypo- and hyperthyroidism. Our results are thus consistent with a previous report, which showed that hyperthyroidism suppresses Dio2 mRNA in ﬁsh.30 It is also worth noting that diﬀerent responses of Dio1 and Dio2 were observed in the present study. The upregulation of Dio1 gene expressions could possibly be explained as increased TH degradation and downregulation of Dio2 in response to hyperthyroidism. However, previous reports have also shown mixed results with Dio1 and Dio2 gene expressions and demonstrated that deiodinase genes are expressed in tissue and developmental stage speciﬁc patterns,23 especially during early development in zebraﬁsh.25 Therefore the role of deiodinase enzymes as mediators of TH action in ﬁsh remains to be investigated. In our study, diﬀerent UGT1 and TTR gene expressions were also observed at 5 and 10 dpf. The up-regulation of UGT1 at 10 dpf is consistent with those in adult ﬁsh toward increasing T4 clearance and glucuronidation. The temporally diﬀerential response of this gene expression is not clear. However, several previous studies did not ﬁnd a strong correlation between serum T4 levels and T4-UGT activity following PBDE exposure.3,33 In addition, albumin and thyroxine-binding globulin (TBG) can also bind to THs in ﬁsh plasma.9 Therefore, other mechanisms may need to be examined in order to understand the role of TH-binding proteins. In the present study, we also investigated developmental toxicity, TH levels, and related gene expressions with continued DE-71 treatment of the F1 larvae. This experiment allowed the discrimination between a situation where PBDEs could have a direct eﬀect and that where only the parental eﬀects could play a role. Our results showed similar impacts on TH levels, suggesting that these eﬀects could be due to a transgenerational eﬀect on thyroid endocrine disruption. However, continued treatment until 5 and 10 dpf seemed to have contributed a further increase in T4, while an increasing trend of T3 was also observed. PBDEs have been shown to alter TH levels and clearance and can compete for the thyroid hormone binding protein. Our data suggest that there is a decreased binding or reduced clearance of THs in F1 upon further exposure to low doses of DE-71, as marked downregulation of UGT gene and TTR gene expressions was observed in the group with continued DE-71 treatment. It should also be noted that the malformation rate was further increased in the F1 larvae under continued DE-71 treatment compared with those without DE-71 exposure, indicating an increased sensitivity to toxic eﬀects of PBDEs compared with their parental embryos/larvae (F0). THs have an important role in the regulation of early ﬁsh development. Both beneﬁcial34 and harmful35 eﬀects of increased levels of maternal T3 on subsequent larval development and survival have been reported in ﬁsh species. The reasons for the conﬂicting results are unexplained but may be associated with diﬀerences in the dose and mode of administration of the THs used in the experiments. Exogenous TH induces premature diﬀerentiation of the zebraﬁsh pectoral ﬁns; in particular, they inhibit the development of scales and pigment pattern, and impair the growth of both pectoral and pelvic ﬁns.36,37 In 10657

dx.doi.org/10.1021/es2026592 |Environ. Sci. Technol. 2011, 45, 10652–10659

Environmental Science & Technology zebraﬁsh, excessive exogenous T4 (30 nM) is toxic and causes severe developmental defects.38 A signiﬁcant growth inhibition was also observed in F1 larvae in our study, which was consistent with the results of previous report indicating that PBDEs disruption of TH production in rat F1 oﬀspring causes growth impairment.39 In our study, signiﬁcant levels of DE-71 were detected in the eggs, indicating that DE-71 was transferred from exposed adult ﬁsh to their oﬀspring. Furthermore, these PBDEs can also directly disrupt thyroid function and development in the larvae. Although the maternal transfer is complex, and the present study design did not allow for any detailed analysis of the transfer processes, it clearly indicates that hydrophobic compounds are more eﬃciently transferred than less hydrophobic compounds,16 resulting in increased prevalence of adverse health signs in the oﬀspring. Taken together, our results highlighted the thyroid hormone disruption in the adults, as well as transgenerational thyroid disruption following parental exposure to PBDEs.

’ ASSOCIATED CONTENT

bS

Supporting Information. Text S1 outlines thyroid hormone extraction and assay; Text S2 is a description of gene expression method; Text S3 is a description of PBDE analysis and QA/QC procedures; Table S1 shows primer sequence; Table S2 contains F0 growth and condition factor; Table S3 shows gene expression in adult ﬁsh; Table S4 shows gene expression in F1. This information is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Tel: 852-2788 7681; fax: 852- 2788 7406; e-mail: bhpksl@cityu. edu.hk (P. K. S. L.). Tel: 86-27-68780042; fax: 86-27-68780123; e-mail: [email protected] (B. Z.). )

Notes

Co-ﬁrst authors.

’ ACKNOWLEDGMENT This work was supported by grants from the NSFC (20890113), the Chinese Academy of Sciences (KCZCX2-YW-Q02-05), and FEBL (2008FBZ10). ’ REFERENCES (1) International Institute for Sustainable development (IISD). Summary of the Fourth Conference of the Parties to the Stockholm Convention on Persistent Organic Pollutants; Geneva, Switzerland, 48 May 2009; http://www.unido.org/ﬁleadmin/user_ media/Services/Environmental_ Management/Stockholm_Convention/POPs/ SummaryRepotrofCOP4_01.pdf. (2) Zhou, T.; Ross, D. G.; DeVito, M. J.; Crofton, K. M. Eﬀects of short-term in vivo exposure to polybrominated diphenyl ethers on thyroid hormones and hepatic enzyme activities in weanling rats. Toxicol. Sci. 2001, 61, 76–82. (3) Zhou, T.; Taylor, M. M.; DeVito, M. J.; Crofton, K. M. Developmental exposure to brominated diphenyl ethers results in thyroid hormone disruption. Toxicol. Sci. 2002, 66, 105–116. (4) Stoker, T. E.; Laws, S. C.; Crofton, K. M.; Hedge, J. M.; Ferrell, J. M.; Cooper, R. L. Assessment of DE-71, a commercial polybrominated diphenyl ether (PBDE) mixture, in the EDSP male and female pubertal protocols. Toxicol. Sci. 2004, 78, 144–155. (5) Ellis-Hutchings, R. G.; Cherr, G. N.; Hanna, L. A.; Keen, C. L. Polybrominated diphenyl ether (PBDE)-induced alterations in vitamin

ARTICLE

A and thyroid hormone concentrations in the rat during lactation and early postnatal development. Toxicol. Appl. Pharmacol. 2006, 215, 135–145. (6) Richardson, V. M.; Staskal, D. F.; Ross, D. G.; Diliberto, J. J.; DeVito, M. J.; Birnbaum, L. S. Possible mechanisms of thyroid hormone disruption in mice by BDE 47, a major polybrominated diphenyl ether congener. Toxicol. Appl. Pharmacol. 2008, 226, 244–250. (7) Szabo, D. T.; Richardson, V. M.; Ross, D. G.; Diliberto, J. J.; Kodavanti, P. R.; Birnbaum, L. S. Eﬀects of perinatal PBDE exposure on hepatic phase I, phase II, phase III, and deiodinase 1 gene expression involved in thyroid hormone metabolism in male rat pups. Toxicol. Sci. 2009, 107, 27–39. (8) Tomy, G. T.; Palace, V. P.; Halldorson, T.; Braekevelt, E.; Danell, R.; Wautier, K.; Evans, B.; Brinkworth, L.; Fisk, A. T. Bioaccumulation, biotransformation, and biochemical eﬀects of brominated diphenyl ethers in juvenile lake trout (Salvelinus namaycush). Environ. Sci. Technol. 2004, 38, 1496–1504. (9) Lema, S. C.; Dickey, J. T.; Schultz, I. R.; Swanson, P. Dietary exposure to 2,20 ,4,40 -tetrabromodiphenyl ether (PBDE-47) alters thyroid status and thyroid hormone-regulated gene transcription in the pituitary and brain. Environ. Health Perspect. 2008, 116, 1694–1699. (10) Yu, L.; Deng, J.; Shi, X.; Liu, C.; Yu, K.; Zhou, B. Exposure to DE-71 alters thyroid hormone levels and gene transcription in the hypothalamic-pituitary-thyroid axis of zebraﬁsh larvae. Aquat. Toxicol. 2010, 97, 226–233. (11) Wu, J. P.; Luo, X. J.; Zhang, Y.; Luo, Y.; Chen, S. J.; Mai, B. X.; Yang, Z. Y. Bioaccumulation of polybrominated diphenyl ethers (PBDEs) and polychlorinated biphenyls (PCBs) in wild aquatic species from an electronic waste (e-waste) recycling site in South China. Environ. Int. 2008, 34, 1109–1113. (12) Lundstedt-Enkel, K.; Asplund, L.; Nylund, K.; Bignert, A.; Tysklind, M.; Olsson, M.; Orberg, J. Multivariate data analysis of organochlorines and brominated ﬂame retardants in Baltic Sea guillemot (Uria aalge) egg and muscle. Chemosphere 2006, 65, 1591–1599. (13) Verreault, J.; Villa, R. A.; Gabrielsen, G. W.; Skaare, J. U.; Letcher, R. J. Maternal transfer of organohalogen contaminants and metabolites to eggs of Arctic-breeding glaucous gulls. Environ. Pollut. 2006, 144, 1053–1060. (14) Kodavanti, P. R.; Coburn, C. G.; Moser, V. C.; MacPhail, R. C.; Fenton, S. E.; Stoker, T. E.; Rayner, J. L.; Kannan, K.; Birnbaum, L. S. Developmental exposure to a commercial PBDE mixture, DE-71: neurobehavioral, hormonal, and reproductive eﬀects. Toxicol. Sci. 2010, 116, 297–312. (15) Schecter, A.; Johnson-Welch, S.; Tung, K. C.; Harris, T. R.; P€apke, O.; Rosen, R. Polybrominated diphenyl ether (PBDE) levels in livers of U.S. human fetuses and newborns. J. Toxicol. Environ. Health A 2007, 70, 1–6. (16) Nyholm, J. R.; Norman, A.; Norrgren, L.; Haglund, P.; Andersson, P. L. Maternal transfer of brominated ﬂame retardants in zebraﬁsh (Danio rerio). Chemosphere 2008, 73, 203–208. (17) Ostrach, D. J.; Low-Marchelli, J. M.; Eder, K. J.; Whiteman, S. J.; Zinkl, J. G. Maternal transfer of xenobiotics and eﬀects on larval striped bass in the San Francisco Estuary. Proc. Natl. Acad. Sci. U.S.A. 2008, 105, 19354–19359. (18) Mazdai, A.; Dodder, N. G.; Abernathy, M. P.; Hites, R. A.; Bigsby, R. M. Polybrominated diphenyl ethers in maternal and fetal blood samples. Environ. Health Perspect. 2003, 111, 1249–1252. (19) Hites, R. A. Polybrominated diphenyl ethers in the environment and in people: A meta-analysis of concentrations. Environ. Sci. Technol. 2004, 38, 945–956. (20) Zhang, S.; Bursian, S. J.; Martin, P. A.; Chan, H. M.; Tomy, G.; Palace, V. P.; Mayne, G. J.; Martin, J. W. Reproductive and developmental toxicity of a pentabrominated diphenyl ether mixture, DE-71, to ranch mink (Mustela vison) and hazard assessment for wild mink in the Great Lakes region. Toxicol. Sci. 2009, 110, 107–116. (21) Kuiper, R. V.; Vethaak, A. D.; Canton, R. F.; Anselmo, H.; Dubbeldam, M.; van den Brandhof, E. J.; Leonards, P. E.; Wester, P. W.; van den Berg, M. Toxicity of analytically cleaned pentabromodiphenylether after prolonged exposure in estuarine European ﬂounder (Platichthys 10658

dx.doi.org/10.1021/es2026592 |Environ. Sci. Technol. 2011, 45, 10652–10659

Environmental Science & Technology

ARTICLE

ﬂesus), and partial life-cycle exposure in fresh water zebraﬁsh (Danio rerio). Chemosphere 2008, 73, 195–202. (22) Hakk, H.; Larsen, G.; Klasson-Wehler, E. Tissue disposition, excretion and metabolism of 2,20 ,4,40 ,5-pentabromodiphenyl ether (BDE99) in the male Sprague-Dawley rat. Xenobiotica 2002, 32, 369–382. (23) Power, D. M.; Llewellyn, L.; Faustino, M.; Nowell, M. A.; Bjornsson, B. T.; Einarsdottir, I. E.; Canario, A. V.; Sweeney, G. E. Thyroid hormones in growth and development of ﬁsh. Comp. Biochem. Physiol., Part C: Toxicol. Pharmacol. 2001, 130, 447–459. (24) Crane, H. M.; Pickford, D. B.; Hutchinson, T. H.; Brown, J. A. Developmental changes of thyroid hormones in the fathead minnow, Pimephales promelas. Gen. Comp. Endrocrinol. 2004, 139, 55–60. (25) Walpita, C. N.; Van der Geyten, S.; Rurangwa, E.; Darras, V. M. The eﬀect of 3,5,30 -triiodothyronine supplementation on zebraﬁsh (Danio rerio) embryonic development and expression of iodothyronine deiodinases and thyroid hormone receptors. Gen. Comp. Endrocrinol. 2007, 152, 206–214. (26) De Groef, B.; Van der Geyten, S.; Darras, V. M.; K€uhn, E. R. Role of corticotropin-releasing hormone as a thyrotropin-releasing factor in non-mammalian vertebrates. Gen. Comp. Endrocrinol. 2006, 146, 62–68. (27) Geven, E. J. W.; Flik, G.; Klaren, P. H. M. Central and peripheral integration of interrenal and thyroid axes signals in common carp (Cyprinuscarpio L.). J. Endocrinol. 2009, 200, 117–123. (28) Birnbaum, L. S.; Staskal, D. F. Brominated ﬂame retardants: Cause for concern? Environ. Health Perspect. 2004, 112, 9–17. (29) Hallgren, S.; Darnerud, P. O. Polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs) and chlorinated paraﬃns (CPs) in rats-testing interactions and mechanisms for thyroid hormone eﬀects. Toxicology 2002, 177, 227–243. (30) Orozco, A.; Valverde, R. C. Thyroid hormone deiodination in ﬁsh. Thyroid 2005, 15, 799–813. (31) Van der Geyten, S.; Byamungu, N.; Reyns, G. E.; Kuhn, E. R.; Darras, V. M. Iodothyronine deiodinases and the control of plasma and tissue thyroid hormone levels in hyperthyroid tilapia (Oreochromis niloticus). J. Endocrinol. 2005, 184, 467–479. (32) Walpita, C. N.; Crawford, A. D.; Janssens, E. D.; Van der Geyten, S.; Darras, V. M. Type 2 iodothyronine deiodinase is essential for thyroid hormone-dependent embryonic development and pigmentation in zebraﬁsh. Endocrinology 2009, 150, 530–539. (33) Kuriyama, S. N.; Wanner, A.; Fidalgo-Neto, A. A.; Talsness, C. E.; Koerner, W.; Chahoud, I. Developmental exposure to low-dose PBDE-99: Tissue distribution and thyroid hormone levels. Toxicology 2007, 242, 80–90. (34) Ayson, F. G.; Lam, T. J. Thyroxine injection of female rabbitﬁsh (Siganus guttatus) broodstock: Changes in thyroid hormone levels in plasma, eggs, and yolk-sac larvae, and its eﬀect on larval growth and survival. Aquaculture 1993, 109, 83–93. (35) Mylonas, C. C.; Sullivan, C. V.; Hinshaw, J. M. Thyroid hormones in brown trout (Salmo trutta) reproduction and early development. Fish Physiol. Biochem. 1994, 13, 485–493. (36) Brown, D. D. The role of thyroid hormone in zebraﬁsh and axolotl development. Proc. Natl. Acad. Sci. U. S. A. 1997, 94, 13011–13016. (37) Wendl, T.; Lun, K.; Mione, M.; Favor, J.; Brand, M.; Wilson, S. W.; Rohr, K. B. Pax2.1 is required for the development of thyroid follicles in zebraﬁsh. Development 2002, 129, 3751–3760. (38) Liu, Y. W.; Chan, W. K. Thyroid hormones are important for embryonic to larval transitory phase in zebraﬁsh. Diﬀerentiation 2002, 70, 36–45. (39) Blake, C. A.; McCoy, G. L.; Hui, Y. Y.; LaVoie, H. A. Perinatal exposure to low-dose DE-71 increases serum thyroid hormones and gonadal osteopontin gene expression. Exp. Biol. Med. 2011, 236, 445–455.

10659

dx.doi.org/10.1021/es2026592 |Environ. Sci. Technol. 2011, 45, 10652–10659

ARTICLE pubs.acs.org/est

Bioassay-Directed Identification of Novel Antiandrogenic Compounds in Bile of Fish Exposed to Wastewater Effluents Pawel Rostkowski,† Julia Horwood,† Janice A. Shears,‡ Anke Lange,‡ Francis O. Oladapo,† Harrie T. Besselink,§ Charles R. Tyler,‡ and Elizabeth M. Hill*,† †

Department of Biology and Environmental Science, University of Sussex, BN1 9QG Brighton, U.K. School of Biosciences, University of Exeter, EX4 4PS Exeter, U.K. § BioDetection Systems BV (BDS), Science Park 406, 1098 XH Amsterdam, Netherlands ‡

bS Supporting Information ABSTRACT: The widespread occurrence of feminized male ﬁsh downstream of some UK Wastewater Treatment Works (WwTWs) has been associated with exposure to estrogenic and potentially antiandrogenic (AA) contaminants in the eﬄuents. In this study, proﬁling of AA contaminants in WwTW eﬄuents and ﬁsh was conducted using HPLC in combination with in vitro androgen receptor transcription screens. Analysis of extracts of wastewater eﬄuents revealed complex proﬁles of AA activity comprising 2153 HPLC fractions. Structures of bioavailable antiandrogens were identiﬁed by exposing rainbow trout to a WwTW eﬄuent and proﬁling the bile for AA activity using yeast (anti-YAS) and mammalianbased (AR-CALUX) androgen receptor transcription screens. The predominant fractions with AA activity in both androgen receptor screens contained the germicides chlorophene and triclosan, and together these contaminants accounted for 51% of the total anti-YAS activity in the ﬁsh bile. Other AA compounds identiﬁed in bile included chloroxylenol, dichlorophene, resin acids, napthols, oxybenzone, 4-nonylphenol, and bisphenol A. Pure standards of these compounds were active in the androgen receptor screens at potencies relative to ﬂutamide of between 0.1 and 13.0. Thus, we have identiﬁed, for the ﬁrst time, a diverse range of AA chemicals in WwTWs that are bioavailable to ﬁsh and which need to be assessed for their risk to the reproductive health of these organisms and other aquatic biota.

’ INTRODUCTION The widespread occurrence of feminized male ﬁsh downstream of discharge from UK wastewater treatment works (WwTWs) has led to substantial interest from environmental biologists, government organizations, and industry.1,2 The phenomenon of feminized ﬁsh has further been observed in freshwater and marine environments throughout the world. Feminized responses include the formation of the egg-yolk precursor, vitellogenin, within male and juvenile animals as well as histopathological changes in reproductive organs such as testes-ova and reductions in sperm count and motility.1,3 These eﬀects have been attributed to exposure to environment estrogens, particularly steroidal estrogens present in WwTWs eﬄuents.4 However, a survey of UK WwTWs has revealed that the majority of the eﬄuents sampled contain antiandrogenic (AA) activity (between 21.31231 μg of ﬂutamide equivalents/L) as well as estrogenic activity (between 0.442.7 ng of estradiol (E2) equivalents/L).5 Feminization of ﬁsh at river sites was correlated with their predicted exposure to both antiandrogens and estrogens or to antiandrogens alone.6 There is good evidence that exposure of ﬁsh to some antiandrogens in the laboratory can cause feminization and sexual disruption. For instance exposure of ﬁsh to either environmental or clinical antiandrogens induced intersexuality in male and ovarian r 2011 American Chemical Society

atresia in female medaka7 and resulted in a reduction of sperm count and male secondary sexual features in guppies and fathead minnows. It also inhibited androgen-induced spiggin production in the stickleback.8 The widespread occurrence of antiandrogens in WwTW eﬄuents may contribute (alongside estrogens) to the gonadal disruption of ﬁsh in UK rivers. Therefore, it is essential that antiandrogens present in wastewaters are identiﬁed and to allow for their possible associated risks to wildlife health to be properly established. The nature of AA compounds in UK eﬄuents is currently unknown. Structures of chemicals containing androgen receptor antagonist properties can be extremely diverse and include environmental contaminants such as insecticides or their metabolites (e.g., pp0 -DDE and certain pyrethroids), fungicides (vinclozolin and procymidone), herbicides (linuron and prochloraz), components of sunscreen products, some industrial contaminants such as selected PCB congeners and pharmaceuticals (e.g., ﬂutamide and cyproterone acetate).2 It is unlikely that many of these environmental antiandrogens would be present in UK Received: March 16, 2011 Accepted: November 2, 2011 Revised: October 26, 2011 Published: November 02, 2011 10660

dx.doi.org/10.1021/es202966c | Environ. Sci. Technol. 2011, 45, 10660–10667

Environmental Science & Technology WwTW eﬄuents which arise primarily from domestic (sewage) inputs. Therefore, analytical approaches such as bioassay-directed fractionation and analysis are needed to identify unknown compounds with AA activity in wastewater samples. Identiﬁcation of key active compounds in wastewaters is diﬃcult as they can often be present at subnanogram per liter concentrations. However, a wide range of xenobiotics can bioconcentrate in ﬁsh bile at concentrations tens of thousands greater than in the eﬄuent itself, facilitating the structural identiﬁcation of bioavailable contaminants present in the ambient environment.9,10 Bioassaydirected fractionation of bile has proven to be eﬀective in studies identifying estrogens,9 androgens, and some antiandrogens taken up into ﬁsh from WWTW eﬄuents.11 However in that previous work, due to complexity of the bile fractions and the limited amount of bile available to purify and detect antiandrogens, only two antiandrogens were identiﬁed; dichlorophene and 9,10di(chloromethyl)anthracene, and together these accounted for less than 5% of the AA activity in the bile samples.11 In this study, we reveal the complex proﬁles of AA activity that are present in eﬄuent samples from WwTWs, and we identify a comprehensive array of AA contaminants bioavailable (in bile) from trout exposed to a WwTW eﬄuent using a bioassay-directed analytical approach. A composite bile sample from mature trout held in a tank and exposed to eﬄuent was used to determine the structures of AAs in the ﬁsh, and an additional exposure, using juvenile trout, was used to conﬁrm the identiﬁcation of the key bioavailable AAs contributing to the AA activity in the bile. To identify AA structures, bile extracts were fractionated using reverse phasehigh performance liquid chromatography (RP-HPLC) and the fractions interrogated for antiandrogen activity using two in vitro assays: a yeast and a mammalian cell-based androgen receptor transcription assay, anti-YAS and AR-CALUX, respectively. Fractions containing antiandrogen activity were analyzed by GC-MS to identify key structures.

’ EXPERIMENTAL SECTION Materials see Supporting Information (SI). WwTW Sites. Three grab samples (12.5 L) of final effluent from each of the 3 WwTWs (A, B, and C) were collected in solvent-rinsed glass containers. The population equivalents of the 3 WwTWs were between 47,200 and 142,370 and the influent consisted of primarily domestic input, and the effluent had been processed by both primary and secondary treatments (details on the treatment works are given in SI, Table S1). Methanol (final concentration 3%) and acetic acid (final concentration 1%) were added to the samples which were stored overnight at 4 °C prior to solid phase extraction (SPE) of the samples. Fish Exposure. Trout were raised in the Hatherley Laboratories, University of Exeter, UK. The fish were exposed to undiluted effluent from only WwTW C in two tanks (1 m3, flow rate of 10 L/min) for a period of 10 days. The exposure protocol consisted of two tanks: one tank contained 9 mature female rainbow trout (Oncorhynchus mykiss) that were 2+ years old (in their third year of growth) (mean length ( SEM of 37.1 ( 0.6 cm and weight 465.3 ( 22.9 g) in order to obtain enough bile for method development and structural analyses of AAs. The other tank contained 15 smaller juvenile fish that were 1+ year old (in their second year of growth) (length 26.0 ( 0.5 cm and weight of 121.1 ( 1.4 g, n = 15). The effluent exposures were not replicated; however, samples from these juvenile fish were used to obtain analytical replicates in order to confirm the identity of

ARTICLE

AA structures in the bile. A similar population of mature and juvenile trout were held in dechlorinated tap water in the laboratory for 10 days. Trout were fed daily on a commercial trout food until 2 days prior to the end of the experiment to maximize bile production. At the end of the exposure periods, fish were anaesthetized and sacrificed, and the bile sacs were removed for analysis of antiandrogens. Extraction and Purification of AA Compounds. In order to prepare enough bile for preparation of semipurified fractions and unequivocal chemical identification of antiandrogenic structures, a composite sample (1.8 mL) was prepared from bile of mature trout exposed to effluent C. Aliquots of this sample were used for method development and for structural identification of AA compounds in the bile. Three different composite samples of bile were prepared from juvenile trout exposed to effluent C in another tank, and these were used to replicate the chemical analyses; each sample (200 μL volume) represented a composite from 5 different fish. Similar composite bile samples were prepared from control trout held in dechlorinated tap water. Metabolites in the composite bile samples were deconjugated as described elsewhere.10,11 Extraction of wastewater and bile samples by SPE was similar to that described earlier11 with the modification that samples were eluted sequentially from the cartridges with methanol, dichloromethane (instead of ethyl acetate used in previous work11), and hexane. The use of dichloromethane ensured that acceptable recoveries of chlorinated antiandrogens (identified previously11) were achieved from the SPE. The efficiency of SPE was tested using trout bile spiked with standard androgens and antiandrogens (for analytical details see the SI). In order to ensure all AA compounds in the bile were extracted by Oasis HLB, the solution eluting from the SPE cartridge during sample loading, as well as the washes, were collected from the cartridge and re-extracted on ion exchange SPE (details in the SI). The SPE eluents were dried down under vacuum and redissolved in ethanol for bioassay in the anti-YAS. After bioassay analysis, extracts were dried down and redissolved in acetonitrile:water (90:10, v:v) for RP-HPLC fractionation. RP-HPLC Fractionation. Extracts of two replicate grab samples of each WwTW effluent, and analytical replicates of composite bile samples from mature fish (two replicates) and juvenile fish (three replicates) were fractionated by HPLC using an acetonitrile/water system (see the SI for details). HPLC fractions of bile or wastewater samples were collected every minute for analysis in receptor bioassays. With some bile fractions, where subsequent GC-MS analyses indicated they contained a complex mixture of xenobiotics, the fractions were repurified on HPLC using water: acetonitrile gradients between 70:30% to 10:90% (30 min) and retested in the anti-YAS. GC-MS Analysis. The identities of chemicals in fractions with AA activity were investigated using gas chromatograph ion trap mass spectrometer (GC-MS) analyses (see the SI for details). Identified compounds were quantified with a four point linear regression calibration curve using a ratio of internal standard and selected ions for each compound of interest. Steroid Receptor Screens. Anti-YAS. AA activities of bile or effluent samples were quantified using a recombinant yeast screen that contains androgen receptor (YAS). This assay has been validated for a range of AA contaminants.12 Samples used for total AA activity measurements (before HPLC fractionation) and pure standards were serially diluted in ethanol, which was evaporated to dryness before addition of culture media. To test for receptor antagonist activity, the agonist (DHT) was added to 10661

dx.doi.org/10.1021/es202966c |Environ. Sci. Technol. 2011, 45, 10660–10667

Environmental Science & Technology

ARTICLE

Figure 1. Proﬁles of the antiandrogenic activity in extracts of ﬁnal eﬄuent (μg FEq/L) from WwTW A, B, C and in extract of bile from trout exposed to eﬄuent WwTW C (μg FEq/mL). Proﬁles are representative of 2 replicate grab samples of each of the wastewaters and 2 replicate analyses of one composite bile sample prepared from mature trout held in a tank exposed to WwTW C. No antiandrogenic activity was detected in blank workup samples of dechlorinated tap water or in proﬁles from bile of trout held in clean water. Dotted lines indicate limit of detection (LOD) values.

the yeast medium at a concentration giving a 65% submaximal response of the assay. AA activity was quantified as flutamide standard equivalents (FEq). Samples showing toxicity which resulted in poor yeast growth (monitored at 620 nm and in comparison with blank samples containing only ethanol and media) were not quantified. The toxicity of concentrations of standards of putative antiandrogens was also assessed by analysis of the 620 nm response in the YAS without addition of DHT to the media. In some cases, where toxicity of standard compound was observed only at the highest concentrations, then additional DHT (27.4 ng/mL) was added back to the wells containing cell incubations with the serial dilutions of test compound. This allowed the detection of any latent toxicity (measured as a reduced response to the DHT agonist) in the dilution curve. AR-CALUX. Bile fractions from HPLC analysis (mature fish) and pure standards of putative antiandrogens revealed by the anti-YAS and GC-MS analysis were retested in the AR-CALUX bioassay (BioDetection Systems, Netherlands) as a confirmation of androgen receptor antagonist activity. The AR-CALUX assay is a reporter gene assay consisting of a human osteoblast cell line that contains a luciferase gene under transcriptional control of an androgen responsive element. The assay was performed according to refs 13 and 14 using a 24 h exposure. Bile fractions in dimethyl sulfoxide (DMSO) were analyzed in triplicate in the presence of the EC50 of DHT (2 1010 M). The amount of

activity (FEq) was determined by interpolating the response of the sample into the concentrationresponse of the reference compound. Cell toxicity was identified if cell cultures were found to be detached from the multiwell plate surface.

’ RESULTS AND DISCUSSION Profiles of Antiandrogen Activity (Anti-YAS) in Wastewaters and in Fish Bile. Anti-YAS analysis of AA activity of the

SPE extracts of three replicate grab samples of each wastewater revealed that effluents A, B, and C contained 260 ( 40 μg, 468 ( 51 μg, and 214 ( 27 μg FEq/L, respectively. Extracts from two of the replicate grab samples from each WwTW effluent were fractionated by HPLC, and the fractions were analyzed for AA activity using the anti-YAS. This revealed that the profiles of AA activity in the three WwTW effluents were complex, with each sample containing clusters of polar, moderately polar, and nonpolar fractions (Figure 1). The effluent profiles contained between 21 (WwTWs A and B) and 53 (WwTW C) fractions with AA activity at or above the LOD value of the method. In contrast, previous studies have revealed that estrogenic profiles in wastewaters are dominated by far fewer active fractions.11,15 In this study, the influent of the three WwTWs plants was primarily domestic origin and was treated by biological aerated filter and/ or activated sludge processing (see Table S1). However the 10662

dx.doi.org/10.1021/es202966c |Environ. Sci. Technol. 2011, 45, 10660–10667

Environmental Science & Technology effluent of WwTWs A and B were also subjected to additional treatments including sand filtering (WwTW A) or denitrification stages (WwTW B) which may account for the lower number of AA fractions detected in these effluents compared with WwTW C. The determination of the structures of the AA chemicals in an effluent extract could be challenging due to the possibility of low concentrations of such a complex number of AA chemicals. Therefore, the profiles of AA activity in bile from rainbow trout exposed to an effluent containing the most AA fractions (WwTW C) were analyzed. Development of Analytical Methods To Profile Antiandrogen Activity in Fish Bile. SPE methods were used to extract AA compounds in bile extracts. Using this methodology, standard androgens (dihydrotesterone and testosterone) and AA compounds (flutamide, p,p0 -DDE, dichlorophene, bisphenol A) were spiked into bile samples from control fish which were extracted using Oasis HLB SPE. Analysis of the bile extracts by GC-MS revealed high SPE recoveries of the standards of between 7489% (Table S2). Three analytical replicates from a composite bile sample prepared from effluent-exposed mature fish were used to determine the recoveries of AA activity from SPE. Recoveries of bile extracts were assessed by the comparison of the total AA activity of samples of bile (after hydrolysis) before and after Oasis HLB SPE and after additional serial extractions of the sample solution using two types of ion exchange SPEs. The majority of AA activity in the original bile sample was extracted with Oasis HLB cartridge (97.4 ( 1.6%,), with a very small part recovered with Oasis WAX and MCX SPE (0.6 ( 0.1% and 1.5 ( 0.1%, respectively). There was no detectable AA activity in blank workup samples that were carried over all steps of analytical procedure or in bile from the control population of trout held in dechlorinated tap water (LOD < 2.1 μg of FEq/mL of bile). However, the total AA activity in bile of effluent-exposed mature trout was 1840 ( 140 μg FEq/mL compared with 214 ( 27 μg FEq/L in the effluent, which indicated an average bioconcentration of AA chemicals (in bile) of 8,600 fold, similar to levels estimated in previous studies.11 Analysis of bile extracts from the effluent-exposed mature trout revealed a simpler profile than in effluent WwTW C used in the exposures, with about 30 mainly moderately polar and nonpolar fractions eluting from the HPLC between 16 and 61 min (Figure 1). Differences between the bile and effluent profiles could be due to a number of factors; that either some of the polar or nonpolar AA chemicals that were present in the effluent were not bioavailable to fish or that some bioavailable AAs were not detected in the bile as they were metabolized and deactivated or excreted in the urine. Xenobiotics can be present in bile as phase 1 metabolites,16 and these structures may have little or no AA activity compared to the parent structures. Similarly, although glucuronide and sulfate conjugates were hydrolyzed back to the parent structures, other phase II metabolites of AAs, such as glutathione conjugates, may not have been detected in the anti-YAS. Also polar or low molecular weight structures tend to be excreted in urine rather than the bile.17 Nevertheless, this study indicates that a variety of xenobiotic compounds bioconcentrate in the bile of the fish allowing for the identification of bioavailable AA structures arising from exposure to the effluent. The most active AA fractions in bile of mature ﬁsh eluted at retention times of 40, 41, and 46 min (Figure 1). Similar proﬁles of AA activity were observed in bile samples obtained from exposure of juvenile ﬁsh to WwTW C where some fractions eluted one min earlier than proﬁles of mature ﬁsh. (Table 1). The total

ARTICLE

AA in bile of juvenile ﬁsh was 2210 ( 230 μg FEq/mL, and the recovery of AA activity after HPLC proﬁling of the bile was 119.0 ( 11.2% (n = 3 analytical replicates). Together, these studies indicated that all the AA activity present in ﬁsh bile was recovered using the analytical methodologies of HLB SPE and RP-HPLC fractionation. Our studies indicate that steroidal androgens, if present in the bile extracts, would have been extracted by SPE. The presence of any androgen receptor agonists may have reduced the estimation of total AA activity in the bile extracts as HPLC fractionation did increase the amount of total AA activity estimated from the summed fractions. Identification of Key Antiandrogen Structures in Fish Bile. GC-MS analyses of HPLC fractions with antiandrogen activity derived from a composite bile sample from mature fish resulted in detection of a variety of contaminants with AA activity (Table S3). The identities of the antiandrogens were also confirmed by analysis of corresponding fractions with AA activity in bile from juvenile fish (Table 1). The antiandrogens identified in the bile fractions included the following: biocides such as triclosan, chlorophene, dichlorophene, and chloroxylenol. Isomers of resin acids (abietic acid, pimaric acid, isopimaric acid, neoabietic acid) were also identified as AA as well as bisphenol A (a plastic monomer), a chlorinated anthracene compound and metabolites of PAHs (hydroxypyrene, naphthols). Other compounds with AA activity included 4-nonylphenol (a surfactant product), oxybenzone, an ingredient used in sunscreen filters, and (in juvenile fish) dihydroxybiphenyl. These compounds had the same retention times on HPLC and GC-MS and the same mass spectra as the commercially available standards (see examples in Figure S1A, B). Additional contaminants were tentatively identified in other fractions containing AA activity, and these were positional isomers of hydroxypyrene, chlorinated and dichlorinated derivatives of bisphenol A, a methoxy metabolite of chlorophene, additional isomers of resin acids, and a triclosan analogue diclosan. However, due to lack of available commercial standards for these compounds, it was not possible to confirm their exact structure or their AA activity in the androgen receptor screens. Mefenamic acid was also detected in bile fractions from juvenile and mature fish but the pure standard was toxic in the anti-YAS. Analysis of pure commercial standards of the identiﬁed structures in the ﬁsh bile conﬁrmed that the compounds were antiandrogenic in the anti-YAS bioassay. Examples of the dose response curves are shown in Figure 2 and potency values relative to the ﬂutamide standard are given in Tables 1 and S3. To exclude the possibility that toxicity was responsible for the inhibition in the anti-YAS, two tests were performed. First, yeast growth in the YAS (without DHT coexposure) was monitored during incubation with the standard compounds using absorbance values at 620 nm to measure turbidity. If toxicity was observed at the highest concentrations, standards were measured again in YAS in the presence of an excess of DHT. That led to an increase absorbance (measured at the wavelength of 540 nm), and these values were compared with those of a blank positive control (ethanol only plus media with DHT). This allowed detection of the inhibition in the original anti-YAS which could be due to toxicity rather than receptor antagonist activity (examples shown in Figure S2). Using this approach, none of the standards except for 1-hydroxypyrene showed toxicity in the YAS at concentrations up to 10-fold higher than the corresponding IC50 antagonist values in the anti-YAS. In the case of 1-hydroxypyrene, reduction in yeast growth and in the response 10663

dx.doi.org/10.1021/es202966c |Environ. Sci. Technol. 2011, 45, 10660–10667

Environmental Science & Technology

ARTICLE

Table 1. Antiandrogenic Compounds Identiﬁed in the Bile of Juvenile Trout Exposed to Wastewater Eﬄuent from WwTW C and Their Contribution to the Total Antiandrogenic Activity in the Bile As Measured by the Anti-YASa,b

Two asterisks denote the following: antiandrogenic potency possibly overestimated due to toxicity in the anti-YAS. b Data are mean ( s.d. values from 3 analytical replicates, each prepared from composite samples of diﬀerent ﬁsh held in one exposure tank. In this HPLC analysis, active fractions and AA standards eluted 1 min earlier compared to proﬁles presented in Figure 1 and Table S3. The sum of all the compounds identiﬁed and tested in the antiYAS accounted for 60.3 ( 5.7% of the total antiandrogenic activity measured in all the bile fractions. The LOD of the anti-YAS = 20 ( 7 μg FEq/mL of bile. Additional compounds (diclosan and methoxy metabolite of chlorophene) were identiﬁed in fraction 42, but standards were not available to test in the anti-YAS. Mefenamic acid was identiﬁed in fractions 40, 41, but a pure standard was toxic in the anti-YAS. a

to addition of excess DHT was detected at concentrations g1.4 106 M which were 2-fold higher than the IC50 values in the antiYAS. Therefore, the androgen receptor antagonist activity of this compound in the yeast screen cannot be veriﬁed.

AR-CALUX Analysis of Bile Antiandrogens. Due to the limited permeable nature of the yeast cell wall which comprises complex polysaccharides, there was the possibility that not all of the AA xenobiotics present in the bile of effluent-exposed trout 10664

dx.doi.org/10.1021/es202966c |Environ. Sci. Technol. 2011, 45, 10660–10667

Environmental Science & Technology

ARTICLE

Contribution of Identified Structures to the Total AA Activity in Fish Bile. The contribution of the antiandrogens

Figure 2. Doseresponse curves of pure standard compounds tested in the anti-YAS Mean ( s.d. IC50 values (n = 3) of pure standard compounds are given in brackets (M): ﬂutamide (7.07 106 ( 9.47 109), abietic acid (1.77 106 ( 1.06 107), triclosan (1.47 106 ( 9.89 108), 1-hydroxypyrene (7.14 107 ( 5.61 108), chlorophene (5.43 107 ( 3.15 108), dichlorophene (1.50 106 ( 9.09 108), 4-nonylphenol (2.36 105 ( 2.483 106), bisphenol A (1.18 105 ( 5.58 107). Yeast cell toxicity was detected at g1.4 106 M concentrations of 1-hydroxypyrene.

were detected in the anti-YAS. Therefore, bile fractions from the composite sample from mature fish and pure standards of the identified AA compounds were retested in the AR-CALUX assay which is based on an osteosarcoma cell line. Both in vitro assays resulted in a similar profile of AA activity in the bile, and the two predominant fractions with AA activity in the anti-YAS were also principal AA fractions in the AR-CALUX (Figure S3). However, there were a number of minor antiandrogen fractions that were only detected in one of the assays. This observation may be due to the compound specific differences in sensitivity between the two screens. The AR-CALUX is 10-fold more sensitive to flutamide than the anti-YAS (The IC50 of flutamide in the anti-YAS is 1.21.8 mg/L, whereas in the AR-CALUX it is 0.090.15 mg/L.). Commercial standards of all the antiandrogens previously identified in the anti-YAS analysis of bile were also AA in the ARCALUX, and none (including hydroxypyrene) showed toxicity in the AR-CALUX (Figure S4). However, although the potencies (relative to flutamide) of 4-nonylphenol and bisphenol A were similar in both screens, the potencies of triclosan, chlorophene, dichlorophene, and abietic acid were all higher in the anti-YAS compared with the AR-CALUX. The difference in potencies between the two assays for certain chemicals may be due to a number of factors, including metabolic transformation in the osteosarcoma cell line used in the AR-CALUX. It is unlikely that the difference in potencies is due to the solubility of the chemicals in different carrier solvents. Work in our laboratory revealed that regardless of the solvent used in the anti-YAS, the potencies of the chemicals relative to flutamide were similar: 4.8 ( 0.3 and 4.1 ( 0.8 for triclosan (mean ( s.d., n = 3) and 13.0 ( 0.3 and 15.8 ( 1.2 for chlorophene in ethanol and DMSO, respectively. Therefore, even using the DMSO as the same carrier solvent, there were still differences in potency for these compounds between the two assays. The permeability of the cell wall or membrane could also be a factor, but studies have indicated that the yeast cell wall is permeable to small (<600 g/mol) polar or nonpolar chemicals.18 Our findings highlight that two widely used in vitro assays for screening antiandrogenic activity, although producing similar qualitative analyses, can produce results that differ quantitatively, and this is something that warrants further investigations to enable harmonization of data across studies using the different screens.

identified in effluent-exposed juvenile or mature fish to the AA activity in the HPLC fractions and in the bile extracts were estimated from their concentrations determined by GC-MS and their relative potency values to flutamide in the yeast screen (Table 1 and S3). In addition, the corresponding fraction in the effluent extracts from WwTWs A, B, and C were analyzed by GCMS to determine the presence of the antiandrogen in the effluent samples. Chlorophene contributed to 27% of the total AA activity in bile from juvenile or mature trout and was the most potent antiandrogen detected in the bile with a potency relative to ﬂutamide of 13. It was also detected in the analogous HPLC fraction from proﬁles of the three WwTWs eﬄuents at concentrations between 32 and 311 ng/L. Chlorophene is a germicide that is widely used in disinfectant products, and concentrations of 142850 ng/L have been reported in ﬁnal WwTW eﬄuents in the UK and Spain.19,20 To our knowledge it is the ﬁrst report in which its AA properties have been recognized. Triclosan, another germicide, contributed 24% and 15% to the total AA activity in the bile of juvenile and mature ﬁsh, respectively, and was detected in all three eﬄuents at concentrations between 50 and 100 ng/L. Triclosan is used in cosmetics and other personal care products and is widely detected in environmental samples.21 Triclosan has been reported previously as an androgen antagonist in vitro22 and also to disrupt thyroid hormone signaling and decrease serum thyroxine in male juvenile rats.23 The analysis of HPLC fractions of the eﬄuent extracts revealed that although all three WwTWs contained chlorophene and triclosan, the concentrations detected in both compounds would have been below or close to the LOD of the proﬁling methods used for analyses of the eﬄuent extracts (Figure 1). A number of other contaminants were identiﬁed in bile from eﬄuent-exposed trout which individually contributed <4% to the total AA activity in the bile (Table 1). These included other chlorinated phenolics, such as dichlorophene, an antimicrobial agent reported previously,11 and chloroxylenol (para-chlorometa-xylenol) which is actively used in disinfectants and antiseptics and has been used as a preservative in pharmaceuticals and cosmetic products. Both these compounds were detected in the three WwTW eﬄuents at concentrations of 10450 ng/L (dichlorophene) and 19140 ng/L (chloroxylenol). The AA activity of chloroxylenol has not been reported previously; however, together with chlorophene and triclosan, these germicides have been detected in the bile of bream (Abramis brama) collected from rivers in The Netherlands.24 Hydroxypyrene is recognized as a biomarker of exposure to polycyclic aromatic hydrocarbons (PAHs) and is commonly detected in ﬁsh bile.25 It was not detected in HPLC fractions of the WwTWs eﬄuents indicating that it was a product of ﬁsh metabolism. Although 1-hydroxypyrene was toxic in the antiYAS, it appeared to be a possible androgen antagonist in the ARCALUX screen. We also identiﬁed for the ﬁrst time naphthol isomers as other weakly active PAH metabolites in the anti-YAS assay. A number of xenobiotics with known estrogenic activity were also identiﬁed contributing to the androgen receptor antagonist activity of the bile. Oxybenzone is the chemical found in most commercial sunscreens as well as other cosmetics and plastic mixtures and was detected in eﬄuent of WwTW C at concentrations of 8 ng/L. In vitro studies have shown that oxybenzone is 10665

dx.doi.org/10.1021/es202966c |Environ. Sci. Technol. 2011, 45, 10660–10667

Environmental Science & Technology estrogenic as well antiestrogenic and antiandrogenic.26 Bisphenol A, a component of polycarbonate plastics and food packaging, contributed <1% to the AA activity of the bile and was detected in the WwTWs eﬄuents at concentrations between <2.5 to 25 ng/L although concentrations of up to 7 μg/L have been reported at times for some WwTWs.27 Similarly the contribution of 4-nonylphenol to the AA activity of the bile was insigniﬁcant. 4-Nonylphenol can be derived from degradation of 4-nonylphenol ethoxylates surfactants in WwTW, and it was detected in eﬄuent from WwTW C at concentration of 0.42 μg/L. The levels of 4-nonylphenol in the environment have declined since the use of such compounds have been banned or strictly monitored in many countries, although it can still be found at concentrations of up to 4 μg/L in some river waters.28 Both bisphenol A and 4-nonylphenol have been reported to be estrogen receptor agonists and androgen receptor antagonists when tested in steroid receptor transcription screens.12,29 A mixture of resin acids, including abietic acid and its analogues, were detected in HPLC fractions of bile and also in the eﬄuent of WwTW C at a concentration of 510 ng/L. Resin acids are tricyclic diterpenes that occur naturally in the resin of tree wood and bark and are transferred to process waters during pulping operations. In paper mill eﬄuents they have been detected at concentrations between 0.02 to 12 mg/L,30 and the detection of abietic acid in the WwTW eﬄuent in this study indicates that they are also present in eﬄuent from domestic sources. We report for the ﬁrst time that the three resin acid structures tested were androgen receptor antagonists in both the anti-YAS and AR-CALUX screens, although they did not give a full response in the latter. Using androgen receptor bioassay-directed analysis, this study has revealed that wastewater eﬄuents of primarily domestic origin contain a complex mixture of compounds with AA activity. The germicides chlorophene and triclosan were identiﬁed as the predominant androgen receptor antagonists bioconcentrating in bile from ﬁsh exposed to one of the eﬄuents, and these compounds are currently being investigated for antiandrogenic properties in further in vivo tests in ﬁsh. Further work is needed to replicate this study using independent exposures with this and other WwTW eﬄuents. Antiandrogens can aﬀect sexual development and function, and there is a reasonable likelihood that exposure to the mixtures of antiandrogens taken up into ﬁsh may result in disruption of sexual diﬀerentiation or exacerbate the feminizing eﬀects of estrogen exposure in male ﬁsh. This would have signiﬁcant implications for risk assessment of wastewater eﬄuents on aquatic wildlife, particularly as many (UK) WwTW eﬄuents contain AA activity.5 In this study germicidal chemicals were the predominant bioavailable antiandrogens identiﬁed in ﬁsh exposed to a domestic wastewater; however, it is likely that the proﬁle of bioavailable AA contaminants will diﬀer in other eﬄuent types. As an example, the resin acids detected in our study are likely to contribute signiﬁcantly to AA activity in ﬁsh exposed to paper mill eﬄuents. Recent work has revealed that the principal contaminants with AA activity in eﬄuents from oil production platforms were napthenic acids, PAHs, and alkylphenols.31 Thus the nature of bioavailable antiandrogens present in contaminated ﬁsh are likely to be complex and variable reﬂecting the nature and sources of the eﬄuent. Adding to concern about antiandrogens, a growing number of environmental chemicals are being recognized to demonstrate androgen receptor antagonist activity.32,33 It should also be highlighted that to date bioassay-directed analytical studies used to identify environmental

ARTICLE

antiandrogens have relied on human androgen receptor transcription screens, and the parity between the ﬁsh and the human androgen receptor in aﬃnity for environmental chemicals still remains to be investigated. Our work demonstrates the importance of establishing the full spectrum of bioavailable endocrine disrupting chemicals present in the environment, and highlights further the need for testing environmentally relevant mixtures for establishing their potential for deleterious eﬀects on ﬁsh reproductive health.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional information available as noted in the text. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 44-01273-67-8382. Fax: 44-01273- 877586. E-mail: [email protected].

’ ACKNOWLEDGMENT This work was funded by the UK Natural Environmental Research Council (NE/E017363/1 and NE/E016634/1) to E.H. and C.R.T. ’ REFERENCES (1) Gross-Sorokin, M. Y.; Roast, S. D.; Brighty, G. C. Assessment of feminization of male ﬁsh in English rivers by the Environment Agency of England and Wales. Environ. Health Perspect. 2006, 114 (Suppl 1), 147–51. (2) Hotchkiss, A. K.; Rider, C. V.; Blystone, C. R.; Wilson, V. S.; Hartig, P. C.; Ankley, G. T.; Foster, P. M.; Gray, C. L.; Gray, L. E. Fifteen years after 00 Wingspread00 - Environmental endocrine disrupters and human and wildlife health: Where we are today and where we need to go. Toxicol. Sci. 2008, 105 (2), 235–259. (3) Jobling, S.; Beresford, N.; Nolan, M.; Rodgers-Gray, T.; Brighty, G. C.; Sumpter, J. P.; Tyler, C. R. Altered sexual maturation and gamete production in wild roach (Rutilus rutilus) living in rivers that receive treated sewage eﬄuents. Biol. Reprod. 2002, 66 (2), 272–281. (4) Jobling, S.; Williams, R.; Johnson, A.; Taylor, A.; Gross-Sorokin, M.; Nolan, M.; Tyler, C. R.; van Aerle, R.; Santos, E.; Brighty, G. Predicted exposures to steroid estrogens in UK rivers correlate with widespread sexual disruption in wild ﬁsh populations. Environ. Health Perspect. 2006, 114, 32–39. (5) Johnson, I.; Hetheridge, M. J.; Tyler, C. Assessment of (Anti-) Oestrogenic and (Anti-) Androgenic Activities of Final Eﬄuents from Sewage Treatment Works; SC020118/SR; Environment Agency: Bristol, 2007. (6) Jobling, S.; Burn, R. W.; Thorpe, K.; Williams, R.; Tyler, C. Statistical modeling suggests that antiandrogens in eﬄuents from wastewater treatment works contribute to widespread sexual disruption in ﬁsh living in English rivers. Environ. Health Perspect. 2009, 117 (5), 797–802. (7) Kiparissis, Y.; Metcalfe, T. L.; Balch, G. C.; Metcalfe, C. D. Eﬀects of the antiandrogens, vinclozolin and cyproterone acetate on gonadal development in the Japanese medaka (Oryzias latipes). Aquat. Toxicol. 2003, 63 (4), 391–403. (8) Jolly, C.; Katsiadaki, I.; Morris, S.; Le Belle, N.; Dufour, S.; Mayer, I.; Pottinger, T. G.; Scott, A. P. Detection of the antiandrogenic eﬀect of endocrine disrupting environmental contaminants using in vivo and in vitro assays in the three-spined stickleback. Aquat. Toxicol. 2009, 92 (4), 228–239. 10666

dx.doi.org/10.1021/es202966c |Environ. Sci. Technol. 2011, 45, 10660–10667

Environmental Science & Technology (9) Gibson, R.; Smith, M. D.; Spary, C. J.; Tyler, C. R.; Hill, E. M. Mixtures of estrogenic contaminants in bile of ﬁsh exposed to wastewater treatment works eﬄuents. Environ. Sci. Technol. 2005, 39 (8), 2461–71. (10) Fenlon, K. A.; Johnson, A. C.; Tyler, C. R.; Hill, E. M. Gas liquid chromatographytandem mass spectrometry methodology for the quantitation of estrogenic contaminants in bile of ﬁsh exposed to wastewater treatment works eﬄuents and from wild populations. J. Chromatogr., A 2010, 1217 (1), 112–118. (11) Hill, E. M.; Evans, K. L.; Horwood, J.; Rostkowski, P.; Oladapo, F. O.; Gibson, R.; Shears, J. A.; Tyler, C. R. Proﬁles and some initial identiﬁcations of (anti)androgenic compounds in ﬁsh exposed to Wastewater Treatment Works eﬄuents. Environ. Sci. Technol. 2010, 44 (3), 1137–1143. (12) Sohoni, P.; Sumpter, J. P. Several environmental oestrogens are also antiandrogens. J. Endocrinol. 1998, 158 (3), 327–339. (13) Sonneveld, E.; Jansen, H. J.; Riteco, J. A.; Brouwer, A.; van der Burg, B. Development of androgen- and estrogen-responsive bioassays, members of a panel of human cell line-based highly selective steroidresponsive bioassays. Toxicol. Sci. 2005, 83 (1), 136–48. (14) van der Burg, B.; Winter, R.; Man, H. Y.; Vangenechten, C.; Berckmans, P.; Weimer, M.; Witters, H.; van der Linden, S. Optimization and prevalidation of the in vitro AR CALUX method to test androgenic and antiandrogenic activity of compounds. Reprod. Toxicol. 2010, 30 (1), 18–24. (15) Gibson, R.; Smith, M.; Spary, C.; Tyler, C.; Hill, E. Mixtures of estrogenic contaminants in bile of ﬁsh exposed to wastewater treatment works eﬄuents. Environ. Sci. Technol. 2005, 39 (8), 2461–2471. (16) James, M. O. Conjugation of Organic Pollutants in Aquatic Species. Environ. Health Perspect. 1987, 71, 97–103. (17) Glickman, A. H.; Hamid, A. A. R.; Rickert, D. E.; Lech, J. J. Elimination and Metabolism of Permethrin Isomers in Rainbow-Trout. Toxicol. Appl. Pharmacol. 1981, 57 (1), 88–98. (18) Denobel, J. G.; Barnett, J. A. Passage of Molecules through Yeast-Cell Walls - a Brief Essay-Review. Yeast 1991, 7 (4), 313–323. (19) Kasprzyk-Hordern, B.; Dinsdale, R. M.; Guwy, A. J. The occurrence of pharmaceuticals, personal care products, endocrine disruptors and illicit drugs in surface water in South Wales, UK. Water Res. 2008, 42 (13), 3498–3518. (20) Bueno, M. J.; Aguera, A.; Gomez, M. J.; Hernando, M. D.; Garcia-Reyes, J. F.; Fernandez-Alba, A. R. Application of liquid chromatography/quadrupole-linear ion trap mass spectrometry and time-ofﬂight mass spectrometry to the determination of pharmaceuticals and related contaminants in wastewater. Anal. Chem. 2007, 79 (24), 9372–84. (21) Kolpin, D. W.; Furlong, E. T.; Meyer, M. T.; Thurman, E. M.; Zaugg, S. D.; Barber, L. B.; Buxton, H. T. Pharmaceuticals, hormones, and other organic wastewater contaminants in U.S. streams, 19992000: a national reconnaissance. Environ. Sci. Technol. 2002, 36 (6), 1202–11. (22) Ahn, K. C.; Zhao, B.; Chen, J.; Cherednichenko, G.; Sanmarti, E.; Denison, M. S.; Lasley, B.; Pessah, I. N.; Kultz, D.; Chang, D. P.; Gee, S. J.; Hammock, B. D. In vitro biologic activities of the antimicrobials triclocarban, its analogs, and triclosan in bioassay screens: receptorbased bioassay screens. Environ. Health Perspect. 2008, 116 (9), 1203–10. (23) Zorrilla, L. M.; Gibson, E. K.; Jeﬀay, S. C.; Crofton, K. M.; Setzer, W. R.; Cooper, R. L.; Stoker, T. E. The eﬀects of triclosan on puberty and thyroid hormones in male Wistar rats. Toxicol. Sci. 2009, 107 (1), 56–64. (24) Houtman, C. J.; van Oostveen, A. M.; Brouwer, A.; Lamoree, M. H.; Legler, J. Identiﬁcation of Estrogenic Compounds in Fish Bile Using Bioassay-Directed Fractionation. Environ. Sci. Technol. 2004, 38 (23), 6415–6423. (25) Hurlbert, S. H. Pseudoreplication and the Design of Ecological Field Experiments. Ecol. Monogr. 1984, 54 (2), 187–211. (26) Kunz, P.; Fent, K. Multiple hormonal activities of UV ﬁlters and comparison of in vivo and in vitro estrogenic activity of ethyl-4aminobenzoate in ﬁsh. Aquat. Toxicol. 2006, 79 (4), 305–324. (27) Hohne, C.; Puttmann, W. Occurrence and temporal variations of the xenoestrogens bisphenol A, 4-tert-octylphenol, and tech.

ARTICLE

4-nonylphenol in two German wastewater treatment plants. Environ. Sci. Pollut. Res. 2008, 15 (5), 405–416. (28) Soares, A.; Guieysse, B.; Jeﬀerson, B.; Cartmell, E.; Lester, J. N. Nonylphenol in the environment: A critical review on occurrence, fate, toxicity and treatment in wastewaters. Environ. Int. 2008, 34 (7), 1033–1049. (29) Lee, H. J.; Chattopadhyay, S.; Gong, E. Y.; Ahn, R. S.; Lee, K. Antiandrogenic eﬀects of bisphenol A and nonylphenol on the function of androgen receptor. Toxicol. Sci. 2003, 75 (1), 40–6. (30) Quinn, B. P.; Booth, M. M.; Delﬁno, J. J.; Holm, S. E.; Gross, T. S. Selected resin acids in eﬄuent and receiving waters derived from a bleached and unbleached kraft pulp and paper mill. Environ. Toxicol. Chem. 2003, 22 (1), 214–8. (31) Thomas, K. V.; Langford, K.; Petersen, K.; Smith, A. J.; Tollefsen, K. E. Eﬀect-directed identiﬁcation of naphthenic acids as important in vitro xeno-estrogens and antiandrogens in North Sea oﬀshore produced water discharges. Environ. Sci. Technol. 2009, 43 (21), 8066–8071. (32) Fang, H.; Tong, W.; Branham, W. S.; Moland, C. L.; Dial, S. L.; Hong, H.; Xie, Q.; Perkins, R.; Owens, W.; Sheehan, D. M. Study of 202 Natural, Synthetic, and Environmental Chemicals for Binding to the Androgen Receptor. Chem. Res. Toxicol. 2003, 16 (10), 1338–1358. (33) Vinggaard, A. M.; Niemela, J.; Wedebye, E. B.; Jensen, G. E. Screening of 397 chemicals and development of a quantitative structureactivity relationship model for androgen receptor antagonism. Chem. Res. Toxicol. 2008, 21 (4), 813–823.

10667

dx.doi.org/10.1021/es202966c |Environ. Sci. Technol. 2011, 45, 10660–10667

ARTICLE pubs.acs.org/est

Physicochemical and Toxicological Properties of Commercial Carbon Blacks Modified by Reaction with Ozone Brian C. Peebles,† Prabir K. Dutta,*,† W. James Waldman,‡ Frederick A. Villamena,§ Kevin Nash,§ Michael Severance,† and Amber Nagy‡ †

Department of Chemistry, The Ohio State University, 100 West 18th Avenue, Columbus, Ohio 43210, United States Department of Pathology, The Ohio State University, 4160 Graves Hall, 333 West 10th Avenue, Columbus, Ohio 43210, United States § Department of Pharmacology, The Ohio State University, 390 Biomedical Research Tower, 460 West 12th Avenue, Columbus, Ohio 43210, United States ‡

bS Supporting Information ABSTRACT: Ozonation of two commercial carbon blacks (CBs), Printex 90 (P90) and Flammruss 101 (F101), was carried out and changes in their morphology, physical properties, and cytotoxicity were examined. The hypothesis examined was that diﬀerent methods of manufacture of CBs inﬂuence their chemical reactivity and toxicological properties. Structural changes were examined by X-ray photoelectron spectroscopy, infrared spectroscopy, Raman spectroscopy, and electron paramagnetic resonance spectroscopy (EPR). Introduction of surface oxygen functionality upon ozonation led to changes in surface charge, aggregation characteristics, and free radical content of the CBs. However, these changes in surface functionality did not alter the cytotoxicity and release of inﬂammation markers upon exposure of the CBs to murine macrophages. Interaction of macrophages with F101 resulted in higher levels of inﬂammatory markers than P90, and the only structural correlation was with the higher persistent radical concentration on the F101.

’ INTRODUCTION Carbon black (CB) is the product of incomplete combustion of hydrocarbon feedstock and is an important technological material. It is used as a pigment in paints, inks, and toners, a reinforcing agent in rubber and polymers, a conductivity enhancer in polymers, an adsorbent, and a support for catalysts. The many industrial uses of CB and carbonaceous materials make it necessary to tailor their physical and chemical properties. Surface functionalization of CB is an active area of research; in particular, using ozone to introduce oxygen functionalities.17 Carbon black is also chemically similar to the carbonaceous component of environmentally relevant particles such as soot and coal ﬂy ash.2,8 There is considerable research activity in the use of nanoparticles in biological systems,9 and the physiological eﬀects of CBs, particularly their toxic and inﬂammatory response, are of interest.10,11 In this study, two commercial CBs, Flammruss 101, a lamp black (F101, prepared by burning liquids with a restricted air supply and quenched by deposition on a cool surface), and Printex 90, a furnace black (P90, made by partial combustion of residual aromatic oils, and quenched by water) were examined. The hypothesis behind this study is that methods of manufacture of CBs determine their chemical reactivity, as evidenced by oxidative functionalization and toxicology properties, as measured by interaction of the CBs with murine macrophages. Because of their technological relevance, there have been numerous studies on toxicity and eﬀects of workplace exposure to CBs. Of particular interest is the report of a lack of r 2011 American Chemical Society

lung tumor formation in mice upon inhalation exposure to P90, as compared to controls.12 For intratracheally administered F101 and P90 in rats, the tumor incidence was higher in rats exposed to P90 at 21% (10 out of 48 rats) compared to 8% (4 out of 48 rats) exposed to F101.13 In this study, we report that the evolution of surface oxygen functionality upon ozonation inﬂuences surface charge, morphology, size, and toxicity. The cytotoxicity and inﬂammatory properties of CBs and correlations with the surface chemistry are also relevant to occupational and environmental exposure.

’ EXPERIMENTAL SECTION Physicochemical Characterization. F101 and P90 CB particles (from Degussa) were characterized as obtained, and after ozonolysis. Ozonolysis of F101 and P90 was performed over a 4-h period using an Enaly EOZ-300Y corona discharge ozone generator (Ozone Solutions, Hull, Iowa) that produces approximately 40 mg of ozone per hour in a stream of 20% oxygen. The generator was supplied with a flow of approximately 100 mL/minute compressed air that had been passed through a hydrocarbon trap and a desiccant. The ozone generator was connected to the tip of a buret, which was loaded with 500 mg of CB (fluid bed geometry). Received: March 17, 2011 Accepted: November 4, 2011 Revised: October 31, 2011 Published: November 04, 2011 10668

dx.doi.org/10.1021/es202984t | Environ. Sci. Technol. 2011, 45, 10668–10675

Environmental Science & Technology

ARTICLE

Figure 1. TEM images of (a) untreated P90 and (b) untreated F101.

X-ray photoelectron spectroscopy was performed using a Kratos Ultra Axis spectrometer with a monochromatic Al Kα source. Spectral analysis was performed using CasaXPS software. Ultraviolet photoelectron spectroscopy was performed using a He (I) ultraviolet light source with a helium pressure of ∼5 108 Torr. The EPR spectrometer (Bruker EMX X-band) was operated at the X-band (9.87 GHz) with 10 mW microwave power, 1 G modulation amplitude, receiver gain of 1.0 105, 21.5 s scan time, and 42 s time constant with a 120 G sweep width. A Perkin-Elmer Spectrum 400 FTIR spectrometer was ﬁtted with a Pike Technologies 100-mm gas cell with calcium ﬂuoride windows. The carbon samples were coated on the IR window and exposed to ozone. Baseline-corrected spectra were translated to zero at 2000 cm1, a point at which all of the spectra had approximately the same slope and no spectral features. Powders were pressed onto the surfaces of Teﬂon discs and placed on the stage of a Renishaw InVia Raman microprobe equipped with a 633 nm HeNe laser. Particles were imaged by transmission electron microscopy using a Tecnai F20 transmission electron microscope. Dynamic light scattering (DLS) and zeta potential measurements were made using a Malvern ZetaSizer Nano instrument (sample loading 10 mg/L, 30 s sonication prior to use). Macrophage Toxicity Studies. The murine alveolar macrophage cell line (MH-S) was purchased from the American Type Culture Collection (Manassas, VA), and propagated as described previously.14 Cells were plated in 24 well plates at a density of 1 105 cells/well. Carbon blacks were suspended in PBS, sonicated for 30 s and added to confluent murine macrophage monolayers at a concentration ranging from 2.5 to 10 μg/cm2. Cells treated with 1% Triton-100 (Sigma) served as a positive control for lactate dehydrogenase (LDH) release assays, while cells treated with E. coli lipopolysaccharide (LPS, Sigma) served as the positive control for tumor necrosis factor-alpha (TNF-α) ELISAs. Supernatants were collected from each well and clarified by centrifugation at 16 000 RCF for 2 min to pellet uninternalized carbon nanoparticles. Lactate dehydrogenase and TNF-α secretion was measured via ELISA assays as described previously.14 ELISA assays for IL-1β were also carried out according to manufacturer’s instructions (R&D Systems, Minneapolis, MN). There was no interference of CB on the assays, and these results are shown in Supporting Information.

’ RESULTS Characterization of CB and Its Reaction with Ozone. Morphology. Transmission electron microscope images of the two

Figure 2. Changes in particle diameter (dashed) and zeta potential (solid) for (a) P90 and (b) F101 upon exposure to ozone (particles dispersed in water).

carbon samples are shown in Figure 1. Flammruss 101 comprises well-defined primary particles that have an average diameter of ∼100 nm, with little surface “roughness”, whereas P90 has more “roughness” on its surface. These micrographs are consistent with the Heckman and Harling model of CB structure.15 The morphologies of P90 and F101 were not noticeably changed after 4 h of ozone exposure (Figure S1, Supporting Information). The surface areas of P90 and F101 measured by BET were 320 and 33 m2/g, respectively. Upon ozonation for 4 h, the surface areas changed to 316 and 44 m2/g for P90 and F101, respectively. Size and Surface Charge. Figure 2a shows that the average hydrodynamic diameter of untreated Printex 90 was ∼180 ( 0.7 nm in water at pH 5, with an average zeta potential of ∼26.3 ( 1.9 mV. The positive zeta potential reflects a basic carbon black surface.16 This basicity is also reflected as an increase in the pH of ultrapure water from pH 6.7 to 7.1 with suspended P90 (50 mg/mL). For F101 (Figure 2b), the initial 10669

dx.doi.org/10.1021/es202984t |Environ. Sci. Technol. 2011, 45, 10668–10675

Environmental Science & Technology

ARTICLE

Figure 3. (a) Raman spectrum of P90 after ozonolysis, showing bands characteristic of commercial carbon blacks. (b) Relative area comparison of Raman bands of ozonated and untreated P90 and F101.

size in water was ∼724 ( 71 nm and the zeta potential was negative at ∼ 4 ( 1.2 mV. This is indicative of an acidic surface,16 and reflected in the decrease in pH from 6.7 to 4.2 when 50 mg/mL F101 was added to ultrapure water. These samples were highly polydisperse (polydispersity index PDI ranging from 0.2 for P90 to 0.5 for F101). Though TEM indicates primary sizes of both the CB were <100 nm, the CBs are significantly agglomerated. Upon exposure to ozone, dramatic changes were observed in the surface charge and size. With increasing exposure to ozone, the zeta potential of P90 became less positive, and the agglomerate size increased and reached a peak of 1.8 ( 0.43 μm in the 4080 min exposure time. With further ozonation, the size decreased to ∼110 ( 3.5 nm and the zeta potential reached ∼ 35.3 ( 1.6 mV. Upon ozonation of F101 (Figure 2b), the zeta potential fell from ∼ 4 to ∼ 30.5 ( 0.4 mV in the ﬁrst 20 min and the average agglomerate diameter decreased from ∼724 to ∼598 ( 112 nm. The zeta potential remained nearly constant upon further ozone treatment as the average diameter of the particle agglomerates stabilized to ∼468 ( 40 nm. The ozonated F101 samples were highly polydisperse with PDI ranging from 0.4 to 0.6, whereas for ozonated P90, the PDIs ranged from 0.1 to 0.3, except for the highly agglomerated sample at 60 min, which had a PDI of 0.9 (indicating a very broad particle distribution). Raman Spectroscopy. Raman spectra of P90 and F101 were generally similar, with the spectrum for ozonated P90 shown in Figure 3a. The most prominent features of the spectrum are bands at ∼1350 and 1580 cm1, the “D1” band and “G” band, respectively. Three other bands used for curve fitting are at 1180 (D4), 1500

Figure 4. (a) Diﬀerence spectra of P90 exposed to ozone, taken at 40min intervals, starting at 40 min (front) and ﬁnishing at 240 min (back). (b) Curve ﬁtting of the infrared diﬀerence spectrum of P90 exposed to ozone for 240 min. (c) Plot of the areas of integrated bands of P90 diﬀerence spectra vs ozonolysis time.

(D3), and 1620 (D2) cm1. Raman spectra of CBs have been investigated in detail and there is agreement in band assignments.8,17,18 Disorder or defect (D) bands are at 1620 cm1 (D2), assigned to surface graphene disorder, the 1500 cm1 (D3) band assigned to carbon with functional groups, 1350 cm1 (D1) band assigned to graphene layer edges, and the 1180 cm1 (D4) band assigned to CdC vibrations of polyene-like structures. The band at 1580 cm1 (G) arises from the ideal graphitic lattice vibration. Figure 3b plots the relative intensities of these five bands for P90 and F101 before and after treatment with ozone. More marked changes were observed for P90, with significant increase in D3 due to formation of surface functional groups, at the expense of graphene bands (G and D2). Infrared Spectroscopy. The infrared spectra of P90 and F101 were similar. There are three bands: a weak band at ∼1730 cm1, and sharper bands at ∼1590 and ∼1240 cm1. The band around 10670

dx.doi.org/10.1021/es202984t |Environ. Sci. Technol. 2011, 45, 10668–10675

Environmental Science & Technology 1730 cm1 is assigned to a mix of carbonyl functionalities, and the band at 1590 cm1 corresponds to aromatic ring stretching.19,20 The band at 1240 cm1 can arise from lactones, ethers, and the symmetric bend of hydrogen atoms on adjacent double-bonded carbon atoms.5,21 To evaluate the changes taking place during ozonation, diﬀerence spectra of P90 and F101 were obtained between the ozonated samples and the starting material. The data for P90 is shown in Figure 4. Data for F101 is shown after 240 min of ozonolysis in Figure S2, and the evolution of bands as a function of reaction time with ozone is shown for P90 in Figure 4a. The changes from ozonation were more marked in P90 as compared to F101 (the band at 16001650 cm1 is ﬁve times as intense in P90 as compared to F101), and could be readily followed upon ozonation.

Figure 5. Relative area of C 1s XPS peaks before and after ozonolysis.

ARTICLE

The spectra were deconvoluted into 8 bands as shown in Figure 4b. The major bands are at 1775, 1720, 1630, 1605, 1320, and 1288 cm1. The band at 1775 cm1 has been assigned to carbonyl groups on strained functionality like lactones and acid anhydrides.19,22,23 The band centered at ∼1720 cm1 has been assigned to carboxylic acids.19,22 The bands at 1630 and 1605 cm1 represent conjugated carbonyls and the carbon backbone, respectively, with the band at 1600 cm1 being enhanced by the carboxyl group. Intensity in the ∼1600 cm1 region can also arise from CdO vibrations conjugated with large ring structures (>25). Vicinal OH groups are also reported to promote intensity of this band.19,22 Bands formed at 1320 and 1288 cm1 were assigned to CO single bonds in strained and unstrained cyclic ethers, respectively.19,20 Figure 4c is a plot of the intensity of the four major bands as a function of ozonation time. The 1630 and 1720 cm1 bands increase with similar slopes, which might be expected since they are related. The strained carbonyl groups at 1775 cm1 appear rapidly but then saturate, indicating that the sites at which initial carbonylation takes place are the most reactive. Surface Spectroscopy. X-ray photoelectron spectra (XPS) in the C 1s region were similar for P90 and F101. The changes upon ozonation for P90 are shown in Figure S3. The O 1s shows a broad peak at 532.5 eV (shown in the left inset in Figure S3). The C 1s region was deconvoluted into five peaks. They represent CC and CH bonded carbon, CO and ether carbon, carbonyl carbon, carboxyl carbon, and a πfπ* shakeup satellite,1,24,25 with binding energies of ∼284.8, 285.4, 287.5, 288.9, and 291.0 eV, respectively (seen clearly in the inset for the last four peaks in Figure S3). The πfπ* shakeup satellite peak arises from the interaction of the excited C1s electron with the π electrons of the aromatic structure. The quantitative distribution of the various functionalities is shown in Figure 5 for P90 and F101, before and after ozonation. Upon reaction with ozone, there was an increase in CdO functionalities at

Figure 6. X-Band spectra of untreated solid (a) F101and (d) P90, ozonated solid (b) F101and (e) P90, and ozonated samples exposed to PBS buﬀer (c) F101and (f) P90. 10671

dx.doi.org/10.1021/es202984t |Environ. Sci. Technol. 2011, 45, 10668–10675

Environmental Science & Technology

ARTICLE

Figure 7. X-band spectra of (a) carbon-centered adduct of DEPMPO generated from ozonated aqueous F101 (DEPMPOR, 95%, aN = 14.4, aH = 21.4, aP = 46.1, g = 2.0056); (b) hydroxyl adduct of DEPMPO generated from ozonated aqueous P90 (DEPMPOOH, aN = 14.0, aH = 13.2, aP = 47.2, g = 2.0056); (c) α-hydroxy ethyl adduct of DEPMPO generated from ozonated aqueous P90 in ethanol (DEPMPOR, 98%, aN = 14.1, aH = 20.9, aP = 46.4, g = 2.0056). (a)0 , (b)0 , and (c)0 are the corresponding simulations from which the hyperﬁne constants were obtained.

the expense of the πfπ* shakeup peak, indicating that the graphitic surface was getting oxidized. The oxygenation of the surface was also reflected in the 9-fold O:C ratio increase for P90 and 5.7-fold increase for F101. The decrease in graphitic character upon ozonation was also reflected in the ultraviolet photoelectron spectra (UPS). Figure S4 shows the UPS data with a shoulder on the C 2p band at ∼3 eV attributable to pπ character, which disappeared on ozonation. EPR Spectroscopy. The EPR spectra of P90 and F101 asobtained and after treatment with ozone are shown in Figure 6 (raw data are shown in Table S1, Supporting Information). Stable free radicals were detected on F101 with spin content of 2.8 ( 0.8 108 spins/mg (Figure 6a), whereas for P90, the signal was very weak (<2.2 ( 0.8 103 spins/mg) (Figure 6d). Upon ozonation, in both CBs, the free radical content increased, for F101 to 5.8 ( 0.6 108 spins/mg (Figure 6b), and for P90 to 4.4 ( 0.9 104 spins/mg (Figure 6e). In the case of ozonated F101, the EPR signal of the solid decreased upon exposure to PBS buffer to 1.1 ( 0.1 108 spins/mg (Figure 6c), whereas in P90, the signal increased to 1.7 ( 0.1 106 spins/mg (Figure 6f). The g values in all cases were in the range of 2.00222.0028, typical of graphitic compounds.19 To examine whether the freshly ozonated CB samples release soluble radicals upon contact with PBS buffer, spin trapping experiments were carried out, using DEPMPO as the spin trap. 26,27 As shown in Figure 7, there is a difference in the spin pattern for the radicals observed in

solution between the ozonated F101 and P90. In the case of F101, the data is best simulated by a C-centered radical adduct (aN = 14.4, aH = 21.4, aP = 46.1, g = 2.0056) and for P90 an O-centered radical adduct (aN = 14.0, aH = 13.2, aP = 47.2, g = 2.0056). Upon addition of ethanol to P90, the DEPMPO spin pattern suggests the formation of α-hydroxyethylradical (Ccentered: aN = 14.1, aH = 20.9, aP = 46.4, g = 2.0056), indicating that 3 OH radicals are being formed upon exposure of ozonated P90 to water. Figure S5 shows that the as-obtained CBs also generate radicals upon contact with PBS, but 1530 fold less than the ozonated samples (P90: 0.22 μM/mg, ozonated P90: 6.1 μM/mg; F101: 0.62 μM/mg, ozonated F101: 10.9 μM/mg, procedure described in Supporting Information along with Figure S6). Biological Activity of CB. Optical microscopy shows the ready uptake of CBs in the macrophages (a typical micrograph with F101 and P90 is shown in Figure S7). In the RPMI + FBS media used in these assays, the zeta potentials for ozonated P90 and F101 were 11.3 ( 0.2 and 12.6 ( 0.5 mV, with sizes of 218 ( 4 and 554 ( 9 nm, respectively. For the nonozonated P90 and F101 in media, zeta potentials were 11.2 ( 0.4 and 11.5 ( 0.2 mV and sizes were 262 ( 5 and 511 ( 9 nm, respectively (Figure S8 shows a characteristic light scattering and zeta potential data for F101 in media). Protein adsorption from the media is bringing about significant change in the surface charge, especially for P90, which in the as-obtained form in water exhibited a positive charge. Similar changes in charge were reported for a series 10672

dx.doi.org/10.1021/es202984t |Environ. Sci. Technol. 2011, 45, 10668–10675

Environmental Science & Technology

Figure 8. TNF-α assays for macrophages given the same treatment with P90 (top) and F101 (bottom) with and without ozonation (LPS serves as positive control).

of oxide nanoparticles suspended in cell culture media, all of which exhibited comparable negative zeta potentials.28 Cytotoxicity of the CBs toward macrophages was measured with the lactate dehydrogenase assay. The results are shown in Figure S9. The assay indicated that untreated and ozonated samples of Printex 90 and Flammruss 101 were only slightly cytotoxic. At exposures of 2.510 μg F101 or P90 per cm2 plate area, with or without treatment of particles with ozone, cell death was less than 10%. Upon interaction with macrophages, Printex 90 caused a release of 2040 pg of the inflammatory cytokine TNF-α per mL of cell media, and F101 resulted in the release of 4070 pg TNF-α per mL, as shown in Figure 8. For comparison, levels of TNF-α produced by cells in response to stimulation with LPS were 680 pg per mL. Ozone treatment of the CBs had minimal influence on the inflammatory response of exposed cells, as measured by production of TNF-α. IL-1β production by cells treated with either CB ( ozone was negligible (data not shown). In all cases, the supernatants were examined for the biological assays after the removal of the CB, and Figure S10 demonstrates that, in this procedure, the CB did not interfere with the cytotoxicity assay.

’ DISCUSSION Two facts emerge from this study: the enhanced reactivity of P90 to ozone, as indicated by Raman, IR, and XPS, and the stronger inﬂammatory response to F101 upon exposure to macrophages. These results form the basis of this discussion.

ARTICLE

Chemical Reactivity. The Raman data shown in Figure 3b suggest that the graphene layers are the primary reaction sites (decrease of intensity of the bands G and D2), resulting in increase of band D3. The decrease in intensity of the ππ* shake up peaks shown by XPS also confirms that the reaction is happening on the graphene layers, and changes in peaks in the C 1s region indicate that the reaction products are primarily oxygen functionalities (Figure 5). Though both of the CBs have graphene content, the diﬀerent behavior of P90 must stem from its more reactive graphene structures. The UPS data indicate higher intensity ππ* peaks (Figure S4) in P90. Graphene layers are known to have basic properties, and explain the positive zeta potential in as-obtained P90, and graphene has also been proposed as sites for ozone adsorption.29 TEM and the higher surface area of P90 indicate more disorder and thus possible higher reactivity (Figure 1). Infrared spectroscopy shows that the strained carbonyls on P90 have higher intensity after ﬁrst 30 min of ozonation as compared to the other carbonyl groups (Figure 4b, c), indicating that the sites responsible for the strained carbonyls are more reactive. Eventually, the production of carboxylic acid groups overtook the production of strained carbonyl species, consistent with previous chemical titration results.7 Because the oxygen functionality development upon ozonation takes place at diﬀerent rates on the P90 surface, it indicates that there is a distribution of adsorption sites.7,30 The edges of structural units might functionalize more easily than the faces, and may be responsible for the strained carbonyl groups; whereas, the conversion of the faces to carbonyl groups is slower and continues with time (Figure 4c). Development of the negatively charged functionalities upon ozonation altered the zeta potential to more negative values for F101 and caused some deagglomeration (due to electrostatic repulsion between particles) as seen in Figure 2b. For P90, increase in negatively charged functionality with ozonation balanced the positively charged functionality in the starting material leading to a more neutral particle and increase in hydrodynamic diameter. Eventually, the zeta potential became more negative with ozonation and the particle size decreased (Figure 2a). In both F101 and P90, there is an increase in free radical concentration upon ozonation, in contrast to studies on soot.31 As compared to freshly prepared soot (1015 spins/mg), the spin concentrations in F101 (∼108) and P90 (<103) are considerably smaller. With freshly prepared soot, the spin concentration was proposed to decrease upon ozonation because of spin pairing interactions with paramagnetic O2, and does not appear to be taking place with the CBs. Instead, an increase in free radical content is observed with ozonation of the CBs which is consistent with reaction of ozone with aromatic and oleﬁnic compounds. Both F101 and P90 contain graphene. These functionalities will react with ozone to generate ozonides, leading to the “Criegee” intermediate and possible radical formation.32,33 Upon reaction with water, both CBs should form hydroxyl radicals. In the case of F101, the 3 OH appear to be reacting with the CB surface and releasing C-centered radicals into solution. In the case of P90, only the 3 OH radicals are observed. It is unclear as to why there is this diﬀerence, both CBs have low levels of adsorbed PAH’s, though F101is reported to have slightly higher levels than P90 (F101 < 0.1%, P90 < 0.04%).13 Biological Properties. Neither P90 nor F101, in the parent form or in the oxidized form, elicited a strong cytotoxic response from exposed macrophages, with fewer than 10% of the cells 10673

dx.doi.org/10.1021/es202984t |Environ. Sci. Technol. 2011, 45, 10668–10675

Environmental Science & Technology dying. There is, however, an inflammatory response for both parent and oxidized particles. The lack of cytotoxic response and the mild inflammatory response are consistent with previous studies.11,34 Any dependence of the cytotoxicity or inflammatory response for P90 based on initial surface charge (positive to negative upon ozonation) could not be determined since protein adsorption (from the media) on the CB made the surface charge similar for both ozonated and nonozonated samples. F101 provokes a stronger inflammatory response from macrophages than P90. The only structural correlation we find is the EPR data (Figure 6,7), which shows higher inherent free radical contents in F101 in its native form, upon ozonation, and exposure to water, being at least 2 orders of magnitude higher compared to P90. Persistent free radical content and its influence on toxicity of airborne fine particulates have been noted, with the proposed mechanism being redox cycling of semiquinone radicals.35,36 Both F101 and P90 in their native form release free radicals upon contact with water and the ozonated samples release an order of magnitude more radicals in solution. The solids also retain free radicals with spin content on F101 > P90. The role of the solution radicals released by the CBs upon contact with water on the biological effects observed here is unclear. It is likely that the radicals formed in solution immediately react with the media, and may not make it into the macrophages. However, the solid samples, on the other hand, will carry the persistent free radicals into the cell, and the higher levels of inflammatory markers in F101 over P90 correlate with the higher spin content of F101. In conclusion, the method of manufacture has a profound inﬂuence on the CB. With the furnace black sample (P90), there is more structural disorder and a more reactive surface. Its method of manufacture involves a water quenching step, which may play a role in its surface properties. The lampblack sample (F101) formed via deposition on a cool surface has more entrapped persistent free radicals, is more ordered and less reactive.

’ ASSOCIATED CONTENT

bS

Supporting Information. Detailed structural and spectroscopic data on the CBs and details of the biological activity. This information is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Tel.: (614) 292-4532. E-mail: [email protected].

’ ACKNOWLEDGMENT We acknowledge funding from the NIOSH Grant R01 OH009141 and USDA/NIFA (2011-67021-30360). ’ REFERENCES (1) Chen, X.; Farber, M.; Gao, Y.; Kulaots, I.; Suuberg, E. M.; Hurt, R. H. Mechanisms of surfactant adsorption on non-polar, air-oxidized, and ozone-treated carbon surfaces. Carbon 2003, 41, 1489–1500. (2) Gao, Y.; K€ulaots, I; Chen, X.; Aggarwal, L; Mehta, A.; Suuberg, E. M.; Hurt, R. H. Ozonation for the chemical modiﬁcation of carbon surfaces in ﬂy ash. Fuel 2001, 80, 765–768. (3) Cataldo, F. Ozone reaction with carbon nanostructures. 2: The reaction of ozone with milled graphite and diﬀerent carbon black grades. J. Nanosci. Nanotechnol. 2007, 7, 1446–1454.

ARTICLE

lvarez, P. M.; Jaramillo, J.; Beltran, F. J. (4) Gomez-Serrano, V.; A Formation of oxygen complexes by ozonation of carbonaceous materials prepared from cherry stones. I. Thermal eﬀects. Carbon 2002, 40, 513–522. (5) Mul, G.; Neeft, J. P. A.; Kapteijn, F.; Moulijn, J. A. The formation of carbon surface oxygen complexes by oxygen and ozone. The eﬀect of transition metal oxides. Carbon 1998, 9, 1269–1276. (6) Lee, G.; Lee, B.; Kim, J.; Cho, K. Ozone adsorption on graphene: Ab initio study and experimental validation. J. Phys. Chem. C 2009, 113, 14225–14229. (7) Valdes, H.; Sanchez-Polo, M.; Rivera-Utrilla, J; Zaror, C. A. Eﬀect of ozone treatment on surface properties of activated carbon. Langmuir 2002, 18, 2111–2116. (8) Liu, Y.; Liu, C.; Ma, J.; Ma, Q.; He, H. Structural and hygroscopic changes of soot during heterogeneous reaction with O3. Phys. Chem. Chem. Phys. 2010, 12, 10896–10903. (9) Stark, W. J. Nanoparticles in Biological Systems. Angew. Chem. Int. Ed. 2011, 50, 1242–1258. (10) Grassian, V. H., Ed. Nanoscience and Nanotechnology: Environmental and Health Impacts; John Wiley & Sons: Hoboken, NJ, 2008. (11) Donaldson, K.; Born, P. J. A.; Oberdorster, G.; Pinkerton, K. E.; Stone, V.; Tran, C. L. Concordance between in vitro and in vitro delivery in the proinﬂammatory eﬀects of low-toxicity, low-solubility particles: the key role of the proximal alveolar region. Inhalat. Toxicol. 2008, 20, 53–62. (12) Heinrich, U.; Fuhst, R.; Rittinghausen, R. Chronic inhalation exposure of Wistar rats and two diﬀerent strains of mice to diesel exhaust, carbon black and titanium dioxide. Inhalat. Toxicol. 1995, 7, 533–556. (13) Pott, F.; Roller, M. Carcinogenicity Studies with nineteen granular dusts in rats. Eur. J. Oncol. 2005, 10, 249–281. (14) Peebles, B. C.; Dutta, P. K.; Nagy, A.; Waldman, W. J. Fenton activity and cytotoxicity studies of iron-loaded carbon particles. Environ. Sci. Technol. 2010, 44, 6887–6892. (15) Donnet, J.-B.; Custodero, E. La microscopie a eﬀet tunnel des poudres, application aux noirs de carbone. C. R. Acad. Sci., Ser. II 1992, 314 (6), 579–584. (16) Xu, R.; Wu, C.; Xu, H. Particle size and zeta potential of carbon black in liquid media. Carbon 2007, 45, 2806–2809. (17) Ferrari, A. C.; Robertson, J. Interpretation of Raman spectra of disordered and amorphous carbon. Phys. Rev. B 2000, 61 (20), 14095–14107. (18) Sadezky, A.; Muckenhuber, H.; Grothe, H.; Neissner, R.; P€oschl, U. Raman microscopy of soot and related carbonaceous materials: Spectral analysis and spectral information. Carbon 2005, 43, 1731–1742. (19) Akhter, M. S.; Chughtai, A. R.; Smith, D. M. The structure of hexane soot I: Spectroscopic studies. Appl. Spectrosc. 1985, 39 (1), 143–153. (20) Fanning, P. E.; Vannice, M.; Albert, A. DRIFTS study of the formation of surface groups on carbon by oxidation. Carbon 1993, 31 (5), 721–730. (21) Fuente, E.; Menendez, J. A.; Díez, M. A.; Suarez, D.; MontesMoran, M. A. Infrared spectroscopy of carbon materials: A quantum chemical study of model compounds. J. Phys. Chem. B 2003, 107, 6350–6359. (22) Prest, W. M., Jr.; Mosher, R. A. Fourier transform IR spectroscopic characterization of the functional groups on carbon black. ACS Symp. Ser. 1982, 200, 225–247. (23) Mawhinney, D. B.; Yates, J. T. FTIR study of the oxidation of amorphous carbon by ozone at 300K direct COOH formation. Carbon 2001, 39, 1167–1173. (24) Huang, C. D.; Xiong, Z. T.; Lin, J. Y.; Tan, K. L. Investigation of the electrochemical behavior of heat-treated carbon blacks and Pt/C catalysts. Surf. Rev. Lett. 2002, 9 (3,4), 1443–1452. (25) Lee, W. H.; Kim, J. Y.; Ko, Y. K.; Reucroft, P. J.; Zondlo, J. W. Surface analysis of carbon black waste materials from tire residues. Appl. Surf. Sci. 1999, 141, 107–113. 10674

dx.doi.org/10.1021/es202984t |Environ. Sci. Technol. 2011, 45, 10668–10675

Environmental Science & Technology

ARTICLE

(26) Frejaville, C.; Karoui, H.; Tuccio, B.; Moigne, F. L.; Culcasi, M.; Pietri, S.; Lauricella, R.; Tordo, P. 5-(Diethoxyphosphoryl)-5-methyl-1pyrroline N-oxide: A New Eﬃcient Phosphorylated Nitrone for the in Vitro and in Vivo Spin Trapping of Oxygen-Centered Radicals. J. Med. Chem. 1995, 38 (2), 258–265. (27) Villamena, F.; Hadad, C. M.; Zweier, J. L. Kinetic Study and Theoretical Analysis of Hydroxyl Radical Trapping and Spin Adduct Decay of Alkoxycarbonyl and Dialkoxycarbonyl Nitrones in Aqueous Media. J. Phys. Chem. A 2003, 107, 4407–4414. (28) Limbach, L. K.; Yuchun, L.; Grass, R. N.; Brunner, T. J.; Hintermann, M. A.; Gunther, D.; Stark, W. J. Oxide nanoparticle uptake in human lung ﬁbroblasts: Eﬀects of particle size, agglomeration and diﬀusion at low concentrations. Environ. Sci. Technol. 2005, 39, 9370–9376. lvarez, P. M.; Masa, F. J.; Jaramillo, J.; Beltran, F. J.; G (29) A omezSerrano, V. Kinetics of ozone decomposition on activated carbon. Ind. Eng. Chem. Res. 2008, 47, 2545–2553. (30) Schr€oder, A.; Kl€uppel, M.; Schuster, R. H.; Heidberg, J. Surface energy distribution of carbon black measured by static gas adsorption. Carbon 2002, 40, 207–210. (31) Chugtai, A. R.; Atteya, M. M. O.; Kim, J.; Konowalchuk, B. K.; Smith, D. M. Adsorption and adsorbate interaction at soot particle surfaces. Carbon 1998, 36, 1573–1589. (32) Alam, M. S.; Camredon, M.; Rickard, A. R.; Carr, T.; Wyche, K. P.; Hornsby, K. E.; Monks, P. S.; Bloss, W. J. Total radical yields from tropospheric ethane ozonolysis. Phys. Chem. Chem. Phys. 2011, 13, 11002–11015. (33) Mysak, E. R.; Smith, J. D.; Ashby, P. D.; Newberg, J. T.; Wilson, K. R.; Bluhm, H. Competitive reaction pathways for functionalization and volatilization in the heterogeneous oxidation of coronene thin ﬁlms by hydroxyl radical and ozone. Phys. Chem. Chem. Phys. 2011, 13, 7554–7564. (34) Donaldson, K.; Brown, D.; Clouter, A.; Duﬃn, R.; MacNee, W.; Renwick, L.; Tran, L.; Stone, V. The pulmonary toxicology of ultraﬁne particles. J. Aerosol Med. 2002, 15, 213–220. (35) Dellinger, B.; Pryor, W. A.; Cueto, R.; Squadrito, G. L.; Hegde, V.; Deutsch, W. A. Role of free radicals in the toxicity of airborne ﬁne particulate matter. Chem. Res. Toxicol. 2001, 14, 1371–1377. (36) Khachatryan, L.; Vejerano, E.; Lomnicki, S.; Dellinger, B. Environ. Sci. Technol. 2011, 45, 8559–8566.

10675

dx.doi.org/10.1021/es202984t |Environ. Sci. Technol. 2011, 45, 10668–10675

ARTICLE pubs.acs.org/est

Predicting Arsenic Relative Bioavailability in Contaminated Soils Using Meta Analysis and Relative Bioavailability Bioaccessibility Regression Models Albert L. Juhasz,* John Weber, and Euan Smith Centre for Environmental Risk Assessment and Remediation, University of South Australia, Mawson Lakes, South Australia 5095, Australia

bS Supporting Information ABSTRACT: A number of in vitro assays are available for the determination of arsenic (As) bioaccessibility and prediction of As relative bioavailability (RBA) to quantify exposure for sitespeciﬁc risk assessment. These data are usually considered in isolation; however, meta analysis may provide predictive capabilities for source-speciﬁc As bioaccessibility and RBA. The objectives of this study were to predict As RBA using previously published in vivo/in vitro correlations and to assess the inﬂuence of As sources on As RBA independent of geographical location. Data representing 351 soils (classiﬁed based on As source) and 514 independent bioaccessibility values were retrieved from the literature for comparison. Arsenic RBA was predicted using published in vivo/in vitro regression models, and 90th and 95th percentiles were determined for each As source classiﬁcation and in vitro methodology. Diﬀerences in predicted mean As RBA were observed among soils contaminated from diﬀerent As sources and within source materials when various in vitro methodologies were utilized. However, when in vitro data were standardized by transforming SBRC intestinal, IVG, and PBET data to SBRC gastric phase values (through linear regression models), predicted As RBA values for As sources followed the order CCA posts g herbicide/pesticide > mining/smelting > gossan soils with 95th percentiles for predicted As RBA of 78.0, 78.4, 67.0, and 23.7%, respectively.

’ INTRODUCTION Arsenic (As) occurs naturally in soil and is the 20th most abundant element in the earth’s crust. However, enrichment of soil As may occur as a result of anthropogenic processes, including (but not limited to) pesticide/herbicide manufacture and use, mining, smelting, and wood preservation. Where anthropogenic inputs have occurred, elevated As concentrations in soils may exceed regulatory guidelines, with potential impacts on human and environmental health.1 4 Contamination of soil with As was ranked the most common inorganic constituent in the National Priority List of Sites in the United States.5 Because of the eﬀects associated with As exposure, such as the development of numerous health disorders in addition to carcinogenesis,6,7 there is concern regarding the potential risk associated with soilborne As to human and environmental health. To deﬁne remediation goals for As-contaminated sites, sitespeciﬁc data are required to ensure an accurate assessment of potential risk at the site. Site-speciﬁc data are also warranted to reﬁne default risk variables that, as a result of their conservative nature, may result in unnecessarily low remediation goals, use of additional remediation resources, and deliver unwarranted remediation costs. A parameter that may be utilized to reﬁne site-speciﬁc r 2011 American Chemical Society

remediation goals is As relative bioavailability (RBA). Arsenic RBA is a measure of the amount of As that is absorbed into systemic circulation (with comparison to a reference dose) as a result of contaminated soil exposure. It is dependent on As mineralogy, the inﬂuence of soil properties (e.g., iron [Fe] content), and the residence time of As in the soil.8 As a result of these factors, As RBA may be less than the conservative assumption of 100%, and this may have a signiﬁcant inﬂuence on human exposure and risk assessment. Adjustments to As RBA may be achieved by conducting in vivo bioavailability or in vitro bioaccessibility studies. Currently, in vivo assays (e.g., swine) are the method of choice; however, these assays are complicated, expensive, and time-consuming.9 Because of their simplicity, speed, and aﬀordability, in vitro assays (measuring bioaccessibility) that simulate conditions in the gastrointestinal tract may be an attractive alternative for predicting As RBA.10 However, for in vitro assays to be used as Received: May 29, 2011 Accepted: November 7, 2011 Revised: November 1, 2011 Published: November 07, 2011 10676

dx.doi.org/10.1021/es2018384 | Environ. Sci. Technol. 2011, 45, 10676–10683

Environmental Science & Technology

ARTICLE

Table 1. Linear Regression Models for Predicting in Vivo As RBA in Contaminated Soils Using SBRC, IVG, and PBET in Vitro Assaysa in vitro assay SBRC

phase gastric

in vivo-in vitro predictive model in vivo relative As bioavailability (%) = 1.656 + (0.992) SBRC-gastric (%)

Pearson correlation 0.868

R2 = 0.754 intestinal

in vivo relative As bioavailability (%) = 5.626 + (1.644) SBRC-intestinal (%)

0.809

R2 = 0.654 IVG

gastric

in vivo relative As bioavailability (%) = 14.323 + (0.853) IVG-gastric (%)

0.757

R2 = 0.573

PBET

intestinal

in vivo relative As bioavailability (%) = 13.971 + (1.105) IVG-intestinal (%)

0.753

gastric

R2 = 0.567 in vivo relative As bioavailability (%) = 10.096 + (1.162) PBET-gastric (%)

0.799

R2 = 0.638 intestinal

in vivo relative As bioavailability (%) = 5.682 + (1.762) PBET-intestinal (%)

0.816

R2 = 0.665 a

Modiﬁed from Juhasz et al.13

a surrogate measurement of contaminant RBA, the correlation between bioaccessibility and RBA is a mandatory prerequisite for scientiﬁc as well as regulatory acceptance. A limited number of studies have established the relationship between in vivo As RBA and in vitro As bioaccessibility.11 14 Recently, Juhasz et al. 13 determined As bioaccessibility in contaminated soils (n = 12) using four commonly utilized in vitro assays (Solubility Bioaccessibility Research Consortium assay [SBRC], in vitro gastrointestinal extraction method [IVG], physiologically based extraction test [PBET], and German standardized in vitro assay [DIN]). In vitro results were compared to in vivo As RBA data (swine assay) to ascertain which methodologies best correlate with in vivo data. Comparison of in vitro and in vivo results demonstrated that the in vitro assay encompassing the SBRC gastric phase (SBRC-G) provided the best prediction of in vivo As RBA (R2 = 0.75, Pearson correlation = 0.87). However, As RBA could also be predicted using gastric or intestinal phases of IVG, PBET, and DIN assays with varying degrees of conﬁdence (R2 = 0.53 0.67, Pearson correlation = 0.73 0.82). Although in vivo As RBA studies are limited, a considerable amount of in vitro data has been published reporting As bioaccessibility in contaminated soils from varying geographical regions and As sources.10,11,14 25 Although in vivo/in vitro correlations are not available for some of these studies, it may be possible to utilize other published in vivo/in vitro and in vitro/ in vitro relationships to assess the variability in predicted As RBA for contaminated soils with diﬀerent As sources. Consequently, the objectives of this study were to retrieve As bioaccessibility data from the literature, predict As RBA using previously published in vivo/in vitro correlations, and assess the inﬂuence of As sources on As RBA independent of geographical location.

’ MATERIALS AND METHODS Data for As Bioaccessibility Meta Analysis. Arsenic bioaccessibility data was retrieved from the literature following key word searches in ISI Web of Knowledge (to March 2010), in addition to data from research and consultancy reports.10,11,14 25 Only data utilizing SBRC, IVG, and PBET methodologies were retrieved because of the availability of As RBA bioaccessibility regression models utilizing these in vitro assays (from Juhasz et al.13). Although Juhasz et al.13 developed an As RBA bioaccessibility regression model for the DIN assay, limited in

vitro data were present in the literature utilizing this methodology, and therefore, they were not utilized for meta analysis. Available data were sorted into four categories (herbicide/ pesticide, mining/smelting, copper chrome arsenate (CCA)treated posts, geogenic) on the basis of the source of the As contamination. These source categories are representative of the common sources of As soil contamination reported in the literature. Arsenic Relative Bioavailability Bioaccessibility Regression Models. To transform bioaccessibility data from the literature into As RBA values, As RBA bioaccessibility linear regression models, developed by Juhasz et al.13 were utilized (Table 1). These models were developed following the assessment of As RBA in 12 As-contaminated soils using an in vivo swine assay with companion bioaccessibility analysis using gastric and intestinal phases of the SBRC, IVG, and PBET in vitro assays. Although other in vivo/in vitro data is available in the literature,11,14,22 this information was not included in the development of linear regression models because of the lack of data that demonstrate that As RBA is congruent across rabbit,22 monkey,22 and swine models;11,13,14 the difference in As RBA end points for swine data (urine11,14 versus blood analysis13); and the difference in swine dosing matrix (dough11,14 versus gavage13) and length of dosing. After consideration of the above, only the predictive models of Juhasz et al.13 were utilized, although it is conceded that any model chosen may introduce some experimental bias into the analysis. The relationship between in vitro bioaccessibility methods was also determined from data presented in Juhasz et al.13 to transform As bioaccessibility values between methodologies. Because SBRC-G provided the best estimate for As RBA (see Table 1), linear regression models were developed for in vitro assays in order to predict SBRC-G As bioaccessibility from SBRC intestinal (SBRC-I), IVG, and PBET (gastric and intestinal) data (Table 2 and Supporting Information Figure S1). Incorporation of SBRC-G transformed data into the in vivo-SBRC-G linear regression model may propagate additional uncertainty, although with the application of two successive regression equations it provided a comparison of predicted As RBA for diﬀerent As sources irrespective of bioaccessibility methodology. Statistical Analysis. Linear regression models were developed using SPSS 16.0.1 (2007). Comparison of As bioaccessibility and RBA means were determined using unpaired t tests or one-way 10677

dx.doi.org/10.1021/es2018384 |Environ. Sci. Technol. 2011, 45, 10676–10683

Environmental Science & Technology

ARTICLE

Table 2. Linear Regression Models for Predicting SBRC-G As Bioaccessibility from SBRC-I, IVG and PBET (gastric and intestinal phase) Data in vitro assay

phase

SBRC

intestinal

As bioaccessibility (SBRC gastric; %) = (1.670) SBRC-intestinal (%) + 3.771

0.939

gastric

R2 = 0.882 As bioaccessibility (SBRC gastric; %) = (0.837) IVG-gastric (%) + 13.272

0.848

IVG

in vivo/in vitro predictive model

Pearson correlation

R2 = 0.720 intestinal

As bioaccessibility (SBRC gastric; %) = (1.102) IVG-intestinal (%) + 12.610

0.858

R2 = 0.737 PBET

gastric

As bioaccessibility (SBRC gastric; %) = (1.137) PBET-gastric (%) + 9.172

0.894

R2 = 0.800 intestinal

As bioaccessibility (SBRC gastric; %) = (1.801) PBET-intestinal (%) + 3.660

0.953

R2 = 0.908

Table 3. Summary of As Concentrations in Soil Used in the As RBA Bioaccessibility Meta Analysis As (mg kg 1)a no. of As source herbicide and pesticide

25 025

1124

32

23

319

164

163

mining and smelting geogenic

164 40

24 13

68 924 422

4771 87

911 52

all soils

351

68 924

2663

327

CCA posts

a

samples minimum maximum mean median 115

8.4

8.4

270

Concentration of As in the <250 μm soil particle size fraction.

ANOVA. The 90th and 95th percentiles for As RBA were generated in GraphPad Prism (version 5.0).

’ RESULTS AND DISCUSSION Arsenic Contaminated Soil. Table 3 provides a summary of As contaminated soils used in the meta analysis. The 351 soils were divided into four categories on the basis of As source: As contamination arising from herbicide/pesticide application, use of CCA treated timber, mining/smelting activities, and naturally occurring elevated concentrations of As (geogenic). Minimum, maximum, mean, and median As concentrations are presented for each source category, in addition, for all soils (Table 3). Soil As concentrations ranged from 8.4 (herbicide/pesticide impacted) to 68 924 mg kg 1 (mining/smelting impacted), with mean and median As concentrations of 2663 and 327 mg kg 1, respectively. Arsenic Bioaccessibility. As highlighted in Table S1 and Figure S2 (Supporting Information), As bioaccessibility in the 351 soils varied considerably and was dependent on the As source and the methodology used for its assessment, including the in vitro phase. When SBRC-G was used to assess herbicide and pesticide impacted soils, As bioaccessibility ranged from <1 to 89.0%, with a mean of 15.8% (n = 103). In a much smaller data set (n = 12) assessed using IVG gastric (IVG-G) and intestinal (IVG-I) phases, the mean As bioaccessibility was significantly different from SBRC-G values (p < 0.01); however, there was no significant difference between As bioaccessibility when IVG-G or IVG-I were utilized (p = 0.83; Figure 1). Only the IVG methodology was utilized for the assessment of As bioaccessibility in soils surrounding CCA-treated posts (n = 32). Arsenic

bioaccessibility ranged from 15.8 to 63.6% and from 17.0 to 66.3% following assessment using IVG-G and IVG-I, respectively, with no significant difference in the means for each in vitro phase (p = 0.23; Figure 1). Mining- and smelting-impacted soils have attracted considerable attention in terms of As bioaccessibility research. Both gastric and intestinal phases of SBRC, IVG, and PBET methods have been utilized for assessing As bioaccessibility in contaminated soils ranging in concentration from 24 to 68 924 mg As kg 1. Arsenic bioaccessibility ranged from 0.1 to 66.0% with mean values ranging from 7.9% (n = 105; PBET gastric [PBET-G] phase) to 23.1% (n = 6; SBRC-I). Signiﬁcant diﬀerences between mean As bioaccessibility were observed between SBRC-G and IVG-I (p < 0.05), SBRC-G and PBET-G (p < 0.001), SBRC-I and PBET-G (p < 0.05), and PBET-I and PBET intestinal (PBET-I) phases (p < 0.001) (Figure 1). In gossan soils (n = 40), As bioaccessibility ranged from 1.0 to 21.0% with mean values <7.0%, irrespective of sample location and methodology utilized for its assessment (Figure 1). For SBRC and PBET (gastric and intestinal phases) methodologies, there was no signiﬁcant diﬀerence (p = 0.67) between mean As bioaccessibility for gossan soils. Due to the inﬂuence of soil properties and aging eﬀects as a result of the dominance of surface adsorbed As (due to leaching or surface application), As bioaccessibility ranged considerably for CCA- and herbicide/pesticide-impacted soils. A number of researchers have identiﬁed the importance of soil Fe in controlling As sorption and bioaccessibility.10,26 28 In spiked soil studies conducted by Yang et al.,28 ∼75% of the variability in arsenate bioaccessibility was attributable to the soil’s pH and Fe oxide content. Similarly, Juhasz et al.10 showed that total As, Fe, and free Fe (dithionite citrate bicarbonate-extractable Fe) were variables best able to describe As bioaccessibility in herbicide/ pesticide-impacted soils. Iron is an important variable controlling As bioaccessibility due to its marked inﬂuence on As sorption in pesticide/herbicide-contaminated soils.29,30 In addition, Fendorf et al.31 suggested that contaminant aging decreases bioaccessibility because of changes in surface phase complexes. A decrease in As bioaccessibility in CCA- and herbicide/pesticide-impacted soils may result from the development of inner sphere complexes, surface diﬀusion within micropores, or surface precipitates32 with increasing As-soil residence time. In mining and smelting-impacted soils and gossans, As bioaccessibility is inﬂuenced by mineralogical composition. Studies undertaken with pure minerals and impacted soils have shown decreasing As bioaccessibility with As sulﬁdes (e.g., arsenopyrite, 10678

dx.doi.org/10.1021/es2018384 |Environ. Sci. Technol. 2011, 45, 10676–10683

Environmental Science & Technology

ARTICLE

Figure 1. Box plot of As bioaccessibility determined using gastric and intestinal phases of SBRC, IVG, and PBET in vitro methods. Each box represents the lower and upper quartiles, and the band within the box represents the median As bioaccessibility value for the respective in vitro assay. Whiskers represent minimum and maximum values for data determined using each in vitro assay. Panels represent all soils used in the meta analysis (A), herbicide/ pesticide (B), mining/smelting (C), and gossan (0) and CCA (9)-impacted soils (D).

realgar), iron arsenates (e.g., scorodite, kankite, pharmacosiderite), As-bearing Fe oxyhydroxides (e.g., goethite, lepidocrocite, akaganeite, roaster), iron oxides (e.g., hematite, maghemite), As sulfates (e.g., tooeleite, jarosite, schwertmannite), clay minerals, and calcium iron arsenates (e.g., yukonite).33 Higher As bioaccessibility may be associated with calcinated samples in which As is associated with the more amorphous/less dense/more weathered Fe oxides.3 Unlike discrete mineral phases, these weathered As Fe oxide associations may undergo dissolution when exposed to pH conditions encountered during gastric phase extraction. Arsenic bioaccessibility may be <1% when As sulphides are the predominant As mineral phase.8,33 In addition, As bioaccessibility may be inﬂuenced by the encapsulation or coating of grains due to a reduction in the dissolution of surfacebound As or the exterior portion of As-bearing grains.8 Arsenic Relative Bioavailability. When in vivo/in vitro regression equations developed by Juhasz et al.13 were utilized to convert in vitro data from 351 soils (Supporting Information Table S1) into As RBA data (Supporting Information Figure S3 and Table S2), predicted values were higher than the corresponding As bioaccessibility values. This was particularly evident for soils assessed using SBRC-I, IVG and PBET methodologies because of the poorer fit of the in vivo/in vitro regression equations (i.e., the slope being >1 or due to large y intercepts) (Table 1). In some cases, predicted As RBA values were in excess of 100% because not all the variability was accounted for in the in vivo/in vitro linear regression

models. However, as illustrated in Figure 2, some differences in predicted mean As RBA were observed between soils contaminated from different As sourced and within source materials when different in vitro methodologies were utilized. For herbicide/pesticide-impacted soils assessed using the SBRC-G assay, the predicted mean As RBA was signiﬁcantly lower (p < 0.002) compared with soils assessed using the IVG assay in which there was no signiﬁcant diﬀerence in means when gastric or intestinal phase predictions were undertaken (p = 0.33). Similarly, for mining/smelting impacted soils, there was no signiﬁcant diﬀerence (p = 0.67) in predicted mean As RBA when IVG-G and IVG-I were utilized. However, diﬀerences in SBRC and PBET predicted mean As RBA's were signiﬁcant (p < 0.005) when gastric and intestinal phases of the respective assays were compared. In contrast, there was no signiﬁcant diﬀerence (p = 0.63) between mean As RBA when either phase of the PBET assay was utilized for gossan soil predictions, although predicted mean As RBA was signiﬁcantly lower (p < 0.005) when assessed using SBRC-G. For soils contaminated as a result of CCA post usage, As RBA was signiﬁcantly higher (p < 0.005) when predictions utilized IVG-I data compared with IVG-G. Table 4 shows the predicted As RBA 90th and 95th percentiles for herbicide/pesticide, mining/smelting, gossan, and CCAimpacted soils determined using in vivo/in vitro linear regression models outlined in Table 1 and the corresponding SBRC, IVG, or PBET in vitro data. For all in vitro assay predictions, As RBA 10679

dx.doi.org/10.1021/es2018384 |Environ. Sci. Technol. 2011, 45, 10676–10683

Environmental Science & Technology

ARTICLE

Figure 2. Box plot of predicted As RBA determined using gastric and intestinal phase data form SBRC, IVG, and PBET in vitro assays and in vivo/ in vitro regression models of Juhasz et al.13 Each box represents the lower and upper quartiles, and the band within the box represents the median predicted As relative bioavailability value for the respective in vitro assay. Whiskers represent the 95th percentile for predicted As RBA determined using each in vitro assay. Data above and below these percentiles are represented by a circle (b). Panels represent all soils used in the meta analysis (A), herbicide/pesticide (B), mining/smelting (C), and gossan (0) and CCA (9)-impacted soils (D).

90th and 95th percentiles were higher when intestinal phase data was utilized; however, the diﬀerence between calculated values using gastric and intestinal phase data was variable and was not speciﬁc to individual in vitro assays or As sources (Table 4). For example, when PBET data was utilized to calculate the 95th percentile for As RBA in gossan and mining/smelting-impacted soils, predicted values using gastric and intestinal phase data were similar for gossan soils, but values determined using intestinal phase data were 2.1-fold higher for mining/smelting soils compared with gastric phase predicted values. In contrast, predicted As RBA 95th percentile values for mining- and smelting-impacted soils were similar when calculated using IVG gastric and intestinal phase data; however, values generated using intestinal phase data were ∼1.3-fold greater for CCA-impacted soils. For mining and smelting impacted soils, however, 95th percentile values for predicted As RBA were comparable across SBRC, IVG, and PBET assays for values calculated using gastric phase data (42.9, 39.6, and 40.8%, respectively) (Table 4). Standardization of in Vitro Data and Predicted As Relative Bioavailability. In the study of Juhasz et al.,13 the in vitro method

incorporating SBRC-G was shown to be the best predictor of in vivo As RBA. To “standardize” As bioaccessibility data from the literature, in vitro data generated from SBRC-I, IVG, and PBET (gastric and intestinal phases) extractions were transformed to SBRC-G values using linear regression models outlined in Table 2. Linear regression models were developed using bioaccessibility data generated for the 12 As-contaminated soils in Juhasz et al.13 For arsenic bioaccessibility for the 12 soils determined using SBRC-G yielded values that were 1.16 1.98 times higher than those generated using other in vitro methods or phases, however, in vitro/in vitro relationships were linear with variances of 0.720 0.908 compared with the model equations (Table 2). When As bioaccessibility data was transformed to SBRC-G values, < 2% of data produced values in excess of 100% because not all the variability was accounted for in the model equations. For example, the maximum IVG-I bioaccessibility value for herbicide/pesticide-impacted soils (90.0%) produced a transformed SBRC-G bioaccessibility value of 111.8%. Although As 10680

dx.doi.org/10.1021/es2018384 |Environ. Sci. Technol. 2011, 45, 10676–10683

Environmental Science & Technology

ARTICLE

Table 4. Predicted As RBA for Herbicide/Pesticide, Mining/Smelting, Gossan and CCA-Impacted Soils Determined Using Linear Regression Models Outlined in Table 1 and SBRC, IVG, and PBET in Vitro Dataa predicted as relative bioavailability (%) SBRC assay As source herbicide/pesticide CCA posts mining/smelting gossan all soils

percentile

SBRC-G

SBRC-I

IVG assay IVG-G

PBET assay

IVG-I

PBET-G

PBET-I

SBRC-G standardization

90th

39.9

75.9

110.0

50.6

95th

46.9

79.2

113.4

78.4

62.3 68.6

79.3 87.2

71.4 78.0

90th 95th 90th

38.0

64.2

39.2

39.5

29.2

73.1

52.1

95th

42.9

64.2

39.6

39.9

40.8

87.2

67.0

90th

21.9

22.9

23.7

23.0

95th

23.5

22.9

23.9

23.7

90th

38.6

64.2

66.4

83.1

29.2

73.1

59.8

95th

44.3

64.2

68.6

91.0

40.8

87.2

70.7

a

90th and 95th percentiles are shown for As source and in vitro assay. 90th and 95th percentiles are also shown for each As source following standardization of IVG and PBET data to SBRC-G values (according to Table 2) and recalculation of As RBA using the in vivo SBRC-G linear regression model.

bioaccessibility cannot exceed 100%, these values were retained (see Supporting Information Table S1) to illustrate the potential limitations of the model equations. The inclusion of data in excess of 100% had a modest impact on mean As bioaccessibility calculations, with little impact on 90th and 95th percentile values. Table S1 (Supporting Information) provides an overview of transformed SBRC-G As bioaccessibility values for gossan soils in addition to herbicide/pesticide, CCA post, and mining/smelting impacted soils. Gossan soils (n = 40) yielded the lowest mean As bioaccessibility value (12.9 ( 6.2%), with an upper 95% conﬁdence interval (CI) of the mean of 14.9%. In contrast, CCAimpacted soils yielded the highest As bioaccessibility value of 48.2%, with a standard deviation of 15.1% (upper 95% CI of the mean of 51.9%). For herbicide/pesticide- and mining/meltingimpacted soils, there was no signiﬁcant diﬀerence (p > 0.05) between mean As bioaccessibility values: 21.9 ( 22.8% (upper 95% CI of the mean of 25.9%) and 26.3 ( 18.4% (upper 95% CI of the mean of 28.5%), respectively, of the SBRC-G transformed As bioaccessibility data. Figure 3 and Table S2 (Supporting Information) show predicted As RBA values for contaminated soils calculated using SBRC-G standardized in vitro data (from Table S2 of the Supporting Information) and the in vivo SBRC-G linear regression model (Table 1). Mean As RBA ranged from 14.4 ( 6.2% (gossan soils) to 49.4 ( 15.0% (CCA-impacted soils) but there was no signiﬁcant diﬀerence (p = 0.59) between the means for herbicide/pesticide- and mining/smelting-impacted soils (∼25%). The 90th and 95th percentiles for As RBA, calculated using SBRC-G standardized in vitro data, are presented in Table 4. Predicted As RBA percentiles for As sources followed the order CCA posts g herbicide/pesticide > mining/smelting > gossan. However, for the majority of samples included in the study, the As source was not a good predictor of As RBA because of the wide range of values reported. For example, the relative standard deviation of predicted As RBA for herbicide/pesticide and mining/smelting sources (representing ∼80% of the data) were 104 and 70%, respectively. In contrast, predicted As RBA for gossan soils was typically low because of the inﬂuence of As mineralogy,

Figure 3. Box plot of predicted As RBA for each As source following conversion of SBRC-I, IVG and PBET data to SBRC-G values (according to Table 4) and calculation of As RBA using the in vivo SBRC-G linear regression model. Each box represents the lower and upper quartiles, and the band within the box represents the median predicted As RBA value. Whiskers represent the 95th percentile for predicted As RBA, and data above and below these percentiles are represented by a circle (b).

which controls As solubility in the gastric phase. In soils contaminated as a result of As pesticide/herbicide application or CCA post usage, As was mainly associated with amorphous Fe oxyhydroxide phases1,34 which are more soluble in the gastric phase than As associated with mineral phases. Solubility in the gastric phase primarily determines the free ion concentration potentially available for absorption in the intestinal phase,35 which was highlighted by the fact that predicted As RBA in gossan soils was low compared with the 90th and 95th percentiles for pesticide/herbicide, CCA post and mining/smelting impacted soils (Table 4). 10681

dx.doi.org/10.1021/es2018384 |Environ. Sci. Technol. 2011, 45, 10676–10683

Environmental Science & Technology When SBRC-G standardized predicted As RBA 95th percentiles were compared with values generated using in vitro-speciﬁc in vivo/in vitro regression models, similar values were observed for gossan soils. The 95th percentile for all SBRC-G standardized predictions was 23.7%, compared with 23.5, 22.9, and 23.9% for SRBC-G only, PBET-G, and PBET-I data, respectively. However, for herbicide/pesticide, CCA post, and mining/smelting impacted soils, standardized predicted 95th percentiles were generally more conservative than values generated using individual gastric phase data or less conservative than some individual intestinal phase data. The diﬀerence in predicted As RBA 95th percentiles (standardized versus in vitro method speciﬁc) is reﬂective of the model ﬁt associated with in vivo/in vitro or in vitro/in vitro regressions. However, 95th percentiles for standardized predicted As RBA were 78.0, 78.4, 67.0, and 23.7% for CCA posts, herbicide/pesticide, mining/smelting, and gossan soils, respectively. On the basis of the meta analysis, the As source may not provide useful predictions of As bioaccessibility or As RBA. Although predicted As RBA values (90th or 95th percentiles) for speciﬁc As sources could be used as a ﬁrst screen to determine the impact of this parameter on human health risk assessment, for the majority of As sources, this may provide minimal impact when reﬁning human health exposure. However, these values may provide further justiﬁcation for additional site investigation to determine site-speciﬁc As bioaccessibility. In addition, for a number of sites, the source of As contamination may be unknown, or contamination may have arisen from multiple sources. In these situations, the assessment of site-speciﬁc As bioaccessibility is recommended. Another factor that needs to be considered when predicting As RBA values (90th or 95th percentiles) for speciﬁc As sources is the inﬂuence of analytical methodologies on the determination of total As concentration, measurement and calculation of As bioaccessibility and their use for the prediction of As RBA. For example, the determination of total As concentration in soil may vary depending on the digestion matrix (e.g aqua-regia versus concentrated nitric acid versus hydroﬂuoric acid) and procedure used.36 Arsenic bioaccessibility may also vary for a given in vitro methodology, depending on the laboratory where the analysis is undertaken. Round robin studies have identiﬁed considerable interlaboratory variability for As bioaccessibility with relative standard deviations ranging up to 46% (Koch Pers. comms.). These analytical factors will introduce variability when As bioaccessibility is determined and during As RBA predictions from data sourced from the literature. In addition, uncertainty in model predictions may be improved through the inclusion of additional As bioaccessibility data sets and by improvements in As in vivo/in vitro correlations; a worldwide bioavailability bioaccessibility research priority.

’ ASSOCIATED CONTENT

bS

Supporting Information. Graphs and tables showing the relationship between As bioaccessibility determined using SBRC-G and other in vitro methods (Figure S1); As bioaccessibility in herbicide/pesticide, CCA, mining/smelting, and gossan soils determined using gastric and intestinal phases of the SBRC, IVG, and PBET assays (Figure S2, Table S1); and predicted As RBA in herbicide/pesticide, CCA, mining/ smelting, and gossan soils determined using gastric and intestinal phase in vitro data and bioaccessibility-RBA regression models

ARTICLE

(Figure S3, Table S2). This information is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Address: Centre for Enironmental Risk Assessment and Remediation (CERAR), University of South Australia, Building X1-17, Mawson Lakes Campus, Adelaide, SA 5095, Australia. Phone: (61 8) 8302 5045; fax: (61 8) 8302 3057; e-mail: Albert. [email protected]

’ ACKNOWLEDGMENT The authors acknowledge the support of the Centre for Environmental Risk Assessment and Remediation, University of South Australia for this research. ’ REFERENCES (1) Smith, E.; Smith, J.; Smith, L.; Biswas, T.; Correll, R.; Naidu, R. Arsenic in Australian environment: an overview. J. Environ. Sci. Health, Part A 2003, 38, 223–239. (2) Townsend, T.; Solo-Gabriele, H.; Tolaymat, T.; Stook, K.; Hosein, N. Chromium, copper, and arsenic concentrations in soil underneath CCA-treated wood structures. Soil Sed. Contam. 2003, 12, 779–798. (3) Smith, E.; Juhasz, A.; Naidu, R. Human availability of arsenic at mining waste areas in central Victoria. Rep. Victoria EPA, 2002. (4) Wang, S.; Mulligan, C. N. Occurrence of arsenic contamination in Canada: sources, behaviour and distribution. Sci. Total Environ. 2006, 366, 701–721. (5) Agency for Toxic Substances and Disease Registry. Priority list of hazardous substances 2007, http://www.atsdr.cdc.gov/cercla/07list. html. (6) Lien, H. C.; Tsai, T. F.; Lee, Y. Y.; Hsiao, C. H. Merkel cell carcinoma and chronic arsenicism. J. Am. Acad. Dermatol. 2001, 41, 641–643. (7) Mandal, B. K.; Suzuki, K. T. Arsenic round the world: a review. Talanta 2002, 58, 201–235. (8) Ruby, M. V.; Schoof, R.; Brattin, W.; Goldade, M.; Post, G.; Harnois, M.; Mosby, D. E.; Casteel, S. W.; Berti, W.; Carpenter, M.; Edwards, D.; Cragin, D.; Chappell, W. Advances in evaluating the oral bioavailability of inorganics in soil for use in human health risk assessment. Environ. Sci. Technol. 1999, 33, 3697–3705. (9) Rees, M.; Sansom, L.; Rofe, A.; Juhasz, A. L.; Smith, E.; Weber, J.; Naidu, R.; Kuchel, T. Principles and application of an in vivo swine assay for the determination of arsenic bioavailability in contaminated matrices. Environ. Geochem. Health 2009, 31, 167–177. (10) Juhasz, A. L.; Smith, E.; Weber, J.; Rees, M.; Rofe, A.; Kuchel, T.; Sansom, L.; Naidu, R. In vitro assessment of arsenic bioaccessibility in contaminated (anthropogenic and geogenic) soils. Chemosphere 2007, 69, 69–78. (11) Basta, N. T.; Foster, J. N.; Dayton, E. A.; Rodriguez, R. R.; Casteel, S. W. The eﬀect of dosing vehicle on arsenic bioaccessibility in smelter-contaminated soil. J. Environ. Health Sci., Part A 2007, 42, 1275–1281. (12) Juhasz, A. L.; Smith, E.; Weber, J.; Rees, M.; Rofe, A.; Kuchel, T.; Sansom, L.; Naidu, R. Comparison of in vivo and in vitro methodologies for the assessment of arsenic bioavailability in contaminated soils. Chemosphere 2007, 69, 961–966. (13) Juhasz, A. L.; Weber, J.; Smith, E.; Naidu, R.; Rees, M.; Rofe, A.; Kuchel, T.; Sansom, L. Assessment of four commonly employed in vitro bioaccessibility assays for predicting in vivo relative arsenic bioavailability in contaminated soils. Environ. Sci. Technol. 2009, 43, 9487–9494. 10682

dx.doi.org/10.1021/es2018384 |Environ. Sci. Technol. 2011, 45, 10676–10683

Environmental Science & Technology (14) Rodriguez, R. R.; Basta, N. T.; Casteel, S. W.; Pace, L. W. An in-vitro gastrointestinal method to assess bioavailable arsenic in contaminated soils and solid media. Environ. Sci. Technol. 1999, 33, 642–649. (15) Carrizales, L.; Razo, I.; Tellez-Hernandez, J. I.; Torres-Nerio, R.; Torres, A.; Batres, L. E.; Cubillas, A.-C.; Díaz-Barriga, F. Exposure to arsenic and lead of children living near a copper-smelter in San Luis Potosi, Mexico: importance of soil contamination for exposure of children. Environ. Res. 2006, 101, 1–10. (16) Cave, M. R.; Wragg, J.; Palumbo, B.; Klinck, B. A. Measurement of the bioaccessibility of arsenic in UK soil. R&D Technical Report P5-062/TR02; British Geological Survey: Keyworth, Nottingham, UK, 2003. (17) Cutler, W. Bioaccessible arsenic in soils of the island of Hawaii. Ph.D. Dissertation, University if Hawai’I, 2011. (18) Devesa-Rey, R.; Paradelo, R.; Díaz-Fierros, F.; Barral, M. T. Fractionation and bioavailability of arsenic in the bed sediment of the Anllons River (NW Spain). Water Air Soil Pollut. 2008, 195, 189–199. (19) Girouard, E.; Zagury, G. J. Arsenic bioaccessibility in CCAcontaminated soils: inﬂuence of soil properties, arsenic fractionation and particle-size fraction. Sci. Total Environ. 2009, 407, 2576–2585. (20) Oomen, A. G.; Hack, A.; Minekus, M.; Zeijdner, E.; Cornelis, C.; Schoeters, G.; Verstraete, W.; Van de Wiele, T.; Wragg, J.; Rompelberg, C. J. M.; Sips, A. J. A. M.; Van Wijnen, J. H. Comparison of ﬁve in vitro digestion models to study the bioaccessibility of soil contaminants. Environ. Sci. Technol. 2002, 36, 3326–3334. (21) Pouschat, P.; Zagury, G. J. In vitro gastrointestinal bioavailability of arsenic in soil collected near CCA-treated utility poles. Environ. Sci. Technol. 2006, 40, 4317–4323. (22) Ruby, M. V.; Davis, A.; Schoof, R.; Eberle, S.; Sellstone, C. M. Estimation of lead and arsenic bioavailability using a physiologically based extraction test. Environ. Sci. Technol. 1996, 30, 422–430. (23) Sarkar, D.; Makris, K. C.; Parra-Noonan, M. T.; Datta, R. Eﬀect of soil properties on arsenic fractionation and bioaccessibility in cattle and sheep dipping vat soils. Environ. Int. 2007, 33, 164–169. (24) Williams, T. M.; Rawlins, B. G.; Smith, B.; Breward, N. In-vitro determination of arsenic bioavailability in contaminated soil and mineral beneﬁciation waste from Ron Phibum, Southern Thailand: a basis for improved human risk assessment. Environ. Geochem. Health 1998, 20, 169–177. (25) Wragg, J.; Cave, M.; Nathanail, P. A. Study of the relationship between arsenic bioaccessibility and its solid-phase distribution in soils from Wellingborough, UK. J. Environ. Sci. Health, Part A 2007, 42, 1303–1315. (26) Smith, E.; Naidu, R.; Weber, J.; Juhasz, A. L. The impact of sequestration on the bioaccessibility of arsenic in long-term contaminated soils. Chemosphere 2008, 71, 773–780. (27) Tang, X. Y.; Zhu, Y. G.; Shan, X. Q.; McLaren, R.; Duan, J. The ageing eﬀect on bioaccessibility and fractionation of arsenic in soils from China. Chemosphere 2007, 66, 1183–1190. (28) Yang, J.; Barnett, M. O.; Jardine, P. M.; Basta, N. T.; Casteel, S. W. Adsorption, sequestration, and bioaccessibility of As(V) in soils. Environ. Sci. Technol. 2002, 36, 4562–4569. (29) Dixit, S.; Hering, J. G. Comparison of arsenic(V) and arsenic(III) sorption onto iron oxide minerals: implications for arsenic mobility. Environ. Sci. Technol. 2003, 37, 4182–4189. (30) Smith, E.; Naidu, R.; Alston, A. M. Sorption of arsenate and arsenite by four Australian soils. J. Environ. Qual. 1999, 28, 1719–1726. (31) Fendorf, S.; La Force, M. J.; Li, G. Temporal changes in soil partitioning and bioaccessibility of arsenic, chromium and lead. J. Environ. Qual. 2004, 33, 2049–2055. (32) Aharoni, C.; Sparks, D. L. Kinetics of soil chemical reactions a theoretical treatment. In Rates of Soil Chemical Processes; Sparks, D. L., Suarez, D. L., Eds.; SSSA: Madison, WI, 1991; pp 1 19. (33) Meunier, L.; Walker, S. R.; Wragg, J.; Parsons, M. B.; Koch, I.; Jamieson, H. E.; Reimer, K. J. Eﬀects of soil composition and mineralogy on the bioaccessibility of arsenic from tailings and soil in gold mine districts of Nova Scotia. Environ. Sci. Technol. 2010, 47, 2667– 2674.

ARTICLE

(34) Bowell, R. J. Sorption of arsenic by iron oxides and oxyhydroxides in soils. Appl. Geochem. 1994, 9, 279–286. (35) Hoﬀman, A. F. Regulation of metal absorption in the gastrointestinal tract. Gut 1996, 39, 625–628. (36) Loska, K.; Wiechuza, D. Comparison of sample digestion procedure for the determination of arsenic in bottom sediment using hydride generation AAS. Microchim. Acta 2006, 154, 235–240.

10683

dx.doi.org/10.1021/es2018384 |Environ. Sci. Technol. 2011, 45, 10676–10683

ARTICLE pubs.acs.org/est

Temporal Dynamics of Periphyton Exposed to Tetracycline in Stream Mesocosms Erin L. Quinlan,† Christopher T. Nietch,‡ Karen Blocksom,† James M. Lazorchak,*,† Angela L. Batt,† Richard Griﬃths,§ and Donald J. Klemm† †

Oﬃce of Research and Development, National Exposure Research Laboratory, Ecological Exposure Research Division, U.S. Environmental Protection Agency, 26 West Martin Luther King Drive, Cincinnati, Ohio 45268, United States ‡ Oﬃce of Research and Development, National Risk Management Research Laboratory, Water Supply Water Resources Division, U.S. EPA, 26 West Martin Luther King Drive, Cincinnati, Ohio 45268, United States § Oﬃce of Research and Development, National Risk Management Research Laboratory, Land Remediation and Pollution Control Division, U.S. EPA, 26 West Martin Luther King Drive, Cincinnati, Ohio 45268, United States

bS Supporting Information ABSTRACT: Signiﬁcant amounts of antibiotics enter the environment via point and nonpoint sources. We examined the temporal dynamics of tetracycline exposure to stream periphyton and associated organisms across a logarithmically dosed-series of experimental mesocosms, designed to mimic natural conditions. Target in-stream tetracycline exposures were based on environmentally relevant concentrations in aquatic ecosystems throughout the United States (<1100 μg L1). Signiﬁcant changes in the stream biotic community were observed within 7 days with in-stream tetracycline concentrations as low as 0.5 μg L1, including signiﬁcant changes in antibiotic resistance, bacteria abundance and productivity, algae biomass, cyanobacteria, organic biomass, and nematodes. These eﬀects were magniﬁed with increased exposure time and dosing concentration. Recovery of the periphyton community after 28 days of exposure was dependent upon the tetracycline dose. At the highest doses, 10 and 100 μg L1, bacteria productivity recovered; however, bacteria, algae, and nematode abundance did not recover at the same rate and remained low even after a 28-day recovery period (of nondosing). This study demonstrates that tetracycline exposure under near-natural conditions and at concentrations currently observed in aquatic environments may have important consequences for the structure and function of stream periphyton and, potentially, public health via increasing resistance of naturally occurring bacteria.

’ INTRODUCTION Widespread use of antibiotics to reduce the severity and duration of many diseases and infections has resulted in increased detection in streams and rivers as well as an increase in antibioticresistant bacteria.1 A signiﬁcant amount of antibiotics are not metabolized but excreted from the body (up to 75% of the original active form).2 Antibiotics detected downstream of human-use areas are the result of incomplete antibiotic removal in the wastewater treatment process (i.e., wastewater treatment plants, hospital onsite systems, septic systems).3 An alternative pathway, via nonpoint discharge, stems from extensive use of antibiotics in agriculture and aquaculture (i.e., rainfall-runoﬀ from feedlots, liquid manure spreading, and seepage from animal waste lagoons).46 The presence of these antibiotics in the environment may complicate the treatment and management of disease by selecting for antibiotic-resistant microorganisms79 and potentially contaminate drinking water supplies.10,11 Recent work theorizes that low concentrations of antibiotic compounds play important roles as chemical cues in the structuring of microbial bioﬁlm communities.12 This article not subject to U.S. Copyright. Published 2011 by the American Chemical Society

Although there is clinical evidence of the impacts of antibiotics, particularly with bacteria cultures, there is a paucity of studies examining the potential structural and functional impacts of antibiotics in aquatic ecosystems,13 particularly to periphyton communities. Periphyton adheres to underwater surfaces in aquatic ecosystems and represents a diverse matrix of bacteria, fungus, algae, protozoans, microinvertebrates, macroinvertebrates, and organic and nonorganic detritus. Periphyton is an essential component in aquatic ecosystems, providing community structure and primary productivity that support higher trophic levels, and is a signiﬁcant contributor to carbon ﬁxation and nutrient cycling.14 Antibiotic exposure may elicit changes in the periphyton community, including shifting the overall biology, chemistry, and energy ﬂow of aquatic ecosystems (i.e., carbon storage, denitriﬁcation, decomposition, and nutrient recycling), similar to Received: June 13, 2011 Accepted: November 3, 2011 Revised: November 3, 2011 Published: November 03, 2011 10684

dx.doi.org/10.1021/es202004k | Environ. Sci. Technol. 2011, 45, 10684–10690

Environmental Science & Technology other periphyton stressors, including nutrients, metals, and temperature.1517 Due to the importance of periphyton to stream function, it is imperative to determine baseline estimates of antibiotic exposure and periphyton response. Tetracycline (TC) and its derivatives are among the most extensively used human and animal antibiotics.18 Due to the potential for leaching19 and sorption to soil compounds,20 TC may have far reaching impacts on aquatic trophic structure and function owing to extended half-lives and persistence in the environment under speciﬁc environmental conditions.21 We hypothesized that aquatic ecosystems exposed to environmentally relevant concentrations of TC would be impacted, focusing on the stream periphyton community (i.e., bacteria, algae, microinvertebrates, and macroinvertebrates) as a proximal recipient of TC exposure based on sources and modes of entry into the aquatic environment. Speciﬁcally, we hypothesized that with an increase in TC exposure, the bacteria community would respond with an increase in antibiotic resistance, a reduction in total bacteria, and a reduction in bacteria productivity. We also predicted that these bacterial end point responses may illicit a response in other aspects of the periphyton. Using a stream mesocosm experimental design, we were able to assess the supposition that the severity of the impact and recovery of the community to TC exposure would be based on both concentration and length of exposure.

’ MATERIALS AND METHODS Site Description. Five mesocosms within the U.S. Environmental Protection Agency (U.S. EPA) Experimental Stream Facility (ESF) were used to examine the short- and long-term impacts of TC exposure on the periphyton community from August to October 2007. Conditions in the mesocosms were semicontrolled and designed to mimic the conditions found in natural streams. Source water for this study was the Lower East Fork of the Little Miami River (LEFLMR), Clermont County, OH, USA. Additional details are available in the Supporting Information (SI) for this paper. Experimental Conditions. Background water quality conditions were continuously measured for each mesocosm and included dissolved oxygen, pH, specific conductance, water temperature, oxidation reduction potential, and turbidity. Surface and intergravel water samples were collected and analyzed for carbon and nutrient species every 14 days. Water temperature declined signiﬁcantly across experimental periods during the experiment due to seasonal changes, while other climate and surface water parameters remained statistically similar. Throughout the study, irradiance represented ca. 15% of open canopy conditions. Phosphorus within the intergravel space of the streambed generally increased as the experiment progressed from colonization through recovery. Intergravel ammonium also increased, while nitrite-nitrate decreased and total nitrogen remained unchanged across periods. Trends for inorganic nutrient species in the mesocosm streambeds were reﬂective of sediment deposition within the gravel matrix, creating lower redox conditions. For mean background mesocosm conditions during colonization, dosing, and recovery, see Table S1. Tetracycline Dosing. After a 21-day colonization period to allow for the natural colonization of periphyton and associated organisms, experimental mesocosms were exposed to TC continuously for 28 days, so that individual mesocosm concentrations of 0.0, 0.5, 1, 10, and 100 μg L1 were achieved.

ARTICLE

TC concentrations were monitored every 7 days during the experiment using LC/MS/MS. These concentrations were measured at the LEFLMR influent, the ESF effluent, and the 5 mesocosm head and tail tanks to determine potential loss of TC due to absorption and degradation. TC concentrations in the mesocosms during dosing remained within the 20% of the target concentrations (Table S1). LEFLMR influent and control mesocosm concentrations of TC were below detection limits (0.002 μg L1 TC) for all dates sampled. Degradation products of the parent TC compound, anhydrotetracycline and epi-tetracycline, were not detected in the background water or the mesocosms at any time during the experiment. See the SI for detailed methods. Periphyton Development. Periphyton end points were collected from unglazed ceramic tiles (surface area 8870 mm2, n = 3) and included ash-free dry mass, chlorophyll a, bacteria abundance, bacteria productivity (as measured by leucine uptake), antibiotic-resistant bacteria, algae abundance, and algae composition. Invertebrate end points were collected from gravel-filled trays (2.0-cm median particle diameter, n = 2) and included nematodes and macroinvertebrates. Separation of the two substrates within the mesocosms allowed for the periphyton to respond to the exposures without top-down grazing pressures of the invertebrate community. Tiles and trays were randomly sampled from each mesocosm every 7 days during colonization, dosing, and recovery periods from each mesocosm every 7 days during colonization, dosing, and recovery periods. Details on periphyton sampling, preservation, and end point analysis are provided in the SI. Statistical Analysis. Box-Cox transformations were used to find potentially nonlinear transformations of end points (Table S1). Significant differences between treatments were analyzed in SAS version 9.1.3 (SAS Institute Inc., Cary, NC). A separate generalized linear model was run for each response variable and each time period (i.e., dosing, recovery). Time, treatment, and time-treatment interaction effects were included in the model, and a Dunnett’s test was used posthoc to test for differences between treatments and the control (Table S1). Dunnett’s tests are specifically designed for situations where all treatments are to be compared against one 00 reference00 group or the control treatment for each sampling event. By comparing the treatments each week to the control, allowed for the changes in end points based on temperature, seasonal, and sediment load which changed throughout the experiment. In addition, colonization within mesocosms was evaluated prior to TC dosing to determine if further analysis would be considered. If there was a significant difference in organism colonization among mesocosms prior to TC dosing for any end point, all or part of that data were not used in any further statistical analysis. End point data were represented by the weekly mean ( one standard error of the mean (SEM).

’ RESULTS AND DISCUSSION Pharmaceuticals in the environment have been documented since the mid-1980s;22 however, this issue was not highlighted within the scientiﬁc and resource management community until after the detection of a wide-range of pharmaceuticals in streams throughout the United States.23 Consequences of pharmaceuticals, particularly antibiotics, in aquatic ecosystems remain unclear.13 Although there have been several investigations on the fate and transport of antibiotic resistant genes (ARG), and speciﬁcally the migration of ARG into the bioﬁlms,8,9 the paucity of fundamental data on the organismal eﬀects of antibiotics in these ecosystems delays the understanding of potential risks for 10685

dx.doi.org/10.1021/es202004k |Environ. Sci. Technol. 2011, 45, 10684–10690

Environmental Science & Technology

ARTICLE

Figure 1. Mean bacteria abundance ( 107 cells cm2), ash-free dry mass (AFDM) (mg cm2), and ﬁlamentous cyanobacteria (%) ((SEM) for mesocosms exposed to tetracycline (TC) (Dosing) and after TC dosing was discontinued (Recovery). Signiﬁcant diﬀerence relative to control, Dunnett’s test (†, n = 15).

both humans and the environment.3 This study addresses the comprehensive ecological eﬀects of environmentally relevant concentrations of a single widely used antibiotic at an experimental scale applicable to real exposure settings occurring in the environment. Tetracycline represents only one compound in a soup of anthropogenic compounds of emerging concern or emerging contaminants. Evaluation of the eﬀects of mixtures will be of utmost importance if we hope to truly assess the environment as it now exists in many places. Tetracycline Dosing Period. In aquatic communities, periphyton is often one of the first indicators of environmental changes and exposure to contaminants. The examination of the periphyton community after a long-term antibiotic exposure, including changes in antibiotic resistance, bacteria abundance and productivity, algae biomass, cyanobacteria, organic biomass, nematodes, and macroinvertebrates, provided some of the first insights of the consequences of pharmaceuticals in the environment. Periphyton biomass (AFDM, Chl a, bacteria, and cyanobacteria) increased throughout the experiment in the control mesocosms (Figure 1 and Figure S1). This increase in biomass follows similar trends in natural occurring periphyton mats. Seasonal aspects of the experiment are reviewed in the SI.

The ﬁrst targets of antibiotic exposure in streams are aquatic bacteria. A primary human health concern is the potential for increases in antibiotic resistant bacteria with antibiotic exposure.1 We observed signiﬁcant increases in antibiotic resistant culturable bacteria with increases in TC doses and length of exposure time (Table S1). Mean tetracycline resistant bacteria (TRB) colonies for TC treatments 0.5, 1, 10, and 100 μg L1 after 28 d of dosing were 71, 48, 53, and 62% of the control colonies respectively. After dosing was discontinued, TRB colonies for TC treatments 0.5, 1, 10, and 100 μg L1 after 28 d of recovery were 47, 42, 50, and 66% of the control colonies respectively (Figure S2). The signiﬁcant increase in TC resistance is relatively conservative considering that only cultured bacteria was tested for resistance. Culturable bacteria may only account for <1 to 5% of the total bacteria.24 TC-resistant bacteria (TRB) colonies were observed in the control mesocosm, which suggests two potential mechanisms the rise of a spontaneous antibiotic resistant mutation from plating on TC-infused plates and/or prior TC exposure in the LEFR, providing inoculums of mobile genetic elements, such as plasmids and integrons25 to the organisms colonizing the ESF mesocosms. Two wastewater treatment plants and numerous septic 10686

dx.doi.org/10.1021/es202004k |Environ. Sci. Technol. 2011, 45, 10684–10690

Environmental Science & Technology systems upstream of the ESF intake are probable sources of TC exposure and may represent a selection factor for TC-resistant genes in the LEFR.8 However, despite the presence of probable sources of TC in the source water supplied to the ESF mesocosms, the observed TRB in the control was low compared to that reported in other studies.23,2527 Another component of the bacteria community, abundance, was also examined during and after TC exposure. Bacteria abundance increased signiﬁcantly during the TC dosing period in the control mesocosm, particularly by 21 days. These densities were similar to bacteria densities reported for natural periphyton communities.2830 However, all TC treated mesocosms had signiﬁcantly lower bacteria densities relative to the control (Table S1, Figure 1), even after only 7 days of exposure. Bacteria abundance in the 0.5, 1, 10, and 100 μg L1 was less than 31, 20, 9, and 9% of the bacteria abundance in the control mesocosm (74.9 107 cells cm2), respectively, by the end of the dosing period (Figure 1). Statistical analysis also indicated a signiﬁcant time TC eﬀect, suggesting that longer exposures may signiﬁcantly impact bacteria density (Table S1). Other shorter-term studies have reported eﬀects on abundance at concentrations an order of magnitude greater (mg L1).13,31 Our data suggest that the low concentrations of TC reported in the literature for natural streams and rivers (<1 μg 3 L1) 23 have a signiﬁcant eﬀect on periphytic bacteria abundance if exposures are at minimum 7-day durations. Although TC resistance increased with TC exposure, resistant bacteria did not oﬀset the loss of nonresistant bacteria in the higher dosed mesocosms. Consequences of reducing the number of bacteria in stream periphyton was not inconsequential and could have ecosystem-wide consequences, including shifting food-web dynamics and altering decomposition rates,31 ultimately changing the carbon source/sink strength of the periphyton community.32 Eﬀects of TC on bacteria production were also examined via leucine uptake during protein synthesis. Bacterial protein synthesis, as measured by the incorporation of leucine, provides a direct estimate of bacterial carbon production. Signiﬁcant decreases in cell-speciﬁc leucine uptake were observed in mesocosms dosed at 10 and 100 μg L1. The mode of action by which TC serves as an antibiotic is by binding the 30s subunit of the ribosome, preventing protein synthesis. This mechanism is bacteriostatic but at high concentrations can become bacteriocidal. Under a static condition (10 μg L1), signiﬁcant decreases in total bacteria protein synthesis have been reported.33 Under the ﬂowing conditions with constant renewal simulated at ESF, we observed a similar response to bacteria protein synthesis when the leucine incorporation was normalized by bacteria abundance at 10 and 100 μg L1 (Table S1, Figure S2). In terms of periphyton dynamics and function, the trend in AFDM (Figure 1) and Chlorophyll a (Chl-a) (Figure S1), a proxy for algal biomass, increased over the dosing period in the control, 0.5, and 1 μg L1 during the dosing period. The mean nonorganic content (ash) in the control and low dose mesocosms (0.5 and 1 μg L1) over the dosing period was 87.9% ( 0.5 SEM of the dry weight. In contrast, ash in the higher level TC treatments (10 and 100 μg L1) decreased to 84 and 83.5% ( 1.0 SEM of the dry weight, respectively. A reduction in the ash content indicates a shift in the sediment retaining abilities of the periphyton mat, which in turn can be linked to the composition of the algae. Algae composition in the mesocosms was dominated by diatoms throughout the experiment, speciﬁcally Melosira varians

ARTICLE

Agardh, Synedra ulna Ehrenberg, Nitzschia ﬁliformis (Smith) Van Heurck, and Cocconeis spp. Ehrenberg. On the last day of colonization (day 0), diatoms composed 9097% of the total algal composition based on algal units. Remaining algae (310%) was composed of cyanobacteria, chlorophytes, phytoﬂagellates, and dinoﬂagellates. Cyanobacteria increased in relative abundance in the control during the dosing period, reaching 27% of the total algae composition in 14 days, and was characterized by ﬁlamentous cyanobacteria, including Oscillatoria spp. Vaucher ex Gomont, Lyngbya spp. Gomont, Anabaena spp. Bory de Saint-Vincent ex Bornet & Flahault, Spirulina spp. Turpin ex Gomont, and Phormidium spp. K€utzing ex Gomont. However, there were signiﬁcantly lower abundances of cyanobacteria observed in all TC treatments (Figure 1, Table S1). Most notably, cyanobacteria was absent from the community in the 100 μg L1 mesocosm after 28 days of TC dosing, even after counting was extended to 2000 algal units. The algal periphyton response to TC exposure was signiﬁcant in that both an overall reduction in biomass (i.e., Chl-a) was observed, and there was a shift in the dominant taxa. Diatoms and ﬁlamentous cyanobacteria dominated the periphyton in the control. As TC exposure increased in duration and concentration, the algal community in the treatments greater than 1 μg L1 shifted to one dominated almost exclusively by diatoms. Similar observations were observed in static phytoplankton mesocosms exposed to a mixed tetracycline treatment of greater than 10 μg L1.34 We observed a decline in cyanobacteria abundance at TC concentrations currently observed in the environment (i.e., 1 μg 3 L1).23 The reduction of cyanobacteria can be attributed to the same mechanism observed for bacteria, as the 30s ribosome is found in all prokaryotic organisms, including cyanobacteria. There was no signiﬁcant diﬀerence in biovolumes of the individual taxa with TC dose. Loss of the ﬁlamentous cyanobacteria fraction of the periphyton changed the physical functionality in the stream ecosystem. Periphyton mats trap and retain sediment and organic particles and regulate dissolved gas and nutrient exchange between the bed and water column. A periphyton mat dominated by diatoms tends to be relatively more porous than cyanobacteria-dominated mats.35 Mat porosity aﬀects the exchange rate of materials between the bed and water column, with investigators noting how mat architecture aﬀects biogeochemical ﬂuxes and solute hydrodynamics.36,37 As periphyton mats develop, the matrix becomes less interspersed with channels and exchange decreases, presumably leading to build-up of mineralized material in the bed (i.e., higher ash content). TC exposure, particularly at prolonged high concentrations, prevented the succession of periphyton development, resulting in a thin, diﬀuse mat composed of ﬁlamentous diatoms and greatly reduced habitat complexity for invertebrates. To assess the relative importance of this eﬀect, the results from the nutrient concentrations measured in the intergravel space were analyzed. There was no signiﬁcant diﬀerence measured among TC treatments in the concentrations of nitrogen and phosphorus species in the intergravel space. This suggests that the relative magnitude of TC eﬀects on periphyton structure was not such to aﬀect gravel-bed to water column nutrient exchange under conditions represented in our experiments. The decline and eventual loss of an important ecological component of the periphyton, the ﬁlamentous cyanobacteria at higher concentrations of TC, also resulted in the potential loss of bacteria habitat in the mucilage in the cyanobacteria. Mucilage provides both substrate and nutrients for heterotrophic bacteria. So that in addition to the direct eﬀect of TC, the loss of the 10687

dx.doi.org/10.1021/es202004k |Environ. Sci. Technol. 2011, 45, 10684–10690

Environmental Science & Technology

ARTICLE

Figure 2. Mean abundance of nematodes per m2 ((SEM) during tetracycline (TC) exposure (Dosing) and after TC dosing was discontinued (Recovery). Signiﬁcant diﬀerence relative to control, Dunnett’s (†).

cyanobacteria community may have also resulted in an indirect eﬀect of substrate loss in heterotrophic bacteria. TC exposure also had a negative impact on the invertebrate community, which subsequently begs the question: is this a direct or indirect aspect of TC exposure. Periphyton functions as habitat and a food source for stream invertebrates. Signiﬁcant losses in four aspects of the periphyton - bacteria, algae, AFDM, and nonorganic matter (ash) - may have led to a cascade eﬀect in higher trophic levels, as both micro- and macroinvertebrates in this study declined. In terms of microinvertebrates, nematode abundance increased in the control and at lower doses of TC (0.5 and 1 μg L1) over the dosing period, while nematode abundance was signiﬁcantly lower in the 100 μg L1 treatment after 21- and 28-day dosing (Table S1; Figure 2). Many nematodes utilize a symbiotic relationship with bacteria, cultivating a “bacteria garden” in surrounding mucus trails and then revisiting these trails to consume both bacteria and bacterial enzymes.38 Loss of this food source may explain the decline in nematodes at higher TC concentrations. TC exposure has also been shown to reduce the endosymbiotic bacteria levels associated with parasitic nematodes, resulting in skewed sex ratios and stunted reproduction in nematodes.39 Without further analysis it is diﬃcult to determine if the eﬀect of TC on nematode abundance was direct or indirect; however, this study suggests evidence of a reduction in nematode abundance with concentrations of TC at or greater than 100 μg L1. Macroinvertebrate abundance also decreased with increasing doses of TC (Figure S3). By the end of the colonization period, the macroinvertebrate community was primarily composed of aquatic insects (55% of the total macroinvertebrates), mollusks (7.3%), annelids (9%), and crustaceans (7%). Among the insects, the family Chironomidae dominated the community (70% of all insects during colonization), while the remaining insects were represented by the orders Ephemeroptera, Plecoptera, and Trichoptera taxa (EPT). As stated in the statistical methods, the dynamic nature of ﬂow-through, continuously inoculated mesocosms, necessitated the need for statistically similar mesocosms in terms of organisms prior to our analysis during and post TC dosing. Signiﬁcant differences in the macroinvertebrate community resulted in the elimination of these organisms from the analysis of the TC dosing response. One of the overlying patterns in the aquatic insect data is an overall decline in abundance as the experiment progressed. This phenomenon has been observed in previous ESF studies conducted at the same time of year40,41 and can be attributed to seasonal changes in the source water from late summer to early

fall, in which aquatic insects decline signiﬁcantly with cooler weather. Despite general declines in insects for all treatments during the dosing period, insect densities in the 10 and 100 μg L1 treatments were lower relative to the control (Table S1; Figure S3). Periphyton Recovery after TC Dosing. Differences in the recovery of periphyton after exposure to TC were an important aspect of this experiment. At the highest TC concentrations (10 and 100 μg L1), cell-specific leucine uptake recovered to control conditions within 7 days (Table S1; Figure S1); however, bacteria abundance, AFDM, Chl-a, cyanobacteria, and nematodes required longer periods of recovery, and certain organismal (bacteria, algae, microinvertebrates) and functional end points (AFDM, ash content) at the higher doses of TC (100 μg L1) did not recover within 28 d (Table S1; Figures 1 and 2, Figures S1 and S2). The recovery aspect of the bacteria and cyanobacteria abundance was unexpected due to the experimental design, which promoted continuous colonization from upstream sources, similar to natural systems. Although there are seasonal aspects of bacteria and cyanobacteria abundances in the source water, there was sufficient inoculum to allow for recovery concentrations to within the control levels within 7 days for both bacteria and cyanobacteria. Low recovery rates suggest lingering effects of TC exposure to the periphyton community, although no TC was detected in the outflow after dosing was discontinued. It has been reported that TC adsorbed to soils retains its antimicrobial activity.42 Lingering effects observed in this experiment adds support to the potential for remnant TC activity to continue to impact the periphyton community after TC dosing is discontinued and is a likely scenario in streams impacted by pulses of antibiotic contaminants. The relatively rapid response and subsequent slow recovery of periphytic bacteria and cyanobacteria after TC exposures in the range of 10 to 100 μg L1 implies that pulses of antibiotic exposure due to episodic events, such as high rainfall-related breaching of waste lagoons or agricultural runoﬀ, could have long-lasting impacts on the periphyton community. Reiterating the importance of periphyton to material ﬂuxes in streams, these long-lasting reductions after TC exposure could translate to signiﬁcant losses in stream function, particularly with respect to carbon ﬂux through the food-webs, as signiﬁcant amounts of periphyton carbon are incorporated into higher trophic levels in streams.43 The mesoscale attention of this antibiotic dosing experiment allowed for the determination of detailed, diverse, and prolonged responses of a stream periphyton community to TC exposures 10688

dx.doi.org/10.1021/es202004k |Environ. Sci. Technol. 2011, 45, 10684–10690

Environmental Science & Technology under conditions simulating environmentally relevant concentrations with continuous water renewal and, therefore, constant sources of biological seed material to support natural colonization rates. The results suggest that several aspects of stream periphyton were directly aﬀected by, and were proportional to, the TC dose, including bacteria resistance, bacteria abundance, cell-speciﬁc bacteria productivity, AFDM, and algal abundance and composition; coinciding responses were also observed in the micro- and macroinvertebrate communities at TC doses simulating the higher end of the range of environmentally relevant concentrations. This study demonstrates that TC exposure under near-natural conditions and at concentrations currently observed in aquatic environments may have important consequences for the structure and function of stream periphyton and, potentially, public health via increasing resistance of naturally occurring bacteria.

’ ASSOCIATED CONTENT

bS

Supporting Information. Detailed materials and methods discussion; a table describing the statistics (Table S1); a table describing background mesocosm conditions during each experimental period (Table S2); and ﬁgures showing chlorophyll a concentrations, relative percentage of tetracycline-resistant bacteria and cell-speciﬁc leucine uptake, and macroinvertebrate composition and abundance (Figures S1S3). This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 513-569-7076. E-mail: [email protected]. Corresponding author address: USEPA Facilities, 26 West Martin Luther King Drive, Mail Code: 642, Cincinnati, OH 45268.

’ ACKNOWLEDGMENT The United States Environmental Protection Agency through its Oﬃce of Research and Development funded and managed the research described here. It has been subjected to the Agency’s administrative review and approved for publication. We acknowledge the following for their support: Balaji Ramakrishnan, Jackie Tompkins, Don Brown, Adam Lehmann, Stacy Pfaller, Herman Haring, Rajib Sinha, Mike Elovitz, Brian Hill, Colleen Elonen, Leo Fichter, Joel Allen, Paul Wernsing, Kimberly Thiesen, Susanna DeCelles, and William Thoeny. ’ REFERENCES (1) Biyela, P. T.; Lin, J.; Bezuidenhout, C. C. The role of aquatic ecosystems as reservoirs of antibiotic resistant bacteria and antibiotic resistance genes. Water Sci. Technol. 2004, 50 (1), 45–50. (2) Elmund, G. K.; Morrison, S. M.; Grant, D. W.; Nevins, M. P. Role of excreted chlortetracycline in modifying the decomposition process in feedlot waste. Bull. Environ. Contam. Toxicol. 1971, 6 (2), 129132; DOI 10.1007/BF01540093. (3) K€ummerer, K. Signiﬁcance of antibiotics in the environment. J. Antimicrob. Chemother. 2003, 52 (1), 57; DOI 10.1093/jac/dkg293. (4) Hamscher, G.; Sczesny, S.; H€oper, H.; Nau, H. Determination of persistent tetracycline residues in soil fertilized with liquid manure by high-performance liquid chromatography with electrospray ionization tandem mass spectrometry. Anal. Chem. 2002, 74 (7), 15091518; DOI 10.1021/ac015588m.

ARTICLE

(5) Lipsitch, M; Singer, R. S.; Levin, B. R. Antibiotics in agriculture: when is it time to close the barn door? Proc. Natl. Acad. Sci. U.S.A. 2002, 99 (9), 57525754; DOI 10.1073/pnas.092142499. (6) Thurman, E. M.; Dietz, J. E.; Scribner, E. A. Occurrence of antibiotics in water from ﬁsh hatcheries; USGS Fact Sheet 120-02; U.S. Geological Survey: Lawrence, KS, 2003. http://ks.water.usgs.gov/pubs/ fact-sheets/fs.120-02.pdf (accessed month day, year). (7) Kumar, K.; Gupta, S. C.; Chander, Y.; Singh, A. K. Antibiotic use in agriculture and its impact on the terrestrial environment. Adv. Agron. 2005, 87,154; DOI 10.1016/S0065-2113(05)87001-4. (8) Engemann, C. A.; Keen, P. L.; Knapp, C. W.; Hall, K. J.; Graham, D. W. Fate of tetracycline resistance genes in aquatic systems; migration from the water column to peripheral bioﬁlms. Environ. Sci. Technol. 2008, 42 (14), 51315136; DOI 0.1021/es800238e. (9) Knapp, C. W.; Engemann, C. A.; Hanson, M. L.; Keen, P. L.; Hall, K. J.; Graham, D. W. Indirect evidence of transposon-mediated selection of antibiotic resistance genes in aquatic systems at low-level oxytetracycline exposures. Environ. Sci. Techol. 2008, 42 (14), 53485353; DOI 10.1021/es703199g. (10) Webb, S.; Ternes, T.; Gibert, M.; Olejniczak, K. Indirect human exposure to pharmaceuticals via drinking water. Toxicol. Lett. 2003, 142 (3), 157167; DOI 10.1016/S0378-4274(03)00071-7. (11) Chuanwu, X.; Zhang, Y.; Marrs, C. F.; Ye, W.; Simon, C.; Foxman, B.; Nriagu, J. Prevalence of antibiotic resistance in drinking water treatment and distribution systems. Appl. Environ. Micro. 2009, 75 (17), 57145718; DOI 10.1128/AEM.00382-09. (12) Mlot, C. Antibiotics in nature: beyond biological warfare. Science 2009, 324 (5935), 16371639; DOI 10.1126/science.324_1637. (13) Halling-Sørensen, B.; Sengeløv, G.; Tjørnelund, J. Toxicity of tetracyclines and tetracycline degradation products to environmentally relevant bacteria, including selected tetracycline-resistant bacteria. Arch. Environ. Contam. Toxicol. 2002, 42 (3), 263271; DOI 10.1007/ s00244-001-0017-2. (14) Azim, M. E.; et al. Periphyton and aquatic production: an introduction. In Periphyton: Ecology, Exploitation, and Management; Azim, M. E., Verdegem, M. C. J., van Dam, A. A., Beveridge, M. C. M., Eds.; CABI: Cambridge, MA, 2005; p 319. (15) McCormick, P. V.; Rawlik, P. S.; Lurding, K.; Smith, E. P.; Sklar, F. H. Periphyton-water quality relationships along a nutrient gradient in the northern Florida Everglades. J. N. Am. Benth. Soc. 1996, 15 (4), 433–449. (16) Hill, B. H.; Herlihy, A. T.; Kaufmann, P. R.; Stevenson, R. J.; McCormick, F. H.; Burch Johnson, C. Use of periphyton assemblage data as an index of biotic integrity. J. N. Am. Benth. Soc. 2000, 19 (1), 50–67. (17) Boivin, M.-E. Y.; Massieux, B.; Breure, A. M.; van den Ende, F. P.; Greve, G. D.; Rutgers, M.; Admiraal, W. Eﬀects of copper and temperature on aquatic bacterial communities. Aquat. Toxicol. 2005, 71 (4), 345356; DOI 10.1016/j.aquatox.2004.12.004. (18) Sarmah, A. K.; Meyer, M. J.; Boxall, A. B. A. A global perspective on the use, sales, exposure pathways, occurrence, fate and eﬀects of veterinary antibiotics (VAs) in the environment. Chemosphere, 2006, 65, 725-759; DOI 10.1016/j.chemosphere.2006.03.026. (19) Hamscher, G.; Pawelzick, H. T.; H€oper, H.; Nau, H. Diﬀerent behavior of tetracyclines and sulfonamides in sandy soils after repeated fertilization with liquid manure. Environ. Toxicol. Chem. 2005, 24 (4), 861868; DOI 10.1897/04-182R.1. (20) Pils, J. R. V.; Laird, D. A. Sorption of tetracycline and chlortetracycline on K- and Ca-saturated soil clays, humic substances, and clayhumic complexes. Environ. Sci. Technol. 2007, 41 (6), 19281933; DOI 10.1021/es062316y. (21) Hirsch, R.; Ternes, T.; Haberer, K.; Kratz, K.-L. Occurrence of antibiotics in the aquatic environment. Sci. Total Environ. 1999, 225 (1), 109118; DOI 10.1016/S0048-9697(98)00337-4. (22) K€ummerer, K. Pharmaceuticals in the Environment: Sources, Fate, Eﬀects and Risks, 3rd ed.; Springer-Verlag: Berlin, 2008. (23) Kolpin, D. W.; Furlong, E. T.; Meyer, M. T.; Thurman, E. M.; Zaugg, S. D.; Barber, L. B.; Buxton, H. T. Pharmaceuticals, hormones, 10689

dx.doi.org/10.1021/es202004k |Environ. Sci. Technol. 2011, 45, 10684–10690

Environmental Science & Technology and other organic wastewater contaminants in U.S. streams, 19992000: a national reconnaissance. Environ. Sci. Technol. 2002, 36 (6), 12021211; DOI 10.1021/es011055j. (24) Liesack, W.; et al. Microbial diversity in soil: the need for a combined approach using molecular and cultivation techniques. In Modern soil microbiology; van Elsas, J. D., Trevors, J. T., Wellington, E. M. H., Eds.; Marcel Dekker: New York, 1997; p 375. (25) Hellweger, F. L.; Ruan, X.; Sanchez, S. A simple model of tetracycline antibiotic resistance in the aquatic environment (with application to the Poudre River). Int. J. Environ. Res. Public Health 2001, 8, 480497; DOI 10.3390/ijerph8020480. (26) Go~ni-Urriza, M.; Capdepuy, M.; Arpin, C.; Raymond, N.; Caumette, P.; Quentin, C. Impact of an urban eﬄuent on antibiotic resistance of riverine Enterobacteriaceae and Aeromonas spp. Appl. Environ. Microbiol. 2000, 66 (1), 125–132. (27) Pathak, S. P.; Bhattacherjee, J. W.; Ray, P. K. Seasonal variation in survival and antibiotic resistance among various bacterial populations in a tropical river. J. Gen. Appl. Microbiol. 1993, 39 (1), 4756; DOI 10.2323/jgam.39.47. (28) Gillies, J. E.; Kuehn, K. A.; Francoeur, S. N.; Neely, R. K. Application of the [3H] leucine incorporation technique for quantiﬁcation of bacterial secondary production associated with decaying wetland plant litter. Appl. Environ. Microbiol. 2006, 72 (9), 59485956; DOI 10.1128/AEM.00696-06. (29) Olapade, O. A.; Leﬀ, L. G. Seasonal dynamics of bacterial assemblages in epilithic bioﬁlms in a northeastern Ohio stream. J. N. Am. Benth. Soc. 2004, 23 (4), 686700; DOI 10.1899/08873593(2004)023<0686:SDOBAI>2.0.CO;2. (30) Carr, G. M.; Morin, A.; Chambers, P. A. Bacteria and algae in stream periphyton along a nutrient gradient. Freshwater Biol. 2005, 50 (8), 13371350; DOI 10.1111/j.1365-2427.2005.01401.x. (31) J€urgens, K.; Sala, M. M. Predation-mediated shifts in size distribution of microbial biomass and activity during detritus decomposition. Oikos 2000, 91 (1), 2940; DOI 10.1034/j.1600-0706.2000.910103.x. (32) Findlay, S.; Pace, M. L.; Lints, D.; Howe, K. Bacterial metabolism of organic carbon in the tidal freshwater Hudson Estuary. Mar. Ecol.: Prog. Ser. 1992, 89, 147–153. (33) Verma, B.; Robarts, R. D.; Headley, J. V. Impacts of tetracycline on planktonic bacterial production in prairie aquatic systems. Microb. Ecol. 2006, 54 (1), 5255; DOI 10.1007/s00248-006-9170-5. (34) Wilson, C. J.; Brain, R. A.; Sanderson, H.; Johnson, D. J.; Bestari, K. T.; Sibley, P. K.; Solomon, K. R. Structural and functional responses of plankton to a mixture of four tetracyclines in aquatic microcosms. Environ. Sci. Technol. 2004, 38 (23), 64306439; DOI 10.1021/es049766f. (35) Wetzel, R. G.. Periphyton in the aquatic ecosystem and food webs. In Periphyton: Ecology, Exploitation, and Management; Azim, M. E., Verdegem, M. C. H., van Dam, A. A., Beveridge, M. C. M., Eds.; CABI: Cambridge, MA, 2005; pp 5166. (36) Battin, T. J.; Kaplan, L. A.; Newbold, J. D.; Cheng, X.; Hansen, C. Eﬀects of current velocity on the nascent architecture of stream microbial bioﬁlms. Appl. Environ. Microbiol. 2003, 69 (9), 54435452; DOI 10.1128/AEM.69.9.5443-5452.2003. (37) Besemer, K.; Singer, G.; Limberger, R.; Chlup, A.-K.; Hochedlinger, G.; H€odl, I.; Baranyi, C.; Battin, T. J. Biophysical controls on community succession in stream bioﬁlms. Appl. Environ. Microbiol. 2007, 73 (15), 49664974; DOI 10.1128/AEM.00588-07. (38) Riemann, F.; Helmke, E. Symbiotic relations of sedimentagglutinating nematodes and bacteria in detrital habitats: the enzymesharing concept. Mar. Ecol. 2002, 23 (2), 93113; DOI 10.1046/j.14390485.2002.02765.x. (39) McCall, J. W.; Jun, J. J.; Bandi, C. Wolbachia and the antiﬁlarial properties of tetracycline. An untold story. Ital. J. Zool. 1999, 66 (1), 710; DOI 10.1080/11250009909356230. (40) Taulbee, W. T.; Nietch, C. T.; Brown, D.; Ramakrishnan, B.; Tompkins, M. J. Ecosystem consequences of contrasting ﬂow regimes in an urban eﬀects stream mesocom study. J. Am. Water Resour. Assoc. 2009, 45 (4), 907927; DOI 10.1111/j.1752-1688.2009.00336.x.

ARTICLE

(41) Belanger, S. E.; Bowling, J. W.; LeBlanc, E. M.; Price, B. B.; Herzog, R. R.; Bozso, E. Comprehensive assessment of aquatic community responses to a new anionic surfactant, high-solubility alkyl sulfate. Ecotoxicol. Environ. Saf. 2005, 62 (1), 7592; DOI 10.1016/ j.ecoenv.2004.12.025. (42) Chander, Y.; Kumar, K.; Goyal, S. M.; Gupta, S. C. Antibacterial activity of soil-bound antibiotics. J. Environ. Qual. 2005, 34 (6), 19521957; DOI 10.2134/jeq2005.0017. (43) Meyer, J. L. The microbial loop in ﬂowing waters. Microb. Ecol. 1994, 28 (2), 195199; DOI10.1007/BF00166808.

10690

dx.doi.org/10.1021/es202004k |Environ. Sci. Technol. 2011, 45, 10684–10690

ARTICLE pubs.acs.org/est

Associations between Levels of Serum Perfluorinated Chemicals and Adiponectin in a Young Hypertension Cohort in Taiwan Chien-Yu Lin,†,‡ LiLi Wen,§ Lian-Yu Lin,|| Ting-Wen Wen,^ Guang-Wen Lien,^ Chia-Yang Chen,# Sandy H.J. Hsu,r Kuo-Liong Chien,O Fung-Chang Sung,[ Pau-Chung Chen,*,^ and Ta-Chen Su*,|| †

Department of Internal Medicine, En Chu Kong Hospital, New Taipei City 237, Taiwan School of Medicine, Fu Jen Catholic University, Taipei County 242, Taiwan § Department of Clinical Laboratory, En Chu Kong Hospital, New Taipei City 237, Taiwan Department of Internal Medicine, National Taiwan University Hospital, Taipei 100, Taiwan ^ Institute of Occupational Medicine and Industrial Hygiene, College of Public Health, National Taiwan University, Taipei 100, Taiwan # Institute of Environmental Health, College of Public Health, National Taiwan University, Taipei 100, Taiwan r Department of Laboratory Medicine, National Taiwan University Hospital, Taipei, Taiwan O Institute of Epidemiology and Preventive Medicine, College of Public Health, National Taiwan University, Taipei 100, Taiwan [ Institute of Environmental Health, College of Public Health, China Medical University, Taichung 404, Taiwan

)

‡

bS Supporting Information ABSTRACT: In animals, perﬂuorinated chemicals (PFCs), speciﬁcally perﬂuorooctanoic acid (PFOA) and perﬂuorooctane sulfate (PFOS), function as peroxisome proliferator-activated receptor (PPAR) alpha agonists. However, the relevance of animal (primarily rodent) data to humans is unresolved. While plasma adiponectin level is very responsive to PPAR gamma agonist drugs, it has not been determined whether adiponectin level is related to serum PFCs concentrations. In the present study, 287 subjects (1230 years of age) were recruited to determine the relationship between serum level of PFCs and serum level of adiponectin. The results showed males had higher serum PFOS concentrations than females and that those with metabolic syndrome had lower serum PFOA than controls. Besides, it showed regional elevations of the perﬂuoroundecanoic acid (PFUA) (median concentration: 7.11 ng/mL) in the study subjects. No relationship of PFOA, PFOS, PFUA, and the sum of all four PFCs was found to glucose homeostasis, adiponectin level, lipid proﬁle, and inﬂammatory markers. The median and the range of perﬂuorononanoic acid (PFNA) concentration (in ng/mL; for four categories corresponding to the <50, 5074, 7589, and g90th percentiles) were 0.38 (0.381.68), 3.22 (1.734.65), 5.85 (4.758.29), 10.56 (8.4025.40), respectively. After controlling for confounding factors, multiple linear regression analysis revealed that the mean natural log-transformed level of adiponectin increased signiﬁcantly across categories of PFNA (in ng/mL; 8.78, 8.73, 9.06, 9.36; P for trend = 0.010 in the full model). In conclusion, higher serum PFNA concentration is associated with elevated serum adiponectin concentration.

’ INTRODUCTION The toxicity of PFOS and PFOA has been linked to their functions as peroxisome proliferator-activated receptor (PPAR) agonists, including their ability to alter the expression of genes involved in peroxisome proliferation, cell cycle control, and apoptosis. In addition, other PFCs have also been shown to act as strong peroxisomal β-oxidation inducers.1 Moreover, recent studies using advanced technologies in genomics and bioinformatics have shown that several categories of genes are commonly altered by some PFCs including those for peroxisome proliferation, fatty acid metabolism, lipid transport, cholesterol synthesis, proteosome activation and proteolysis, cell communication, and inﬂammation.2 A recent study in mice found low-dose r 2011 American Chemical Society

developmental exposure to PFOA led to greater weight in adulthood and increased serum leptin and insulin levels. Animals exposed to higher doses of PFOA, on the other hand, had decreased birth weight.3 Epidemiological studies in human beings, unlike studies in animals,4 do not report hypocholesteremic eﬀects. In occupational populations, several studies have failed to establish a deﬁnite association between exposure to PFCs and adverse Received: June 18, 2011 Accepted: November 2, 2011 Revised: October 29, 2011 Published: November 02, 2011 10691

dx.doi.org/10.1021/es201964x | Environ. Sci. Technol. 2011, 45, 10691–10698

Environmental Science & Technology health eﬀects.57 A few cross-sectional and longitudinal occupational studies have proposed a positive correlation between PFOA and serum lipid and liver enzymes levels but without clinical relevance.8,9 Examination of a nonworker population versus the rest of the U.S. population after exposure to PFOA through contaminated drinking water found an association of elevated levels of PFOA and PFOS to increased cholesterol10,11 and uric acid.12 In the general U.S. population, analysis of data from the National Health and Nutrition Examination Survey (NHANES) also found a positive association between PFCs and cholesterol, despite much lower levels of exposure to PFOA.13 Whether serum PFCs are associated with diabetes mellitus or glucose homeostasis in humans remains a mystery. A casecontrol study in a community whose members were exposed to PFOA-contaminated groundwater found no association of serum PFOA level and type II diabetes either self-reported or veriﬁed from medical records.14 Using data from the NHANES database in 19992000 and 20032004, we found that serum PFCs concentrations were associated with glucose homeostasis and indicators of metabolic syndrome,15 but a similar analysis by Nelson et al. of data from the 20032004 NHANES database found no such association.13 Adiponectin is a protein hormone that modulates a number of metabolic processes, including glucose regulation and fatty acid catabolism.16 In addition, plasma adiponectin level is highly responsive to treatment with PPAR gamma agonist drugs, and the increase in plasma adiponectin level following pioglitazone treatment is much greater than would be predicted from the change in insulin sensitivity.17 On the basis of the animal study and humans studies mentioned above,3,15 it is possible that PFCs might modulate glucose sensitivity or insulin tolerance. Moreover, some of the toxicity of PFCs has been linked to their functions as PPAR agonists. While plasma adiponectin is very responsive to PPARgamma agonist drugs, the relationship between serum PFCs and adiponectin has not yet been determined. We hypothesize that PFCs might be associated with serum adiponectin levels, glucose, and lipid homeostasis. We conducted a cross-sectional study in a community-based sample of adolescents and young adults in Taiwan selected on the basis of results from a mass urine screening for proteinuria, glucosuria, or hematuria.

’ MATERIALS AND METHODS Participants and Study Design. From 1992 to 2000, the Chinese Foundation of Health in Taipei, Taiwan, conducted an annual urine screening of approximately 2 615 000 to 2 932 000 school age children in grades 112 in Taiwan. A urine strip was used for the screening. School-age children with positive results on two tests for proteinuria, glucosuria, or hematuria underwent a third urine screening test and a general health check-up following the same protocol. The check-up included anthropometric measures, fasting blood tests, and blood pressure (BP). Overall, 131 547 students who participated in the general checkup were referred to their physicians for further diagnosis and follow-up care. This campaign was described in a previous report.18 From 2006 to 2008 we established a cohort based on these hypertensive students and students without hypertension selected from the mass urine screening population. Letters of invitation were mailed to eligible students in the Taipei area and in Taichung City. After 35 days, 12 trained assistants and nurses conducted telephone interviews inviting these subjects to

ARTICLE

Figure 1. Algorithm used to select the participants with PFCs.

come in for a follow-up health examination. Normotensive students were not contacted by telephone. In all, 347 of the 707 subjects with elevated blood pressure (EBP) and 641 of the 6390 subjects with normal BP completed follow-up health questionnaires, for a response rate of 49% and 10%, respectively. The detailed information is available in a recent report.19 In this current study, we selected 790 subjects who lived in the Taipei area whose serum samples were available for further analysis. The interview/check-up was conducted in National Taiwan University Hospital. The study protocol was approved by the ethics committee of National Taiwan University Hospital. Each participant gave her/his inform consent before she/he joined the cohort follow-up study. Of the potential subjects, 144 were eliminated due to lack of suﬃcient serum sample volume. Another 346 subjects were eliminated due to budgetary limitations that prevented the measurement of serum adiponectin levels. An additional 13 subjects were eliminated because they were taking medication for diabetes. They were excluded because numerous studies demonstrated that various agents (e.g., thiazolidinediones and sulphonylureas) increase adiponectin levels in subjects with diabetes.20,21 Therefore, a ﬁnal number of 287 participants were selected for the ﬁnal analysis. A ﬂowchart of the selection process is shown in Figure 1. Anthropometric and Biochemical Data. Sociodemographic information (i.e., age, gender, history of medication, and household income) was elicited by interview. The household income was dichotomized into above versus below 50 000 new Taiwan dollars (NTD) per month. The degree of alcohol intake (as determined by questionnaire) was dichotomized into current versus no consumption of alcohol. Smoking status was either active smoker, environmental tobacco smoke, or never smoked as determined by serum cotinine level and as described on the questionnaire.22 Serum cotinine concentration was measured by the DRI Cotinine Assay for urine (Microgenics Corp., Fremont, CA) on a Dimension RXL analyzer (Siemens Healthcare Diagnostics, Inc., Tarrytown, NY). Weight and height were measured by standard methods, and BMI was calculated. Two seated blood pressure 10692

dx.doi.org/10.1021/es201964x |Environ. Sci. Technol. 2011, 45, 10691–10698

Environmental Science & Technology and heart rate measurements were made at least 1 min apart after 3 min of rest by using a mercury manometer and the appropriate cuff size. Blood was drawn from all study participants after they had fasted more than 8 h. Levels of triglyceride (TG), plasma total cholesterol, low density lipoprotein (LDL), high density lipoprotein (HDL), and glucose were measured by an autoanalyzer (Technician RA 2000 Autoanalyzer; Bayer Diagnostic, Michawaka, IN), and their units are mg/dL . Commercially available kits were used to measure levels of serum insulin (Immulite 2000; Siemens Healthcare Diagnostics), serum adiponectin (Human Adiponectin/Acrp30 Immunoassay; R&D Systems, Minneapolis, MN), and serum high sensitivity (hs)-C-reactive protein (CRP) (chemiluminescent enzymelabeled immunometric assay, Immulite C-Reactive Protein; Diagnostic Products Co., Los Angeles, CA). The homeostasis model assessment (HOMA) of insulin resistance (HOMA-IR) index (the product of basal glucose and insulin levels divided by 22.5) is regarded as a simple, inexpensive, and reliable surrogate measure of insulin resistance.23 Diabetes mellitus was deﬁned as the presence of fasting serum glucose g126 mg/dL or a selfreported current use of oral hypoglycemic agents or insulin. Hypertension was deﬁned as self-reported current use of antihypertensive medication or an average BP greater than 140/90 mmHg. Childhood EBP was deﬁned as systolic blood pressure, diastolic blood pressure, or both greater than or equal to values meeting the modiﬁed sex- and age-speciﬁc standards.24 Measurement of PFCs Concentration. Plasma samples were stored at 80 C before analysis. In all, 8 of the 12 PFCs analyzed in our study were undetectable using a Waters ACQUITY UPLC system (Waters Corporation, Milford, MA) coupled with a Waters Quattro Premier XE triple quadrupole mass spectrometer (Waters Corporation). Therefore, we measured plasma levels of only PFOA, PFOS, perfluorononanoic acid (PFNA), and perfluoroundecanoic acid (PFUA). Details of the analytical method have been described elsewhere.25 Briefly, frozen samples were thawed at room temperature and vortexed for 30 s to ensure homogeneity. Each plasma sample (100 μL) was vortexed for 30 s with 100 μL of 1% formic acid (pH = 2.8), then mixed with 80 μL of methanol and 20 μL of 0.375 ng/mL internal standard (13C8-PFOA) in methanol, sonicated for 20 min, and then centrifuged at 14 000 g for 20 min. The collected supernatant (about 150 μL) was then filtered through a 0.22 μm polyvinyl difluoride (PVDF) syringe filter before instrumental analysis. The standard solutions used for calibration were prepared in 100 μL of bovine plasma and went through the same procedure used for sample preparation; the concentrations of all analytes were equivalent to 0.05300 ng/mL bovine plasma containing a fixed amount of internal standard (75 ng/mL). The limits of quantitation for PFOA, PFOS, PFNA, and PFUA were 1.5, 0.22, 0.75, and 1.5 ng/mL, respectively. In blank samples, trace background amounts of PFOA (up to 1.5 ng/mL), PFNA (up to 0.75 ng/mL), and PFUA (up to 3 ng/mL) but no PFOS was detected; consequently, the reported PFOA, PFNA, and PFUA concentrations are concentrations corrected by subtraction of background levels found in the blank. For concentrations below the detection limits (49.5% for PFOA, 1.7% for PFOS, 44.3% for PFNA, and 26.5% for PFUA), a proxy value of half the detection limit was used. All laboratory analyses were conducted by investigators blinded to the characteristics of study subjects. Total PFCs were determined by summing the four individual PFC congeners (raw or imputed) for each study subject.

ARTICLE

Definition of Metabolic Syndrome. For subjects g20 years old, metabolic syndrome (MS) was defined according to the updated guidelines of the National Cholesterol Education Program Third Adult Treatment Panel (NCEP/ATP III).26 At least three of the following criteria had to be satisfied: waist measurement g 80 cm for women and g 90 cm for men (Asian criteria); serum triglyceride g 1.7 mmol/L; serum HDL-C < 1.03 mmol/L in men and < 1.29 mmol/L in women; systolic blood pressure g 130 mmHg or diastolic blood pressure g 85 mmHg or a selfreport of taking antihypertensive medications; fasting glucose g 5.6 mmol/L or a self-report of taking antihyperglycemic medications. To define MS in participants (age, 1219 years), we used a modification of the definition proposed in the NCEP/ATP III. The participants had to meet 3 of the following 5 criteria: serum concentration of triglyceride g 1.24 mmol/L; HDL-C < 1.03 mmol/L; waist circumference g 92nd percentile for males and 72nd percentile for females; SBP or DBP g 92nd percentile for males and 72nd percentile for females or a self-report of taking antihypertensive medications; glucose concentration g 5.6 mmol/L or a self-report of taking antihyperglycemic medications.27 Statistical Analysis. SPSS for Windows (version 16.0, SPSS Inc., Chicago, IL) was used for all statistical analyses. Because exposure was relatively low in most people and variance was considerably greater at the higher exposure end, PFC concentrations were expressed as the median with range (difference between the maximum and minimum value). The relation of PFC variables to categorical variables was tested using the MannWhitney U test or KruskalWallis test (if 3 or more groups). We also divided PFCs in four categories, with cut points at the 50th (the reference category), 75th, and 90th percentiles in a linear regression model. Analyses were conducted using linear regression with adiponectin, glucose homeostasis, lipid profile, and inflammatory markers as the outcome variables. Natural log transformation was performed for adiponectin, CRP, HOMA-IR, triglyceride levels with significant deviation from the normal distribution before further analyses. All the log-transformed data in the study had a normal distribution (by KolmogorovSmirnov test). Because adiponectin concentration depends on many endogenous and exogenous factors, including age, gender, smoking, alcohol consumption, and renal function,2830 we used an extended model approach for covariates to adjust for potential confounders in multiple linear regression models. Model 1 adjusted for age and gender. Model 2 adjusted for age, gender, life-style factors (smoking status, drinking status, and household income). Model 3 = model 2 adjusted for measurement data (waist measurement, systolic blood pressure, total cholesterol, HOMA-IR, and creatinine). To avoid “model-dependent association”, an association was considered significant only when it remained statistically significant in all models. Each PFC was modeled separately. When the independent variables were the derivatives of the plasma glucose and the serum insulin, no adjustment for either plasma glucose or serum insulin was made in the model. The HOMA-IR was adjusted in the final model if the independent variable was not an arithmetic combination of the plasma glucose and the serum insulin.

’ RESULTS The basic demography of the sample population (121 males and 166 females) is summarized in Table 1. Males had a signiﬁcantly higher median serum PFOS level than females (P < 0.001), and subjects with metabolic syndrome had signﬁcantly lower 10693

dx.doi.org/10.1021/es201964x |Environ. Sci. Technol. 2011, 45, 10691–10698

Environmental Science & Technology

ARTICLE

Table 1. Basic Demographics of the Sample Subjects Including Median (range) of PFC Concentrations no.

PFOA (ng/mL)

PFNA (ng/mL)

PFUA (ng/mL)

287

2.39 (27.38)

8.93 (67.14)

1.68 (25.03)

7.11 (83.96)

males

121

0.75 (27.38)

11.82 (67.14)a

females

1.52 (19.26)

6.07 (65.30)

166

2.45 (22.76)

8.10 (28.34)a

1.81 (25.03)

7.49 (83.96)

1219

78

0.75 (19.62)

8.85 (28.67)

1.76 (19.26)

6.83 (58.91)

2030

209

2.67 (28.38)

8.93 (67.14)

1.64 (25.03)

7.20 (83.96)

<50 000 NT dollar per month

123

0.75 (13.35)

8.54 (67.14)

1.68 (25.03)

6.74 (83.96)

g50 000 NT dollar per month

164

3.02 (27.38)

9.08 (32.14)

1.69 (19.26)

7.22 (65.30)

never smoked

162

0.75 (27.38)

8.94 (67.14)

1.64 (25.03)

7.69 (83.96)

environmental smoker

24

4.07 (13.35)

11.71 (27.45)

2.43 (19.26)

7.60 (52.84)

active smoker

101

2.60 (22.76)

8.51 (33.05)

1.68 (13.04)

5.93 (44.90)

yes

25

3.66 (9.48)

8.41 (67.14)

0.38 (13.25)

5.02 (44.90)

no

262

0.75 (27.38)

9.03 (33.05)

1.75 (25.03)

7.11 (83.96)

<24

206

2.78 (22.76)

8.48 (67.14)

1.65 (25.03)

6.51 (84.00)

g24

81

0.75 (27.38)

10.91 (30.4)

1.78 (13.08)

8.63 (65.30)

yes

17

0.75 (10.00)

9.16 (26.81)

0.38 (6.55)

8.31 (44.91)

no

270

2.55 (27.38)

8.86 (67.14)

1.71 (25.03)

6.86 (83.96)

yes

15

0.75 (7.23)b

8.28 (26.81)

0.38 (11.61)

8.31 (65.30)

no

272

2.67 (27.38) b

8.93 (67.14)

1.71 (25.03)

7.10 (37.16)

total

PFOS (ng/mL)

Sex

Age in Years

Household Income

Smoking Status

Current Alcohol Consumption

2

Body Mass Index (kg/m )

Current Hypertension

Metabolic Syndrome

a

Males have a higher median concentration of PFOS than females (P < 0.001). b PFOA levels were signiﬁcantly lower in subjects with metabolic syndrome (P = 0.009).

serum PFOA level (P = 0.009). Concentrations of the four PFCs were similar in other groups. Some of the PFCs were not correlated with one another (see Supporting Information Table S1). PFNA and PFUA were strongly correlated (Spearman correlation coeﬃcient, 0.62; P < 0.001). Glucose homeostasis, adiponectin level, lipid proﬁle, and level inﬂammatory markers were not signiﬁcantly associated with categories of serum PFOA, PFOS, PFUA, and total PFCs in the simple model (Supporting Information Tables S25). The association of categories of serum PFNA concentration with glucose homeostasis, adiponectin level, lipid proﬁle, and levels of inﬂammatory markers after the adjustment for other potential covariates is summarized in Table 2. The log transformed mean adiponectin level (ng/mL) was signiﬁcantly increased across categories of PFNA in all three models (8.78, 8.73, 9.06, and 9.36; P for trend = 0.010 in the full model). Insulin level and HOMAIR were signiﬁcantly decreased with PFNA in model 1 and model 2 but not in the full model. The PFNA concentration was not associated with levels of glucose, HDL, triglyceride, or CRP. Linear regression coeﬃcients (SEs) of the relationship of serum adiponectin level with categories of PFNA deﬁning different subpopulations of the sample are shown in Table 3. The association between serum adiponectin and PFNA levels was signiﬁcant in subjects with male gender, age 2030 years old, and

higher HOMA-IR. The relationship between serum adiponectin level and quartiles of BMI in diﬀerent subpopulations of the sample is also shown (Supporting Information Table S6).

’ DISCUSSION In this study, increased serum PFNA concentration was associated with elevated serum adiponectin concentration. Studying the potential health consequences of an environmental exposure in adolescents and young adults rather than in older adults might provide greater insight because the number of underlying factors confounding the associations is likely to be smaller. In this study, the median concentration of PFUA was 6.07 ng/mL in males and 7.49 ng/mL in females. Our ﬁndings are compatible with those of another study that analyzed perﬂuorinated chemicals in umbilical cord blood in Taiwan.25 These concentrations are 10 times higher than those previously reported for PFUA while concentrations of PFOA, PFOS, and PFNA are comparable to those of another report in a diﬀerent country.31 Taiwan is the only country thus far reported with high PFUA serum concentrations, possibly from an environmental source. One study which investigated the inﬂuence of discharge by the semiconductor and electronics industries on contamination of river water with PFCs in Taiwan detected PFUA in wastewater but not the 10694

dx.doi.org/10.1021/es201964x |Environ. Sci. Technol. 2011, 45, 10691–10698

Environmental Science & Technology

ARTICLE

Table 2. Mean and Standard Error of Adiponectin, Glucose Homeostasis, Lipid Proﬁle, and Inﬂammatory Marker across Diﬀerent Categories of Serum PFNA Level in Linear Regression Modelsa PFNA percentile %b log-adiponectin (ng/mL) glucose (mg/dL) log-insulin (pmol/L) log-HOMA-IR HDL (mg/dL) log-TG (mg/dL) log-CRP (mg/L) Model 1 <50th

8.76 (0.08)

84.44 (0.54)

1.07 (0.08)

0.50 (0.08)

48.38 (0.77)

4.33 (0.04)

1.59 (0.09)

50th74th 75th89th

8.64 (0.12) 9.14 (0.16)

84.88 (0.77) 84.03 (1.00)

1.09 (0.11) 0.70 (0.14)

0.48 (0.12) 0.87 (0.15)

49.23 (1.10) 48.80 (1.42)

4.35 (0.06) 4.38 (0.07)

1.70 (0.13) 1.27 (0.17)

g90th

9.51 (0.19)

84.20 (1.23)

0.75 (0.18)

0.82 (0.18)

48.89 (1.75)

4.29 (0.09)

1.21 (0.21)

P for trend

<0.001

0.913

0.051

0.057

0.936

0.887

0.138

Model 2 <50th

8.75 (0.14)

84.51 (0.86)

0.97 (0.13)

0.60 (0.13)

47.53 (1.26)

4.31 (0.07)

1.49 (0.15)

50th74th

8.64 (0.16)

85.07 (1.03)

0.99 (0.15)

0.58 (0.16)

48.38 (1.50)

4.34 (0.08)

1.63 (0.18)

75th89th

9.16 (0.19)

84.05 (1.21)

0.59 (0.18)

0.99 (0.18)

48.01 (1.75)

4.35 (0.09)

1.16 (0.21)

g90th

9.48 (0.22)

83.84 (1.38)

0.67 (0.21)

0.91 (0.21)

47.92 (2.01)

4.26 (0.10)

1.11 (0.25)

P for trend

<0.001

0.787

0.054

0.057

0.937

0.981

0.068

Model 3 <50th

8.78 (0.10)

84.77 (0.58)

0.98 (0.08)

0.66 (0.11)

48.64 (0.89)

4.28 (0.04)

1.47 (0.10)

50th74th

8.73 (0.12)

85.33 (0.75)

0.87 (0.10)

0.77 (0.13)

50.20 (1.14)

4.25 (0.05)

1.46 (0.12)

75th89th

9.06 (0.16)

85.51 (0.98)

0.73 (0.13)

0.90 (0.15)

48.20 (1.48)

4.37 (0.07)

1.28 (0.16)

g90th

9.36 (0.19)

85.18 (1.15)

0.88 (0.15)

0.76 (0.17)

47.70 (1.75)

4.31 (0.08)

1.27 (0.19)

P for trend

0.010

0.852

0.328

0.363

0.504

0.479

0.549

a

Model 1: adjusted for age, gender. Model 2: adjusted for age, gender, lifestyle (smoking status, drinking status, household income). Model 3: adjusted for age, gender, lifestyle (smoking status, drinking status, household income) and measurement data (SBP, waist, HOMA-IR, total cholesterol, creatinine). b The median and the range concentrations of the PFNA for the cut points used (<50th, 5074th, 7589th, and g90th percentiles) were listed below: PFNA 0.38 (0.381.68) ng/mL, 3.22 (1.734.65) ng/mL, 5.85 (4.758.29) ng/mL, 10.56 (8.4025.40) ng/mL, respectively.

Table 3. Adjusted Mean and Standard Error of Natural Log Adiponectin (ng/mL) across Diﬀerent Serum Levels of PFNA in SubPopulations of the Sample Subjectsa no.

<50th

50th74th

75th89th

g90th

P for trend

males

121

8.59 (0.16)

8.43 (0.21)

9.01 (0.25)

9.51 (0.23)

0.001

females

166

8.92 (0.20)

8.85 (0.23)

9.04 (0.27)

9.05 (0.35)

0.864

1219

78

8.70 (0.31)

8.36 (0.36)

8.95 (0.40)

8.81 (0.46)

0.504

2030

209

8.82 (0.14)

8.86 (0.17)

9.14 (0.19)

9.41 (0.23)

0.032

206 81

9.05 (0.16) 8.42 (0.19)

9.00 (0.20) 8.19 (0.21)

9.32 (0.23) 8.55 (0.26)

9.61 (0.25) 8.70 (0.35)

0.073 0.356

never smoked

162

8.78 (0.22)

8.73 (0.28)

8.97 (0.30)

9.45 (0.29)

0.084

has smoked

125

8.84 (0.13)

8.72 (0.16)

9.09 (0.20)

9.26 (0.29)

0.148

e0.57

144

9.02 (0.22)

9.19 (0.24)

9.33 (0.25)

9.52 (0.31)

0.310

>0.57

143

8.58 (0.15)

8.27 (0.19)

8.83 (0.26)

9.22 (0.26)

0.005

Sex

Age in Years

Body Mass Index (kg/m2) e24 >24 Smoking Status

HOMA-IR

a

Adjusted for full model.

major PFCs.32 The source of exposure is not clear and needs to be further investigated. Like the eight-carbon PFOA, the nine-carbon PFNA is a developmental toxicant and an immune system toxicant in animals.33 However, longer chain PFCs are considered more bioaccumulative and toxic. PFNA has been shown to be a strong peroxisome

β-oxidation inducer in animals.1 Fibrates (amphipathic carboxylic acids that activate PPAR alpha) can decrease the triglycerides level, normalize the LDL cholesterol proﬁle, and increase HDL cholesterol.34 However, most epidemiological studies in humans1012 unlike studies in animals4 report hypercholesteremic eﬀects. Duration of exposure, dose (concentration), and 10695

dx.doi.org/10.1021/es201964x |Environ. Sci. Technol. 2011, 45, 10691–10698

Environmental Science & Technology interspecies diﬀerences may account for the inconsistent cholesterol ﬁndings between animal and human studies. Moreover, human liver tissue expresses PPAR alpha at approximately 10% of rodent levels, and the liver tissue of humans and other primates (compared with that of rodents) is refractory or less responsive to PPAR alpha agonists.35 In a cross-sectional study of 20032004 NHANES samples, the authors observed a positive association of PFOS, PFOA, and PFNA concentrations with total and non-high-density cholesterol concentrations, and the opposite for perﬂuorohexane sulfonic acid (PFHxS).13 The strongest, most consistent cholesterol results were seen for PFNA, despite lower serum concentrations versus the other PFCs in the NHANES population. However, no signiﬁcant correlation between lipid proﬁle and serum PFCs concentration was found in the current study. Race/ ethnicity diﬀerences between studies and the existence of a higher frequency of aberrant serum lipid proﬁles in adults than in younger people (children and adolescents) may account for this discrepancy between studies. Unlike PFOA and PFOS, which do not activate mouse or human PPAR gamma,36 PFNA activates both PPAR-alpha and PPAR-gamma.33 In the latter study,33 the PFNA (0, 1, 3, or 5 mg/ kg/day) was administered via gastric gavage for 2 weeks. PPAR gamma (along with PPAR alpha) was upregulated in the thymus by 1 mg but not by 3 or 5 mg/kg/day, suggesting maximal transcriptional activity occurs at a dose no higher than 1 mg/kg/ day. Maximal PPAR gamma activity may be induced in mice by a much lower level than the environmental levels to which humans are exposed. Unlike PPAR alpha receptor, PPAR gamma receptor is highly expressed in rodents and humans. PPAR gamma1 is highly expressed in the adipose tissue of adult humans.37 PPAR gamma ligands increases the number of small adipocytes, thereby increasing the adiponectin level and insulin sensitivity and reducing the levels of tumor necrosis factor alpha, which induce insulin resistance.16 The evidence from the above animal study and the present study suggests that PFNA may be involved in the pathway of PPAR-gamma activation. Nonetheless, interaction with insulin and other modulators may also be involved in this mechanism since adiponectin knockout mice are known to develop moderate insulin resistance and glucose intolerance. In this study, increased serum PFNA concentration was associated with elevated serum adiponectin. An approximately 10 ng/mL PFNA diﬀerence in concentration between the 50th percentile and 90th percentile resulted in a 7 μg/mL diﬀerence in mean adiponectin concentration in the present study. In a recent meta-analysis of 13 studies, average adiponectin level ranged from 5 to 15 μg/mL.38 Our study shows the magnitude of the eﬀect if the association between PFNA and adiponectin has etiologic implications,. A nonstatistically signiﬁcant trend was observed with insulin and HOMA-IR with decreased PFNA. This association, however, is similar to the association we found between PFCs and glucose homeostasis in our previous study using 19992000 and 2003 2004 NHANES data.15 In that study, increased serum PFNA concentration in adolescents was associated with decreased blood insulin. However, this was not found by Nelson et al. using the 20032004 NHANES databases. Although there were isolated suggestive trends, such as a signiﬁcant positive trend between PFNA level and HOMA-IR in adult females, they found no association between PFNA concentration and HOMA-IR on the whole study population.13 It is diﬃcult to compare the two studies, inasmuch as we analyzed two additional years of data and

ARTICLE

used diﬀerent criteria for inclusion of subjects. PFNA may increase serum adiponectin level and decrease serum insulin level by acting as an agonist to PPAR gamma, and thereby yielding correlations of PFNA level with serum HDL and MS. However, we found only the association between PFNA and adiponectin but not between PFNA and decreased insulin and insulin resistance. Insulin resistance, type II diabetes, obesity, and myocardial infarction are inversely related to increasing plasma adiponectin level.38 The data presented would suggest that increased PFNA levels in the study population might be inversely associated with these pathological conditions because it causes plasma adiponectin level to increase. In subgroup analysis, the association between serum PFNA and adiponectin was more evident in subjects with male gender, older age, and high insulin resistance. Serum adiponectin concentrations reveal a sexual dimorphism developed during the progression of puberty, with females having higher levels than males.39 BMI was also signiﬁcantly related to circulating levels.39 It is possible that the relationship between PFNA and adiponectin was also inﬂuenced by gender and age. Whether these results are a statistical artifact or have a true underlying biological basis needs further study. Our study had several limitations. First, the cross-sectional design does not permit any causal inference. Second, we did not include other environmental chemicals which may be important covariates or explanatory variables aﬀecting the outcomes of our study. Third, since our study population was adolescents and young adults, the same association might not be found in older adults. Finally, a common physiology could inﬂuence both serum PFCs and adiponectin levels rather than exposure aﬀecting outcome. In conclusion, higher serum PFNA concentration was associated with higher serum adiponectin concentration in a sample of adolescents and young adults in Taiwan. Although the relationship between PFNA and adiponectin has little biological signiﬁcance and the eﬀect of this relationship on health is uncertain in the Taiwan population, our data suggest that it would be prudent to monitor the adiponectin level of people with low, environmentally relevant exposure to PFNA. Further studies are needed to conﬁrm these ﬁndings and to clarify whether these associations are causal.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional tables. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +886-2-3322-8088 (P.-C.C); +886-2-23123456-66719 (T.-C.S.). Fax: +886-2-358-2402 (P.-C.C.); +886-2-23712361 (T.-C.S.). E-mail: [email protected] (P.-C.C.); tachensu@ ntu.edu.tw (T.-C.S.).

’ ACKNOWLEDGMENT This study was supported by grants from the National Health Research Institute of Taiwan (NHRI-EX97-9721PC, EX979821PC, and X97-9921PC) and (NHRI-EX95-9531PI, EX959631PI, and EX95-9731PI) and from the National Science Council of Taiwan (99-2314-B-385-001-MY3). No funding 10696

dx.doi.org/10.1021/es201964x |Environ. Sci. Technol. 2011, 45, 10691–10698

Environmental Science & Technology organization or sponsor played any part in the design or conduct of the study; in the analysis or interpretation of the data; or preparation, review, or approval of the manuscript. The authors declare no competing interests.

’ REFERENCES (1) Kudo, N.; Bandi, N.; Suzuki, E.; Katakura, M.; Kawashima, Y. Induction by perﬂuorinated fatty acids with diﬀerent carbon chain length of peroxisomal b-oxidation in the liver of rats. Chem. Biol. Interact. 2000, 124, 119–132. (2) Guruge, K. S.; Yeung, L. W. Y.; Yamanaka, N.; Miyazaki, S.; Lam, P. K. S.; Giesy, J. P.; Jones, P. D.; Yamashita, N. Gene expression proﬁles in rat liver treated with perﬂuorooctaonic acid (PFOA). Toxicol. Sci. 2006, 89, 93–107. (3) Hines, E. P.; White, S. S.; Stanko, J. P.; Gibbs-Flournoy, E. A.; Lau, C.; Fenton, S. E. Phenotypic dichotomy following developmental exposure to perﬂuorooctanoic acid (PFOA) in female CD-1 mice: Low doses induce elevated serum leptin and insulin, and overweight in midlife. Mol. Cell. Endocrinol. 2009, 304, 97–105. (4) Loveless, S. E.; Finlay, C.; Everds, N. E.; Frame, S. R.; Gillies, P. J.; O’Connor, J. C.; Powley, C. R.; Kennedy, G. L. Comparative responses of rats and mice exposed to linear/branched, linear, or branched ammonium perﬂuorooctanoate (APFO). Toxicology 2006, 220, 203–217. (5) Olsen, G. W.; Burris, J. M.; Burlew, M. M.; Mandel, J. H. Epidemiologic assessment of worker serum perﬂuorooctanesulfonate (PFOS) and perﬂuorooctanoate (PFOA) concentrations and medical surveillance examinations. J. Occup. Environ. Med. 2003, 45, 260–270. (6) Olsen, G. W.; Burlew, M. M.; Marshall, J. C.; Burris, J. M.; Mandel, J. H. Analysis of episodes of care in a perﬂuorooctanesulfonyl ﬂuoride production facility. J. Occup. Environ. Med. 2004, 46, 837–846. (7) Olsen, G. W.; Zobel, L. R. Assessment of lipid, hepatic, and thyroid parameters with serum perﬂuorooctanoate (PFOA) concentrations in ﬂuorochemical production workers. Int. Arch. Occup. Environ. Health. 2007, 81, 231–246. (8) Sakr, C. J.; Kreckmann, K. H.; Green, J. W.; Gillies, P. J.; Reynolds, J. L.; Leonard, R. C. Cross-sectional study of lipids and liver enzymes related to a serum biomarker of exposure (ammonium perﬂuorooctanoate or APFO) as part of a general health survey in a cohort of occupationally exposed workers. J. Occup. Environ. Med. 2007, 49, 1086–1096. (9) Sakr, C. J.; Leonard, R. C.; Kreckmann, K. H.; Slade, M. D.; Cullen, M. R. Longitudinal study of serum lipids and liver enzymes in workers with occupational exposure to ammonium perﬂuorooctanoate. Occup. Environ. Med. 2007, 49, 872–879. (10) Steenland, K.; Tinker, S.; Frisbee, S.; Ducatman, A.; Vaccarino, V. Association of perﬂuorooctanoic acid and perﬂuorooctane sulfonate with serum lipids among adults living near a chemical plant. Am. J. Epidemiol. 2009, 170, 1268–1278. (11) Frisbee, S. J.; Shankar, A.; Knox, S. S.; Steenland, K.; Savitz, D. A.; Fletcher, T.; Ducatman, A. M. Perﬂuorooctanoic cid, perﬂuorooctanesulfonate, and serum lipids in children and adolescents: results from the c8 health project. Arch. Pediatr. Adolesc. Med. 2010, 164, 860–869. (12) Steenland, K.; Tinker, S.; Shankar, A.; Ducatman, A. Association of perﬂuorooctanoic acid (PFOA) and perﬂuorooctanesulfonate (PFOS) with uric acid among adults with elevated community exposure to PFOA. Environ. Health Prospect. 2010, 118, 229–233. (13) Nelson, J.; Hatch, E.; Webster, T. Exposure to polyﬂuoroalkyl chemicals and cholesterol, body weight, and insulin resistance in the general U.S. population. Environ. Health Perspect. 2010, 118, 197–202. (14) MacNeil, J.; Steenland, N. K.; Shankar, A.; Ducatman, A. A cross-sectional analysis of type II diabetes in a community with exposure to perﬂuorooctanoic acid (PFOA). Environ. Res. 2009, 109, 997–1003. (15) Lin, C. Y.; Chen, P. C.; Lin, Y. C.; Lin, L. Y. Association among serum perﬂuoroalkyl chemicals, glucose homeostasis and metabolic syndrome in adolescents and adults. Diabetes Care 2009, 32, 702–707.

ARTICLE

(16) Matsuzawa, Y.; Funahashi, T.; Kihara, S.; Shimomura, I. Adiponectin and metabolic syndrome. Arterioscler., Thromb., Vasc. Biol. 2004, 24, 29–33. (17) Rasouli, N.; Yao-Borengasser, A.; Miles, L. M.; Elbein, S. C.; Kern, P. A. Increased plasma adiponectin in response to pioglitazone does not result from increased gene expression. Am. J. Physiol.: Endocrinol. Metab. 2006, 290, E42–E46. (18) Wei, J. N.; Sung, F. C.; Lin, C. C.; Lin, R. S.; Chiang, C. C.; Chuang, L. M. National surveillance for type 2 diabetes mellitus in Taiwanese children. JAMA, J. Am. Med. Assoc. 2003, 290, 1345–1350. (19) Tsai C. W.; Kuo, C. C.; Wu, C. F.; Chien, K. L.; Wu, V. C.; Chen, M. F.; Sung, F. C.; Su. T. C. Associations of renal vascular resistance with albuminuria in adolescents and young adults. Nephrol., Dial., Transplant. 2011 Mar 28. [Epub ahead of print]. (20) Chu, C. S.; Lee, K. T.; Lee, M. Y.; Su, H. M.; Voon, W. C.; Sheu, S. H.; Lai, W. T. Eﬀects of rosiglitazone alone and in combination with atorvastatin on nontraditional markers of cardiovascular disease in patients with type 2 diabetes mellitus. Am. J. Cardiol. 2006, 97, 646–650. (21) Drzewoski, J.; Zurawska-Klis, M. Eﬀect of gliclazide modiﬁed release on adiponectin, interleukin-6, and tumor necrosis factor-alpha plasma levels in individuals with type 2 diabetes mellitus. Curr. Med. Res. Opin. 2006, 22, 1921–1926. (22) Weitzman, M.; Cook, S.; Auinger, P.; Florin, T. A.; Daniels, S.; Nguyen, M.; Winickoﬀ, J. P. Tobacco smoke exposure is associated with the metabolic syndrome in adolescents. Circulation 2005, 112, 862–869. (23) Wallace, T. M.; Levy, J. C.; Matthews, D. R. Use and abuse of HOMA modeling. Diabetes Care 2004, 27, 1487–1495. (24) Goodfriend, T. L.; Sowers, J. R.; Messerli, F. H.; Cutler, J. A.; Sheps, S. G.; Vidt, D. G. Hypertension Primer: The Essential of High Blood Pressure; American Heart Association : Dallas, 2003. (25) Lien, G. W.; Wen, T. W.; Hsieh, W. S.; Wu, K. Y.; Chen, C. Y.; Chen, P. C. Analysis of perﬂuorinated chemicals in umbilical cord blood by ultra-high performance liquid chromatography/tandem mass spectrometry. J. Chromatogr., B: Analyt. Technol. Biomed. Life. Sci. 2011, 879, 641–646. (26) Grundy, S. M.; Brewer, H. B.; Cleeman, J. I.; Smith, S. C.; Lenfant, D. Deﬁnition of metabolic syndrome: report of the National, Heart, Lung, and Blood Institute/American Heart Association conference on scientiﬁc issues related to deﬁnition. Circulation 2004, 109, 433–438. (27) Jolliﬀe, C. J.; Janssen, I. Development of age-speciﬁc adolescent metabolic syndrome criteria that are linked to the adult treatment panel III and international diabetes federation criteria. J. Am. Coll. Cardiol. 2007, 49, 891–898. (28) Cnop, M.; Havel, P. J.; Utzschneider, K. M.; Carr, D. B.; Sinha, M. K.; Boyko, E. J.; Retzlaﬀ, B. M.; Knopp, R. H.; Brunzell, J. D.; Kahn, S. E. Relationship of adiponectin to body fat distribution, insulin sensitivity and plasma lipoproteins: evidence for independent roles of age and sex. Diabetologia 2003, 46, 459–469. (29) Abbasi, F.; Farin, H. M.; Lamendola, C.; McLaughlin, T.; Schwartz, E. A.; Reaven, G. M.; Reaven, P. D. The relationship between plasma adiponectin concentration and insulin resistance is altered in smokers. J. Clin. Endocrinol. Metab. 2006, 91, 5002–5007. (30) Guebre-Egziabher, F.; Bernhard, J.; Funahashi, T.; Hadj-Aissa, A.; Fouque, D. Adiponectin in chronic kidney disease is related more to metabolic disturbances than to decline in renal function. Nephrol., Dial., Transplant 2005, 20, 129–134. (31) Calafat, A. M.; Wong, L. Y.; Kuklenyik, Z.; Reidy, J. A.; Needham, L. L. Polyﬂuoroalkyl chemicals in the U.S. population: Data from the National Health and Nutrition Examination Survey (NHANES) 20032004 and comparisons with NHANES 19992000. Environ. Health Perspect. 2007, 115 (11), 1596–1602. (32) Lin, A. Y.; Panchangam, S. C.; Lo, C. C. The impact of semiconductor, electronics and optoelectronic industries on downstream perﬂuorinated chemical contamination in Taiwanese rivers. Environ. Pollut. 2009, 157 (4), 1365–1372. (33) Fang, X.; Zhang, L.; Feng, Y.; Zhao, Y.; Dai, J. Immunotoxic eﬀects of perﬂuorononanoic acid on BALB/c mice. Toxicol. Sci. 2008, 105, 312–321. 10697

dx.doi.org/10.1021/es201964x |Environ. Sci. Technol. 2011, 45, 10691–10698

Environmental Science & Technology

ARTICLE

(34) Paumelle, R.; Staels, B. Cross-talk between statins and PPARalpha in cardiovascular diseases: clinical evidence and basic mechanisms. Trends. Cardiovasc. Med. 2008, 18, 73–78. (35) Klaunig, J. E.; Babich, M. A.; Baetcke, K. P.; Cook, J. C.; Corton, J. C.; David, R. M.; DeLuca, J. G.; Lai, D. Y.; McKee, R. H.; Peters, J. M.; Roberts, R. A.; Fenner-Crisp, P. A. PPAR agonist-induced rodent tumors: modes of action and human relevance. Crit. Rev. Toxicol. 2003, 33, 655–780. (36) Takacs, M. L.; Abbott, B. D. Activation of mouse and human peroxisome proliferator-activated receptors (alpha, beta/delta, gamma) by perﬂuorooctanoic acid and perﬂuorooctane sulfonate. Toxicol. Sci. 2007, 95, 108–117. (37) Abbott, B. D. Review of the expression of peroxisome proliferator-activated receptors alpha (PPAR alpha), beta (PPAR beta), and gamma (PPAR gamma) in rodent and human development. Reprod. Toxicol. 2009, 27, 246–257. (38) Li, S.; Shin, H. J.; Ding, E. L.; Van Dam, R. M. Adiponectin levels and risk of type 2 diabetes: a systematic review and meta-analysis. JAMA, J. Am. Med. Assoc. 2009, 302, 179–188. (39) B€ottner, A.; Kratzsch, J.; M€uller, G.; Kapellen, T. M.; Bl€uher, S.; Keller, E.; Bl€uher, M.; Kiess, W. Gender diﬀerences of adiponectin levels develop during the progression of puberty and are related to serum androgen levels. J. Clin. Endocrinol. Metab. 2004, 89, 4053–4061.

10698

dx.doi.org/10.1021/es201964x |Environ. Sci. Technol. 2011, 45, 10691–10698

ARTICLE pubs.acs.org/est

Bacteriophage Lysis of Enterococcus Host Strains: A Tool for Microbial Source Tracking? Sarah E. Purnell,* James E. Ebdon, and Huw D. Taylor Environment & Public Health Research Unit, School of Environment and Technology, Cockcroft Building, University of Brighton, Lewes Road, Brighton, BN2 4GJ, United Kingdom ABSTRACT: This paper describes the isolation of Enterococcus host strains, for potential use as simple bacteriophage (phage)-based microbial source tracking (MST) tools. Presumptive Enterococcus host strains were isolated from cattle feces, raw municipal wastewater, agricultural runoﬀ, and waters impacted by farms or wastewater treatment works (WWTW) in southern England, United Kingdom (UK). All enterococcal host strains (n = 390) were ﬁrst screened for their ability to detect phage in samples of raw municipal wastewater and fecal material from cattle, pigs, and sheep. Host strains that detected phage (n = 147) were ranked according to both their speciﬁcity to a particular fecal source and also the number of phages (expressed as plaqueforming units, PFU) that they detected per milliliter of sample. Host strains that demonstrated host speciﬁcity and which detected phages at levels greater than 100 PFU/mL (n = 29) were further tested using additional fecal samples of human and nonhuman origin. The speciﬁcity and sensitivity of the enterococcal host strains were found to vary, ranging from 44 to 100% and from 17 to 83%, respectively. Most notably, seven strains exhibited 100% speciﬁcity to either cattle, human, or pig samples. Isolates exhibiting speciﬁcity to cattle were identiﬁed as belonging to the species Enterococcus casseliﬂavus, Enterococcus mundtii, or Enterococcus gallinarum, while human and pig isolates were members of either Enterococcus faecium or Enterococcus faecalis. The high speciﬁcity of phages infecting Enterococcus hosts and the simplicity and relatively low cost of the approach collectively indicate a strong potential for using this method as a tool in MST.

’ INTRODUCTION Contamination of surface waters with feces of human and nonhuman (domestic, agricultural, and wild animal) origin leads to increased public health risk of exposure to pathogens through drinking water supply, aquaculture, and recreational activities.1 3 Of the two, human fecal contamination is considered to be the greater risk to public health, largely because viruses that commonly cause illness are highly host speciﬁc, but feces from nonhuman sources also pose a potential risk of infection by zoonotic pathogens, such as Escherichia coli O157, Giardia spp., Campylobacter spp., and Cryptosporidium spp.4 Recent research ﬁndings have suggested that the risk to human health attributable to contamination by nonhuman fecal sources is varied and, speciﬁcally, that the risk to human health from fresh cattle feces in recreational waters is not substantially diﬀerent than that from human fecal contamination.5 The Clean Water Act (CWA) and the Water Framework Directive (WFD) regulate surface water quality standards in the United States (US) and the European Union (EU), respectively.6,7 These legislative measures require the identiﬁcation and management of those point and diﬀuse sources of microbial pollution that lead to the impairment (US) and noncompliance (EU) of surface waters. In the US, impaired waters are investigated and actions for remediation are set out by the principle of “total maximum daily loads” (TMDL). TMDL deﬁne the quantity of pollutant that a water body can tolerate while adhering to water r 2011 American Chemical Society

quality standards.6 Similarly, the WFD in the EU requires the establishment of a “program of measures” for identiﬁed river basin districts, in order to achieve a “good status” by 2015.7 Water quality standards are monitored using fecal indicator organisms (FIO), typically intestinal enterococci and E. coli, as it remains technically complicated and expensive to monitor pathogens directly.8 FIO concentrations oﬀer an indication of the presence of fecal contamination and the potential risk to public health, but they do not determine contamination sources. Since certain fecal sources pose a greater risk to human health than others, reliance on FIO concentrations alone could result in either an underestimation or overestimation of the potential risk. Legislative requirements, the need to predict risk to human health, and the implications of source on the selection of appropriate remediation measures, have led to the development of the ﬁeld of microbial source tracking (MST). MST encompasses techniques that aim to distinguish source(s) of fecal contamination in surface and ground waters. MST is a rapidly advancing ﬁeld, but there is currently no single standardized method available to distinguish sources of fecal contamination in surface waters in all situations. 9,10 Received: June 23, 2011 Accepted: November 2, 2011 Revised: October 18, 2011 Published: November 02, 2011 10699

dx.doi.org/10.1021/es202141x | Environ. Sci. Technol. 2011, 45, 10699–10705

Environmental Science & Technology Simple, low-cost, and eﬀective phage-based MST techniques involving the detection and enumeration of phages infecting Bacteroides spp. and the subgroups of F-speciﬁc RNA phage have been used to discriminate human from nonhuman fecal contamination.11,12 Recent research has shown that the inactivation proﬁles of F-speciﬁc RNA phage subgroups diﬀer, with phages of groups III and IV being less resistant to inactivation by natural stressors, thus hindering the interpretation of results when using F-speciﬁc phage in MST.13 Host strains of Bacteroides spp. that are demonstrably restricted to human and nonhuman feces have been isolated from diﬀerent geographical regions14 17 and have been used successfully to distinguish human and nonhuman fecal contamination of surface waters. An advantage of detecting bacteriophage (phage) of Bacteroides spp., as opposed to a phage infecting aerobic or facultatively anaerobic hosts such as E. coli (somatic coliphage), is that the host bacterium requires strict anaerobic conditions and is therefore unlikely to replicate outside the gut environment. However, recent research into the ecology of somatic coliphage, suggests that any replication of their bacterial host in the natural environment is unlikely to have a signiﬁcant impact on the numbers of phage detected.18 With this in mind, it may be advantageous to use an alternative host genus that does not need to be handled under strict anaerobic conditions and which has simpler growth requirements. The intestinal enterococci are an “indicator of choice” in many parts of the world for determining water quality, and phages capable of infecting Enterococcus faecalis have already been proposed as a potential alternative indicator of human fecal contamination.19 Phage-based MST methods have tended to focus on identifying human fecal pollution, but the need to identify diﬀuse nonhuman fecal contamination in order to meet the stipulations of the CWA and WFD is clear. The objectives of this study were therefore (1) to evaluate the ability of phages infecting Enterococcus host strains to identify human and speciﬁc nonhuman sources of fecal contamination and (2) to establish an eﬀective protocol for isolating new Enterococcus host strains suitable for MST application in other parts of the world.

’ MATERIALS AND METHODS Sample Collection. Samples of animal feces, cattle and pig runoff, wastewater, and surface water were collected for the isolation of potential Enterococcus host strains and for subsequent bacteriophage detection. Fecal material from either pooled individual scats (at least 20 individuals) or agricultural runoff was collected from cattle, ducks, geese, goats, horses, pigs, rabbits, seagulls, and sheep on 25 occasions, from 2008 to 2010, using sterile swabs and 100 mL sampling containers (Nalgene) as appropriate. Previous research has shown that not all individuals in a population will excrete selected phage and that the titers may vary,20 so to obtain a sample representative of a population, all individual scat samples were pooled so that they included feces from at least 20 different animals. In the laboratory, scat samples were mixed and homogenized using a Seward Stomacher 400 (Lab System) in sterile onequarter strength Ringer’s solution (Fisher Scientiﬁc). Fecally contaminated runoﬀ water was collected from cattle housing and from the drainage pipes of pig housing at Wales Farm, Plumpton Agricultural College, in East Sussex, UK. Samples of wastewater (raw and treated) were collected from seven biological wastewater treatment works (WWTW) situated in South East England, UK. Population equivalents of the sites ranged from 258 to 53 425. Surface water samples were collected downstream of two

ARTICLE

livestock farms (Wales Farm and Pellingbridge Farm) and downstream of a medium sized WWTW (population equivalent, 55 955). All surface water and wastewater samples were collected in 1 L sterile sample bottles (Nalgene) using an extendable sampling pole, from approximately 30 cm below the surface of the water. Following collection, samples were transported to the laboratory in the dark at 4 °C within 4 h. Samples collected for the isolation of enterococci were processed immediately, whereas samples for phage enumeration were preserved with 10% glycerol at 20 °C21 for analysis within 4 weeks. Isolation of Presumptive Enterococcus Host Strains. Potential host strains were obtained from pooled individual cattle scats (approximately 25), cattle and pig fecal runoff, municipal wastewater, and surface waters impacted by human and nonhuman sources of feces in southern England, UK, and a dilution series was prepared. These samples were then passed through 0.45 μm membrane filters for enumeration of presumptive enterococci in accordance with ISO 7899/2.22 m-Enterococcus Agar (Becton Dickinson Microbiology Systems) was used as a selective medium for the isolation of presumptive enterococci and was incubated at 37 °C ((2 °C) for 44 h ((2 h). Following incubation, filter membranes (Thermo Scientific Nalgene) with between 10 and 60 colony forming units (CFU) were transferred to prewarmed bile esculin agar (Oxoid) plates and incubated for a further 4 h at 44 °C. Distinct colonies demonstrating esculin hydrolysis (blackening of the media) were picked and streaked onto m-Enterococcus Agar in order to obtain pure cultures. Further identification was achieved by undertaking catalase and Gram stain tests (in which morphology was also recorded). Presumptive Enterococcus host strains (Gram positive, coccoid, catalase negative, esculin positive) were then grown at 37 °C ((2 °C) for 24 h ((2 h) in tryptone soya broth (TSB) (Oxoid), mixed with 50% glycerol, and preserved at 80 °C for up to 6 months prior to further testing. Bacteriophage Enumeration. Phages infecting Enterococcus hosts were enumerated using a previously described and ISO standardized double-agar-layer23 method, and the results were expressed as plaque-forming units (PFU) per 100 mL of sample. Different media were tested for the top and bottom agar layers, including m-Enterococcus Agar, nutrient agar, KF Streptococcus Agar, and tryptone soya agar (TSA) (Oxoid). The clearest plaques were obtained using TSA for both layers, so TSA was therefore used for all subsequent phage assays. This observation is in accordance with results from a previous study that focused on phage of E. faecalis.19 The concentrations of agar in top and bottom layers used were the same as those for standardized methods reported elsewhere (ISO 10705/2).24 Homogenized fecal samples were diluted (1:10 w/v) with onequarter strength Ringer’s solution and centrifuged at 3000g for 10 min. The resulting supernatants were ﬁltered through 0.22 μm polyvinylidene diﬂouride membrane syringe ﬁlter units (Millipore). A 1 mL portion of each sample was added to 1 mL of exponentially growing host strain and 2.5 mL of tryptone semisolid agar. The resulting suspension was mixed brieﬂy using a Whirlimixer (Fisher Scientiﬁc) and poured onto previously prepared TSA bottom agar layers in 90 mm diameter Petri plates. Once solidiﬁed, plates were inverted and incubated at 37 °C ((2 °C) for 18 24 h. All fecal samples were tested in triplicate. Clearly visible circular “zones of lysis” in a conﬂuent lawn of enterococcal host were recorded as PFU (Figure 1). Isolate Screening. A tiered approach was designed and implemented in order to reduce in a rational manner the initial large 10700

dx.doi.org/10.1021/es202141x |Environ. Sci. Technol. 2011, 45, 10699–10705

Environmental Science & Technology

Figure 1. Enterococcal host strain Newhaven River Ouse 30 (NRO30), demonstrating clear plaques from a duck and goose mixed fecal sample.

ARTICLE

Tier 2 hosts that showed speciﬁcity to a particular fecal source or source group (tier 3 hosts) and which also detected phage in greater numbers than 100 PFU per 100 mL (tier 4 hosts) were considered to be the host stains with the greatest potential for MST application. Host strains detecting higher concentrations of phage in fecal samples (tier 4 host strains) have greater potential for MST application, because they are more likely to detect phage in surface waters, where phage are likely to be present in much lower numbers. All 25 tier 4 hosts, as well as four tier 3 hosts demonstrating 100% speciﬁcity and clear plaque production, were then subjected to further rigorous testing using inputs representative of those present across the whole study catchment. Host Identification. Host strains meeting the criteria for specificity, sensitivity, and plaque clarity were identified to species level. This process was carried out using the API 20 Strep identification system (BioMerieux) according to the manufacturer’s instructions, and a six step biochemical key described by Manero and Blanch.25 To aid identification using API 20 Strep, tests for yellow pigmentation were also performed. Production of yellow pigmentation was demonstrated by growing the enterococcal hosts on nutrient agar and incubating for 24 h at 37 °C. Following incubation, host strains were checked for yellow pigmentation against a white filter paper. Statistical Analysis. Performance of the enterococcal strains as potential hosts for MST was evaluated in relation to their specificity and sensitivity to a particular fecal source. Specificity and sensitivity were calculated using standard definitions26 and expressed as a percentage. The Spearman’s rank correlation coeﬃcient was used to test the hypothesis that a negative relationship between speciﬁcity and sensitivity existed. The Kruskal Wallis test was used to determine variation of speciﬁcity and phage numbers detected on host strains obtained from diﬀerent sources. Where variation was signiﬁcant, post hoc Mann Whitney tests were used to determine where variation occurred. Statistical tests were performed using the statistical package SPSS 16.0, with the signiﬁcance level set at 5%.

’ RESULTS Figure 2. Tiered approach to the isolation of Enterococcus host strains.

number of enterococcal hosts to a smaller subgroup that would be suitable for phage enumeration and MST (Figure 2). This approach provided a protocol that could rapidly eliminate enterococcal strains that would not be effective hosts and focus efforts on those host strains warranting further investigation. Tier 1 hosts were those strains conﬁrmed as being presumptive Enterococcus spp. that grew well at 37 °C ((2 °C) for 24 h ((2 h) in TSB. All tier 1 host strains were screened against the same battery of reference samples containing phage from seven municipal wastewater samples and ﬁve pooled cattle, sheep, and pig fecal samples. Those enterococcal strains that detected one or more PFU per 100 mL (tier 2 hosts) were ranked according to their speciﬁcity (to a particular source) and to the numbers of phage that they detected (PFU/100 mL). Speciﬁcity was given a higher weighting, as this characteristic was considered essential for MST application. It was important that the phage infecting the hosts produced clear well-deﬁned plaques (Figure 1). Hosts that demonstrated plaques that were unclear, and therefore diﬃcult to identify and enumerate, were not assessed further.

Host Strain Screening. In total, 554 potential enterococcal hosts were isolated, and 390 were confirmed as tier 1 hosts (Table 1). Thirty-eight percent of hosts detected phage in reference samples from cattle, pig, sheep, and raw municipal wastewater (tier 2 hosts). A high percentage of host strains (67% and 68%, respectively) isolated from cattle and pig runoff detected phage in reference samples, but far fewer host strains isolated from both municipal wastewater and surface waters (34% and 35%, respectively) and only one out of 76 hosts obtained from pooled cattle scats detected phage in any of the four reference samples. A sum of 117 tier 2 hosts was restricted to one fecal source or source group (tier 3 hosts) and 25 tier 3 hosts detected phages in numbers greater than 1.0 104 per 100 mL of sample (tier 4 hosts). Sixty-eight percent of tier 4 hosts originated from cattle runoff, 24% from surface waters, 4% from pig runoff, and 4% from raw municipal wastewater. Finally, 25 tier 4 hosts and four additional tier 3 hosts (investigated further because they had 100% specificity and excellent plaque clarity) underwent further investigation, and of the 29 enterococcal hosts, just over 48% were found to be highly specific to a particular source or source group. Our screening process therefore offers a rapid 10701

dx.doi.org/10.1021/es202141x |Environ. Sci. Technol. 2011, 45, 10699–10705

Environmental Science & Technology

ARTICLE

Table 1. Assignment of Potential Enterococcus Host Strains from Various Sources According to the Four Diﬀerent Tier Categories no. of strains host origin

tier 1a

no. of samples

tier 2b(%)

tier 3c (%)

tier 4d (%)

pooled fecal scats from cattle

1

76

1 (1)

1 (1)

0 (0)

liquid runoﬀ from cattle

2

83

56 (67)

52 (63)

17 (20)

liquid runoﬀ from pig

1

31

21 (68)

12 (39)

1 (3)

municipal wastewater

1

112

38 (34)

29 (26)

1 (1)

impacted surface waters

3

88

31 (35)

23 (26)

6 (7)

total

8

390

147 (38)

117 (30)

25 (6)

Tier 1: presumptive Enterococcus (Gram positive, catalase negative, esculin positive, and exhibiting good growth after 24 h at 37 °C in TSB). b Tier 2: tier 1 strains that detect bacteriophage >0 PFU/mL in human, cattle, pig, or sheep reference fecal samples. c Tier 3: tier 2 strains that show potential speciﬁcity to human, cattle, pig, sheep, or animal reference fecal samples. d Tier 4: tier 3 strains detecting bacteriophage in reference fecal samples >100 PFU/mL. a

Table 2. Number of Plaque Forming Units (PFU/100 mL) Detected by Tier 4 Presumptive Enterococcus Host Strains in Pooled Fecal Samples from Diﬀerent Origins PFU/100 mL for samples of diﬀerent origins (no. of samples) host origin

a

host ID

municipal wastewater (n = 7)

cattle runoﬀ (n = 5)

pig runoﬀ (n = 5)

sheep feces (n = 5)

cattle runoﬀ

CR1

0

1.7 104

0

0

cattle runoﬀ

CR4

0

1.4 104

0

0

cattle runoﬀ

CR6

0

2.1 104

0

0

cattle runoﬀ

CR7

0

1.3 104

0

0

cattle runoﬀ

CR15

0

8.3 104

0

0

cattle runoﬀ cattle runoﬀ

CR16 CR17

0 0

4.4 104 7.1 104

0 0

0 0

cattle runoﬀ

CR37

0

1.2 105

0

0

cattle runoﬀ

CR45

0

1.8 104

6.0 102

0

cattle runoﬀ

CR47

0

4.8 104

1.2 103

0

cattle runoﬀ

CR51

0

5.2 10

2.0 103

0

cattle runoﬀ

CR58

0

1.9 104

0

2.0 102

4

cattle runoﬀ

CR61

0

2.2 10

1.7 10

0

cattle runoﬀ cattle runoﬀ

CR63 CR70

0 0

3.4 104 7.3 104

0 1.0 103

0 0

cattle runoﬀ

CR73

0

2.3 104

2.1 103

0

4

3

cattle runoﬀ

CR75

0

4.8 10

1.0 102

3.0 102

municipal wastewater

MW42a

2.7 103

0

0

0

municipal wastewater

MW47a

3.4 103

0

0

0

municipal wastewater

MW96

0

0

>2.0 105

0

Newhaven (River Ouse)

NRO5a

0

0

9.8 103

0

Newhaven (River Ouse) Newhaven (River Ouse)

NRO8a NRO20

3.4 103 0

0 1.0 105

0 0

0 0

4

Newhaven (River Ouse)

NRO24

0

0

1.1 105

0

Newhaven (River Ouse)

NRO30

0

1.2 104

0

0

Newhaven (River Ouse)

NRO39

0

2.5 104

0

0

pig runoﬀ

PR3

0

2.2 104

1.0 102

0

Pellingbridge (River Ouse)

PRO4

0

2.0 104

0

0

Wales Farm stream

WFS7

0

6.2 104

0

0

Tier 3 host strains with 100% speciﬁcity and excellent plaque clarity, but they detected lower plaque counts (<1.0 104) than tier 4 host strains

approach to determining the suitability of potential hosts for MST application. Specificity vs Sensitivity. Following initial screening, 29 host strains emerged as potentially suitable for MST application (Table 2). The suitability of these enterococcal hosts was further

tested by exposing them to 37 additional samples from 10 source groups (Table 3). The specificity of the host strains ranged from 44 to 100%. For MST purposes it is particularly important that the hosts do not detect phage from both human and animal sources (cross-reaction). Further testing revealed that 15 of the 10702

dx.doi.org/10.1021/es202141x |Environ. Sci. Technol. 2011, 45, 10699–10705

Environmental Science & Technology

ARTICLE

Table 3. Percentage of Phage Positive Samples from Diﬀerent Origins Detected Using Four Enterococcus Host Strains % phage positive samples from Enterococcus host strains sample origin cattle

municipal wastewater 47

cattle run-oﬀ 70

Wales Farm stream 7

Newhaven River Ouse 24

15

0

67

33

0

chicken

6

0

0

0

0

ducks and geese

6

0

100

0

0

goat

6

0

0

0

0

horse

6

0

0

0

0

15

0

100

0

20

rabbit

6

0

0

0

0

seagull sheep

6 6

0 0

0 0

0 0

0 0

raw MWa

15

100

0

0

0

ﬁnal MWa

12

25

0

0

0

pig

a

no. of samples

MW = municipal wastewater.

Table 4. Identiﬁcation of Enterococcus Hosts Strains with API 20 Strep and a Biochemical Key host origin

a

ID

API 20 Strep identiﬁcation

biochemical key identiﬁcation

cattle runoﬀ cattle runoﬀ

CR1 CR6

E. casseliﬂavus CNIa

E. mundtii E. gallinarum

cattle runoﬀ

CR7

E. casseliﬂavus

E. casseliﬂavus

cattle runoﬀ

CR15

E. casseliﬂavus

E. casseliﬂavus

cattle runoﬀ

CR16

CNIa

E. gallinarum

cattle runoﬀ

CR17

CNIa

E. gallinarum

cattle runoﬀ

CR70

E. gallinarum

E. gallinarum

municipal wastewater

MW42

E. faecalis

E. faecalis

municipal wastewater Newhaven (River Ouse)

MW47 NRO8

E. faecium E. faecalis

E. faecium E. faecalis

Newhaven (River Ouse)

NRO24

E. faecium

E. faecium

Newhaven (River Ouse)

NRO30

E. casseliﬂavus

E. casseliﬂavus

Newhaven (River Ouse)

NRO39

E. gallinarum

E. faecium

Wales Farm stream

WFS7

E. gallinarum

E. gallinarum

CNI = API 20 Strep could not identify.

hosts cross-reacted with animal feces and human wastewater samples, even though they had previously been restricted to either human or nonhuman sources. This ruled them out from any further analysis. At this stage, 14 potential hosts remained specific to a single human or nonhuman source. Notably, seven host strains were 100% specific to a single source (e.g., cattle, pig, or human). Table 3 illustrates the results for four host strains specific to either cattle, pig, human, or mixed fecal sources. Strains WFS7 and NRO24 isolated from surface waters were 100% specific to cattle and pig fecal samples, respectively. However, the sensitivities of WFS7 and NRO24 were only 33% and 20%, respectively, and, interestingly, they appeared to be restricted to samples originating from specific herds present on Wales Farm, Plumpton, UK. Human specific host strain MW47 had a much higher sensitivity, detecting phage in all raw wastewater samples (100%) tested from six WWTW, one of which has a population equivalent of only 258. However, MW47 was only detected in one-fourth of the treated wastewaters tested, suggesting removal or die-off of phage during treatment. Certain host strains such as strain CR70, while demonstrating lower levels of phage specificity (Table 3), had sensitivity levels much higher

than more specific hosts. Strain CR70 was found almost exclusively in cattle and pig samples (90% specificity), had a sensitivity of 83%, and could be useful as an indicator of nonhuman fecal contamination. Spearman’s correlation coeﬃcient was used to test the relationship between speciﬁcity and sensitivity of the 29 promising enterococcal hosts. Results showed a moderate negative relationship between sensitivity and speciﬁcity (Rs = 0.480, p < 0.01). As speciﬁcity increased, the sensitivities of the hosts also tended to decrease. Ideally, the speciﬁcity and sensitivity of a host for phage lysis would be 100% in each case. Host Strain Origin. As shown in Table 1, potentially useful enterococcal hosts were successfully isolated from five samples originating from a range of sources. In order to analyze variations in host specificity and sensitivity, all enterococcal hosts were classified as being members of one of four source groups [(1) pooled cattle feces, (2) cattle and pig fecal runoff, (3) municipal wastewater, or (4) surface waters]. The Kruskal Wallis test suggested that specificity did not vary significantly between host strains isolated from the different source groups (p > 0.05), though statistically significant variations in the number of phage 10703

dx.doi.org/10.1021/es202141x |Environ. Sci. Technol. 2011, 45, 10699–10705

Environmental Science & Technology in each source group were apparent (p < 0.01). Post hoc Mann Whitney tests revealed that numbers of phage capable of infecting host strains isolated from cattle feces were significantly lower when compared to numbers of phage infecting host strains from the other source groups (p < 0.008). Phage numbers detected by enterococcal hosts isolated from wastewater and surface waters did not vary greatly from one another, but significantly higher numbers of phage were detected by host strains isolated from cattle and pig fecal runoff (p < 0.008). Identification of Enterococcal Host Strains. Fourteen potential hosts demonstrating specificity to a single fecal source or source group were identified (Table 4). Human-specific host strains NRO8 and MW42 were both identified as being members of the species E. faecalis, and MW47 was identified as Enterococcus faecium. Pig-specific host strain (NRO24) was also identified as E. faecium. All cattle specific host strains were identified as Enterococcus mundtii, Enterococcus casseliflavus, or Enterococcus gallinarum. E. casseliflavus and E. gallinarum are commonly isolated from environmental samples, but it has been suggested that these species could be incorporated into the microbiota of the digestive tract following ingestion, and that their presence could therefore be related to diet.27 The species classification of the host strains CR1 and NRO39 differed depending on which of the two identification methods was used. Host strains CR1 and NRO39 were identified as being members of the species E. casseliflavus and E. gallinarum, respectively, by API 20 Strep, whereas the simplified biochemical key of Manero and Blanch25 identified the hosts as being members of the species Enterococcus mundtii and E. faecium, respectively. API 20 Strep failed to provide identification for three isolates, all of which were identified as E. gallinarum by the biochemical key.

’ DISCUSSION The high speciﬁcity of phages infecting Enterococcus hosts (even down to a speciﬁc herd level) as witnessed in this study suggests that Enterococcus hosts have a potential role as MST tools. However, the lower level of sensitivity associated with high speciﬁcity represents a potential problem. Therefore, strains with lower speciﬁcity but with higher sensitivity (such as CR70) may prove more useful for MST applications than more speciﬁc strains (such as WFS7) in surface waters. Ideally a toolbox approach to the detection of human and nonhuman sources, utilizing several strains in parallel, may be advisible for future MST studies. These results also suggest that, when isolating potential enterococcal host strains for phage-lysis based MST applications, it is better to isolate agricultural hosts directly from liquid farmyard runoﬀ, rather than from fresh pooled scats from individual animals. Isolation of enterococcal hosts directly from impacted surface waters also led to the discovery of strains useful for MST application, without negatively impacting host speciﬁcity. This may be due to the selection of more environmentally tolerant phages and enterococcal host strains. The application of the tiered screening approach simpliﬁed and improved the eﬃciency of isolating new host strains and serves as a useful protocol for future isolation of phage hosts. At least one potential host strain was isolated for every sample analyzed (approximately one host strain for every 40 isolates), with the exception of the pooled cattle fecal sample. Although the geographical stability of the host strains isolated in this study was not determined, not all host strains used in phage-based MST techniques have shown geographical stability.14 We therefore

ARTICLE

suggest that the approach presented here can be applied in other countries to isolate host strains suitable for particular regions. The cost of isolating and screening new Enterococcus host strains is considerably lower than that of developing molecular MST techniques, particularly those that are dependent on the construction of a library or database.18 Culture-based phage MST techniques are relatively simple to carry out, do not require specialist expertise, and can be performed in laboratories equipped with basic microbiological apparatus.17 In this study, the use of Enterococcus host strains also proved rapid with visualization of plaques possible within 4 h. In conclusion, the ﬁndings of this study are signiﬁcant in that they oﬀer an insight into host-phage interactions, speciﬁcity, sensitivity, and suitability of phages infecting diﬀerent Enterococcus strains for MST application. The high speciﬁcity of the enterococcal hosts isolated in this study demonstrates that phages infecting strains of Enterococcus spp. possess narrow host ranges similar to those previously reported for anaerobes such as Bacteroides spp. An eﬀective protocol for the isolation of new enterococcal host strains suitable for MST application has been presented here which may be eﬀectively used by others to isolate bacterial hosts for phage-lysis MST work in other parts of the world. Our relatively simple, rapid, and nonmolecular laboratory protocol (utilizing existing ISO phage methods), which is based on the use of a globally used fecal indicator bacterium, has considerable potential to support the identiﬁcation of fecal contamination sources in both recreational waters and sources of raw drinking water. Our approach therefore has the potential to support Water Safety Plans and hence to help sustain global eﬀorts to reduce the burden of waterborne disease in low-income countries.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +44 1273 643455; e-mail: [email protected].

’ ACKNOWLEDGMENT This work was partly funded by the European Regional Development Fund Interreg IVA Program as part of the collaborative project AquaManche. The authors thank Southern Water (UK) for providing wastewater samples. ’ REFERENCES (1) Craun, G. F.; Brunkard, J. M.; Yoder, J. S.; Roberts, V. A.; Carpenter, J.; Wade, T.; Calderon, R. L; Roberts, J. M.; Beach, M. J.; Roy, S. L. Causes of outbreaks associated with drinking water in the United States from 1971 to 2006. Clin. Microbiol. Rev. 2010, 23 (3), 507–+. (2) Le Guyader, F.; Haugarreau, L.; Miossec, L.; Dubois, E.; Pommepuy, M. Three-year study to assess human enteric viruses in shellﬁsh. Appl. Environ. Microbiol. 2000, 66 (8), 3241–3248. (3) Sinclair, R. G.; Jones, E. L.; Gerba, C. P. Viruses in recreational waterborne-disease outbreaks: A review. J. Appl. Microbiol. 2009, 107 (6), 1769–1780. (4) Craun, G. F.; Calderon, R. L.; Craun, M. F. Waterborne outbreaks caused by zoonotic pathogens in the USA. In Waterborne Zoonoses: Identiﬁcation, Causes and Control; Cotruvo, J. A., Dufour, A., Rees, G., Bartram, J., Carr, R., Cliver, D. O., Craun, G. F., Fayer, R., Gannon, V. P. G., Eds.; World Health Organization, IWA Publishing: London, 2004; pp 120 135. (5) Soller, J. A.; Schoen, M. E.; Bartrand, T.; Ravenscroft, J. E.; Ashbolt, N. J. Estimated human health risks from exposure to 10704

dx.doi.org/10.1021/es202141x |Environ. Sci. Technol. 2011, 45, 10699–10705

Environmental Science & Technology recreational waters impacted by human and non-human sources of faecal contamination. Water Res. 2010, 44 (16), 4674–4691. (6) United States Environmental Protection Agency (USEPA). Federal Water Pollution Control Act of 2002. As amended through P. L. 107 303. http://epw.senate.gov/water.pdf. (7) Commission of the European Union (CEU), Directive 2000/ 60/EC of the European Parliament and of the Council of the 23 October 2000 establishing a framework for community action in the ﬁeld of water policy. Oﬀ. J. Eur. Communities 2000, L327, 1 72. (8) Field, G. K.; Samadpour, M. Fecal source tracking, the indicator paradigm, and managing water quality. Water Res. 2007, 41 (16), 3517–3538. (9) Domingo, J. W. S.; Bambic, D; G.; Edge, T. A.; Wuertz, S. Quo vadis source tracking? Towards a strategic framework for environmental monitoring of fecal pollution. Water Res. 2007, 41 (16), 3539–3552. (10) Meays, C. L.; Broersma, K.; Nordin, R.; Mazumder, A. Source tracking fecal bacteria in water: A critical review of current methods. J. Environ. Manage. 2004, 73 (1), 71–79. (11) Jofre, J.; Bosch, A.; Lucena, F.; Girones, R.; Tartera, C. Evaluation of Bacteroides fragilis bacteriophages as indicators of the virological quality of water. Water Sci. Technol. 1986, 18 (10), 167–173. (12) Gourmelon, M.; Caprais, M. P.; Mieszkin, S.; Marti, R.; Wery, N.; Jarde, E.; Derrien, M.; Jadas-Hecart, A.; Communal, P. Y.; Jaﬀrezic, A.; Pourcher, A. M. Development of microbial and chemical MST tools to identify the origin of the faecal pollution in bathing and shellﬁsh harvesting waters in France. Water Res. 2010, 44 (16), 4812–4824. (13) Maite, M.; Payan, A.; Moce-Llivina, L.; Blanch, A. R.; Jofre, J. Diﬀerential persistence of of F-speciﬁc RNA phage subgroups hinders their use as single tracers for faecal source tracking in surface water. Water Res. 2009, 43 (6), 1559–1564. (14) Payan, A.; Ebdon, J.; Taylor, H.; Gantzer, C.; Ottoson, J.; Papageorgiou, G. T.; Blanch, A. R.; Lucena, F.; Jofre, J.; Muniesa, M. Method for isolation of Bacteroides bacteriophage host strains suitable for tracking sources of fecal pollution in water. Appl. Environ. Microbiol. 2005, 71 (9), 5659–5662. (15) Ebdon, J.; Maite, M.; Taylor, H. The application of a recently isolated strain of Bacteroides (GB-124) to identify human sources of faecal pollution in a temperate river catchment. Water Res. 2007, 41 (16), 3683–3690. (16) Vijayavel, K.; Fujioka, R.; Ebdon, J.; Taylor, H. Isolation and characterization of Bacteroides host strain HB-73 used to detect sewage speciﬁc phages in Hawaii. Water Res. 2010, 44 (12), 3714–3724. (17) Gomez-Do~nate, M.; Payan, A.; Cortes, I.; Blanch, A. R.; Lucena, F.; Jofre, J.; Muniesa, M. Isolation of bacteriophage host strains of Bacteroides species suitable for tracking sources of animal faecal pollution in water. Environ. Microbiol. 2011, 13 (6), 1622–1631. (18) Jofre, J. Is the replication of somatic coliphages in water environments signiﬁcant? J. Appl. Microbiol. 2009, 106 (4), 1059–1069. (19) Bonilla, N.; Santiago, T.; Marcos, P.; Urdaneta, M.; Domingo, J. S.; Toranzos, G. A. Enterophages, a group of phages infecting Enterococcus faecalis, and their potential as alternate indicators of human faecal contamination. Water Sci. Technol. 2010, 61 (2), 293–300. (20) Grabow, W.; Neubrech, T. E.; Holtzhausen, C. S.; Jofre, J. Bacteroides fragilis and Escherichia coli bacteriophages: Excretion by humans and animals. Water Sci Technol. 1995, 31, 223–230. (21) Mendez, J.; Jofre, J.; Lucena, F.; Contreras, N.; Mooijman, K.; Araujo, R. Conservation of phage reference materials and water samples containing bacteriophages of enteric bacteria. J. Virol. Methods 2002, 106 (2), 215–224. (22) Anon. ISO 7899-2: Water Quality. Detection and Enumeration of Intestinal Enterococci—Part 2: Membrane Filtration Method; International Organisation for Standardisation: Geneva, Switzerland, 2000. (23) Adams, M. H. Bacteriophages; Wiley Interscience: New York, 1959. (24) Anon. ISO 10705-2: Water Quality. Detection and Enumeration of Bacteriophages—Part 2: Enumeration of Somatic Coliphages. International Organisation for Standardisation: Geneva, Switzerland, 2001. (25) Manero, A.; Blanch, A. R. Identiﬁcation of Enterococcus spp. with a biochemical key. Appl. Environ. Microbiol. 1999, 65 (10), 4425–4430.

ARTICLE

(26) Gawler, A. H.; Beecher, J. E.; Brandao, J.; Carroll, N. M.; Falcao, L.; Gourmelon, M.; Masterson, B.; Nunes, B.; Porter, J.; Rince, A.; Rodrigues, R.; Thorp, M.; Walters, J. M.; Meijer, W. G. Validation of host-speciﬁc bacteriodales 16S rRNA genes as markers to determine the origin of faecal pollution in Atlantic Rim countries of the European Union. Water Res. 2007, 41 (16), 3780–3784. (27) Layton, B. A.; Walters, S. P.; Lam, L. H.; Boehm, A. B. Enterococcus species distribution among human and animal hosts using multiplex PCR. J. Appl. Microbiol. 2010, 109 (2), 539–547. (28) Malakoﬀ, D. Microbiologists on the trial of polluting bacteria. Science 2002, 295 (5564), 2352–2353.

10705

dx.doi.org/10.1021/es202141x |Environ. Sci. Technol. 2011, 45, 10699–10705

ARTICLE pubs.acs.org/est

Differential Effects in Mammalian Cells Induced by Chemical Mixtures in Environmental Biota As Profiled Using Infrared Spectroscopy Valon Llabjani,† John D. Crosse,‡ Abdullah A. Ahmadzai,† Imran I. Patel,† Weiyi Pang,† Julio Trevisan,†,§ Kevin C. Jones,† Richard F. Shore,‡ and Francis L. Martin†,* †

Centre for Biophotonics, Lancaster Environment Centre, Lancaster University, Bailrigg, Lancaster LA1 4YQ, U.K. NERC Centre for Ecology and Hydrology, Lancaster Environment Centre, Library Avenue, Bailrigg, Lancaster LA1 4AP, U.K. § Department of Communication Systems, Lancaster University, Lancaster LA1 4WA, U.K. ‡

bS Supporting Information ABSTRACT: Environmental contaminants accumulate in many organisms and induce a number of adverse eﬀects. As contaminants mostly occur in the environment as mixtures, it remains to be fully understood which chemical interactions induce the most important toxic responses. In this study, we set out to determine the eﬀects of chemical contaminants extracted from Northern Gannet (Morus bassanus) eggs (collected from the UK coast from three sampling years (1987, 1990, and 1992) on cell cultures using infrared (IR) spectroscopy with computational data handling approaches. Gannet extracts were chemically analyzed for diﬀerent contaminants, and MCF-7 cell lines were treated for 24 h in a dose-related manner with individual-year extracts varying in their polybrominated diphenyl ether (PBDE) to polychlorinated biphenyl (PCB) ratios. Treated cellular material was then ﬁxed and interrogated using attenuated total reﬂection Fourier-transform IR (ATR-FTIR) spectroscopy; resultant IR spectra were computationally analyzed to derive dose-response relationships and to identify biomarkers associated with each contaminant mixture treatment. The results show distinct biomarkers of eﬀect are related to each contamination scenario, with an inverse relationship with dose observed. This study suggests that speciﬁc contaminant mixtures induce cellular alterations in the DNA/RNA spectral region that are most pronounced at low doses. It also suggests alterations in the “biochemical-cell ﬁngerprint” of IR spectra can be indicative of mixture exposures.

’ INTRODUCTION Environmental contaminants of concern such as persistent organic pollutants (POPs) include the legacy organochlorine (OC) substances dichloro-diphenyl-trichloroethanes (DDTs), polychlorinated biphenyls (PCBs), and emerging polybrominated diphenyl ether (PBDE) ﬂame retardants.1 Although the production of DDT and PCBs was banned in most countries during the 1970s and exposure levels in the environment have decreased, these compounds are still found all over the world, and there is still a concern that these chemicals or their metabolites, combine with other substances and induce adverse eﬀects in humans and wildlife.2 4 PBDE production started in large quantities during 1980s as a result of ﬁre regulations. Production of some brominated diphenyl ether (BDE) congener mixtures (penta-BDE and octa-BDE) ceased in the European Union and America by 2004 due to the concerns over toxic eﬀects similar to those associated with PCBs.5 Decabromodiphenyl ether (decaBDE) mixtures has been regulated in some states in the U.S. but is currently being used without restriction for non-electronic/electrical uses in the E.U. region.6 Many of these contaminants bioaccumulate in biota due to their lipophilic r 2011 American Chemical Society

nature and reach their highest concentrations in high trophic levels species.7 Predatory birds are good bioindicators of environmental contamination because they are at the top of the food chain, accumulate a wide range of contaminants, and are susceptible to the eﬀects of pollution.8,9 Birds’ eggs are a favorable matrix for analyzing contaminant concentrations as they are easily collected and reﬂect local pollution levels at the time of laying.10 12 POPs, individually or in combination, induce various adverse eﬀects in most species, including mammals, birds, reptiles, and amphibians;13 15 they are mainly thought to alter the endocrine system, but immunosuppressive, genotoxic, and neurological eﬀects have also been observed.16 18 POPs may disrupt the endocrine system by binding to estrogen, androgen, or thyroid receptors and mimic the function of that hormone,19 as well as by altering hormone metabolism via binding to the aryl hydrocarbon Received: July 25, 2011 Accepted: October 31, 2011 Revised: October 31, 2011 Published: October 31, 2011 10706

dx.doi.org/10.1021/es202574b | Environ. Sci. Technol. 2011, 45, 10706–10712

Environmental Science & Technology receptor (AhR) and activating the cytochrome P450 family of enzymes (e.g., CYP1A1 and CYP1B1).20 The AhR complex is found in most mammals, birds, reptiles, amphibians, ﬁsh, and invertebrates,21 and its activation is viewed as a critical step required for toxic substances such as dioxin (2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) to induce biological eﬀects.22 In vitro cell cultures possessing AhR activity have been used to predict the toxicity of newly emerging chemical contaminants, and explain how toxic substances impinge on diﬀerent species. For instance, it was shown that PBDE congeners produce a similar rank order of relative potencies (REPs) to AhR in cells of human, rat, chick and rainbow trout, when assessed by the EROD (ethoxyresoruﬁn-O-deethylase) assay.23 Similarly, PCB-induced hormetic eﬀects (estrogenic and antiestrogenic) have been observed in vitro in cell lines (MCF-7 cells and primary cultures of rainbow trout),24 as well as in in vivo rat studies.25 Because chemical-induced characteristics in vitro reﬂect some processes that occur in vivo, cell culture techniques have been successfully applied to toxicity testing and provide some mechanistic understanding of chemical actions.26 MCF-7 cells have been used in many studies for chemical testing because they are well established, hormonally responsive and metabolically active.27 Laboratory experiments showing biological responses to toxic substances based on adverse eﬀects induced by high concentrations may not reﬂect how a biological organism responds to low concentrations.28 We have previously used infrared (IR) spectroscopy as a novel tool to examine how MCF-7 cells respond to environmental levels (10 12 M) of chemical pollutants individually or in a binary mixture, and to ﬁngerprint speciﬁc exposure patterns.29 32 IR spectroscopy works on the basis that biomolecules absorb mid-IR (λ = 2 25 μm), and when a biological sample is interrogated a signature ﬁngerprint in the form of an absorbance spectrum correlating with structure and function can be derived. Diﬀerent biomolecules absorb diﬀerent regions of IR and a biochemical-cell ﬁngerprint region (1800 900 cm 1) can be approximately divided into regions associated mostly with lipids (1800 1700 cm 1), proteins (1690 1480 cm 1) and DNA/RNA (1425 900 cm1).33 In particular, secondary structure protein conformations are detected at 1650 cm 1 (Amide I), 1550 cm 1 (Amide II) or 1250 cm 1 (Amide III), DNA/RNA alterations at 1225 cm 1 (asymmetric phosphate; νasPO2 ) and 1080 cm 1 (symmetric phosphate; νsPO2 ), peaks for glycogen content (1030 cm 1), and protein phosphorylation (940 cm 1).33 Interrogation of cellular material with attenuated total reﬂection Fourier-transform IR (ATR-FTIR) spectroscopy leads to the generation of large data sets with hundreds of variables that are best analyzed with a multivariate analysis technique such as linear discriminant analysis (LDA).34 In this study, we set out to examine biochemical alterations induced by real-world environmental contaminant scenarios in MCF-7 cells using IR spectroscopy coupled with multivariate analysis. Particularly, our aim was to investigate if diﬀerent concentration ratios of environmentally relevant contaminants (PCB-to-BDE) produce distinct eﬀects that can be detected using ATR-FTIR spectroscopy combined with LDA. To investigate this, Northern gannet (Morus bassanus) eggs from diﬀerent years that varied in their BDE-to-PCB ratio were chemically extracted and MCF-7 cells were treated with these extracts in a dose-related manner. In addition, we compared the biochemical alterations induced by these chemical mixtures with those observed following treatment with BDE and PCB analytical standard treatment, in order to determine whether eﬀects associated with gannet egg extracts are driven by these compounds.

ARTICLE

’ EXPERIMENTAL SECTION Cell Culture. The human MCF-7 cell line was grown in Dulbecco’s modified essential medium supplemented with 10% heat-inactivated fetal calf serum (FCS), penicillin (100 U/mL) and streptomycin (100 μg/mL). Cells were grown in 5% CO2 in air at 37°C in a humidified atmosphere and were disaggregated with trypsin (0.05%)/EDTA (0.02%) solution before incorporation into experiments. Cell culture consumables were obtained from Invitrogen Life Technologies (Paisley, UK), unless otherwise stated. Extracts and Analysis. Fresh gannet eggs were taken from nests in each sampling year (1987, 1990, and 1992) at a Scottish colony Ailsa Craig (OS Grid Reference NX019997) by licensed egg collectors. The egg contents were homogenized and stored in glass jars at 20°C until use. A subgroup of five eggs from each sampling year for each colony was selected and analyzed for PBDE, PCB, mercury (Hg), DDE, and diedrin (HEOD) concentrations. The analytical methods for OCs, PCBs, and Hg were conducted as previously described.35 For PBDEs briefly, a 0.5 g subsample of each egg homogenate was thawed, weighed accurately, dried with anhydrous sodium sulfate and sochlet-extracted for 16 h in dichloromethane. Lipid content was determined gravimetrically and samples were cleaned by using an alumina glass column packed with 15 g acidified silica eluted with 300 mL hexane and then by gel permeation chromatography, before being evaporated under nitrogen. The extract was analyzed by gas chromatography mass spectrometry (GC-MS, ThermoFinnigan Trace) fitted with a ThermoQuest AS2000 autosampler and using a 30 m CPSIL-8 CB pesticide column (0.25 mm diameter, 0.12 μm internal diameter), as previously described.36 Stock solutions of gannet egg extracts in hexane were solvent exchanged, ﬁrst in ethanol and then dimethyl sulfoxide (DMSO) to minimize any sample loss (no ﬂuctuation of PBDE concentrations were apparent after solvent exchange). Serial dilutions in DMSO were added to cell incubates. The maximum DMSO concentration per incubate was 1% (v/v). Egg extract controls (EC) consisted of solvent only and were run at the same time as egg extracts and DMSO controls consisting of the same amount of DMSO as in treatments. Cell Treatments and ATR-FTIR Spectroscopy. Egg extracts from these years (1987, 1990, and 1992) were chosen to treat cell cultures because they varied in their BDE-to-PCB concentration ratios. The years selected were defined based on their BDE-toPCB levels: 1987 as medium in BDE and medium in PCB (MM); 1990 as low in BDE and high in PCB (LH); and 1992 as high in BDE and low in PCB (HL). Five chicken egg (CH) extracts (no BDE and PCB were detected) were used as controls. Mean concentrations of BDE and PCBs for MM extracts were 18.26 and 1260 ng/g Wwt (1:68), whereas in LH extracts they were 11.96 and 2280 ng/g Wwt (1:190) and for HL extract were 47.33 and 577.59 ng/g Wwt (1:12), respectively. Chemical concentrations for individual PBDE congeners, Hg, DDE, and HEOD are shown in Supporting Information (SI) Tables S1 and S2. MCF-7 cells were disaggregated, resuspended in complete medium and then seeded in T25 ﬂasks, whereupon they were concentrated in S-phase (grown for 24 h) prior to treatment with or without egg extracts for a further 24 h. Five eggs per year were analyzed. MCF-7 cells were treated separately in triplicate with each individual egg extract in a dose-related manner, at 5 mg, 10 mg, or 25 mg equivalents, as well as with DMSO and EC. Following treatment, cells were disaggregated and the cell 10707

dx.doi.org/10.1021/es202574b |Environ. Sci. Technol. 2011, 45, 10706–10712

Environmental Science & Technology suspensions were immediately ﬁxed with 70% ethanol and stored at 4°C until analyzed. Cellular material was then applied to IRreﬂective “Low E” glass slides (Kevley Technologies, Chesterland, OH), allowed to air-dry and then desiccated. IR spectra were obtained using a Bruker Vector 27 FTIR spectrometer with Helios ATR attachment containing a diamond crystal (Bruker Optics Ltd., Coventry, U.K.). For each experimental condition (i.e., each slide), 10 spectra were acquired (32 coadditions, ≈3.85 cm 1 wavenumber spacing, 2.2 kHz mirror speed). The ATR crystal was cleaned with sodium dodecyl sulfate (SDS; Sigma Chemical Co.) and a new background spectrum was collected prior to analysis of a new sample. Computational Analysis for Chemical Data. Principal component analysis (PCA) was employed in order to determine if the three selected sampling years differed in their chemical profiles. The data were first log-transformed and then PCA was carried out using total BDE, PCB, DDT, HEOD, and Hg chemical concentrations (ng/g Wwt) as input variables from each year (n = 15, 5 samples per year). In PCA, each variable (e.g., chemical) becomes a single point in n-dimensional space and, using PCs 1, 2, and 3 as coordinates, the data were analyzed for clustering. Spectra Processing and Multivariate Data Analysis. Raw IR spectra obtained from interrogated samples were preprocessed prior to computational analysis. Spectra were cut at the “biochemical-cell fingerprint” region (1800 900 cm 1), baseline-corrected (rubberband) using OPUS software and normalized to the Amide I peak (1650 cm 1). Spectra from five individual extract treatments were combined into separate groups (MM, LH, HL, or CH) and each treatment contained: n = 150 spectra, 10 spectra per experiment slide triplicate experiments 5 extracts per group. Variables in spectral data were reduced from 235 to 117 absorption intensities at different wavenumbers by averaging every two wavenumbers in order to ensure that the rate between number of spectra and number of variables is >5.33 Following on, LDA was applied. LDA is a linear transformation and thus generates new variables as linear combinations (i.e., weighted sums) of the original absorption intensities. Each new variable is called a “factor”. The weights for each factor are represented by a vector called a “loadings vector”. The loadings vectors are successive orthogonal solutions to the problem stated as “maximize the between-class variance over the within-class variance of the factor”.37 Each scalar value of each factor is called a “score”, which may be visualized through 1-, 2-, or 3-dimensional scatter plots—called “scores plots”—where clusters may be identified. LDA allows for a visualization called “cluster vectors plots” which may be used to identify biomarkers (i.e., wavenumbers) associated to specific treatment conditions. A cluster vector is a geometric construction whereby a vector is drawn from the center of a reference class (i.e., the “Control” group) to the center of a treatment class in the vector space spanned by the LDA loadings vectors.30 Thus, each cluster vector is a linear combination of the loadings vectors and it can be plotted as y-values having the wavenumbers as x-values. We applied a peak detection algorithm to identify the ten most prominent peaks from each cluster vector and plotted the location of the detected peaks along the wavenumber line using marker symbols whose size are proportional to the height of their corresponding peaks. Absolute values were used to measure the height of the peaks from cluster vector plots in order include both the negative and positive values. Repeated-measures one-way analysis of variance (ANOVA) with Dunnett’s post hoc tests were used to examine whether the

ARTICLE

Figure 1. Scores plots from principal component analysis (PCA) of chemical data using concentrations (ng/g Wwt) of contaminants in gannet eggs from diﬀerent years collected from Ailsa Craig as input variables (n = 15, 5 per year); MM (Medium BDE, Medium PCB), LH (Low BDE, High PCB), and HL (High BDE, Low PCB). Principal components 1, 2, and 3 account for 92.7%, 6.4%, and 0.75% of variance, respectively.

mixture-induced eﬀects observed in LD1 space diﬀered signiﬁcantly between treated vs control cell populations; there was no need to transform data to meet the underlying assumptions of homogeneity of variance between groups and normality of residuals. Apart from baseline correction (conducted using OPUS software) and ANOVA (conducted using GraphPad Prism), all data handling and visualization was performed in MATLAB r2009b using our in-house IRTools toolbox, which is freely available for download at http://biophotonics.lancs.ac.uk/software/irtools.

’ RESULTS AND DISCUSSION PCA using chemical contaminant concentrations as inputting variables showed a good separation in chemical concentrations between years, indicating diﬀerences in their chemical mixture content (Figure 1). When LDA was carried out for the MCF-7 cells treated with gannet egg extracts on the spectral data grouped by year with each year containing all the treatment concentrations (combination of 5 mg, 10 mg, and 25 mg equivalents), there was good separation between years and control cell populations [DMSO, EC or CH (Figure 2A)]. Separation of data based on diﬀerent mg-equivalent treatments indicated that segregation between the controls and years was greater. When the eﬀects of 5 mg equivalent treatments were compared to the controls there was a clear segregation between 5 mg equivalents treatments and all control cell population including DMSO (Figure 2B), CH (Figure 2C) and EC (Figure 2D). Similarly, good separation from controls was also apparent when MCF-7 cells were treated with 10 or 25 mg equivalent extracts (see SI Figures S1 and S2). Data analysis indicated that lowest-dose treatment (5 mg equivalent extracts) produced greater eﬀects on MCF-7 cells compared to 10 or 25 mg equivalent extracts in all treatment years (Figure 4 for MM extracts; SI Figures S3 and S4 for LH and HL extracts, respectively); hence, we examined the 5 mg equivalents to compare the eﬀects on cells between diﬀerent years. The eﬀects of 5 mg equivalent MM, LH or HL extracts were compared by measuring the distance in LD1 space from the mean of CH control extracts (Figure 3; for 10 or 25 mg equivalents see SI Figure S5 and S6, respectively). Distance in LDA enables one to determine which treatment produces the most alterations in overall cellular structures compared to the 10708

dx.doi.org/10.1021/es202574b |Environ. Sci. Technol. 2011, 45, 10706–10712

Environmental Science & Technology

ARTICLE

Figure 2. LDA scores plots of infrared (IR) spectra derived from MCF-7 cells treated with gannet egg extracts compared to control cell populations. (A) Eﬀects of all concentrations combined vs control cell populations. IR spectra from diﬀerent concentration treatments were combined based on class and each class contained the following number of spectra; for MM (Medium BDE, Medium PCB), LH (Low BDE, High PCB), and HL (High BDE, Low PCB), and chicken egg extract (CH) [n = 450 spectra (10 spectra per experiment slide triplicate experiments 5 extracts per year 3 concentrations)] and for corresponding controls combined DMSO or EC (extract control) (n = 600 spectra); (B) Eﬀects of 5 mg equivalent extracts (n = 150 spectra, 10 spectra per experiment slide triplicate experiments 5 extracts per year) vs DMSO combined (n = 600 spectra); (C) EC combined (n = 600 spectra); or, (D) 5 mg equivalent CH extracts (n = 150 spectra, 10 spectra per experiment slide triplicate experiments 5 extracts).

control egg matrix, and cluster vectors plots identify which biomolecules are altered with diﬀerent treatments.30 The highest distance from control in 5 mg equivalent extracts was observed for MM extracts, followed by LH and HL extracts, and ANOVA showed that the eﬀects induced by these contaminants resulted in spectral alterations in treated-cell populations that were signiﬁcantly diﬀerent (P e0.01) compared to those exposed to CH control extracts. Cluster vector plots (Figure 3B) showed that all of the mixtures induced their main eﬀects in the DNA/ RNA region (1425 900 cm 1), but some protein conformational changes were also evident. MM extracts in general produced the most pronounced eﬀects compared to LH or HL extracts in all spectral regions. Speciﬁc alterations associated with MM extracts included secondary protein structure conformation alterations in Amide I and Amide II (≈1650 and ≈1550 cm 1), glycogen content (≈1000 cm 1), protein phosphorylation (≈940 cm 1) and, in the νasPO2 and νsPO2 (≈1225 cm 1 and ≈1080 cm 1). LH eﬀects were associated with Amide II (≈1550 cm 1), CdO stretching of proteins (≈1400 cm 1), Amide III (≈1280 cm 1), carbohydrate (≈1180 cm 1), νasPO2 (≈1225 cm 1), and glycogen content (≈1025 cm 1). Less distinct biochemical changes were induced by HL extracts, although alterations in 1400 cm 1 (proteins), carbohydrate (≈1180 cm1), νsPO2 (≈1080 cm 1) and glycogen content (≈1030 cm 1) were observed. Given that MM extracts induced the most pronounced eﬀects on MCF-7 cultures, we tested how cells responded with this

extract in a dose response manner. An inverse dose-response relationship was apparent that gave rise to spectral alterations that were signiﬁcant at all concentrations (P e0.01), where the highest eﬀect was observed when cells were treated with 5 mg equivalent extracts and the lowest eﬀect was seen following treatment with 25 mg equivalent extract (Figure 4A). Examination of diﬀerences in signatures (Figure 4B) associated with each dose (5 mg, 10 mg, or 25 mg equivalent) also showed that all doses induced their main alterations in the DNA/RNA region (1425 900 cm 1); however, the lowest concentration produced no alterations in the lipid or protein region (1800 1480 cm 1) and the bands associated with νsPO2 (1080 cm 1) are predominantly altered following exposure to 5 mg equivalent extracts but not with 10 or 25 mg equivalent extracts. The results suggest that these chemical mixtures produced eﬀects in cells by genotoxic-associated mechanisms that were most pronounced with low doses. The end points observed in the current study we believe, rather than reﬂecting the amount of induced damage, indicate the cellular responses following contaminant-mediated eﬀects at environmentally relevant concentrations. Environmentally relevant concentrations of POP mixtures can induce negative eﬀects in hightrophic-level avian species; for instance, glaucous gulls (Larus hyperboreus) breeding in highly OC-polluted sites have been shown to have lower levels of thyroid hormones compared to birds from less exposed colonies.38 However, adverse eﬀects may not necessarily be associated with the highest POP 10709

dx.doi.org/10.1021/es202574b |Environ. Sci. Technol. 2011, 45, 10706–10712

Environmental Science & Technology

Figure 3. LDA comparing the eﬀects of gannet eggs extracts between diﬀerent years. (A) LD1 scores plot following LDA of infrared (IR) spectra derived from MCF-7 cells treated with 5 mg equivalent gannet egg extracts; MM (Medium BDE, Medium PCB), LH (Low BDE, High PCB), and HL (High BDE, Low PCB), and 5 mg equivalent CH (chicken) egg extracts (n = 150 spectra per class, 10 spectra per experiment slide triplicate experiments 5 extracts per year); and, (B) Corresponding cluster vector plots showing wavenumbers discriminating classes. The sizes of marker symbols in cluster vector plots are proportional to the height of their corresponding peaks. The dashed line represents a typical IR spectrum of the biochemical-ﬁngerprint region (1800 900 cm 1). ANOVA with Dunnett’s post hoc test was used to test signiﬁcance between contaminant-exposed vs cell populations treated with CH extracts (**, P e0.01).

concentration. A study comparing sex hormones (testosterone and 17β-estradiol) levels in glaucous gull from three contaminated sites that varied in their POP concentrations showed an association between pollution and alteration of hormone levels, but only at the least polluted site.39 Whereas PCB-induced eﬀects such as endocrine and immune disruptions have been well documented in predatory birds such as the American kestrel (Falco sparverius),18,40,41 it has also been shown that exposure of these species to PBDEs induced endocrine-related eﬀects, egg thinning and histopathological end points with diminished reproductive success.42,43 Given that organisms are continuously and variously exposed to low levels of diﬀerent agents, interactions between chemical mixtures that induce adverse eﬀects are likely to be complex and elusive. In the current study, the choice of eggs was based on existing knowledge of PCB and PBDE concentrations. However, gannets are exposed to a mixture of chemical pollutants and it is diﬃcult to identify which compounds in the egg extracts are responsible for biochemical alterations noted in MCF-7 cells. In previous studies, we treated MCF-7 cells with endocrine active (e.g., 17βestradiol or lindane) and DNA-reactive substances [e.g., benzo[a]pyrene (B[a]P) or 2-amino-1-methyl-6-phenylimidazo[4,5b]pyridine (PHIP)] and used ATR-FTIR spectroscopy to ﬁngerprint these speciﬁc exposures, as well as obtain biomarkers

ARTICLE

Figure 4. Dose-response eﬀects of MM (Medium BDE, Medium PCB) gannet egg extracts on MCF-7 cells. (A) LD1 scores plot following LDA of infrared (IR) spectra derived from MCF-7 cells treated with egg extracts (n = 150 spectra per treatment, 10 spectra per experiment slide triplicate experiments 5 extracts per treatment group) and EC (extract control) combined (n = 600 spectra); and (B) Corresponding cluster vector plots showing wavenumbers responsible for discriminating of discriminating eﬀects at doses tested. The sizes of marker symbols in cluster vector plots are proportional to the height of their corresponding peaks. The dashed line represents a typical IR spectrum of the biochemical-ﬁngerprint region (1800 900 cm 1). ANOVA with Dunnett’s post hoc test was used to test signiﬁcance between contaminantexposed vs EC-treated cell populations (**, P e0.01).

associated with technical mixtures of PCBs and PBDEs.29 32 We showed that BDE and PCB congeners, individually or in combination, induce alterations of cellular structures that are most apparent at low doses (10 12 M to 10 9 M). In particular an inverse dose-response relationship was observed with BDE-47, which is one of the most abundant congeners in environment.44 When we compared the biomarkers of eﬀects observed in the current study with those noted in previous studies, we can identify some similarity in the ﬁngerprint to that of PCB- and BDE-induced eﬀects, as well as some DNA/RNA eﬀects similar to B[a]P treatment. However, most of the eﬀects observed in the present study are distinct and mainly found in the DNA/RNA region, indicating an eﬀect that is mainly driven by genotoxic-associated mechanisms. In particular, it is interesting to note that these mixtures alter the lipid or Amide I regions (1800 1650 cm 1) minimally, and they induce their most pronounced eﬀects in the DNA/RNA region (1400 900 cm 1). These results indicate that eﬀects are likely to be driven via a speciﬁc combination of contaminants rather than one chemical compound or constituent present at the highest level. This response is clearest when MCF-7 cells are treated with MM extracts where it is evident that the lowest concentration gives rise to the largest eﬀect. In this study, ATR-FTIR spectroscopy was employed to identify if real world contamination scenarios induce cellular alterations in 10710

dx.doi.org/10.1021/es202574b |Environ. Sci. Technol. 2011, 45, 10706–10712

Environmental Science & Technology MCF-7 cells and to obtain biomarkers of eﬀect associated with representative chemical mixtures that may be found in the environment. We showed that ATR-FTIR spectroscopy distinguished eﬀects associated with diﬀerent contaminant mixtures and ﬁngerprinted real environmental pollution scenarios. Comparison of biomarkers previously identiﬁed using the same techniques showed that there may be some contribution of BDEs and PCBs to total observed eﬀects. However, chemical mixtures in this study seem to produce “new” signatures of eﬀects, most apparent at low doses, and of a genotoxic nature. This may be as a result of speciﬁc combination-driven eﬀects of known contaminants or some uncharacterized pollution. Whether these mixtures are able to induce such eﬀects in vivo remains to be validated and stresses that there is a need of future monitoring of chemical pollutants and their eﬀects in predatory birds.

’ ASSOCIATED CONTENT

bS

Supporting Information. Contaminant concentrations and additional dose-related eﬀects of gannet egg extracts from Ailsa Craig, Scotland. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +44 1524 510206; e-mail: [email protected].

’ ACKNOWLEDGMENT V.L. is a NERC-CEH algorithm student (NE/F008643/1). ’ REFERENCES (1) Muir, D. C. G.; de Wit, C. A. Trends of legacy and new persistent organic pollutants in the circumpolar arctic: Overview, conclusions, and recommendations. Sci. Total Environ. 2010, 408 (15), 3044–3051. (2) Eﬀects of pesticides on wildlife. In Handbook of Pesticide Toxicology; Cooper, K., Ed.; Academic Press: San Diego/New York, 1991; p 463 496. (3) Safe, S. H. Polychlorinated biphenyls (PCBs): environmental impact, biochemical and toxic responses, and implications for risk assessment. Crit. Rev. Toxicol. 1994, 24, 87–149. (4) Brustad, M.; Sandanger, T. M.; Nieboer, E.; Lund, E. 10th Anniversary Review: when healthy food becomes polluted-implications for public health and dietary advice. J. Environ. Monit. 2008, 10, 422–427. (5) Hale, R. C.; La Guardia, M. J.; Harvey, E.; Gaylor, M. O.; Mainor, T. M. Brominated ﬂame retardant concentrations and trends in abiotic media. Chemosphere 2006, 64, 181–186. (6) European Brominated Flame Retardant Industry Panel. The ESR and Deca-BDE. http://www.ebfrip.org/main-nav/european-regulatory-centre/the-existing-substances-regulation-esr/the-esr-and-deca-bde (accessed). (7) Hop, H.; Borga, K.; Gabrielsen, G. W.; Kleivane, L.; Skaare, J. U. Food web magniﬁcation of persistent organic pollutants in poikilotherms and homeotherms from the Barents Sea. Environ. Sci. Technol. 2002, 36, 2589–2597. (8) Ratcliﬀe, D. A. Decrease in eggshell weight in certain birds of prey. Nature 1967, 215, 208–210. (9) Fry, D. M. Reproductive eﬀects in birds exposed to pesticides and industrial chemicals. Environ. Health Perspect. 1995, 103, 165–171. (10) Becker, P. H.; Thyen, S.; Mickstein, S.; Sommer, U.; Schmieder, K. R. Monitoring Pollutants in Coastal Bird Eggs in the Wadden Sea, Wadden Sea Ecosystem No. 8; Common Wadden Sea Secretariat: Wilhelmshaven, 1998; pp 59 101.

ARTICLE

(11) Dell’Omo, G.; Costantini, D.; Wright, J.; Casagrande, S.; Shore, R. F. PCBs in the eggs of eurasian kestrels indicate exposure to local pollution. Ambio 2008, 37, 452–456. (12) Lam, J. C.; Tanabe, S.; Wong, B. S. F.; Lam, P. K. Trace element residues in eggs of Little Egret (Egretta garzetta) and Black-crowned Night Heron (Nycticorax nycticorax) from Hong Kong, China. Mar. Pollut. Bul. 2004, 48, 390–396. (13) Fisk, A. T.; de Wit, C. A.; Wayland, M.; Kuzyk, Z. Z.; Burgess, N.; Letcher, R.; Braune, B.; Norstrom, R.; Blum, S. P.; Sandau, C. An assessment of the toxicological signiﬁcance of anthropogenic contaminants in Canadian arctic wildlife. Sci. Total Environ. 2005, 351, 57–93. (14) Tanabe, S. Contamination and toxic eﬀects of persistent endocrine disrupters in marine mammals and birds. Mar. Pollut. Bull. 2002, 45, 69–77. (15) Hayes, T. B.; Case, P.; Chui, S.; Chung, D.; Haeﬀele, C.; Haston, K.; Lee, M.; Mai, V. P.; Marjuoa, Y.; Parker, J. Pesticide mixtures, endocrine disruption, and amphibian declines: are we underestimating the impact? Environ Health Perspect. 2006, 114, 40–50. (16) Østby, L.; Gabrielsen, G. W.; Krøkje, Å. Cytochrome P4501A induction and DNA adduct formation in glaucous gulls (Larus hyperboreus), fed with environmentally contaminated gull eggs. Ecotoxicol. Environ. Saf. 2005, 62, 363–375. (17) Grandjean, P.; Landrigan, P. J. Developmental neurotoxicity of industrial chemicals. Lancet 2006, 368, 2167–2178. (18) Smits, J. E.; Fernie, K. J.; Bortolotti, G. R.; Marchant, T. A Thyroid hormone suppression and cell-mediated immunomodulation in American kestrels (Falco sparverius) exposed to PCBs. Arch. Environ. Contam. Toxicol. 2002, 43, 338–344. (19) Peakall, D. B. Pesticide-induced enzyme breakdown of steroids in birds. Nature 1967, 216, 505–506. (20) Van den Berg, M.; Birnbaum, L.; Bosveld, A. T.; Brunstr€om, B.; Cook, P.; Feeley, M.; Giesy, J. P.; Hanberg, A.; Hasegawa, R.; Kennedy, S. W. Toxic equivalency factors (TEFs) for PCBs, PCDDs, PCDFs for humans and wildlife. Environ. Health Perspect. 1998, 106, 775–792. (21) Hahn, M. E. The aryl hydrocarbon receptor: A comparative perspective. Comp. Biochem. Physiol., Part C: Pharmacol., Toxicol. Endocrinol. 1998, 121, 23–53. (22) Nowicki, H. G.; Norman, A. W. Enhanced hepatic metabolism of testosterone, 4-androstene-3,17-dione, and estradiol-17 in chickens pretreated with DDT or PCB. Steroids 1972, 19, 85–99. (23) Chen, G.; Konstantinov, A. D.; Chittim, B. G.; Joyce, E. M.; Bols, N. C.; Bunce, N. J. Synthesis of polybrominated diphenyl ethers and their capacity to induce CYP1A by the Ah receptor mediated pathway. Environ. Sci. Technol. 2001, 35, 3749–3756. (24) Andersson, P. L.; Blom, A.; Johannisson, A.; Pesonen, M.; Tysklind, M.; Berg, A. H.; Olsson, P. E.; Norrgren, L. Assessment of PCBs and hydroxylated PCBs as potential xenoestrogens: in vitro studies based on MCF-7 cell proliferation and induction of vitellogenin in primary culture of rainbow trout hepatocytes. Arch. Environ. Contam. Toxicol. 1999, 37, 145–150. (25) Jansen, H. T.; Cooke, P. S.; Porcelli, J.; Liu, T. C.; Hansen, L. G. Estrogenic and antiestrogenic actions of PCBs in the female rat: in vitro and in vivo studies. Reprod. Toxicol. 1993, 7, 237–248. (26) Zucco, F.; De Angelis, I.; Testai, E.; Stammati, A. Toxicology investigations with cell culture systems: 20 years after. Toxicol. In Vitro 2004, 18, 153–163. (27) Jiao, H.; Allinson, S. L.; Walsh, M. J.; Hewitt, R.; Cole, K. J.; Phillips, D. H.; Martin, F. L. Growth kinetics in MCF-7 cells modulate benzo[a]pyrene-induced CYP1A1 up-regulation. Mutagenesis 2007, 22, 111–116. (28) Martin, F. L. Complex mixtures that may contain mutagenic and/or genotoxic components: a need to assess in vivo target-site eﬀect (s) associated with in vitro-positive(s). Chemosphere 2007, 69, 841–848. (29) Llabjani, V.; Jones, K. C.; Thomas, G. O.; Walker, L. A.; Shore, R. F.; Martin, F. L. Polybrominated diphenyl ether-associated alterations in cell biochemistry as determined by attenuated total reﬂection Fouriertransform infrared spectroscopy: A comparison with DNA-reactive and/or endocrine-disrupting agents. Environ. Sci. Technol. 2009, 43, 3356–3364. 10711

dx.doi.org/10.1021/es202574b |Environ. Sci. Technol. 2011, 45, 10706–10712

Environmental Science & Technology

ARTICLE

(30) Llabjani, V.; Trevisan, J.; Jones, K. C.; Shore, R. F.; Martin, F. L. Binary mixture eﬀects by PBDE congeners (47, 153, 183, or 209) and PCB congeners (126 or 153) in MCF-7 cells: biochemical alterations assessed by IR spectroscopy and multivariate analysis. Environ. Sci. Technol. 2010, 44, 3992–3998. (31) Ukpebor, J.; Llabjani, V.; Martin, F. L.; Halsall, C. J. Sublethal genotoxicity and cell alterations by organophosphorus pesticides in MCF-7 cells: implications for environmentally relevant concentrations. Environ. Toxicol. Chem. 2011, 30, 632–639. (32) Llabjani, V.; Trevisan, J.; Jones, K. C.; Shore, R. F.; Martin, F. L. Derivation by infrared spectroscopy with multivariate analysis of bimodal contaminant-induced dose-response eﬀects in MCF-7 cells. Environ. Sci. Technol. 2011, 45, 6129–6135. (33) Martin, F. L.; Kelly, J. G.; Llabjani, V.; Martin-Hirsch, P. L.; Patel, I. I.; Trevisan, J.; Fullwood, N. J.; Walsh, M. J. Distinguishing cell types or populations based on the computational analysis of their infrared spectra. Nat. Protoc. 2010, 5, 1748–1760. (34) Walsh, M. J.; Singh, M. N.; Stringfellow, H. F.; Pollock, H. M.; Hammiche, A.; Grude, O.; Fullwood, N. J.; Pitt, M. A.; Martin-Hirsch, P. L.; Martin, F. L. FTIR microspectroscopy coupled with two-class discrimination segregates markers responsible for inter-and intra-category variance in exfoliative cervical cytology. Biomark. Insights 2008, 3, 179–189. (35) Pereira, M. G.; Walker, L. A.; Best, J.; Shore, R. F. Long-term trends in mercury and PCB congener concentrations in gannet (Morus bassanus) eggs in Britain. Environ. Pollut. 2009, 157, 155–163. (36) Crosse, J. D.; Shore, R. F.; Jones, K. C.,; Pereira, M. G., Long term trends of PBDEs in gannet (Morus bassanus) eggs from two UK colonies, Environ. Pollut. 2012, 161, 93 100. (37) Hastie, T.; Tibshirani, R.; Friedman, J. H. the Elements of Statistical Learning: Data Mining, Inference, and Prediction: Springer Verlag: New York, 2009. (38) Verreault, J.; Skaare, J. U.; Jenssen, B. M.; Gabrielsen, G. W. Eﬀects of organochlorine contaminants on thyroid hormone levels in Arctic breeding glaucous gulls, Larus hyperboreus. Environ. Health Perspect. 2004, 112, 532–537. (39) Fisher, S. A.; Bortolotti, G. R.; Fernie, K. J.; Bird, D. M.; Smits, J. E. Brood patches of American kestrels altered by experimental exposure to PCBs. J. Toxicol. Environ. Health, Part A 2006, 69, 1603–1612. (40) Fernie, K. J.; Smits, J. E.; Bortolotti, G. R.; Bird, D. M. In ovo exposure to polychlorinated biphenyls: Reproductive eﬀects on secondgeneration American kestrels. Arch. Environ. Contam. Toxicol. 2001, 40, 544–550. (41) Fernie, K. J.; Shutt, J. L.; Mayne, G.; Hoﬀman, D.; Letcher, R. J.; Drouillard, K. G.; Ritchie, I. J. Exposure to polybrominated diphenyl ethers (PBDEs): changes in thyroid, vitamin A, glutathione homeostasis, and oxidative stress in American kestrels (Falco sparverius). Toxicol. Sci. 2005, 88, 375–383. (42) Fernie, K. J.; Shutt, J. L.; Letcher, R. J.; Ritchie, I. J.; Bird, D. M. Environmentally relevant concentrations of DE-71 and HBCD alter eggshell thickness and reproductive success of American kestrels. Environ. Sci. Technol. 2009, 43, 2124–2130. (43) Verboven, N.; Verreault, J.; Letcher, R. J.; Gabrielsen, G. W.; Evans, N. P. Maternally derived testosterone and 17β-estradiol in the eggs of Arctic-breeding glaucous gulls in relation to persistent organic pollutants. Comp. Biochem. Physiol C. Pharmacol. Toxicol. and Endocrinol. 2008, 148, 143–151. (44) Darnerud, P. O.; Eriksen, G. S.; Johannesson, T.; Larsen, P. B.; Viluksela, M. Polybrominated diphenyl ethers: occurrence, dietary exposure, and toxicology. Environ. Health Perspect. 2001, 109, 49–68.

10712

dx.doi.org/10.1021/es202574b |Environ. Sci. Technol. 2011, 45, 10706–10712

ARTICLE pubs.acs.org/est

Silver Nanoparticles and Total Aerosols Emitted by Nanotechnology-Related Consumer Spray Products Marina E. Quadros and Linsey C. Marr* Department of Civil and Environmental Engineering, Virginia Tech, 418 Durham Hall, Blacksburg, Virginia 24061, United States

bS Supporting Information ABSTRACT: Products containing silver nanoparticles are entering the market rapidly, but little is known about the potential for inhalation exposure to nanosilver. The objectives of this work were to characterize the emissions of airborne particles from consumer products that claim to contain silver nanoparticles or ions, determine the relationship between emissions and the products’ liquid characteristics, and assess the potential for inhalation exposure to silver during product use. Three products were investigated: an antiodor spray for hunters, a surface disinfectant, and a throat spray. Products emitted 0.24 56 ng of silver in aerosols per spray action. The plurality of silver was found in aerosols 1 2.5 μm in diameter for two products. Both the products’ liquid characteristics and the bottles’ spray mechanisms played roles in determining the size distribution of total aerosols, and the size of silver-containing aerosols emitted by the products was largely independent of the silver size distributions in the liquid phase. Silver was associated with chlorine in most samples. Results demonstrate that the normal use of silver-containing spray products carries the potential for inhalation of silvercontaining aerosols. Exposure modeling suggests that up to 70 ng of silver may deposit in the respiratory tract during product use.

’ INTRODUCTION Policies and regulations for responsible incorporation of nanomaterials into consumer products are still under development,1 4 yet the introduction of new products that have the potential to aerosolize engineered nanoparticles is proceeding swiftly.5 Silver nanoparticles (nanosilver) are increasingly popular additives to consumer products for antimicrobial purposes.6,7 Normal use of several types of nanotechnology-based consumer products, such as sprays, humidiﬁers, and hairdryers, may to lead to inhalation exposure of engineered nanoparticles, but the aerosol emission rates and characteristics—important for assessing risk— are largely unknown. Inhalation exposure to silver can cause respiratory tract irritation and an irreversible bluish-grayish discoloration of the skin (argyria) or eyes (argyrosis).8 10 Silver polishers exposed to silver and other metals developed bronchitis, emphysema, and a reduction in pulmonary volume.11 Occupational exposure to silver dusts and fumes via inhalation is regulated on the basis of mass concentration. The American Conference of Governmental Industrial Hygienists (ACGIH) recommends a threshold limit value (TLV) of 0.1 mg m 3 for metallic silver and 0.01 mg m 3 for soluble silver compounds, and the Occupational Safety and Health Administration (OSHA) has set a permissible exposure limit (PEL) 0.01 mg m 3 for metallic and soluble silver compounds combined. The Environmental Protection Agency (EPA) does not have an inhalation reference r 2011 American Chemical Society

concentration for silver, but its oral reference dose for silver is 0.005 mg kg 1 day 1. It is well established that inhalation of nanoparticles is associated with adverse health eﬀects,5,12 and recent studies justify concern about exposure to nanosilver. Soto et al.13 found that nanoparticle cytotoxicity was greater for cells exposed to nanosilver than to other materials (e.g., titanium dioxide, iron oxide). Sung et al.14 reported chronic alveolar inﬂammation in rats exposed subchronically to airborne silver nanoparticles, with the lungs and liver as the main target organs for silver accumulation. These authors suggested a threshold of 0.1 mg m 3 for no adverse eﬀects in rats for inhalation exposure to ∼20 nm silver nanoparticles. Even in those studies in which only minimal toxicity was found, authors suggested that chronic eﬀects might arise following long-term exposure to nanosilver.15 We are aware of only two studies of aerosols generated from nanotechnology-related consumer products. Hagendorfer et al.16 reported no measurable release of nanoparticles from a product containing silver nanoparticles when it was dispensed from a pump spray bottle; aerosol emissions were only detected when this product was applied from a pressurized spray bottle. Received: August 8, 2011 Accepted: November 9, 2011 Revised: October 31, 2011 Published: November 09, 2011 10713

dx.doi.org/10.1021/es202770m | Environ. Sci. Technol. 2011, 45, 10713–10719

Environmental Science & Technology

Figure 1. Experimental setup and mass balance equations used to determine aerosol emission rates (further described in the Supporting Information).

Norgaard et al.17 assessed the VOC and aerosol emissions from four types of spray products used for surface-coatings (none containing silver), three of which used manual spray pumps and one of which used a pressurized gas can. More data on the rates and characteristics of nanoparticles that are released during the use of consumer products are needed for studies of the toxicity18 and environmental fate and transport of engineered nanomaterials.19 Although toxicity studies have been successful in assessing dose response relationships for speciﬁc types of nanoparticles, the use of laboratory-generated, monodisperse, high purity nanoparticles (for the most part, uncoated) carried by puriﬁed air streams may not represent realistic human exposure scenarios. There are concerns about the inhalation toxicity of silver nanoparticles, and there are consumer products that contain nanosilver, but we do not know the amount or characteristics of aerosols that may be released during use of the products. This gap in knowledge is addressed by the objectives of this work: to characterize the emissions of aerosols from consumer products that claim to contain silver nanoparticles or ions, determine the relationship between aerosol emissions and the products’ liquid characteristics, and assess the potential for inhalation exposure to silver during product use.

’ EXPERIMENTAL METHODS Products Tested. Three spray products were chosen for this study based on manufacturers’ claims that they contain silver and the products’ potential for generating aerosols during normal use. 1 An antiodor spray for hunters (16 ounces, Silver Scent, SilverScent Products, Mechanicsville, VA), which lists “patented nano-silver” and water as the only ingredients. The silver concentration was not advertised. Two diﬀerent bottles were tested. 2 A surface disinfectant (4 ounces, SilverClene 24, Agion, Wakeﬁeld, MA), whose label advertises “electrolytically generated Ag+” (0.003%), citric acid (4.840%) and “other ingredients” (95.157%). The company’s Web site advertises the use of zeolite to deliver silver ions in its products. 3 A throat spray (2 ounces, Wellness Colloidal Silver Throat Spray, Source Naturals, Santa Cruz, CA). “Colloidal silver (30 ppm)” is the only ingredient listed. Real-Time Aerosol Characterization. To characterize aerosol emissions from products, we sprayed them inside a 0.52 m3 polyethylene chamber (Atmosbag, Sigma-Aldrich, St. Louis, MO) (Figure 1). The chamber was initially filled with air that was conditioned though a hydrocarbon trap, two desiccant dryers,

ARTICLE

and a high efficiency particulate air (HEPA) filter to a relative humidity <10% and particle concentration <10 cm 3. We used a scanning mobility particle sizer (3936NL SMPS, TSI, Shoreview, MN), consisting of a long differential mobility analyzer and an ultrafine condensation particle counter (CPC 3025A, TSI), and a six-channel optical particle counter (OPC, Aerotrak, TSI) to measure concentrations and size distributions of aerosols with diameters <0.7 μm and 0.3 10 μm, respectively. We also used a diffusion charger (DC 2000 CE, EcoChem, League City, TX) to measure aerosol surface area. We designed a spraying scheme to maintain aerosol concentrations relatively constant inside the chamber in order to facilitate modeling and calculations. Before each experiment, we placed the product in its original bottle inside the chamber, sealed the chamber, and purged it for at least 2 h. Using the chamber’s built-in gloves, we manually sprayed products initially at a higher frequency to build aerosol levels and then at a lower frequency to maintain aerosol counts at relatively constant levels (relative standard error <10%). The hunter spray, which had a larger spray pump compared to other products, was sprayed once every 5 s for the ﬁrst 2 min and then once every 2 min for the remaining time. The throat spray and the disinfecting spray, which had similarly sized, smaller spray pumps, were sprayed once every 5 s for the ﬁrst 4 min and then once every 1 min for the remaining time. The total spraying time was 30 min for all products. Running all aerosolization characterization methods simultaneously would have resulted in an excessively high ﬂow rate through the chamber, so we repeated the spraying scheme three times for each product, once for each of the following measurement scenarios (described in greater detail in Table S1 in the Supporting Information): (1) concentration and size distribution (0.3 10 μm) with the CPC and OPC, (2) concentration and size distribution (<0.7 μm) with the SMPS, and (3) collection of aerosol samples onto ﬁlters and electron microscopy grids. After each experiment, we wiped the chamber’s interior walls with a moist paper towel and let the chamber airdry overnight. We weighed the product before and after each experiment. As a control, we repeated the experiments using each product’s spray pump with a bottle ﬁlled with ultrapure water. By modeling the chamber as a continuously stirred tank reactor (Figure 1) and using size-resolved aerosol concentrations (C, in mL 1), we developed emission factors describing the total amount of particles emitted per spray action (i.e., a single squeeze of the pump) of each product. They can also be converted to emission factors in terms of volume of product sprayed by dividing by the volume per spray action. We set the inlet ﬂow rate (Qin) to zero for SMPS measurements and equal to the OPC ﬂow rate (2.83 L min 1) for measurements using the OPC and CPC together, such that dV/dt was equal to 300 mL min 1, the SMPS/CPC’s ﬂow rate, in both cases. We assumed that the particle concentration in the inlet air was zero (Cin = 0) and that wall losses followed a ﬁrst-order decay process (L = βVC), where β (in s 1) is the size-speciﬁc aerosol wall loss coeﬃcient, which was determined experimentally for this chamber. We solved the mass balance equation in Figure 1 for the size-speciﬁc aerosol emission term (E, in s 1). Diﬀerent spray pumps may generate diﬀerent droplet size distributions depending on the geometry of the nozzle and other design parameters, and to control for this variability, we also used a laboratory atomizer (3076, TSI, Shoreview, MN) to produce aerosols from the liquid products. This type of atomizer is used in 10714

dx.doi.org/10.1021/es202770m |Environ. Sci. Technol. 2011, 45, 10713–10719

Environmental Science & Technology many in vivo inhalation toxicity studies to produce aerosols from liquid solutions and suspensions. Aerosols were subsequently dried and neutralized and then characterized using the SMPS and the OPC. Silver Characterization in Aerosols. We collected aerosol samples on PTFE (Teflon) filters using a 4-stage cascade impactor (Sioutas, SKC, Eighty Four, PA) and extracted them using a method adapted from Benn and Westerhoff,20 Drake et al.,21 and OSHA ID-206 and ID-125G. We placed the filters in beakers covered with watch glasses and heated them to ∼90 °C in 20 mL of aqua regia—a 1:3 mixture of reagent-grade nitric acid (HNO3, 69%, Sigma-Aldrich) and hydrochloric acid (HCl, 38%, Sigma-Aldrich)—until the solution became clear. Then, we removed and rinsed the watch glasses using dilution acid (5% aqua regia), continued heating the solution until it reduced to <10 mL in volume, and restored it to 10 mL with ultrapure water. We quantified silver by inductively coupled plasma mass spectrometry (ICP-MS, Thermo Electron X-Series ICP-MS, Waltham, MA). We were unable to perform this characterization step with the first bottle of the hunter spray because we had consumed it entirely in other experiments. We also collected aerosol samples onto TEM grids using a custom-built thermophoretic precipitator (TP). The ﬂow rate was 12.2 ( 0.2 mL min 1, and top and bottom temperatures were maintained at 61.6 ( 0.4 °C and 18.2 ( 0.1 °C, respectively. We visualized grids using an environmental scanning electron microscope with electron dispersive X-ray spectrometry capabilities (ESEM, FEI Quanta 600 FEG, Hillsboro, OR), operated under high vacuum, and a transmission electron microscope (TEM, FEI Philips EM420). Liquid Characterization. To separate silver by size, we diluted products 1:10 in ultrapure water (11 ppb total organic carbon, 18.2 MΩ-cm) and successively filtered them through 1, 0.45, and 0.1 μm pore size, 47 mm diameter hydrophilic PTFE filters (Millipore Omnipore, Billerica, MA). We placed 4 mL of 0.1 μm-filtered samples in 3-KDa-cutoff centrifuge filtering units (Millipore Amicon Ultra) in attempt to separate ionic and nanoparticulate silver. We diluted bulk and size-fractionated samples of each product 1:100, added 20% aqua regia, and analyzed them for total silver using ICP-MS. We assessed silver ion sorption (i.e., loss) to filter media by repeating the same procedure with an ionic silver solution, obtained by dissolving ∼30 mg l 1 silver nitrate (Fisher Scientific, Fair Lawn, NJ) in ultrapure water. Loss of ionic silver to PTFE filters was negligible, but loss to the 3-KDa centrifuge filtering membrane was 40 ( 1%. We collected and analyzed all samples in triplicate. We rinsed all glassware in aqua regia, then multiple times in ultrapure water, and let it air-dry before use. Bulk and 1 μm ﬁltered samples were placed in a sonicating bath (85 W, American Scientiﬁc Products, McGaw Park, IL) for 60 s and analyzed by dynamic light scattering (DLS, Zetasizer, Malvern, Worcestershire, UK). The disinfecting spray could not be analyzed by DLS because it was a thick detergent liquid with unknown constituents; a refractive index of the material of interest is needed, and the suspension liquid must be known. Bulk and 1 μm ﬁltered samples were also analyzed on a UV-vis spectrophotometer (Varian Cary 5000, Santa Clara, CA). Aliquots of 10 μL of each product were placed onto holey-carbon coated copper TEM grids (SPI, West Chester, PA) and allowed to air-dry inside a desiccator for approximately 2 h. These grids were then visualized by SEM/EDS and TEM.

ARTICLE

Table 1. Size-Resolved Silver Concentrations (Mean ( Standard Error) in Liquid Media (ppm) size cutoﬀ

hunter spray 2

disinfecting spray

throat spray

>1000 nm

6.5 ( 2.1

0.2 ( 0.5

4.8 ( 0.9

450 1000 nm 100 450 nm

2.8 ( 0.7 0.8 ( 0.2

0.4 ( 0.3 1.6 ( 1.3a

0.7 ( 0.1 0.8 ( 0.2 16.5 ( 0.2

3 KDa 100 nm

1.7 ( 0.1

1.8 ( 0.7

<3 KDa

0.7 ( 0.1

26.6 ( 0.4

0.8 ( 0.1

total (bulk)

12.5 ( 1.8

27.5 ( 0.4

23.7 ( 1.2

a

Negative value resulted from a higher silver concentration in 100 nm ﬁltered samples relative to 450 nm ﬁltered samples. Diﬀerences in concentrations among all ﬁltration steps for this product were within 2 standard deviations of the average value, so nearly all silver was present in ionic form.

’ RESULTS AND DISCUSSION Characterization of Liquid Products. Total silver concentrations in the liquid phase were 12.5 ( 1.8 ppm (mean ( standard error) for the hunter spray, 27.5 ( 0.4 ppm for the disinfecting spray, and 23.7 ( 1.2 ppm for the throat spray. These values were 8% and 21% lower than the advertised concentrations of 30 ppm for the disinfecting spray and the throat spray, respectively. The hunter spray did not advertise its silver concentration. The silver mass distribution by particle size in liquid media varied greatly between products (Table 1). While most silver in the hunter spray was associated with relatively large particles (>450 nm), that in the disinfecting spray was nearly all ionic (<3 KDa). Silver in the throat spray was bimodally distributed; nearly 70% was nanosized (between the 3 KDa and 100 nm cutoffs), and ∼20% was associated with large particles (>1000 nm). Results indicate that the percentage of ionic silver lost due to sorption to the 3-KDa centrifuge filtering membrane is not easily transferable to more complex silver-containing solutions, such as the products tested, because correcting the disinfecting spray’s 3-KDa filtrate for 40% loss would have produced a concentration much higher than the total measured in the bulk liquid. There was no detectable UV-vis absorbance with the hunter spray and the disinfecting spray. The throat spray’s absorbance peaked at 419 nm, which is indicative of silver nanoparticles in suspension. Size distributions obtained by DLS indicated that a wide range of particle sizes were present in the products. The high polydispersity indexes (0.63 for the throat spray and 0.43 for the hunter spray) render the results highly uncertain, since DLS is a technique more suitable for unimodal, monodisperse suspensions. EDS spectra from liquid samples analyzed by TEM and SEM showed that silver was associated with chlorine in all products; most particles did not contain elements other than silver and chlorine. There was no microscopic or spectroscopic evidence of the presence of zeolite in the disinfecting spray. TEM and SEM images of particles in the liquid products, as well as size distributions obtained by DLS and UV vis absorption spectra, are presented in Supporting Information Figures S2 S20. Total Aerosol Emissions. Spraying produces aerosols that consist of water, solutes, and possibly solids, and some of the water will evaporate until the aerosols reach equilibrium, which depends on their initial diameter, ambient humidity, and solute concentrations. Figure 2 shows size distributions of total aerosols, not just those containing silver. The three products emitted aerosols spanning a wide range of sizes, from nanoscale up to 10 μm, and it is possible that they also produced larger droplets 10715

dx.doi.org/10.1021/es202770m |Environ. Sci. Technol. 2011, 45, 10713–10719

Environmental Science & Technology

ARTICLE

that settled quickly. The disinfecting spray and the first bottle of hunter spray produced bimodal size distributions, peaking around 20 nm and then again around 500 nm. The second bottle of hunter spray and the throat spray produced similar size distributions, with most aerosols larger than 500 nm. Interestingly, two diﬀerent bottles of the hunter spray produced diﬀerent aerosol size distributions. We acquired the bottles approximately 5 months apart and noticed that they had diﬀerent appearances; the liquid in the ﬁrst bottle was clear, while that in the second was milky-white, even though the packaging, labeling, and directions were identical. The diﬀerence in size distributions between the two bottles of hunter spray, which had the exact same type of nozzle, is evidence that both the product’s liquid characteristics and the spray nozzle play roles in determining the aerosol size distribution. The results serve as an example of the great variability that can be expected within consumer products, which makes generalizing results, and thus, regulating products on a case-by-case basis, challenging. Table 2 shows the total aerosol emission factors per spray action and per volume sprayed for each product, as well as the median diameter of the size distributions shown in Figure 2 and the amount of product dispensed per spray action. On the basis of spray action, the throat spray had the lowest emission factors. Across all products, emission factors varied by less than a factor of 10 for aerosols <0.75 μm and by less than a factor of 3 for aerosols 0.3 10 μm. On the basis of the volume sprayed, emission factors varied by a factor of 14 for aerosols <0.75 μm and by a factor of 7 for aerosols 0.3 10 μm. The disinfecting spray and the ﬁrst bottle of hunter spray produced aerosols with the smallest

Figure 2. Size distributions of total (silver and other) aerosols released per spray action (standard errors shown).

median diameters, whereas the second bottle of hunter spray and the throat spray produced aerosols with similar, larger median diameters. Because aerosol concentrations were very low in the chamber (100 400 cm 3), surface area concentrations were below the limit of detection. Aerosols generated by the atomizer were very diﬀerent from those generated by the products’ original spray bottles. Atomized aerosols had unimodal size distributions, with median diameters between 1.8 and 2.8 times smaller than their spray bottle counterparts (Table S2 in the Supporting Information). The median diameters of the atomized aerosols were similar across diﬀerent products (∼78 85 nm), probably because the size of aerosols generated is more dependent on the air pressure applied at the atomizer nozzle than on solute concentrations. The aerosol number concentration, however, varied greatly between products. This is probably because the number of aerosols generated is more dependent on the solute concentrations than on the pressure applied at the atomizer nozzle. The disinfecting spray produced over 10 times more aerosols compared to the hunter spray and the throat spray. These results conﬁrm that the diﬀerences observed in aerosol emission rates between products were not exclusively due to the use of diﬀerent spray pumps. During the experiment, particle concentrations in the 14 750 nm size range were elevated by 99 405 cm 3 above background levels inside the chamber. In the 300 500 nm size range, concentrations were 29 85 cm 3 above background. These concentrations were similar to those measured by Norgaard at al.17 but their aerosols were more evenly distributed over a wide range of sizes. Comparing size distributions, and, whenever possible, normalized emission rates (i.e., emission factors) between products and between diﬀerent studies is more appropriate than comparing concentrations, which are subject to experimental circumstances. Emission factors can also be aﬀected by variables such as temperature and humidity, so adoption of a set of standard environmental conditions for the investigation of nanoparticle emissions would be useful. Silver in Aerosolized Products. Silver aerosol emission factors in mass of silver emitted per spray action, segregated by particle size, are shown in Table 3. For the hunter spray and the throat spray, silver was present over a wide range of aerosol sizes, with the mode at 1 2.5 μm. Silver aerosol emissions were very low for the disinfecting spray, and there was no dominant size. Even though total silver concentrations in the liquid media varied by a factor of only 2.2 between products, the amount aerosolized by the throat spray was 4.6 and 230 times higher than by the hunter spray and disinfecting spray, respectively. These emission factors can be converted to mass per volume of product used by dividing by the volume of product dispersed per spray action in Table 2.

Table 2. Aerosol emission factors and characteristics for each product (mean ( standard error) characteristic

hunter spray 1

hunter spray 2

disinfecting spray

throat spray

Emissions Per Spray Action aerosols <0.75 μm

(5.6 ( 0.5) 107

(1.5 ( 0.1) 107

(4.0 ( 0.6) 107

(7.3 ( 0.9) 106

aerosols 0.3 10 μm

(4.4 ( 0.4) 10

(8 ( 3) 10

(7 ( 1) 10

(3 ( 1) 106

aerosols <0.75 μm

(7.5 ( 0.6) 107

(1.9 ( 0.1) 107

(2.6 ( 0.4) 108

(4.1 ( 0.5) 107

aerosols 0.3 10 μm

(5.9 ( 0.6) 10

(1.4 ( 0.2) 10

(4.4 ( 0.6) 10

(2.5 ( 0.4) 107

6

6

6

Emissions Per mL of Product

median diameter (nm) product volume per spray action (ml)

167 ( 9 0.747 ( 0.008

6

7

7

217 ( 23

150 ( 12

219 ( 27

0.778 ( 0.013

0.157 ( 0.005

0.181 ( 0.001

10716

dx.doi.org/10.1021/es202770m |Environ. Sci. Technol. 2011, 45, 10713–10719

Environmental Science & Technology

ARTICLE

In Tables 1 and 2, the size cutoﬀs refer to particles that may be only partially composed of silver; other elements and water may also be associated with the particles. The size of silver-containing aerosols emitted by the products was largely independent of the silver size distributions in the liquid phase. Rather, the aerosol size distribution produced from the spray pump itself governed the release of silver. For liquid products that contained mainly particulate silver (hunter spray and throat spray), over 90% of aerosolized silver was found in aerosols larger than 500 nm in size. The disinfecting spray contained mostly ionic silver and produced very low amounts of silver-containing aerosols that could be sampled using our impactor. The silver content of aerosols in terms of mass of silver per volume of particles could be estimated based on total aerosol size distributions (Figure 2 and Table 3), but the diﬀerent sampling methods (impactor for silver mass v. SMPS and OPC for number and size) and coarse resolution of the impactor, which necessitated assumptions about a representative aerosol size collected on each stage, introduced considerable uncertainty to the calculation. The resulting silver contents as a function of aerosol size spanned several orders of magnitude but were centered Table 3. Size-resolved silver emissions in aerosol, per spray action (ng) size cutoﬀ

hunter spray 2

disinfecting spray

throat spray 16.6 ( 2.8

>2.5 μm

2.4 ( 0.3

0.04 ( 0.02

1 2.5 μm

6.1 ( 1.8

0.04 ( 0.02

24.9 ( 7.3

0.5 1 μm

2.6 ( 0.8

0.10 ( 0.08

10.2 ( 3.1

0.25 0.5 μm after ﬁltera

0.7 ( 0.3 0.1 ( 0.1

0.04 ( 0.01 0.03 ( 0.02

2.8 ( 0.9 1.2 ( 0.2

12.0 ( 2.7

0.24 ( 0.12

55.6 ( 8.2

total

A Teﬂon ﬁlter that is expected to collect all particles that pass the 0.25-μm cutoﬀ with near 100% eﬃciency. a

around 0.5 50 ng mm 3, compared to the expected values of 12.5 27.5 ng mm 3 found in the products’ liquid phases (Table 1). If aerosol emissions were related to liquid characteristics, evaluation of new products would be greatly simpliﬁed. However, products’ liquid characteristics were suﬃcient for predicting some only some aerosol characteristics. The product that contained mostly ionic silver in liquid form, the disinfecting spray, yielded aerosols with very low silver concentrations, even though the total aerosol emissions (not just silver) were of the same order of magnitude as those of the other products. It is likely that this product’s spray pump produced mostly large droplets, which settled within seconds, too quickly and to be sampled and detected, and too quickly to present much of an inhalation risk. Products that contained particulate silver, either nanosized (disinfecting spray) or micrometer-sized (hunter spray), yielded silver-containing aerosols mostly in the 1 2.5 μm range, small enough to be inhaled. Figure 3 shows TEM and SEM images of aerosols, or more precisely the particles that remain when aerosols are exposed to a vacuum, generated by the hunter spray. They were spherical and multifaceted and ranged in size from 13 to 400 nm; most were smaller than 100 nm. All of the particles >100 nm that were observed seemed to be agglomerates of sub-100 nm unit particles. It is likely that the products’ manual spray pumps produced large droplets containing multiple nanoparticles, which agglomerated as the droplets dried and shrank. As can be seen in the EDS spectra, some of the particles contained silver and chlorine. Due to the low collection eﬃciency of the thermophoretic precipitator and low aerosol concentrations in the chamber throughout the experiments (∼100 cm 3), samples contained insuﬃcient numbers of particles to permit a sizedistribution count based on TEM or SEM images for the disinfecting spray and the throat spray. Inhalation Exposure to Silver. The aerosol emission factors can be used to estimate inhalation exposure to silver. Applying a

Figure 3. Hunter spray aerosols: (a) TEM image and selected area diﬀraction pattern, with d-spacings roughly matching those of silver crystals, (b) SEM images with EDS spectra of three selected aggregates (EDS spot size of 1 μm outlined), and (c) size distribution of the particles observed (n = 28). 10717

dx.doi.org/10.1021/es202770m |Environ. Sci. Technol. 2011, 45, 10713–10719

Environmental Science & Technology mass balance in a well-mixed indoor environment, we can calculate the total mass of silver inhaled associated with episodic emissions during product use.22 For example, if a person pumps the hunter spray 20 times at a rate of 1 spray action per second in a 30 m3 (4 m 3 m 2.5 m) room with an air exchange rate of 0.5 h 1,23 and remains in the room for a total of 10 min, she would inhale 0.62 ng of silver. Based on the size distribution of the silver aerosols (Table 3) and size-dependent deposition efficiency in various regions of the respiratory system,24 we estimate that 0.38 ng of silver would deposit in the respiratory system, with 77% of this in the nasopharyngeal region, 6% in the tracheobronchial region, and 17% in the alveolar region. Of course, actual exposure will differ because emissions will not mix instantaneously and homogeneously throughout the room. In fact, because the hunter spray is supposed to be sprayed onto the body, actual concentrations in the breathing zone are likely to be higher than modeled, so the resulting inhalation dose of silver is likely to be at least of the order of magnitude of 1 ng for this scenario. Less than 1% of the silver deposited by mass is expected to be nanoscale. If the particles are 100 nm in diameter, this mass corresponds to a number dose of ∼300 silver nanoparticles, or more if the actual diameter is smaller. Models of near-source concentrations of aerosol emissions from personal care products, currently not available to our knowledge, would greatly aid evaluation of the inhalation risk posed by spray products containing engineered nanoparticles. We omit the disinfecting spray from modeling because it did not contain silver nanoparticles. The throat spray’s recommended dose is 1 to 2 sprays per day into the mouth, and if under the worst-case scenario all of the aerosols produced were inhaled, 70 ng of silver would deposit in the respiratory system, with 82% of this falling in the nasopharyngeal region, 2% in the tracheobronchial region, and 16% in the alveolar region. Using the suggested maximum subchronic exposure level for Sprague Dawley rats of 100 μg m 3 of ∼20 nm silver nanoparticles, suggested by Sung et al.,14 rat body masses from the same reference, and minute ventilation rates from Sung et al.,25 we calculate a maximum inhalation exposure dose of 0.002 mg kg 1 day 1, above which alveolar inﬂammatory responses were observed. The silver dose from the use of the throat spray would be ∼16 times below the maximum recommended dose for a 70 kg adult and ∼3 times below it for a 15 kg child. This comparison does not account for interspecies variation. Measurements of personal exposure associated with the use of nanotechnology-based products under real-world conditions are needed to complement model-based predictions. Argyria is also of concern with exposure to silver. Since the throat spray is supposed to be sprayed directly into the mouth, we can assume that nearly all of it will enter the body, either by inhalation or ingestion. Considering the silver concentration in the product and the volume emitted per spray, we calculate the total silver exposure to be 0.009 mg day 1, which is ∼40 times below the EPA’s reference dose of 0.005 mg kg 1 day 1 for the development of argyria for a 70 kg adult, but only ∼9 times below the reference dose for a 15 kg child. As silver-nanotechnology products grow in popularity, exposure from multiple types of products can be expected and might lead to cumulative exposure levels above the reference dose for argyria. Evaluation of the health and environmental impacts of nanotechnology requires an accurate description of particles that are released at all stages of a product’s life cycle. During product use, the nanoparticles emitted are not necessarily in the same form in which they are added to the product. In addition to engineered

ARTICLE

nanoparticles, nanotechnology-based products may contain other ingredients such as surfactants, shelf stabilizers, or odor enhancers that may interact with the nanoparticles. When liquid products are sprayed, droplets are emitted, some of the solvent may evaporate, and except with pure suspensions, complex aerosols remain. The engineered nanomaterials may agglomerate with each other, become coated, and/or aggregate with other components of the product that solidify as the droplets dry. Therefore aerosolized nanoparticles’ physical and chemical characteristics may not be the same as those of the virgin nanomaterial. For future work, we recommend a more intense electron microscopy approach, including the use of single-particle chemical characterization for the aerosols using high-resolution TEM coupled with EDS. Additionally, single-particle aerosol mass spectrometry would enable in situ quantiﬁcation of the size and silver content of airborne particles. Because aerosol characteristics diﬀer greatly with use of a laboratory atomizer versus the consumer products’ own bottles, we advise future inhalation toxicity studies to consider carefully the aerosolization method. We also recommend that future studies assess the water content of the aerosols; the low aerosol concentrations in the chamber (100 400 cm 3) in this study were not conducive to the use of a diﬀusion dryer for such an investigation. The aerosol emission rates and size distributions presented in this work can serve as input to risk assessment models, such as the one developed by Lorenz et al.26 Results can be used to guide the selection of relevant particle doses in nanotoxicity testing, to predict exposure to emissions from nanotechnology-based product in indoor air quality models, and to develop regulations to ensure consumer safety.

’ ASSOCIATED CONTENT

bS

Supporting Information. Description of the experimental setup. Physical characteristics of aerosols generated by an atomizer. DLS size distributions, UV-vis absorbance spectra, TEM and SEM micrographs, micrograph-based primary particle size distributions, and EDS spectra of the liquid products. This material is available free of charge via the Internet at http://pubs. acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: (540) 231-6071; fax: (540) 231-7916; e-mail: lmarr@ vt.edu.

’ ACKNOWLEDGMENT This material is based upon work supported by the National Science Foundation (NSF) and the Environmental Protection Agency (EPA) under NSF Cooperative Agreement EF-0830093, Center for the Environmental Implications of NanoTechnology (CEINT). Virginia Tech’s Institute for Critical Technology and Applied Science (ICTAS) also provided support for this work. We thank the anonymous reviewers for their helpful comments and suggestions. ’ REFERENCES (1) Nowack, B.; Krug, H. F.; Height, M., 120 years of nanosilver history: Implications for policy makers. Environ. Sci. Technol. 2011, 47 (7), 3189 3189; DOI: 10.1021/es103316q. 10718

dx.doi.org/10.1021/es202770m |Environ. Sci. Technol. 2011, 45, 10713–10719

Environmental Science & Technology (2) USEPA. Nanotechnology White Paper; United States Environmental Protection Agency: Washington, DC, December 2, 2005; p 134; http://www.epa.gov/OSA/pdfs/EPA_nanotechnology_white_paper_ external_review_draft_12-02-2005.pdf. (3) Advisory Committee on Hazardous Substances. Report on Nanosilver; Department for Environment, Food and Rural Aﬀairs: London, October, 2009; p 7. (4) Kessler, R., Engineered nanoparticles in consumer products: Understanding a new ingredient. Environ. Health Perspect. 2011, 119 (3), 121 125; DOI: 10.1289/ehp.119-a120. (5) Quadros, M. E.; Marr, L. C., Environmental and human health risks of aerosolized silver nanoparticles. J. Air Waste Manage. Assoc. 2010, 60 (7), 770 781; DOI: 10.3155/1047-3289.60.7.770. (6) Marambio-Jones, C.; Hoek, E. M. V., A review of the antibacterial eﬀects of silver nanomaterials and potential implications for human health and the environment. J. Nanopart. Res. 2010, 12 (5), 1531 1551; DOI: 10.1007/s11051-010-9900-y. (7) Majestic, B. J.; Erdakos, G. B.; Lewandowski, M.; Oliver, K. D.; Willis, R. D.; Kleindienst, T. E.; Bhave, P. V. A review of selected engineered nanoparticles in the atmosphere, sources, transformations, and techniques for sampling and analysis. Int. J. Occup. Environ. Health 2010, 16 (4), 488–507. (8) Drake, P. L.; Hazelwood, K. J., Exposure-related health eﬀects of silver and silver compounds: A review. . 2005, 49 (7), 575 585; DOI: 10.1093/annhyg/mei019 . (9) Hill, W. R.; Pillsbury, D. M. Argyria: The Pharmacology of Silver; The Williams & Wilkins company, 1939. (10) Johnston, H. J.; Hutchison, G.; Christensen, F. M.; Peters, S.; Hankin, S.; Stone, V., A review of the in vivo and in vitro toxicity of silver and gold particulates: Particle attributes and biological mechanisms responsible for the observed toxicity. Crit. Rev. Toxicol. 2010, 40 (4), 328 346; DOI: 10.3109/10408440903453074. (11) Barrie, H. J.; Harding, H. E. Argyro-siderosis of the lungs in silver ﬁnishers. Br. J. Ind. Med. 1947, 4 (225), 5. (12) Oberdorster, G.; Oberdorster, E.; Oberdorster, J., Nanotoxicology: An emerging discipline evolving from studies of ultraﬁne particles. Environ. Health Perspect. 2005, 113 (7), 823 839; DOI: 10.1289/ehp.7339. (13) Soto, K. F.; Carrasco, A.; Powell, T. G.; Garza, K. M.; Murr, L. E., Comparative in vitro cytotoxicity assessment of some manufactured nanoparticulate materials characterized by transmission electron microscopy. J. Nanopart. Res. 2005, 7 (2 3), 145 169; DOI: 10.1007/ s11051-005-3473-1. (14) Sung, J. H.; Ji, J. H.; Park, J. D.; Yoon, J. U.; Kim, D. S.; Jeon, K. S.; Song, M. Y.; Jeong, J.; Han, B. S.; Han, J. H.; Chung, Y. H.; Chang, H. K.; Lee, J. H.; Cho, M. H.; Kelman, B. J.; Yu, I. J., Subchronic inhalation toxicity of silver nanoparticles. Toxicol. Sci. 2009, 108 (2), 452 461; DOI: 10.1093/toxsci/kfn246. (15) Stebounova, L. V.; Adamcakova-Dodd, A.; Kim, J. S.; Park, H.; O’Shaughnessy, P. T.; Grassian, V. H.; Thorne, P. S., Nanosilver induces minimal lung toxicity or inﬂammation in a subacute murine inhalation model. Particle and Fibre Toxicology 2011, 8 (5), 1 12; DOI: 10.1186/ 1743-8977-8-5. (16) Hagendorfer, H.; Lorenz, C.; Kaegi, R.; Sinnet, B.; Gehrig, R.; Goetz, N. V.; Scheringer, M.; Ludwig, C.; Ulrich, A., Size-fractionated characterization and quantiﬁcation of nanoparticle release rates from a consumer spray product containing engineered nanoparticles. J. Nanopart. Res. 2010, 12 (7), 2481 2494; DOI: 10.1007/s11051-009-9816-6. (17) Norgaard, A. W.; Jensen, K. A.; Janfelt, C.; Lauritsen, F. R.; Clausen, P. A.; Wolkoﬀ, P., Release of VOCs and particles during use of nanoﬁlm spray products. Environ. Sci. Technol. 2009, 43 (20), 7824 7830; DOI: 10.1021/es9010468. (18) Christensen, F. M.; Johnston, H. J.; Stone, V.; Aitken, R. J.; Hankin, S.; Peters, S.; Aschberger, K., Nano-silver—Feasibility and challenges for human health risk assessment based on open literature. Nanotoxicology 2010, 4 (3), 284 295; DOI: 10.3109/17435391003690549. (19) Gottschalk, F.; Nowack, B., The release of engineered nanomaterials to the environment. J. Environ. Monit. 2011, 13 (5), 1145 1155; DOI: 10.1039/c0em00547a.

ARTICLE

(20) Benn, T. M.; Westerhoﬀ, P., Nanoparticle silver released into water from commercially available sock fabrics. Environ. Sci. Technol. 2008, 42 (11), 4133 4139; DOI: 10.1021/es7032718. (21) Drake, P. L.; Marcy, A. D.; Ashley, K., Evaluation of a standardized method for determining soluble silver in workplace air samples. J. Environ. Monit. 2006, 8 (1), 134 139; DOI: 10.1039/ b511150a. (22) Nazaroﬀ, W. W., Inhalation intake fraction of pollutants from episodic indoor emissions. Build. Environ. 2008, 43 (3), 269 277; DOI: 10.1016/j.buildenv.2006.03.021. (23) Murray, D. M.; Burmaster, D. E., Residential air exchange rates in the United States—Empirical and estimated parametric distributions by season and climatic region. Risk Anal. 1995, 15 (4), 459 465; DOI: 10.1111/j.1539-6924.1995.tb00338.x. (24) Hinds, W. C., Aerosol Technology., 2 ed.; Wiley-Interscience: New York, 1999; p 483. (25) Sung, J. H.; Ji, J. H.; Yoon, J. U.; Kim, D. S.; Song, M. Y.; Jeong, J.; Han, B. S.; Han, J. H.; Chung, Y. H.; Kim, J.; Kim, T. S.; Chang, H. K.; Lee, E. J.; Lee, J. H.; Yu, I. J., Lung function changes in Sprague-Dawley rats after prolonged inhalation exposure to silver nanoparticles. Inhalation Toxicol. 2008, 20 (6), 567 574; DOI: 10.1080/08958370701874671. (26) Lorenz, C.; Goetz, N. V.; Scheringer, M.; Wormuth, M.; Hungerb€uhler, K., Potential exposure of German consumers to engineered nanoparticles in cosmetics and personal care products. Nanotoxicology 2010, 5 (1), 18; DOI: 10.3109/17435390.2010.484554.

10719

dx.doi.org/10.1021/es202770m |Environ. Sci. Technol. 2011, 45, 10713–10719

ARTICLE pubs.acs.org/est

New Evidence for Toxicity of Polybrominated Diphenyl Ethers: DNA Adduct Formation from Quinone Metabolites Yongquan Lai, Minghua Lu, Xiang Gao, Hanzhi Wu, and Zongwei Cai* Department of Chemistry, Hong Kong Baptist University, 224 Waterloo Road, Kowloon Tong, Hong Kong SAR, China

bS Supporting Information ABSTRACT: This study investigated the formation of DNA adducts of polybrominated diphenyl ethers (PBDEs) and the possible mechanisms. DNA adduction was conducted by in vitro reaction of deoxyguanosine (dG) and DNA with PBDEquinone (PBDE-Q) metabolites, and DNA adducts were characterized by using electrospray ionization tandem mass spectrometry. The results suggested DNA adduction involved Michael Addition between the exocyclic NH2 group at the N-2 position of dG and the electron-deﬁcient carbon of quinone, followed by reductive cyclization with loss of (bromo-)1-hydroperoxy-benzene or water to form a type I or type II adduct. PBDE-Q with substituted bromine on the quinone ring was proven to be a favorable structure to form a type I adduct, while the absence of bromine on the quinone ring resulted in a type II adduct. Lower reactivity of adduction was also observed with increasing the number of bromine atoms on the phenoxyl ring. Our data clearly demonstrated PBDEs could covalently bind to DNA mediated by quinone metabolites, depending on the degree of bromine substitution. This study opened a new view on the mechanism of toxicity of PBDEs and reported the structure of PBDEDNA adducts, which might be valuable for the evaluation on potential in vivo formation of PBDEDNA adducts.

’ INTRODUCTION Polybrominated diphenyl ethers (PBDEs) have been used as ﬂame retardants in many commercial and household products including polyurethane foam, textiles, furniture, and electronics.13 The noncovalent binding of PBDEs in polymers allows the chemicals to enter the environment easily from the product surface during use.4 PBDEs have become contaminants of worldwide concern because of their widespread use, ubiquitous environmental distribution, great bioaccumulation potential, and toxicity.5 It is of particular worth to note that the concentrations of PBDEs in human blood, breast milk, and other body tissues have been increasing with a doubling time of approximately 46 years over the last 30 years.6,7 PBDEs could induce phase I enzyme activities, such as cytochrome P450 isozymes,810 and could also be the substrates for the isozymes with the formation of several metabolites in vivo.3,1113 It has been reported that PBDE phase I metabolism occurred via three possible metabolic pathways: oxidation, debromination,14 and oxidative debromination.15 For the oxidation metabolism, PBDEs were initially metabolized to arene oxides, followed by cytochrome P450 enzyme-catalyzed hydroxylation to form monohydroxylated PBDEs (OH-PBDEs) or dihydroxylated PBDEs (diOH-PBDEs).1618 For example, OH-PBDEs have been determined in pooled serum from humans living or working at municipal waste disposal site.19 OH-PBDEs have received great attention due to their contribution to the recognized toxic eﬀects of PBDEs such as endocrine disruption and development neurotoxicity. r 2011 American Chemical Society

It is of particular concern to note that metabolic activation of PBDEs could result in hydroquinone and catechol metabolites (diOH-PBDEs with two hydroxyl group at para and ortho positions) that might possess more toxic eﬀects than PBDEs and OH-PBDEs. However, little attention has been paid to the toxicity of hydroquinone and catechol metabolites. Hydroquinone metabolites might undergo peroxidase-catalyzed oxidation to quinones that could create a variety of the hazardous eﬀects in biological systems, including acute cytotoxicity, immunotoxicity, and carcinogenesis.20 The structure change of DNA as a result of covalent binding to carcinogens or their active metabolites was considered as an early critical step in chemical carcinogenesis. If not being repaired before DNA replication, DNA adducts can cause misrepairing, resulting in mutations and chromosomal alternations.21 To take as an example of polycyclic aromatic hydrocarbon (PAH) carcinogenesis, it has been shown that metabolic activation of PAH occurred via a two-step process involving cytochrome P450-mediated formation of hydroquinone metabolites and their further peroxidase-dependent oxidation conversion to PAH-derived quinones. The formed PAHquinones played a critical role in the initiation of cancer by covalent binding to DNA.20 Based on the mechanism for PAH carcinogenesis, it Received: September 2, 2011 Accepted: November 3, 2011 Revised: October 26, 2011 Published: November 03, 2011 10720

dx.doi.org/10.1021/es203068f | Environ. Sci. Technol. 2011, 45, 10720–10727

Environmental Science & Technology was postulated that PBDE-derived quinones (PBDEQs) might covalently bind to DNA, and the DNA adducts might become biomarkers for the potential of PBDEs to initiate cancer. Although 32P-postlabeling is one of the most sensitive methods commonly used for the analysis of DNA adducts, this method provides no structural information. Mass spectrometry (MS) has become a powerful and alternative technology to characterize and in most cases determine DNA adducts with the sensitivity comparable to that of the other analytical approaches. In addition, the LC-MS method is more speciﬁc compared to 32 P-postlabeling, because adducts can be identiﬁed by their molecular weights and LC retention times. To date, there has been no report on the analysis of DNA adducts formed by PBDEQ metabolites. This study investigated the covalent binding of PBDEQs to DNA by using electrospray ionization tandem mass spectrometry (ESI-MS/MS). Identiﬁcation and characterization of the chemical structures of the DNA adducts may facilitate eﬀorts to establish the occurrence of DNA adduction and its biological signiﬁcance in complex biological system.22

’ EXPERIMENTAL SECTION Chemicals and Reagents. 20 -Deoxyguanosine (dG), nuclease

P1, phosphodiesterase I, alkaline phosphatase, calf thymus DNA, and 2,4-dibromophenol were obtained from Sigma (St. Louis, MO, USA). 2-Bromobenzoquinone (2BrBQ) was obtained from Tokyo Chemical Industrial Co., Ltd. (Tokyo, Japan). 2,6-Bromobenzoquinone (26BrBQ) was purchased from APIN Chemicals Ltd. (Abingdon, UK). CDCl3, phenol, 4-bromophenol, and 2,4,6-tribromophenol were purchased from Acros Organics (Geel, Belgium). Horseradish peroxidase (HRP) was purchased from Wako Pure Chemical Industries Ltd. (Osaka, Japan). β-Nicotinamideadenine dinucleotide phosphate, reduced (NADPH) was obtained from Oriental Yeast Co., Ltd. (Tokyo, Japan). 60 -OHBDE-17, 30 -OH-BDE-7, and 6-OH-BDE-47 were received from Dr. Michael H.W. Lam (Department of Biology and Chemistry, City University of Hong Kong, Hong Kong). Precoated thinlayer chromatographic (TLC) plates (DC-Fertigplatten SIL G-25 UV254) were purchased from Macherey-Nagel (D€uren, Germany). Dimethyl sulfoxide (DMSO) in analytical grade was purchased from AJAX Chemicals (Sydney, Australia). Water was purified by employing a Milli-Q reagent water system (Millipore, Billerica, MA, USA). Synthesis of PBDEQs. PBDEQs were synthesized from the reaction of phenol or bromophenol with bromobenzoquinones. A solution of phenol or bromophenol (0.4 mmol) in DMF (0.5 mL) was added to the solution of bromobenzoquinone (0.25 mmol) in DMF (1 mL). The reaction was initiated by adding Na2HPO4 (0.25 mmol) and K2CO3 (0.08 mmol). The mixture was stirred at room temperature for 3 h. The reaction mixture was pour into 40 mL of ice cold H2O and extracted with ethyl acetate. The organic layer was washed with H2O and brine, dried (Na2SO4), and removed in vacuo. The crude product was purified on a TLC plate (CH2Cl2hexane, v/v, 1:1). The purified sample was submitted to NMR and GC-MS analysis. The 1H and 13C NMR spectra were recorded on a Bruker Avance-III spectrometer (at 400 and 100 MHz, respectively) in CDCl3. GCMS spectra were recorded on an Agilent 6890N GC coupled to an Agilent 5973 mass spectrometer (70 eV). More information on the GC-MS conditions and data is available in the Supporting Information.

ARTICLE

2-(20 ,40 -Bromophenoxyl)-benzoquinone (20 40 BrPhO-BQ). H NMR (400 MHz, CDCl3): δ = 5.62 (d, J = 2.4 Hz, 1 H), δ = 6.75 (dd, J = 2.4 and 10.4 Hz, 1 H), δ = 6.84 (d, J = 10.0 Hz, 1 H), δ = 7.05 (d, J = 8.4 Hz, 1 H), δ = 7.53 (dd, J = 2.4 and 8.8 Hz, 1 H), δ = 7.83 (d, J = 2.4 Hz, 1 H) ppm. 13C NMR (100 MHz, CDCl3): δ = 111.47, 116.55, 120.48, 124.03, 132.50, 134.66, 136.72, 137.02, 148.70, 156.51, 180.58, 187.04 ppm. MS (EI) m/z (relative intensity) = 358 (8) [M], 277 (100) [M Br], 249 (48) [M Br CO]. 2-Phenoxyl-6-bromo-benzoquinone (PhO-6-BrBQ). 1H NMR (400 MHz, CDCl3): δ = 5.74 (d, J = 2.4 Hz, 1 H), δ = 7.08 (d, J = 8.8 Hz, 2 H), δ = 7.21 (d, J = 2.0 Hz, 1 H), δ = 7.297.47 (m, 3 H) ppm. 13C NMR (100 MHz, CDCl3): δ = 110.99, 120.82, 126.91, 130.50, 134.49, 138.26, 152.41, 158.28, 174.40, 184.72 ppm. MS (EI) m/z (relative intensity) = 280 (97) [M], 199 (25) [M Br], 171 (100) [M Br CO]. 2-(20 -Bromophenoxyl)-6-bromo-benzoquinone (20 BrPhO-6BrBQ). 1H NMR (400 MHz, CDCl3): δ = 5.76 (d, J = 2.4 Hz, 1 H), δ = 6.98 (d, J = 8.8 Hz, 2 H), δ = 7.22 (d, J = 2.0 Hz, 1 H), δ = 7.57 (d, J = 8.8 Hz, 2 H) ppm. 13C NMR (100 MHz, CDCl3): δ = 111.29, 120.09, 122.60, 133.64, 134.54, 138.26, 151.44, 157.75, 174.14, 184.41 ppm. MS (EI) m/z (relative intensity) = 358 (90) [M], 277 (90) [M Br], 249 (100) [M Br CO]. 2-(20 ,40 -Bromophenoxyl)-6-bromo-benzoquinone (20 40 BrPhO6-BrBQ). 1H NMR (400 MHz, CDCl3): δ = 5.63 (d, J = 2.4 Hz, 1 H), δ = 7.04 (d, J = 7.2 Hz, 1 H), δ = 7.23(d, J = 2.4 Hz, 1 H), δ = 7.52 (dd, J = 2.4 and 8.4 Hz, 1 H,), δ = 7.82 (dd, J = 2.4 Hz, 1 H) ppm. 13C NMR (100 MHz, CDCl3): δ = 111.53, 116.47, 120.73, 123.92, 132.60, 134.65, 136.81, 138.28, 148.72, 156.11, 173.56, 184.19 ppm. MS (EI) m/z (relative intensity) = 438 (17) [M], 357 (100) [M Br], 329 (48) [M Br CO]. 2-(20 ,40 ,60 -Bromophenoxyl)-6-bromo-benzoquinone (20 40 60 BrPhO-6-BrBQ). 1H NMR (400 MHz, CDCl3): δ = 5.62 (d, J = 2.0 Hz, 1 H), δ = 7.25 (d, J = 2.0 Hz, 1 H), δ = 7.78 (s, 2 H) ppm. 13C NMR (100 MHz, CDCl3): δ = 111.33, 117.56, 121.07, 134.77, 135.76, 138.32, 146.28, 154.19, 173.07, 184.06 ppm. MS (EI) m/z (relative intensity) = 516 (10) [M], 435 (100) [M Br], 407 (35) [M Br CO]. Reaction of dG with PBDEQs. PBDEQs (0.3 μmol in DMSO, 5 μL) and dG (0.6 μmol in DMSO, 5 μL) were added into 300 μL of potassium phosphate buffer (0.1 M, pH = 7.4). The resulted mixtures were incubated in a shaking water bath at 37 °C for 2 h. The reaction mixture was extracted with ethyl acetate (2 0.5 mL), and the collected organic fractions were washed with DI water. The solvent was removed under a stream of nitrogen at room temperature. The residues were reconstituted with 0.5 mL of 0.1% formic acid in ACN before being submitted to ESI-MS/MS analysis. For control samples, PBDEQ or dG was absent in the incubation reactions. Microsomal-Mediated Metabolism of OH-PBDEs. In addition to chemically synthetic PBDEQs, the biotransformed PBDEquinone metabolites via microsomal-mediated metabolism of OH-PBDEs were also investigated. Liver microsomes from phenobarbital exposed male SpragueDawley rats were prepared as described previously.23 OH-PBDE congeners (0.3 μmol) dissolved in DMSO were added to 1.2 mL of potassium phosphate buffer (pH 7.4, 100 mM) containing 5 mM MgCl2 and microsomal protein (2.0 mg/mL). Incubations were prewarmed for 5 min at 37 °C in a shaking water bath before adding NADPH (2 mg) to initiate the reactions. After incubation for 1 h, the reaction mixtures were freeze-dried in a vacuum chamber, and the residues were extracted with ethyl acetate (2 0.5 mL). 1

10721

dx.doi.org/10.1021/es203068f |Environ. Sci. Technol. 2011, 45, 10720–10727

Environmental Science & Technology The combined extracts were dried under a stream of nitrogen at room temperature. Control reactions were carried out in the same manner, except that the NADPH or OH-PBDE congener was excluded. Adduction of dG with PBDE Metabolites. The residues of the microsomal-generated PBDE metabolites were dissolved in potassium phosphate buffer (pH 7.4, 100 mM) containing dG (0.3 μmol), HRP (0.2 U), and H2O2 (1 mM) to a final volume of 200 μL. Reactions were carried out for 2 h at 37 °C in a shaking water bath. The reaction mixtures were extracted with ethyl acetate (2 0.5 mL). The solvent was removed under a stream of nitrogen at room temperature, and the residues were reconstituted with 0.2 mL of 0.1% formic acid in ACN. Reaction of Calf Thymus DNA with 2BrBQ, 26BrBQ, and PBDEQs. DNA adduct formation in calf thymus DNA treated with bromobenzoquinone or PBDEQ was measured as follows. Briefly, quinone (0.15 μmol in DMSO, 5 μL) and calf thymus DNA (50 μg in H2O, 50 μL) were added into 250 μL of potassium phosphate buffer (0.1 M, pH = 7.4). The mixture was incubated in a shaking water bath at 37 °C for 12 h. Control incubation of quinone or DNA alone was carried out under the same conditions. The DNA was precipitated by adding 30 μL of 3 M sodium acetate solution (1/10 of the incubation buffer volume) and 0.9 mL of ice-cold ethanol. The precipitated DNA was recovered by centrifugation at 14 000 g for 30 min and rinsed with 0.9 mL of 70% ethanol to remove excess salts. After centrifugation at 14 000 g for 30 min, the wash liquid was removed, and the pellets were further dried under a nitrogen flow. The resulting DNA pellets were sequentially digested by nuclease P1, alkaline phosphatase, and phosphodiesterase using a previously published method.2426 More information regarding DNA digestion is available in the Supporting Information. The digestion solution was freeze-dried in a vacuum chamber, and the residue was reconstituted with 0.1% formic acid in the ACN before submitted to ESI-MS/MS analysis. ESI-MS/MS Analysis. Accurate mass values of DNA adducts were acquired in positive ion mode using a QTOF mass spectrometer (API QStar Pulsar i, Applied Biosystems, Foster City, USA) equipped with a TurboIonspray source. Experiments were performed at an ion spray voltage of 5000 V, declustering potential I (DP I) of 30 V, declustering potential II (DP II) of 15 V, and focusing potential (FP) of 80 V. The mass detection range was set to m/z 100900. The ion source gas I (GS I), gas II (GS II), curtain gas (CUR), and collision gas (CAD) were set at 30, 15, 30, and 3 psi, respectively. The temperature of GS II was set at 300 °C. Samples were directly infused to into the ESI source at a flow rate of 5 μL/min. Data acquisition and processing were performed by using Analyst QS software. The DNA adducts were analyzed by using triple-quadrupole mass spectrometry equipped with an ESI source (Waters ACQUITY TQ Detector, Waters Corporation, Milford, MA, USA). Separations of PBDEDNA adducts were performed on an UPLC (Waters ACQUITY UPLC system, Waters Corporation, Milford, MA, USA) with a reversed-phase ethylene bridged hybrid phenyl column (2.1 mm 150 mm, 1.7 μm). The mobile phase consisted of two components: A (H2O) and B (ACN). The mobile phase gradient was from 95% A to 5% A over 2.0 min, at which point it was held for 1.0 min. The column was then allowed to re-equilibrate back to the starting mobile phase of 95% A for 2.0 min before the next injection. An injection volume of 10 μL was selected with a ﬂow rate of 0.3 mL/min. PBDEDNA adducts were detected by monitoring their precursor-product

ARTICLE

transition ions in selected reaction monitoring (SRM) mode according to the results obtained from the ESI-QTOFMS. The optimized MS parameters were described as follows: the capillary voltage was 3000 V; the dwell time was 0.05 s; the extractor voltage was 2.5 V; the temperatures of the negative ESI source and desolvation gas were 118 and 500 °C, respectively; the cone gas and the desolvation gas ﬂows were 40 and 650 L/h, respectively. Instrument operation and data acquisition were processed by using the Waters MassLynx V4.1 SCN562 software package.

’ RESULTS AND DISCUSSION Previous studies suggested polychlorinated biphenyls (PCBs) could bind to DNA with an apparent preference for the guanine residue, which was mediated by the corresponding quinone metabolites.21,27,28 It was also found that PCB-derived paraquinones rather than ortho-quinones were involved in the major DNA adduction. Due to the structural analogy to PCBs, it was reasonable to envisage a PBDEDNA adduction mediated by PBDEquinone metabolites. Motivated by the need for understanding biological conversion of PBDEs and the causes for their potential genotoxicity, we examined the reactivity of PBDEQs with dG and calf thymus DNA. Understanding their interactions with DNA will provide useful information to guide future toxicological studies of PBDEs in vivo. In an eﬀort to better understand the reactivity of PBDEQs, a series of PBDE-derived para-quinones were synthesized (Figure 1). The availability of these compounds has allowed us to examine whether the chemical structures of PBDEQs aﬀect their ability to bind to DNA at physiological pH. The structures of puriﬁed products were conﬁrmed from GC-MS and NMR analysis (Figures S1S10 in the Supporting Information). For structural characterization of DNA adducts by ESI-MS/ MS, the reaction of PBDEQ with dG was carried out to produce respectable quantities synthetically. The analysis of the reaction mixture by ESI-QTOFMS clearly indicated that a mixture of dG adducts was produced. The observed adducts could be divided into two major types based on the reaction mechanism (Table 1 and Figure 2). Benzoquinone that is a di-α,β-unsaturated carbonyl compound has been found to undergo Michael Addition resulting in the formation of nucleoside adduction.29 PBDEQs have the same core structure as that of benzoquinone. Therefore, it was proposed that PBDEQs might also undergo DNA adduction at the quinone moiety, preferentially with dG followed by reductive cyclization with loss of a small molecule. According to the mechanism of Michael Addition, PBDEQs have four electron-deﬁcient carbons (C-2, C-3, C-5, and C-6) available for nucleophilic attack by the exocyclic NH2 group at position N-2 of dG, which in turn can result in the formation of multiple adducts (Figure 2). As for the formation of a type I adduct, Michael Addition was initiated between the electrondeﬁcient C-2 from the quinone and the N-2 from the dG, and led to the formation of unstable intermediate. Another nucleophilic attack on the carbonyl group of the intermediate generated a ﬁvemember ring, which was followed by the loss of a molecule of (bromo-)1-hydroperoxy-benzene before forming an aromatic system that stabilized the molecule. Similarly, a type II adduct was formed by nucleophilic attack of N-2 from the dG at C-5 position of quinone followed by reductive cyclization with loss a molecule of water. Table 1 summarized the accurate mass values of the identiﬁed dG adducts by ESI-QTOFMS, and the 10722

dx.doi.org/10.1021/es203068f |Environ. Sci. Technol. 2011, 45, 10720–10727

Environmental Science & Technology

ARTICLE

Figure 1. Chemical structures of OH-PBDEs (30 -OH-BDE-7, 60 -OH-BDE-17, and 6-OH-BDE-47), bromobenzoquinones (2BrBQ and 26BrBQ), and synthetic PBDEQs (20 40 BrPhO-BQ, PhO-6-BrBQ, 20 BrPhO-6-BrBQ, 20 40 BrPhO-6-BrBQ, and 20 40 60 BrPhO-6-BrBQ).

Table 1. Protonated Molecular Ions of the Identiﬁed PBDEdG Adducts by ESI-QTOFMS PBDEdG adduct ([M + H]+) type I adduct PBDEQ

expt

calcd

2 4 BrPhO-BQ

358.1142

358.1151

PhO-6-BrBQ

436.0237

436.0257

20 BrPhO-6-BrBQ

438.0235 436.0261

0 0

20 40 BrPhO-6-BrBQ 20 40 60 BrPhO-6-BrBQ

type II adduct error (ppm)

expt

calcd

error (ppm)

2.51

607.9652

607.9603

8.06

4.59

528.0461

528.0519

10.98

438.0236 436.0257

0.23 0.92

530.0454 607.9588

530.0498 607.9603

8.30 2.47

438.0257

438.0236

4.80

436.0252

436.0257

1.15

685.8634

685.8708

10.79

438.0226

438.0236

2.28

687.8667

687.8668

0.14

not detected

436.0224

436.0257

7.57

438.0237

438.0236

0.23

mass errors were within 11 ppm. The molecular weights of type I and II adducts were consistent with M (dG + PBDEQ (bromo-)1-hydroperoxy-benzene) and M (dG + PBDEQ H2O), respectively, which supported the proposed mechanism for the formation of dG adducts. It was clear to note that a type I adduct with two isotopic protonated ions at m/z 436 and 438 was observed from PhO-6-BrBQ, 20 BrPhO-6-BrBQ, 20 40 BrPhO-6BrBQ, and 20 40 60 BrPhO-6-BrBQ due to the same substitution pattern of bromine at the quinone ring (Table 1 and Figure 1). For comparison, the above identiﬁed DNA adducts were conﬁrmed from UPLC-ESI-MS/MS analysis. A fragmentation pattern of loss of the deoxyribose moiety from the molecular ion was observed for each adduct. The unique product ions generated by loss of deoxyribose moiety were chosen for the SRM analysis with increased detection sensitivity. The results from UPLC-MS/MS analysis were similar with those from the Q-TOF-MS/MS analysis, except that the UPLC-MS/MS in SRM mode provides much better sensitivity. A type I adduct with retention time at 1.80 min was detected from PhO-6-BrBQ,

20 BrPhO-6-BrBQ, 20 40 BrPhO-6-BrBQ, and 20 40 60 BrPhO-6-BrBQ by monitoring the same transition ions at m/z 436 > 320 (Figure 3). In addition, no type II adduct was observed from 20 40 60 BrPhO-6-BrBQ, which indicated that higher PBDEQs exhibited less reactivity compared to lower congeners. It was also demonstrated that all PBDEQs could form a type I adduct, which might be due to the more facile cleavage group of bromophenol compared to hydrogen. Theoretically, nucleophilic attack of N-2 from the dG at the C-3 or C-5 positions of quinone will generate two isomers of type II adducts. For example, two isomers of a type II adduct were observed from PhO-6-BrBQ, and they shared the same molecular ion and product ion with loss of a deoxyribose moiety from the molecular ion (Figure 3B). However, only one peak was identiﬁed as a type II adduct for 20 BrPhO-6-BrBQ and 20 40 BrPhO-6BrBQ. One possibility was that the bromophenoxyl group, unlike a nonsubstituted phenoxyl group, exerted a regioselective eﬀect on the Michael addition of quinone. It was proposed that nucleophilic attack at C-5 (para to the bromophenoxyl group) formed a 10723

dx.doi.org/10.1021/es203068f |Environ. Sci. Technol. 2011, 45, 10720–10727

Environmental Science & Technology

ARTICLE

Figure 2. Proposed mechanism for the formation of PBDEdG adducts. Michael addition was ﬁrst initiated between the electron-deﬁcient carbon of quinone and the exocyclic nitrogen, NH2, at the N-2 position of dG. Another nucleophilic attack formed a ﬁve-member ring, which was followed by loss of water or (bromo-)1-hydroperoxy-benzene before forming an aromatic system that stabilized the molecule.

Figure 3. UPLC-ESI-SRM chromatograms of PBDEdG adducts. For type I adducts of PhO-6-BrBQ, 20 BrPhO-6-BrBQ, 20 40 BrPhO-6-BrBQ, and 20 40 60 BrPhO-6-BrBQ, the chromatograms were obtained in negative ion mode with a collision energy value of 30 V. The type I adduct of 20 40 BrPhO-BQ and type II adducts of 20 40 BrPhO-BQ, PhO-6-BrBQ, 20 BrPhO-6-BrBQ, and 20 40 BrPhO-6-BrBQ were detected in positive ion mode with a collision energy value of 15 V.

sterically favorable adduct, whereas nucleophilic attack at C-3 (ortho to the bromophenoxyl group) generated a sterically unfavorable adduct due to the steric hindrance from the bulky bromophenoxyl group (Figure 2).22 Similar results were also obtained from the analysis of the type II adduct of 20 40 BrPhO-BQ (Figure 3A). To further conﬁrm the mechanism for the DNA adduction and the structures of the identiﬁed adducts, multiple-stage tandem mass spectra were obtained by using ESI-QTOFMS. By taking the type I adduct of PhO-6-BrBQ as an example, the mass values and isotopic distribution of its protonated molecular ions agreed well with the theoretical ones (Figure 4A). Two major

isotopic protonated ions were observed at m/z 436.0237 and 438.0235 (calcd. m/z 436.0257 and 438.0236) with mass errors of 4.59 and 0.23 ppm, respectively. In addition, the MS2 spectrum exhibited one characteristic fragment ion at m/z 322 after the release of 116 mass units from the molecular ion at m/z 438 (Figure 4B). This was corresponding to the loss of deoxyribose with proton transferring from the sugar to the guanine moiety. This also revealed that the DNA adduct was modiﬁed on the base moiety.22 As depicted in Figure 4C, other fragments of m/z 293.9869, 266.9779, 241.0670, 186.0638, and 134.0511 were consistent with the losses from either the guanine or the quinone portion of the product. Structural information on other adducts 10724

dx.doi.org/10.1021/es203068f |Environ. Sci. Technol. 2011, 45, 10720–10727

Environmental Science & Technology

Figure 4. ESI-QTOFMS (A), MS2 (B and C) spectra of a type I adduct generated by incubating PhO-6-BrBQ with dG. For the MS2 spectrum (B), a MS/MS experiment was carried out on a selected molecular ion at m/z 438.0236 by collision-induced dissociation (CID) with a collision energy value of 10 V. For the MS2 spectrum (C), a MS/MS experiment with a higher collision energy (45 V) was performed on the ion at m/z 321.9847 that was produced from the molecular ion at m/z 438.0236 through “source fragmentation”.

was also obtained from the corresponding tandem mass spectra (Figures S11S14 in the Supporting Information). Conﬁrmation of the PBDEDNA structures by NMR should be carried out if suﬃcient amount of adducts could be collected. In addition to the DNA adducts formed from chemically synthetic PBDEQs, the possibility of DNA adduct formation from microsomal-mediated quinone metabolites of PBDEs was also investigated. Metabolic activation of PBDEs has been reported to result in mono- and dihydroxylated metabolites.1618 It has also been reported that OH-PBDEs are more potent than PBDEs regarding the disruption of thyroid function and alteration of steroidogenesis.23,30 Compared to PBDEs and OHPBDEs, little attention has been paid to the toxicity of hydroquinone and catechol metabolites. PBDEs are ﬁrst metabolized to OH-PBDEs that can be further oxidized to dihydroxylated metabolites by microsomal cytochrome P450s (Figure 5A).

ARTICLE

Figure 5. Reaction pathway and UPLC-ESI-MS/MS analysis of dG adducts formed by PBDEQ metabolites of 60 -OH-BDE-17. Reaction pathway for the biotransformation of 60 -OH-BDE-17 and adduction of dG with PBDEQ metabolites (A); UPLC-ESI-SRM chromatograms of 60 -OH-BDE-17 and diOH-BDE metabolites (B), type I and type II DNA adducts (C). The collision energy values for type I and II adducts were 30 and 15 V, respectively. The UPLC and MS conditions used for the analysis of 60 -OH-BDE-17 and diOH-BDE metabolites were the same as that described in Figure S15 in the Supporting Information.

For example, four diOH-PBDEs were detected as metabolites of 60 -OH-BDE-17 as shown in Figure 5B. It is of particular concern to note that hydroquinone and catechol metabolites might be oxidized by peroxidases to quinones that are electrophilic and capable of reacting with DNA and sulfur nucleophiles such as glutathione and N-acetylcysteine.28 UPLC-ESI-MS/MS analysis indicated that one type I adduct with deprotonated molecular at m/z 436 and two type II adducts with protonated molecular ions at m/z 686 were detected from the reaction of dG and diOHPBDE metabolites of 60 -OH-BDE-17 in the presence of HRP and H2O2 (Figure 5C). Similarly, 30 -OH-BDE-7 produced a type II adduct with a protonated molecular ion at m/z 608 after sequential oxidation by microsomal cytochrome P450s and peroxidase (Figure S15 in the Supporting Information). Further more, a type I adduct was also observed from the reaction of dG with diOH-PBDE metabolites of 6-OH-BDE-47 (Figure S16 in the Supporting Information). It should be noted that 6-OH-BDE-47 was detected with the highest level among the reported OHPBDE congeners.1,31 This evidence suggested PBDE metabolites could be oxidized to species capable of covalently binding to DNA. 10725

dx.doi.org/10.1021/es203068f |Environ. Sci. Technol. 2011, 45, 10720–10727

Environmental Science & Technology

Figure 6. Eﬀects of chemical structures of PBDEQs on the formation of DNA adducts. After treatment of DNA with 2BrBQ, 26BrBQ, PhO-6BrBQ, 20 BrPhO-6-BrBQ, and 20 40 BrPhO-6-BrBQ, the same DNA adduct with transition ions at m/z 436 and 320 was detected from the digestion mixture of DNA by using UPLC-ESI-MS/MS. N.D. = not detected. The experiment was repeated twice.

After the analysis of dG adducts of PBDEQs was completed, the adduct formation in calf thymus DNA was examined. The generated DNA adducts were not detected in Q-TOF-MS analysis, probably due to the problem of detection limit. UPLC-ESIMS/MS analysis, however, provided good detection on the DNA adducts because much better sensitivity was obtained compared to Q-TOF-MS/MS. The UPLC-ESI-MS/MS analysis of the digestion mixture of DNA treated with PhO-6-BrBQ, 40 BrPhO6-BrBQ, and 20 40 BrPhO-6-BrBQ revealed that only one type I adduct was detected and conﬁrmed (Figure S17 in the Supporting Information), probably because the substituted bromine exists at the C-6 position of the quinone ring in these PBDEQs. On the other hand, one type II adduct was observed for 20 40 BrPhOBQ that does not have bromine on the quinone ring (Figure S17(F) in the Supporting Information). Thus, PBDEQs with a substituted bromine on the quinone ring is favored for forming a type I adduct, while the absence of bromine on the quinone ring resulted in a type II adduct. 2BrBQ and 26BrBQ were found to form the same adduct with the protonated isotopic ions at m/z 436 and 438 as the adduct generated from the incubation of dG with PhO-6-BrBQ, 40 BrPhO-6-BrBQ, or 20 40 BrPhO-6-BrBQ.26 In order to further examine the eﬀects of the bromophenoxyl group on the DNA adduction from PBDEQs, 2BrBQ and 26BrBQ were also incubated with calf thymus DNA under the same conditions. The obtained results shown in Figure 6 clearly demonstrated that chemical structure exerted eﬀects on the formation of the DNA adducts. 2BrBQ without the bromine substitution at the C-6 position was not as reactive as 26BrBQ. Compared to 2BrBQ and PhO-6-BrBQ, much more DNA adducts were detected from 40 BrPhO-6-BrBQ, suggesting that the occurrence of a bromophenoxyl group increased the reactivity of PBDEQs. Furthermore, 40 BrPhO-6-BrBQ was found to have higher reaction yield for the adduct formation than 20 40 BrPhO-6-BrBQ. No adduct was detected from the reaction with 20 40 60 BrPhO-6-BrBQ. Thus, the increase in the number of bromine substitutions in the phenoxyl group greatly lowered the reactivity of adduct formation due to the steric hindrance, which was consistent with results obtained from the DNA adduct formation from PCB-derived quinones.21,27,28 This data has important implications for understanding the mechanism of metabolic activation of PBDEs to genotoxins.

ARTICLE

In summary, the results obtained from this investigation clearly demonstrated that PBDEs could covalently bind to DNA in vitro to form DNA adducts mediated by quinone metabolites. The mechanism for the DNA adduct formation involved the Michael addition followed by reductive cyclization with loss of a molecule of water or (bromo-)1-hydroperoxy-benzene. Our data also suggested that ESI-MS/MS was a powerful analytical tool for the detection of suspected DNA adducts. The identiﬁed adducts might provide validated biomarkers for future in vivo PBDE risk assessment study. In addition, the implications that these adducts have for the cytotoxicity and genotoxicity of PBDEs are not known. Given the exposure potential to PBDEs and PBDEQ metabolites, however, the likelihood of genetic abnormality is suﬃciently serious that it cannot be dismissed without further study.

’ ASSOCIATED CONTENT

bS

Supporting Information. Details of the identiﬁcation of synthetic PBDEQs by GC-MS and NMR, DNA digestion procedures, ESI-QTOF-MS and UPLC-ESI-MS conditions, and analysis of PBDEDNA adducts, OH-PBDEs, and diOHPBDEs by ESI-MS/MS. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 852-34117070; fax: 852-34117348; e-mail: zwcai@ hkbu.edu.hk.

’ ACKNOWLEDGMENT The authors thank the special postgraduate studentship program of “Persistent Toxic Substances” from Research Grant Council, University Grants Committee of Hong Kong SAR. We also thank Dr. Michael H.W. Lam (Department of Biology and Chemistry, City University of Hong Kong, Hong Kong) for providing the 30 -OH-BDE-7, 60 -OH-BDE-17, and 6-OH-BDE-47. ’ REFERENCES (1) Valters, K.; Li, H. X.; Alaee, M.; D’Sa, I.; Marsh, G.; Bergman, A.; Letcher, R. J. Polybrominated diphenyl ethers and hydroxylated and methoxylated brominated and chlorinated analogues in the plasma of ﬁsh from the Detroit River. Environ. Sci. Technol. 2005, 39 (15), 5612–5619. (2) Alaee, M.; Arias, P.; Sjodin, A.; Bergman, A. An overview of commercially used brominated ﬂame retardants, their applications, their use patterns in diﬀerent countries/regions and possible modes of release. Environ. Int. 2003, 29 (6), 683–689. (3) Qiu, X. H.; Bigsby, R. M.; Hites, R. A. Hydroxylated metabolites of polybrominated diphenyl ethers in human blood samples from the United States. Environ. Health Perspect. 2009, 117 (1), 93–98. (4) de Wit, C. A. An overview of brominated ﬂame retardants in the environment. Chemosphere 2002, 46 (5), 583–624. (5) Wan, Y.; Wiseman, S.; Chang, H.; Zhang, X. W.; Jones, P. D.; Hecker, M.; Kannan, K.; Tanabe, S.; Hu, J. Y.; Lam, M. H. W.; Giesy, J. P. Origin of hydroxylated brominated diphenyl ethers: natural compounds or man-made ﬂame retardants. Environ. Sci. Technol. 2009, 43 (19), 7536–7542. (6) Hites, R. A. Polybrominated diphenyl ethers in the environment and in people: A meta-analysis of concentrations. Environ. Sci. Technol. 2004, 38 (4), 945–956. (7) Vonderheide, A. P. A review of the challenges in the chemical analysis of the polybrominated diphenyl ethers. Microchem. J. 2009, 92 (1), 49–57. 10726

dx.doi.org/10.1021/es203068f |Environ. Sci. Technol. 2011, 45, 10720–10727

Environmental Science & Technology (8) Chen, G. S.; Konstantinov, A. D.; Chittim, B. G.; Joyce, E. M.; Bols, N. C.; Bunce, N. J. Synthesis of polybrominated diphenyl ethers and their capacity to induce CYP1A by the Ah receptor mediated pathway. Environ. Sci. Technol. 2001, 35 (18), 3749–3756. (9) Sanders, J. M.; Burka, L. T.; Smith, C. S.; Black, W.; James, R.; Cunningham, M. L. Diﬀerential expression of CYP1A, 2B, and 3A genes in the F344 rat following exposure to a polybrominated diphenyl ether mixture or individual components. Toxicol. Sci. 2005, 88 (1), 127–133. (10) Pacyniak, E. K.; Cheng, X. G.; Cunningham, M. L.; Crofton, K.; Klaassen, C. D.; Guo, G. L. The ﬂame retardants, polybrominated diphenyl ethers, are pregnane X receptor activators. Toxicol. Sci. 2007, 97 (1), 94–102. (11) Orn, U.; Klasson-Wehler, E. Metabolism of 2,20 ,4,40 -tetrabromodiphenyl ether in rat and mouse. Xenobiotica 1998, 28 (2), 199–211. (12) Sandholm, A.; Emanuelsson, B. M.; Wehler, E. K. Bioavailability and half-life of decabromodiphenyl ether (BDE-209) in rat. Xenobiotica 2003, 33 (11), 1149–1158. (13) Malmberg, T.; Athanasiadou, M.; Marsh, G.; Brandt, I.; Bergmant, A. Identiﬁcation of hydroxylated polybrominated diphenyl ether metabolites in blood plasma from polybrominated diphenyl ether exposed rats. Environ. Sci. Technol. 2005, 39 (14), 5342–5348. (14) Stapleton, H. M.; Letcher, R. J.; Baker, J. E. Debromination of polybrominated diphenyl ether congeners BDE 99 and BDE 183 in the intestinal tract of the common carp (Cyprinus carpio). Environ. Sci. Technol. 2004, 38 (4), 1054–1061. (15) Hakk, H.; Huwe, J.; Low, M.; Rutherford, D.; Larsen, G. Tissue disposition, excretion and metabolism of 2,20 ,4,40 ,6-pentabromodiphenyl ether (BDE-100) in male Sprague-Dawley rats. Xenobiotica 2006, 36 (1), 79–94. (16) Hakk, H.; Larsen, G.; Klasson-Wehler, E. Tissue disposition, excretion and metabolism of 2,20 ,4,40 ,5-pentabromodiphenyl ether (BDE-99) in the male Sprague-Dawley rat. Xenobiotica 2002, 32 (5), 369–382. (17) Qiu, X. H.; Mercado-Feliciano, M.; Bigsby, R. M.; Hites, R. A. Measurement of polybrominated diphenyl ethers and metabolites in mouse plasma after exposure to a commercial pentabromodiphenyl ether mixture. Environ. Health Perspect. 2007, 115 (7), 1052–1058. (18) Lupton, S. J.; McGarrigle, B. P.; Olson, J. R.; Wood, T. D.; Aga, D. S. Human liver microsome-mediated metabolism of brominated diphenyl ethers 47, 99, and 153 and identiﬁcation of their major metabolites. Chem. Res. Toxicol. 2009, 22 (11), 1802–1809. (19) Yu, Z. Q.; Zheng, K. W.; Ren, G. F.; Zheng, Y. Y.; Ma, S. T.; Peng, P. G.; Wu, M. H.; Sheng, G. Y.; Fu, J. M. Identiﬁcation of hydroxylated octa- and nona-bromodiphenyl ethers in human serum from electronic waste dismantling workers. Environ. Sci. Technol. 2010, 44 (10), 3979–3985. (20) Bolton, J. L.; Trush, M. A.; Penning, T. M.; Dryhurst, G.; Monks, T. J. Role of quinones in toxicology. Chem. Res. Toxicol. 2000, 13 (3), 135–160. (21) Zhao, S. X.; Narang, A.; Ding, X. X.; Eadon, G. Characterization and quantitative analysis of DNA adducts formed from lower chlorinated PCB-derived quinones. Chem. Res. Toxicol. 2004, 17 (4), 502–511. (22) Zhao, S.; Narang, A.; Gierthy, J.; Eadon, G. Detection and characterization of DNA adducts formed from metabolites of the fungicide ortho-phenylphenol. J. Agric. Food Chem. 2002, 50 (11), 3351–3358. (23) Meerts, I. A. T. M.; van Zanden, J. J.; Luijks, E. A. C.; van Leeuwen-Bol, I.; Marsh, G.; Jakobsson, E.; Bergman, A.; Brouwer, A. Potent competitive interactions of some brominated ﬂame retardants and related compounds with human transthyretin in vitro. Toxicol. Sci. 2000, 56 (1), 95–104. (24) Crain, P. F. Preparation and enzymatic hydrolysis of DNA and RNA for mass spectrometry. Methods Enzymol. 1990, 193, 782–790. (25) Embrechts, J.; Lemiere, F.; Van Dongen, W.; Esmans, E. L.; Buytaert, P.; Van Marck, E.; Kockx, M.; Makar, A. Detection of estrogen DNA-adducts in human breast tumor tissue and healthy tissue by combined nano LC-nano ES tandem mass spectrometry. J. Am. Soc. Mass Spectrom. 2003, 14 (5), 482–491.

ARTICLE

(26) Lai, Y. Q.; Lu, M. H.; Lin, S. H.; Wu, H. Z.; Cai, Z. W. Electrospray ionization tandem mass spectrometric characterization of DNA adducts formed by bromobenzoquinones. Rapid Commun. Mass Spectrom. 2011, 25 (19), 2943–2950. (27) McLean, M. R.; Robertson, L. W.; Gupta, R. C. Detection of PCB adducts by the P-32-Postlabeling technique. Chem. Res. Toxicol. 1996, 9 (1), 165–171. (28) Oakley, G. G.; Robertson, L. W.; Gupta, R. C. Analysis of polychlorinated biphenyl-DNA adducts by P-32-postlabeling. Carcinogenesis 1996, 17 (1), 109–114. (29) Jowa, L.; Witz, G.; Snyder, R.; Winkle, S.; Kalf, G. F. Synthesis and characterization of deoxyguanosine-benzoquinone adducts. J. Appl. Toxicol. 1990, 10 (1), 47–54. (30) Hallgren, S.; Darnerud, P. O. Polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs) and chlorinated paraﬃns (CPs) in rats - testing interactions and mechanisms for thyroid hormone eﬀects. Toxicology 2002, 177 (23), 227–243. (31) Wan, Y.; Choi, K.; Kim, S.; Ji, K.; Chang, H.; Wiseman, S.; Jones, P. D.; Khim, J. S.; Park, S.; Park, J.; Lam, M. H. W.; Giesy, J. P. Hydroxylated polybrominated diphenyl ethers and bisphenol A in pregnant women and their matching fetuses: Placental transfer and potential risks. Environ. Sci. Technol. 2010, 44 (13), 5233–5239.

10727

dx.doi.org/10.1021/es203068f |Environ. Sci. Technol. 2011, 45, 10720–10727

ARTICLE pubs.acs.org/est

Emissions Impacts of Wind and Energy Storage in a Market Environment Ramteen Sioshansi*,† †

Department of Integrated Systems Engineering, The Ohio State University, Columbus, Ohio, United States

bS Supporting Information ABSTRACT: This study examines the emissions impacts of adding wind and energy storage to a market-based electric power system. Using Texas as a case study, we demonstrate that market power can greatly eﬀect the emissions beneﬁts of wind, due to most of the coal-ﬁred generation being owned by the two dominant ﬁrms. Wind tends to have less emissions beneﬁts when generators exercise market power, since coal-ﬁred generation is withheld from the market and wind displaces natural gas-ﬁred generators. We also show that storage can have greater negative emissions impacts in the presence of wind than if only storage is added to the system. This is due to wind increasing on- and oﬀ-peak electricity price diﬀerences, which increases the amount that storage and coal-ﬁred generation are used. We demonstrate that this eﬀect is exacerbated by market power.

’ INTRODUCTION Recent years have seen increased interest in renewable electricity in the U.S. and elsewhere. This interest has been driven by several factors, one of which is the emissions and environmental impact of conventional fossil-fueled generation. Wind has provided the bulk of renewable capacity expansion, due to its currently being the lowest-cost technology and the abundance of wind resources. Energy storage is often discussed as an enabling technology that can ease the integration and improve the economic and technical characteristics of wind.113 Denholm et al.14 examine the emissions of a wind generator that uses storage to provide baseload energy. They show that life cycle greenhouse gas emissions from such a baseload wind system can be less than 20% of a combined-cycle natural gasﬁred generator. Denny and O’Malley15 estimate the emissions impacts of wind and storage in the Irish system. Their analysis is based on a perfectly competitive model, in which the operation of conventional generation is co-optimized with wind and storage to minimize system costs. They focus on the impact of wind uncertainty and part-load operation of conventional generators on system emissions. Their results show that wind will have much greater eﬀects in reducing CO2 emissions as compared to SO2 and NOx. One limitation of these analyses is that they neglect interactions between wind, storage, and the market. Since wind participates in wholesale electricity markets,16 a wind generator may prefer using storage to maximize energy revenues as opposed to providing baseload energy. Indeed, storage analyses assume such operations to maximize revenues from charging and discharging energy, an activity referred to as energy arbitrage.1721 Adding wind and storage to a system together can increase this use of storage, since wind tends to suppress r 2011 American Chemical Society

energy prices.2224 This price suppression is due to wind displacing higher-cost generation. Another factor that can inﬂuence the emissions impacts of wind and storage is the competitiveness of the generation sector. Generating ﬁrms exercise market power by withholding capacity from the market.25,26 Thus depending on the ownership of generation and the extent to which diﬀerent ﬁrms have market power, the actual mix of generators used and the type of generation that wind or storage displace can vary. This study examines the emissions eﬀects of wind and storage when accounting for market price eﬀects on storage use. We consider two cases, one with a perfectly competitive generation sector and another in which the two dominant ﬁrms exercise market power. We use an optimization model to represent the interactions between conventional generators, wind, and storage, which is used to derive the dispatch of the system over a one-year period.24 The optimized dispatch is combined with emissions rates estimates to model generator emissions of CO2, SO2, and NOx with and without wind and storage.

’ METHODS Our analysis is based on the Electricity Reliability Council of Texas (ERCOT) system in 2005. ERCOT had about 2 GW of wind installed in 2005, which are included in the base system. We compare the base system to systems with up to 10 GW of added wind and up to 10 GW of storage with up to 20 h of charging capacity. For purposes of comparison, ERCOT had about 83 GW Received: March 3, 2011 Accepted: November 1, 2011 Revised: October 28, 2011 Published: November 01, 2011 10728

dx.doi.org/10.1021/es2007353 | Environ. Sci. Technol. 2011, 45, 10728–10735

Environmental Science & Technology

ARTICLE

of generation capacity installed and a peak load of 60 GW in 2005. Ownership and Market Structure. ERCOT had about 81 GW of conventional (e.g. thermal and hydroelectric) generation installed in 2005, of which about 16 GW were coal-fired, 60 GW natural gas-fired, and the remaining used other fuels.27 These assets were divided between 53 firms. Of these, two firms— Luminant and Texas Genco—owned a large share of about 18% and 14% (on a capacity basis), respectively. Between them, these two firms owned about 65% of the coal-fired capacity in the system. Analyses of the ERCOT market suggest that Luminant and Texas Genco have historically had a greater tendency to exercise market power than the other ﬁrms.28,29 Thus we model wind and storage impacts under two market competitiveness cases: the ﬁrst, which we refer to as the competitive case, assumes that all 53 generating ﬁrms behave perfectly competitively; the other, referred to as the oligopoly case, assumes that Luminant and Texas Genco behave as proﬁt-maximizers while the remaining 51 ﬁrms behave competitively. Further details regarding the breakdown of generation ownership and the market competitiveness cases considered are given in the Supporting Information. Market Operation. In both the competitive and oligopoly cases, we assume that the generating firms submit supply functions, qi,t(p), to a market operator. The function qi,t(p) specifies the maximum amount of energy that firm i is willing to supply in hour t as a function of price. In the competitive case, the supply functions are the inverse of the firms’ marginal cost functions. In the oligopoly case, Luminant and Texas Genco’s supply functions are found by solving a profit-maximization problem, while the remaining firms submit supply functions equal to the inverse of their marginal cost functions. The derivation of these supply functions do not take into account dynamics of conventional generators, such as ramping limits, minimum load constraints, and startup costs. Each firm’s cost function is estimated based on the heat rates of the generators that it owns and fuel prices. Heat rate and fuel price data are obtained from Global Energy Decisions and Platts Energy. We use stepped heat rate functions, which capture differences in a generator’s efficiency as a function of its output. Modeling Wind and Storage. Letting Dt be the system load and Xt net energy sales from wind and storage30 in hour t, the market operator sets the hour-t price of energy as pt ðXt Þ ¼ minfpj p

N

∑ qi, t ðpÞ ¼ Dt Xt g i¼1

ð1Þ

where N is the number of generating firms. Equation 1 defines the price such that it induces exactly enough supply from the conventional generating firms to serve the load net of wind and storage sales. The proﬁt of wind and storage over a T-hour time horizon is given by T

∑

t ¼1

pt ðXt Þ 3 Xt

ð2Þ

We model the behavior of wind and storage by maximizing this proﬁt, subject to technical constraints on the storage plant and the availability of wind energy. Thus even in the competitive generation case, we assume the wind and storage choose their net sales to maximize proﬁts. This allows us to capture the emissions

impacts of competitiveness of the generation sector, without diﬀerences in the assumed behavior of wind and storage confounding the results. Storage constraints include roundtrip eﬃciency losses of the storage system, which we assume to have an 80% roundtrip eﬃciency, and power and energy capacity limits.20,24 This proﬁtmaximization model does not impose any restrictions that only energy from wind be stored. Thus net wind and storage sales could be negative, which would imply that energy is purchased from the market and stored. Further details of this proﬁtmaximization model are given in the Supporting Information. Estimating Emissions. We model emissions associated with the combustion of fossil fuels in generators only. We therefore assume that there are no emissions directly associated with storage use or wind generation. The amount of energy that generating firm i must supply in hour t is given by qi,t(p*t(Xt)). Generator emissions are estimated based on these hourly generation levels using input-based emissions rates, which give kg of each pollutant released per GJ of fuel burned. This is in contrast to output-based emissions rates, which give kg of each pollutant released per MWh of electricity generated. Using input-based emissions rates better accounts for differences in generator heat rates caused by operating a generator at part-load. CO2 emissions rates are assumed to be constant for each generator. To account for the impact of part-load operations on the eﬀectiveness of emissions controls, we assume that the SO2 and NOx emissions rates of each generator can vary as a function of generating load. We approximate these emissions rates using a Nadaraya-Watson kernel estimator.3133 We use separate kernel estimates for NOx emissions rates during an ozone season, which is from May to September, and a nonozone season, which covers the remaining months. This assumption reﬂects the possibility of more stringent NOx restrictions being in place during the ozone season, since NOx is an ozone precursor. Such restrictions may result in greater use of emissions controls. The emissions rates are estimated using continuous emissions monitoring system (CEMS) data for the year 2005 obtained from the U.S. Environmental Protection Agency. The CEMS data record GJ of fuel burned and kg of CO2, SO2, and NOx released by each generator on an hourly basis. Wind Data. We use modeled wind generation data developed by 3TIER for the National Renewable Energy Laboratory’s Western Wind and Solar Integration Study (WWSIS) to model wind generation. This data set provides wind data for 2005 at sites across Texas. We model the 2 GW of wind capacity that were installed in Texas in 2005 by associating each actual wind generator to a location in the WWSIS data set, based on geographic distance. We assume that the additional wind generators that we model are located at the same sites as actual wind generators that were or are planned to be installed between 2005 and 2011. We use the location of these planned installations to associate the incremental wind capacity with locations in the WWSIS data set.

’ RESULTS Emissions Impacts of Wind. Coal is a less costly generation fuel than natural gas. Thus in the competitive case, wherein all of the generators submit cost-based supply functions to the market, coal is used as baseload generation and natural gas is used for any additional load above the capacity of coal-fired plants. The system is not dispatched on the basis of cost in the oligopoly case, however, because the two dominant firms submit supply 10729

dx.doi.org/10.1021/es2007353 |Environ. Sci. Technol. 2011, 45, 10728–10735

Environmental Science & Technology

ARTICLE

Figure 1. Total annual reduction in generator emissions of CO2, SO2, and NOx when wind is added to the system. CO2 reductions are reported in Mt, and SO2 and NOx reduction are reported in kt. Part a shows emissions reductions in the competitive case and part b shows emissions reductions in the oligopoly case.

Figure 2. Total annual increase in generator emissions of CO2, SO2, and NOx when energy storage with four hours of charging capacity is added to the base system with no additional wind. CO2 increases are reported in kt, and SO2 and NOx increases are reported in t.

functions that are above their marginal costs. These above-cost supply functions act to withhold some of the dominant firms’ generating capacity from the market, forcing the market operator to use higher-cost generation which increases energy prices and firm profits. Because the dominant firms own much of the coalfired generation, this withholding causes differences in the breakdown of the generation load. In the base system, coal constitutes about 46.1% of fossil-fueled generation in the competitive case as opposed to 45.8% in an oligopoly. The withholding of coal-fired generation occurs during low-load periods, in which the dominant firms’ natural gas-fired generators are shutdown. By submitting above-cost bids for their coal-fired generators, the dominant firms force the market operator to use more (of the dominant or competitive firms’) natural gas-fired generation, increasing the price of energy. This greater use of coal

gives higher emissions in the competitive case due to the higher emissions rates of coal—CO2, SO2, and NOx emissions are 235 Mt, 461 kt, and 202 kt, respectively, in the competitive case as opposed to 230 Mt, 448 kt, and 171 kt in an oligopoly. These diﬀerences in the dispatch also aﬀect the emissions reductions when wind is added to the system. Figure 1 shows annual emissions reductions when wind is added to the base system. The ﬁgure shows that CO2 and NOx reductions are roughly linear in the amount of wind added to the system and that the emissions reductions are comparable between the competitive and oligopoly cases. Marginal SO2 reductions are, on the other hand, increasing in the amount of wind added to the system. Wind also has a greater impact in reducing SO2 emissions in the competitive case. The diﬀerences in SO2 reductions are due to the impact of wind on natural gas- as opposed to coal-ﬁred generation. Because of capacity withholding in an oligopoly, there are fewer hours, compared to the competitive case, in which coal-ﬁred generation is marginal and displaced by wind. Thus, the ﬁrst 5.5 GW of wind added to the system have a relatively modest eﬀect with an average of 111 MWh of coal-ﬁred generation being displaced annually per MW of added wind capacity. The same 5.5 GW of wind have a much greater impact in the competitive case, with 896 MWh of coal-ﬁred generation being displaced on average per MW of wind. Additional wind beyond the ﬁrst 5.5 GW have a greater impact, however, since at suﬃciently high penetrations coal-ﬁred generation will increasingly be marginal and displaced. Each additional MW of wind beyond the ﬁrst 5.5 GW results in annual coal-ﬁred generation reductions of between 145 MWh and 389 MWh in the oligopoly case. This incremental wind has an even greater impact in the competitive case, however, with annual coal-ﬁred generation reductions of between 1,097 MWh and 1,208 MWh per MW of added wind. Emissions Impacts of Storage. Figure 2 shows annual emissions increases when storage with four hours of charging capacity is added to the system and used for arbitrage. The trends are similar for different numbers of charging hours. In all of the 10730

dx.doi.org/10.1021/es2007353 |Environ. Sci. Technol. 2011, 45, 10728–10735

Environmental Science & Technology

Figure 3. Hourly output of a 10 GW wind plant, net sales from a 5 GW storage plant with four hours of charging capacity, and resulting changes in conventional generation on 15 January. Storage use is assumed to be co-optimized with wind to maximize total proﬁt. Bars show the change in total natural gas-ﬁred generation between the wind-only and wind-andstorage cases, as a percentage of the change in total conventional generation between the two cases.

cases CO2 and SO2 emissions increase when storage is added, whereas NOx emissions decrease in some cases. The emissions increases are due to two effects. One is that more energy must be generated, due to roundtrip efficiency losses of storage. The other is that storage is used to arbitrage price differences between on- and off-peak hours. In the competitive case, much of this arbitraging is done between coal and natural gas-fired generation. Coal-fired generators provide between 38% and 50% of the incremental generation when energy is charged into storage, whereas more than 98% of the generation displaced when storage is discharged is natural gas-fired. Due to the exercise of market power, coal-fired generation provides less than 5% of the energy stored in the oligopoly case. In this case, storage is largely arbitraging price differences between more-efficient combinedcycle and less-efficient simple-cycle natural gas plants. This difference in the generation used to provide the charging energy explains the significantly higher SO 2 increases in the competitive case. These ﬁndings of increased emissions and the sensitivity to the generating fuels used and displaced when storage is charged and discharged are consistent with other storage analyses. Denholm and Holloway34 examine the emissions impact of compressed-air energy storage (CAES) in Ohio. They show that since CAES could be charged using coal-ﬁred generation and displace natural gas-ﬁred generation when discharged, the net emissions of CAES could be greater than a new coal-ﬁred generator. They also show that CAES could have signiﬁcantly lower emissions if storage is charged using cleaner generation, such as nuclear, renewable, or new coal that meets 2004 Clean Air Act standards. Figure 2 also shows that emissions will not necessarily be monotone in the power capacity of the storage plant. This is because storage will aﬀect the output of marginal generators, which can change their emissions rates. For instance, there is an approximately 1 t reduction in SO2 emissions in the oligopoly case between 5 GW and 5.5 GW of storage. This diﬀerence is due to a 5.5 GW storage plant doing more arbitrage than a 5 GW

ARTICLE

plant on 30 August. This increased arbitraging results in two of the coal-ﬁred plants shifting their generation between hours with diﬀerent emissions rates. This type of emissions ﬂuctuation is likely speciﬁc to the 2005 data used in our case study and should not be interpreted as a general result that will occur in all years. Storage has similar eﬀects on NOx and decreases NOx emissions in some cases. This is because the shifting of generating loads results in marginal generators having lower emissions rates. These lower rates yield a NOx reduction, which outweighs the emissions increase caused by greater generation and the arbitraging eﬀect. Joint Emissions Impacts of Wind and Storage. Adding wind and storage to a system together increases storage use compared to the storage-only case. This is because wind suppresses energy prices by displacing high-cost generation from the market. Since this price effect is associated with wind availability and hourly wind availability can be highly variable, wind increases hourly price differences and arbitrage opportunities. Our analysis assumes joint ownership of wind and storage; however, the same effects persist in a disjoint-ownership case and storage use and emissions impacts will largely be the same in the two cases. This is because wind will have the same price-suppressing impact regardless of storage ownership. Figure 3 shows, as an example, the operation of 5 GW of storage with four hours of charging capacity on a sample day in a system with 10 GW of added wind. It shows hourly wind generation and total net sales from wind and storage when storage use is optimized to maximize energy revenues. Comparing the hourly wind output and net sales proﬁles shows that storage is used extensively on this day. About 11 GWh of energy are stored in the morning and afternoon when wind suppresses energy prices. This energy is later discharged in the late morning and evening when wind generation is lower and prices are higher. The ﬁgure also shows the breakdown of the change in hourly conventional generation caused by storage. This is shown as the change in natural gas-ﬁred generation between the wind-only and wind-and-storage cases, as a percentage of the change in total conventional generation between these two cases. Coal-ﬁred generation provides roughly a third of the incremental energy when storage is charged on this day in the competitive case, as opposed to only 13% in an oligopoly. There are also diﬀerences in the generation that is displaced when storage is discharged— roughly 88% of the displaced generation is natural gas-ﬁred in the competitive case as opposed to 99% in an oligopoly. These types of diﬀerences persist throughout the year and with diﬀerent wind and storage penetrations. Figure 4 shows the marginal eﬀect of charging storage on conventional generation. The ﬁgure shows the change in natural gas-ﬁred generation between wind-only and wind-and-storage cases during hours of the year in which storage is charged, as a percentage of the total change in conventional generation during those hours. The shading of the circles and squares is based on the energy capacity of the added storage—lighter shading indicates more storage. The ﬁgure shows that coal-ﬁred generation tends to provide more of the incremental generation when storage is charged in the competitive case, due to it being marginal in more hours. The diﬀerences in the composition of the charging load between the competitive and oligopoly cases decreases as more wind is added, however. This is because adding more wind in the oligopoly case will increasingly displace coal-ﬁred generation, making coal the marginal generating fuel in more low-price hours. There are also small diﬀerences in the breakdown of the 10731

dx.doi.org/10.1021/es2007353 |Environ. Sci. Technol. 2011, 45, 10728–10735

Environmental Science & Technology

ARTICLE

Figure 4. Incremental natural gas-ﬁred generation when storage is charged, as a percentage of change in total conventional generation when storage is charged. Generation changes shown are between wind-only and wind-and-storage cases. The shading of the circles and squares are based on the energy capacity of the added storage—lighter shading indicates more storage. Part a shows generation changes in the competitive case and part b shows generation changes in the oligopoly case.

Figure 5. Total annual increase in generator emissions of CO2, SO2, and NOx when energy storage with four hours of charging capacity is added to a system with 10 GW of added wind. Emissions increases are relative to the wind-only case. CO2 increases are reported in kt, and SO2 and NOx increases are reported in t.

conventional generation that is displaced when storage is discharged. Between 84% and 99% of the generation displaced when storage is discharged is natural gas-ﬁred in the competitive case, whereas this number is always above 97% in an oligopoly. Figure 5 shows the eﬀect of these diﬀerences in incremental generation when storage is charged and discharged on the net emissions impact of adding storage and wind to a system together. The ﬁgure shows emissions increases between a case with 10 GW of added wind only and a case with 10 GW of wind and storage with four hours of charging capacity. With the exception of NOx in the oligopoly case, the combination of wind and storage causes all emissions to increase. The ﬁgure shows that the emissions impacts of wind and storage together are highly sensitive to amount of wind and storage added. For

instance, there are greater SO2 increases in the competitive as opposed to oligopoly case if less than 8 GW of storage is added, whereas these impacts are reversed for larger amounts of storage. This is because in an oligopoly with less than 5 GW of storage, coal-ﬁred generation provides about 21% of the incremental energy when storage is charged. As storage capacity increases, coal’s share of charging energy drops to below 16%. In the competitive case, however, coal always provides between 25% and 29% of the added load when storage is charged. Thus as more storage is added to the system, storage is decreasingly arbitraging between coal- and natural gas-ﬁred generation in an oligopoly, reducing its SO2 impact relative to the competitive case. Comparing the range of emissions increases in Figure 2 and Figure 5 shows that the emissions impacts of storage in a system 10732

dx.doi.org/10.1021/es2007353 |Environ. Sci. Technol. 2011, 45, 10728–10735

Environmental Science & Technology

ARTICLE

Figure 6. Total annual increase in generator emissions of CO2, SO2, and NOx when storage is added to a system with 10 GW of wind. The ﬁgures show the increase between wind-and-storage and wind-only cases, less the emissions increases between storage-only and base cases. CO2 increases are reported in kt and SO2 and NOx increases are reported in t.

with wind are, in some cases, 2 orders of magnitude greater compared to if there is no added wind. This is because wind signiﬁcantly increases arbitrage opportunities for storage. For instance, 5 GW of storage with four hours of charging capacity

stores about 321 GWh and 529 GWh of energy in the competitive and oligopoly cases, respectively, if no wind is added to the system. The same storage plant stores about 3.4 TWh and 3.6 TWh of energy in the competitive and oligopoly cases, 10733

dx.doi.org/10.1021/es2007353 |Environ. Sci. Technol. 2011, 45, 10728–10735

Environmental Science & Technology

ARTICLE

respectively, if 10 GW of wind is also added to the system. Thus the combination of wind and storage yields superadditive emissions impacts relative to the impacts of introducing the two technologies to the system individually. We can measure this superadditive eﬀect by deﬁning Mu, a

ð3Þ

as the annual emissions of pollutant u under deployment scenario a, where a denotes either the base (a = B), storage-only (a = S), wind-only (a = W), or wind-and-storage (a = WS) case. We can then measure the superadditive eﬀect of adding wind and storage together as ξu ¼ ðMu, WS Mu, W Þ ðMu, S Mu, B Þ

ð4Þ

which is the emissions increase between the wind-and-storage and wind-only cases, less the emissions increase between storageonly and base cases. Thus ξu measures the extent to which storage impacts generator emissions due to the increased arbitrage opportunities created by wind. Figure 6 summarizes the values of ξu with diﬀerent amounts of storage and 10 GW of wind. The increases in the competitive case are relatively small compared to the emissions beneﬁts of wind, representing up to 8% of the emissions savings from introducing wind to the system. The combination of wind and storage have much more pronounced eﬀects in the oligopoly case, however. The SO2 increases represent up to 24% of the SO2 reductions from wind, thus the combination of wind and storage can eliminate close to a quarter of the SO2 savings from wind. On the other hand, wind and storage together reduce NOx emissions compared to the wind-only case. This is because the changes in conventional generator loads improves the NOx emissions rates of marginal generators, yielding a net NOx reduction that outweighs increased emissions due to higher generating loads.

’ DISCUSSION Although storage is discussed as a technology to improve the characteristics of wind, these results show that storage and wind can interact in ways that increase the emissions impact of storage. Conventional generator ownership, market competitiveness, and the penetration of wind and storage can substantively change the emissions impacts of these technologies individually and together. Our analysis assumes up to 10 GW of wind is added to a base system with 2 GW of wind. ERCOT has close to 10 GW of wind installed today, thus the impacts of an additional 10 GW on top of this would be diﬀerent than our estimates. For example, it is likely that wind would have a greater impact on SO2 emissions, since the relatively high wind penetrations would result in coalﬁred generation being marginal in signiﬁcantly more hours. Although our case study is based on the ERCOT system, our results should be viewed as illustrative. This is because the ERCOT market is not perfectly competitive, nor do the two dominant ﬁrms fully behave as proﬁt-maximizers. Thus the competitive and oligopoly cases should be viewed as providing bounds on the impacts of wind and storage. Some of the emissions ﬂuctuations (e.g. nonsmooth and nonmonotone emissions impacts of wind and storage) are possibly speciﬁc to the 2005 data that we base our analysis on and may not be general results. Nevertheless, the ﬁndings regarding shifting of generation between generating fuels and technologies would likely occur in other systems. This is because marginal generating technologies and emissions rates can diﬀer by time of day and

also be sensitive to market competitiveness. For instance, California has virtually no coal-ﬁred generation. Nevertheless, hourly marginal emissions rates can vary depending on whether combined- or simple-cycle natural gas-ﬁred generation is marginal.35 Our analysis assumes joint ownership of wind and storage, because storage is considerably more valuable to a wind generator than to a standalone storage operator or conventional generator.24,36 As noted before, storage use and emissions impacts would largely be the same with disjoint ownership of wind and storage. Our joint-ownership assumption should not, however, be taken to suggest that wind and storage must or should be jointly owned. Our analysis further assumes that wind and storage are owned by a single proﬁt-maximizing ﬁrm. Although wind ownership was rather concentrated in 2005 (Table S3 in the Supporting Information summarizes wind ownership), this assumption may overstate the extent to which wind and storage can exercise market power by adjusting net sales to maximize proﬁts. Relaxing this assumption would not aﬀect wind generation, since wind is never curtailed in our model. Storage use could increase, however, since it is proﬁt-maximizing to reduce storage use from a competitive level to maintain higher price diﬀerences between on- and oﬀ-peak periods.20,36 Based on our ﬁndings, it is likely that this greater use of storage would yield higher generator emissions. Our model does not consider operational impacts that wind and storage can have on power systems. Storage can provide valuable renewable integration services, such as reducing the need for transmission expansions and wind curtailment.13,37 These types of interactions between wind and the power system arise due to dynamics of conventional generators, such as ramping limits, minimum load constraints, and startup costs, that cannot be accommodated in the model that we use. Some of these services can decrease system emissions. For example, wind generation variability can require ineﬃcient fast-ramping generators, that often have high emissions rates, to follow wind supply.38 If storage can reduce the variability of wind, this can reduce the need for such generation and the associated emissions. Storage can also increase the proﬁtability of a wind generator, which could spur or encourage further wind capacity to enter the market.24 Since these types of beneﬁts are not directly captured in our model, such gains should be weighed against the impacts that we estimate here.

’ ASSOCIATED CONTENT

bS

Supporting Information. A summary overview of the ERCOT market structure and more detailed description of our market and storage modeling methodology and assumptions. This material is available free of charge via the Internet at http:// pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT Thank you to Paul Denholm, Armin Sorooshian, the editor, and four anonymous reviewers for helpful comments and discussions. Tony Grasso of the Public Utility Commission of Texas provided invaluable assistance in data collection. 10734

dx.doi.org/10.1021/es2007353 |Environ. Sci. Technol. 2011, 45, 10728–10735

Environmental Science & Technology

’ REFERENCES (1) Sørensen, B. A combined wind and hydro power system. Energy Policy 1981, 9, 51–55. (2) Cavallo, A. J. High-Capacity Factor Wind Energy Systems. J. Solar Energy Eng. 1995, 117, 137–143. (3) Study of Electric Transmission in Conjunction with Energy Storage Technology. Lower Colorado River Authority: Austin, Texas, 2003; Prepared for Texas State Energy Conservation Oﬃce. (4) Paatero, J. V.; Lund, P. D. Eﬀect of energy storage on variations in wind power. Wind Energy 2005, 8, 421–441. (5) DeCarolis, J. F.; Keith, D. W. The economics of large-scale wind power in a carbon constrained world. Energy Policy 2006, 34, 395–410. (6) Succar, S.; Greenblatt, J. B.; Denkenberger, D. C.; Williams, R. H. An Integrated Optimization Of Large-Scale Wind With Variable Rating Coupled To Compressed Air Energy Storage 2006. (7) Greenblatt, J. B.; Succar, S.; Denkenberger, D. C.; Williams, R. H.; Socolow, R. H. Baseload wind energy: modeling the competition between gas turbines and compressed air energy storage for supplemental generation. Energy Policy 2007, 35, 1474–1492. (8) Swider, D. J. Compressed Air Energy Storage in an Electricity SystemWith SigniﬁcantWind Power Generation. IEEE Trans. Energy Convers. 2007, 22, 95–102. (9) Black, M.; Strbac, G. Value of Bulk Energy Storage for ManagingWind Power Fluctuations. IEEE Trans. Energy Convers. 2007, 22, 197–205. (10) Abbey, C.; Joos, G. Supercapacitor Energy Storage for Wind Energy Applications. IEEE Trans. Ind. Appl. 2007, 43, 769–776. (11) García-Gonzalez, J.; de la Muela, R. M. R.; Santos, L. M.; Gonzalez, A. M. Stochastic Joint Optimization of Wind Generation and Pumped-Storage Units in an Electricity Market. IEEE Trans. Power Syst. 2008, 23, 460–468. (12) Arsie, I.; Marano, V.; Rizzo, G.; Moran, M. Integration of Wind Turbines with Compressed Air Energy Storage 2009. (13) Denholm, P.; Sioshansi, R. The value of compressed air energy storage with wind in transmission-constrained electric power systems. Energy Policy 2009, 37, 3149–3158. (14) Denholm, P.; Kulcinski, G. L.; Holloway, T. Emissions and energy eﬃciency assessment of baseload wind energy systems. Environ. Sci. Technol. 2005, 39, 1903–1911. (15) Denny, E.; O’Malley, M. Wind Generation, Power System Operation, and Emissions Reduction. IEEE Trans. Power Syst. 2006, 21, 341–347. (16) Sioshansi, R.; Hurlbut, D. Market Protocols in ERCOT and Their Eﬀect onWind Generation. Energy Policy 2010, 38, 3192–3197. (17) Graves, F.; Jenkin, T.; Murphy, D. Opportunities for Electricity Storage in Deregulating Markets. Electr. J. 1999, 12, 46–56. (18) Figueiredo, F. C.; Flynn, P. C.; Cabral, E. A. The Economics of Energy Storage in 14 Deregulated Power Markets. Energy Studies Rev. 2006, 14, 131–152. (19) Walawalkar, R.; Apt, J.; Mancini, R. Economics of electric energy storage for energy arbitrage and regulation in New York. Energy Policy 2007, 35, 2558–2568. (20) Sioshansi, R.; Denholm, P.; Jenkin, T.; Weiss, J. Estimating the Value of Electricity Storage in PJM: Arbitrage and Some Welfare Eﬀects. Energy Econ. 2009, 31, 269–277. (21) Sioshansi, R.; Denholm, P.; Jenkin, T. A. Comparative Analysis of the Value of Pure and Hybrid Electricity Storage. Energy Econ. 2011, 33, 56–66. (22) Green, R. J.; Vasilakos, N. Market behaviour with large amounts of intermittent generation. Energy Policy 2010, 38, 3211–3220. (23) Twomey, P.; Neuhoﬀ, K. Wind power and market power in competitive markets. Energy Policy 2010, 38, 3198–3210. (24) Sioshansi, R. Increasing the Value of Wind with Energy Storage. Energy J. 2011, 32, 1–30. (25) Borenstein, S.; Bushnell, J. B. An Empirical Analysis of the Potential for Market Power in California’s Electricity Industry. J. Ind. Econ. 1999, 47, 285–323.

ARTICLE

(26) Borenstein, S.; Bushnell, J. B.; Knittel, C. R. Market Power in Electricity Markets: Beyond Concentration Measures. Energy J. 1999, 20, 65–89. (27) Based on values reported in Form 860 data reported by the U.S. Department of Energy’s Energy Information Administration. (28) Sioshansi, R.; Oren, S. How good are supply function equilibrium models: an empirical analysis of the ERCOT balancing market. J. Regul. Econ. 2007, 31, 1–35. (29) Hortac-su, A.; Puller, S. L. Understanding Strategic Bidding in Multi-Unit Auctions: A Case Study of the Texas Electricity Spot Market. RAND J. Econ. 2008, 39, 86–114. (30) The term ‘net energy sales’ emphasizes the fact that wind and storage may have negative sales if more energy is charged into storage than the amount of wind generated. This can occur because we do not impose any restrictions that only wind generation be stored. (31) Nadaraya, E. A. On Estimating Regression. Theory Probab. Its Appl. (Engl. Transl.) 1964, 9, 141–142. (32) Watson, G. S. Smooth Regression Analysis. Sankhya, Ser. A 1964, 26, 359–372. (33) Sioshansi, R.; Denholm, P. Emissions Impacts and Beneﬁts of Plug-in Hybrid Electric Vehicles and Vehicle to Grid Services. Environ. Sci. Technol. 2009, 43, 1199–1204. (34) Denholm, P.; Holloway, T. Improved Accounting of Emissions from Utility Energy Storage System Operation. Environ. Sci. Technol. 2005, 39, 9016–9022. (35) McCarthy, R.; Yang, C. Determining marginal electricity for near-term plug-in and fuel cell vehicle demands in California: Impacts on vehicle greenhouse gas emissions. J. Power Sources 2010, 195, 2099– 2109. (36) Sioshansi, R. Welfare Impacts of Electricity Storage and the Implications of Ownership Structure. Energy J. 2010, 31, 189–214. (37) Tuohy, A.; O’Malley, M. Pumped storage in systems with very high wind penetration. Energy Policy 2011, 39, 1965–1974. (38) Katzenstein, W.; Apt, J. Air Emissions Due To Wind And Solar Power. Environ. Sci. Technol. 2009, 43, 253–258. (39) Klemperer, P. D.; Meyer, M. A. Supply Function Equilibria in Oligopoly Under Uncertainty. Econometrica 1989, 56, 1243–1277. (40) Green, R. J.; Newbery, D. M. Competition in the British Electricity Spot Market. J. Political Econ. 1992, 100, 929–953. (41) Green, R. J. Increasing Competition in the British Electricity Spot Market. J. Ind. Econ. 1996, 44, 205–216. (42) Newbery, D. M. Competition, Contracts, and Entry in the Electricity SpotMarket. RAND J. Econ. 1998, 29, 726–749. (43) Green, R. J. The Electricity Contract Market in England and Wales. J. Ind. Econ. 1999, 47, 107–124. (44) Green, R. J. Carbon trading or carbon taxes: the impact on generators’ risks. Energy J. 2008, 29, 67–89.

10735

dx.doi.org/10.1021/es2007353 |Environ. Sci. Technol. 2011, 45, 10728–10735

ARTICLE pubs.acs.org/est

Impact of Aviation Non-CO2 Combustion Effects on the Environmental Feasibility of Alternative Jet Fuels Russell W. Stratton, Philip J. Wolfe, and James I. Hileman* Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts, 02139

bS Supporting Information ABSTRACT:

Alternative fuels represent a potential option for reducing the climate impacts of the aviation sector. The climate impacts of alternatives fuel are traditionally considered as a ratio of life cycle greenhouse gas (GHG) emissions to those of the displaced petroleum product; however, this ignores the climate impacts of the non-CO2 combustion eﬀects from aircraft in the upper atmosphere. The results of this study show that including non-CO2 combustion emissions and eﬀects in the life cycle of a Synthetic Paraﬃnic Kerosene (SPK) fuel can lead to a decrease in the relative merit of the SPK fuel relative to conventional jet fuel. For example, an SPK fuel option with zero life cycle GHG emissions would oﬀer a 100% reduction in GHG emissions but only a 48% reduction in actual climate impact using a 100-year time window and the nominal climate modeling assumption set outlined herein. Therefore, climate change mitigation policies for aviation that rely exclusively on relative well-to-wake life cycle GHG emissions as a proxy for aviation climate impact may overestimate the beneﬁt of alternative fuel use on the global climate system.

1. INTRODUCTION Aviation plays a key role in the global economy and transportation systems. Projections indicate that over the next 2 decades, the demand for aviation within the U.S. will grow at roughly 2% per annum.1 Mitigating climate change from the aviation sector can be simpliﬁed to consuming less energy, through improvements in aircraft technology or operational eﬃciency, and reducing the climate impacts of the energy source, through the use of alternative fuels. Synthetic Paraﬃnic Kerosene, as deﬁned herein, has been certiﬁed for jet aircraft use up to a blend ratio of 50% and can be created via (1) Gasiﬁcation of coal, natural gas, or biomass to form synthesis gas, which is processed using FischerTropsch (FT) synthesis, and subsequently upgraded to a product slate that includes a synthetic jet fuel; and (2) Hydroprocessing of renewable oils, such as those created from jatropha, camelina, and algae, among many others, to create a hydroprocessed renewable jet (HRJ) fuel.2,3 These fuels, and others being considered for aviation, diﬀer from conventional jet fuel in that they are comprised solely of paraﬃnic hydrocarbons and contain neither aromatic compounds nor sulfur 4,5 The potential of a particular fuel to reduce greenhouse gas emissions (GHG) is generally assessed through a comparison of the life cycle GHG emissions inventory of the alternative fuel with that of the fuel it is intended to displace.2,6 A life cycle GHG r 2011 American Chemical Society

emissions inventory encompasses emissions from recovery and transportation of the feedstock to the production facility, processing of these materials into fuels, transportation and distribution of the fuel to the aircraft tank, and ﬁnally, the combustion of the fuel in the aircraft. The term “well-to-wake” is used to describe the life cycle GHG inventory of aviation fuels. Life cycle analyses of biobased, ground transportation fuels assume that the emissions from fuel combustion are equal and opposite to the emissions absorbed from the atmosphere during growth of the feedstock.2,58 However, this approach neglects non-CO2 combustion emissions and eﬀects, namely, soot and sulfate aerosols, water vapor, and NOx. Aviation also causes contrails and induced cirrus clouds called contrail cirrus. Such products will exist even if the net GHG emissions from the fuel life cycle are zero. Figure 1 schematically demonstrates the life cycle impacts pathway of aviation related climate change using a biobased fuel starting with emissions from fuel production and fuel combustion in the engine and culminating in societal impacts. Received: May 23, 2011 Accepted: October 3, 2011 Revised: September 25, 2011 Published: November 22, 2011 10736

dx.doi.org/10.1021/es2017522 | Environ. Sci. Technol. 2011, 45, 10736–10743

Environmental Science & Technology

ARTICLE

Figure 1. Aviation climate change impacts pathway (adapted from ref 9).

Soot and sulfate aerosols generate atmospheric warming and cooling, respectively, and have lifetimes on the order of days to weeks.10 Contrails and contrail cirrus sustain only for hours to days and cause atmospheric warming;9,11 however, their impact is the most uncertain of all aviation induced climate forcing.9 NOx results in both short-term warming and long-term cooling. In the months following a pulse of NOx in the upper atmosphere, ozone production is stimulated causing a short-term warming. NOx emissions also stimulate the production of additional OH, acting as a sink for CH4. The corresponding reduction in CH4, which is an important ozone precursor, leads to a long-term reduction in ozone. Both the long-term reduction in CH4 and ozone cool the atmosphere and decay with a lifetime of approximately 11 years.12 The purely paraﬃnic nature and lack of sulfur present in SPK fuels has been shown to cause changes in the combustion emissions from gas turbine engines;1316 hence, the purpose of this paper is 2-fold: (1) develop ratios by which the CO2 from combustion can be scaled to include the climate forcing from non-CO2 combustion eﬀects of conventional jet fuel and SPK, and (2) quantify how including non-CO2 combustion species within the fuel life cycle changes the merit of alternative jet fuels relative to conventional jet fuel from the perspective of climate change. Select alternative jet fuel life cycle GHG inventories, as developed by Stratton et al.2 and documented in the Supporting Information (SI), have been leveraged herein as examples of how non-CO2 combustion eﬀects change the climate change mitigation potential of alternative fuel options. The SI also contains additional details on the underlying methodology adopted in the creation of these life cycle GHG inventories; however, the conclusions of this work are independent of the life cycle GHG inventories to which the climate forcing from non-CO2 combustion eﬀects are added.

2. MATERIALS AND METHODS This paper implements a modiﬁed version of the climate impacts module of the Aviation Portfolio Management Tool (APMT) to establish a “basis of equivalence” between emissions of diﬀerent species, such that the climate impacts of non-CO2 combustion emissions and eﬀects can be related to those of CO2. This process is described in further detail in Sections 2.1 through 2.4. 2.1. Modeling Framework. The APMT climate module has been documented and tested in the literature.1719 The model is based on the Bern carbon-cycle impulse response function with a simplified analytical temperature change model to estimate climate impacts for aviation CO2 and non-CO2 effects. The temporal resolution is limited to one year while the spatial resolution is an aggregated global mean level. Inputs to APMT Climate are a background emissions scenario, a demand scenario for aviation fuel burn and corresponding emissions inventories for CO2 and NOx. Radiative forcing estimates for NOx were obtained by linearly scaling radiative forcing estimates from the literature based on NOx emissions because the shortlived nature of the species inhibits a well-deﬁned gas-cycle like that for carbon.12,20,21 All short-lived eﬀects (aerosols, H2O, contrails and contrail cirrus) are scaled linearly with fuel burn levels based on radiative forcing estimates from the literature.2224 Outputs from the model are temporal proﬁles of RF and change in global mean average temperature from each forcing agent. It is important to note the treatment of climate forcing from aircraft induced cloudiness (AIC). AIC encompasses both contrails and contrail cirrus and its treatment is based on results from Lee et al.22 Instead of having distinct contrail and contrail cirrus climate forcings, results are given for contrails and for AIC. This treatment stems from a coupling between the uncertainty levels of contrails and total AIC. 10737

dx.doi.org/10.1021/es2017522 |Environ. Sci. Technol. 2011, 45, 10736–10743

Environmental Science & Technology

ARTICLE

Table 1. Normalized Emissions Characteristics of SPK Fuel Relative to Conventional Jet Fuel. All Are on a Mass Basis except Energy Density, Which Is on a Volumetric Basis fuel characteristics

SPK relative to conventional jet fuel

speciﬁc energy

1.023

energy density

0.963

combustion CO2

0.98

water vapor

1.11

sulfate aerosols soot aerosols

0.0 0.050.4

nitrogen oxides

0.91.0

contrails AIC

1.0

2.2. Model Uncertainty. Monte Carlo methods were used to capture uncertainties in inputs and model parameters. This requires expressing inputs and parameters as distributions where possible. As described by Mahashabde et al.,18 in order to extract meaningful insights about the possible costs and benefits of a policy, it is helpful if the analysis options are synthesized into a set of predefined combinations of inputs and assumptions. These combinations of inputs and model parameters each describe a particular point of view or perspective on analysis. Each of these combinations is designated as a lens as it symbolizes a particular viewpoint through which one can assess environmental and economic impacts. There are currently three lenses implemented in the APMT climate module; namely, low impacts, nominal impacts and high impacts. Each lens corresponds to the use of different values or distributions for the most influential parameters in the climate module. Parameters captured in the lenses are the projected growth of aviation, climate forcing of each nonCO2 effect, climate sensitivity and a climate damage coefficient. While APMT would normally allow the background CO2 concentration to vary between lenses, this work uniformly adopted a constant background CO2 concentration of 378 ppm to maintain consistency with existing IPCC GWP calculations (even though the current atmospheric CO2 concentration has subsequently risen to 390 ppm) In a manner that parallels the lenses described above, Stratton et al.2,6 developed a new methodological approach for constructing life cycle GHG inventories of transportation fuels. In it, key parameters were identiﬁed through examination of the GHG emissions resulting from each life cycle step. Optimistic, nominal and pessimistic sets of these key parameters were developed and used to formulate corresponding low GHG inventories, baseline or nominal GHG inventories, and high GHG inventories using attributional life cycle analysis (LCA). The low lens, nominal lens and high lens of APMT, which were used to assess tank-to-wake combustion emissions, mirror the formulation of the low, baseline and high well-to-tank GHG emissions inventories. 2.3. Combustion of SPK Fuel Compared to Conventional Jet Fuel. The purely paraffinic nature and lack of sulfur in SPK fuels result in increased specific energy, decreased energy density and changes to the emissions characteristics of CO2, H2O, soot, sulfates and NOx.4,1316 Therefore, independent non-CO2 ratios are required for conventional jet fuel and SPK fuel. The changes in combustion properties between conventional jet fuel and SPK fuel are summarized in Table 1 and were implemented into the APMT climate impacts module.

Figure 2. Non-CO2 ratios disaggregated by species and time window for conventional jet fuel and SPK fuel.

A detailed characterization of conventional and synthetic jet fuel by Hileman et al.4 was used to determine changes in speciﬁc energy, energy density and CO2 emissions. Water vapor emissions were modiﬁed based on the carbon to hydrogen ratio of Jet A and SPK fuel. Synthetic fuels contain negligible quantities of sulfur so all sulfate emissions were eliminated. Changes in soot and NOX emissions were represented as probabilistic distributions to compliment the existing lens framework of APMT and reﬂect reduced certainty. The formation of contrails and contrail cirrus were assumed unchanged by the use of SPK fuel. Further characterization of combustion emissions and eﬀects of SPK fuel relative to conventional jet fuel is available in the SI. 2.4. Climate Metrics and Functional Unit. Well-to-wake GHG emissions are presented per unit of energy (lower heating value) consumed by the aircraft. The life cycle emissions of a fuel pathway can be presented either with or without the inclusion of climate impacts from non-CO2 combustion emissions and effects. When non-CO2 combustion emissions and effects are ignored, the emissions inventory is a pure GHG emissions inventory composed of CO2, CH4 and N2O. Merging non-CO2 combustion emissions and effects into a fuel life cycle GHG inventory requires them to be presented per megajoule (LHV) of fuel consumed by the aircraft. This work developed non-CO2 ratios to scale the CO2 from combustion to account for the climate forcing from non-CO2 combustion emissions. This approach draws from the process and results developed for conventional jet fuel by Dorbian et al.25 Although non-CO2 combustion emissions and effects have climate impacts that have been represented in terms of CO2, they are not themselves greenhouse gases (with the exception of water vapor). As such, integrating the non-CO2 combustion emissions and effects into a GHG life cycle inventory results in a combination of a GHG inventory and an impact analysis. The terminology “well-to-wake (+)” is introduced here to identify the combination of CO2 and non-CO2 effects from fuel combustion in aircraft. Global warming potentials (GWP) are commonly used to express the impact of long-lived gases such as CH4 and N2O in terms of a carbon dioxide equivalent.24,26,27 The time window over 10738

dx.doi.org/10.1021/es2017522 |Environ. Sci. Technol. 2011, 45, 10736–10743

Environmental Science & Technology

ARTICLE

which the radiative forcing is integrated is a value judgment rather than a matter of science, although 100 years is commonly chosen. Other metrics are also available to relate the climate forcing of one substance to another.26,27 Clearly stating the chosen metric and time window is essential in all situations. This is particularly important when assessing the climate impacts of non-CO2 combustion emissions and eﬀects within a greenhouse gas inventory that includes CH4 and N2O. For example, the use of 100-year global warming potentials to express the CH4 and N2O in terms of CO2 equivalent should require the use of 100year integrated radiative forcing to assess the non-CO2 combustion emissions eﬀects.

The IPCC provides GWP values for CH4 and N2O over time windows of 20 years, 100 years and 500 years.24 Maintaining consistency with the treatment of life cycle CH4 and N2O limits the assessment of non-CO2 combustion emissions and eﬀects in this work to integrated-radiative forcing over 20 year, 100 year, and 500 year time windows. The non-CO2 ratio for a given time window, Δt, and its integration with a fuel life cycle GHG inventory to form a well-to-wake (+) inventory is given by the equations below. Each RF element is the aggregated radiative forcing from all direct and indirect mechanisms through which that species or eﬀect inﬂuences the climate system.

ðCO2 eÞwell-to-wake ¼ ðCO2 þ CH4 3 GWPCH4 þ N2 O 3 GWPN2 O ÞFuel

Production

þ ðCO2 ÞCombustion

ðCO2 eÞwell-to-wakeðþÞ ¼ ðCO2 þ CH4 3 GWPCH4 þ N2 O 3 GWPN2 O ÞFuel Production þ ðCO2 ÞCombustion 3 ðnon-CO2 ratioÞ Z t0 þ Δt ½RFCO2 ðtÞ þ RFNOx ðtÞ þ RFsoot þ RFsulf ates ðtÞ þ RFAIC ðtÞ þ RFH2 O ðtÞdt t0 non-CO2 ratio ¼ Z t0 þ Δt ½RFCO2 ðtÞdt t0

The challenge in treating non-CO2 combustion emissions and eﬀects lies in reconciling the wide range of atmospheric lifetimes ranging from centuries for CO2 to hours for contrails. Long lived gases become well mixed in the atmosphere; however, short-lived emissions can remain concentrated near ﬂight routes, mainly in the northern midlatitudes; hence, these emissions can lead to regional perturbations to radiative forcing.27 The impact of aircraft on regional climate could be important, but is currently beyond the capability of the models used in this work; hence, the results from APMT Climate may not be applied to individual ﬂights. Despite these limitations, assessing short-lived eﬀects on a globally averaged basis gives an indication of the contribution from non-CO2 forcing agents to climate change for the purposes of policy making.24

3. RESULTS 3.1. Non-CO2 Ratios for Conventional and SPK Fuel. The non-CO2 ratios derived for conventional jet fuel and SPK fuel are given in Figure 2 for time windows of 500 years, 100 years, and 20 years. Each time window has ratios derived using the low, nominal, and high impact lens to capture climate-modeling uncertainties of the APMT climate model. The bars correspond to results using the nominal lens while the whiskers show the results using the low and high lens. Shorter time windows emphasize the climate forcing of short-lived effects. Note that total non-CO2 ratios are given for a case where only contrails are included as well as a case where total AIC is included (recall that contrails and AIC are mutually exclusive within the formulation of APMT). All calculations subsequently presented herein will be using the non-CO2 ratio including total AIC. Although these results indicate that both the time horizon and choice of lens play an important role in assessing the climate impacts of non-CO2 combustion emissions and eﬀects, Figure 2 does not convey that the use of diﬀerent background emissions scenarios has an important inﬂuence on the resulting non-CO2 ratios. The default background scenarios used in the APMT climate lenses are inconsistent with the assumptions used by the IPCC in calculating GWP values. Speciﬁcally, the low, nominal

and high lenses typically assume SRES B2, SRES A1B, and SRES A2 background scenarios, respectively, whereas the IPCC adopts a constant background CO2 concentration of 378 ppm in all GWP calculations. While a constant CO2 background concentration assumption is unrealistic, it was adopted in this work for consistency with the IPCC treatment of CH4 and N2O. For example, the 100-year non-CO2 ratios for conventional jet fuel calculated using the APMT prescribed low, nominal and high lens SRES background scenarios are 37% lower, 11% higher and 24% higher, respectively, than those calculated using the constant background scenario. These diﬀerences are ampliﬁed under longer time windows (i.e., 500 year), while reduced under shorter time windows (i.e., 20 year). The combustion of conventional jet fuel emits 73.2 g CO2/MJ. In the nominal case (100-year time window and nominal lens), the results from Figure 2 show that accounting for the climate impacts of all combustion products from aircraft is equivalent to emitting 2.07 times the amount of CO2 actually produced from the combustion process, or 151.3 gCO2e/MJ instead of 73.2 gCO2/MJ. Similarly, the combustion of SPK fuel emits 70.4 g CO2/MJ. When compared to conventional jet fuel, the elimination of sulfates and the increase in water vapor lead to warming while the reduction in NOx and soot lead to cooling. Using the low lens under the 100-year and 500-year time windows, the reduction in NOx instead leads to a small warming eﬀect. In the nominal case, Figure 2 shows that accounting for the climate impacts of the actual combustion products from aircraft consuming SPK fuel is equivalent to emitting 2.22 times the amount of CO2 actually produced from the combustion process, or 156.1 gCO2e/MJ instead of 70.4 gCO2/MJ. The contribution of contrails and contrail cirrus to the total non-CO2 ratio is suﬃciently large that even a small change of either is more signiﬁcant than a large change in other species; hence, the total non-CO2 ratios are most sensitive to error in these two forcing agents. The reader is reminded that the radiative forcing from contrails and contrail cirrus were assumed unchanged per unit mass of fuel burned. As discussed in the SI, more detailed models and empirical measurements are needed to 10739

dx.doi.org/10.1021/es2017522 |Environ. Sci. Technol. 2011, 45, 10736–10743

Environmental Science & Technology

ARTICLE

Figure 3. Well-to-wake (+) emissions for select alternative jet fuel pathways, normalized by conventional jet fuel.

determine what, if any, changes may actually occur to contrails and contrail-cirrus with a change in fuel composition. 3.2. Well to Wake (+) Emissions Inventories. The results from Figure 2 can be combined with well-to-wake GHG emissions inventories to complete the framework required to include non-CO2 combustion emissions and effects in a fuel life cycle. The non-CO2 ratios developed using the low, nominal and high impact lenses of the APMT climate model were paired with select life cycle GHG inventories developed by Stratton et al.2 These GHG inventories were created by identifying key parameters through examination of the GHG emissions resulting from each life cycle step. Optimistic, nominal and pessimistic sets of these key parameters formed the basis for corresponding low GHG inventories, baseline or nominal GHG inventories, and high GHG inventories. The select fuel pathways used herein are conventional jet fuel, FischerTropsch jet fuel from switchgrass, FischerTropsch jet fuel from coal without carbon capture and sequestration (CCS), and hydroprocessed jet fuel from rapeseed oil. These fuel pathways were chosen to span a wide range of total GHG emissions and also the relative fractions of methane and nitrous oxide. Details regarding the analysis of these fuels are given in the SI. All results and discussion in this section are limited to a 100-year time window because of its prevailing use in the scientific community for global warming potentials. The sensitivity of results to choice of time window is addressed in Section 3.3. Life cycle analysis is fundamentally a comparative tool; hence, results are normalized by the life cycle GHG emissions of the baseline, conventional jet fuel. The range in life cycle GHG emissions of conventional jet fuel is primarily driven by the origin of the crude oil and the reﬁning process. Speciﬁcally, the low, baseline and high GHG emissions results of conventional jet fuel correspond to straight run jet fuel from domestically sourced crude oil, the 2008 average U.S. reﬁning eﬃciency with a weighted average of all crude oil fed into U.S. reﬁneries, excluding Canadian

oil sands, and hydroprocessed jet fuel from Nigerian crude oil, respectively.2 As noted in Table 1, the speciﬁc energy of SPK is 2.3% higher than that of conventional jet fuel. Therefore, the use of SPK leads to a reduced takeoﬀ weight and hence a reduced quantity of fuel needed to complete a given mission. Hileman et al.4 quantiﬁed this eﬀect on a ﬂeet wide basis, ﬁnding that a 0.3% reduction in ﬂeet wide fuel energy use would result from the use of all SPK; therefore, a 0.3% improvement in energy eﬃciency was applied to SPK fuels when normalizing by conventional jet fuel. Figure 3 compares the well-to-wake and well-to-wake (+) emissions inventories normalized by the corresponding baseline of conventional jet fuel. The upper half shows only the well-to-wake inventories as presented by Stratton et al.2 and documented in the SI. The lower half shows the well-to-wake (+) version of the same GHG inventories. The solid error bars, which have lines on their ends, correspond to combining the 100-year nominal nonCO2 ratios with the well-to-wake low, baseline and high GHG emissions scenarios. The dashed error bars, which have circles on their ends, correspond to pairing the 100-year low impact, nominal and high impact non-CO2 ratios with the well-to-wake low, baseline and high GHG emissions scenarios, respectively. Presenting the results in this manner separates the variability introduced by the well-to-tank GHG inventories from the climate modeling uncertainty of the non-CO2 ratios. The uncertainty of the non-CO2 ratios has a larger inﬂuence than the internal variability of the GHG inventories on the range of normalized well-to-wake (+) emissions for each fuel pathway. Relative to the normalized well-to-wake GHG inventories shown in the upper half of Figure 3, intrapathway variability is increased with the inclusion of non-CO2 combustion emissions and eﬀects while interpathway variability in normalized well-towake (+) emissions is reduced. For these pathways, when only GHG emissions are considered, the range in baseline life cycle GHG emissions is 0.22.2 times those of baseline conventional 10740

dx.doi.org/10.1021/es2017522 |Environ. Sci. Technol. 2011, 45, 10736–10743

Environmental Science & Technology

ARTICLE

Figure 4. Sensitivity to time window of well-to-wake GHG emissions inventories of baseline conventional jet fuel from U.S. crude oil, baseline conventional jet fuel from Nigerian crude oil, baseline rapeseed oil to HRJ and the combustion emissions and eﬀects of conventional jet fuel. Time windows of 500 years, 100 years, and 20 years were considered using the nominal lens.

jet fuel. When GHG emissions and non-CO2 combustion emissions and eﬀects are included, the range in baseline well-to-wake (+) emissions is reduced to 0.61.7 times those of baseline conventional jet fuel. Hence, the inclusion of non-CO2 combustion emissions and eﬀects in the fuel life cycle increases the absolute emissions range of each fuel pathway but reduces the overall range in the life cycle emissions of alternative fuels relative to conventional jet fuel. Interest is primarily focused on fuel options with well-to-wake GHG emissions lower than those of conventional jet fuel. Because of the similar magnitude of the non-CO2 combustion emissions and eﬀects of SPK and conventional jet fuel, the normalized well-to-wake (+) emissions of fuels in this category are higher than their normalized well-to-wake GHG emissions. Hence, a percentage reduction in life cycle GHG emissions of the jet fuel mix is less than the actual percentage reduction in aviation related climate impacts. For example, an SPK fuel option with zero life cycle GHG emissions would oﬀer a 100% reduction in GHG emissions but only a 48% reduction in actual climate impact using a 100-year time window and the nominal lens. Therefore, aviation GHG reduction scenarios (e.g., emissions wedge charts) that rely exclusively on relative changes in GHG emissions may overestimate the beneﬁt of alternative fuel use on the global climate system. Only a percentage reduction in fuel burn is equivalent to the same percentage reduction in aviation related climate forcing. The degree of overestimation is dependent on the assumptions used for the climate impact analysis of non-CO2 combustion emissions and eﬀects. Conversely, aviation GHG reduction scenarios that rely on absolute changes in GHG

emissions (e.g., mass of CO2e/MJ) will yield similar results for the impact of alternative fuel use on the global climate system regardless of whether well-to-wake or well-to-wake (+) emissions are used. The discrepancy between conventional jet fuel and SPK fuels caused by combustion CO2 emissions and the non-CO2 ratios is small by comparison to the variability in well-to-tank GHG emissions among the fuel options of Figure 3. While the results of this work were developed using the best currently available data, climate forcing from contrails and contrail cirrus remains uncertain, especially for SPK fuel; therefore, these results should be used with caution until further research is available on how SPK fuel use changes their impacts. 3.3. Sensitivity of Results to Time Window. The results of Figure 3 were created using a 100-year time window to express all species and effects in terms of carbon dioxide equivalent; however, other time windows (i.e., 500-year and 20-year) could have been chosen for the non-CO2 ratios and GWP values of CH4 and N2O. The sensitivity of the well-to-wake and well-to-wake (+) emissions inventories to the choice of time window was examined through the use of alternative jet fuel case studies. Specifically, the well-to-wake GHG emissions inventories of baseline conventional jet fuel from U.S. crude oil, baseline conventional jet fuel from Nigerian crude oil and baseline rapeseed oil to HRJ were chosen to span a fuel pathway dominated by CO2, by CH4, and by N2O. Methane and nitrous oxide have atmospheric perturbation lifetimes of 12 and 114 years, respectively; therefore, a 20year time window more heavily weights methane while a 100 year time window more heavily weights nitrous oxide. Additionally, the CO2 and non-CO2 combustion emissions and effects of 10741

dx.doi.org/10.1021/es2017522 |Environ. Sci. Technol. 2011, 45, 10736–10743

Environmental Science & Technology conventional jet fuel were evaluated using the nominal lens with 500-year, 100-year, and 20-year time windows. The sensitivity of each GHG inventory and the combustion emissions and effects to time window selection is shown in Figure 4. The life cycle GHG inventory of conventional jet fuel from U.S. crude oil is largely composed of CO2 emissions. Only a small fraction is from CH4 or N2O; therefore, this inventory is insensitive to the choice of time window chosen to represent CH4 and N2O in terms of CO2. The choice of time window causes only a 1.3 gCO2e/MJ change in the well-to-wake GHG emissions the fuel. Substantial methane emissions result from the venting processes used in Nigeria for crude oil extraction. The global warming potential of CH4 varies by approximately an order of magnitude when evaluated using a 20-year time window versus a 500-year time window. This variation is carried through to the life cycle GHG inventories of fuels where CH4 is an important contributor to the total. Choosing diﬀerent time windows to assess the life cycle GHG emissions of baseline conventional jet fuel from Nigerian crude oil results in a range of 33.6 gCO2e/MJ. The life cycle GHG inventory of HRJ from rapeseed oil is strongly inﬂuenced by N2O emissions from direct and indirect conversion of synthetic nitrogen applied to the ﬁeld and nitrogen rich crop residues. The global warming potential for N2O is less sensitive to time window than that of CH4; however, it still varies by approximately a factor of 2 and this variation is carried through to the life cycle GHG inventories of fuels where N2O is an important contributor. Baseline HRJ from rapeseed oil (assuming no GHG emissions from land use change) is subject to a range of 13.5 gCO2e/MJ with the use of 500-year and 20-year time windows. While both the time horizon and choice of lens are important when assessing the climate impacts of non-CO2 combustion emissions and eﬀects, the focus of this section is the inﬂuence of the time window; therefore, the nominal lens was used as a representative assumptions set. The combustion emissions and eﬀects from conventional jet fuel vary by 267.9 gCO2e/MJ, from 96.6 gCO2e/MJ to 364.5 gCO2e/MJ, with the use of a 500-year or 20-year time window, respectively. SPK fuels are subject to the same inﬂuence because of the similarity in the non-CO2 ratios of SPK fuel and conventional jet fuel. As a result, all SPK fuels used by aviation are aﬀected by the substantial inﬂuence of time window choice on well-to-wake (+) emissions. The scope is not limited to fuel pathways that are strong in a particular type of emissions and the magnitude is several times larger than that of CH4 or N2O. Despite this undesirable variability, the time window used to assess non-CO2 combustion emissions and eﬀects should be constrained to the same as that used in the global warming potentials of well-to-tank CH4 and N2O; hence, the need for consistency serves as a constraint in choosing the appropriate time window.

4. DISCUSSION This work implemented a modiﬁed version of the APMT climate impacts module to develop ratios to scale the CO2 emissions from fuel combustion to account for the additional climate forcing from the non-CO2 combustion emissions and eﬀects of SPK fuel and conventional jet fuel. The results indicated that including non-CO2 combustion eﬀects from SPK fuel use can lead to an important decrease in the relative merit of the SPK fuel relative to conventional jet fuel. This is because contrails and contrail-cirrus clouds dominate the climate impact of the non-

ARTICLE

CO2 eﬀects from conventional fuel and the analysis herein assumed that SPK fuel use would not change the prevalence or impact of these eﬀects. Additional work should be devoted to better understanding how changes in fuel composition aﬀect the formation of contrails and contrail-cirrus clouds as their impact may be reduced with a change in fuel composition. The decrease in relative merit of SPK fuels means that methods of tracking climate change mitigation that rely exclusively on relative well-to-wake life cycle GHG emissions as a proxy for aviation climate impact may overestimate the impact of alternative fuel use on the global climate system. Furthermore, the variability introduced into the results by including non-CO2 combustion emissions and eﬀects highlights the broad challenges faced in collapsing multiple attributes into a single metric for comparison when assessing any new energy technology options. Determining an absolute “better or worse” requires an evaluation system, which is usually accomplished through a weighting scheme, monetization or time windowing with one or more metrics. Greenhouse gases are a convenient metric of comparison because their cause and environmental eﬀect are both important and readily quantiﬁed. Many other factors have less quantiﬁable impacts. The need for absolute comparisons requires deﬁning a “basis of equivalence” that introduces signiﬁcant variability into the result. This work indicates that aviation has an opportunity space extending beyond improving fuel eﬃciency and burning alternative fuels to reduce its climate impact. Technologies that reduce GHG emissions from fuel production, combustion CO2 emissions, and non-CO2 combustion emissions and eﬀects can all be considered simultaneously. Currently, these areas are largely examined in isolation. A holistic analysis framework is needed that examines alternative fuels for reduced well-to-wake (+) emissions, aircraft design and operations for reduced fuel consumption, and changes to operational procedures for reduced contrails and contrail-cirrus impacts; however, an equitable comparison of the climate mitigation options for aviation requires consistent accounting of the climate impacts of non-CO2 combustion emissions and eﬀects.

’ ASSOCIATED CONTENT

bS

Supporting Information. Information on the combustion emissions of SPK fuel and the development of the life cycle GHG inventory used here is provided. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 617-452-2879; e-mail: [email protected].

’ ACKNOWLEDGMENT This work was made possible by funding from the Federal Aviation Administration and Air Force Research Laboratories through the PARTNER Center of Excellence under Award Number 06-C-NE-MIT, Amendment Nos. 012 and 021. We thank Mr. Chris Dorbian, Mrs. Hsin Min Wong, Prof. Steven Barrett, Prof. Jessika Trancik, and Prof. Ian Waitz for their help in improving the quality of the work presented herein as well as Warren Gillette and Lourdes Maurice, of FAA, and Tim Edwards and Bill Harrison, both of AFRL, for their leadership in managing this project. 10742

dx.doi.org/10.1021/es2017522 |Environ. Sci. Technol. 2011, 45, 10736–10743

Environmental Science & Technology

’ REFERENCES (1) FAA. Federal Aviation Administration. FAA Aerospace Forecast: Fiscal Years 2010203d0; U.S. Department of Transportation, FAA: Washington, DC, 2009; http://www.faa.gov/data_research/aviation/ aerospace_forecasts/2010-2030/media/2010%20Forecast%20Doc.pdf (accessed April 14, 2010). (2) Stratton, R. W.; Wong, H. M.; Hileman, J. I. Life Cycle Greenhouse Gas Emissions from Alternative Jet Fuels, Partnership for Air Transportation Noise and Emissions Reduction (PARTNER) Report, Number 2010-001; Massachusetts Institute of Technology: Cambridge, MA, 2010; http://web.mit.edu/aeroastro/partner/reports/proj28/partnerproj28-2010-001.pdf (accessed July 28, 2010) (3) ASTM D7566-09. Standard Speciﬁcation for Aviation Turbine Fuel Containing Synthesized Hydrocarbons; ASTM International: West Conshohocken, PA, 2009; DOI: 10.1520/D7566-09, http://www.astm.org/ Standards/D7566.htm (accessed May 19, 2011). (4) Hileman, J. I.; Stratton, R. W.; Donohoo., P. E. Energy Content and Alternative Fuel Viability. J. Propul. Power 2010, 26 (6), 1184–1195. (5) Hileman, J. I.; Stratton, R. W. Alternative Jet Fuel Feasibility, Transp. Policy, November 2010 (accepted). (6) Stratton, R. W.; Wong, H. M.; Hileman, J. I. Quantifying variability in life cycle greenhouse gas inventories of alternative middle distillate transportation fuels. Environ. Sci. Technol. 2011, DOI: 10.1021/ es102597f. (7) Edwards, R.; Larive, J.; Mahieu, V.; Rouveirolles, P. Well-toWheels Analysis of Future Automotive Fuels and Powertrains in the European Context, Tank-to-Wake Report Version 2c; EUCAR, CONCAWE, & JRC: Ispra, Italy, 2007; http://ies.jrc.ec.europa.eu/WTW (accessed April 14, 2010). (8) Broch, A.; Hoekman, S. K.; Gertler, A.; Robbins, C.; Natarajan, M. Biodistillate Transportation fuels 3Life cycle impacts. SAE Int. 2009, 2009–012768. (9) Wuebbles, D.; Gupta, M.; Ko, M. Evaluating the Impacts of Aviation on Climate Change. EOS 2007, 88 (14), 157–168. (10) Jacobson, M. Z. Short-term eﬀects of controlling fossil-fuel soot, biofuel soot and gases, and methane on climate, Arctic ice, and air pollution health, J. Geophys. Res. 2010, 115 (D14209) (11) Stuber, N.; Forster, P.; Shine, K. The importance of the diurnal and annual cycle of air traﬃc for contrail radiative forcing. Nat. Lett. 2006, 44 (15), 864–867. (12) Stevenson, S. D.; Doherty, R. M.; Sanderson, M. G.; Collins, W. J.; Johnson, C. E.; Derwent, R. G. Radiative forcing from aircraft NOx emissions: Mechanisms and seasonal dependence. J. Geophys. Res. 2004, 109 (D17307). (13) Bester, N. Yates, A. Assessment of the operational performance of FischerTropsch synthetic paraﬃnic kerosene in a T63 gas turbine compared to conventional jet A-1 fuel, GT200960333. In Power for Land, Sea and Air, Proceedings of ASME Turbo Expo 2009, Orlando, FL, June 812, 2009. (14) Bulzan, D.; Howard, R.; Corporan, E.; et al. Gaseous particulate emissions results of the NASA alternative aviation fuel experiment (AAFEX). In Power for Land, Sea and Air, Proceedings of ASME Turbo Expo 2010, Glasgow, UK, June 1418, 2010. (15) Dewitt, M. J.; Corporan, E.; Graham, J.; Minus, D. Eﬀects of aromatic type and concentration in FischerTropsch fuels on emissions production and material compatibility. Energy Fuels 2008, 22, 2411–2418. (16) Donohoo, P. Scaling Air quality eﬀects from alternative jet fuel in aircraft and ground support equipment, M.Sc. Thesis, Massachusetts Institute of Technology, Cambridge, MA, 2010. (17) Marais, K.; Lukachko, S. P.; Jun, M.; Mahashabde, A.; Waitz, I. A. Assessing the impact of aviation on climate. Meteorol. Z. 2008, 17, 157–172. (18) Mahashabde, A.; Wolfe, P.; Ashok, A.; Dorbian, C.; He, Q.; Fan, A.; Lukachko, S.; Mozdzanowska, A.; Wollersheim, C.; Barrett, S. R. H.; Locke, M.; Waitz, I. A. Assessing the environmental impacts of aircraft noise and emissions. Prog. Aerosp. Sci. 2011, 47, 15–52. (19) APMT, Aviation Portfolio Management Tool, Climate Impacts Methodology, 2010; http://web.mit.edu/aeroastro/partner/apmt/ climateimpact.html (accessed April 14, 2010).

ARTICLE

(20) Wild, O.; Prather, M. J.; Akimoto, H. Indirect long-term global radiative cooling from NOx emissions. Geophys. Res. Lett. 2001, 28 (9), 1719–1722. (21) Hoor, P.; Borken-Kleefeld, J.; Caro, D.; Dessens, O.; Endresen, O.; Gauss, M.; Grewe, V.; Hauglustaine, D.; Isaksen, I. S. A.; €Jockel, P.; Lelieveld, J.; Myhre, G.; Meijer, E.; Olivie, D.; Prather, M.; Schnadt Poberaj, C.; Shine, K. P.; Staehelin, J.; Tang, Q.; van Aardenne, J.; van Velthoven, P.; Sausen, R. The impact of traﬃc emissions on atmospheric ozone and OH: results from QUANTIFY. Atmos. Chem. Phys. 2009, 9, 3113–3136. (22) Lee, D. S.; Pitari, G; Grewe, V; Gierens, K; Penner, J. E.; Petzold, A; Prather, M. J.; Schumann, U; Bais, A; Berntsen, T.; Iachetti, D.; Lim, L. L.; Sausen, R. Transport impacts on atmosphere and climate: Aviation. Atmos. Environ. 2010, 44, 4678–4734. (23) Hansen, J.; Sato, M.; Ruedy, R. and et al. Eﬃcacy of climate forcings. J. Geophys. Res., [Atmos.] 2005, 110 (D18104). (24) Solomon, S.; Qin, D.; Manning, M.; Chen, Z.; Marquis, M.; Averyt, K. B.; Tignor, M.; Miller, H. L. Contribution of Working Group I to the Fourth Assessment Report. In Intergovernmental Panel on Climate Change; Cambridge University Press, Cambridge, UK, 2006. (25) Wuebbles, D. J.; Yang, H.; Herman, R. Climate Metrics and Aviation: Analysis of Current Understanding and Uncertainties, Subject Speciﬁc white paper on Metrics for Climate Impacts; University of Illinois and Western Illinois University: Urbana and Macomb, IL, 2008. (26) Shine, K. P.; Fuglestvedt, J. S.; Hailemariam, K.; Stuber, N. Alternatives to the global warming potential for comparing climate impacts of emissions of greenhouse gases. Clim. Change 2005, 68 (3), 281–302. (27) Penner, J. E.; Lister, D. H.; Griggs, D. J.; Dokken, D. J.; McFarland, M. Aviation and the global atmosphere: A special report in collaboration with the Scientiﬁc Assessment Panel to the Montreal Protocol on Substances that Deplete the Ozone Layer. In Intergovernmental Panel on Climate Change; Cambridge University Press: Cambridge, UK, 1999.

10743

dx.doi.org/10.1021/es2017522 |Environ. Sci. Technol. 2011, 45, 10736–10743

ARTICLE pubs.acs.org/est

Comparison of PM Emissions from a Commercial Jet Engine Burning Conventional, Biomass, and Fischer Tropsch Fuels Prem Lobo,* Donald E. Hagen, and Philip D. Whiteﬁeld Center of Excellence for Aerospace Particulate Emissions Reduction Research, Missouri University of Science and Technology, Rolla, Missouri 65409, United States

bS Supporting Information ABSTRACT: Rising fuel costs, an increasing desire to enhance security of energy supply, and potential environmental beneﬁts have driven research into alternative renewable fuels for commercial aviation applications. This paper reports the results of the ﬁrst measurements of particulate matter (PM) emissions from a CFM567B commercial jet engine burning conventional and alternative biomass- and, Fischer Tropsch (F-T)-based fuels. PM emissions reductions are observed with all fuels and blends when compared to the emissions from a reference conventional fuel, Jet A1, and are attributed to fuel properties associated with the fuels and blends studied. Although the alternative fuel candidates studied in this campaign oﬀer the potential for large PM emissions reductions, with the exception of the 50% blend of F-T fuel, they do not meet current standards for aviation fuel and thus cannot be considered as certiﬁed replacement fuels. Over the ICAO Landing Takeoﬀ Cycle, which is intended to simulate aircraft engine operations that aﬀect local air quality, the overall PM number-based emissions for the 50% blend of F-T fuel were reduced by 34 ( 7%, and the mass-based emissions were reduced by 39 ( 7%.

’ INTRODUCTION The anticipated growth in commercial air traﬃc, rising costs of fuel, an increasing desire to enhance the security of energy supply, and potential environmental beneﬁts have recently driven feasibility and viability assessment studies of alternative renewable fuels for commercial aviation applications, with a particular focus on fuels derived from biomass or synthesis from coal and natural gas via the Fischer Tropsch (F-T) process.1,2 Several ﬂight demonstrations of commercial aircrafts burning various blends of conventional jet fuel and either biomass or synthetic F-T fuels have been conducted recently.3 Speciﬁcations for aviation turbine fuels are established by American Society for Testing and Materials (ASTM) and United Kingdom Ministry of Defense (MOD). Other speciﬁcations for jet fuel exist but these are similar to those of ASTM and MOD. ASTM D16554 includes speciﬁcations for Jet A and Jet A-1 fuels used for commercial aviation within the United States. The MOD’s DEF STAN 91-915 outlines the speciﬁcation for Jet A-1 used in Europe. ASTM recently adopted a new speciﬁcation D75666 for up to 50:50 blends of synthetic fuel produced from the F-T process and conventional jet fuel. Since 1999, Sasol’s Semi Synthetic Jet Fuel (SSJF), a blend of up to a 50% of synthetic fuel, made from coal by F-T synthesis and conventional jet fuel, has been supplied to the Johannesburg, South Africa Airport.7 Eﬀorts are ongoing to certify Sasol’s Fully Synthetic Jet Fuel (FSJF) for commercial aviation use.8 The United States Department of Defense (DoD) has evaluated the r 2011 American Chemical Society

use of F-T fuels in military gas-turbine and diesel engines as a Battleﬁeld-Use Fuel to reduce reliance on foreign crude oil sources.9 The U.S. Air Force is currently in the process of certifying a 50:50 blend of conventional and synthetic F-T fuel for use in its entire ﬂeet by 2011, and a 50:50 blend of conventional and biofuel by 2013.10 Until recently, almost all of the studies on the performance and emissions characteristics of alternative fuels in gas turbine engines have been limited to military engine applications.11 13 This paper presents the results of a comparison of particulate matter (PM) emissions from a commercial jet engine burning several alternative biomass- (fatty acid methyl ester, FAME) and F-T-based fuels. The Missouri University of Science and Technology (Missouri S&T) along with Aerodyne Research, Inc. (ARI) and Air Force Research Lab (AFRL) at Wright Patterson Air Force Base participated in a ﬁeld campaign in November 2007 to characterize and compare the PM and gas-phase emissions of a CFM56-7B engine burning several alternative fuels and/or their blends with Jet A1. The measurements were performed at Test Site 3B at the GE Engine Test Facility in Peebles, OH. The results of the physical characterization of PM

Received: June 3, 2011 Accepted: November 1, 2011 Revised: September 29, 2011 Published: November 01, 2011 10744

dx.doi.org/10.1021/es201902e | Environ. Sci. Technol. 2011, 45, 10744–10749

Environmental Science & Technology

ARTICLE

Table 1. Properties of Fuelsa and Fuel Blends Testedb

fuel tested

fuel identiﬁer

speciﬁc

kinematic

EI CO2d

heat of

aromatic

oxygen

gravity @15 °C

viscosityc @ 20 °C (mm2/s)

(g/kg fuel burned)

combustion (kJ/kg)

H/C ratio

content (vol %)

content (wt%)

Jet-A

Jet A

0.803

5.78

3155

43302

1.92

12.3

Jet-A1

Jet A1

0.797

4.27

3155

43300

1.92

18.5

0

20% FAME/80% Jet-A1

20% FAME

0.808

4.74

3045

42000

1.94

14.8

3.4

40% FAME/60% Jet-A1

40% FAME

0.825

5.62

2942

40300

1.94

11.1

6.6

50% F-T/50% Jet-A1

50% F-T

0.776

4.4

3127

43600

2.04

100% F-T

100% F-T

0.755

4.65

3100

44100

2.17

9.25 <0.2

0

0 0

a

Fuels were provided by GE Aviation (Jet A, Jet A1), Boeing (FAME), and AFRL (F-T). b Fuel analysis was performed and reported by AFRL, unless otherwise stated. c Kinematic viscosity data were provided by GE Aviation. d EI CO2 data were provided by Boeing.

emissions are presented here while the gas-phase emissions results are presented elsewhere.14

’ EXPERIMENTAL SECTION Sample Acquisition. Exhaust samples were extracted from the core exhaust flow of a CFM56-7B engine mounted in an open air test cell and transported out of the exhaust environment to a mobile diagnostic facility located adjacent to the test cell and within 30 m of the engine. Sample extraction and transport were achieved using a PM probe and rake assembly coupled to a sample train similar to that used in previous CFM56-type engine field studies.15,16 The engine was operated over a range of power settings, including the operating points specified in the ICAO Landing Takeoff (LTO) cycle.17 The LTO Cycle is intended to simulate aircraft engine operations that affect local air quality (aircraft operations below 3000 feet altitude) and comprises the following power settings: 7%, 30%, 85%, and 100% corresponding to idle/taxi, approach, climb-out, and takeoff, respectively. The sampling rake and associated PM probes were positioned within 1 m of the exhaust nozzle exit plane, and approximately 5 cm oﬀ the vertical center line of the engine, in its core exhaust ﬂow. The PM probes were designed to introduce dilution gas at the probe tip such that sample dilutions >10:1 could be achieved at the point of sample capture, thereby reducing and/or eliminating the onset of any condensation, agglomeration, and gas-toparticle conversion processes in the sampling system.18 The rake assembly and probes were water cooled to protect them from thermal degradation during testing. The PM samples were conducted to the mobile diagnostic facilities through unheated 19.05 mm o.d. (16 mm i.d.) stainless steel sample lines. The entire sample train from probe tip to diagnostic suite was calibrated for size-dependent line loss as in previous studies.16,19 Sample Analysis. The PM emissions size distributions and total concentrations, associated combustion CO2 concentrations, and atmospheric conditions were measured with state-of-the-art high-frequency, real-time instrumentation. The instrumentation included the Cambustion DMS500, a fast particulate spectrometer to gather real-time PM size distributions, with a fast data acquisition rate yielding size distributions from 5 to 1000 nm at up to a 10 Hz frequency;20 a TSI Condensation Particle Counter (TSI model 3025) to measure total number concentration with a sampling frequency of 1 Hz; a fast-response CO2 detector (Sable Systems model CA-2A with a sampling frequency of 1 Hz) to establish emission factors and quantify sample dilution factors; and a weather station to monitor ambient conditions including

temperature, relative humidity, and pressure (sample frequency 0.2 Hz). Test Matrix. The engine was cycled through a matrix of reproducible engine power settings where for each power setting steady-state emissions and engine data were recorded. The engine power settings selected were as follows: 3%, 7%, 30%, 45%, 65%, 85% and 100% rated thrust. The fuels studied along with relevant physical, thermodynamic, and compositional characteristics are presented in Table 1. Jet A1 has been selected as the reference fuel for this analysis, and thus the PM emissions characteristics of the alternative fuels were compared to those from Jet A1.

’ DATA ANALYSIS Line Loss. Modification of the PM size spectrum due to line loss is an artifact associated with extractive sampling and must be taken into account. In this study a size-dependent line loss function determined from line loss calibrations was applied to the instrument data yielding an estimate of the PM size distributions at the point of entry into the sampling system. The average correction factors applied to the data set for number-based measurements were 42% at 7% power and decreased linearly to 16% at 100% power. The average correction factors applied to mass-based measurements were also found to vary linearly with power yielding an 18% correction at 7% power and a 6% correction at 100% power. The low size cutoff for the PM size distribution data was selected to be 10 nm. Emission Indices. PM number and mass concentrations were converted to number- and mass-based emission indices (EIn and EIm, respectively) to allow quantification of emissions per kilogram of fuel burned.19 The emission indices data are presented in normalized form, achieved by dividing the EIn and EIm values by a fixed normalization factor (same for all fuels), since the emissions data are proprietary. Heats of Combustion Corrections. Each fuel and fuel blend tested has different heats of combustion (Table 1). To accomplish intercomparison between fuels at the same engine power settings, the measured fuel flow rates at each test condition were adjusted to account for these different heats of combustion. The resulting Jet A1 equivalent fuel flow rates assured intercomparisons were made for the same energy per unit time for a given engine power setting. Uncertainty Estimation. Measurement uncertainties in PM parameters are taken to be the repeat measurement standard deviation (1σ) + 5%. These measurement uncertainties are calculated for each parameter for each fuel. A percent difference for PM parameters referenced to Jet A1 was calculated and the 10745

dx.doi.org/10.1021/es201902e |Environ. Sci. Technol. 2011, 45, 10744–10749

Environmental Science & Technology

ARTICLE

Figure 1. Normalized EIn size distributions as a function of engine power setting, measured at the engine exit plane for Jet A1 (a) and 100% F-T (b) fuels.

Table 2. Geometric Mean Diameters for All Alternative Fuels Compared to Those for Jet A1 at Engine Power Settings Corresponding to the LTO Cycle GMDalternative fuel/GMDJet A1 alternative fuel

7%

30%

85%

100%

20% FAME 40% FAME

1.07 0.98

1.12 0.98

1.16 1.02

1.05 0.92

50% F-T

0.93

0.94

1.01

0.93

100% F-T

0.80

0.84

0.95

0.88

fractional uncertainty was estimated by taking the square root of the sum of the squares of the fractional uncertainty in the difference and fractional uncertainty for Jet A1. The fractional uncertainty was multiplied by the percent difference to yield the uncertainty in the percent difference. The resulting uncertainty value was used to ascertain statistical significance of the PM parameters for alternative fuels and blends relative to Jet A1.

’ RESULTS AND DISCUSSION Size Distributions and Emission Indices. The PM emissions intercomparisons for the fuels in this study are based in part on PM emission size distributions data acquired as a function of engine power setting. From these measurements, the geometric mean diameters (GMDs), geometric standard deviations (GSDs), number-based emission indices can be derived and with assumptions on particle sphericity and density (1 g/cm3 was used in this study), the mass-based emission indices can be calculated. For all fuels studied, the PM emission size distribution at all engine power settings was found to be log-normal. For a given fuel, GMD, EIn, and EIm increased linearly with engine power setting from 7% to 100% rated thrust. Figure 1 presents the normalized EIn size distributions as a function of engine power setting, measured at the engine exit plane for Jet A1 and 100% F-T fuels, respectively. The distributions demonstrate a general correlation to both engine power setting and fuel type. These results are consistent with those reported for a T701C military helicopter engine burning a JP-8 and a 100% F-T fuel.13 The geometric mean diameter for the four alternative fuels relative to Jet A1 at engine power settings corresponding to the

LTO cycle are presented in Table 2. The uncertainty in the GMD ratio is 5%, i.e., any diﬀerence between the alternative fuels and Jet A1 is statistically signiﬁcant if it is outside the range 0.95 1.05. The GMDs for the 20% FAME fuel are consistently higher than those for Jet A1. For all other fuels, GMDs were either lower than or equal to those for Jet A1. The most signiﬁcant diﬀerences in GMD are observed for the 50% F-T and 100% F-T fuels. For all fuels studied, EIn and EIm were found to increase with increasing engine power setting, with minimums observed at 7% power and maxima observed at 100% power. Reductions in both EIn and EIm were observed when burning the alternative fuels compared to the baseline Jet A1. The resulting relative changes in EIn and EIm and their uncertainties are summarized in Figure 2a and b. Most of the reductions in number and massbased emission indices were found to be statistically signiﬁcant (based on the uncertainty estimation) except for the change in EIn for 20% FAME at 7% rated thrust and EIm for 20% FAME at 30% rated thrust. Generally, the measured reductions in PM were largest at idle, and smallest at maximum rated thrust. For low engine power settings, the trend in emissions reduction is 20% FAME < 40% FAME < 50% F-T < 100% F-T, i.e., the emissions reduction is greater as the relative amount of alternative fuel content in the fuel is increased. The overall reduction in PM number and mass reductions over the LTO cycle for the diﬀerent alternative fuels studied are presented in Table 3. It should be noted that during the Jet A1 measurements the ambient temperature was ∼4 °C and for the other fuels tested it was ∼0 °C. Some of the observed reduction in emissions could be attributed to the change in temperature. A more recent study, AAFEX (Alternative Aviation Fuels EXperiment) speciﬁcally examined the impact of temperature on the emissions at the engine exit for a ﬁxed fuel, albeit for a diﬀerent model of the CFM56 engine and diﬀerent fuel.21 The AAFEX results suggest that the eﬀects of the 4 °C temperature diﬀerence should be small (<8%) compared to the reductions observed in this study. Impact of Fuel Properties on PM Emissions Reduction. The pure fuels and blends studied here vary significantly in their aromatic content (see Table 1). An examination of the PM emissions vs aromatic content was performed. Figure 3a and b present the normalized EIn and EIm, respectively, for all the fuels as a function of fuel aromatic content and engine power settings corresponding to the LTO cycle. Both EIn and EIm were found to decrease with decreasing fuel aromatic content at each of the four operational points in the LTO cycle. The largest reduction 10746

dx.doi.org/10.1021/es201902e |Environ. Sci. Technol. 2011, 45, 10744–10749

Environmental Science & Technology

ARTICLE

Figure 2. Percent reduction in EIn and EIm compared to Jet A1, as a function of Jet A1 equivalent fuel ﬂow rate.

Table 3. Overall PM Number and Mass Reductions over the LTO Cycle Achieved Using Candidate Alternative Fuels alternative fuel

a

PM number reduction

PM mass reduction

20% FAME

22% ( 7%

20% ( 8%

40% FAMEa

35% ( 6%

52% ( 5%

50% F-T

34% ( 7%

39% ( 7%

100% F-T

52% ( 4%

62% ( 4%

Estimated PM number and mass reduction.

in EIn was observed when emissions at 7% power for the Jet A1 and 100% F-T fuels were compared, where the aromatic contents were 18.5% and ∼0% respectively. These results are consistent with previous studies on military engines burning F-T fuels where the PM emissions reduction observed was attributed to the low aromatic content of the F-T fuel.12 Decreasing the aromatic content of a fuel will typically increase the fuel’s hydrogen/carbon (H/C) ratio, but because aviation fuels are complex mixtures of both aromatic and aliphatic hydrocarbons the relationship is not a simple inverse proportionality. For example, in Table 1 the H/C ratios for Jet A and Jet A1 are essentially identical but the aromatic content for Jet A is 33% lower than that for JetA1. Thus it is reasonable to conclude both fuel aromatic content and H/C ratio can inﬂuence PM emissions. However, the limited data set available from this study does not permit the deconvolution of these coupled impacts.

The 20% FAME and 40% FAME fuels are diﬀerent from the other fuels tested in that they contain oxygen. In studies on combustion of methyl esters in diesel engines, the observed reductions in PM emissions were attributed to the higher oxygen content of the methyl esters compared to diesel.22 However, the higher oxygen content of the methyl esters permits decarboxylation which leads to the formation of CO2 and thereby reduces the eﬀect of fuel-bound oxygen in removing carbon from soot precursor reactions.23,24 The 20% FAME and 40% FAME fuels are also more viscous than Jet A1. Increased fuel viscosity impacts both the spray pattern and the size of the fuel droplets in the combustor, leading to incomplete combustion.25,26 These factors mitigate the reduction of PM associated with reduced fuel aromatic content for the 20% FAME and 40% FAME fuels. Implications of Alternative Fuel Use in the Commercial Fleet. This study demonstrates the potential for large PM emissions reduction at the exit plane of the exhaust nozzle should alternative aviation fuels with low aromatic content, high H/C ratio, and low viscosity be adopted by the commercial fleets. These emissions reductions have major implications for local and regional air quality at airports since such fuels could greatly reduce the PM emissions inventory for aircraft operating in all modes of the LTO cycle. The level of emissions reductions observed for the CFM56-7B engine in this study may or may not be achieved with other engines in the commercial fleet, as their emissions characteristics are known to vary.27 10747

dx.doi.org/10.1021/es201902e |Environ. Sci. Technol. 2011, 45, 10744–10749

Environmental Science & Technology

ARTICLE

Emissions Reduction (PARTNER)—a FAA Transport Canada National Aeronautics and Space Administration (NASA)-sponsored Center of Excellence under Grant 07-C-NE-UMR Amendments 003 and 004 (Carl Ma, Project Manager). Any opinions, ﬁndings, and conclusions or recommendations expressed in this paper are those of the authors and do not necessarily reﬂect the views of the FAA. We acknowledge the following individuals and organizations for substantial contributions to the successful execution of this study: CFM International for sponsoring the engine test; Art Johnson and colleagues at General Electric Aviation; Jim Girton and colleagues at GE’s Peebles Test Site 3B; Will Dodds at GE Aviation; Dave Daggett and Steve Baughcum at Boeing; and ARI and AFRL teams.

’ REFERENCES

Figure 3. Normalized EIn (a) and EIm (b) as a function of fuel aromatic content at the power settings corresponding to the LTO cycle.

It is important to note that although the alternative fuel candidates studied in this campaign oﬀer the potential for large PM emissions reductions; the 20% FAME, 40% FAME, and 100% F-T fuels do not meet current ASTM standards for aviation fuel4,6 and thus cannot be considered as certiﬁed replacement fuels. Furthermore, fuels with low aromatic content (<10%) are known to aﬀect seal-swell in aircraft fuel delivery systems which can result in seal failures and fuel leakage.7 The FAME fuels have much lower energy content than the other fuels tested because the oxygen in the fuel molecule does not contribute any energy during combustion and can lead to reduced ﬂight range. Despite these limitations, this study clearly indicates the potential for air quality beneﬁts should alternative fuels be used in the development of future aviation fuels.

’ ASSOCIATED CONTENT

bS

Supporting Information. A description of the sample line loss characterization methodology and size-dependent line loss function for the sampling system employed in this study. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT This work was funded by the Federal Aviation Administration (FAA) through the Partnership for AiR Transportation for Noise and

(1) Daggett, D.; Hadaller, O.; Hendricks, R.; Walther, R. Alternative Fuels and Their Potential Impact on Aviation; NASA/TM-2006-214365; Cleveland, OH, October 2006. (2) Hileman, J. I.; Stratton, R. W.; Donohoo, P. E. Energy Content and Alternative Jet Fuel Viability. J. Propul. Power 2010, 26, 1184–1196. (3) Blakey, S.; Rye, L.; Wilson, C. W. Aviation gas turbine alternative fuels: A Review. Proc. Combust. Inst. 2011, 33, 2863–2885. (4) Standard Speciﬁcation for Aviation Turbine Fuels; ASTM D1655; ASTM International: West Conshohocken, PA. (5) Def Stan 91-91. Turbine Fuel, Aviation Kerosine Type, Jet A-1 NATO Code: F-35, Joint Service Designation: AVTUR www.dstan. mod.uk. (6) Standard Speciﬁcation for Aviation Turbine Fuel Containing Synthesized Hydrocarbons; ASTM D7566; ASTM International: West Conshohocken, PA. (7) Hemighaus, G.; Boval, T.; Bosley, C.; Organ, R.; Lind, J.; Brouette, R.; Thompson, T.; Lynch, J.; Jones, J. Alternative Jet Fuels: A supplement to Chevron’s Aviation Fuels Technical Review; Chevron Corporation: San Ramon, CA, 2006. (8) Moses, C. A.; Roets, P. N. J. Properties, Characteristics and Combustion Performance of Sasol Fully Synthetic Jet Fuel. J. Eng. Gas Turbines Power 2009, 131, 041502. (9) Lamprecht, D. Fischer Tropsch Fuel for Use by the U.S. Military as Battleﬁeld-Use Fuel of the Future. Energy Fuels 2007, 21, 1448–1453. (10) Haskin, H.; Rumizen, M. Paving the way for alternative fuels. Int. Airport Rev. 2010, 3, 29–31. (11) Corporan, E.; Reich, R.; Monroig, O.; DeWitt, M.; Larson, V.; Aulich, T.; Mann, M.; Seames, W. Impact of Biodiesel on Pollutant Emissions of a JP-8-Fueled Turbine Engine. J. Air Waste Manage. Assoc. 2005, 55, 940–949. (12) Corporan, E.; DeWitt, M.; Belovich, V.; Pawlik, R.; Lynch, A. C.; Gord, J. R.; Meyer, T. R. Emissions Characteristics of a Turbine Engine and Research Combustor Burning a Fischer Tropsch Jet Fuel. Energy Fuels 2007, 21, 2615–2626. (13) Corporan, E.; DeWitt, M. J.; Klingshirn, C. D.; Striebich, R.; Cheng, M.-D. Emissions Characteristics of Military Helicopter Engines with JP-8 and Fischer Tropsch Fuels. J. Propul. Power 2010, 26, 317–324. (14) Timko, M. T.; Herndon, S. C.; de la Rosa Blanco, E.; Wood, E. C.; Yu, Z.; Miake-Lye, R. C.; Knighton, W. B.; Shafer, L.; DeWitt, M.; Corporan, E. Combustion Products of Petroleum Jet Fuel, a Fischer Tropsch Fuel, and a Biomass Fatty Acid Ester Fuel for a Gas Turbine Engine. Combust. Sci. Technol. 2011, 183, 1039–1068. (15) Wey, C. C.; Anderson, B. A.; Wey, C.; Miake-Lye, R. C.; Whiteﬁeld, P.; Howard, R. Overview on the Aircraft Particle Emissions eXperiment (APEX). J. Propul. Power 2007, 23, 898–905. (16) Lobo, P.; Whiteﬁeld, P. D.; Hagen, D. E.; Herndon, S. C.; Jayne, J. T.; Wood, E. C.; Knighton, W. B.; Northway, M. J.; Miake-Lye, R. C.; Cocker, D.; Sawant, A.; Agrawal, H.; Miller, J. W The Development of Exhaust Speciation Proﬁles for Commercial Jet Engines; ARB Contract 10748

dx.doi.org/10.1021/es201902e |Environ. Sci. Technol. 2011, 45, 10744–10749

Environmental Science & Technology

ARTICLE

No. 04-344 Report; California Air Resources Board: Sacramento, CA, October 2007. (17) ICAO Annex 16. International standards and recommended practices, Environmental protection, Volume II, Aircraft engine emissions, 2nd ed.; International Civil Aviation Organization: Montreal, 1993. (18) Schmid, O.; Hagen, D. E.; Whiteﬁeld, P. D.; Trueblood, M. B.; Rutter, A. P.; Lilenfeld, H. V. Methodology for Particle Characterization in the Exhaust Flows of Gas Turbine Engines. Aerosol Sci. Technol. 2004, 38, 1108–1122. (19) Lobo, P.; Hagen, D. E.; Whiteﬁeld, P. D.; Alofs, D. J. Physical characterization of aerosol emissions from a Commercial Gas Turbine Engine. J. Propul. Power 2007, 23, 919–929. (20) Reavell, K.; Hands, T.; Collings, N. A Fast Response Particulate Spectrometer for Combustion Aerosols; SAE Technical Paper 2002-012714, 2002. (21) Anderson, B. E.; Beyersdorf, A. J.; Hudgins, C. H.; Plant, J. V.; Thornhill, K. L.; Winstead, E. L.; Ziemba, L. D.; Howard, R.; Corporan, E.; Miake-Lye, R. C.; Herndon, S. C.; Timko, M.; Wood, E.; Dodds, W.; Whiteﬁeld, P.; Hagen, D.; Lobo, P.; Knighton, W. B.; Bulzan, D.; Tacina, K.; Wey, C.; Vander Wal, R.; Bhargava, A.; Kinsey, J.; Liscinsky, D. S. Alternative Aviation Fuel Experiment (AAFEX); NASA/TM-2011217059; Hanover, MD, February 2011. (22) Lapuerta, M.; Armas, O.; Rodríguez-Fernandez, J. Eﬀect of biodiesel fuels on diesel engine emissions. Prog. Energy Combust. Sci. 2008, 34, 198–223. (23) Westbrook, C. K.; Pitz, W. J.; Curran, H. J. Chemical Kinetic Modeling Study of the Eﬀects of Oxygenated Hydrocarbons on Soot Emissions from Diesel Engines. J. Phys. Chem. A 2006, 110, 6912–6922. (24) Szybist, J. P.; Song, J.; Alam, M.; Boehman, A. L. Biodiesel combustion, emissions and emission control. Fuel Process. Technol. 2007, 88, 679–691. (25) Lefebvre, A. H. Gas Turbine Combustion; Taylor and Francis: Philadelphia, 1999. (26) Pucher, G.; Allan, W.; LaViolette, M.; Poitras, P. Emissions From a Gas Turbine Sector Rig Operated With Synthetic Aviation and Biodiesel Fuel. J. Eng. Gas Turbines Power 2011, 133, 111502. (27) Timko, M. T.; Onasch, T. B.; Northway, M. J.; Jayne, J. T.; Canagaratna, M. R.; Herndon, S. C.; Wood, E. C.; Miake-Lye, R. C.; Knighton, W. B. Gas Turbine Engine Emissions—Part II: Chemical Properties of Particulate Matter. J. Eng. Gas Turbines Power 2010, 132, 061505.

10749

dx.doi.org/10.1021/es201902e |Environ. Sci. Technol. 2011, 45, 10744–10749

ARTICLE pubs.acs.org/est

Integration of Calcium and Chemical Looping Combustion using Composite CaO/CuO-Based Materials Vasilije Manovic and Edward J. Anthony* CanmetENERGY, Natural Resources Canada, 1 Haanel Drive, Ottawa, Ontario, Canada K1A 1M1 ABSTRACT: Calcium looping cycles (CaL) and chemical looping combustion (CLC) are two new, developing technologies for reduction of CO2 emissions from plants using fossil fuels for energy production, which are being intensively examined. Calcium looping is a two-stage process, which includes oxy-fuel combustion for sorbent regeneration, i.e., generation of a concentrated CO2 stream. This paper discuss the development of composite materials which can use copper(II)-oxide (CuO) as an oxygen carrier to provide oxygen for the sorbent regeneration stage of calcium looping. In other words, the work presented here involves integration of calcium looping and chemical looping into a new class of postcombustion CO2 capture processes designated as integrated CaL and CLC (CaLCLC or CaCu looping cycles) using composite pellets containing lime (CaO) and CuO together with the addition of calcium aluminate cement as a binder. Their activity was tested in a thermogravimetric analyzer (TGA) during calcination/reduction/oxidation/carbonation cycles. The calcination/reduction typically was performed in methane (CH4), and the oxidation/carbonation stage was carried out using a gas mixture containing both CO2 and O2. It was conﬁrmed that the material synthesized is suitable for the proposed cycles; with the very favorable ﬁnding that reduction/oxidation of the oxygen carrier is complete. Various schemes for the CaCu looping process have been explored here that would be compatible with these new composite materials, along with some diﬀerent possibilities for ﬂow directions among carbonator, calciner, and air reactor.

1. INTRODUCTION Carbon dioxide is a major greenhouse gas responsible for climate changes.1 The negative environmental eﬀects of such emissions represent a growing problem as the utilization of fossil fuels such as coal is increasing and can be expected to do so for the near- to medium-term future.2,3 Therefore, technologies associated with CO2 capture and storage (CCS) are increasingly considered to be likely contributors to reduce these emissions.1,4 Three main scenarios for CO2 separation and capture are postcombustion processes for traditional coal-ﬁred power plants, precombustion processes for gasiﬁcation or reforming, and oxyfuel combustion.57 Some proposed technologies incorporate processes from diﬀerent scenarios, and one example of this is the calcium looping cycle, which is in general regarded as a postcombustion CO2 capture technology. However, about 30% of the heat generated in a power plant integrated with a calcium looping cycle CO2 capture system is generated in the sorbent regenerator (calciner) employing oxy-fuel combustion.7 Calcium looping cycles (CaL) represent a new important class of technology which is based on the reversible chemical reaction between lime (CaO) and CO2. CaOðsÞ þ CO2ðgÞ ¼ CaCO3ðsÞ

ΔHr0

¼ 179 kJ=mol ð1Þ

The forward reaction, namely carbonation, is an exothermic reaction, while the reverse reaction, calcination, represents sorbent regeneration which is an endothermic process requiring heat. Published 2011 by the American Chemical Society

That heat can be supplied to the reactor (calciner) directly or indirectly from a combustor, or generated “in situ”, usually by combustion of a gas (CH4) or biomass.8 To produce a highlyconcentrated CO2 stream, nearly pure oxygen is used (oxy-fuel combustion), which is an expensive step because air separation is highly energy intensive. Another class of solid looping cycles is chemical looping combustion (CLC), a combustion technology with inherent CO2 separation.9 In this technology an oxygen carrier, typically a metal oxide, transfers the oxygen from the air to the fuel such that the combustion air and the fuel are never mixed, and the obtained ﬂue gas is a concentrated CO2 stream. Gaseous fuels such as syngas from coal gasiﬁcation or natural gas are preferred fuels, but combustion of solids is also feasible.10,11 Metal oxides in the oxidized form (MexOy) are the source of oxygen for the fuel oxidation, which occurs in a fuel reactor. ð2n þ mÞMex Oy þ Cn H2m f ð2n þ mÞMey Oy1 þ nCO2 þ mH2 O

ð2Þ

O2 carriers are usually oxides of Fe, Ni, Cu, or Mn supported by inert materials such as Al2O3, ZrO2, TiO2, or MgO. After reaction with the fuel, the reduced form of the O2 carrier Received: July 4, 2011 Accepted: October 24, 2011 Revised: September 15, 2011 Published: October 24, 2011 10750

dx.doi.org/10.1021/es202292c | Environ. Sci. Technol. 2011, 45, 10750–10756

Environmental Science & Technology

ARTICLE

(MeyOy1) is transported to the air reactor, and oxidized by air. Mey Oy1 þ 0:5O2 f Mex Oy

ð3Þ

The oxidized form of the O2 carrier is returned to the fuel reactor for a new cycle. In summary, sorbent regeneration in CaL occurs through combustion of a suitable fuel using pure oxygen, which has been demonstrated at a pilot-scale facility.12 However, the expense involved in providing pure oxygen diminishes the economic advantages of the technology.12,13 Moreover, a high temperature (>900 °C) is required for CaCO3 decomposition to produce a highly concentrated CO2 stream which unfavorably aﬀects sorbent activity over multiple cycles.14 An interesting possibility is to provide the calciner with the necessary heat generated in a CLC process and this is a study dealing with that concept for CaL. Practically, this means supplying heat for an endothermic reaction by a CLC process, which is an idea presented by Lyon and Cole,15 and more recently considered by Abanades et al. 16,17 for steam reforming in particular, and their work adds the critical reﬁnement that the heat from the copper oxide reduction drives the calcination step, thus ensuring that a pure CO2 stream can be produced in the calciner without the use of oxygen. It should also be noted that Abanades et al. have also proposed that this approach can be used for CCS, based on process simulations and have considered a large number of alternatives to the copper system in their patent application.17 What has not been done so far is to demonstrate this approach experimentally. Recently we have studied the preparation of CaO-based pellets for CO2 capture,18,19 and found that aluminate cements are suitable binders. These pellets show a high long-term CO2 capture activity and strength,20 which we attributed to the favorable inﬂuence of alumina compounds in the pellet structure. Moreover, the pellet preparation procedure, using a mixture of hydrated lime and aluminate cement,19 allows doping of pellets during preparation by other suitable compounds which can act as catalysts and/or oxygen carriers. Therefore, those pellets are an excellent candidate for integration of CaL with CLC. Oxidation of the O2 carrier in a CLC is exothermic, while the reaction in the fuel reactor, i.e., reduction, can be either endothermic or exothermic, depending on the type of O2 carrier and fuel. In practice, CuO serves as a suitable oxygen carrier for integration with CaL because both its oxidation and reduction reactions by methane (CH4) are exothermic. 2CuðsÞ þ O2ðgÞ ¼ 2CuOðsÞ

ΔHr0 ¼ 156 kJ=mol

ð4Þ

4CuOðsÞ þ CH4ðgÞ ¼ 4CuðsÞ þ CO2ðgÞ þ 2H2 OðgÞ ΔHr0 ¼ 178 kJ=mol

ð5Þ

The exothermic reduction of CuO can provide heat for the endothermic calcination. The other advantages of this well investigated oxygen carrier21,22 are a high oxygen transport capacity, favorable kinetics and thermodynamics which enable complete conversion of CH4 into CO2 and H2O, and ﬁnally its cost (it is one of the cheaper metal oxygen carriers). It has been shown that CuO-based oxygen carriers need a support, and the most promising performance has been demonstrated using alumina (Al2O3).21,22 Moreover, if such a system were to be developed for ﬂuidized bed application, the support ought to provide resistance to attrition to minimize the production of ﬁne particulates, since copper is potentially somewhat

toxic. Therefore, aluminate pellets, for which we have already developed an inexpensive preparation procedure and which we have extensively investigated,1820 are a good candidate to simultaneously provide support for the CuO carrier and sorbent for CO2 capture. The main objective of this study was to synthesize a new class of pellets, based on CaO and CuO with calcium aluminate cements as a binder/support. These pellets were explored as a means of integration of CaL and CLC in order to prove the proposed concept of CaCu looping cycles.

2. MATERIALS AND METHODS Cadomin (CD) limestone from Canada (see elsewhere19 for its elemental analysis), with a particle size 0.251.4 mm, was used as a natural CaO-containing material. The high content of SiO2 (5.47 wt %) and Al2O3 content (1.54 wt %) in CD limestone indicate the presence of silicate and aluminosilicate impurities. A powder of CuO, 98% of particles <5 μm, produced by Aldrich, was used as a CuO-containing material. A commercial calcium aluminate cement, CA-14, (71 wt % Al2O3 and 28 wt % CaO), produced by Almatis Inc., was used as binder for pelletization. It was supplied as a very ﬁne powder with >80% of the particles <45 μm. The pellets were prepared using calcined powdered limestone (lime), the powder of CuO, and the aluminate cement to obtain the ﬁnal CaO/CuO/Al2O3 mass ratio of 45:45:10 in the pellets. The limestone was calcined at 850 °C for 2 h before hydration, and weighed amounts of lime, CuO, and cement were mixed in a glass beaker. Water was slowly added, with stirring, to minimize the eﬀects of heat release due to the exothermic hydration process. The paste obtained was extruded through a 1.0-mm sieve to obtain uniform pellet diameters, and the resulting pellets were then air-dried for 24 h. The calcination/reduction/oxidation/carbonation cycles were performed in a Perkin-Elmer TGA-7 thermogravimetric analyzer apparatus using typically ∼3 mg samples suspended in a quartz tube (i.d. 20 mm) on a platinum pan (i.d. 5 mm). One series of 10 cycles was done with 100 mg of sample to collect enough material for X-ray diﬀraction (XRD) analysis. The gas ﬂow rate was 0.04 dm3/min (0.1 dm3/min in the case of the 100 mg sample) and the temperature and gas used were controlled by Pyris software. Data on sample mass during the experiments were monitored and conversions were calculated on the basis of mass change, assuming that mass change occurs only due to formation/decomposition of CaCO3 and reduction/oxidation of CuO. The calcination/reduction steps were done during heating from 600 to 800 °C in an atmosphere of CH4. The oxidation/carbonation steps were performed during cooling from 800 to 600 °C, and carbonation continued for 12 min at 600 °C. The heating/cooling rate was 50 °C/min. These conditions were typical, but some variation was incorporated in particular runs in order to separate the calcination and reduction steps, as well as the oxidation and carbonation steps (this is described for particular runs and presented in subsequent sections). The sample morphologies were observed with a Hitachi S3400 scanning electron microscope (SEM) with 20 kV of accelerating voltage under high vacuum. The microscope was equipped with energy dispersive X-ray (EDX) analyzers, which enabled the determination of elemental composition at points of interest on the sample surface. The samples were coated with gold/palladium before SEM examination and images obtained by secondary electrons are presented here. 10751

dx.doi.org/10.1021/es202292c |Environ. Sci. Technol. 2011, 45, 10750–10756

Environmental Science & Technology

Figure 1. Calcination/reduction/oxidation/carbonation cycle of CaO/ CuO-based pellets, 1 mm particle size (20% steam present in reacting gas): (a) separated steps, (b) simultaneous calcination-reduction and oxidation-carbonation.

The original pellets after calcination were submitted to nitrogen adsorption/desorption analyses of pore surface area. Pore surface areas (BET) and pore volume distributions (BJH) were determined using a Micromeritics TriStar II 3020 V1.02 N2 sorption analyzer. Identiﬁcation/quantiﬁcation of compounds in the original, calcined, reduced, and pellet samples after 10 calcination/reduction/oxidation/carbonation cycles were done using the XRD method. XRD data were collected on a Bruker D500TT diffractometer over the angular range 1070° (2-theta) in 0.02° steps and 20 s per step. The phases were identiﬁed and quantitative analyses of the samples were obtained using alpha-alumina (Al2O3) as an internal spiking standard.

3. RESULTS AND DISCUSSION Prepared CaO/CuO-based pellets were tested in the TGA to explore their activity during calcination of CaCO3, reduction of CuO, oxidation of Cu, and carbonation of CaO during cycles. Figure 1 shows the results from two types of tests and it should be mentioned that both were done with 20% steam present in the reactive gas. The ﬁrst run (Figure 1a) represents calcination in a nitrogen atmosphere, reduction by CH4, oxidation by air, and carbonation by a gas mixture containing 20% CO2. The purpose of this test was to separately explore activity of pellets during calcination and reduction, as well as during oxidation and carbonation. It can be seen that a slight mass loss occurs at the beginning of the run, which represents release of a small amount of the remaining water adsorbed on the sample and dehydration of some hydrates. The ﬁrst more intensive mass loss occurred at 400450 °C, which is a result of calcination of Ca(OH)2. Calcination of CaCO3 ﬁnished at 800 °C, and after that sample

ARTICLE

mass remained constant. This run continued isothermally at 800 °C, but N2 was switched with CH4, which caused rapid mass loss due to reduction of CuO. The calcined/reduced sample was ﬁrst oxidized by air, which resulted now in fast and quantitative oxidation of Cu, which ﬁnished in less than 3 min. This fast oxidation of Cu is in good agreement with the literature data23 where detailed kinetic studies of oxidation of Cu have demonstrated that complete conversion takes place in less than 1 min. The sample was then carbonated and the carbonation proﬁle obtained is characteristic for synthetic sorbents,19 with a signiﬁcant conversion rate during the diﬀusion controlled stage. Finally, one more calcination/reduction was completed and the reaction proﬁles obtained were similar with those for the fresh material. The test presented in Figure 1a clearly demonstrates the capability of CaO/CuO-based composite materials to be used for CO2 capture with subsequent regeneration of the sorbent employing the heat obtained by reduction of the CuO oxygen carrier. A very important ﬁnding here is that both reduction of CuO and oxidation of Cu are fast and quantitative reactions, which highlights the potential practical application of these CaOCuO pellets. Moreover, it is also signiﬁcant that CaO reached carbonation conversion ∼65% in the ﬁrst cycle, but with the potential for a further increase. A new class of cycles for postcombustion CO2 capture that we are proposing in this study includes CO2 capture by a calcined CaO/CuO-based composite material. Taking into account that a ﬂue gas always contains some oxygen and that the test presented in Figure 1a showed rapid oxidation of Cu it was reasonable to suppose that oxygen contained in ﬂue gas can oxidize Cu during the carbonation step. Moreover, it has been also demonstrated in the literature23 that oxidation of Cu in similar Cu-containing materials was completed in ∼20 s at 650 °C in a gas mixture containing only 1.2% O2. In addition, the rapid reduction of CuO (Figure 1a) can provide heat for CaCO3 decomposition, which means that calcination and reduction can occur simultaneously in the same pellet. Finally, this also opens the possibility that this material can also remove O2 from the ﬁnal ﬂue gas, thus reducing eﬀorts to purify CO2 before it is cleaned, compressed, and sequestered. Therefore, in order to simulate conditions expected in the proposed process, two-step cycles have been designed, which include (i) calcinationreduction and (ii) oxidation carbonation steps. The proﬁle obtained during calcinationreduction of the fresh CaO/CuO-based pellets followed by one further cycle is presented in Figure 1b. It can be seen that after an initial sample mass loss, a mass increase starts at a temperature of ∼480 °C, which can be explained by the steam reforming reaction of CH4 in the presence of CaO. CH4ðgÞ þ CaOðsÞ þ 2H2 OðgÞ ¼ CaCO3ðsÞ þ 4H2ðgÞ ΔHr0 ¼ 14 kJ=mol

ð6Þ

The further increase of temperature above 720 °C caused reduction of CuO by CH4 and, at the same time, thermodynamic conditions favored calcination of CaCO3. Both reactions ﬁnished by ∼2 min, i.e., when temperature reached 800 °C. After that, the cooling step began and, at the same time, CH4 was switched with a gas mixture containing 5% O2 and 20% CO2, which simulated a ﬂue gas. The ﬁrst ramp in the proﬁle is due to oxidation of Cu and, after that, the sample mass was relatively constant. According to the thermodynamic equilibrium of the CaO/CO2/CaCO3 system, 20% CO2 in the reactive gas should allow carbonation 10752

dx.doi.org/10.1021/es202292c |Environ. Sci. Technol. 2011, 45, 10750–10756

Environmental Science & Technology

ARTICLE

Figure 2. Oxidationcarbonationreductioncalcination cycles with CaO/CuO-based pellets. Conditions are the same as those in the run presented in Figure 1b.

Figure 3. Results of nitrogen physisorption test of CaO/CuO-based pellets.

even at >800 °C. However, in this experiment carbonation took place only when the temperature dropped below 700 °C, which is, on ﬁrst consideration, an unexpected result. However, it should be noted here that oxidation of Cu is a rapid and highly exothermic reaction, which rapidly releases heat into the pellet particles causing their heating above the programmed temperature, i.e., the temperature measured by thermocouple. That is, the thermocouple measures the temperature in the gas around the pellets, but the reaction occurs through the volume of “superheated” pellet particles. It appears that the temperature inside the pellet particles is more than 100 °C higher than that in the surrounding gas, which is important to consider in the design of any process. Namely, this phenomenon can cause undesired sintering in pellet particles (or even agglomeration), and its eﬀects should be mitigated in practice to avoid “superheating”. On the other hand, this eﬀect also oﬀers the scenario in which the pellets are oxidized by ﬂue gas (smaller O2 concentration), which means slower conversion rate and heat release (when compared with that during oxidation by air), and as noted earlier helps purify the CO2 stream produced. Furthermore, the direction of solid material transport from calciner to air reactor becomes more desirable than that from carbonator to air reactor (this scheme and further discussion is presented below). The TGA tests presented in Figure 1 show that the carbonation of pellets previously reduced can provide some beneﬁts. Therefore, further runs aimed at testing pellet activity during a multicycle operation were done under conditions used in the run

Figure 4. SEM images of interior of CaO/CuO-based pellet: (a) original, (b) calcinedreduced, and (c) after 10 calcination reductionoxidationcarbonation cycles.

for Figure 1b. Furthermore, there is evidence that H2O(g) enhances carbonation,24 and its presence is expected during a real CaCu looping cycle process in both the carbonator (ﬂue gas usually contains steam) and calciner (reduction of CuO by CH4 generates steam). Therefore, steam eﬀects were also tested, i.e., the cycles were performed with steam and without steam present. The conversion proﬁles obtained during three-cycle runs are presented in Figure 2. As is expected, steam enhances carbonation and it is more pronounced with cycle number. Oxidation and reduction are not noticeably aﬀected by steam, and complete oxidation conversions are reached in all three cycles, which is important for practical application. The same result was obtained in a 10-cycle run (not presented here), which is in agreement with results presented in the literature.23 As has been seen for other CaO-based sorbents,14,1820 carbonation conversions decreased with the cycle number, but were still high after the third cycle (∼50% with steam present). It can be seen that both O2 and CO2 carrying capacity (mass of O2 and CO2 per 10753

dx.doi.org/10.1021/es202292c |Environ. Sci. Technol. 2011, 45, 10750–10756

Environmental Science & Technology

ARTICLE

Table 1. EDX Analyses of CalcinedReduced CaO/ CuO-Based Pellet (For Which SEM Image Is Presented in Figure 4b) site 1

site 2

site 3

site 4

site 5

Al2O3 [wt%]

5.73

3.66

2.15

11.56

8.78

CaO [wt%]

17.03

10.38

6.96

76.20

83.63

CuO [wt%]

77.24

85.96

90.89

12.25

5.59

Figure 6. Schematic representation of CaL integrated with CLC (CaLCLC).

Figure 5. XRD spectra of CaO/CuO-based pellet samples: original, calcined at 800 °C in N2, reduced at 800 °C in CH4, and after 10 calcination/reduction/oxidation/carbonation cycles.

100 g of sorbent) are comparable with those obtained with materials used solely for transport of O2 in CLC or for transport of CO2 in CaL processes. Moreover, the synthesis procedure of CaO/CuO-based pellets allows managing the CaO/CuO ratio, i.e., varying O2 carrying capacity depending on a demand for O2 in a particular process. Pore surface area and pore volume distribution of the tested pellets were determined by nitrogen physiosorption and results are presented in Figure 3. The BET pore surface area of the calcined sample is 1.69 m2/g, which is lower than is expected for this kind of material. The possible reasons for this are a high content of CuO in the sample (45 wt %), and a preparation procedure which is a type of wet mechanical mixing. Namely, a powder of CuO, which is a lightly porous material (see SEMEDX analyses below) was used for the synthesis. Somewhat higher pore surface areas, 6.53 and 10.31 m2/g, have been reported for two oxygen carriers (82.5 wt % CuO supported by Al2O3) prepared by coprecipitation.25 The pore size distribution has a peak at ∼3 nm, which corresponds to the ﬁrst peak of the BJH pore volume distribution. Another very pronounced peak of pore volume distribution is placed at ∼100 nm. Figure 4 presents SEM analyses of CaO/CuO-based pellets taken from the interior of the particles. The original sample has low porosity; however, after calcinationreduction it becomes a more porous material. Two types of morphology are noticeable (they were analyzed by EDX), and results are presented in Table 1. It can be seen that sites which contain larger grains

(1, 2, and 3) are rich in Cu, and sites with smaller grains are mainly CaO. This conﬁrms that the high content of CuO in the pellets is a cause for their low pore surface area (Figure 3). However, the lower porosity of CuO sites is not critical because TGA experiments (Figures 1 and 2) conﬁrmed high reactivity of the oxygen carrier. A more developed porous pattern can be seen for CaO sites and their morphology is similar to that seen for calcium aluminate pellets.19 It should be noted that the smaller contact area between CuO and surrounding CaO/Al2O3 material may reduce formation of copper aluminate (CuAl2O4) during cycles. Finally, as is usual with other CaO-based sorbents for CO2 capture, cyclic calcination/carbonation causes sintering; therefore, larger grains and pores are present in the residue after 10 cycles (Figure 4c). XRD analysis of CaO/CuO-based pellets conﬁrms the results of TGA tests. The XRD diﬀractograms obtained, with the most intense peaks designated, are presented in Figure 5. As expected, the main compounds present in the crystal phase of the original samples are CaCO3 (calcite and aragonite), Ca(OH)2 (portlandite), CuO (copper oxide), and CaAl4O7 (grossite). During calcination, the calcium compounds decomposed forming CaO (lime). It is interesting to note that copper aluminates are not formed after calcination, and moreover, their presence is not identiﬁed either in the reduced or in the cycled sample. Despite the fact that copper aluminates are fully reducible,25 their absence in this case is desirable because Al2O3 remains available to form Ca12Al14O33 (mayenite) which improves the performances of the CaO sorbent.19 The main change after reduction is the formation of elemental copper (Cu). Moreover, it should be noted here that the presence of Cu2O (cuprite) is also seen in the sample after these cycles. This is an indicator that these pellets were “preheated” during reduction as well during cycles. Most likely their temperature was higher due to larger mass of “XRD sample” (100 mg) than was the case in other TGA tests (∼3 mg). Namely, the formation of Cu2O is characteristic for both reduction of CuO and oxidation of Cu at temperatures ∼800 °C or higher.23 This implies that operation at lower temperatures and its control during the process is important to maintain desired sorbent performances. Finally, it should be mentioned that some Cu2O detected by XRD may be formed due to partial oxidation of Cu by air during handling with the samples. The research presented here has demonstrated that integration of CaL with CLC is a real possibility with practical application. The composite materials containing CaO and CuO are good candidates for this process. A block diagram of the proposed process that is compatible with the composite materials we have developed is presented in Figure 6. As can be seen, this system for 10754

dx.doi.org/10.1021/es202292c |Environ. Sci. Technol. 2011, 45, 10750–10756

Environmental Science & Technology postcombustion CO2 capture has three reactors: carbonator, calciner, and air reactor. Three main possibilities for the direction of solids circulation are worthy of attention: a Carbonator f calciner f air reactor f carbonator b Carbonator f calciner carbonator c Carbonator f air reactor f calciner f carbonator The ﬁrst route (a) is represented by solid arrows, and it describes a basic cycle which starts with CO2 capture in the carbonator. Carbonatedoxidized sorbent (CaCO3/CuO) circulates from the carbonator to the calciner where calcination takes place due to heat of exothermic reduction of CuO (reaction 5). Calcinedreduced solid material is then transferred to the air reactor to oxidize Cu (reaction 4). These regenerated pellets are now ready for a new CO2 capture step in the carbonator. The second route (b) is interesting because only the carbonator and calciner are required for the process: oxygen is always present in the ﬂue gas. This means that regenerated solid carrier is directly returned from the calciner to the carbonator where, apart from CO2 capture, oxidation of Cu by oxygen from ﬂue gas takes place. This route provides an interesting opportunity for ﬂue gases with a high O2/CO2 ratio. It is also possible to employ a high air/fuel ratio in the combustor, which can enhance combustion eﬃciency and supply excess oxygen to the carbonator for oxidation of Cu. In this case one should be aware that CO2 capture eﬃciency can be diminished (more N2 in the carbonator coming with the excess air in the combustor). However, this type of cycle does not require an air reactor—since the carbonator becomes at the same time an air reactor. A combination of routes (a) and (b) is also very interesting and should be considered as a valuable possibility. Namely, the ﬂow of regenerated pellets from the calciner can be divided into two streams: one entering the air reactor and the other entering the carbonator. The ratio of ﬂow rates of these two streams can be regulated according to the amount of oxygen available in the carbonator and/or according to the desired distribution of heat between the air reactor and carbonator. The third route (c) is the reverse of the ﬁrst scenario (a), which means the transfer of carbonated sorbent to the air reactor. After oxidation of Cu, the pellets transfer to the calciner, and after regeneration/reduction they continue to the carbonator. This ﬂow direction requires strict control of the temperature in the air reactor; otherwise a signiﬁcant amount of CaCO3 can be calcined at an elevated temperature diminishing CO2 capture eﬃciency (CO2 leaves air reactor with the N2 stream). This route is also discussed in detail by Abanades et al.16,17 and it should be noted that the direction which includes oxidation of CaCO3/ Cu appears to be the only possible one in the case of sorptionenhanced reforming (SER) because copper is always in reduced form in the reformer and must be ﬁrst oxidized before CaCO3 calcination. A technoeconomic study can explore which route is most promising. However, it is shown here that a new class of looping cycles for postcombustion CO2 capture is feasible and worthy of consideration, and it is experimentally feasible to employ CaL and CLC instead of the integration of CaL with oxy-fuel combustion. The Cu-based oxygen carrier is a promising candidate for CLC due to its exothermic reduction, and even more so, in that CaO and CuO can be used in a mixed pellet, having advantages such as heat release/consumption in the same pellet particle. Moreover, CaO/Al2O3 acts as a solid porous support for CuO. That also has the beneﬁt that another support for CuO is not required, reducing total solids ﬂow. Fluidized bed (FBC)

ARTICLE

reactors are most likely to be employed in the proposed process, but other options, such as ﬁxed beds, are also open to consideration. Finally, the possibility that this type of system may also oﬀer a method of ﬁnal puriﬁcation of the CO2 from oxygen is a valuable consideration which has not been explored here for reasons of space.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]; phone: (613) 996-2868; fax: (613) 992-9335.

’ REFERENCES (1) Intergovernmental Panel on Climate Change (IPCC). Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change; IPCC: Geneva, Switzerland, 2007. (2) Mohr, S. H.; Evans, G. M. Forecasting Coal Production until 2100. Fuel 2009, 88, 2059–2067. (3) Hook, M.; Aleklett, K. Historical Trends in American Coal Production and a Possible Future Outlook. Int. J. Coal Geol. 2009, 78, 201–216. (4) Herzog, H. What Future for Carbon Capture and Sequestration? Environ. Sci. Technol. 2001, 35, 148–153. (5) Yang, H.; Xu, Z.; Fan, M.; Gupta, R.; Slimane, R. B.; Bland, A. E.; Wright, I. Progress in Carbon Dioxide Separation and Capture: A Review. J. Environ. Sci. 2008, 20, 14–27. (6) Blamey, J.; Anthony, E. J.; Wang, J.; Fennell, P. S. The Use of the Calcium Looping Cycle for Post-Combustion CO2 Capture. Prog. Energy Combust. Sci. 2010, 36, 260–279. (7) Anthony, E. J. Solid Looping Cycles: A New Technology for Coal Conversion. Ind. Eng. Chem. Res. 2008, 47, 1747–1754. (8) Abanades, J. C.; Anthony, E. J; Wang, J.; Oakey, A. Fluidized Bed Combustion Systems Integrating CO2 Capture with CaO. Environ. Sci. Technol. 2005, 39, 2861–2866. (9) Buhre, B. J. P.; Elliot, L. K.; Sheng, C. D.; Gupta, G. P.; Wall, T. F. Oxy-fuel Combustion Technology for Coal-Fired Power Generation. Prog. Energy Combust. Sci. 2005, 31, 283–307. (10) Shen, L.; Wu, J.; Xiao, J. Experiments on Chemical Looping Combustion of Coal with a NiO Based Oxygen Carrier. Combust. Flame 2009, 156, 721–728. (11) Berguerand, N.; Lyngfelt, A. Design and Operation of a 10 kWth Chemical-Looping Combustor for Solid Fuels - Testing with South African Coal. Fuel 2008, 87, 2713–2726. (12) Lu, D. Y.; Hughes, R. W.; Anthony, E. J. Ca-Based Sorbent Looping Combustion for CO2 Capture in Pilot-Scale Dual Fluidized Beds. Fuel Process. Technol. 2008, 89, 1386–1395. (13) Abanades, J. C.; Grasa, G.; Alonso, M.; Rodriguez, N.; Anthony, E. J.; Romeo, L. M. Cost Structure of a Postcombustion CO2 Capture System using CaO. Environ. Sci. Technol. 2007, 41, 5523–5527. (14) Manovic, V.; Charland, J.-P.; Blamey, J.; Fennell, P. S.; Lu, D.; Anthony, E. J. Inﬂuence of Calcination Conditions on Carrying Capacity of CaO-Based Sorbent in CO2 Looping Cycles. Fuel 2009, 10, 1893–1900. (15) Lyon, R. K.; Cole, J. A. Unmixed Combustion: An Alternative to Fire. Combust. Flame 2000, 121, 249–261. (16) Abanades, J. C.; Murillo, R.; Fernandez, J. R.; Grasa, G.; Martinez, I. New CO2 Capture Process for Hydrogen Production Combining Ca and Cu Chemical Loops. Environ. Sci. Technol. 2010, 44, 6901–6904. (17) Abanades, J. C.; Murillo, R. et al. Method of Capturing CO2 by Means of CaO and the Exothermal Reduction of a Solid. European Patent Application EP2305366 A1, 2009. (18) Manovic, V.; Anthony, E. J. Screening of Binders for Pelletization of CaO-Based Sorbents for CO2 Capture. Energy Fuels 2009, 23, 4797–4804. 10755

dx.doi.org/10.1021/es202292c |Environ. Sci. Technol. 2011, 45, 10750–10756

Environmental Science & Technology

ARTICLE

(19) Manovic, V.; Anthony, E. J. CaO-Based Pellets Supported by Calcium Aluminate Cements for High-Temperature CO2 Capture. Environ. Sci. Technol. 2009, 43, 7117–7122. (20) Manovic, V.; Anthony, E. J. Long-Term Behavior of CaO-Based Pellets Supported by Calcium Aluminate Cements in Long Series of CO2 Capture Cycles. Ind. Eng. Chem. Res. 2009, 48, 8906–8912. (21) Adanez, J.; Gayan, P.; Celaya, J.; de Diego, L. F.; GarciaLabiano, F.; Abad, A. Chemical Looping Combustion in a 10 kWth Prototype using a CuO/Al2O3 Oxygen Carrier: Eﬀect of Operating Conditions on Methane Combustion. Ind. Eng. Chem. Res. 2006, 45, 6075–6080. (22) Cho, P.; Mattisson, T.; Lyngfelt, A. Comparison of Iron-, Nickel-, Copper and Manganese-Based Oxygen Carriers for ChemicalLooping Combustion. Fuel 2004, 83, 1215–1225. (23) Chuang, S. Y.; Dennis, J. S.; Hayhurst, A. N.; Scott, S. A. Kinetics of the Oxidation of a Co-precipitated Mixture of Cu and Al2O3 by O2 for Chemical-Looping Combustion. Energy Fuels 2010, 24, 3917–3927. (24) Manovic, V.; Anthony, E. J. Carbonation of CaO-Based Sorbents Enhanced by Steam Addition. Ind. Eng. Chem. Res. 2010, 49, 9105–9110. (25) Chuang, S. Y.; Dennis, J. S.; Hayhurst, A. N.; Scott, S. A. Development and Performance of Cu-Based Oxygen Carriers for Chemical-Looping Combustion. Combust. Flame 2008, 154, 109–121.

10756

dx.doi.org/10.1021/es202292c |Environ. Sci. Technol. 2011, 45, 10750–10756

ARTICLE pubs.acs.org/est

Modeling the Relative GHG Emissions of Conventional and Shale Gas Production Trevor Stephenson,*,† Jose Eduardo Valle,‡ and Xavier Riera-Palou† † ‡

Shell Global Solutions (U.K.), Shell Technology Centre Thornton, P.O. Box 1, Chester CH1 3SH, United Kingdom Shell Global Solutions International B.V., Kessler Park 1, Rijswijk 2288 GS, The Netherlands

bS Supporting Information ABSTRACT: Recent reports show growing reserves of unconventional gas are available and that there is an appetite from policy makers, industry, and others to better understand the GHG impact of exploiting reserves such as shale gas. There is little publicly available data comparing unconventional and conventional gas production. Existing studies rely on national inventories, but it is not generally possible to separate emissions from unconventional and conventional sources within these totals. Even if unconventional and conventional sites had been listed separately, it would not be possible to eliminate sitespeciﬁc factors to compare gas production methods on an equal footing. To address this diﬃculty, the emissions of gas production have instead been modeled. In this way, parameters common to both methods of production can be held constant, while allowing those parameters which diﬀerentiate unconventional gas and conventional gas production to vary. The results are placed into the context of power generation, to give a 00 well-to-wire00 (WtW) intensity. It was estimated that shale gas typically has a WtW emissions intensity about 1.82.4% higher than conventional gas, arising mainly from higher methane releases in well completion. Even using extreme assumptions, it was found that WtW emissions from shale gas need be no more than 15% higher than conventional gas if ﬂaring or recovery measures are used. In all cases considered, the WtW emissions of shale gas powergen are signiﬁcantly lower than those of coal.

’ INTRODUCTION 00

00

Tight Gas. A gas well is called tight if it requires stimulation before gas production can begin, or if it needs stimulation to maintain production. Tight sands, shale gas, and coal bed methane are all examples of 00 tight00 gas production and are collectively described as 00 unconventional00 gas. Tight gas production is characterized by low rock permeability. Conventional gas occurs in rocks with a permeability of more than 1000 microdarcy, whereas tight sands have a permeability of 1100 microdarcy and shale permeability is 1 microdarcy or less.1 Where the permeability is low, gas can only be collected within a small radius of the well bore. As a result, more or longer wells must be drilled and the rock must be fractured to access the gas. Unconventional gas drilling diﬀers from conventional in the large amounts of water used for hydraulic fracturing, approximately 24 million gallons (750015 000 m3) of water per well.2 The ﬂuid pumped into the well consists mainly of water and sand (∼98%) with various chemicals (ﬂow improvers to keep the sand in suspension, friction reducers, surfactants, corrosion inhibitors, acids, etc.). Much of this water ﬂows back to the surface following fracturing. In addition to these chemicals, ﬂowback water contains salt and other minerals. Some ﬂowback water can be recovered for reuse but, unless r 2011 American Chemical Society

there is an opportunity to reinject the water locally, it must be treated before disposal. There are no systematic diﬀerences in gas composition between unconventional and conventional gas reservoirs. Both are equally likely to contain high or low levels of contaminants such as CO2 or H2S. Production of water over the life of the well varies considerably among rock formations. Coal bed methane typically produces a lot of water; shale gas is typically quite dry. Shale Gas. Unconventional gas now makes up about 50% of North American gas production and is predicted to rise to 64% by 2020.3,4 For the world as a whole, EIA estimates that 00 adding the identified shale gas resources to other gas resources increases total world technically recoverable gas resources by over 40% to 22 600 trillion cubic feet00 .5 The most signiﬁcant trend in U.S. natural gas production is the rapid rise in production from shale formations. This is largely attributable to advances in horizontal drilling and well stimulation technologies and improvements in their cost eﬀectiveness. Received: July 13, 2011 Accepted: November 1, 2011 Revised: October 21, 2011 Published: November 15, 2011 10757

dx.doi.org/10.1021/es2024115 | Environ. Sci. Technol. 2011, 45, 10757–10764

Environmental Science & Technology

ARTICLE

Figure 1. Simpliﬁed well-to-wire (WtW) pathway.

’ METHODOLOGY: COMPARING CONVENTIONAL AND SHALE GAS Inventory versus Modeling Approach. It is possible to estimate the emissions intensity of the gas industry from the total emissions reported for a territory. For example, Jaramillo6 estimated the emissions intensity of natural gas powergen using sector emissions data from the 2004 U.S. EPA Inventory Report.7 At that time, it was not possible to distinguish unconventional gas and conventional gas production within the totals. The EPA’s 2011 Inventory Report8 attempted to separate the two, but the underlying data do not contain enough detail to do this accurately and their results could only be based on estimates.9 However, even if individual production sites had been listed separately, it would not be possible to eliminate site-specific factors (such as gas composition) to compare unconventional gas and conventional gas production methods on an equal footing. To overcome these diﬃculties, the emissions of gas production can be modeled. In this way, those parameters which are common to both methods of production can be held constant, while allowing those parameters which characterize unconventional gas and conventional gas production to vary. Unlike an inventory approach, where data ranges can only be identiﬁed for diﬀerent producers, ranges of uncertainty can be evaluated for individual parameters within the production process. This modeling work adopted a four-stage approach to understanding the diﬀerences between conventional and shale gas production. First, two generic base cases were deﬁned: conventional gas and shale gas, deﬁned on the basis of typical parameters. The results for conventional gas production provide a simulated, but realistic yardstick against which shale gas production can be judged. Second, a sensitivity analysis was conducted by varying the production parameters one by one within the likely range of variation at actual production sites. The results show which parameters have most inﬂuence on GHG emissions intensity and which are relatively unimportant. Third, a “worst case” analysis was conducted. Whereas the sensitivity analysis varied one parameter at a time, the worst case analysis looked at the cumulative eﬀect of these changes. The results show how high the GHG emissions might be in the most unfavorable circumstances.

Finally, the relevance of the ﬁndings to shale gas production in the U.S. and elsewhere is discussed. The ﬁndings are compared against the results of other recent publications. Scope. Unconventional gas production requires large numbers of wells and those wells may be reworked during the lifetime of the project. Well drilling and completion emissions are potentially significant compared to the total emissions over the lifecycle of a project. Furthermore, the EPA recently increased its estimate of the fugitive emissions from unconventional well completions and workovers. Well completion emissions were therefore also included in the scope. The following items were included in the analysis: • CO2, CH4, and N2O emissions associated with combustion of fuels at every stage of the lifecycle, applying 2007 IPCC AR4 factors for 100-year global warming potential. (Some authors have considered 20-year global warming potential factors, but use of these is not widely accepted.) • Venting, ﬂaring, and fugitive (VFF) emissions from gas production facilities and transport pipelines. • Lifecycle emissions of imported fuels, e.g., production of diesel fuel and the fuels used for grid-supplied power. • Emissions from the transport and treatment of produced and ﬂowback water. The analysis does not include the following: • Non-GHG environmental impacts. • Land use change emissions associated with access road and well pad construction (these were assessed but found to make no material contribution to WtW emissions). • Exploration and appraisal of new gas ﬁelds. Relatively few wells (∼1%) are drilled in the appraisal phase compared to the total number of producing wells. • Lifecycle emissions of the chemicals used in fracturing or gas treatment. • Emissions associated with construction or end-of-life disposal of equipment. Functional Unit. EIA data for 2009 show that 94% of U.S. coal is used for powergen10 so, when comparing the life cycles of gas and coal, it is appropriate to consider 1 kWh of electricity to be the functional unit of the life cycle analysis and compare WtW rather than combustion emissions. Gas powergen typically has higher efficiency than coal powergen, which tends to reduce the WtW emissions of gas relative to coal. 10758

dx.doi.org/10.1021/es2024115 |Environ. Sci. Technol. 2011, 45, 10757–10764

Environmental Science & Technology Allocation of Emissions to Gas and Coproducts. A gas production project may have multiple coproducts: not only gas, but also condensate, ethane, and LPG. To calculate the emissions intensity, the emissions total must be divided between the sales gas and other coproducts. Allocating emissions in this way removes some of the uncertainty introduced by differences in gas composition at various locations. Condensate and LPG carry a proportionate share of the total emissions so the results remain comparable with locations that produce more or less condensate and LPG. Emissions were allocated to coproducts in proportion to their energy content. The allocation was the same for conventional and shale gas because the same gas composition was used in both models.

’ CONSTRUCTION OF THE MODEL To show the relative importance of gas production emissions in the context of the fuel life cycle, it is assumed that gas is transported by pipeline to a power station and used to generate electricity. The gas lifecycle has been simpliﬁed to four steps, as shown in Figure 1: extraction, gas treatment, pipeline transmission, and combustion at power station to generate electricity (excluding transmission losses). Choice of Model Parameters. Once the gas has been gathered, there are no essential differences in subsequent treatment. This study focused on the parameters that are necessarily different between unconventional and conventional gas in the wells and gathering system. Extensive data are available for North America, via the U.S. Energy Information Administration (EIA), U.S. Environmental Protection Agency (EPA), and other government and industry bodies. For this reason, the model reﬂects unconventional gas production in a North American context, but the insights generated can be applied to unconventional gas production elsewhere. Shale gas is the major source of growth in unconventional gas production and therefore the analysis looked mainly to shale gas sources. Model Gas Composition. Data from the 2011 EPA Inventory Report8 show that there is no systematic variation in the CO2 content of conventional and unconventional gas wells. The data show an almost complete overlap, with both types of gas ranging from nearly zero to more than 7%. A single gas composition was therefore used to model both conventional and unconventional production. Data from EIA and EPA reports were used to derive a composition typical of average U.S. gas composition (Supporting Information, S2.1). Gas Treatment: Common Elements. Gas treatment consisted of the following elements (Supporting Information, S2.2): well water handling, condensate separation and treatment, acid gas removal by amine treatment, dehydration, dewpointing and LPG fractionation, export compression, sulfur recovery, flare stack, and fugitive emissions. Fugitive emissions were calculated using the facility-level factors from the 2009 API Compendium:11 0.17% of the gas is lost to fugitives from onshore production and 0.18% is lost to fugitive emissions from gas processing. (The EPA 2011 Inventory Report8 estimated that total methane emissions from the natural gas industry were 10 535 kilotonnes in 2009, of which gas processing accounted for 8%, or roughly 0.21% of total gas production, which is in good agreement with the API value.) The emissions allocated to sales gas in this process amount to 4.18 gCO2e/MJ, of which 34% is fugitive methane emissions.

ARTICLE

Transmission Pipeline. It was assumed that gas was transported 900 miles (1440 km) from gas field to power plant, as used in a recent NETL study of the emissions intensity of Natural Gas Combined Cycle (NGCC) powergen.12 Based on EIA data13 it was calculated that 1.4% of the gas is consumed as fuel for the compressor stations (Supporting Information, S2.3). Fugitive emissions were calculated using facility-level factors for transmission pipelines from the 2009 API Compendium:11 0.066% of the gas is lost to fugitive emissions over 900 miles. The emissions intensity of pipeline transmission is then 1.94 gCO2e/ MJ. (The EPA’s 2011 Inventory Report8 estimated that total methane emissions from the natural gas industry were 10 535 kilotonnes in 2009, of which transmission and storage accounted for 20%, or roughly 0.52% of total gas production. If this ﬁgure were used, the emissions intensity of pipeline transmission would rise to 3.99 gCO2e/MJ.) Power Station. It was assumed that natural gas is burned in the average U.S. power station. Jaramillo6 quoted 2003 U.S. EIA data which showed natural gas power plant efficiencies ranging from 28% to 58%. The average efficiency (total power/total fuel) was 38.7%.14 EIA data15 show that by 2009 (the last complete year for which data are available) natural gas generating capacity had increased by 41% and the average eﬃciency increased to 43.0%. By contrast, coal generating capacity fell by 11% over the same period and its eﬃciency remained almost constant: 33.1% in 2003 and 32.8% in 2009. The emissions intensity of gas powergen has improved relative to coal since Jaramillo’s paper appeared (Supporting Information, S2.4). For Life Cycle Assessment, emissions are conventionally reported per MJ of lower heating value (LHV). For the model, the eﬃciency of powergen was assumed to be 43.0%, or 47.6% on an LHV basis: 2.10 MJ of gas is needed to generate 1 MJ of electricity. The emissions intensity of the power station is then 122 gCO2e/MJ or 440 gCO2e/kWh. Well-to-Wire Totals: Common Elements. The total WtW emissions (excluding well head operations) amount to 485.2 gCO2e/kWh. This value is in good agreement with data for gas powergen published by Jaramillo6 and NETL12 (Supporting Information, S2.5) and therefore offers a reasonable baseline for comparison of the remaining elements of conventional and unconventional gas production.

’ RESULTS: CONVENTIONAL GAS Production Profile. Both conventional and shale gas wells start from a high initial flow rate, which declines over time. Production from a field is maintained by drilling new wells. Real wells exhibit a wide variation in behavior, so a simplified approach was needed. For this analysis, it was only necessary to calculate the emissions intensity per unit of gas; the time profile is unimportant. The well drilling and completion emissions were divided by the estimated ultimate recovery (EUR). The same EUR value of 2.0 Bcf was used for both conventional and shale gas so the results would show the relative contribution of the diﬀerent production methods without distortions due to diﬀerences in well productivity. The eﬀect of changing this assumption was addressed in the sensitivity analysis. Well Drilling. The reference case assumed an annual average production 750 mmscf/d (21.1 Msm3/d) from 500 wells and, following the simplified model above, 137 wells must be drilled 10759

dx.doi.org/10.1021/es2024115 |Environ. Sci. Technol. 2011, 45, 10757–10764

Environmental Science & Technology each year to maintain production at this level. It was assumed that no fracturing was required. Estimates of methane releases during well completion range from the EPA/GRI factor of 0.71 tCH4 per completion (based on 1992 data16) to the API Compendium onshore well factor of 25.9 tCH4 per completion-day11 (based on EIA data from 2000). Allowing two days for conventional well completion, methane emissions would be 0.20% of lifetime production. Well Completion Emissions Abatement. In an exploration situation, there may be no opportunity to recover methane releases for sale. In these circumstances, methane releases are usually flared (although new opportunities may become available using micro-LNG or electric powergen systems). In a production situation, where a gas pipeline exists, it is technically possible to recover methane releases. For example, Williams Inc.17 estimated 91% recovery at Picecance basin. In 2010, the EPA estimated that that 51% of well completion/ workover emissions were ﬂared or recovered.9 Flaring of completions and workovers is required in Wyoming; however, it is not required in Texas, New Mexico, or Oklahoma. EPA assumed no completions were ﬂared in those states and then took the ratio of unconventional wells in Wyoming to the unconventional wells in all four states to estimate the percentage of well completions and workovers that are ﬂared. EPA assumed that this sample was indicative of the rest of the U.S. For the generic conventional gas base case, it was therefore assumed that 51% of the gas used for powergen originates from wells where completion emissions were ﬂared and the rest vented. Production Emissions Intensity. For the conventional gas base case, diesel used for well drilling added 0.30 gCO2e/MJ to the common elements. Methane releases from well completion would add another 0.010.65 gCO2e/MJ if vented. If flared at a typical efficiency of 98%, methane release would add only 0.0010.08 gCO2e/MJ. Allowing for an average 51% ﬂaring of methane released in well completion, the model gives an emissions intensity ranging from 4.49 to 4.84 gCO2e/MJ, of which 3136% is fugitive methane emissions from well completion and gas treatment. It is common in the industry to express the intensity of production as the total direct emissions divided by total hydrocarbon production, without allocation to individual products. On this basis, the emissions intensity of conventional hydrocarbon production ranges from 0.211 to 0.228 tCO2e/tHC (where HC = gas, condensate, and LPG). WtW Emissions Intensity. The WtW emissions intensity of the conventional base case ranges from 487.5 to 490.2 gCO2e/ kWh, of which 2.73.2% is methane. Combustion at the power station makes up 89.5% of the total, pipeline transport 3.0%, common elements 6.5%, and well drilling and completion 1.0%. Well drilling makes up a relatively small part of the WtW emissions in this conventional gas model.

’ RESULTS: SHALE GAS Production Profile. A survey of unconventional wells (Supporting Information, S4.1) shows that unconventional wells commonly show a steep decline in production, so that the estimated ultimate recovery (EUR) is typically about 3 years but can be as little as 1 year. A Shell internal rule of thumb for shale gas is that EUR is 12001500 times the initial production (IP) per day. Data from U.S. Geological Survey indicate that an ultimate recovery of 2 Bcf per well is typical of

ARTICLE

horizontal shale gas wells.18 An initial production rate of 1.5 mmscf/d was considered typical for wells of this size. Well Drilling. The reference case assumed an annual average production of 750 mmscf/d (21.1 Msm3/d) from 500 wells and that 137 wells must be drilled each year to maintain production. The scatter on Shell’s correlation suggests that the ultimate recovery could vary between 1 and 3 Bcf, or 91273 wells per year, and USGS data show a similar spread. It was assumed that methane releases were equal to the new EPA factor of 177 tCH4 per well completion.9 The EPA completion factor is an average value used for inventory calculations and is independent of well size. These estimates have been challenged as “dramatically” overstated and “not credible” by IHS CERA19 and ongoing data collection exercises by EPA and API may result in revised values in future. Nevertheless, it is interesting to see what eﬀect methane releases of this scale might have. The factor 177 tCH4 corresponds to 7 days ﬂowback at the assumed initial production rate of 1.5 mmscf/d and 85% methane content. The API factor of 25.9 tCH4/completionday would result in emissions of 181.3 tCH4/completion over 7 days—not dissimilar to the EPA’s estimate. Methane emissions on this scale would amount to 0.46% of lifetime production in the base case (2 Bcf recovery) and 0.92% for the low ultimate recovery case (1 Bcf). Well Fracturing. It was assumed that the well is fractured immediately after drilling and thereafter no further fracturing is conducted. It was assumed that a total of 15 fracturing operations are needed per well, each requiring 2 h of water injection at maximum pressure. (Approximately 35 fracturing operations are conducted each day, allowing for turnaround time between operations). Fracturing fluid was assumed to flow at 50 bbl/min at 10 000 psi (8 m3/min and 689 bar) corresponding to a 00 hydraulic horsepower00 of 12 250 HP. Fracturing Water. NETL quotes a range of 24 million gallons of water needed to fracture each well.3 Water for fracturing may be obtained from various sources. In order of increasing cost they are as follows: pipeline from local river (may need holding ponds to cover periods of low flow); drilling for water from nearby aquifer; use of waste (00 gray00 ) water from cities; and trucking of water by road tanker. New York State’s Supplemental Generic Environmental Impact Statement (SGEIS) for potential natural gas drilling activities in the Marcellus Shale formation estimated that haulage of all materials (including water) for hydraulic fracturing totaled 15 74023 040 truck-miles for a one-well project20 and about 20% less per well for a 10-well pad. At 5 mpg, this is equivalent to 31484608 gallons of diesel. For this analysis, the ﬁgure was rounded up to 5000 gallons. Flowback Water Treatment. Typically, 3070% of the water used will flow back in the days following the fracturing operation.3 Of this volume, some can be recovered for reuse and the rest is sent for water treatment. For this analysis it was assumed that 4 million gallons of water are used per well, of which 50% flows back and is sent for treatment. Treatment was assumed to consist of trucking of water for disposal a round trip distance of 150 miles by road at 100 bbl load per truck at a fuel economy of 5 miles per gallon of diesel and treatment by a relatively energy intensive method: reverse osmosis and evaporation or freezethaw evaporation at 2 kWh/bbl.21 Total emissions from one fracturing operation are therefore of the order of 228 tCO2e, made up of 15 000 gallons of diesel for trucking, and 100 MWh of electricity for water treatment. 10760

dx.doi.org/10.1021/es2024115 |Environ. Sci. Technol. 2011, 45, 10757–10764

Environmental Science & Technology

ARTICLE

Table 1. Shale Gas Operations: Summary of Sensitivity Cases parameter

low emissions

base case

high emissions

ultimate recovery

3 Bcf

2 Bcf

produced water

WGR = 0.1

WGR = 0.4

WGR = 0.8

completion/workover emissions

51.8 tCH4 (API)

177 tCH4 (EPA)

385 tCH4 (EPA)

methane emissions abatement

98%

51%

0%

Wellhead pressure

60 bar

40 bar

20 bar

ﬂowback water for treatment

1 million gal

2 million gal

4 million gal

number of fractures per well

10 per well

15 per well

24 per well + workover

Production Emissions Intensity. Shale gas drilling adds 0.44 gCO2e/MJ to the common elements (more than conventional gas because of the need for hydraulic fracturing). Sourcing and treatment of fracturing water adds another 0.17 gCO2e/MJ. Methane releases during completion are higher than for conventional wells and would add another 2.26 gCO2e/MJ if vented. If flared at a typical efficiency of 98%, methane release would add only 0.29 gCO2e/MJ. Added to the common elements of 4.18 gCO2e/MJ, the emissions allocated to shale gas range from 5.08 to 7.05 gCO2e/ MJ. Allowing for 51% ﬂaring of methane released in well completion, the model gives an emissions intensity for shale gas of 6.02 gCO2e/MJ, of which 54% is fugitive methane emissions from well completion and gas treatment. The intensity of production expressed as the total emissions divided by total hydrocarbon production (i.e., without allocation) was 0.280 tCO2e/tHC. WtW Emissions Intensity. The WtW emissions intensity of the shale gas base case is 499.2 gCO2e/kWh (1.82.4% higher than the conventional gas base cases), of which 4.3% is methane. Although well drilling, fracturing, wastewater disposal, and fugitive emissions have a significant impact on the emissions intensity of production, their effect on WtW emissions is relatively small because the total is dominated by emissions from the power station, pipeline, and common elements. Only 2.8% of the total is made up of wellhead operations upstream of the common elements. The largest unknown is the amount of fugitive emissions, but even if none of the methane releases from well completion were ﬂared, the WtW emissions would be only 507.4 gCO2e/kWh, which is 3.54.0% higher than conventional gas powergen.

’ RESULTS: SENSITIVITY ANALYSIS The model considered a set of parameters representing generic conventional and shale gas production. It was seen that shale gas production has higher emissions, but that these emissions do not add signiﬁcantly to the WtW emissions intensity of powergen. However, future conventional and shale gas production might depart signiﬁcantly from these initial assumptions, so a sensitivity analysis was conducted. Shale Gas Operations. To describe more or less difficult unconventional gas production, the following parameters were varied above and below the base case: • Ultimate recovery: the more gas is recovered from a well, the smaller the contribution of well drilling and fracturing to the emissions intensity per unit of gas produced. A range of 13 Bcf was explored. • Produced water treatment: a watergas ratio of 0.8 represents a doubling of the base case, included for interest. In fact, shale gas is relatively dry shale and the lower value of

1 Bcf

Figure 2. Tornado plot. Sensitivity of WtW emissions intensity to best/ worst parameter settings changed one at a time about a 2 Bcf base case.

WGR = 0.1 is a more likely scenario (see Supporting Information). Water treatment (desalination) was assumed to require 0.5 kWh/bbl.21 • Well completion emissions: the EPA’s estimated emissions ranged from 13 to 385 tCH4 per completion or workover9 (16 rather than 7 days ﬂowback time). The API value of 51.8 tCH4 used for the conventional gas base case was taken as the lower bound. • Completion/workover emissions abatement: in the worst case, methane is vented; in the best case, ﬂaring is typically 98% eﬃcient. • Wellhead pressure: wellhead pressure could fall quickly after initial production and compression would then be needed in the gathering system. • Flowback water: it was assumed that this might vary from half to twice as much as the base case. • Fractures per well: A report from the Tyndall Centre described a scenario in which wells were refractured once and outputs from these are 25% higher than unfractured wells.22 There will be a second methane release equal to the original completion emissions. This is equivalent to 1.6 times as much fracturing per unit of gas produced, or 24 fractures per well rather than 15. Conventional Gas Operations. The ultimate recovery of the conventional gas base case was also varied from 1 to 3 Bcf, for both EPA and API methane release factors for well completion. A summary of the parameters varied is given in Table 1. The results of these sensitivity cases on WtW emissions are shown as a tornado plot in Figure 2 below. Although large changes in production emissions are seen, the changes in WtW emissions are not as large because production emissions make up only a small part of the total. In Figure 2, each 10761

dx.doi.org/10.1021/es2024115 |Environ. Sci. Technol. 2011, 45, 10757–10764

Environmental Science & Technology

Figure 3. “Worst case” plot. Cumulative eﬀect of best/worst parameter settings on top of a 2 Bcf base case.

gridline represents a 1% change in WtW emissions intensity. The following factors can increase emissions by more than 1% above the base case: • Ultimate recovery: determines the number of wells drilled per unit of gas. This aﬀects both the amount of diesel used for drilling and the fugitive emissions per well. • The amount of fugitive emissions per well completion: high methane releases obviously increase the WtW emissions. • If the fugitive emissions of well workovers and completions are not ﬂared or recovered, then these can also signiﬁcantly increase WtW emissions. • The number of fracturing operations carried out per well is relatively unimportant, unless the well is refractured during its life, in which case the fugitives emissions from well workover become signiﬁcant. The following factors can increase WtW emissions by less than 1% above the base case: • Produced water: a high watergas ratio increases the energy required for water treatment and transport. • Low wellhead pressure increases the energy needed for compressing the gas between wellhead and gas treatment. • Treatment of ﬂowback water after fracturing is relatively insigniﬁcant. These operations are conducted only once or twice in the life of a well and contribute little per unit of gas produced.

’ RESULTS: WORST CASE ANALYSIS In the sensitivity cases explored above, the various parameters were varied one at a time about the base case. In reality, individual producers may be aﬀected by multiple factors, e.g., high initial methane release followed by rapid decline, leading to low ultimate recovery. The following analysis applied the best and worst assumptions cumulatively to derive best and worst cases. Although an ultimate recovery of 13 Bcf was considered suitable to describe generic shale gas, USGS data show that individual wells can have much higher or lower yields. The range was widened to 0.54.5 Bcf for this “worst case” analysis. Otherwise, the best and worst parameters from Table 1 were applied unchanged. The assumptions generate extreme results, as shown in Figure 3. In the best case, shale gas could have WtW emissions slightly lower than the base case. In the worst case, the WtW

ARTICLE

emissions intensity of shale gas could be nearly 60% higher than conventional gas. Some emissions abatement would be possible even in the most unfavorable situations. For a given geology and location, little could be done about gas quality, pressure, ﬂow rate, or water production but the following abatement options may still be possible, individually or all together. Methane releases could be reduced with better working practices and equipment. Flaring or recovery of well completion/workover emissions can be put in place, even where it is not mandatory. It may be possible to avoid treatment of produced water by reinjecting it. Figure 3 also shows the eﬀect of ﬂaring of well completion emissions at 98% eﬃciency. Even in the most unfavorable case, eﬀective abatement of methane releases could reduce the WtW emissions from shale gas to less than 15% higher than conventional gas. Note that in all cases considered, shale gas has lower WtW emissions than coal powergen, which has approximately twice the WtW emissions of conventional gas powergen.11

’ DISCUSSION Shale Gas versus Conventional Gas. The findings of the modeling work were that the WtW emissions of shale gas were approximately 1.82.4% higher than conventional gas for the base cases considered and that individual producers might have a higher WtW emissions intensity but with efficient flaring of methane releases, WtW emissions need be no more than 15% higher, even in the most unfavorable circumstances modeled. Two trends are apparent in the model results. First, that the emissions intensity is strongly aﬀected by the ultimate recovery from a well. Second, that methane releases during well completion can signiﬁcantly increase the emissions intensity of production unless abated through ﬂaring or recovery. It is the methane released rather than the diesel fuel burned which contributes most to GHG emissions from drilling, fracturing, or refracturing of unconventional gas wells. For the generic conventional gas base case, an ultimate recovery of 2.0 Bcf was assumed. However, for wells in the U.S., EUR has been in decline for many years and a value of 1.0 Bcf might be more realistic (based on EIA23 and EPA8 data, as described in Supporting Information, S3.1). By contrast, shale gas production in the U.S. has begun with large productive wells, so that it is possible that the emissions of well drilling and completion are currently lower per unit of shale gas than older, less productive conventional wells. Ultimately, production from shale gas may also decline, reducing the gap again. The uncertainty around the EPA’s latest estimate of methane emissions highlights the need for better understanding. Starting in 2011, the U.S. Greenhouse Gas (GHG) Reporting Program will collect comprehensive actual emissions data from major sources across the United States petroleum and natural gas industry. Using these data, the EPA will be able to reﬁne their emissions factors in future inventory calculations. Natural Gas versus Coal Powergen. Studies by NETL compared new-build gas24 with existing and new-build coal powergen25,26 in the U.S. NETL’s results showed that the WtW emissions of conventional gas powergen are 5358% lower than coal, when new-build gas powergen is compared against existing or new-build coal powergen. The higher production emissions of shale gas production amount to only a few percent over the WtW life cycle and do not close the gap between gas and coal by any significant amount. 10762

dx.doi.org/10.1021/es2024115 |Environ. Sci. Technol. 2011, 45, 10757–10764

Environmental Science & Technology Comparison with Other Studies. The modeling study predicted that development and completion of a shale gas well results in production emissions 1.21.5 gCO2e/MJ higher than conventional gas and that WtW emissions were 1.82.4% higher over the life cycle of gas powergen. It is interesting to compare these findings with other, recently published results. IEA recently estimated27 that 00 ...total emissions from shale gas from production through to use (well-to-burner) are only 3.5% higher in the best case (ﬂaring the gas) than the equivalent ﬁgure for conventional gas, and about 12% higher in the worst case (venting the gas).” A study of Marcellus shale gas by Carnegie Mellon University28 concluded that 00 ...development and completion of a typical Marcellus shale well results in roughly ...1.8 gCO2e/MJ of gas produced, assuming conservative estimates of the production lifetime of a typical well. This represents ... a 3% increase relative to the life cycle emissions when combustion is included.00 The Tyndall Institute22 evaluated shale gas emissions in the context of production in the United Kingdom and concluded that 00 ...the additional emissions from the shale gas extraction processes identiﬁed represent only 0.22.9% of combustion.” There is broad agreement among the studies, despite the diﬀerences in approach and assumptions. A signiﬁcant outlier is a publication by Howarth et al.29 which concluded that, “Compared to coal, the footprint of shale gas is at least 20% greater and perhaps more than twice as great on the 20-year horizon and is comparable when compared over 100 years.” The apparent contradiction can be traced to signiﬁcant diﬀerences in the underlying data and methodology. First, methane releases were overestimated: evidence for ﬂaring and recovery was disregarded in their estimate of ﬂowback emissions;17,30,31 “venting and ﬂaring” was interpreted to mean 100% vented;32 lost and unaccounted for gas (LUG) in pipelines (mostly used as fuel) was mistakenly assumed to be lost to leaks;33 and ﬁnally, lost gas was assumed to be 100% methane. Second, the diﬀerence between coal and shale gas was increased by the following: use of 20-year rather than the accepted 100-year basis for global warming potential, use of non-IPCC factors for global warming potential, and by failing to account for diﬀerences in power station eﬃciency. Applying 2007 IPCC AR4 GWP factors, 32.8% coal and 43.0% gas powergen HHV eﬃciency, and assuming 51% ﬂaring of methane, Howarth’s worst case of 7.9% methane leakage translates into shale gas emissions ∼30% lower than coal (not dissimilar to the worst case in Figure 3) and if average methane emissions are assumed to be 2.6% (in line with the EPA inventory report for 20098), then gas has half the WtW emissions of coal, in line with the consensus view of Jaramillo,6 CMU,28 and NETL12 (Supporting Information, S7). In conclusion, this modeling study shows that emissions from shale gas are not as high as some alarmist articles have claimed and that, so long as control or abatement of methane emissions are in place, shale gas WtW emissions are comparable with conventional gas and signiﬁcantly lower than coal when used for powergen.

’ ASSOCIATED CONTENT

bS

Supporting Information. Details of the parameters used in the model and additional ﬁgures. This material is available free of charge via the Internet at http://pubs.acs.org

ARTICLE

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT Shell Global Solutions is a network of independent technology companies in the Shell Group. In this publication, the expression “Shell Global Solutions” or “Shell” is sometimes used for convenience where reference is made to these companies in general, or where no useful purpose is served by identifying a particular company. ’ REFERENCES (1) Three main sources of unconventional gas; Total website; http:// www.total.com/en/our-energies/natural-gas-/exploration-and-production/ our-skills-and-expertise/unconventional-gas/presentation/speciﬁc-ﬁelds201900.html. (2) Availability, economics, and production potential of North American unconventional natural gas supplies; INGAA Foundation, Inc., 2008; http://www.ingaa.org/File.aspx?id=7878. (3) Modern Shale Gas Development in the United States: A Primer; National Energy Technology Laboratory, 2009; http://www.netl.doe. gov/technologies/oil-gas/publications/epreports/shale_gas_primer_ 2009.pdf. (4) Annual Energy Outlook 2011; U.S. Energy Information Administration (EIA), 2011; http://www.eia.doe.gov/forecasts/aeo/. (5) World Shale Gas Resources: An initial assessment of 14 regions outside the United States; U.S. Energy Information Administration (EIA), April 2011; http://www.eia.gov/analysis/studies/worldshalegas/. (6) Jaramillo, P.; Griﬃn, W. M.; Matthews, H. S. Comparative Life Cycle Air Emissions of Coal, Domestic Natural Gas, LNG, and SNG for Electricity Generation. Environ. Sci. Technol. 2007, 41 (17), 6290–6296. (7) Inventory of US Greenhouse Gas Emissions and Sinks:19902002; U.S. EPA, Oﬃce of Global Warming: Washington, DC, 2004; http:// www.epa.gov/climatechange/emissions/usgginv_archive.html. (8) Inventory of US Greenhouse Gas Emissions and Sinks: 1990 2009; U.S. EPA, Oﬃce of Global Warming: Washington, DC, 2011; http://www.epa.gov/climatechange/emissions/downloads11/US-GHGInventory-2011-Complete_Report.pdf. (9) Greenhouse Gas Emissions Reporting from the Petroleum and Natural Gas Industry: Background Technical Support Document; U.S. EPA, Oﬃce of Global Warming: Washington, DC, 2010; http://www. epa.gov/climatechange/emissions/downloads10/Subpart-W_TSD.pdf. (10) U.S. Coal Consumption by End Use Sector; U.S. Energy Information Administration (EIA); DOE/EIA-0584; 2009; http://205.254.135.24/ cneaf/coal/page/acr/table26.html. (11) Compendium of Greenhouse Gas Emissions Methodologies for the Oil and Gas Industry; American Petroleum Institute (API), 2009; http:// www.api.org/ehs/climate/new/upload/2009_GHG_COMPENDIUM.pdf. (12) Life Cycle Analysis: Power Studies Compilation Report; DOE/NETL2010/1419; National Energy Technology Laboratory, 2010; http:// www.netl.doe.gov/energy-analyses/pubs/PowerLCA_Comp_Rep.pdf. (13) Natural Gas Compressor Stations on the Interstate Pipeline Network: Developments Since 1996; Energy Information Administration, Oﬃce of Oil and Gas, 2007; http://www.eia.doe.gov/pub/oil_gas/ natural_gas/analysis_publications/ngcompressor/ngcompressor.pdf. (14) Combined (Utility, Non-Utility, and Combined Heat&Power Plant) Database in Excel Format; U.S. DOE, 2003; http://www.eia. doe.gov/cneaf/electricity/page/eia906_920.html. (15) Combined (Utility, Non-Utility, and Combined Heat&Power Plant) Database in Excel Format; U.S. DOE. 2009; http://www.eia. gov/Ftproot/pub/electricity/f923_2009.zip. (16) Methane emissions from the natural gas industry, Volume: 7 Blow and purge activities; GRI/EPA, 1996; http://www.epa.gov/gasstar/ documents/emissions_report/7_blowandpurge.pdf. 10763

dx.doi.org/10.1021/es2024115 |Environ. Sci. Technol. 2011, 45, 10757–10764

Environmental Science & Technology

ARTICLE

(17) Reducing methane emissions during completion operations; Natural Gas STAR Producer’s Technology Transfer Workshop; U.S. EPA, Oﬃce of Global Warming: Washington, DC, 2007; http://epa.gov/ gasstar/documents/workshops/glenwood-2007/04_recs.pdf (18) Assembling probabilistic performance parameters of shale-gas wells; U.S. Geological Survey Open-File Report 2010-1138; http:// pubs.usgs.gov/of/2010/1138/. (19) Mismeasuring Methane: Estimating Greenhouse Gas Emissions from Upstream Natural Gas Development; HIS CERA, 2011; www.ihs. com/images/MisMeasuringMethane082311.pdf. (20) Draft Supplemental Generic Environmental Impact Statement on the Oil, Gas and Solution Mining Regulatory Program; New York State, Department of Environmental Conservation, 2011; http://www.dec.ny. gov/docs/materials_minerals_pdf/ogsgeisapp3.pdf. (21) Produced Water from Oil and Gas Operations in the Onshore Lower 48 States; National Energy Technology Laboratory, 2004; http:// www.netl.doe.gov/KMD/cds/disk23/D-Water%20Management% 20Projects/Produced%20Water%5CNT00249%20ProducedWaterReport%20NGC103.pdf. (22) Shale gas: a provisional assessment of climate change and environmental impact; Tyndall Centre for Climate Research: Manchester, 2011; http://www.co-operative.coop/PageFiles/344773080/The-Co-operativeShale-Gas-Report-120111.pdf. (23) U.S. Natural Gas Production Summary; U.S. Energy Information Administration (EIA), 2011; http://www.eia.gov/dnav/ng/ng_sum_ lsum_dcu_nus_a.htm. (24) Life Cycle Analysis: Natural Gas Combined Cycle (NGCC) Power Plant; DOE/NETL-403-110509; National Energy Technology Laboratory, 2010; http://www.netl.doe.gov/energyanalyses/refshelf/ PubDetails.aspx?Action=View&PubId=353. (25) Life Cycle Analysis: Existing Pulverized Coal (EXPC) Power Plant; DOE/NETL-403-110809; http://www.netl.doe.gov/energy-analyses/ refshelf/PubDetails.aspx?Action=View&PubId=351. (26) Life Cycle Analysis: Supercritical Pulverized Coal (SCPC) Power Plant; DOE/NETL-403-110609; National Energy Technology Laboratory, 2010; http://www.netl.doe.gov/energy-analyses/refshelf/PubDetails.aspx?Action=View&PubId=354. (27) Are we entering a golden age of gas; World Energy Outlook 2011, Special Report; International Energy Agency; http://www.iea.org/weo/ docs/weo2011/WEO2011_GoldenAgeofGasReport.pdf. (28) Jiang, M; Griﬃn, W. M.; Hendrickson, C.; Jaramillo, P.; VanBriesen, J.; Venkatesh, A. Life cycle greenhouse gas emissions of Marcellus shale gas. Environ. Res. Lett. 2011, 6 (July-September), 034014. (29) Howarth, R. W.; Santoro, R.; Anthony Ingraﬀea, A. Methane and the greenhouse gas footprint of natural gas from shale formations. Climatic Change; DOI 10.1007/s10584-011-0061-5; http://www.eeb. cornell.edu/howarth/Howarth%20et%20al%20%202011.pdf. (30) Emission reduction strategies in the greater natural buttes; Anadarko Petroleum Corporation; EPA Gas STAR, Producers Technology Transfer Workshop, 2010; http://www.epa.gov/gasstar/documents/ workshops/vernal-2010/03_anadarko.pdf. (31) Reduced emission completions in DJ basin and natural buttes; EPA/GasSTAR Producers Technology Transfer Workshop; 2008; http://www.epa.gov/gasstar/documents/workshops/2008-tech-transfer/ rocksprings5.pdf. (32) Federal oil and gas leases: opportunities exist to capture vented and ﬂared natural gas, which would increase royalty payments and reduce greenhouse gases; U.S. General Accountability Oﬃce, Washington DC.; GAO-11-34, 2010; http://www.gao.gov/new.items/d1134.pdf. (33) Update on “lost and unaccounted for” natural gas in Texas; Basin Oil & Gas Magazine, July/August 2010; http://fwbog.com/index. php?page=article&article=248.

10764

dx.doi.org/10.1021/es2024115 |Environ. Sci. Technol. 2011, 45, 10757–10764

ARTICLE pubs.acs.org/est

Carbon Consequences and Agricultural Implications of Growing Biofuel Crops on Marginal Agricultural Lands in China Zhangcai Qin,*,† Qianlai Zhuang,†,‡ Xudong Zhu,† Ximing Cai,§ and Xiao Zhang§ †

Department of Earth & Atmospheric Sciences, Purdue University, West Lafayette, Indiana 47907, United States Department of Agronomy, Purdue University, West Lafayette, Indiana 47907, United States § Ven Te Chow Hydrosystems Laboratory, Department of Civil and Environmental Engineering, University of Illinois at UrbanaChampaign, Urbana, Illinois 61801, United States ‡

bS Supporting Information ABSTRACT: Using marginal agricultural lands to grow energy crops for biofuel feedstocks is a promising option to meet the biofuel needs in populous China without causing further food shortages or environmental problems. Here we quantify the eﬀects of growing switchgrass and Miscanthus on Chinese marginal agricultural lands on biomass production and carbon emissions with a globalscale biogeochemical model. We ﬁnd that the national net primary production (NPP) of these two biofuel crops are 622 and 1546 g C m2 yr1, respectively, whereas the NPP of food crops is about 600 g C m2 yr1 in China. The net carbon sink over the 47 Mha of marginal agricultural lands across China is 2.1 Tg C yr1 for switchgrass and 5.0 Tg C yr1 for Miscanthus. Soil organic carbon is estimated to be 10 kg C m2 in both biofuel ecosystems, which is equal to the soil carbon levels of grasslands in China. In order to reach the goal of 12.5 billion liters of bioethanol in 2020 using crop biomass as biofuel feedstocks, 7.98.0 Mha corn grain, 4.36.1 Mha switchgrass, or 1.42.0 Mha Miscanthus will be needed. Miscanthus has tremendous potential to meet future biofuel needs, and to beneﬁt CO2 mitigation in China.

1. INTRODUCTION With increasing economic, political, and environmental concerns, the fossil-fuel-supported society is seeking renewable energy sources for future sustainable development.1,2 Along with solar, wind, and hydropower energy, bioenergy draws a lot of attention both scientiﬁcally and economically. Bioenergy made available from biological sources could potentially substitute fossil fuels as a major energy source, and help to mitigate climate change by reducing CO2 emissions by sequestrating carbon into agroecosystems.1,3,4 With Brazil and U.S. as pioneers in bioethanol, and Europe in biodiesel, many regions and countries are accelerating and commercializing bioenergy and replacing fossil fuels.4,5 Many countries have set voluntary or mandatory bioenergy targets for substituting petroleum fuels with biofuels; considerable bioenergy production is expectable by the 2020s according to the short- and long-term biofuel goals.4 Because of the concerns of food security and land availability, bioenergy development is still relatively slow, and its future is not very clear in China.6 On the one hand, the large population and booming economy directly stimulate bioenergy demand. With one-ﬁfth of the world’s population, and a GDP growing at a rapid pace above 9%, China’s energy consumption has doubled over the last two decades, and ranks among the top of the world’s largest energy consumers.7 Further, the current energy structure and consumption style has caused a series of environmental problems, threatening human health and environmental sustainability.8 r 2011 American Chemical Society

In this context, bioenergy, one of the important clean and renewable energy sources, could potentially solve the puzzle of energy demand and environmental pollution. However, the perception that substantial bioenergy production requires sizable cropland and would threaten food security holds back any aggressive bioenergy plan.6,9 No more than 10% of the world’s cropland10 is available in China to feed its large population, and the arable land area is at risk of decreasing due to expansion of built-up land, natural disasters, land degradation, and the restructuring of land use patterns.11 It is well accepted that the bioenergy industry in China must not compete with food crops for land, and must not sacriﬁce foodbased grain, oil, and sugar for biofuels.13 Besides the issues of food security and land availability, food-based biofuels can also lead to ecological and environmental problems. Studies indicated that ﬁrst generation or conventional biofuels converted from food and oil crops contribute directly to monoculture and deforestation, threatening biodiversity and ecosystem services,4,1214 and potentially resulting in net greenhouse gas (GHG) emissions through indirect land use change eﬀects.12,15 Currently, China produces a relatively limited amount of biofuels compared with Brazil and the U.S., and mostly produces biofuels from food-based Received: July 19, 2011 Accepted: November 15, 2011 Revised: October 25, 2011 Published: November 15, 2011 10765

dx.doi.org/10.1021/es2024934 | Environ. Sci. Technol. 2011, 45, 10765–10772

Environmental Science & Technology feedstocks.6,16 Without reliable and plentiful biological feedstocks, it is diﬃcult for China to reach its biofuel goal of 10 million metric tons (12.5 billion liters) bioethanol and 2 million metric tons biodiesel in 2020,17 or make any further aggressive longterm targets. However, using marginal agricultural lands to grow energy crops for biofuel feedstocks could be a promising option to meet the biofuel need without causing further food and environment problems. According to Fargione et al.,4 three categories of direct source of lands are available for biofuel production: crop switching by growing biofuel crops on existing cropland, previous cropped land, and land conversion from other land uses. As mentioned above, it is not practical for China to grow biofuel crops by means of crop switching. And, converting natural ecosystems to biofuel crops will undoubtedly lead to environmental problems, such as increasing CO2 emissions and losing biodiversity,2,4,18 besides which all of the eﬀorts that China has dedicated to switching intensiﬁed agroecosystems back into natural ecosystems during the past several years will be wasted. However, bringing previously cropped land back into production could provide the desired land for growing biofuel feedstocks, mitigate GHG emissions, and possibly enhance biodiversity with proper agronomic management.4,19 Using marginal land, deﬁned primarily as abandoned cropland or used land with poor natural conditions for agriculture, for biofuel production prevent competition with food crops for land, but would be a way of capably satisfying environmental requirements for biofuel crops with relatively less resource input than food crops. Owing to their low-nutrient requirements and high water use eﬃciency, energy crops like switchgrass and Miscanthus are capable of growing on the sterile soils where food crops cannot survive.4,20,21 The hypothesis of this study is that it may be proﬁtable to develop second-generation biofuels on marginal agricultural lands in China. This can be justiﬁed from the following assessment. First of all, China possesses a large amount of marginal land covering a vast area over the nation. Previous assessments show that, in China, there are 7150 Mha marginal lands, depending on diﬀerent deﬁnitions and data sources.6,13,16,2225 Of these marginal lands, 1060% could be utilized for bioenergy production.23,25 From the perspective of land suitability for bioenergy production, Cai et al.24 estimated that there are about 130 Mha of mixed crop and vegetation lands and cropland in China, with marginal quality, which is distributed primarily across eastern China. Together with marginal grasslands and other marginal lands (discounting pasture land), China has about 152 Mha marginal lands, 14% of the global total marginal land area.24 As reported, there are at least 64 species of oilseed crops, starch-producing crops, sugar-producing crops and lignocellulosic crops that can be used as energy crops in China.8 Of these, four species could be selected as the primary energy crops: Barbados nut (Jatropha curcas L.), Jerusalem artichoke (Helianthus tuberosus L.), sweet sorghum (Sorghum bicolor L.), and Chinese silvergrass (Miscanthus sinensis Anderss).8 Considering that lignocellulosic crops provide the entire aboveground biomass, rather than just harvested fruits (e.g., Jatropha) as biofuel feedstocks, switchgrass (Panicum virgatum) and Miscanthus (e.g., Miscanthus giganteus) are preferred for growth on the marginal lands of China, where water, nutrients and even harvest machinery are limiting factors for energy crop production.6 To date, biofuel biomass production on marginal lands, and the consequences of this practice on the carbon balance due to the land use change for energy crops are not well-studied, especially

ARTICLE

in China. Most previous studies about biomass or bioenergy production used the bookkeeping approach with site-level observations and marginal land area to estimate total production.16,22,23,25 These estimates are subject to further study using mechanism models for evaluating regional biomass production.6 Energy-crop-related land use change will impact the ecosystem carbon balance in regions producing biofuel feedstocks; it is critical to account for regional environmental outcomes of producing bioenergy on marginal land.19,26 In this study, using a biogeochemical model, we estimate regional biofuel biomass production, bioenergy, and carbon consequences. Assuming growth of switchgrass or Miscanthus on marginal land in China, we present a regional, spatially explicit estimate of net primary production (NPP), net ecosystem production (NEP) and soil carbon of energy crop ecosystems, using historical climate data, marginal land distribution data and a global-scale process-based ecosystem model. The spatially resolved results are then combined with energy conversion eﬃciency and carbon mitigation information to determine the bioenergy production and potential environmental impacts of bioenergy cropping.

2. MATERIALS AND METHODS 2.1. Selected Energy Crops. Two well-tested, productive lignocellulosic crops in Europe27,28 and the U.S.,20 switchgrass and Miscanthus, are selected in this study as potential energy crops to grow on marginal lands for producing bioenergy feedstocks in China (Supporting Information (SI)). 2.2. Marginal Agricultural Land for Energy Crops. Marginal agricultural land is usually defined as land with little or no potential for agricultural productivity, and often has poor soil or other undesirable qualities with regard to agricultural use.22,24 A variety of land types are included in the definition of marginal agricultural land, such as abandoned or degraded agricultural lands, waste land, idle land23,24 and even forest areas or grassland with marginal productivity13,25 (SI Table S1). Marginal land available for energy crops, however, discounts those lands with great importance to local environment and ecology, and those with extremely poor conditions for cultivation.24 Additionally, by taking into consideration natural conditions that may limit growth of specific energy crops or agricultural management, such as water availability, and economic factors, such as labor and transportation, the actual usable marginal land is constrained to reclaimable marginal land (SI Table S1). In this study, available and reclaimable marginal land for energy crops is derived from an estimation by Cai et al.24 A total of 213 Mha of marginal land is included in the “largest-area” scenario of Cai et al.,24 which covers cropland, mixed crop and vegetation land, grassland and other lands with marginal productivity.24 However, to ensure food security and positive environmental impacts, a large proportion of cropland and most natural ecosystem lands, including grassland, are not included in this study, and are excluded by calculating as a scenario discounting environmentally sensitive land and pasture land in Cai et al.24 Finally, discounting grid cells with less than 1% marginal land cover by area, a total of 78.3 Mha of available marginal land are extracted from Cai et al.,24 for mainland China (SI Figure S1). A high reclaimable index of 60% (SI Table S1) is used in this study to account for actual reclaimable marginal land for growing energy crops, since cellulosic crops, such as switchgrass and Miscanthus in this study, normally have greater resistance to poor environmental conditions than regular crops like sweet potato, 10766

dx.doi.org/10.1021/es2024934 |Environ. Sci. Technol. 2011, 45, 10765–10772

Environmental Science & Technology rapeseed, and sweet sorghum as studied in SI Table S1.6 Therefore, a total of 47.0 Mha of recalimable marginal land is expected to grow switchgrass and Miscanthus in China (SI Table S1). The available marginal land is distributed across a vast area of east China (SI Figure S1), covering a majority proportion of the wasteland and marginal cropland estimated by Kou et al.,23 and the marginal grassland and woodland estimated by Zhuang et al.25 2.3. Model Description and Parametrization. The Terrestrial Ecosystem Model (TEM) is a process-based global-scale ecosystem model, which estimates carbon (C) and nitrogen (N) fluxes and pool sizes in terrestrial ecosystems at a monthly time step using spatial information on climate, soil, vegetation, and land use.2931 Of the fluxes, net primary production (NPP) represents the biomass of the ecosystem produced, and is usually used to calculate the harvestable biomass of crops in an agroecosystem,32 and net ecosystem production (NEP) represents the total amount of net organic carbon in an ecosystem, and is a comprehensive measure of net carbon accumulation by ecosystems.33,34 In this study, TEM is modified and parametrized to quantify the carbon dynamics of switchgrass and Miscanthus ecosystems. For each crop, TEM is calibrated against driving data, and the ratelimiting parameters for several biogeochemical processes, including gross primary production, autotrophic respiration and heterotrophic respiration, are obtained from the parametrization (SI Table S2). See Supporting Information for details on this section. 2.4. TEM Application in China. TEM and calibrated vegetation-specific parameters are used to estimate C fluxes and pool sizes of energy crop ecosystems in China. Assuming that switchgrass and Miscanthus will be grown on marginal agricultural land in China, we conduct a regional simulation for each ecosystem. The spatially referenced information on climate, elevation, soil and marginal land distribution used in TEM are organized at a 150 latitude 150 longitude resolution. Specifically, the driving climate data, including the monthly air temperature, precipitation and cloudiness, use the correspondingly averaged values from 1990 to 1999 based on CRU.35 The elevation data are derived from the Shuttle Radar Topography Mission (SRTM)36 and resampled to the same resolution as the climate data.37 For soil texture, data are based on the Food and Agriculture Organization/Civil Service Reform Committee (FAO/CSRC) digitization of FAO/UNESCO soil map of the World (1971). For switchgrass and Miscanthus, specific vegetation data describing crop distribution are used for TEM simulation, assuming that energy crops will be grown on the 78.3 Mha of available marginal land in China. The global database of finer spatial resolution in Cai et al.24 is reorganized to a resolution of 150 latitude 150 longitude. To conduct regional simulations, we ﬁrst run TEM to estimate C dynamics at a grid cell level with a monthly time step from 1990 to 1999 (see protocols in McGuire et al.30 and Melillo et al.38). TEM is run to equilibrium for each grid cell using longterm averaged monthly climatic data and annual CO2 concentrations from 1900 to 2000. The equilibrium C and N pools are then used as the initial conditions for transient simulations.37 Each grid cell in TEM is assigned a certain ecosystem type according to the vegetation data, and calculated separately for switchgrass and Miscanthus. The regional estimations of carbon ﬂuxes and pool sizes are estimated for each ecosystem. The decadal average carbon ﬂuxes and pools of the 1990s are presented. 2.4. Implications of Land Use Change and Carbon Mitigation of Growing Biofuels on Marginal Lands. By growing energy crops on marginal agricultural lands, energy crops are compared with major food crops with respect to NPP in China.

ARTICLE

Following Hicke et al.32 and Monfreda et al.,10 food crop NPP is derived from the corresponding crop economic yield according to NPPi ¼

EYi Di C ðRSi þ 1Þ HIi

ð1Þ

where i is the specific crop of rice, corn, wheat and sugar cane, EY is the economic yield based on annual reported yield from FAOSTAT, and NPP is the net primary production. HI refers to the harvest index, which measures the proportion of total aboveground biological yield allocated to the economic yield of the crop. D is the dry proportion of the EY, and C is the carbon content in the dry matter (C = 0.45). RS is the root-to-shoot ratio, which indicates the ratio of below to aboveground biomass. The value of these parameters differs among different crops (SI Table S3). The food crop NPP used for comparison is averaged NPP over during the 1990s. Bioenergy expected from energy crops could be quantiﬁed either in the form of biofuels produced21 or electricity generated6 from the biomass feedstocks. To quantify biomass feedstocks available for bioethanol use, we derive harvestable biomass from net primary production determined by TEM. As previously developed,39 by using aboveground to belowground biomass ratios of 1.4 and 2.5 for switchgrass and Miscanthus respectively,40 and assuming 90% aboveground biomass been harvested, we can roughly calculate the energy crops’ harvestable biomass. A value of 0.78 is used as the dry proportion of the yield for both corn and biofuel crops.41 For bioethanol produced from biomass, conversion eﬃciency of dry biomass into bioethanol varies among diﬀerent feedstocks. Conversion technology for corn is relatively well established and achieves a theoretical maximum yield; the current and potential yield is about 416 and 424 L Mg1, respectively, for corn grain.4,42 However conversion of biomass to ethanol is still immature, and its current yield of 282 L Mg1 can be ampliﬁed to a potential yield of 399 L Mg1.4,43 Following Clifton-Brown et al.28 and Sang & Zhu6 for electricity generated from biomass, we assume the eﬃciency of combustion and conversion into thermal energy is 35%.44

3. RESULTS 3.1. Net Primary Production. Energy crops grown on marginal agricultural lands, especially Miscanthus, have generally higher NPP than food crops in China. Nationally, switchgrass produces a mean annual NPP of 622 g C m2 yr1, which is much higher than that of corn and wheat, two major food crops in China. Its NPP falls into the range of the average NPP produced by 13 major crops over the mainland China45 and the NPP of productive rice (Table 1). For Miscanthus, the annual NPP yield more than doubles that of switchgrass, and almost equals that of productive sugar cane which is grown in limited areas of Southern China (Table 1). By making use of the 47.0 Mha of marginal land in China, a total NPP of 292 and 727 of Tg C yr1 can be produced by growing switchgrass and Miscanthus, respectively (Table 1). For regional distribution, both switchgrass and Miscanthus seem to prefer southern regions to northern regions, especially 34° southwards along the Yangtze River region, and in Southwest China, where warm temperatures and a moist climate are favorable for crops (Figure 1a, b). In comparison, Miscanthus produces 4001200 g C m2 yr1 more NPP than switchgrass, from north to south. 10767

dx.doi.org/10.1021/es2024934 |Environ. Sci. Technol. 2011, 45, 10765–10772

Environmental Science & Technology

ARTICLE

Table 1. Annual Net Primary Production of Food Crops and Biofuel Crops in the 1990s over China national average (g C m2 yr1)

crop

national total (Tg C yr1)

method and reference

switchgrassa

622 ((43)

292 ((20)

Estimated by TEM

Miscanthusa

1546 ((139)

727 ((66)

estimated by TEM

riceb

631

201

estimated from yield of FAOSTAT (2011)45

408

94

estimated from yield of FAOSTAT (2011)45

b

corn

b

378

113

estimated from yield of FAOSTAT (2011)45

b

1721

20

estimated from yield of FAOSTAT (2011)45

c

613

513

estimated by statistical data46

wheat

sugar cane

food crops a

Estimated by the Terrestrial Ecosystem Model (TEM) assuming growing biofuel crops on 47.0 Mha of marginal agricultural land over China; values are presented as “mean ((standard deviation)” (same hereafter). b Estimated from statistical data of economic food yield45 according to eq 1. c Estimated from statistical data for 13 major crops over mainland China.46

Figure 1. Annual net primary production of biofuel crops grown on marginal agricultural land in China, as determined by TEM. Values are land area weighted net primary production (g C m2 yr1) for (a) switchgrass and (b) Miscanthus.

Figure 2. Annual net ecosystem production of biofuel crops grown on marginal agricultural land in China, as determined by TEM. Values are land area weighted net ecosystem production (g C m2 yr1) for (a) switchgrass and (b) Miscanthus.

Our estimated NPP for switchgrass and Miscanthus in China are comparable with results from other ﬁeld experiment and regional estimates. An annual switchgrass yield of 212 t ha1, or 1701000 g C m2 yr1 NPP was reported for switchgrass trials in the U.S.,47,48 and 628 t ha1, or 5002400 g C m2 yr1 NPP in China.4952 Miscanthus, however, has a much higher yield at 1040 t ha1 (about 7003000 g C m2 yr1 NPP), observed in Europe and the U.S.,28,53,54 and 1544 t ha1 (10003000 g C m2 yr1 NPP) in China.49,55 The energy crop yield varies among regions and countries with diﬀerent environmental conditions like water, nutrients and climate resources. The NPP of switchgrass and Miscanthus in China are very close to that simulated in the U.S.39 It is estimated that a switchgrass NPP of 596668 g C m2 yr1 and a Miscanthus NPP of 13541588 g C m2 yr1 can be produced in the conterminous U.S., depending on the cropland type used for energy crops.39

3.2. Ecosystem Carbon Balance and Soil Carbon Sequestration. Nationally, marginal land in China acts as net carbon

sink if used to grow energy crops, either switchgrass or Miscanthus; however, Miscanthus ecosystems have relatively greater NEP than switchgrass ecosystems. The national average NEP is estimated to be 4.4 g C m2 yr1 for growing switchgrass and 10.2 g C m2 yr1 for growing Miscanthus. By growing these biofuel crops on the 47.0 Mha of marginal land, switchgrass and Miscanthus ecosystems could reach a net carbon sink of 2.1 and 4.8 Tg C yr1, respectively. The estimated heterotrophic respiration is relatively lower than NPP at a national scale, and therefore, growth of these biofuels results in a net carbon sink (SI Figure S2). Water limits the soil respiration when growing biofuel crops on marginal agricultural lands, but does not reduce the NPP of these high water-use-efficiency species.56,57 The NEP distribution, however, varies among different locations. Both switchgrass and 10768

dx.doi.org/10.1021/es2024934 |Environ. Sci. Technol. 2011, 45, 10765–10772

Environmental Science & Technology

ARTICLE

Table 2. Harvestable Biomass Production, Potential Bioethanol Production and Land Needed for Diﬀerent Bioenergy Feedstocks to Reach 12.5 Billion Liters Bioethanol Goal in 2020 net primary production

harvestable biomass

ethanol production

land needed for ethanol

harvested cropland in 2000

(g C m2)

(t ha1)a

(L ha1)b

(Mha)c

(%)d

feedstock corn grain

3.7

1558† 1588‡

8.0 7.9

5.7 5.6

corn stover

4.1

1162†

10.8

7.6

1644‡

7.6

5.4

2720†

4.6

3.3

3232‡

3.9

2.7

2046 ((141)†

6.1 ((0.4)

4.3 ((0.3)

2895 ((200)‡

4.3 ((0.3)

3.1 ((0.2)

6228 ((562)† 8812 ((794)‡

2.0 ((0.2) 1.4 ((0.1)

1.4 ((0.1) 1.0 ((0.1)

corn total switchgrass Miscanthus

408 622 ((43) 1546 ((139)

7.9 7.3 ((0.5) 22.1 ((2.0)

a

Ratio of harvestable biomass to aboveground biomass is assumed to be 1.0 for corn grain, and 0.9 for corn stover, switchgrass and Miscanthus. Calculated according to current (†) and potential (‡) biofuel conversion eﬃciency, respectively. c Cropland area needed to produce 12.5 billion liters (10.0 billion metric tons) ethanol in 2020. d Percentage of land needed to total harvested cropland in 2000.67 b

Miscanthus ecosystems act as net carbon sinks in most parts of the northeast and the west parts of China, but act as net carbon sources in the central areas, especially around the North China Plain (Figure 2a, b). This is largely driven by local temperatures, together with precipitation. Relatively high temperature and low precipitation in this area constrains biomass formation at a larger magnitude than heterotrophic respiration, and therefore determines the net ecosystem carbon balance in a negative way. Soil organic carbon stocks in the upper 1 m soil are estimated to be 4.5 and 4.7 Pg C, with average densities (carbon stock per area) of 9.6 and 10.0 kg C m2 for switchgrass and Miscanthus ecosystems on marginal agricultural land, respectively. The soil organic carbon density under energy crops is similar to that of grassland in China.58,59 In Chinese croplands, the soil organic carbon density is 3.5 kg C m2 in the top 30 cm of the soil,60,61 and that is about 6.5 kg C m2 for the top 1 m of soils according to the vertical distribution of soil carbon.62,63 Assuming that marginal agricultural lands have the same soil organic carbon density as croplands, there will be an increase of 3.1 and 3.5 kg C m2 in soils when growing switchgrass and Miscanthus, respectively. If soil reaches carbon equilibrium in 50 years,64,65 a soil carbon sequestration rate of 0.62 and 0.70 t C ha1 yr1 can be achieved by switchgrass and Miscanthus ecosystems, respectively. A total of 1.51.6 Pg C will be sequestrated in the soils of energy crop ecosystems, and that is about 2% of the total soil organic carbon stock in the whole of China,58,66 or 1/6 of the total soil organic carbon stock in Chinese cropland.60

4. DISCUSSION 4.1. Importance of Growing Biofuel Crops on Marginal Lands. From the perspective of bioethanol production, switch-

grass and Miscanthus are highly productive and may meet China’s long-term biofuel goal. Following Heaton et al.,20 we evaluate the harvestable biomass production and potential bioethanol production for different energy crops (Table 2). Besides switchgrass and Miscanthus, corn was also included as a possible energy crop since it is currently serving as a major biomass feedstock source for bioethanol production in China.6 With a higher energy conversion efficiency, corn produces slightly higher ethanol per unit area than switchgrass, both under current or potential biofuel

Figure 3. Estimates of bioenergy yield and production from marginal land. (1) Marginal land area (Mha), (2) biofuel yield (0.1 t ha1) and (3) biofuel production (10 Mt) are derived from results of this study and others.22,23,25.

conversion technology (Table 2). Miscanthus, however, has a much higher ethanol yield than corn and switchgrass. Under the current conversion efficiency, Miscanthus produces three times more ethanol per unit area than corn grain, and 1.3 times more than corn total (Table 2). With improved biofuel conversion technology, Miscanthus produces 4.5 times more ethanol per unit area than corn grain, and 1.7 times more than corn total (Table 2). To meet the 12.5 billion liters bioethanol goal in 2020, 7.98.0 Mha cropland will be needed if corn grain is used to provide bioenergy feedstocks. By doing so, 38 Mt corn grain is used for energy over food, accounting for almost 1/3 of the total corn production in 2000.45 But by growing Miscanthus instead, only 1.42.0 Mha land is needed. That is, China could reach its bioethanol goal without affecting food production by growing Miscanthus on 34% of its reclaimable marginal agricultural lands. In total, 47.0 Mha of marginal land can produce 293 billion liters bioethanol under the current conversion efficiency, and 414 billion liters under the potential conversion efficiency; this is far beyond the biofuel goal in 2020. Other crops, such as sweet potato, cassava and sweet sorghum, could also serve as energy crops and be grown on marginal land; however, these crops have much lower land use eﬃciencies compared with Miscanthus (Figure 3). Generally, these crops could produce 2.53.5 t ha1 biofuel (bioethanol or biodiesel), depending on the plant species and location.22,23,25 This is 10769

dx.doi.org/10.1021/es2024934 |Environ. Sci. Technol. 2011, 45, 10765–10772

Environmental Science & Technology

ARTICLE

Table 3. Estimates of Electricity Power Generation, Coal Displacement and Carbon Mitigation from Miscanthus to Be Grown on Marginal Land in China land for

biomass

electricity

coal

soil carbon

biomass

energy

production

generation

displacement

sequestration

total C mitigation

total CO2 mitigation

(t ha1 yr1)

crops (Mha)

(Mt yr1)

(TW h yr1) b

(Mt C yr1)b

(Mt C yr1)c

(Mt C yr1)

(Mt yr1)

Sang & Zhu6

10.0a

100

1000

1458

405

50

455

1668

6

Sang & Zhu

20.0a

100

2000

2916

810

100

910

3337

This study

22.1

47

1038

1513

420

33

453

1662

reference

Sang & Zhu et al.6 sets average biomass yield of 10 and 20 t dry biomass ha1 as short- and long-term goals of Miscanthus production. b Electricity generation and coal displacement are determined according to Clifton-Brown et al.28 and Sang & Zhu6 where dry biomass used to generate electricity prevents C being emitted from coal, and assuming that Miscanthus substitutes for coal in electricity generation. c In Sang & Zhu,6 soil carbon sequestration is assumed to be 0.5 and 1.0 t C ha1 yr1 under low- (10t ha1 yr1) and high-yield (20 t ha1 yr1 or above) scenarios, respectively. a

relatively higher than switchgrass, but less than Miscanthus. Among several estimates with similar marginal land area, Miscanthus produces the highest biofuel yield and production (Figure 3). 4.2. Carbon Emission Consequences. Biofuel itself could be clean and decrease possible CO2 emissions from what would otherwise be fossil-fuel derived sources; but life cycle assessments have reported that the use of corn and perennial grasses for ethanol would result in more carbon emissions to the atmosphere than simple combustion of gasoline, taking into consideration both direct and indirect land use change effects.2,12,15 However, the results in this study and others6,20,68 presents a much more positive perspective on the environmental benefits of growing energy crops on marginal lands for bioenergy feedstocks. By using marginal lands, the food crops and croplands are not jeopardized for bioenergy, and there are no carbon emissions due to indirect land use change effects. Moreover, energy crops like switchgrass and Miscanthus could accumulate a large amount of carbon into soils,20 about 60110 kg C m2 yr1 according to field observations.68 The biomass produced from marginal lands could be converted to bioenergy and that will reduce carbon emissions by acting as substitutes for coal or fossil fuels. Sang & Zhu6 estimated a total carbon mitigation of 455 Mt C yr1 under a low-yield Miscanthus production scenario, and 910 Mt C yr1 under a high-yield scenario, by assuming that all biomass from Miscanthus is converted to electricity (Table 3). Soil carbon sequestration is assumed to be 0.5 and 1.0 t C ha 1 yr1 under the low- and high-yield scenarios, respectively.6 Under the similar hypotheses of biomass-based electricity generation, our results show that by growing energy crops on 60% of available marginal agricultural lands in China, 1038 million tons of Miscanthus could generate 1513 TW h electricity and save 420 Mt C of coal; together with the 0.70 t C ha1 yr1 of soil carbon sequestration, it would mitigate 453 Mt of carbon, or 1662 Mt CO2 emissions from coal power (Table 3), which accounts for a half of the total CO2 emissions in China in the year 2000.69 With well-selected crop species and improved agronomic management, Miscanthus could have tremendous potential to meet future biofuel needs and beneﬁt carbon mitigation in China.6,8 However, to better understand the carbon footprint of the whole process of bioenergy production, it will be necessary to use a life cycle assessment to incorporate the energy consumption and corresponding carbon balance during biomass production and harvest, biofuel conversion, and transportation, and then to evaluate the diﬀerence in CO2 emissions between biofuels and fossil fuels.

’ ASSOCIATED CONTENT

bS

Supporting Information. Detailed information of selected energy crops, TEM model description and parametrization, model parameters and marginal land distribution are provided. This material is available free of charge via the Internet at http://pubs. acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +1 765 496 2409; fax +1 765 496 1210; e-mail: qin9@ purdue.edu.

’ ACKNOWLEDGMENT This study is supported through projects funded by the NASA Land Use and Land Cover Change program (NASANNX09AI26G), Department of Energy (DE-FG02-08ER64599), the NSF Division of Information & Intelligent Systems (NSF1028291), and the NSF Carbon and Water in the Earth Program (NSF-0630319). We thank Rosen Center for Advanced Computing (RCAC) at Purdue University for computing support, Jayne Piepenburg for proofreading and editing and three anonymous reviewers for insightful comments and suggestions. ’ REFERENCES (1) Kim, H.; Kim, S.; Dale, B. E. Biofuels, land use change, and greenhouse gas emissions: Some unexplored variables. Environ. Sci. Technol. 2009, 43 (3), 961–967. (2) Melillo, J. M.; Reilly, J. M.; Kicklighter, D. W.; Gurgel, A. C.; Cronin, T. W.; Paltsev, S.; Felzer, B. S.; Wang, X.; Sokolov, A. P.; Schlosser, C. A. Indirect emissions from biofuels: How important? Science 2009, 326 (5958), 1397–1399. (3) Beringer, T. I. M.; Lucht, W.; Schaphoﬀ, S. Bioenergy production potential of global biomass plantations under environmental and agricultural constraints. GCB Bioenergy 2011, 3 (4), 299–312. (4) Fargione, J.; Plevin, R. J.; Hill, J. D. The ecological impact of biofuels. Annu. Rev. Ecol. Evol. Syst. 2010, 41 (1), 351–377. (5) Carriquiry, M. A.; Du, X.; Timilsina, G. R. Second generation biofuels: Economics and policies. Energy Policy 2011, 39 (7), 4222–4234. (6) Sang, T.; Zhu, W. China’s bioenergy potential. GCB Bioenergy 2011, 3 (2), 79–90. (7) Crompton, P.; Wu, Y. Energy consumption in China: Past trends and future directions. Energy Econ. 2005, 27 (1), 195–208. (8) Li, X.; Hou, S.; Su, M.; Yang, M.; Shen, S.; Jiang, G.; Qi, D.; Chen, S.; Liu, G. Major energy plants and their potential for bioenergy development in China. Environ. Manage. 2010, 46 (4), 579–589. 10770

dx.doi.org/10.1021/es2024934 |Environ. Sci. Technol. 2011, 45, 10765–10772

Environmental Science & Technology (9) Ma, H.; Oxley, L.; Gibson, J.; Li, W. A survey of China’s renewable energy economy. Renewable Sustainable Energy Rev. 2010, 14 (1), 438–445. (10) Monfreda, C.; Ramankutty, N.; Foley, J. A. Farming the planet: 2. Geographic distribution of crop areas, yields, physiological types, and net primary production in the year 2000. Global Biogeochem. Cycles 2008, 22, GB1022. (11) Lal, R. Soil carbon sequestration in China through agricultural intensiﬁcation, and restoration of degraded and desertiﬁed ecosystems. Land Degrad. Dev. 2002, 13 (6), 469–478. (12) Searchinger, T.; Heimlich, R.; Houghton, R. A.; Dong, F.; Elobeid, A.; Fabiosa, J.; Tokgoz, S.; Hayes, D.; Yu, T. H. Use of US croplands for biofuels increases greenhouse gases through emissions from land-use change. Science 2008, 319 (5867), 1238–1240. (13) Yan, L.; Zhang, L.; Wang, S.; Hu, L. Potential yields of bioethanol from energy crops and their regional distribution in China. Trans. Chin. Soc. Agric. Eng. 2008, 24 (5), 213–216. (14) Pimentel, D.; Marklein, A.; Toth, M. A.; Karpoﬀ, M. N.; Paul, G. S.; McCormack, R.; Kyriazis, J.; Krueger, T. Environmental and economic costs of biofuels. Hum. Ecol. 2010, 4, 349–369. (15) Lapola, D. M.; Schaldach, R.; Alcamo, J.; Bondeau, A.; Koch, J.; Koelking, C.; Priess, J. A. Indirect land-use changes can overcome carbon savings from biofuels in Brazil. Proc. Natl. Acad. Sci. U.S.A. 2010, 107 (8), 3388–3393. (16) Tian, Y.; Zhao, L.; Meng, H.; Sun, L.; Yan, J. Estimation of unused land potential for biofuels development in (the) People’s Republic of China. Appl. Energy 2009, 86 (1), S77–S85. (17) NDRC (the National Development and Reform Commission). Meddle- and long-term development plan for renewable energy in China. http://www.sdpc.gov.cn/zcfb/zcfbtz/2007tongzhi/t20070904_ 157352.htm (accessed July 15, 2011). (18) Fargione, J.; Hill, J.; Tilman, D.; Polasky, S.; Hawthorne, P. Land clearing and the biofuel carbon debt. Science 2008, 319 (5867), 1235–1238. (19) Tilman, D.; Hill, J.; Lehman, C. Carbon-negative biofuels from low-input high-diversity grassland biomass. Science 2006, 314 (5805), 1598–1600. (20) Heaton, E. A.; Dohleman, F. G.; Long, S. P. Meeting US biofuel goals with less land: The potential of Miscanthus. Global Change Biol. 2008, 14 (9), 2000–2014. (21) Stewart, J.; Toma, Y. O.; Fernandez, F. G.; Nishiwaki, A. Y. A.; Yamada, T.; Bollero, G. The ecology and agronomy of Miscanthus sinensis, a species important to bioenergy crop development, in its native range in Japan: A review. GCB Bioenergy 2009, 1 (2), 126–153. (22) Tang, Y.; Xie, J. S.; Geng, S. Marginal land-based biomass energy production in China. J. Integr. Plant Biol. 2010, 52 (1), 112–121. (23) Kou, J. P.; Bi, Y. Y.; Zhao, L. X.; Gao, C. Y.; Tian, Y. S.; Wei, S. Y.; Wang, Y. J. Investigation and evaluation on wasteland for energy crops in China. Renewable Energy Resour 2008, 26 (6), 3–9. (24) Cai, X.; Zhang, X.; Wang, D. Land availability for biofuel production. Environ. Sci. Technol. 2011, 45 (1), 334–339. (25) Zhuang, D.; Jiang, D.; Liu, L.; Huang, Y. Assessment of bioenergy potential on marginal land in China. Renewable Sustainable Energy Rev. 2011, 15 (2), 1050–1056. (26) Hill, J.; Nelson, E.; Tilman, D.; Polasky, S.; Tiﬀany, D. Environmental, economic, and energetic costs and beneﬁts of biodiesel and ethanol biofuels. Proc. Natl. Acad. Sci. U.S.A. 2006, 103 (30), 11206–11210. (27) Kahle, P.; Beuch, S.; Boelcke, B.; Leinweber, P.; Schulten, H. R. Cropping of Miscanthus in Central Europe: Biomass production and inﬂuence on nutrients and soil organic matter. Eur. J. Agron. 2001, 15 (3), 171–184. (28) Clifton-brown, J. C.; Stampﬂ, P. F.; Jones, M. B. Miscanthus biomass production for energy in Europe and its potential contribution to decreasing fossil fuel carbon emissions. Global Change Biol. 2004, 10 (4), 509–518. (29) Raich, J. W.; Rastetter, E. B.; Melillo, J. M.; Kicklighter, D. W.; Steudler, P. A.; Peterson, B. J.; Grace, A. L.; Moore Iii, B.; Vorosmarty,

ARTICLE

C. J. Potential net primary productivity in South America: Application of a global model. Ecol. Appl. 1991, 1 (4), 399–429. (30) McGuire, A. D.; Melillo, J. M.; Joyce, L. A.; Kicklighter, D. W.; Grace, A. L.; Moore Iii, B.; Vorosmarty, C. J. Interactions between carbon and nitrogen dynamics in estimating net primary productivity for potential vegetation in North America. Global Biogeochem. Cycles 1992, 6 (2), 101–124. (31) McGuire, A. D.; Sitch, S.; Clein, J. S.; Dargaville, R.; Esser, G.; Foley, J. A.; Heimann, M.; Joos, F.; Kaplan, J.; Kicklighter, D. W.; Meier, R. A.; Melillo, J. M.; Moore Iii, B.; Prentice, I.; Ramankutty, N.; Reichenau, T.; Schloss, A. Carbon balance of the terrestrial biosphere in the twentieth century: Analyses of CO2, climate and land use eﬀects with four process-based ecosystem models. Global Biogeochem. Cycles 2001, 15 (1), 183–206. (32) Hicke, J. A.; Lobell, D. B.; Asner, G. P. Cropland area and net primary production computed from 30 years of USDA agricultural harvest data. Earth Interact. 2004, 8 (10), 1–20. (33) Randerson, J. T.; Chapin III, F. S.; Harden, J. W.; Neﬀ, J. C.; Harmon, M. E. Net ecosystem production: A comprehensive measure of net carbon accumulation by ecosystems. Ecol. Appl. 2002, 12 (4), 937–947. (34) Lovett, G. M.; Cole, J. J.; Pace, M. L. Is net ecosystem production equal to ecosystem carbon accumulation? Ecosystems 2006, 9 (1), 152–155. (35) Mitchell, T. D.; Jones, P. D. An improved method of constructing a database of monthly climate observations and associated highresolution grids. Int. J. Climatol. 2005, 25 (6), 693–712. (36) Farr, T. G.; Rosen, P. A.; Caro, E.; Crippen, R.; Duren, R.; Hensley, S.; Kobrick, M.; Paller, M.; Rodriguez, E.; Roth, L.; Seal, D.; Shaﬀer, S.; Shimada, J. The shuttle radar topography mission. Rev. Geophys. 2007, 45 (2), 1–43. (37) Lu, Y.; Zhuang, Q.; Zhou, G.; Sirin, A.; Melillo, J.; Kicklighter, D. Possible decline of the carbon sink in the Mongolian Plateau during the 21st century. Environ. Res. Lett. 2009, 4, 045023. (38) Melillo, J. M.; McGuire, A. D.; Kicklighter, D. W.; Moore, B.; Vorosmarty, C. J.; Schloss, A. L. Global climate change and terrestrial net primary production. Nature 1993, 363 (6426), 234–240. (39) Qin, Z.; Zhuang, Q.; Chen, M. Impacts of land use change due to biofuel crops on carbon balance, bioenergy production, and agricultural yield, in the conterminous United States. GCB Bioenergy 2011, DOI: 10.1111/j.1757-1707.2011.01129.x. (40) Meyer, M. H.; Paul, J.; Anderson, N. O. Competive ability of invasive Miscanthus biotypes with aggressive switchgrass. Biol. Invasions 2010, 12 (11), 3809–3816. (41) Hicke, J. A.; Lobell, D. B. Spatiotemporal patterns of cropland area and net primary production in the central United States estimated from USDA agricultural information. Geophys. Res. Lett. 2004, 31 (20), L20502. (42) DOE (U.S. Department of Energy). Breaking the biological barriers to cellulosic ethanol: A joint research agenda. Report from the December 2005 workshop, DOE/SC-0095. http://www.genomicscience. energy.gov/biofuels/ (accessed May 25, 2011). (43) Lynd, L. R.; Laser, M. S.; Bransby, D.; Dale, B. E.; Davison, B.; Hamilton, R.; Himmel, M.; Keller, M.; McMillan, J. D.; Sheehan, J. How biotech can transform biofuels. Nat. Biotechnol. 2008, 26 (2), 169–172. (44) Cannell, M. G. R. Carbon sequestration and biomass energy oﬀset: Theoretical, potential and achievable capacities globally, in Europe and the UK. Biomass Bioenergy 2003, 24 (2), 97–116. (45) FAOSTAT. FAOSTAT. http://www.faostat.fao.org/ (accessed September 20, 2011). (46) Huang, Y.; Zhang, W.; Sun, W.; Zheng, X. Net primary production of Chinese croplands from 1950 to 1999. Ecol. Appl. 2007, 17 (3), 692–701. (47) McLaughlin, S. B.; Adams Kszos, L. Development of switchgrass (Panicum virgatum) as a bioenergy feedstock in the United States. Biomass Bioenergy 2005, 28 (6), 515–535. (48) Schmer, M. R.; Mitchell, R. B.; Vogel, K. P.; Schacht, W. H.; Marx, D. B. Spatial and temporal eﬀects on switchgrass stands and yield in the Great Plains. BioEnergy Res. 2010, 3 (2), 159–171. 10771

dx.doi.org/10.1021/es2024934 |Environ. Sci. Technol. 2011, 45, 10765–10772

Environmental Science & Technology

ARTICLE

(49) Fan, X.; Hou, X.; Zuo, H.; Wu, J.; Duan, L. Biomass yield and quality of three kinds of bioenergy grasses in Beijing of China. Sci. Agric. Sin. 2010, 43 (16), 3316–3322. (50) Fan, X.; Hou, X.; Zuo, H.; Wu, J.; Duan, L. Eﬀect of marginal land types and transplanting methods on the growth of switchgrass seedlings. Pratacultural Sci. 2010, 27 (1), 97–102. (51) Yang, X.; Li, Y.; Wu, T.; Cheng, X. Biomass formation for switchgrass (Panicum virgatum) in the semiarid loess hilly-gully regions. Acta Ecol. Sin. 2008, 28 (12), 6043–6050. (52) Wu, Q.; Chang, X.; Cheng, X. Dynamic research on relationship between switchgrass (Panicum virgatum) biomass and soil water in hilly region on loess plateau. J. Yangzhou Univ. (Agricultural and Life Science Edition) 2005, 26 (4), 70–73. (53) Heaton, E.; Voigt, T.; Long, S. P. A quantitative review comparing the yields of two candidate C4 perennial biomass crops in relation to nitrogen, temperature and water. Biomass Bioenergy 2004, 27 (1), 21–30. (54) Fischer, G.; Prieler, S.; van Velthuizen, H.; Lensink, S. M.; Londo, M.; de Wit, M. Biofuel production potentials in Europe: Sustainable use of cultivated land and pastures. Part I: Land productivity potentials. Biomass Bioenergy 2010, 34 (2), 159–172. (55) Xie, X.; Zhou, F.; Zhao, Y.; Lu, X. A summary of ecological and energy-producing eﬀects of perennial energy grasses. Acta Ecol. Sin. 2008, 28 (5), 2329–2342. (56) Foereid, B.; de Neergaard, A.; H gh-Jensen, H. Turnover of organic matter in a Miscanthus ﬁeld: Eﬀect of time in Miscanthus cultivation and inorganic nitrogen supply. Soil Biol. Biochem. 2004, 36 (7), 1075–1085. (57) Yazaki, Y.; Mariko, S.; Koizumi, H. Carbon dynamics and budget in a Miscanthus sinensis grassland in Japan. Ecol. Res. 2004, 19 (5), 511–520. (58) Xie, Z.; Zhu, J.; Liu, G.; Cadisch, G.; Hasegawa, T.; Chen, C.; Sun, H.; Tang, H.; Zeng, Q. Soil organic carbon stocks in China and changes from 1980s to 2000s. Global Change Biol. 2007, 13 (9), 1989–2007. (59) Yang, Y.; Fang, J.; Ma, W.; Smith, P.; Mohammat, A.; Wang, S.; Wang, W. Soil carbon stock and its changes in northern China’s grasslands from 1980s to 2000s. Global Change Biol. 2010, 16 (11), 3036–3047. (60) Song, G.; Li, L.; Pan, G.; Zhang, Q. Topsoil organic carbon storage of China and its loss by cultivation. Biogeochemistry 2005, 74 (1), 47–62. (61) Sun, W.; Huang, Y.; Zhang, W.; Yu, Y. Carbon sequestration and its potential in agricultural soils of China. Global Biogeochem. Cycles 2010, 24, GB3001. (62) Jobbagy, E. G.; Jackson, R. B. The vertical distribution of soil organic carbon and its relation to climate and vegetation. Ecol. Appl. 2000, 10 (2), 423–436. (63) Wang, S.; Huang, M.; Shao, X.; Mickler, R. A.; Li, K.; Ji, J. Vertical distribution of soil organic carbon in China. Environ. Manage. 2004, 33, 200–209. (64) Lal, R. Soil carbon sequestration to mitigate climate change. Geoderma 2004, 123 (12), 1–22. (65) Metting, F.; Smith, J.; Amthor, J.; Izaurralde, R. Science needs and new technology for increasing soil carbon sequestration. Clim. Change 2001, 51 (1), 11–34. (66) Wang, S.; Zhou, C.; Li, K.; Zhu, S.; Huang, F. Estimation of soil organic carbon reservoir in China. J. Geog. Sci. 2001, 11 (1), 3–13. (67) Liu, J.; Liu, M.; Tian, H.; Zhuang, D.; Zhang, Z.; Zhang, W.; Tang, X.; Deng, X. Spatial and temporal patterns of China’s cropland during 19902000: An analysis based on Landsat TM data. Remote Sens. Environ. 2005, 98 (4), 442–456. (68) Clifton-Brown, J. C.; Breuer, J.; Jones, M. B. Carbon mitigation by the energy crop. Miscanthus. Global Change Biol. 2007, 13 (11), 2296–2307. (69) United Nations. Millennium development goals indicators. The oﬃcial United Nations site for the DMG Indicators. http:// mdgs.un.org/unsd/mdg/SeriesDetail.aspx?srid=749&crid (accessed July 20, 2011). 10772

dx.doi.org/10.1021/es2024934 |Environ. Sci. Technol. 2011, 45, 10765–10772

ARTICLE pubs.acs.org/est

Effects of Diesel Particle Filter Retrofits and Accelerated Fleet Turnover on Drayage Truck Emissions at the Port of Oakland Timothy R. Dallmann,† Robert A. Harley,*,† and Thomas W. Kirchstetter‡ † ‡

Department of Civil and Environmental Engineering, University of California, Berkeley, California 94720-1710, United States Atmospheric Science Department, Lawrence Berkeley National Laboratory, Berkeley, California 94720, United States ABSTRACT: Heavy-duty diesel drayage trucks have a disproportionate impact on the air quality of communities surrounding major freight-handling facilities. In an attempt to mitigate this impact, the state of California has mandated new emission control requirements for drayage trucks accessing ports and rail yards in the state beginning in 2010. This control rule prompted an accelerated diesel particle ﬁlter (DPF) retroﬁt and truck replacement program at the Port of Oakland. The impact of this program was evaluated by measuring emission factor distributions for diesel trucks operating at the Port of Oakland prior to and following the implementation of the emission control rule. Emission factors for black carbon (BC) and oxides of nitrogen (NOx) were quantiﬁed in terms of grams of pollutant emitted per kilogram of fuel burned using a carbon balance method. Concentrations of these species along with carbon dioxide were measured in the exhaust plumes of individual diesel trucks as they drove by en route to the Port. A comparison of emissions measured before and after the implementation of the truck retroﬁt/replacement rule shows a 54 ( 11% reduction in the ﬂeet-average BC emission factor, accompanied by a shift to a more highly skewed emission factor distribution. Although only particulate matter mass reductions were required in the ﬁrst year of the program, a signiﬁcant reduction in the ﬂeet-average NOx emission factor (41 ( 5%) was observed, most likely due to the replacement of older trucks with new ones.

’ INTRODUCTION Heavy-duty diesel trucks are a signiﬁcant source of ﬁne particulate matter (PM2.5), black carbon (BC), and oxides of nitrogen (NOx) emissions.1,2 Diesel NOx emissions are a precursor to secondary air pollutants including ozone, particulate nitrate, and nitric acid. Exposure to diesel PM has been associated with a variety of adverse health eﬀects.3,4 This is of particular concern to populations in close proximity to highly traﬃcked roadways,5 including communities near major freight-handling facilities such as ports and rail yards.6 Air quality impacts may be exacerbated by older trucks with higher pollutant emissions in drayage service at port and rail yards. For example, as shown in Figure 1 for the Port of Oakland as of late 2008, 17% of drayage trucks had 1993 or older model engines, and only 6% were 2004 or newer.7 Recognizing the air quality impacts of diesel truck emissions, the California Air Resources Board (CARB) implemented a drayage truck emission control regulation at ports and intermodal rail yards statewide that took eﬀect in 2010 and will become increasingly stringent over time.8 Unlike current national emission standards that require low emission levels from the new heavy-duty engines sold each year, the drayage truck regulation focuses on achieving reductions in emissions from older engines and accelerating turnover of the in-use truck ﬂeet. Key features of CARB’s regulation include (1) an outright ban on 1993 and older engine model years which are not suitable for retroﬁtting, (2) diesel particle ﬁlter (DPF) retroﬁt requirements for more recent engines, and (3) incentives r 2011 American Chemical Society

to replace older trucks with 2007+ model year trucks that meet the most stringent exhaust PM emission standards currently in force. The retroﬁt schedule imposed by the regulation requires installation of DPF systems on trucks with 19942003 engine model years by 2010 and retroﬁts of 200406 truck engines in stages from 2010 to 2013. All drayage trucks are required to meet the 2007 engine emission standard by the end of 2013. This approach is expected to reduce exhaust PM emissions from drayage trucks much more rapidly than what could be achieved by relying on natural ﬂeet-turnover alone. CARB estimates that by 2014 this program will reduce PM emissions from the state drayage truck ﬂeet 86% from 2007 baseline levels.9 Various methods have been employed to investigate the air quality impact of port-related heavy-duty diesel truck activity in California, with most research eﬀorts focused on the Ports of Los Angeles and Long Beach. Minguillon et al. applied source apportionment techniques to PM2.5 samples collected in communities surrounding these ports and found that vehicular sources were the dominant contributor to measured PM2.5 concentrations.10 High levels of port-related diesel truck activity have also been linked to elevated diesel-related pollutant concentrations measured Received: July 27, 2011 Accepted: October 31, 2011 Revised: October 27, 2011 Published: October 31, 2011 10773

dx.doi.org/10.1021/es202609q | Environ. Sci. Technol. 2011, 45, 10773–10779

Environmental Science & Technology

Figure 1. Port of Oakland drayage truck engine age distributions. The 2008 distribution is based on survey data collected by the Port of Oakland.7 An analysis of compliance inspection records for Port trucks35 was used to develop the 2010 distribution.

downwind of freeways and arterial roadways in the vicinity of the ports.6 Remote sensing of individual diesel trucks operating at the Port of Los Angeles showed a 33% reduction in the ﬂeetaverage NOx emission factor between 2008 and 2009.11 This decrease was attributed to the introduction of new trucks into the port ﬂeet in response to California’s drayage truck control rule. A decrease in average plume opacity was also reported; however, the opacity measurement is diﬃcult to relate to absolute PM mass emission rates. In contrast to programs at the Ports of Los Angeles and Long Beach, where truck replacement was the primary approach used to reduce emissions, both diesel particle ﬁlter retroﬁts and truck replacement were part of the response at the Port of Oakland in the San Francisco Bay area. The use of DPF systems, installed as original equipment on new engines, or retroﬁt on older engines, is a key element in control of diesel PM emissions. All DPF systems use a ﬁlter (also referred to as a particle trap) to physically remove particles from the exhaust stream. There are various approaches to ﬁlter regeneration, whereby trapped carbonaceous particles are oxidized to prevent excessive particle accumulation and back-pressure in the exhaust system. In actively regenerated systems, the ﬁlter is heated (e.g., by electrical heating when trucks are parked at night, or by periodic injection of unburned diesel fuel while the engine is running) to promote the oxidation of trapped particles. In contrast, passively regenerated systems utilize catalysts to promote oxidation of trapped particles at lower temperatures. In these systems the ﬁlter is continuously regenerated during normal engine operation. In a commonly used approach, an oxidation catalyst is installed upstream of the ﬁlter to convert nitric oxide (NO) present in diesel exhaust to nitrogen dioxide (NO2). NO2 is then used as the oxidizing agent for ﬁlter regeneration.12 DPF systems have been shown to reduce PM mass emissions by >90%.1316 Systems utilizing oxidation catalysts are also capable of achieving similar reductions in carbon monoxide and hydrocarbon emissions.14,15 DPF systems with high levels of catalytic loading have been shown to increase the NO2/NOx exhaust ratio,15,17,18 which can exacerbate existing urban ozone and NO2 air quality problems.19,20 This may be of particular concern in the case where older engines with higher baseline NOx emissions undergo DPF retroﬁt, creating the potential for high NO2 emissions. Other concerns include the potential for formation of nitrated polycyclic aromatic hydrocarbons21 and

ARTICLE

Figure 2. Mobile laboratory parked on Bay Street overpass. Air sampling inlet is positioned above the vertical exhaust stacks of diesel trucks driving westbound on 7th Street toward the Port of Oakland.

questions regarding their eﬀectiveness in reducing particle number emissions from diesel engines.13,22 The objectives of this study were to measure drayage truck emissions at the Port of Oakland and quantify emission changes due to the retroﬁt and renewal of the truck ﬂeet in response to California’s drayage truck rule. The results will inform those involved in California’s drayage truck rule and elsewhere where measures to clean up port-related air pollution are being considered. Our results are also meaningful in light of California’s plans to extend similar engine retroﬁt/replacement requirements to all heavy-duty trucks operating anywhere in the state.23

’ METHODS Field Sampling Site. Measurements of exhaust emissions from diesel trucks driving to the Port of Oakland were made using a mobile laboratory equipped with a suite of pollutant analyzers. The mobile lab was positioned on the Bay Street overpass above 7th Street in West Oakland, as shown in Figure 2. The roadway below is a 4-lane arterial connecting the Port of Oakland with nearby Interstate 880 and West Oakland, and is characterized by high volumes of port-related truck activity. Truck exhaust was sampled above the westbound lanes of 7th Street, where trucks heading toward the Port were observed to be cruising at steady speed or accelerating from a traffic light ∼50 m to the east. The roadway grade is level around the location where truck emissions were measured. Truck emissions were measured on selected weekdays during November 2009 and June 2010, before and after the implementation of the drayage truck rule. Pre-1994 engines were banned from the port eﬀective January 1, 2010. Retroﬁt or replacement of 19942003 engines was also required on the same schedule. However, backlogs in retroﬁtting trucks with DPFs led to deadline extensions of several months; the retroﬁt work was ∼95% complete by June 2010.24 Air Pollutant Measurements. From the mobile laboratory parked on the overpass above 7th Street, an air sampling line was extended over the edge of the bridge and down directly (∼13 m) above the vertical exhaust stacks of trucks driving below. During sampling, air was drawn continuously through 8 m of flexible 10774

dx.doi.org/10.1021/es202609q |Environ. Sci. Technol. 2011, 45, 10773–10779

Environmental Science & Technology

ARTICLE

aluminum ducting (7.6 cm diameter) to a manifold located inside the mobile laboratory. A portion of the flow through the manifold was drawn through short (<1 m) Teflon and conductive silicone sampling lines to gas- and particle-phase pollutant analyzers, respectively. Analyzers were operated with 1-s time resolution in order to measure rapidly changing pollutant concentrations when trucks passed by. Pollutant measurements included a nondispersive infrared gas analyzer for CO2 (LI-COR, Lincoln, NE; model LI-820); a chemiluminescent analyzer for NOx (ECO Physics, Ann Arbor, MI; model CLD-64); an Aethalometer for BC (Magee Scientiﬁc, Berkeley, CA; model AE-16); and an aerosol photometer (TSI, Shoreview, MN; model DustTrak II 8530) equipped with a size-selective impactor inlet for quantifying PM2.5. A condensation particle counter (TSI, Shoreview, MN; model 3007) was included in the suite of instrumentation deployed during ﬁeld sampling. However, particle concentrations in the exhaust plumes of sampled trucks often exceeded the upper limit of the instrument (105 particles cm3), preventing the quantiﬁcation of particle number emissions. CO2 and NOx concentrations were logged using a laptop computer, while BC and PM2.5 concentrations were logged internally by the respective analyzers and downloaded at the end of each sampling day. Internal clocks for all instruments were synchronized prior to the start of measurements each day. A video camera was used to record vehicle activity on 7th Street to identify times when individual trucks passed by the sampling site. The Aethalometer used in this study measures light attenuation through a ﬁlter on which particles deposit. BC mass concentration is calculated from light-attenuation measurements using a massspeciﬁc attenuation cross-section speciﬁed by the manufacturer. Previous studies have shown that this attenuation coeﬃcient is not constant with ﬁlter loading, and consequently the Aethalometer incorrectly reports lower BC mass concentrations as the ﬁlter becomes increasingly loaded.25,26 This eﬀect is especially pronounced for the highly absorbing aerosols typical of diesel exhaust. A relationship developed by Kirchstetter and Novakov to account for this eﬀect was used to adjust raw BC concentrations26 BC ¼

BC0 0:88expð ATN=100Þ þ 0:12

ð1Þ

where BC and BCo refer to corrected and raw BC concentrations, respectively, and ATN is the instrument-reported attenuation. This correction has previously been applied to Aethalometer measurements of BC in the exhaust of diesel vehicles, and timeintegrated adjusted BC concentrations were shown to be in agreement with measurements of BC determined by thermaloptical analysis of simultaneously collected PM2.5 samples.27 The DustTrak aerosol photometer used to measure PM2.5 concentrations is an optical instrument in which light scattered by particles is measured and converted to a mass concentration using an empirical calibration factor. The amount of light scattered by an aerosol is a function of particle number concentration, size distribution, and chemical composition. The factory calibration for the DustTrak is derived using a standard test dust, which consists of minerals rather than carbon particles.28 PM2.5 mass concentrations based on the mineral dust-derived calibration factor were measured during this study; this may result in systematic bias for diesel exhaust PM2.5 emissions that consist mainly of carbon particles and that often include a high BC mass fraction. Therefore, when presenting PM2.5 emission factors, the

focus is on the relative change in emissions between the November 2009 and June 2010 sampling periods rather than on absolute PM2.5 emission factors. Data Analysis. The recorded video was analyzed to determine exact times when trucks passed by the air sampling inlet. Pollutant measurements including CO2 were used to calculate BC, PM2.5, and NOx emission factors for individual trucks. A peak in measured CO2 concentration with a rise of at least 7% above baseline levels was used to indicate successful capture of an exhaust plume from a passing truck. This threshold was selected based on a sensitivity analysis of the baseline CO2 concentrations (∼500 ppm) measured at our sampling location and corresponds to three times the relative standard deviation (noise) in the baseline CO2 signal. Emission factors were not calculated for trucks with CO2 peaks below this threshold. When multiple trucks passed the sampling inlet in rapid succession, the emissions from individual trucks were not resolvable. Rather an emission signature from a combined group of trucks was measured, as described in more detail below. Corresponding peaks in CO2, BC, PM2.5, and NOx concentration time series indicated the co-occurrence of these species in an exhaust plume. BC, PM2.5, and NOx emission factors were calculated for individual trucks with valid plume measurements using a carbon balance method. In this approach, concentrations of pollutants measured in the exhaust plume are normalized to concentrations of CO2, the main carbon-containing species present in diesel exhaust. Knowledge of the weight fraction (wc = 0.87) of carbon in diesel fuel allows for the calculation of fuelspeciﬁc emission factors with units of g pollutant emitted per kg of fuel burned.27,29 Z t2 ð½Pt ½Pt1 Þdt EFp ¼ Z t2 t1 ð2Þ wc ð½CO2 t ½CO2 t1 Þdt t1

Here, EFP is the emission factor for pollutant P. The interval t1 e t e t2 represents the time period that instruments were sampling the exhaust plume of an individual truck. This time window characterizes the peak width and is typically 410 s. ([P]t [P]t1) is the baseline-subtracted concentration (μg m3) of pollutant P at time t, and similarly for [CO2] (mg C m3). In a previous application of this method to characterize BC and particle number emission factors for heavy-duty diesel trucks, t1 and t2 were determined by identifying inﬂection points to the left and right of the CO2 concentration peak, respectively.27 A similar method was used here with peak widths determined individually for all species to account for diﬀerences in the time response of individual instruments to the exhaust plume. Concentrations of additional carbon-containing compounds in exhaust plumes (e.g., carbon monoxide, volatile organic compounds) were not measured in this study, and are thus excluded from the denominator of eq 2. These species are typically present in relatively low concentrations relative to CO2 in diesel exhaust,30 therefore only a small positive bias (<5% for most trucks) in emission factor calculations is expected due to omission of these species.

’ RESULTS AND DISCUSSION Truck Activity. Truck activity was similar during the November 2009 and June 2010 sampling periods, when truck volumes averaged 250 and 230 per hour, respectively (Table 1). 10775

dx.doi.org/10.1021/es202609q |Environ. Sci. Technol. 2011, 45, 10773–10779

Environmental Science & Technology

ARTICLE

levels for successful captures during each sampling period (191 ( 31 ppm and 196 ( 23 ppm for November 2009 and June 2010 sampling, respectively). The diﬀerence in temperature and relative humidity between sampling periods (see Table 1) is not expected to signiﬁcantly impact measured emission factors. Temperature and humidity eﬀects on the formation of NOx in

Approximately 85% of observed trucks consisted of a tractor with an attached trailer or chassis trailer and container. The remaining trucks consisted of tractors without an attached trailer and tractors with an unloaded chassis trailer. Other than selection of a Portfocused sampling site, no further attempts were made as part of this study to identify and separate port trucks from other trucks that are not subject to the drayage truck emission control rules. However, a recent survey of truck activity on 7th Street indicates the majority (∼70%) of trucks passing the sampling location used in this study are drayage trucks.7 Plume Captures. Figure 3 shows three examples of baselinesubtracted pollutant concentration time series recorded when trucks drove by our sampling location. Note that clear peaks are seen in the concentrations of all measured species in Figure 3a and c, indicating that the passing trucks emitted significant quantities of NOx, BC, and PM2.5. In contrast, Figure 3b corresponds to the passage of a truck with low particulate matter emissions, as evidenced by clearly defined peaks in concentrations of CO2 and NOx but not BC and PM2.5. The relative frequency of successful truck plume captures with no measurable accompanying PM2.5 and BC emissions increased to 11 and 16%, respectively, in June 2010 from 2 and 5% in November 2009. This increase is consistent with cleaner trucks in the later sampling period. The double peaks shown in Figure 3c resulted from two trucks passing the sampling inlet in close succession. During all sampling periods, there were frequent instances of multiple trucks passing by the sampling site simultaneously (i.e., two trucks driving side-by-side in the westbound lanes) or in rapid succession. These truck-cluster events resulted from the high levels of port-related truck activity on 7th Street and the grouping of trucks at nearby traﬃc lights. In these cases, plumes for individual trucks were not resolvable and emission factors were instead calculated for each cluster. Combined emissions from 384 trucks were observed in 100 clustered events during November (representing 44% of the total truck sample). During June, emissions from 626 trucks were observed in 181 clustered events (23% of the total truck sample). As noted above, emission factors were not calculated for trucks when measured CO2 concentrations did not rise clearly above baseline levels. The percentage of unsuccessful plume captures was greater during June sampling (61% of 2687 trucks) compared to November (36% of 863 trucks). As shown in Table 1, average wind speeds were approximately two times higher in June than in November. Higher wind speeds contributed to more rapid dilution of exhaust plumes during June sampling, and this likely explains the higher percentage of unsuccessful plume captures. Although increased wind speeds during June sampling led to a lower rate of successful captures, the dilution for successful plume captures was similar for both sampling periods. This is indicated by a similar mean rise in CO2 peak concentrations above baseline

Figure 3. Baseline-subtracted concentrations of PM2.5, BC, NOx, and CO2 in the exhaust plumes of (a) a representative truck, (b) a truck with low BC and PM emissions, and (c) two trucks passing the sampling inlet in close succession.

Table 1. Meteorological and Truck Sample Size Data sampling

sampling

temperaturea

relative

wind

total

individually resolved truck

trucks with combined

trucks with no plume

date

time

(°C)

humiditya (%)

speeda (m s1)

trucks

plumes N (%)

plumes N (%)

captureb N (%)

11/19/2009

12:0015:30

13

56

2.4

863

172 (20)

384 (44)

307 (36)

6/15/2010

12:0014:45

17

60

5.2

614

131 (21)

115 (19)

368 (60)

6/16/2010

11:0015:30

21

44

5.0

953

157 (16)

151 (16)

645 (68)

6/17/2010

10:0014:30

20

50

4.2

1120

132 (12)

360 (32)

628 (56)

Average values for each sampling period measured at Oakland International Airport, located ∼13 km SE of our ﬁeld sampling site. b CO2 concentration rise <7% above baseline levels. a

10776

dx.doi.org/10.1021/es202609q |Environ. Sci. Technol. 2011, 45, 10773–10779

Environmental Science & Technology

ARTICLE

Table 2. Fleet-Average Emission Factors for Trucks Operating at the Port of Oaklanda November 2009

June 2010

% changed

individual plumes BC

1.07 ( 0.18 (169)

0.49 ( 0.08 (418)

54 ( 11

NOxb

25.9 ( 1.8 (172)

15.4 ( 0.9 (405)

41 ( 5

combined plumesc BC

1.16 ( 0.27 (100)

0.59 ( 0.10 (180)

49 ( 15

NOxb

25.7 ( 1.8 (100)

16.4 ( 1.0 (178)

36 ( 6

a

Emission factors reported in units of g pollutant emitted per kg of fuel burned. Table entries show mean (95% conﬁdence interval with total number of trucks (individual plumes) or truck cluster events (combined plumes) shown in parentheses. b NOx mass emission factors reported as NO2 equivalents. c For combined plumes, emission factors were calculated for each truck cluster. Unweighted average emission factors for all truck cluster events are reported here. The average number of trucks in each cluster was 3.8 and 3.5 for November 2009 and June 2010, respectively. d Statistical signiﬁcance of the diﬀerences in emission factors was evaluated using two-tail t tests with signiﬁcance set at p < 0.05 for NOx emission factor data and log-transformed BC emission factor data. For all reported changes in ﬂeet-average emission factors, p < 0.0001.

Figure 4. Emission factor distributions for (a) BC, (b) PM2.5, and (c) NOx.

diesel engines and its measurement using chemiluminescent techniques were found to be minor (<3%) and similar for each day of sampling.3133 BC and PM2.5 Emission Factors. Histograms showing BC emission factor distributions for successful individual truck plume captures are presented in Figure 4a. Low-emitting trucks with zero or negative emission factors calculated using eq 2 are included in the leftmost (lowest) bin of the BC emission factor distributions. BC emission factors span a wide range of values; note the use of logarithmic axes for emission factors in Figure 4a. In November 2009, the distribution of BC emission factors was log-normal, leaving aside the small fraction of trucks with emission factors <0.01 g kg1. These trucks were likely equipped with particle filters.

The BC emission factor distribution for individual trucks measured in June 2010 is shifted toward lower values relative to November 2009 (see Figure 4a). The shift in the BC emission factor distribution is characterized by a substantial decrease (from 36 to 11%) in the fraction of trucks with emission factors greater than 1 g kg1 and a corresponding 3-fold increase in the fraction of very low-emitting trucks included in the ﬁrst bin of the distribution. The changes in the distribution of BC emissions from November to June correspond to a decrease of ∼50% in average BC emission factors for trucks at the Port of Oakland (see Table 2). As shown in Table 2, the decrease in average BC emission factor computed from analysis of exhaust plumes of individual trucks is consistent with the decrease in the average emission factor calculated based on combined plume events. Histograms showing PM2.5 emission factor distributions for successful plume captures are shown in Figure 4b. Note these histograms show emission factors calculated using uncalibrated PM2.5 mass concentration data and should not be interpreted as absolute measurements of PM2.5 mass emission rates. Similar to BC, the PM2.5 emission factor distribution is shifted toward lower-emitting trucks in June 2010 relative to the November 2009 distribution. Additionally there is a 5-fold increase in the number of trucks with no measurable PM2.5 emissions (leftmost bin of distribution) between November 2009 and June 2010. This evidence, combined with measured decreases in the BC emission factor for Port trucks, suggests there was a substantial decrease in PM2.5 mass emissions from Port trucks between November 2009 and June 2010. However, due to uncertainties about instrument response to diesel exhaust emissions, the PM2.5 mass emission decrease will not be quantiﬁed here. Results from this study show signiﬁcant reductions in BC emissions from Port of Oakland trucks over a period of only seven months. By comparison, vehicle emissions measured at a nearby traﬃc tunnel (Caldecott tunnel on Highway 24) showed a similar reduction in the ﬂeet-average BC emission factor for diesel trucks of 39 ( 26% over a period of nine years between 1997 and 2006.34 Emission reductions observed in the highway 10777

dx.doi.org/10.1021/es202609q |Environ. Sci. Technol. 2011, 45, 10773–10779

Environmental Science & Technology tunnel study were driven by natural ﬂeet turnover. In contrast, emission changes reported here for trucks at the Port of Oakland are attributable to the large-scale retroﬁt of trucks with DPFs, as well as to accelerated replacement of older trucks with newer and cleaner models. Insight into the factors contributing to emission reductions observed in this study is gained by comparing truck age distributions prior to and following the implementation of the drayage truck rule, as shown in Figure 1.7,35 BC emission factor reductions resulted from (1) removal of trucks with pre-1994 model year engines from the port ﬂeet (17% of total trucks in 2008, 0% in 2010), (2) retroﬁt of trucks with model year 1994 2003 engines with DPFs (53% of truck ﬂeet in 2010), and (3) introduction of trucks with 2007 and newer model year engines already equipped with a particle ﬁlter (2% of total trucks in 2008, 14% in 2010). A large increase in the fraction of 200406 model year engines was also observed at the Port, from 4 to 33% of the ﬂeet. The PM2.5 emission standard for these engines was the same as for 19942003 engines. However, some emissions beneﬁts could still accrue even without retroﬁt of these engines in cases when they replaced pre-1994 models. The results of this study may understate the BC emission reductions due to the drayage truck rule because (1) further emission reductions were expected after June 2010; in particular the 200406 engines must be retroﬁtted or replaced by 2013 and (2) some DPF retroﬁts may have occurred prior to baseline emission measurements in November 2009. On the other hand, DPF systems were only recently installed in the 19942003 truck engines when emissions were measured in June 2010. The durability and maintenance of these emission control systems will be important to preserving BC emission reductions in subsequent years. In the absence of the drayage truck rule requiring replacement/retroﬁt of older engines, a small reduction in Port truck emissions would still have been expected due to “natural” or unforced ﬂeet turnover. Based on measured long-term trends at the Caldecott tunnel, BC emission factors from heavy-duty diesel trucks decreased at a rate of about 4% per year.34 Even assuming no economic slowdown in more recent years, unforced ﬂeet turnover could not have contributed signiﬁcantly to the large (∼50%) emission reductions observed at the Port between November 2009 and June 2010. NOx Emission Factors. In addition to changes in BC, the drayage truck rule also appears to have reduced NOx emissions from trucks operating at the Port of Oakland. As shown in Figure 4c, the NOx emission factor distribution measured in June 2010 shifted toward lower emission levels relative to November 2009. Average NOx emission factors evaluated for both individual trucks and cluster events show a decrease of approximately 40% between November and June (see Table 2). These results may appear surprising given the emphasis at the Port of Oakland on retrofitting existing trucks with DPFs. Whereas the retroﬁt of DPFs and the replacement of older trucks both contributed to BC emissions reductions observed at the Port of Oakland, the change in NOx emissions was likely driven almost entirely by the introduction of trucks with 2004 and newer engines to replace older trucks allowed at the Port. Prior work shows that DPF systems have little to no impact on total NOx emissions from diesel engines in the absence of additional NOx-speciﬁc exhaust after-treatment systems.14,15 Thus, no major changes in NOx emissions are expected from the extensive retroﬁt of model year 19942003 engines that occurred at the Port of Oakland. As shown in Figure 1, the fraction of trucks with 2004 and newer engines operating at the Port of Oakland

ARTICLE

increased from 6 to 47% between 2008 and 2010. Because allowed NOx emissions for 2004 and newer engines are set at lower levels, NOx emission reductions are expected from the accelerated replacement of older trucks. In comparison, a study of NOx emissions from trucks operating at the Port of Los Angeles reported a 33% reduction in mean NOx emission factor between 2008 and 2009.11 This change was attributed to the introduction of many brand new trucks at the Port of Los Angeles due to the drayage truck rule. Air Quality Implications. This study found substantial reductions in exhaust emissions of BC and NOx from trucks operating in the vicinity of the Port of Oakland as a result of the implementation of a retrofit and accelerated truck replacement program. The average BC emission factor for this drayage truck fleet decreased by ∼50% while the average NOx emission factor was reduced by ∼40%. Emission reductions for BC were driven by the retrofit of trucks with DPF systems and the replacement of older model year trucks with newer vehicles; reductions in NOx emissions were mainly the result of truck replacement. Reductions in the average BC emission factor, a major portion of diesel PM2.5, and a measured shift in the PM2.5 emission factor distribution together suggest that exhaust PM2.5 emissions from the Port truck fleet were also reduced. Although these emissions reductions are likely to improve air quality in communities surrounding the Port of Oakland where drayage truck activity is high, a more complete understanding of the air quality impacts of the drayage truck regulation requires measurement of emissions of species not considered here. Specifically, the possibility of increased emissions of NO2 and ultrafine particles from trucks equipped with DPF systems should be investigated as part of future work on this issue.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT We thank Steven DeMartini, Brad Edgar, Drew Gentner, Virginia Lau, Phil Martien, Brian McDonald, and Brent Rudin for their assistance. This research was sponsored by the Bay Area Air Quality Management District and the University of California multicampus research program in sustainable transportation. The statements and conclusions herein are those of the authors and do not necessarily reﬂect the views of the project sponsors. ’ REFERENCES (1) Bond, T. C.; Streets, D. G.; Yarber, K., F.; Nelson, S. M.; Woo, J.-H.; Klimont, Z. A technology-based global inventory of black and organic carbon emissions from combustion. J. Geophys. Res. 2004, 109, D14203, DOI: 10.1029/2003JD003697. (2) Dallmann, T. R.; Harley, R. A. Evaluation of mobile source emission trends in the United States. J. Geophys. Res. 2010, 115, D14305, DOI: 10.1029/2010JD013862. (3) Lloyd, A. C.; Cackette, T. A. Diesel engines: Environmental impact and control. J. Air Waste Manage. Assoc. 2001, 51, 809–847. (4) Brook, R. D.; Rajagopalan, S.; Pope, C. A.; Brook, III; Bhatnagar, J. R.; Diez-Roux, A.; Holguin, A. V.; Hong, F.; Luepker, Y.; Mittleman, R. V.; Peters, A, M. A.; Siscovick, D.; Smith, S. C.; Whitsel, L.; Kaufman, J. D. Particulate matter air pollution and cardiovascular disease. An update to the scientiﬁc statement from the American Heart Association. Circulation 2010, 121 (21), 2331–2378, DOI: 10.1161/CIR.0b013e3181dbece1. 10778

dx.doi.org/10.1021/es202609q |Environ. Sci. Technol. 2011, 45, 10773–10779

Environmental Science & Technology (5) Brugge, D.; Durant, J. L.; Rioux, C. Near-highway pollutants in motor vehicle exhaust: A review of epidemiologic evidence of cardiac and pulmonary health risks. Environ. Health 2007, 6 (23), No. 10.1186/ 1476-069X-6-23. (6) Kozawa, K. H.; Fruin, S. A.; Winer, A. M. Near-road air pollution impacts of goods movement in communities adjacent to the Ports of Los Angeles and Long Beach. Atmos. Environ. 2009, 43, 2960–2970, DOI: 10.1016/j.atmosenv.2009.02.042. (7) West Oakland Truck Survey; Bay Area Air Quality Management District: San Francisco, CA, 2009; http://www.baaqmd.gov/∼/media/ Files/Planning%20and%20Research/CARE%20PrograP/Final%20West% 20Oakland%20Truck%20Survey%20Report%20Dec%202009.ashx. (8) Drayage Truck Regulation; California Air Resources Board: Sacramento, CA, 2010; http://www.arb.ca.gov/regact/2007/drayage07/ ﬁnreg1209.pdf. (9) Regulation to Control Emissions From In-Use On-Road DieselFueled Heavy Duty Drayage Trucks: Technical Support Document; California Air Resources Board: Sacramento, CA, 2007; http://www.arb.ca.gov/ regact/2007/drayage07/tsd.pdf. (10) Minguillon, M. C.; Arhami, M.; Schauer, J. J.; Sioutas, C. Seasonal and spatial variations of sources of ﬁne and quasi-ultraﬁne particulate matter in neighborhoods near the Los Angeles-Long Beach harbor. Atmos. Environ. 2008, 42, 7317–7328, DOI: 10.1016/j.atmosenv.2008.07.036. (11) Bishop, G. A.; Schuchmann, B. G.; Stedman, D. H.; Lawson, D. R.; Li, W.; Saito, D. Trends in heavy-duty diesel truck emissions in the South Coast air basin. Presented at 20th CRC On-Road Vehicle Emissions Workshop, San Diego, CA, March 2224, 2010. (12) van Setten, B. A. A. L.; Makkee, M.; Moulijin, J. A. Science and technology of catalytic diesel particulate ﬁlters. Cat. Rev. Sci. Eng. 2001, 43 (4), 489–564, DOI: 10.1081/CR-120001810. (13) Biswas, S.; Hu, S.; Verma, V.; Herner, J. D.; Robertson, W. H.; Ayala, A.; Sioutas, C. Physical properties of particulate matter (PM) from late model heavy-duty diesel vehicles operating with advanced PM and NOx emission control technologies. Atmos. Environ. 2008, 42, 5622–5634, DOI: 10.1016/j.atmosenv.2008.03.007. (14) Liu, Z. G.; Berg, D. R.; Swor, T. A.; Schauer, J. J. Comparative analysis on the eﬀects of diesel particulate ﬁlter and selective catalytic reduction systems on a wide spectrum of chemical species emissions. Environ. Sci. Technol. 2008, 42, 6080–6085, DOI: 10.1021/es8004046. (15) Herner, J. D.; Hu, S.; Robertson, W. H.; Huai, T.; Collins, J. F.; Dwyer, H.; Ayala, A. Eﬀect of advanced aftertreatment for PM and NOx control on heavy-duty diesel truck emissions. Environ. Sci. Technol. 2009, 43, 5928–5933, DOI: 10.1021/es9008294. (16) Barone, T. L.; Storey, J. M. E.; Domingo, N. An analysis of ﬁeldaged diesel particulate ﬁlter performance: particle emissions before, during, and after regeneration. J. Air Waste Manage. Assoc. 2010, 60, 968–976, DOI: 10.3155/1047-3289.60.8.968. (17) Bishop, G. A.; Peddle, A. M.; Stedman, D. H.; Zhan, T. On-road emission measurements of reactive nitrogen compounds from three California cities. Environ. Sci. Technol. 2010, 44, 3616–3620, DOI: 10.1021/es903722p. (18) Heeb, N. V.; Schmid, P.; Kohler, M.; Gujer, E.; Zennegg, M.; Wenger, D.; Wichser, A.; Ulrich, A.; Gfeller, U.; Honegger, P.; Zeyer, K.; Emmenegger, L.; Petermann, J.-L.; Czerwinski, J.; Mosimann, T.; Kasper, M.; Mayer, A. Impact of low- and high-oxidation diesel particulate ﬁlters on genotoxic exhaust constituents. Environ. Sci. Technol. 2010, 44, 1078–1084, DOI: 10.1021/es9019222. (19) Millstein, D. E.; Harley, R. A. Eﬀects of retroﬁtting emission control systems on in-use heavy diesel vehicles. Environ. Sci. Technol. 2010, 44, 5042–5048, DOI: 10.1021/es1006669. (20) Carslaw, D. C. Evidence of an increasing NO2/NOx emissions ratio from road traﬃc emissions. Atmos. Environ. 2005, 39, 4793–4802, DOI: 10.1016/j.atmosenv.2005.06.023. (21) Heeb, N. V.; Schmid, P.; Kohler, M.; Gujer, E.; Zennegg, M.; Wenger, D.; Wichser, A.; Ulrich, A.; Gfeller, U.; Honegger, P.; Zeyer, K.; Emmenegger, L.; Petermann, J.; Czerwinski, J.; Mosimann, T.; Kasper, M.; Mayer, A. Secondary eﬀects of catalytic diesel particulate ﬁlters: conversion of PAHs versus formation of Nitro-PAHs. Environ. Sci. Technol. 2008, 42, 3773–3779, DOI: 10.1021/es7026949.

ARTICLE

(22) Herner, J. D.; Hu, S.; Robertson, W. H.; Huai, T.; Chang, M.-C. O.; Rieger, P.; Ayala, A. Eﬀect of advanced aftertreatment for PM and NOx reduction on heavy-duty diesel engine ultraﬁne particle emissions. Environ. Sci. Technol. 2011, 45, 2413–2419, DOI: 10.1021/es102792y. (23) Initial Statement of Reasons for Proposed Rulemaking: Proposed Amendments to the Truck and Bus Regulation, the Drayage Truck Regulation and the Tractor-Trailer Greenhouse Gas Regulation; California Air Resources Board: Sacramento, CA, 2010; http://www.arb.ca.gov/regact/2010/truckbus10/truckbus10isor.pdf. (24) Edgar, B. Cleaire, San Leandro, CA, Personal communication, 2010. (25) Weingartner, E.; Saathoﬀ, H.; Schnaiter, M.; Streit, N.; Bitnar, B.; Baltensperger, U. Absorption of light by soot particles: Determination of absorption coeﬃcient by means of aethalometers. J. Aerosol Sci. 2003, 34, 1445–1463. (26) Kirchstetter, T. W.; Novakov, T. Controlled generation of black carbon particles from a diﬀusion ﬂame and applications in evaluating black carbon measurement methods. Atmos. Environ. 2007, 41, 1874–1888, DOI: 10.1016/S0021-8502(03)00359-8. (27) Ban-Weiss, G. A.; Lunden, M. M.; Kirchstetter, T. W.; Harley, R. A. Measurement of black carbon and particle number emission factors from individual heavy-duty trucks. Environ. Sci. Technol. 2009, 43, 1419–1424, DOI: 10.1021/es8021039. (28) Ramachandran, G.; Adgate, J. L.; Pratt, G. C.; Sexton, K. Characterizing indoor and outdoor 15 minute average PM2.5 concentrations in urban neighborhoods. Aerosol Sci. Technol. 2003, 37, 33–45, DOI: 10.1080/02786820300889. (29) Hansen, A. D. A.; Rosen, H. Individual measurements of the emission factor of aerosol black carbon in automobile plumes. J. Air Waste Manage. Assoc. 1990, 40, 1654–1657. (30) Yanowitz, J.; McCormick, R. L.; Graboski, M. S. In-use emissions from heavy-duty diesel vehicles. Environ. Sci. Technol. 2000, 34, 729–740, DOI: 10.1021/es990903w. (31) Matthews, R. D.; Sawyer, R. F.; Schefer, R. W. Interferences in chemiluminescent measurement of NO and NO2 emissions from combustion systems. Environ. Sci. Technol. 1977, 11 (12), 1092–1096. (32) Code of Federal Regulations, Title 40, Part 86. (33) Humidity and Temperature Eﬀects on On-road and Oﬀ-road Emissions and Ozone Formation; ENVIRON International Corporation: Novato, CA, 2004; http://ﬁles.harc.edu/Projects/AirQuality/Projects/ H008B.2003/TH/H8BTH FinalReport.pdf. (34) Ban-Weiss, G. A.; McLaughlin, J. P.; Harley, R. A.; Lunden, M. M.; Kirchstetter, T. W.; Kean, A. J.; Strawa, A. W.; Stevenson, E. D.; Kendall, G. R. Long-term changes in emissions of nitrogen oxides and particulate matter from on-road gasoline and diesel vehicles. Atmos. Environ. 2008, 42, 220–232, DOI: 10.1016/j.atmosenv.2007.09.049. (35) Rudin, B. Compliance and Enforcement Division, Bay Area Air Quality Management District, San Francisco, CA, Personal communication, 2011.

10779

dx.doi.org/10.1021/es202609q |Environ. Sci. Technol. 2011, 45, 10773–10779

ADDITION/CORRECTION pubs.acs.org/est

Correction to Influence of Arsenate Adsorption to Ferrihydrite, Goethite, and Boehmite on the Kinetics of Arsenate Reduction by Shewanella putrefaciens Strain CN-32 Jen-How Huang,* Andreas Voegelin, Silvina A. Pombo, Anna Lazzaro, Josef Zeyer, and Ruben Kretzschmar Environmental Science & Technology DOI: 10.1021/es201503g 2011, 45, 7701–7709 Unfortunately, an error in eq 1 was recently brought to our attention. In our article, the two-site Langmuir equation (eq 1) was given as Seq ¼

KL1 bmax 1 Ceq KL2 bmax 2 Ceq þ 1 þ Ceq 1 þ Ceq

ð1Þ

However, the correct form of the two-site Langmuir equation is Seq ¼

KL1 bmax 1 Ceq KL2 bmax 2 Ceq þ 1 þ KL1 Ceq 1 þ KL2 Ceq

ð2Þ

with model parameters as deﬁned in the article. Equation 2 can be rewritten in the form: Seq ¼

bmax 1 Ceq bmax 2 Ceq þ 1=KL1 þ Ceq 1=KL2 þ Ceq

ð3Þ

Seq ¼

bmax 1 Ceq bmax 2 Ceq þ Kd1 þ Ceq Kd2 þ Ceq

ð4Þ

or

with Kd1 = 1/KL1 and Kd2 = 1/KL2, respectively. The latter form of the two-site Langmuir eq 4 was used in all calculations and the ﬁtted values for Kd1 and Kd2 were reported in Table S1 (Supporting Information). To be consistent with the article, we have now replaced the Kd1 and Kd2 values in Supporting Information Table S1 with the corresponding KL1 and KL2 values used in eq 2. The error reported here had no inﬂuence on the results and conclusions of our article, since a correct form of the two-site Langmuir equation was used in all ﬁts and model calculations.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional materials as noted in the text including a corrected Table S1. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT We acknowledge Dr. Iso Christl, who discovered the error in the article.

Published: November 22, 2011 10780

dx.doi.org/10.1021/es203966x | Environ. Sci. Technol. 2011, 45, 10780–10780

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents