No title

September 2011 Volume 15, Number 9 pp. 375–446 Special Issue: The Genetics of Cognition Editor Stavroula Kousta Executi...

Author: Cell Press

81 downloads 469 Views 2MB 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

September 2011 Volume 15, Number 9 pp. 375–446 Special Issue: The Genetics of Cognition Editor Stavroula Kousta Executive Editor, Neuroscience Katja Brose Journal Manager Rolf van der Sanden Journal Administrators Patrick Scheffmann and Ria Otten Advisory Editorial Board R. Adolphs, Caltech, CA, USA R. Baillargeon, U. Illinois, IL, USA N. Chater, University of Warwick, UK P. Dayan, University College London, UK S. Dehaene, INSERM, France D. Dennett, Tufts U., MA, USA J. Driver, University College, London, UK Y. Dudai, Weizmann Institute, Israel A.K. Engel, Hamburg University, Germany M. Farah, U. Pennsylvania, PA, USA S. Fiske, Princeton U., NJ, USA A.D. Friederici, MPI, Leipzig, Germany O. Hikosaka, NIH, MD, USA R. Jackendoff, Tufts U., MA, USA P. Johnson-Laird, Princeton U., NJ, USA N. Kanwisher, MIT, MA, USA C. Koch, Caltech, CA, USA M. Kutas, UCSD, CA, USA N.K. Logothetis, MPI, Tübingen, Germany J.L. McClelland, Stanford U., CA, USA E.K. Miller, MIT, MA, USA E. Phelps, New York U., NY, USA R. Poldrack, U. Texas Austin, TX, USA M.E. Raichle, Washington U., MO, USA T.W. Robbins, U. Cambridge, UK A. Wagner, Stanford U., CA, USA V. Walsh, University College, London, UK Editorial Enquiries

Editorial

375 Uncovering the genetic underpinnings of cognition

Trevor W. Robbins and Stavroula Kousta

Update FORUM: Science & Society

378 Genetics and criminal responsibility

Stephen J. Morse

Review

381 Genetics of human episodic memory: dealing with complexity

Andreas Papassotiropoulos and Dominique J.-F. de Quervain

388 The genetics of cognitive ability and cognitive ageing in healthy older people

Sarah E. Harris and Ian J. Deary

395 Dissecting the genetic architecture of human personality

Marcus R. Munafò and Jonathan Flint

401 Genetics of emotion

Laura Bevilacqua and David Goldman

409 Genetics of autism spectrum disorders

Daniel H. Geschwind

417 Understanding risk for psychopathology through imaging gene–environment interactions

Luke W. Hyde, Ryan Bogdan and Ahmad R. Hariri

428 The genetics of cognitive impairment in schizophrenia: a phenomic perspective

Robert M. Bilder, Andrew Howe, Nic Novak, Fred W. Sabb and D. Stott Parker

436 The contribution of imaging genetics to the development of predictive markers for addictions

Eva Loth, Fabiana Carvalho and Gunter Schumann

Trends in Cognitive Sciences

Cell Press 600 Technology Square Cambridge, MA 02139, USA Tel: +1 617 397 2817 Fax: +1 617 397 2810 E-mail: [email protected]

Cover: Twin, family and adoption studies have demonstrated that there is a substantial heritable component to all cognitive functions. The articles in this special issue summarize what is currently known about the genetic underpinnings of these functions and their disorders. At the same time, they highlight just how much there is yet to be discovered in this rapidly advancing field. Cover image: Adrian Neal/Lifesize/Getty Images.

Editorial

Special Issue: The Genetics of Cognition

Uncovering the genetic underpinnings of cognition Trevor W. Robbins1 and Stavroula Kousta2 1 2

Department of Experimental Psychology, University of Cambridge, Downing Street, CB2 3EB Cambridge, UK Editor, Trends in Cognitive Sciences, Cell Press, 600 Technology Square, Cambridge MA, 02142, USA

No one today can ignore the genetic approach to cognition and behavior, given the huge achievements of the Human Genome Project (http://www.ornl.gov/sci/techresources/ Human_Genome/home.shtml). The initial impact of studies of individual differences in genetic polymorphisms, such as the catechol-O-methyl transferase (COMT) gene, and their relationship to such core cognitive concepts as working memory (in this case via its modulation by prefrontal dopamine [1]) has been immense, catching the imagination of many cognitive neuroscientists. Here is a way, for example, to evaluate the effect of individual differences in neurotransmitter function without the need to administer drugs. Alternatively, possible heterogeneity in behavioral performance or patterns of neural network activation revealed through functional neuroimaging, may be resolved by taking into account genetic factors. Taken together, the contributions in this special issue address the genetic underpinnings of key aspects of cognition, such as memory, intelligence, reward processing, as well as emotion and personality, in the context of both healthy populations and in key disorders, such as schizophrenia, autism, and addiction. Collectively, the articles discuss evidence from a variety of perspectives and different approaches, including twin, linkage, candidate gene, genome-wide association, imaging genetics, gene environenvironment interaction, and gene expression studies, providing insight into the strengths and challenges for each approach. A key insight arising from several articles in this issue is that variations in complex psychological attributes such as intelligence and psychosis are usually likely to be determined by multiple genes, each exerting a small effect. Moreover, it is becoming increasingly clear that, so as to establish valid behavioral–genetic correlations, it is necessary to test large populations, in order to avoid spurious false positive effects arising from the very large number of contributing genes. Finally, the well-known mantra of the importance of ‘gene environment interaction’ has gained additional complexity from the discovery of effects of imprinted genes and epigenetic factors. As several articles in this special issue make clear, some of the inspiration for the genetic (and epigenetic) approach to cognition comes from our need to understand better the origin of disorders in mental health. The hope is that by understanding their genetic basis, we will be able to identify the causal molecular pathology of different disorders Corresponding authors: Robbins, T.W. ([email protected]); Kousta, S. ([email protected]).

and capitalize on proteomic approaches for treatments. By paying attention to possible intermediate phenotypes (‘endophenotypes’), which may have a biochemical, neural or even cognitive-behavioral basis, we may be able to bridge the present daunting gap between genes and psychopathology. Rapid advances in the field of genetics have also raised several ethical and legal questions. In the first article in this issue, Morse discusses one such question: what are the implications of genetic research for the concept of criminal responsibility? Morse explains that the concept of criminal responsibility as determined by law is based on identifying relevant mental states: was there an intention to commit a crime? And did the individual possess knowledge of wrongdoing? Genetic research, on the other hand, is concerned with mechanistic causation, and causation does not currently constitute an excusing condition in law. Unless it can be shown that biological causation is directly responsible for a particular mental state, it cannot be used either as a mitigating factor or an excuse for a criminal act. But one could argue, and some have, that an individual’s genetic make-up completely determines behaviour, hence suggesting that the concept of criminal responsibility as defined by law should be completely revised: it is biology, not mental states that matter. However, Morse argues, genetic research (or any other research) has provided no evidence to support the idea that mental states are redundant and do not have a causal role in driving behavior. And denying a critical role for mental states would do away with human motivation to do anything whatsoever, not just a criminal act. The four review articles that follow address different aspects of cognitive and emotional processes from a genetic perspective. Papassotiropoulos and De Quervain overview the literature on the genetic underpinnings of human episodic memory, discussing the findings of two key approaches: candidate gene studies based on well-established molecular pathways involved in synaptic plasticity and hypothesis-free genome-wide association studies. Although both types of approaches have yielded promising results, much remains to be discovered. Key in achieving progress, the authors argue, is understanding the phenotypic complexity of episodic memory: episodic memory is not one phenotype, and similarity of a phenotype between studies might not be sufficient for replication purposes. Moreover, most studies (both candidate-gene and genomewide association studies) still perform simple statistics, analyzing genetic variants independently of each other. This approach covers a minor part of the complexity of episodic memory, and new methods are needed to capture

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.006 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

375

Editorial the genetic variability of human memory, for instance, through additive or epistatic interactions. Cognitive ageing affects people’s quality of life and predicts dementia and death. Harris and Deary focus on the genetics of cognitive function in healthy aging, and summarize findings across several different approaches in behavioral and molecular genetics. They discuss a number of findings, best established among which is evidence that the e4 haplotype of the apoliprotein E (APOE) gene is positively associated with cognitive ageing trajectories and with cognitive function in elderly individuals. Several other loci might be implicated in cognitive function and decline with age, however, and the authors argue that welldesigned longitudinal studies, with long follow up times, are required to determine further genetic influences on cognitive ageing in healthy older individuals. Munafo` and Flint critically review the personality genetics literature, as a case study of the missing heritability problem: as with any other complex trait, the genetic architecture of personality traits, the authors argue, is likely to be the result of several hundreds, if not thousands, of small effect loci, which together nonetheless produce substantial heritability. Early candidate gene studies found positive associations of the serotonin transporter gene (SLC6A4) with measures of neuroticism, and the dopamine D4 receptor gene (DRD4) with novelty seeking. Although follow up candidate gene studies, as well as gene by environment interaction studies, abound, meta-analytic studies suggest that these genes, as well as other candidate genes identified, have a very small effect at best. In fact, a number of recent genome-wide association studies have failed to identify any locus with clear genome-wide significance. Munafo` and Flint argue that, in order to achieve progress in this field, it is important to consider issues of power, statistical stringency and independent replication; and to acknowledge that, as with other complex traits, such as blood pressure and height, variation in personality is the outcome of the combined effect of several small effect loci. In a related contribution, Bevilacqua and Goldman discuss the genetic basis of emotional processing, personality and temperament, focusing on functional variants at five genes: COMT, SLC6A4, neuropeptide Y (NPY), a glucocorticoid receptor-regulating cochaperone of stress proteins (FKBP5) and pituitary adenylate cyclase-activating polypeptide (PACAP), as illustrative of the effects of genes on emotion. They address factors that alter or confound the effects of these genes, especially gene environenvironment interactions, as emotionality is strongly influenced by exposure to stress. They acknowledge the issues highlighted in Munafo` and Flint’s contribution—the few genome-wide association studies that have examined the effects on emotion of genes identified in candidate gene studies have indeed failed to yield any significant results for these genes. And they converge with Munafo` and Flint on likely explanations (statistical power and the fact that emotionality is likely to be underpinned by several different loci of small effect). However, despite lack of genomewide significance in the few studies carried out up to now, the authors argue, these genes have large effects on metabolic responses of the brain to emotional stimuli assessed 376

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

in real time by brain imaging, hence validating the effects of these genes in emotion. The final four review articles in this issue address the genetic basis of disorders in which cognitive and emotional dysfunctions are a key aspect. Amongst them, autism genetics may be the area where most progress has been achieved, with known genetic causes for 10–20% of the cases. Geschwind discusses the current state of the art in autism genetics, highlighting the fact that autism spectrum disorders (ASD) are both highly heritable but also etiologically heterogeneous. Moreover, he points out that the autism phenotype is broad and overlaps with many other neurodevelopmental disorders. Geschwind reviews research that has identified common and rare gene variants and explains how these findings verify or refute prevailing genetic models of autism. He explains that new approaches, such as gene expression and epigenetics, have provided novel insight into the genetic basis of ASD. This more recent research has suggested that, despite their heterogeneity, the multiple genetic causes of ASD converge on a few biological pathways affected in most individuals with autism. In the context of psychopathology, Hyde, Bogdan and Hariri discuss recent advances in gene environment interactions and imaging genetics, two approaches that up to the present have been largely independent. They highlight the key strengths as well as challenges for both strands of research and argue that bridging these two approaches will provide crucial insight into risk factors for psychopathology. They lay out the foundations of an integrated approach, imaging gene environment interactions, by examining statistical methods for combining the two approaches and by discussing plausible biological mechanisms, such as epigenetics, that can explain the interplay between genes, experience, and the brain. Imaging gene environment interactions, the authors argue, holds significant promise to elucidate biological mechanisms that underlie the etiology and pathophysiology of psychopathology and for the development of personalized treatment and refined diagnostic criteria. It is well-established that, in addition to psychotic symptoms, cognitive impairment in schizophrenia is severe. Bilder and colleagues’ contribution addresses our current understanding of the genetic basis of cognitive dysfunction in schizophrenia. They conclude that, despite intense research, little progress has been made up to the present in detecting genetic associations for cognitive impairments in schizophrenia. The authors advocate a specific approach, phenomics, as offering significant promise in this field, by emphasizing the simultaneous study of multiple phenotypes across biological scales. This approach might be particularly fruitful if it is the case that the high heritability of schizophrenia and cognitive impairments is due to large numbers of variants with small effects. Bilder et al. present a new collaborative database, CogGene, to share data on genetic associations with cognitive phenotypes. They show how examining results from this database in parallel with results from a similar database on genetic associations in schizophrenia (SZGene) can provide important insight into the genetic underpinnings of cognitive impairment in schizophrenia.

Editorial Abnormalities in reinforcement behaviour (sensitivity to rewards and punishment, inhibitory control and stressrelated emotional vulnerability), particularly during adolescence, have been linked to addiction vulnerability. Loth, Carvalho and Schumann examine the genetic basis of inter-individual differences in reinforcement and reward-related processes by focusing on imaging genetic studies. These studies have addressed associations between one or a few candidate polymorphisms and taskrelated brain activation differences in a few regions of interest, and have revealed effect sizes that manifold exceed the variance typically explained with behavioural traits or cognitive measures. Despite their usefulness, however, results across studies are not always consistent, and the authors advocate strengthening the imaging genetics approach by examining large sample sizes and using

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

longitudinal designs (with repeated testing of the same individuals at different ages). Twin, family and adoption studies have demonstrated that there is a substantial heritable component to all cognitive functions [2]. The articles in this special issue summarize what is currently known about the genetic underpinnings of these functions and their disorders. At the same time, they highlight just how much there is yet to be discovered in this rapidly advancing field. References 1 Egan, M.F. et al. (2001) Effect of COMT Val108/158 Met genotype on frontal lobe function and risk for schizophrenia. Proc. Natl. Acad. Sci. U.S.A. 98, 6917–6922 2 Bouchard, T.J. and McGue, M. (2003) Genetic and environmental influences on human psychological differences. J. Neurobiol. 54, 4–45

377

Update FORUM: Science & Society

Special Issue: The Genetics of Cognition

Genetics and criminal responsibility Stephen J. Morse University of Pennsylvania Law School, Psychiatry Department and Center for Neuroscience and Society, 3400 Chestnut Street, Philadelphia, PA 19104-6204, USA

Some believe that genetics threatens privacy and autonomy and will eviscerate the concept of human nature. Despite the astonishing research advances, however, none of these dire predictions and no radical transformation of the law have occurred. Advocates have tried to use genetic evidence to affect judgments of criminal responsibility. At present, however genetic research can provide little aid to assessments of criminal responsibility and it does not suggest a radical critique of responsibility.

Internal and external interdisciplinary critiques Another discipline could inﬂuence the law by providing either an internal or an external critique. In the former, the general validity of a legal doctrine, practice or institution is accepted but the other discipline tries either to explain the legal phenomenon or to reform it. For example, ﬁndings from other disciplines might suggest that the doctrines of criminal responsibility should be altered in various ways but might not suggest that the concept of responsibility is incoherent. By contrast, an external critique suggests that the doctrine, practice or institution is invalid. Many people believe that discoveries in genetics and other sciences strongly suggest the truth of determinism (or something like it), and that if determinism is true, then no one can be genuinely responsible. For example, The Economist recently warned, ‘Genetics may yet threaten privacy, kill autonomy, make society homogeneous and gut the concept of human nature. But neuroscience could do all of these things ﬁrst’ [1]. Blaming and punishing criminals is thus allegedly unfair because no one deserves such treatment [2]. The conclusion is that current conceptions of criminal justice should be abandoned because they rest on a morally mistaken foundation. External critiques are radical, whereas internal critiques produce incremental change or suggest that the current system is valid. The law’s psychology and general concept of responsibility To understand how genetics research can inﬂuence criminal law requires understanding of the law’s implicit psychology and concept of responsibility. Law is a system of rules and standards that is meant to guide human action by providing agents with reasons to act one way or another [3]. Criminal law, and indeed all law, therefore presupposes the ‘folk psychological’ view, which causally explains behavior in Corresponding author: Morse, S.J. ([email protected]).

378

part by mental states such as desires, beliefs, intentions, volitions, and plans [4]. Other psychological, as well as biological and sociological variables also play a role but folk psychology considers mental states fundamental to a complete explanation of human action. The law’s concept of responsibility follows from the nature of law and the type of creature it addresses. Responsible agents are those who can be adequately guided by the law, which means, roughly, that only conscious, intentional and rational creatures with developed linguistic capacity can be responsible. This explains why young children and some people with mental abnormalities are not considered responsible [5]. Criminal law responsibility criteria Now let us be a bit more speciﬁc about the criteria for criminal liability, which are normative and not scientiﬁc facts. These criteria justify state blame and punishment because offenders who meet them deserve such treatment, and desert is at least a necessary condition for just punishment in the USA. The deﬁnitions of most criminal prohibitions include an intentional action done in a reasonably integrated state of consciousness that is accompanied by another mental state (mens rea) that indicates how culpable the action is. Note that these are both folk psychological criteria. For example, a common deﬁnition of murder is intentional killing conduct done with the purpose to kill. A neuromuscular spasm that causes the death of another is not an action, and if one is driving intentionally but completely carefully, then an entirely accidental killing of a pedestrian would not be done with the purpose to cause death. In both cases, the defendant is not culpable. Even if the agent does the prohibited act with the culpable mental state, the person will not be liable if he or she has an ‘afﬁrmative defense’ because either the agent’s act was right or permissible under the circumstances (a justiﬁcation) or the person was not a responsible agent at the time of the crime (an excuse). Intentionally killing a wrongful aggressor because the agent reasonably believes he must do so to save his own life, self-defense, is an example of the former. If the agent kills because he has the delusional belief that he is in deadly danger, then he has done wrong, but he might be excused by the defense of legal insanity because he was not a rational agent. There are two generic excusing conditions: lack of rational capacity and compulsion. The latter can be caused externally, such as cases involving dreadful ‘do-it-or-else’ threats (e.g. ‘kill or I will kill you’), or internally, such as

Update cases of strong internal desires (e.g. an addict’s desire to seek and use substances). In both cases, the agent acted intentionally but we might think it is unfair to ask him to control himself because it will be so difﬁcult to do so. Note that afﬁrmative defenses also involve folk psychology because they are based on mental states, including desires and beliefs. Translating genetic research for assessing criminal responsibility Genetics concerns mechanistic causation. Genes do not have mental states and do not commit crimes; people do. To make a useful internal contribution to criminal responsibility, the genetic data must be ‘translated’ into the law’s folk psychological responsibility criteria [6]. It must be shown how, precisely, the genetic data are relevant to whether a defendant acted, whether he or she possessed a particular mens rea, and whether the mental states relevant to defenses were present. It is not sufﬁcient to indicate that genetics played a causal role in explaining the criminal behavior, even if that causal role is very powerful. Causation and predictability are not excusing conditions in law and causation is not the equivalent of legal compulsion (most action is not the causal result of dire threats or uncontrollable desires) [7]. If they were, no one would be responsible because we inhabit a causal universe, but we nonetheless hold people responsible. A genetic predisposition to criminal conduct does not per se mitigate or excuse. Causation is relevant only if it tends to show the presence of a genuine excusing condition, but it is the latter that does the legal work. Believing that causation per se mitigates or excuses responsibility is the most pernicious confusion bedeviling the attempt to relate scientiﬁc ﬁndings to criminal responsibility. I have termed it the ‘fundamental psycholegal error’ [8]. In the few legal cases in which genetic information has been used to mitigate responsibility, this error has been common [9,10]. We are reasonably conﬁdent that having a genetically induced MAO-A deﬁciency in interaction with childhood abuse causally increases the risk of criminal and antisocial behavior more than ninefold [11]. Nonetheless, there is no reason to believe that offenders exposed to that interaction did not act or form the required mental states. If exposure to that interaction somehow diminished their rationality or produced some type of uncontrollable internal desire, then mitigation or excuse might be warranted. Such a diminished rationality or control problem would have to be demonstrated independently by evidence other than causation data. Internal contributions unrelated to responsibility Genetic research might contribute internally to criminal law in ways unrelated to responsibility. For example, knowledge about genetic variables that predispose people to crime could enhance the accuracy of dangerousness predictions that affect sentencing and parole and it might enhance the efﬁcacy of interventions to reduce the risk of crime. Note that if it were safe to release an offender early as a result of a successful, genetically based intervention, the reason would be public safety and cost and not because the offender deserved less punishment.

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

External challenges to the concept of responsibility Genetic research might also provide a radical external critique of criminal responsibility if it convincingly demonstrated that no one deserves punishment. It might do this either by lending support to a deterministic metaphysics, or by proving that our mental states play no role in explaining our behavior. The former is simply the familiar determinism and free will issue. Note that contra-causal libertarian free will is not an internal criterion in law and is not even foundational for criminal responsibility. For example, even if determinism is true, some defendants are delusional and most are not. Some offenders act with a gun at their heads but most do not. These are differences that make a moral difference according to theories of fairness we endorse. At the metaphysical level, there is a respectable view, ‘compatibilism,’ which holds that robust responsibility is possible in a deterministic world even if we lack libertarian free will [12]. Determinist ‘incompatibilists’ disagree, of course, but the dispute is not resolvable and the law cannot wait for the metaphysicians. Thus, the external critique based on determinism does not have much legal purchase, although it has proponents. Compatibilism cannot save responsibility from the second external critique because compatibilism presupposes the folk-psychological view of agency that the second critique denies. Some believe that mental states play no causal role – paradoxically, this is a motivating mental state for them – but neither genetics nor any other science at present remotely proves that our mental states are causally inert [6]. As the eminent philosopher of mind, Jerry Fodor, has written, if we are wrong about the importance of mental states, that will be about the ‘wrongest’ we have ever been about anything [13]. Finally, if our mental states, including our reasons for action, are simply an epiphenomenal sideshow the brain somehow constructs, what good reason would we have to do anything? The task of genetics and other sciences should be to explain our intentionality rather than reductively to explain it away. Acknowledgment As always, the author thanks his personal attorney, Jean Avnet Morse, for her sound sober counsel and moral support.

References 1 The Economist 23 May (2002) The ethics of brain science: open your mind, p. 77 2 Greene, J. and Cohen, J. (2006) For the law neuroscience changes nothing and everything. In Law and the Brain (Zeki, S. and Goodenough, O., eds), pp. 207–226, Oxford 3 Shapiro, S. (2000) Law, morality, and the guidance of conduct. Legal Theory 6, 127–170 4 Sifferd, K.L. (2006) In defense of the use of commonsense psychology in the criminal law. Law Philos. 25, 571–612 5 Morse, S.J. (2011) Gene-environment interactions, criminal responsibility and sentencing. In Gene-Environment Interactions in Developmental Psychopathology (Dodge, K.A. and Rutter, M., eds), pp. 207–234, Guilford 6 Morse, S.J. (2011) Lost in translation?: an essay on law and neuroscience. In Law and Neuroscience (Freeman, M., ed.), pp. 529– 562, Oxford 7 Morse, S.J. (2007) The non-problem of free will in forensic psychiatry and psychology. Behav. Sci. Law 25, 203–220 8 Morse, S.J. (1994) Culpability and control. Univ. Pa. Law Rev. 142, 1587–1660 379

Update 9 Rigoni, D. et al. (2010) How neuroscience and behavioral genetics improve. psychiatric assessment: report on a violent murder case. Front. Behav. Neurosci. 4, 1–10 10 Feresin, E. (2009) Lighter sentence for murderer with ‘bad genes’. Nature DOI: 10.1038/news.2009.1050 11 Caspi, A. et al. (2002) Role of genotype in the cycle of violence in maltreated children. Science 297, 851–854

380

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 12 Lenman, J. (2006) Compatibilism and contractualism: the possibility of moral responsibility. Ethics 117, 7–31 13 Fodor, J. (1987) Psychosemantics: The Problem of Meaning in the Philosophy of Mind, MIT Press 1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.06.009 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

Review

Special Issue: The Genetics of Cognition

Genetics of human episodic memory: dealing with complexity Andreas Papassotiropoulos1,2,3 and Dominique J.-F. de Quervain2,4 1

Department of Psychology, Division of Molecular Neuroscience, University of Basel, Basel, Switzerland University Psychiatric Clinics, University of Basel, Basel, Switzerland 3 Department Biozentrum, Life Sciences Training Facility, University of Basel, Basel, Switzerland 4 Department of Psychology, Division of Cognitive Neuroscience, University of Basel, Basel, Switzerland 2

Episodic memory is a polygenic behavioral trait with substantial heritability estimates. Despite its complexity, recent empirical evidence supports the notion that behavioral genetic studies of episodic memory might successfully identify trait-associated molecules and pathways. The development of high-throughput genotyping methods, of elaborated statistical analyses and of phenotypic assessment methods at the neural systems level will facilitate the reliable identification of novel memory-related genes. Importantly, a necessary crosstalk between behavioral genetic studies and investigation of causality by molecular genetic studies will ultimately pave the way towards the identification of biologically important, and hopefully druggable, genes and molecular pathways related to human episodic memory. Heritability of episodic memory as a prerequisite for genetic studies Episodic memory (see Glossary) is a major neurocognitive system that enables conscious recollection of past experiences (e.g. autobiographical episodes and learned material) along with their spatial and temporal contexts [1]. Episodic memory is subserved by a broad network of cortical and subcortical brain regions, is uniquely different from other memory systems, and probably unique to humans [1]. From a genetic standpoint, episodic memory can be deﬁned as a genetically complex behavioral trait with substantial heritability estimates (i.e. genetic factors account for a signiﬁcant proportion of the variance of this phenotype). Several twin studies report heritability values between 30% and 60% [2–8], indicating that naturally occurring genetic variations have an important impact on this cognitive ability. The majority of these studies treat episodic memory as a single uniﬁed construct without breaking it down to its underlying phenotypic structure, which is characterized, for example, by temporally distinct phases as assessed by short-delay recall and long-delay recall. However, recent empirical evidence from twin studies [9] revealed both overlapping and distinct genetic inﬂuences on these temporal components of episodic memory. In brief, twin studies in nonclinical populations clearly Corresponding authors: Papassotiropoulos, A. ([email protected]); de Quervain, D.J.F. ([email protected]).

demonstrate the importance of the genome for this neurocognitive trait. The signiﬁcant proportion of phenotypic variability in episodic memory that is attributable to heritable factors is a prerequisite and a starting point for targeted genetic studies aimed at identifying speciﬁc molecules and the molecular pathways related to speciﬁc components of human episodic memory (Box 1). Human genetic approaches Traditionally, two main research strategies exist for the identiﬁcation of genes related to heritable traits: linkage studies and association studies. In linkage studies, genetic markers of known chromosomal localization are usually examined within large pedigrees. If a certain marker is linked with the trait under study due to linkage disequilibrium with the trait-related genetic variant, then positional cloning, ﬁne-mapping and in-depth resequencing can be performed to identify the trait-associated gene along with its causal variants. This strategy has been, and still is, particularly successful for traits and diseases following Mendelian inheritance but is also applicable to some geGlossary Allele: one of two or more forms of a gene, located on a specific position on a chromosome. Complex trait: a quantifiable property of an organism influenced by both genetic and environmental factors as well as interactions between them. Epigenetics: research field that investigates potentially heritable changes in gene expression caused by mechanisms other than changes in DNA sequence. Episodic memory: a memory system that enables conscious recollection of past experiences (e.g. autobiographical episodes and learned material) along with their spatial and temporal contexts. Mendelian inheritance: patterns of inheritance (i.e. manner in which genes and traits are passed from parents to offspring) following rules described in the 19th century by Gregor Johann Mendel, an Augustinian friar and scientist. Examples of diseases following Mendelian inheritance patterns include autosomal dominant (e.g. Huntington’s disease), autosomal recessive (e.g. cystic fibrosis) and sex chromosome-linked diseases (e.g. hemophilia A). Missing heritability: describes the fact that despite the success of large and dense genome-wide scans in the discovery of trait-associated genetic variants, a large portion of the heritability of complex traits still remains unexplained. Phenotype: physical appearance of an organism with respect to a trait (e.g. blue eye color). Polymorphism: in genetics, a difference in DNA sequence among individuals. A common form of a genetic polymorphism is a SNP, which occurs when a nucleotide — A, T, C or G — differs between individuals. The human genome contains millions of SNPs. Statistical epistasis: a concept defined about a century ago and dealing with the statistical deviation from additive interaction effects between two or more genetic polymorphisms.

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.005 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

381

Review Box 1. Some prerequisites for successful genetic studies of human episodic memory Heritability: significant heritability (i.e. genetic factors account for a significant proportion of this phenotype’s variance) is a prerequisite for the reliable identification of genotype–phenotype correlations. Human episodic memory shows substantial heritability, with values ranging from 30% to 60%. Reliable and valid phenotypic assessment: episodic memory performance can be quantified by using reliable and valid instruments. Episodic memory is not a single phenotype but instead might show distinct molecular profiles, depending on the specific episodic memory task (e.g. word list vs story recall). Control for episodic memory-unrelated phenotypes: factors such as attention, concentration, motivation and working memory could influence episodic memory performance and might bias genetic association findings. Controlling for episodic memoryunrelated phenotypes is therefore crucial. Specific neural correlates: episodic memory has well definable and well quantifiable neural correlates (e.g. memory-related activity of the hippocampus and parahippocampal gyrus). This allows for further corroboration of genetic association findings at the neural systems level. Independent replication: regardless of the statistical models used and however simple or complicated they might be, successful gene identification stands and falls with independent replication of the gene–phenotype associations. This is particularly important in the era of GWAS, which screen for association between heritable traits and millions of genetic variants distributed over the entire genome, thereby introducing a serious type I error burden.

netically complex traits. Regarding episodic memory, linkage studies have successfully identiﬁed genes related to diseases characterized by impaired episodic memory, such as Alzheimer’s disease and other neurodegenerative conditions [10]. Association studies follow a different strategy. Based on theoretical considerations and previous knowledge, and in an unbiased genome-wide manner, the genotype frequencies of common genetic polymorphisms (i.e. with a minor allele frequency 1%) of groups of individuals are compared. Genetic association studies can be performed using both case-control (i.e. comparing unrelated individuals stratiﬁed into distinct phenotypic groups) and family-based designs (i.e. using relatives as a control group). Importantly (and in the case of episodic memory essentially), association studies are particularly applicable to quantitative traits in settings using both unrelated and related individuals. Candidate gene studies Whenever sufﬁcient knowledge about the biological mechanisms underlying the physiology (and in case of disease, the pathophysiology) of a certain trait exists, candidate gene association studies, which assess the association between one or more alleles of a biologically plausible gene (or a set of genes) with the phenotype, could be readily implemented. In the case of episodic memory, sufﬁcient biological knowledge does exist. Several animal studies in both invertebrates and vertebrates over the past three decades have identiﬁed genes and signaling molecules important for memory, including protein kinases and phosphatases, transcription factors, growth factors and receptor complexes [11–15]. These studies suggest that many memory-related molecular mechanisms are conserved across species and that molecules related to learning-related 382

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

synaptic plasticity, long-term potentiation, long-term depression and activity of such brain structures as the hippocampus, parahippocampal gyrus and the amygdala might represent ideal biological candidates for genetic association studies. Indeed, in 2003 two articles supported the notion that the candidate gene approach is promising by showing that functional single nucleotide polymorphisms (SNPs) in the genes encoding the 5-hydroxytryptamine (serotonin) receptor 2A (HTR2A) and the brainderived neurotrophic factor (BDNF) are associated with episodic memory performance [16,17]. Both molecules are highly expressed in the hippocampus and the cerebral cortex and have been implicated in animal learning and memory [18–20]. In humans, a common HTR2A SNP predicts an amino acid substitution (His to Tyr) at residue 452 (H452Y). Compared with carriers of the common His/His variant, heterozygous (His/Tyr) carriers show a blunted receptor response upon pharmacological stimulation with serotonin [21,22] and poorer episodic memory performance in verbal and ﬁgural tasks [16]. BDNF harbors a nonconservative SNP, which is located in the 50 pro-BDNF sequence and results in a Val to Met substitution at codon 66. The Met allele is related to deﬁcits in the cellular distribution and regulated secretion of BDNF and to poorer episodic memory performance [17]. The demonstration of successful identiﬁcation of episodic memory-related molecules through the candidate gene approach prompted subsequent studies, which reported on additional genes related to this cognitive capacity such as COMT [23], GRM3 [24], PRNP [25], CHRFAM7A [26], APOE [27,28], PDYN [29] and CPEB3 [30]. Although the majority of these studies remain to be independently replicated, they do suggest that focusing on pre-existing and well-established biological information could lead to the identiﬁcation of genes related to episodic memory in humans. Genome-wide association studies (GWAS) On the one hand pre-existing biological information clearly facilitates the search for biologically meaningful candidates, on the other hand it could introduce a severe bias towards readily accessible molecular pathways and it deﬁnitely limits the potential of genetic association studies to identify novel genes and molecular pathways. Recent advances in the development of high-density genotyping platforms and analytical software now allow for high-resolution GWAS, which screen for association between heritable traits and millions of genetic variants distributed over the entire genome. This dense screening feature renders GWAS particularly useful in discovering novel molecular pathways of genetically complex traits. The ﬁrst GWAS on episodic memory utilized a genotyping microarray containing more than 500 000 polymorphic markers and identiﬁed KIBRA and CLSTN2 as episodic memory-related genes [31]. Replication studies in European and Asian populations further supported the role of KIBRA in episodic memory (e.g. [32–35]), although not unequivocally [36]. Importantly, a study in two large Scottish samples suggested that KIBRA is related to processes speciﬁc to the conscious recall of item-based material, possibly reﬂecting hippocampal processing [32]. This high degree of speciﬁcity underscores the importance of precise

Review phenotypic assessment in behavioral genetic research, including replication studies. Subsequent GWAS in healthy participants supported the role of additional genes such as CAMTA1 [37] and NRXN1 [38] in episodic memory performance by using a variety of neuropsychological measurements. It is important to stress that the genes identiﬁed through genome-wide screening are involved in memory-related molecular pathways such as protein phosphorylation, calcium-responsive transcriptional activation and, in general, synaptic plasticity. Nevertheless, it is highly unlikely that most of these genes would have been selected a priori in a candidate gene setting, simply because very little was known regarding the biological relevance of these genes before their identiﬁcation in the course of the GWAS. This demonstrates the substantial potential of this approach for the identiﬁcation of novel genes and pathways related to human episodic memory. Our knowledge about the molecular underpinnings of this trait has already increased and it will undoubtedly continue to increase as larger and denser GWAS, utilizing reﬁned analytical methodology, are expected in the near future. Genetic complexity GWAS could powerfully identify components of the genetic basis of complex multigenic traits. In the past few years, GWAS using hundreds of thousands and even millions of polymorphic markers, mainly SNPs, in samples ranging from a few hundred to several thousands of individuals have led and still lead to the identiﬁcation of numerous susceptibility genes and trait-related genomic variants. However, despite this extensive use of genetic and analytical force, which has undoubtedly proven successful, a major portion of the heritability of complex traits still remains unexplained, a phenomenon commonly termed ‘missing heritability’ [39]. Although recent empirical and simulation studies suggest that modeling thousands of SNPs concurrently could explain large proportions of the missing heritability [40,41], the proportion of the observed phenotypic variance explained by the speciﬁc and statistically signiﬁcant trait-associated variants is low. In other words, appropriate modeling shows that a large proportion of the heritability of a complex trait can be estimated by the genomic content of an SNP array [40]. However, when asking how much of the heritability of a complex trait can be explained by the signiﬁcant GWAS SNPs identiﬁed to date, then the answer is generally low, depending on the trait of interest. Gene–gene interactions In addition to parent-of-origin effects and epigenetic factors, causal but nonexamined rare variants, and environmental inﬂuences, the negligence of analytical approaches accounting for gene–gene interaction effects, such as statistical epistasis, could partly explain the phenomenon of ‘missing heritability’ [42]. Indeed, despite the obvious conception that the analysis of genetically complex traits should account for the underlying biological and statistical complexity, the vast majority of large-scale genetic association studies to date are restricted to the use of singlemarker statistics. Clearly, this approach does not fully account for the polygenic nature of the phenotype under

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

study and erroneously implies that the impact of genetic variation is due to independently acting effects. The above also holds true for the study of the genetics of episodic memory. Although the large-scale analysis of statistical epistasis related to complex traits has proven feasible [43], only a few studies have addressed epistatic effects of a limited number of genetic variants on cognitive functions [44–47]. A functional magnetic resonance imaging (fMRI) study of 82 healthy participants demonstrated signiﬁcant nonlinear interaction effects of two dopaminerelated genes. Speciﬁcally, both the COMT Val/Met polymorphism and the functional variable number of tandem repeat polymorphism in the 30 untranslated region of the dopamine transporter gene (SLC6A3) were independently associated with activation differences in the hippocampus during the encoding phase of a hippocampus-dependent memory task. In addition to these independent effects, interaction analyses revealed a nonlinear relationship between the two polymorphisms also at the hippocampal level [46]. Thus, focusing on main SNP effects only and neglecting the study of possible interaction effects most probably leaves a signiﬁcant proportion of variability (and heritability) in episodic memory unexplained. In the light of the complex polygenic nature of episodic memory, studies accounting for the statistical interaction between few (e.g. two) genetic variants should be regarded as proof-of-principle studies and as a starting point for more complex approaches. However, studies involving comprehensive sets of genetic variants and addressing interaction (e.g. epistatic) effects on human memory are lacking. The concept of statistical epistasis was deﬁned about a century ago and deals with the statistical deviation from additive interaction effects between genetic markers [48]. Per deﬁnition, the inclusion of such statistical interaction terms as epistasis exponentially increases the number of statistical tests performed. For example, a two-way interaction analysis between 1000 single markers requires the performance of 499 500 tests. Some strategies attempting to limit the number of necessary tests in analyses of epistasis employ a stepwise procedure by including only those interaction terms for which the corresponding marker showed a signiﬁcant main effect in the ﬁrst step singlemarker analysis. However, this approach is arbitrary because there is no biological rationale for considering only markers with signiﬁcant main effects. Indeed, a recently published report of an exhaustive genome-wide analysis showed that signiﬁcant epistatic interactions would have been missed if SNPs that did not display any main effect had been excluded a priori [49]. It remains to be shown whether these exhaustive approaches will prove feasible and successful in the study of the complex genetic architecture of episodic memory (Box 2). Gene clusters Rather than computing complex gene–gene interactions, another approach to capture the interplay between genetic variants and its impact on episodic memory focuses on the computation of compound gene and SNP clusters based on multilocus analyses. For example, a permutation-based method, termed set association, evaluates sets of polymorphic markers and provides a cluster of signiﬁcant alleles 383

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

and genotypes with a single test statistic. Importantly, such compound analyses deﬁning gene clusters can be used for the calculation of aggregate individual genetic scores, which principally reﬂect a person’s number of trait-associated genetic variants weighted by the effect size of each variant. With regard to episodic memory, such approaches have proven feasible [50] and extendable to conditions of pathological cognition [51]. These multilocus methods represent an extension of the candidate gene approach and deal with sets of genes in biologically meaningful candidate pathways. A preselection of the human homologues of 47 genes with well-established molecular and biological functions in synaptic plasticity and animal memory led to the identiﬁcation of a 7-gene cluster associated with episodic memory [50]. This gene cluster represents important memory-related molecules such as adenylyl cyclases, kinases and glutamate receptors. An aggregate, individual gene score based on the 7-gene cluster was also associated with activations in memory-related brain regions, such as the hippocampus and parahippocampal gyrus (Figure 1). The

Box 2. Dealing with the genetic complexity of episodic memory It is impossible to predict which statistical and bioinformatics approaches will ideally explain the relationship between genetic and phenotypic variability of such complex traits as episodic memory. It will be most probably a combination of several approaches specific to distinct memory-related phenotypes. The obvious genetic complexity of episodic memory, along with studies showing the importance of statistical methods accounting for gene–gene interactions, supports the notion that rare combinations of common variants significantly account for the variability and heritability of this trait. Therefore, the study of the genetic architecture of episodic memory will require a statistical and systems biology framework that accounts for the complexity of the interacting functional molecular networks. The current approaches (e.g. set-association [65], set-based tests [66], multifactor dimensionality reduction [67], gene set enrichment analysis [68], to name a few) are clearly a very good start. However, as more and more methods will become available, study designs based on multidisciplinary research frameworks and focusing on replication of the identified complex genetic structures will prove invaluable for the robust and reliable discovery of parts of the genetic underpinnings of episodic memory.

[(Figure_1)TD$IG] (b)

Contributing loci

ADCY8

GRIN2A

GRM1

CAMK2B

PRKACB

PRKAR2B

PPP3CC

CREB1

GRIN2B

GRM3

CAMK2D

PRKACG

MAPK1

PPP1CA

CREB3

GRIN2C

GRM4

CAMK2G

PRKCD

MAPK3

PPP1CB

CREBBP

GRIN2D

GRM5

PRKCA

PRKCE

FYN

PPP1CC

NCAM1

GRIN3A

GRIA1

PRKCB1

PRKCH

CALM2

ADCY1

GRIN3B

GRIA4

PRKCG

PRKAG1

PPP3CA

ADCY2

Cluster P = 0.00008

20 15 10

5 0

(c)

PRKCA

PPP3CB

GRM3

PRKAR1A

GRIN2A

PRKACA

GRIN2B

CAMK2A

PRKACG

GRINA

ADCY8

GRIN1

% Contribution to gene cluster

25

CAMK2G

(a)

(d)

1.5

1

7 6

4 3 2 1 0

T value

5

Response at [24, -12, -20]

MTL

0.5

0

-0.5

-1 0.8

0.9

1

1.1

1.2 1.3 IMAGS

1.4

1.5 1.6

TRENDS in Cognitive Sciences

Figure 1. Gene score related to human episodic memory and memory-related brain activity. (a) Squares represent a selection of the human homologues of 47 genes with well-established molecular and biological functions in synaptic plasticity and/or animal memory. Squares in red represent the seven genes for which significant association of genetic variations with human episodic memory was found. (b) The seven-SNP cluster was used for the calculation of an individual’s memory-related genetic score, termed individual memory-associated genetic score (IMAGS). (c) Regression analysis revealed a significant positive correlation between the IMAGS and learning-induced brain activations in the medial temporal lobe (MTL), including the hippocampus and parahippocampal gyrus. (d) Scatter plot illustrating the positive correlations between IMAGS and learning-induced brain activations in the hippocampus at coordinate position [24, 12, 20]. Adapted from [50].

384

Review computation of aggregate genetic scores based on genetic clusters has hitherto relied on pre-existing information (i.e. a candidate pathway approach). Recently, similar gene clustering methods have been used for the calculation of genetic risk proﬁles by utilizing GWAS data [52]. Most probably, capitalizing on a combination of GWAS data and gene clustering methods will also facilitate the unbiased identiﬁcation of novel gene clusters related to episodic memory capacity. Phenotypic complexity Episodic memory refers to memory for past experiences (e.g. autobiographical episodes and learned material), which includes information about the content of the experience and the spatial and temporal context in which it occurred [53,54]. It is important to realize that, in terms of neurobiological underpinnings, episodic memory performance could be subserved by distinct molecular proﬁles, depending on the speciﬁc episodic memory task used for quantifying performance. For example, the BDNF Val66Met polymorphism has been shown to be associated with delayed recall from stories of the Wechsler Memory Scale revised version but not with delayed recall of a word list taken from the California Verbal Learning Test [17], yet both tasks are referred to as episodic memory tasks. Furthermore, it is well known that emotionally arousing information is better stored into memory than neutral information and that this phenomenon depends on the activation of noradrenergic transmission [55,56]. Consequently, it has been shown that a functional deletion variant of ADRA2B, the gene encoding the a2b-adrenoceptor, is related to differential episodic memory performance for emotionally arousing pictures but not for neutral pictures [57]. Furthermore, it is important to note that several factors, which are not considered to belong to episodic memory, such as motivation, attention or concentration can have a large impact on performance in episodic memory tasks. Thus, in behavioral genetic studies on episodic memory it is crucial to control for such unspeciﬁc factors. The phenotypic complexity with regard to episodic memory not only has important and obvious implications for replication studies [32] but also for meta- and megaanalyses (i.e. the use of raw data from individual subjects across different studies): by combining several similar but not identical episodic memory phenotypes, genetic studies could identify statistically robust associations pointing to common denominators of these phenotypes but are likely to miss molecules speciﬁcally related to a distinct phenotype with distinct neurobiological features. Imaging genetics Besides the importance of replicating genetic association studies on episodic memory it is important to also consider additional methods for further corroboration and better understanding of the behavioral genetics ﬁndings. In the recent few years, brain-imaging techniques, in particular fMRI, have become increasingly popular for this purpose [31,58–60]. The rationale for combining behavioral genetic and neuroimaging methods is to validate and extend purely behavioral genetic studies by providing insight into the genetic differences in memory processes at the level of

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

neural circuits [61]. So far, the majority of imaging genetic studies has looked at single markers in relation to brain activation differences. However, considering the genetic and phenotypic complexity of human episodic memory, another promising approach involves studying speciﬁc genetic networks (comprising multiple markers in different genes) related to a certain episodic memory phenotype by investigating compound genetic scores in relation to brain activity [50] (Figure 1). The interpretation of genotype-dependent differences in brain activity depends crucially on the fMRI study design and the behavioral ﬁndings of the fMRI sample. Whenever signiﬁcant performance differences across genotype groups exist, a possible outcome is higher activity in memoryrelated brain regions in the genotype group that shows higher memory performance [62]. In cases in which genotype groups of the fMRI sample have been matched for performance, increases in brain activity of the genotype group with low memory performance in the unmatched population can be interpreted as compensatory activity to achieve the same level of performance as the genotype group with high memory performance [31]. Interpretation becomes difﬁcult if the fMRI study reveals genotype-dependent activity differences despite nonsigniﬁcant differences in memory performance in genotype groups unmatched for performance. This situation is common because the number of subjects used in imaging genetic studies reporting signiﬁcant genotype-dependent differences in brain activity is typically between 20 and 60, whereas behavioral genetics studies usually used hundreds or thousands of subjects to consistently produce signiﬁcant results [61]. A possible explanation for this observation is that neural activity is more proximate to the direct effects of functional genetic polymorphisms on gene products and their function, and might therefore be more sensitive in estimating genotypedependent differences in mental processing [61,63,64]. Nevertheless, genotype-dependent differences in brain activity that do not translate to signiﬁcant differences in behavior should be interpreted with caution. Concluding remarks As with any genetically complex trait, neuroscientists focusing on the study of the genetic underpinnings of episodic memory have to deal with many complexity levels. Firstly, the trait (i.e. episodic memory) per se represents an assembly of phenotypes with overlapping but also partially distinct molecular proﬁles. Secondly, complexity at the genetic level is not only related to the polygenic nature of the trait itself but it also reﬂects the complexity of the human genome, which exerts its inﬂuence on the trait not only through simple linear gene effects but also through gene–gene interactions, gene–environment interactions and epigenetic mechanisms. Finally, the relation between genetic and phenotypic variability is not expected to follow simple and general rules applicable to every memoryrelated phenotype. Despite this obvious complexity, empirical evidence supports the notion that behavioral genetic studies of episodic memory successfully identify trait-associated molecules and pathways. As new technological and analytical approaches – both at the genetic and the phenotypic level – emerge, two 385

(Figure_2)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

MTL

GWAS

Candidate gene studies

Chair

7 6 5 4 3 2 1 0

T value

Human behavioral/imaging genetic studies

Animal behavioral & molecular studies Morris water maze

TRENDS in Cognitive Sciences

Figure 2. Crosstalk between behavioral/imaging genetic studies in humans, and behavioral and molecular genetic studies in animals. Human candidate gene studies depend on pre-existing information, which mostly originates from animal and molecular studies. GWAS often identify genes for which the function is not known or poorly understood. In such cases, behavioral and molecular genetic studies in animals could be useful to learn more about the causal role and the function of the gene.

simple and important pillars of reliable genetic research must not be forgotten. Firstly, regardless of the statistical models used and however simple or complicated they might be, successful gene identiﬁcation stands and falls with independent replication of the gene–phenotype associations. Secondly, given the correlative nature of behavioral genetic research, genetic ﬁndings must be corroborated through implementation of additional phenotypic assessment and methods. Whereas fMRI is certainly a powerful tool to gain insight into the genetic differences in memory processes at the level of neural circuits, it is important to note that imaging genetic studies are also correlational in nature and do not allow causal interpretations. Methods, which allow researchers to investigate the causal role of a certain gene or genetic variant, include studies in genetically modiﬁed animals and studies aiming at intervening at the level of gene products. Such studies are especially important to gain more knowledge about genes identiﬁed in GWAS for which the function is not known or poorly understood (Figure 2). This crosstalk between genetic association and investigation of causality will ultimately pave the way towards the identiﬁcation of biologically important, and hopefully druggable, genes and molecular pathways related to human episodic memory. References 1 Tulving, E. (2002) Episodic memory: from mind to brain. Annu. Rev. Psychol. 53, 1–25

386

2 Alarcon, M. et al. (1998) Multivariate path analysis of speciﬁc cognitive abilities data at 12 years of age in the Colorado Adoption Project. Behav. Genet. 28, 255–264 3 Bouchard, T.J., Jr et al. (1990) Genetic and environmental inﬂuences on special mental abilities in a sample of twins reared apart. Acta Genet. Med. Gemellol (Roma) 39, 193–206 4 Finkel, D. et al. (1995) Genetic inﬂuences on memory performance in adulthood: comparison of Minnesota and Swedish twin data. Psychol. Aging 10, 437–446 5 Johansson, B. et al. (1999) Origins of individual differences in episodic memory in the oldest-old: a population-based study of identical and same-sex fraternal twins aged 80 and older. J. Gerontol. B: Psychol. Sci. Soc. Sci. 54, P173–P179 6 McClearn, G.E. et al. (1997) Substantial genetic inﬂuence on cognitive abilities in twins 80 or more years old. Science 276, 1560–1563 7 Swan, G.E. et al. (1999) Differential genetic inﬂuence for components of memory in aging adult twins. Arch. Neurol. 56, 1127–1132 8 Volk, H.E. et al. (2006) Genetic inﬂuences on free and cued recall in long-term memory tasks. Twin Res. Hum. Genet. 9, 623–631 9 Panizzon, M.S. et al. (2011) Genetic architecture of learning and delayed recall: a twin study of episodic memory. Neuropsychology 25, 488–498 10 Papassotiropoulos, A. et al. (2006) Genetics, transcriptomics, and proteomics of Alzheimer’s disease. J. Clin. Psychiatry 67, 652–670 11 Dudai, Y. (2002) Molecular bases of long-term memories: a question of persistence. Curr. Opin. Neurobiol. 12, 211–216 12 Kandel, E.R. (2001) The molecular biology of memory storage: a dialogue between genes and synapses. Science 294, 1030–1038 13 Shobe, J. (2002) The role of PKA, CaMKII, and PKC in avoidance conditioning: permissive or instructive? Neurobiol. Learn. Mem. 77, 291–312 14 Tonegawa, S. et al. (2003) Genetic neuroscience of mammalian learning and memory. Philos. Trans. R. Soc. Lond. B: Biol. Sci. 358, 787–795 15 Waddell, S. and Quinn, W.G. (2001) What can we teach Drosophila? What can they teach us? Trends Genet. 17, 719–726 16 de Quervain, D.J. et al. (2003) A functional genetic variation of the 5HT2a receptor affects human memory. Nat. Neurosci. 6, 1141–1142 17 Egan, M.F. et al. (2003) The BDNF val66met polymorphism affects activity-dependent secretion of BDNF and human memory and hippocampal function. Cell 112, 257–269 18 Poo, M.M. (2001) Neurotrophins as synaptic modulators. Nat. Rev. Neurosci. 2, 24–32 19 Buhot, M.C. (1997) Serotonin receptors in cognitive behaviors. Curr. Opin. Neurobiol. 7, 243–254 20 Sheline, Y.I. et al. (2002) Greater loss of 5-HT(2A) receptors in midlife than in late life. Am. J. Psychiatry 159, 430–435 21 Ozaki, N. et al. (1996) Two naturally occurring amino acid substitutions of the 5-HT2A receptor: similar prevalence in patients with seasonal affective disorder and controls. Biol. Psychiatry 40, 1267–1272 22 Ozaki, N. et al. (1997) A naturally occurring amino acid substitution of the human serotonin 5-HT2A receptor inﬂuences amplitude and timing of intracellular calcium mobilization. J. Neurochem. 68, 2186–2193 23 de Frias, C.M. et al. (2004) COMT gene polymorphism is associated with declarative memory in adulthood and old age. Behav. Genet. 34, 533–539 24 Egan, M.F. et al. (2004) Variation in GRM3 affects cognition, prefrontal glutamate, and risk for schizophrenia. Proc. Natl. Acad. Sci. U.S.A. 101, 12604–12609 25 Papassotiropoulos, A. et al. (2005) The prion gene is associated with human long-term memory. Hum. Mol. Genet. 14, 2241–2246 26 Dempster, E.L. et al. (2006) Episodic memory performance predicted by the 2 bp deletion in exon 6 of the ‘alpha 7-like’ nicotinic receptor subunit gene. Am. J. Psychiatry 163, 1832–1834 27 De Blasi, S. et al. (2009) APOE polymorphism affects episodic memory among non demented elderly subjects. Exp. Gerontol. 44, 224–227 28 Mondadori, C.R. et al. (2007) Better memory and neural efﬁciency in young apolipoprotein E epsilon4 carriers. Cereb. Cortex 17, 1934–1947 29 Kolsch, H. et al. (2009) Gene polymorphisms in prodynorphin (PDYN) are associated with episodic memory in the elderly. J. Neural. Transm. 116, 897–903

Review 30 Vogler, C. et al. (2009) CPEB3 is associated with human episodic memory. Front. Behav. Neurosci. 3, 4 31 Papassotiropoulos, A. et al. (2006) Common Kibra alleles are associated with human memory performance. Science 314, 475–478 32 Bates, T.C. et al. (2009) Association of KIBRA and memory. Neurosci. Lett. 458, 140–143 33 Yasuda, Y. et al. (2010) Association study of KIBRA gene with memory performance in a Japanese population. World J. Biol. Psychiatry 11, 852–857 34 Preuschhof, C. et al. (2010) KIBRA and CLSTN2 polymorphisms exert interactive effects on human episodic memory. Neuropsychologia 48, 402–408 35 Schaper, K. et al. (2008) KIBRA gene variants are associated with episodic memory in healthy elderly. Neurobiol. Aging 29, 1123–1125 36 Need, A.C. et al. (2008) Failure to replicate effect of Kibra on human memory in two large cohorts of European origin. Am. J. Med. Genet. B: Neuropsychiatr. Genet. 147B, 667–668 37 Huentelman, M.J. et al. (2007) Calmodulin-binding transcription activator 1 (CAMTA1) alleles predispose human episodic memory performance. Hum. Mol. Genet. 16, 1469–1477 38 Need, A.C. et al. (2009) A genome-wide study of common SNPs and CNVs in cognitive performance in the CANTAB. Hum. Mol. Genet. 18, 4650–4661 39 Frazer, K.A. et al. (2009) Human genetic variation and its contribution to complex traits. Nat. Rev. Genet. 10, 241–251 40 Makowsky, R. et al. (2011) Beyond missing heritability: prediction of complex traits. PLoS Genet. 7, e1002051 41 Yang, J. et al. (2010) Common SNPs explain a large proportion of the heritability for human height. Nat. Genet. 42, 565–569 42 Eichler, E.E. et al. (2010) Missing heritability and strategies for ﬁnding the underlying causes of complex disease. Nat. Rev. Genet. 11, 446–450 43 Cordell, H.J. (2009) Detecting gene-gene interactions that underlie human diseases. Nat. Rev. Genet. 10, 392–404 44 Greenwood, P.M. et al. (2009) Synergistic effects of genetic variation in nicotinic and muscarinic receptors on visual attention but not working memory. Proc. Natl. Acad. Sci. U.S.A. 106, 3633–3638 45 Stelzel, C. et al. (2009) Effects of dopamine-related gene–gene interactions on working memory component processes. Eur. J. Neurosci. 29, 1056–1063 46 Bertolino, A. et al. (2008) Epistasis between dopamine regulating genes identiﬁes a nonlinear response of the human hippocampus during memory tasks. Biol. Psychiatry 64, 226–234 47 Tan, H.Y. et al. (2007) Epistasis between catechol-O-methyltransferase and type II metabotropic glutamate receptor 3 genes on working memory brain function. Proc. Natl. Acad. Sci. U.S.A. 104, 12536–12541 48 Fisher, R.A. (1918) The correlation between relatives on the supposition of Mendelian inheritance. Philos. Trans. R. Soc. Edinb. 52, 399–404 49 Steffens, M. et al. (2010) Feasible and successful: genome-wide interaction analysis involving all 1.9 10(11) pair-wise interaction tests. Hum. Hered. 69, 268–284

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 50 de Quervain, D.J. and Papassotiropoulos, A. (2006) Identiﬁcation of a genetic cluster inﬂuencing memory performance and hippocampal activity in humans. Proc. Natl. Acad. Sci. U.S.A. 103, 4270–4274 51 Jablensky, A. et al. (2011) Polymorphisms associated with normal memory variation also affect memory impairment in schizophrenia. Genes Brain Behav. 10, 410–417 52 Demirkan, A. et al. (2011) Genetic risk proﬁles for depression and anxiety in adult and elderly cohorts. Mol. Psychiatry 16, 773–783 53 Squire, L.R. and Zola, S.M. (1998) Episodic memory, semantic memory, and amnesia. Hippocampus 8, 205–211 54 Tulving, E. (1972) Episodic and semantic memory. In Organization of Memory (Tulving, E. and Donaldson, W., eds), pp. 381–403, Academic Press 55 Cahill, L. et al. (1994) Beta-adrenergic activation and memory for emotional events. Nature 371, 702–704 56 McGaugh, J.L. and Roozendaal, B. (2002) Role of adrenal stress hormones in forming lasting memories in the brain. Curr. Opin. Neurobiol. 12, 205–210 57 de Quervain, D.J. et al. (2007) A deletion variant of the alpha2badrenoceptor is related to emotional memory in Europeans and Africans. Nat. Neurosci. 10, 1137–1139 58 Krug, A. et al. (2010) The effect of Neuregulin 1 on neural correlates of episodic memory encoding and retrieval. Neuroimage 53, 985–991 59 Schott, B.H. et al. (2006) The dopaminergic midbrain participates in human episodic memory formation: evidence from genetic imaging. J. Neurosci. 26, 1407–1417 60 Urner, M. et al. (2011) Genetic variation of the alpha2b-adrenoceptor affects neural correlates of successful emotional memory formation. Hum. Brain Mapp. DOI: 10.1002/hbm.21171 61 Rasch, B. et al. (2010) Imaging genetics of cognitive functions: focus on episodic memory. Neuroimage 53, 870–877 62 Hariri, A.R. et al. (2003) Brain-derived neurotrophic factor val66met polymorphism affects human memory-related hippocampal activity and predicts memory performance. J. Neurosci. 23, 6690–6694 63 Hariri, A.R. et al. (2006) Imaging genetics: perspectives from studies of genetically driven variation in serotonin function and corticolimbic affective processing. Biol. Psychiatry 59, 888–897 64 Mattay, V.S. et al. (2008) Neurobiology of cognitive aging: insights from imaging genetics. Biol. Psychol. 79, 9–22 65 Hoh, J. et al. (2001) Trimming, weighting, and grouping SNPs in human case-control association studies. Genome Res. 11, 2115–2119 66 Purcell, S. et al. (2007) PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet. 81, 559– 575 67 Hahn, L.W. et al. (2003) Multifactor dimensionality reduction software for detecting gene-gene and gene-environment interactions. Bioinformatics 19, 376–382 68 Zhang, K. et al. (2010) i-GSEA4GWAS: a web server for identiﬁcation of pathways/gene sets associated with traits by applying an improved gene set enrichment analysis to genome-wide association study. Nucleic Acids Res. 38, W90–W95

387

Review

Special Issue: The Genetics of Cognition

The genetics of cognitive ability and cognitive ageing in healthy older people Sarah E. Harris1,2 and Ian J. Deary2 1

Centre for Cognitive Ageing and Cognitive Epidemiology, Medical Genetics Section, University of Edinburgh, Edinburgh, EH4 2XU, UK 2 Centre for Cognitive Ageing and Cognitive Epidemiology, Department of Psychology, University of Edinburgh, 7 George Square, Edinburgh, EH8 9JZ, UK

Determining the genetic influences on cognitive ability in old age and in cognitive ageing are important areas of research in an increasingly ageing society. Heritability studies indicate that genetic variants strongly influence cognitive ability differences throughout the lifespan, including in old age. To date, however, only the genes encoding apolipoprotein E (APOE) and possibly catechol-O-methyl transferase (COMT), brain-derived neurotrophic factor (BDNF) and dystrobrevin binding protein 1 (DTNBP1) have repeatedly been associated in candidate gene studies with cognitive decline or with cognitive ability in older individuals. Genome-wide association studies have identified further potential loci, but results are tentative. Advances in exome and/or whole-genome sequencing, transcriptomics, proteomics and methylomics hold significant promise for uncovering the genetic underpinnings of cognitive ability and decline in old age. Human cognitive ability and non-pathological cognitive ageing To speak of the ‘genetics of’ cognitive ability and cognitive ageing refers to the possibility that genetic differences cause some of the variation in these two phenotypes that is manifest in humans. To proceed, there must be clarification of the difference between cognitive ability and cognitive ageing, and of the phenotypes to be studied within each. Pathological states of cognitive decline (the dementias, other neurodegenerative conditions and prodromal states, such as mild cognitive impairment) are important research areas, and with their own research fields, including genetics. However, in the present overview, we focus on genetic contributions to normal, non-pathological variation in cognitive ageing. This is an important and large research area in its own right, in part because cognitive ageing affects people’s quality of life, predicts dementia and death, and is at this stage more likely to be open to amelioration [1,2]. We present evidence from twin studies that indicate that cognitive ability in old age is highly heritable. We then describe both candidate gene and genome-wide association studies (GWAS; see Glossary) that Corresponding author: Deary, I.J. ([email protected]).

388

have attempted, with limited success, to determine the specific genetic variants that influence cognitive ability and cognitive ageing. Cognitive ability and cognitive ageing Consider the scores obtained by a 75-year-old individual in several mental tests; say, tests of memory, reasoning, speed of processing and so forth. This is an age at which the adult peak in such skills is typically well past [3,4]. Therefore, scores on any one test will be the combination of three things: the previous peak level of the ability, how much age-associated change there has been, and occasionspecific variance (including error variance). To separate the phenotypes of level and change in cognitive abilities, longitudinal studies are required in which people have taken mental tests on more than one occasion. This allows a cognitive trajectory to be computed for each subject. From a sample of such subjects, it is then possible to describe and find determinants of the differences in levels and rates of change in cognitive abilities. These are different things, probably with different determinants.

Glossary Allele: one of two or more versions of a gene. Candidate gene study: a study that investigates associations between genetic variants in a gene suspected of influencing a trait of interest (by virtue of the function or chromosomal location of the gene) and the trait. Cognitive change: the change in cognitive score between occasions of measurement. Note that it is common to see ‘cognitive ageing’ referred to when only one occasion of measurement has taken place. This is a common error. Cognitive level: cognitive score obtained on one occasion. General intelligence (g): operationally, g is typically the first unrotated principal component (or factor) from a battery of mental tests administered to a sample of a population. It describes the near-universal finding that all mental tests tend to correlate positively. Genome-wide association study: a hypothesis-free study whereby up to 1 million SNPs, spread throughout the genome, are investigated to identify genomic regions associated with a particular trait. Intercept and slope: given a number of occasions of cognitive measurements, a regression equation can be applied to the longitudinal data of each individual to discover their starting level of cognitive ability (intercept) and trajectory of cognitive change (slope). Therefore, one can seek the antecedents (including genetics) and outcomes of the slope and trajectory. Note that the slope might be linear, quadratic or even more complex. Single nucleotide polymorphism (SNP): a position on a chromosome where humans are known to differ with regard to which base pair occurs.

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.004 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

mental test capabilities and forms taxonomies of abilities based on correlations among cognitive skills [5] (Box 1). The second approach, called the cognitive or neuropsychological approach, relies more on lesion and brain imaging studies, and tends to focus on mental capabilities within a given domain of function [6]. In the present article, we largely use the differential psychology approach, as this is the approach almost exclusively employed in behaviour genetic studies and the bulk of molecular genetic studies seeking to explain individual differences in cognitive function and change in old age.

Cognitive phenotypes How do we define cognitive phenotypes? If there were just, say, three types of mental work that brains performed, and if there was a perfectly valid test for each of them, this would be simple. One would apply these three tests each time one wished to assess an individual’s cognitive capabilities. This is not the case, however. There is no widely accepted taxonomy of brain functions that is isomorphic with mental tests. Therefore, what is available as phenotypes is a set of cognitive tests that are classified largely by their contents. To an extent, the contents do show isomorphisms with brain activation. Indeed, there tend to be two somewhat separate approaches to cognitive phenotypes, including how they are applied to ageing. The first is called the differential psychology (or psychometric) approach. It studies the multivariate associations among

Factors affecting age-associated cognitive decline There are changes in population mean levels of cognitive test scores as people age, but what we are interested in here is the individual differences in these life-course changes and

Box 1. The differential psychology approach to cognitive ability respect to ageing, an important distinction should be made between relatively age-sensitive cognitive domains (e.g. memory, speed of processing, spatial ability, and reasoning) and relatively age-resistant domains (e.g. verbal and numerical abilities) [50]. Furthermore, one might ask about the stability of cognitive differences, and whether age affects any or all of these three levels of cognitive variation. From youth (20 years) to later middle age (55 years) [51], and even from childhood (11 years) to old age (79 years) [52,53], there is high stability in people’s rank order of cognitive ability differences. In a study of male twins followed from age 20 to 55, genetic factors were responsible for the stability (and the same genetic influences were found at both ages) and non-shared environmental causes explained the changes [51]. With regard to how ageing affects cognitive abilities, a study of over 1200 people subjected to 12 mental tests for up to seven years found that 39% of the effect of age was on general cognitive function, 33% was at the level of domains (i.e. abstract reasoning, spatial visualization, episodic memory and processing speed) and 28% was at the level of the individual test [54]. Other longitudinal [55] and cross-sectional studies [3] also find this distribution of ageing effects across general and specific cognitive skills. To summarise: some ageing effects appear to be general and affect all cognitive skills, and some are more specific.

Put simply, the differential psychology approach asks: do individuals differ in how generally smart they are, or do they tend to differ in their specific cognitive capabilities? Both are true to some extent, and the answer lies somewhere in between. This is based on the finding that all mental tests show positive covariation, and that there are groups of types of test that show higher correlations among themselves than with other mental tests. Therefore, the sources of variance (i.e. the thing that one is trying to explore genetic contributions to) are at least on three levels [5] (Figure I). Individuals differ in how generally mentally able they are; this is often called general intelligence, or g. Approximately half of the population variation in cognitive skills is captured by variation at this level. As long as there is a reasonable number of varied mental tests applied, this g factor ranks people almost identically even when it is based on different tests [47,48]. Individuals differ in how good they are in broad mental domains (e.g. memory, processing speed, visual perception, auditory perception, retrieval ability, etc.). They also differ in their specific cognitive skills. Therefore, one can seek genetic and other contributions to variation at each of these levels. The result is that additive genetic factors contribute approximately 70% or more of the influence on g in adulthood [7,47,49]. Other levels have high heritability too, but much of this is because they are highly correlated with g and so the genetic causes of g partly also cause these lower-level differences. With

[(Box_1)TD$FIG]

g

Fluid intelligence

Test 1

Test 2

Test 3

Crystallised intelligence

Test 1

Test 2

Test 3

Memory and learning

Test 1

Test 2

Test 3

Visual perception

Test 1

Test 2

Test 3

Auditory perception

Test 1

Test 2

Test 3

Retrieval ability

Test 1

Test 2

Test 3

Cognitive speediness

Test 1

Test 2

Test 3

Processing speed

Test 1

Test 2

Test 3

TRENDS in Cognitive Sciences

Figure I. Schematic indicating the hierarchical structure of cognitive abilities. g reflects general intelligence (based on data from [5]).

389

Review what might cause them. Multiple factors have been proposed as possible determinants of individual differences in non-pathological, age-associated cognitive change [1]. A recent and thorough review examined over 100 observational studies, and many randomised controlled trials and systematic reviews [2]. The studies included nutritional, medical, socio-economic, behavioural, toxic and genetic factors. Firm evidence was lacking for most factors, but there appeared to be relatively robust evidence for risk of increased cognitive decline from the apolipoprotein E (APOE) e4 allele, smoking and some medical conditions. There was also some evidence for the protectiveness of physical exercise and cognitive training. With regard to studies of the genetic causes of differences in cognitive ageing, the simplest division of these is into behavioural and molecular genetic studies. Behavioral genetic studies typically examine twins and adoptees. They have been informative about how environmental and genetic causes affect mental test performance in old age. Mostly, they employ quite complex statistical modelling methods, but all are based on the facts that: monozygotic twins have 100% genetic similarity whereas dizygotic twins share, on average, 50%; and that adoptees share environmental but not genetic causes with their rearing family, and the reverse with their biological parents. Behavioral genetic studies have been used to explore whether genetic and environmental causes operate at the level of general intelligence (g) or specific cognitive domains, and have also explored causes of cognitive change as well as cognitive level within old age. It is important to repeat here that an association between a genetic variant and cognitive scores in old age is not the same as studying genetic contributions to cognitive ageing. Any association based on tests taken at a single time point might just represent an association with the life-long, stable trait of intelligence; to study ageing per se, it is important to show some association with individual differences in change across two or more occasions of measurement. Heritability of cognitive ageing The total variance (from genetic and environmental influences) in cognitive ability increases with age, possibly owing to stochastic effects. Although it might drop a little in advanced old age, there is also an increase in the percentage of this total variance that has a genetic component [7]. The contribution of twin studies to cognitive level and change in old age has been summarised extensively by Lee et al. [8]. As is found in earlier periods of life, the majority of the genetic variation is in general cognitive ability. The major cognitive domains also show high heritability, largely because they derive much of their variation from g. Memory in particular tends to have its own genetic causes in addition to those that derive from general cognitive ability [9]. Much of the information about genetic causes of ability differences in old age has come from various samples based on the Swedish Twin Registry. For example, the Swedish Adoption Twin Study of Ageing estimated the heritability of g at approximately 80% in the mid-60 s, and showed a high correlation between g loadings of tests and their heritability [10]. Longitudinal study of this sample tested 390

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

on multiple occasions over 13 years separated the genetic and environmental causes of the level and slope of cognitive functions [11]. The heritability of cognitive level was very high, with most of the rest being caused by unique and non-shared environmental effects. Whereas g still accounted for substantial genetic contributions to the cognitive domains, an interesting finding was that the heritability in fluid and/or spatial ability decreased somewhat with age, whereas that of memory increased. The slope of cognitive function with age had linear and quadratic aspects. The causes of differences in the former were almost entirely the influence of a unique environment, whereas the quadratic aspect (accounting for a much smaller proportion of variance than the linear effect) had some genetic influence. Analyses of the genetic effects on the four cognitive domains in the SATSA tests (verbal, spatial, memory and speed) over a period of up to 16 years found significant genetic influences on the intercepts, but not on the slopes [12]. However, genetic influences did affect cognitive slopes via intercept–slope correlations; that is, there were genetic influences on cognitive level, and individuals starting at different cognitive levels did not decline at the same rates. These complex analyses came to an important conclusion concerning genetic contributions to age-related cognitive changes: that genetic variance in processing speed is largely responsible for the variation that occurs at a later time point in other cognitive factors. Therefore, researchers interested in determining genetic factors that influence cognitive ageing should look for those that influence processing speed. The Swedish study with the oldest subjects (the OctoTwin Study) reported that the heritability of g was 62% at a median age of 82 years, with the heritability of cognitive domains being between 32% and 62% [13]. The non-genetic variance was caused by non-shared environmental factors. Further analysis of this sample found that the genetic contributions to the cognitive domains of verbal ability, spatial ability, speed of processing and memory were largely those that caused differences in g [9]. Similar results and conclusions were derived from studies of the Longitudinal Study of Aging Danish Twins [14– 16]. It should be stressed that the cognitive test battery in this study was small: the number of animals that could be named in 1 min; forward and backward digit span; immediate and delayed recall of a 12-word list. Even so, at initial assessment, when subjects had a mean age of 80 years and were all over 75 years, the additive genetic contribution to the cognitive composite score was 54%, with the remainder owing to non-shared environment, which is similar to that from the comparable Swedish study [13]. Follow-up of this sample to include four waves of cognitive testing showed a heritability of 75% for cognitive level but almost zero for cognitive change [15]. The heritability estimate for slope using data from a subsequent wave was 18%, but still non-significant: most of the contribution to change appeared to come from a non-shared environment [16]. Note that this higher heritability estimate for level is because level is a latent trait summarising four individual waves of testing, each of which had heritabilities typically of approximately 50%; the estimate was lower at a subsequent follow-up [16]. Note also that the total retest time

Review was no more than 6 years, which does not allow much time for any genetic or environmental factors to influence cognitive change. A US study of male twins between 69 and 80 years old reported a heritability of 79% for a latent trait of executive control [17]. The contributing tests to this trait (digit symbol, stroop, trail making and verbal fluency) make it likely to be a reasonable measure of g. In summary, the Swedish Twins-based and other studies have demonstrated that heritability of g and the major cognitive domains is still high in old age and that the heritability of the latter is due substantially to influences on the former. It is important to note that heritability studies have been more revealing for cognitive level than for cognitive change. Cognitive change tends to suffer from a poor phenotype: being studied over too-short periods of time with insufficient assessments [8]. Having established high heritability of cognitive level in old age, and with the jury still being out on genetic contributions to cognitive change, it has been important to seek the specific genes that contribute to cognitive differences in old age. Here, we consider several approaches, including candidate gene studies and GWAS. Several more recent promising approaches are summarised in Box 2. Candidate genes Several approaches have been taken to identify candidate genes to test for association with both cognitive ability in older people and age-related cognitive decline. Generally speaking, selected genes have previously been associated with age-related diseases, traits and mechanisms. Many of these genes have been implicated in cognitive ability and cognitive ageing, but few findings have been replicated. Selected examples of candidate genes, for which there are multiple positive associations, are given below. We first discuss candidate gene studies in the context of normal cognitive function as well as in neurocognitive and psychiatric disease. We then examine findings from two particular approaches, imaging genetics and oxidative stress, which might prove promising in the study of cognitive

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

ability and cognitive decline. A more detailed review of candidate gene studies is provided in [18]. Normal cognitive function As memory is particularly badly affected by ageing, genes previously associated with memory phenotypes have been widely studied with regard to cognitive decline. For example, brain-derived neurotrophic factor (BDNF) has been implicated in hippocampal-dependent learning and memory in humans and non-human species and there is a steady decline in BDNF expression associated with normal ageing [19]. A single functional common polymorphism has been identified in BDNF, a valine to methionine change at amino acid 66. The methionine allele has been implicated in abnormal hippocampal function, lower hippocampal volume and reduced cognitive function. Some, but not all studies have identified an association between BDNF and cognitive function in elderly subjects [18]. Catechol-O-methyl transferase (COMT) degrades the catecholamine neurotransmitters, dopamine, norepinephrine and epinephrine. Similar to BDNF, the gene contains a single common functional polymorphism. This is a valine to methionine substitution at position 158. The methionine allele reduces the thermostability and activity of the enzyme, thus reducing dopamine and other neurotransmitter degradation. COMT has been associated mainly with executive phenotypes in several studies of both younger and elderly individuals [18]. However, a meta-analysis failed to find strong evidence for an association between COMT and cognition, finding only suggestive evidence that methionine homozygotes score slightly higher on cognitive ability tests [20]. Neurocognitive disease The genetic contribution to Alzheimer’s disease (AD) has been widely studied. Genes associated with the rare earlyonset form of the disease include those encoding amyloid precursor protein (APP), presenilin 1 (PS1) and presenilin 2 (PS2), but none has been definitively associated with cognitive function or age-related cognitive decline in

Box 2. The future Exome and/or whole-genome sequencing GWAS are designed to detect, and have been successful at detecting, common genetic variation associated with diseases and traits. However, there is evidence to suggest that a substantial amount of the variation in complex traits, such as cognition and cognitive decline, is influenced by rare genetic variants, present in less than 1% of the population [56]. These variants can only be detected by directly sequencing the DNA of each individual in the study. Until recently, this had been prohibitively expensive. However, technological developments mean that exome (i.e. coding and regulatory region) sequencing is now feasible on the large numbers of individuals required to obtain the necessary power to detect associations between rare variants and complex traits [57]. Over the next few years, it is hoped that the cost of whole-genome sequencing will drop to a level that will make it possible too. This will allow genetic variants in all regions of the genome to be examined. Transcriptomics and proteomics Studying genetic variants in DNA does not directly inform one of their function. DNA is first transcribed into RNA, which is then translated into protein. The potential of genetic variants to alter RNA and protein

levels is of great interest. Where suitable tissue is available, it is now possible to identify genome-wide gene expression (RNA) levels and to correlate these directly with levels of cognitive ability and cognitive decline [58]. The biggest obstacle to this approach is that the most appropriate tissue (the brain) is inaccessible in living subjects. Another approach is to look directly at the abundance of the proteins under the control of multiple genetic variants by using methods such as mass spectrometry. A recent pilot study, looking at cognitive extremes in old age, concluded that this is a promising technique that will help in the understanding of the biological foundations of intelligence differences [59]. Methylomics Epigenetic mechanisms refer to DNA modifications that do not alter the sequence of the DNA. The best-known epigenetic mechanism is DNA methylation. Methylation mainly occurs at CpG islands that tend to be concentrated in gene promoters where transcription is blocked in the presence of a methyl group. Methylation states are known to change with age and be altered in certain age-related diseases [60]. Techniques are now available that allow the methylation status of > 450 000 loci covering > 14 000 genes to be determined [61].

391

Review

[(Figure_2)TD$IG]

non-demented individuals [21]. By contrast, the e4 allele of the APOE gene, associated with the more common lateonset AD (LOAD), was shown by a meta-analysis to be associated also with non-impaired cognitive function, particularly in elderly subjects [22]. e4 carriers performed significantly worse on measures of episodic memory, executive functioning, global cognitive functioning and perceptual speed. In a Scottish sample that was studied for cognitive change from age 11 to age 79 years, and then from age 79 to age 87, individuals with at least one e4 allele declined more in intelligence and memory functions across these two periods of the life course [23,24] (Figure 1). More recently, a poly-t repeat in the neighbouring translocase of outer mitochondrial membrane 40 (TOMM40) gene has been linked to the age of onset of LOAD [25]. Recent GWAS (Figure 2) identified several more genetic variants that increase an individual’s risk of developing LOAD in the genes bridging integrator 1 (BIN1), clusterin (CLU), complement component (3b/4b) receptor 1 (CR1) and phosphatidylinositol binding clathrin assembly protein (PICALM) [26–28]. However, reports on whether these variants are associated with cognitive function in non-demented elderly subjects are mixed [21,29,30].

[(Figure_1)TD$IG]

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Stage 1

Genotype up to 1 million SNPs in several thousand individuals

Stage 2

Genotype top few hundred SNPs in further individuals from the same population

Stage 3

Genotype SNPs that are significantly associated with trait in first population, in an independent population TRENDS in Cognitive Sciences

Figure 2. Genome-wide association study design. The exact number of individuals and single nucleotide polymorphisms (SNPs) typed at each stage will differ from study to study and will depend on the availability of subjects, effect size of individual SNPs and funds.

(a) 104 APOE e4-

IQ score

102

100

98 APOE e4+ 96

94 Age 11

Age 79

Logical memory score

(b) 38 36 34 32 Key:

30

APOE e4+ 28

APOE e4-

26 78

80

82

84

86

88

Age TRENDS in Cognitive Sciences

Figure 1. Apolipoprotein E (APOE) e4 status and ageing of cognition in the Lothian Birth Cohort of 1921: (a) general intelligence from age 11 years to age 79 years; (b) verbal declarative memory from age 79 years to age 87 years. Based on data from [23,24], respectively.

392

Psychiatric disease Cognitive deficits are present in many psychiatric diseases, including schizophrenia and bipolar disorder, and are also found at a higher rate in unaffected family members than in the general population [31]. Therefore, genes previously associated with psychiatric disease are good candidates for cognitive deficits in the absence of psychiatric disease. The gene Disrupted in Schizophrenia 1 (DISC1) was initially linked to psychiatric illness when it was identified at the breakpoint of a balanced translocation segregating within a large Scottish family suffering from multiple forms of mental illness [32]. There is some evidence implicating variation in DISC1 in cognitive function, in patients, unaffected family members and the general population. Elderly Scottish women homozygous for the cysteine allele of a non-synonymous single nucleotide polymorphism (SNP) in exon 11 had significantly lower cognitive ability scores than men, controlling for their childhood cognitive ability. This study suggests that variation in DISC1 affects cognitive ageing specifically in women [33]. However, the findings were not replicated in a second, younger Scottish cohort [34]. Distinct allelic haplotypes have been associated with psychotic and bipolar spectrum disorders along with cognitive impairments in a Finnish bipolar disorder family study [35]. Dystrobrevin binding protein 1 (DTNBP1), initially associated with schizophrenia, has been associated via metaanalysis with cognitive function in several cohorts of varying ages [36]. Imaging endophenotypes Associations between genetic variants and brain structure have been investigated, based on the assumption that individual differences in brain structure are good intermediate

Review phenotypes (so-called ‘endophenotypes’) for cognitive ability. For example, variation in adrenergic, beta-2-, receptor, surface (ADRB2), has been associated both with cognitive ability and white matter integrity (controlling for childhood ability) in a cohort of elderly Scots [37,38]. There was some evidence that white matter integrity mediated the association between ADRB2 and cognitive ability in old age. Mattay et al. [39] also reviewed genes identified by brain imaging genetic studies and related to systemic disease or cardiovascular function [angiotensin I converting enzyme (peptidyl-dipeptidase A) 1 (ACE) and methylenetetrahydrofolate reductase (NAD(P)H), (MTHFR)], or inflammatory processes [interleukin 1, beta (IL1B); tumor necrosis factor (TNF) and caspase 1, apoptosis-related cysteine peptidase (interleukin 1, beta, convertase), (CASP1; ICE)]. They concluded that neuroimaging techniques provide a promising way to identify genes associated with cognitive ageing. Oxidative stress One mechanism that has been used to identify candidate genes is oxidative stress [40]. Oxidative stress occurs during cellular respiration when the defence mechanisms of the cell fail to remove damaging by-products formed during respiration. These so-called ‘free-radicals’ can damage DNA and protein. Oxidative stress is believed to be responsible for many aspects of ageing, including cognitive decline. Therefore, anti-oxidant defence genes are good candidates for influencing cognitive decline. However, to date, no genetic variants within such genes have been definitively associated with either cognitive ability in old age or cognitive decline [40]. Summary of candidate gene studies Despite many studies being published, to date only APOE and possibly COMT, BDNF and DTNBP1 have repeatedly been associated with either cognitive ability in older people or cognitive decline [18,34]. It should also be noted that even when genetic variants are significantly associated with cognitive phenotypes, effect sizes are typically very small (1–2%). Genome-wide association studies As candidate gene studies have failed to identify genetic variants that account for much of the variance in cognitive ability or cognitive ageing, researchers have more recently turned to a hypothesis-free study design. GWAS allow multiple (up to 1 million) SNPs to be investigated in a single study (Figure 2) and have been used to identify genetic variation influencing many common diseases and traits (e.g. http://www.genome.gov/gwastudies; http:// www.gwascentral.org). A small-scale GWAS failed to identify common genetic variants associated with cognitive ability [41,42], although other GWAS have implicated WW and C2 domain containing 1 (WWC1; KIBRA); calmodulin binding transcription activator 1 (CAMTA1) and sodium channel, voltage-gated, type I, alpha subunit (SCN1A) with memory phenotypes [43–45]. A recent GWAS, which included 3500 older individuals, identified no single genetic variant associated with fluid and crystallised intelligence [46]. However, it

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Box 3. Questions for future research Given sufficiently good phenotypic data on age-related cognitive changes, how heritable are these phenotypes? What are the genetic variants that influence cognitive ability and cognitive ageing and can they be identified by deep sequencing individual genomes? What are the gene expression differences at both the RNA and protein level that influence cognitive ability and cognitive ageing? Do epigenetic mechanisms influence cognitive ability and cognitive ageing?

concluded that approximately 50% of the variation in cognitive ability in later life is accounted for by multiple genetic variants in linkage disequilibrium (LD) with common SNPs, each having a very small effect. Concluding remarks In summary, despite behavioural genetic studies consistently showing that the heritability of cognitive ability in old age is high, very few specific genetic variants that influence cognitive ability and cognitive ageing have been identified. To date, the majority of studies have focussed on the identification of relatively common genetic variants that influence these traits. With the introduction of largescale whole-genome sequencing, it is hoped that multiple rare variants influencing cognitive ability and cognitive ageing will be identified. By investigating gene expression levels, researchers will be also able to identify the effects of multiple genetic variants on gene function. Finally, technological advances now allow epigenetic mechanisms to be investigated on a genome-wide scale. This might lead to new insights into mechanisms that influence variation in cognitive ability and cognitive function (Box 3). Acknowledgements The work was undertaken by The University of Edinburgh Centre for Cognitive Ageing and Cognitive Epidemiology, part of the cross council Lifelong Health and Wellbeing Initiative (G0700704/84698). Funding from the Biotechnology and Biological Sciences Research Council (BBSRC), Engineering and Physical Sciences Research Council (EPSRC), Economic and Social Research Council (ESRC) and Medical Research Council (MRC) is gratefully acknowledged.

References 1 Deary, I.J. et al. (2009) Age-associated cognitive decline. Br. Med. Bull. 92, 135–152 2 Plassman, B.L. et al. (2010) Systematic review: factors associated with risk for and possible prevention of cognitive decline in later life. Ann. Intern. Med. 153, 182–193 3 Salthouse, T.A. (2004) Localizing age-related individual differences in a hierarchical structure. Intelligence 32, 541–561 4 Schaie, K.W. (2005) Developmental Inﬂuences on Adult Intelligence, Oxford University Press 5 Carroll, J.B. (1993) Human Cognitive Abilities, Cambridge University Press 6 Lezak, M. et al. (2004) Neuropsychological Testing, (4th edn), Oxford University Press 7 Deary, I.J. et al. (2006) Genetics of intelligence. Eur. J. Hum. Genet. 14, 690–700 8 Lee, T. et al. (2010) Genetic influences on cognitive functions in the elderly: a selective review of twin studies. Brain Res. Rev. 64, 1–13 9 Petrill, S.A. et al. (1998) The genetic and environmental relationship between general and specific cognitive abilities in twins age 80 and older. Psychol. Sci. 9, 183–189 10 Pedersen, N.L. et al. (1992) A quantitative genetic analysis of cognitive abilities during the second half of the lifespan. Psychol. Sci. 3, 346–353 393

Review 11 Reynolds, C.A. et al. (2005) Quantitative genetic analysis of latent growth curve models of cognitive abilities in adulthood. Dev. Psychol. 41, 3–16 12 Finkel, D. et al. (2009) Genetic variance in processing speed drives variation in aging of spatial and memory abilities. Dev. Psychol. 45, 820–834 13 McClearn, G.E. et al. (1997) Substantial genetic influence on cognitive abilities in twins 80 or more years old. Science 276, 1560–1563 14 McGue, M. and Christensen, K. (2001) The heritability of cognitive functioning in very old adults: evidence from Danish twins aged 75 years and older. Psychol. Aging 16, 272–280 15 McGue, M. and Christensen, K. (2002) The heritability of level and rate-of-change in cognitive functioning in Danish twins age 70 years and older. Exp. Aging Res. 28, 435–451 16 McGue, M. and Christensen, K. (2007) Social activity and healthy aging: a study of aging Danish twins. Twin Res. Hum. Genet. 10, 255–265 17 Swan, G.E. and Carmelli, D. (2002) Evidence for genetic mediation of executive control: a study of aging male twins. J. Gerontol. Psychol. Sci. 57B, 133–143 18 Payton, A. (2009) The impact of genetic research on our understanding of normal cognitive ageing: 1995 to 2009. Neuropsychol. Rev. 19, 451–477 19 Tapia-Arancibia, L. et al. (2008) New insights into brain BDNF function in normal aging and Alzheimer disease. Brain Res. Rev. 59, 201–220 20 Barnett, J.H. et al. (2008) Meta-analysis of the cognitive effects of the catechol-O-methyltransferase gene Val158/108Met polymorphism. Biol. Psychiatry 64, 137–144 21 Hamilton, G. et al. (2011) Alzheimer’s disease genes are associated with measures of cognitive ageing in the Lothian Birth Cohorts of 1921 and 1936. Int. J. Alzheimer’s Dis. DOI: 10.4061/2011/505984 22 Wisdom, N.M. et al. (2009) The effects of apolipoprotein E on nonimpaired cognitive functioning: a meta-analysis. Neurobiol. Aging 32, 63–74 23 Deary, I.J. et al. (2002) Cognitive change and the APOE e4 allele. Nature 418, 932 24 Schiepers, O.J.G. et al. (2011) APOE E4 status predicts age-related cognitive decline in the ninth decade: longitudinal follow-up of the DOI: 10.1038/ Lothian Birth Cohort 1921. Mol. Psychiatry mp.2010.137 25 Roses, A.D. et al. (2010) A TOMM40 variable-length polymorphism predicts the age of late-onset Alzheimer’s disease. Pharmacogenomics J. 10, 375–384 26 Harold, D. et al. (2009) Genome-wide association study identifies variants at CLU and PICALM associated with Alzheimer’s disease. Nat. Genet. 41, 1088–1093 27 Lambert, J.C. et al. (2009) Genome-wide association study identifies variants at CLU and CR1 associated with Alzheimer’s disease. Nat. Genet. 41, 1094–1099 28 Seshadri, S. et al. (2010) Genome-wide analysis of genetic loci associated with Alzheimer disease. JAMA 303, 1832–1840 29 Mengel-From, J. et al. (2011) Genetic variations in the CLU and PICALM genes are associated with cognitive function in the oldest old. Neurobiol. Aging 32, 554.e7–11 30 Chibnik, L.B. et al. (2011) CR1 is associated with amyloid plaque burden and age-related cognitive decline. Ann. Neurol. 69, 560– 569 31 Harvey, P.D. et al. (2001) Cognition in schizophrenia. Curr. Psychiatry Rep. 3, 423–428 32 Millar, J.K. et al. (2000) Disruption of two novel genes by a translocation co-segregating with schizophrenia. Hum. Mol. Genet. 9, 1415–1423 33 Thomson, P.A. et al. (2005) Association between genotype at an exonic SNP in DISC1 and normal cognitive aging. Neurosci. Lett. 389, 41–45 34 Houlihan, L.M. et al. (2009) Replication study of candidate genes for cognitive abilities: the Lothian Birth Cohort 1936. Genes Brain Behav. 8, 238–247

394

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 35 Palo, O.M. et al. (2007) Association of distinct allelic haplotypes of DISC1 with psychotic and bipolar spectrum disorders and with underlying cognitive impairments. Hum. Mol. Genet. 16, 2517–2528 36 Zhang, J.P. et al. (2010) Meta-analysis of genetic variation in DTNBP1 and general cognitive ability. Biol. Psychiatry 68, 1126–1133 37 Bochdanovits, Z. et al. (2009) A functional polymorphism under positive evolutionary selection in ADRB2 is associated with human intelligence with opposite effects in the young and the elderly. Behav. Genet. 39, 15–23 38 Penke, L. et al. (2010) White matter integrity in the splenium of the corpus callosum is related to successful cognitive aging and partly mediates the protective effect of a ancestral polymorphism in ADRB2. Behav. Genet. 40, 146–156 39 Mattay, V.S. et al. (2008) Neurobiology of cognitive aging: insights from imaging genetics. Biol. Psychol. 79, 9–22 40 Harris, S.E. et al. (2007) A genetic association analysis of cognitive ability and cognitive ageing using 325 markers for 109 genes associated with oxidative stress or cognition. BMC Genet. 8, 43 41 Need, A.C. et al. (2009) A genome-wide study of common SNPs and CNVs in cognitive performance in the CANTAB. Hum. Mol. Genet. 18, 4650–4661 42 Cirulli, E.T. et al. (2010) Common genetic variation and performance on standardized cognitive tests. Eur. J. Hum. Genet. 18, 815–820 43 Papassotiropoulos, A. et al. (2006) Common Kibra alleles are associated with human memory performancer. Science 314, 475–478 44 Huentelman, M.J. et al. (2007) Calmodulin-binding transcription activator 1 (CAMTA1) alleles predispose human episodic memory performance. Hum. Mol. Genet. 16, 1469 45 Papassotiropoulos, A. et al. (2011) A genome-wide survey of human short-term memory. Mol. Psychiatry 16, 184–192 46 Davies, G. et al. Genome-wide association studies establish that human intelligence is highly heritable and polygenic. Mol. Psychiatry, doi:10.1038/mp.2011.85 47 Johnson, W. et al. (2007) Genetic and environmental influences on the verbal-perceptual-image rotation (VPR) model of the structure on mental abilities in the Minnesota study of twins reared apart. Intelligence 35, 542–562 48 Johnson, W. et al. (2008) Still just 1 g: consistent results from five test batteries. Intelligence 36, 81–95 49 Deary, I.J. et al. (2009) Genetic foundations of human intelligence. Hum. Genet. 126, 215–232 50 Deary, I.J. et al. (2010) The neuroscience of human intelligence differences. Nat. Rev. Neurosci. 11, 201–211 51 Lyons, M.J. et al. (2009) Genes determine stability and the environment determines change in cognitive ability during 35 years of adulthood. Psychol. Sci. 20, 1146–1152 52 Deary, I.J. et al. (2000) The stability of individual differences in mental ability from childhood to old age: follow-up of the 1932 Scottish Mental Survey. Intelligence 28, 49–55 53 Gow, A.J. et al. (2011) Stability and change in intelligence from age 11 to ages 70, 79, and 87: the Lothian Birth Cohorts of 1921 and 1936. Psychol. Aging 26, 232–240 54 Tucker-Drob, E.M. (2011) Global and domain-specific changes in cognition throughout adulthood. Dev. Psychol. 47, 331–343 55 Wilson, R.S. et al. (2002) Individual differences in rates of change in cognitive abilities of older persons. Psychol. Aging 17, 179–193 56 Bodmer, W. and Bonilla, C. (2008) Common and rare variants in multifactorial susceptibility to common diseases. Nat. Genet. 40, 695–701 57 Teer, J.K. and Mullikin, J.C. (2010) Exome sequencing: the sweet spot before whole genomes. Hum. Mol. Genet. 19 (R2), R145–R151 58 Geschwind, D.H. and Konopka, G. (2009) Neuroscience in the era of functional genomics and systems biology. Nature 461, 908–915 59 Lopez, L.M. et al. (2011) A pilot study of urinary peptides as biomarkers for intelligence in old age. Intelligence 39, 46–53 60 Calvanese, V. et al. (2009) The role of epigenetics in aging and agerelated diseases. Ageing Res. Rev. 8, 268–276 61 Bibikova, M. and Fan, J.B. (2009) Genome-wide DNA methylation profiling. Wiley Interdiscip. Rev. Syst. Biol. Med. 2, 210–223

Review

Special Issue: The Genetics of Cognition

Dissecting the genetic architecture of human personality Marcus R. Munafo`1 and Jonathan Flint2 1 2

School of Experimental Psychology, University of Bristol, 12a Priory Road, Bristol BS8 1TU, United Kingdom Wellcome Trust Centre for Human Genetics, University of Oxford, Roosevelt Drive, Oxford OX3 7BN, United Kingdom

The first candidate gene studies of human personality promised much but, in the fifteen years since their publication, have delivered little in the way of clear evidence for the contribution of specific genetic variants to observed variation in personality traits. This is most likely due to the very small effects conferred by individual loci. The advent of genome-wide association studies has brought growing awareness that high levels of statistical stringency, very large sample sizes, and independent replication will be minimum requirements for future genetic studies of personality. At the same time, evidence from other fields indicates that the genetic architecture of personality is likely to consist of the combined effect of many hundreds, if not thousands, of small effect loci.

many conﬂicting claims that have proved so difﬁcult to substantiate, and argue that progress requires greater stringency in study design and the use of much larger sample sizes for robust discovery and replication of genetic association ﬁndings. In particular, we must be prepared for the likelihood that human personality is inﬂuenced by a very large number (possibly thousands) of genetic variants, of individually small effect.

The study of human personality: from traits to genes Thirty years ago the subject matter of personality research was the structure of personality: determining the number of personality traits and their relation to each other and to psychological and biological characteristics. Once it was accepted that self-assessment, using questionnaire instruments, yielded acceptable levels of consistency and prediction [1,2], subsequent research focused on the number of factors that explained behavioural consistency and personality variation, typically extracted from an analysis of questionnaire responses [3]. More recently, the literature has attempted to identify the genes that contribute to observed variation in individual differences in personality, given clear evidence from twin, family and adoption studies for a substantial heritable component to personality (see, for example, [4]). This has proved no less contentious than the earlier discussion about the number of dimensions needed to describe personality. Much of the current direction in personality genetics grew out of interest in testing the theories that gave rise to personality models. Understanding this history may provide clues about current difﬁculties in personality genetics, and suggest potential solutions (Box 1). Here we begin by reviewing genetic association studies of candidate genes (see Glossary) and the results of studies that have looked at the interaction between genes and environment in shaping personality. We then look at the results of more recently introduced hypothesis-free genetic screens: genome-wide association studies and genetic linkage studies. We consider why this ﬁeld has accumulated so

Candidate gene study: Candidate genes are those thought to be involved in a phenotype based on neurobiological or other theories of personality. With a candidate gene in hand, it is now straightforward to use publicly accessible databases to identify sequence variants within the gene, which ideally should result in some change in biological function, and then to determine whether these are correlated with personality variation. Complex trait: A trait is complex when its genetic contribution does not fit simple Mendelian patterns of inheritance (i.e., recessive, dominant or Xlinked). Typically, this means there is evidence for the contribution of multiple loci. The term usually also implies that there is an important environmental contribution. Almost all behavioural traits are complex in this sense. Epistasis: In statistical genetics, epistasis (gene gene interaction) occurs when the joint effect of two (or more) variants differs from the sum of their individual, independent effects. Epistasis in this sense is an interaction term in a linear regression model. Epistasis also has a biological interpretation, where a mutation at one locus masks the effect of another. This would happen, for example, in a biochemical pathway where a mutation would block the production of an intermediate needed for a subsequent reaction. Gene environment interaction: This is perhaps best thought of as genetic susceptibility to the environment. Statistically, a gene by environment interaction (G E) occurs when phenotypic variation is not predicted by the sum of independent environmental and genetic effects. For example, exposure to the same stressful life event could produce very different outcomes in two individuals because they differ in their genetic constitution. Genome-wide association study (GWAS): A genetic association is the statistically significant correlation between allelic and phenotypic variation. Genome-wide association means testing for the presence of genetic association throughout the genome (simply put, testing whether variation in any of the 20,000 or so genes that we all have might contribute to disease susceptibility or trait variation). Linkage study: This employs a within-family design to determine which parts of the genome contribute to variation in a trait or disease susceptibility. Linkage studies set out to find co-segregration of genetic markers and phenotypes in families. Chromosomal regions within which a gene contributing to variation in the phenotype of interest may lie are identified, rather than functional genetic variants themselves. Polymorphism: Variations in DNA sequences are called polymorphisms, and may or may not be functional (that is, they may or may not result in some difference in biological function). When a single base pair is polymorphic, for example such that one chromosome has base ‘T’ and another chromosome has base ‘G’ at the same place (or locus, to use genetic terminology) in the genome, the polymorphism is called a single nucleotide polymorphism or SNP.

Corresponding author: Munafo`, M.R. ([email protected]).

Candidate gene studies Some personality theories suggest the possible involvement of serotoninergic and dopaminergic neurotransmitter Glossary

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.007 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

395

Review Box 1. Personality Models Studies carried out in the mid-1950s were the first to propose a fivefactor framework of personality [46]. This framework is now more or less widely accepted, bolstered by many publications using the NEO Personality Inventory (NEO-PI), a set of scales that can accommodate other questionnaires, such as those developed by Eysenck [47]. The five factors derived from the NEO-PI are openness, conscientiousness, extraversion, agreeableness and neuroticism, sometimes referred to by the acronym OCEAN. However, OCEAN is not the only five-factor model – Zuckerman, for example, introduced his own [48]. Nor is five factors the universally agreed solution – Cloninger, for example, proposed seven factors [49] and Eysenck three [50,51]. These differ in some other respects – Eysenck proposed three very broad ‘superfactors’, which subsumed a number of lower-order trait factors, while Cloninger proposed a hierarchical system with broad factors lying above narrow, more specific factors below. The frameworks proposed by Eysenck, Gray and Cloninger are based on neurobiological theories of the origin of personality. All have their basis, to some extent, in learning theory. Cloninger, for example, defines three higher-order dimensions of personality on the basis of the stimulus-response characteristics of novelty seeking, harm avoidance, and reward dependence. However, the personality dimensions they identify are not necessarily the same as those in other frameworks. Comparison of the Eysenck Personality Questionnaire (EPQ) scales and Costa and McCrae’s NEO inventory reveals that the neuroticism and extraversion factors from the two systems match well [47]. There is less overlap with Cloninger’s Tridimensional Personality Questionnaire (TPQ) scales; harm avoidance correlates 0.59 with neuroticism and novelty seeking 0.44 with extraversion[52]. Further evidence that these classificatory systems are not necessarily just alternative descriptions of the same dimensions of personality comes from twin studies. These show, for example, that Eysenck’s and Cloninger’s dimensions that ‘‘purport to describe the structure of personality in terms of three major dimensions appear instead to jointly assess five or six dimensions of genetic variability and at least six dimensions of environmental variability’’ [53]. Therefore, the identification of genetic influences on personality (and the success of this enterprise) may depend in part on the personality instrument chosen. While it should not be assumed that heritability of a personality trait implies a biological correlate (correlations between genotype and phenotype do not necessarily reflect straightforward variation in a biological process), discovering that variation in personality is associated with variation in a neurotransmitter system would support frameworks arising from relevant neurobiological theories. For example, Cloninger’s claim that harm avoidance is correlated with activity of mesolimbic serotonergic neurons would benefit from finding that genetic variants in serotoninergic genes contribute to variation in personality.

systems in speciﬁc traits (Box 1), which has provided a series of candidate genes for testing. A large literature has developed, brieﬂy reviewed below, with contradictory claims for and against the involvement of these genes emerging over time. Perhaps in part through disappointment with the search for clear individual genetic effects, researchers have examined the possibility that genetic effects might only emerge when explored together with environmental effects. We consider below whether both must be examined in order to detect association. Direct effects on personality Three key publications in 1996 [5–7] investigated whether variation in speciﬁc candidate genes (identiﬁed on the basis of their involvement in relevant neurotransmitter systems) was associated with personality. These were the serotonin transporter (SLC6A4) gene with measures of 396

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

neuroticism [6], and the dopamine D4 receptor (DRD4) gene with novelty seeking [5,7]. Each reported a positive result, and gave rise to the hope that molecular genetic studies were about to open a new chapter in the study of personality. Fifteen years later, it is worth considering how much of this early promise has been realized. Rather than progressing to a deeper understanding of the biology of personality, moving from genetic variant, to gene, to the neurons and neuronal circuitry in which activity constitutes temperament, debate continues as to whether these genes are involved at all and, if they are, what their individual contribution is towards accounting for the roughly 50% heritability of personality traits indicated by twin, adoption and family studies (see, for example, [4]). First, there is still no clear consensus about the involvement of any genetic variant in personality. Studies abound, varying in the degree to which they replicate each other, but combining these using meta-analytic techniques has only resulted in the conclusion that, if the claimed effects are present at all, they must be very small [8–12]. Second, the literature is still dominated by reports on a small number of candidate genes, centered around the function of two neurotransmitters: serotonin and dopamine. By far the most widely-studied candidates are SLC6A4 [11], DRD4 [12] and catechol-O-methyltransferase (COMT) [13,14]. Interestingly, DRD2, while a favourite in the wider psychiatric genetics literature, has not gained purchase in the personality genetics literature. Given that these candidates are identiﬁed on the basis of known (or presumed) neurobiology, conﬁrmatory evidence that they play a role in personality does not seem to provide much more information beyond, for example, ascertaining that serotonergic neurotransmission plays a role in neuroticism. Gene environment interaction studies Faced with the same stressful situation, different people react differently. It is uncontroversial to assert that this variation has, in part, a genetic origin. However, while few doubt that this interplay of genes and environment (gene environment interaction, or G E) exists, its importance has been difﬁcult to assess. The older literature was not optimistic: Jinks and Fulker [15], using the correlation between identical twin intra-pair differences and pair sums, found little evidence for G E effects for cognitive and personality traits. More recently, the tide has turned. In an inﬂuential article from 1994, Bronfenbrenner and Ceci argued strongly that interaction needs to be taken into account in behavioural genetic studies[16]. Could it be that molecular variants will not be found unless G E is taken into account? It may be that modelling the joint effects of genes and environment is necessary to obtain sufﬁcient statistical power to detect the effect. There are several environmental effects to which individuals may be more or less sensitive, such as parental behaviour [17] and stressful life events [18]. Unfortunately, there are few comparable G E studies which have focused on personality as an outcome. However, we can perhaps learn from the parallel literature on psychiatric outcomes. Attempts to ﬁnd G E effects in the wider behaviour genetics literature have met the same problems

Review faced by candidate gene studies [19–21]. A pattern of ﬁndings has emerged similar to that observed in studies of the direct effects of candidate genes on personality: initial positive ﬁndings, followed by a series of inconsistent results, typically reporting effects considerably smaller than suggested by early publications [22]. In addition, G E studies have brought their own, unique, difﬁculties. Two are worth highlighting. The ﬁrst difﬁculty is sample size. Assume a study is carried out searching for the main effect of a genetic variant on a personality trait, which then asks whether there is also an interaction effect. If the interaction is of the same magnitude as the main effect, the sample size must be increased fourfold (to maintain statistical power at the same level) [23]. However, this ‘‘increase[s] dramatically to 100 or greater for more subtle interactions that [are] smaller than 20 per cent of the overall effect’’ [23]. The second difﬁculty is that, once an interaction has been reported, replication does not consist of merely identifying another signiﬁcant interaction for the same environment and genotype. A statistically signiﬁcant interaction can arise from different combinations of factors, some of which are mutually inconsistent. For example, if the ﬁrst study ﬁnds that allele A, in the presence of an environmental stressor, increases a personality measure, but study B ﬁnds that the two combine to decrease the measure, both ﬁndings could be statistically signiﬁcant but one does not replicate the other [20,21]. Unfortunately quite loose deﬁnitions of ‘replication’ are commonplace. Moreover, similar issues exist for the study of gene gene interactions. Genome-wide association studies A few years ago, the state of personality genetic research was comparable to that of other complex traits, such as obesity, type-2 diabetes, and high blood pressure. All were in a similarly parlous condition, each sporting a large literature of claims and counterclaims, until consensus emerged that signals had to exceed a genome-wide threshold for statistical signiﬁcance (of the order of p < 10 7). This cultural change was imposed by the multiple testing inherent in the design of genome-wide association studies (GWAS), which test at least 500,000 loci (and now, more commonly, in excess of 2 million loci). Given a parallel consensus that the effects of common genetic variants on these traits were likely to be small (and much smaller than originally anticipated), it became clear that very large sample sizes would be necessary to reliably detect real genetic effects. For a common quantitative phenotype such as height, each locus, robustly identiﬁed and replicated, explains much less than 0.5% of the phenotypic variance [24]. The Wellcome Trust Case Control Consortium (WTCCC), using 2,000 cases and 3,000 controls, identiﬁed loci for only ﬁve of seven investigated traits [25]. The WTCCC study found no loci that contributed to variation in blood pressure; that required the analysis of 34,433 subjects followed up by genotyping an additional 81,114 subjects [26,27]. Moreover, the WTCCC pointed out that ‘‘for subsamples of 1,000 cases and 1,000 controls, of the 16 loci detected in the full study, we would have been certain of seeing only 2’’ [25]. In other words, the loss of power incurred by reducing

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

the sample size from 3,000 to 1,000 would result in failure to detect almost 90% of the loci. Success here is relative: of the ten loci reported each accounted for between 0.05 to 0.1% of the variance, in total explaining about 1% of the variance. Applying similar stringency to the p-values obtained in the personality genetics literature would remove all ‘signiﬁcant’ ﬁndings at a stroke. In general, personality genetics researchers, still focusing on candidate genes, continue to consider that the results of analyzing a single variant can be considered signiﬁcant if the p-value is less than 0.05. While it is not necessary to insist on genomewide signiﬁcance for a candidate gene study (since by choosing a candidate one is hopefully increasing the prior odds of a genuine association existing), it is necessary to consider how many other genes might conceivably inﬂuence personality. Since this number is presumably large (with, say, at least 10,000 genes expressed in brain), at least some level of enhanced stringency is required to accommodate the implicit (and sometimes explicit) problem of multiple statistical testing. However, this is very rarely done in practice. The effect sizes of loci contributing to variation in personality could be even smaller than those found for blood pressure. All available data point in this direction: as noted above, meta-analyses of the direct association between phenotype and genotype have typically excluded effects greater than 1% [8,9,11,12]. Four genome-wide association studies [28–31] have failed to observe any locus with clear genome-wide signiﬁcance, even when samples in excess of 10,000 individuals are used [31]. Critically, none of the candidate genes from the early personality genetic literature (SLC6A4 and DRD4) have shown evidence of association with any personality traits in these genomewide studies. While the sample sizes were still modest compared to those for blood pressure, the negative results are clear, and can exclude the possibility of individual loci explaining any more than 1% of phenotypic variance. Linkage studies [32–37] deliver exactly the same message: the genetic effects contributing to variation in personality are many and very small (see Box 2). To date, the largest sample size used for genotyping in personality genetics has comprised 88,000 people [38]. In this study, the extremes of the distribution were genotyped (about 3,000 individuals) for the 5-HTTLPR variant of the serotonin transporter gene, which is equivalent to genotyping a total population of about 50,000. No evidence of association with neuroticism or related measures of depression was observed. Yet even this number is too small if the true effect sizes are comparable to those that contribute to blood pressure variation. Assuming a locus accounts for 0.05% of the variance, a sample size of 10,000 has less than 20% power to detect it, at a signiﬁcance threshold of 0.05. A sample size of 50,000 has about 50% power, and sample sizes of 100,000 are needed to obtain power of more than 80%. The lesson from successful complex trait analyses is that the power of a study is more important than the statistical signiﬁcance of its results. Statistical theory demonstrates that nominally signiﬁcant ﬁndings from underpowered studies are likely to be false positives 397

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Box 2. Linkage Studies

Box 3. Questions for Future Research

In family-based linkage studies of personality, about 500 polymorphic markers across the genome are used to identify chromosomal regions that are more similar than expected among siblings with similar personalities. With one exception [33], all sibling pair linkage studies have only studied neuroticism. Two studies collected large community samples (34,371 individuals in one [35], and 88,141 in the other [32]), and then genotyped the extremes of the distribution; one study similarly analysed extremes of two large twin samples, one from Australia and one from the Netherlands [37]; one studied adolescents (measuring personality twice, at ages 14 and 16) [33], one ascertained siblings for concordance on nicotine dependence [36], and another ascertained siblings for concordance on alcohol [34]. Given the evidence for different genetic effects in males and females, some studies, but not all [36], also tested for the presence of sex specific loci. Across these studies, very few regions exceeded a 5% genomewide significance level; at most, six regions have been identified [32], and, with such weak linkage signals, establishing replication is difficult. Two studies found peaks on chromosome 12 [32,33], but there are differences in sample collection (the former recruited subjects over the age of 30, the latter subjects under 17); the 95% confidence intervals of a third study [34] overlap with the peak on chromosome 12 [32], but it remains unclear whether this is coincidence. Linkage studies have therefore failed to conclusively identify regions within which genetic variants contributing to personality may lie.

What is the genetic architecture of personality? An ongoing discussion in complex trait analysis concerns the number of loci involved and their effect sizes. Could there be a relatively small number of variants with large effects or is it all due to a very large number (probably thousands) of variants with small effects? The results of GWAS and linkage studies practically exclude common variants of large effect, but it remains possible that a small number of rare variants (occurring in less than 1% of the population) with large effect (contributing to 5% or more of the phenotypic variation) might be identified [44,54]. How should studies incorporate environmental effects? The dispute over G E has focused on the appropriate study designs for the detection of interaction. One possibility is to acquire samples specifically tailored in advance to find such interactions, for example by pre-selecting groups on the basis of exposure to different environmental stressors, or genotype, or both. Such studies will also require very stringent criteria for declaring statistical significance and replication. Are there new molecular technologies that might shed new light on personality genetics? Without doubt, the most important new development in the past few years has been the emergence of next generation sequencing technologies, which have made population scale sequencing practical as an alternative to genotyping [55]. This methodology makes it possible to detect rare variants, including types of sequence variant that are not accessible to genotyping platforms [56]. Completion of the first phase of the 1,000 Genomes project permits imputation of lowfrequency variants, further increasing the power of sequencing to access genomic variation [57]. These advances mean that sequencing large numbers of individuals is now feasible in the search for variants contributing to personality variation. How should false positives and negatives be dealt with? Many of the problems we have discussed in this article are due to the way results are published: positive findings are more likely to be published in high profile journals, while negative findings less likely to be published at all. If authors submitted, and editors accepted, more negative reports, publication bias would be reduced. At the moment, the rewards for publishing encourage false positives. We should all work to change this structure.

[39,40]. This is because the false positive rate (i.e., the alpha, or signiﬁcance level) will remain constant, while the true positive rate will decrease as statistical power decreases, so that the ratio of true positives to false positives among studies that achieve statistical signiﬁcance will decrease with decreasing power. Since we now know that individual genetic effects on personality is very small, we also know that the vast majority of studies to date are underpowered. Furthermore, it has been suggested for G E studies that low power and low prior probability in favour of interaction effects involving a speciﬁc candidate gene make it likely that most, if not all, of these ﬁndings are false positives [41]. Greater stringency in study design, by means of a higher threshold for declaring statistical signiﬁcance, the use of considerably larger sample sizes, and the requirement for results from a replication sample to be published alongside the original discovery sample, will be necessary if genuine advances in our understanding of personality genetics are to be made [22]. Concluding remarks Our brief overview of personality genetics has highlighted a number of methodological issues, which have resulted in a series of inconsistent ﬁndings and no clear consensus regarding the role of any individual gene in personality variation. What is the way forward (see also Box 3)? One suggestion emerges from the International Schizophrenia Consortium analysis of GWAS data [42]. While very few individual loci achieve conventional genome-wide signiﬁcance, this group has found that there is predictive information even in markers with p-values up to 0.5. This was shown by using results from one GWAS to predict the results in another, independent sample [42]. Wray and Visscher point out that, even with 10,000 case subjects and 10,000 controls, power to detect a variant with a relative risk of 1.05 and a frequency of 0.2, at a low threshold of 398

1 10 6, is only 0.2% [43]. However, they then go on to argue that power in the totality of published schizophrenia GWAS analyses (containing only a few thousand subjects) is such that, 72% of the time, variants of this effect size will feature in the top half of the list of all results. They showed that these single nucleotide polymorphism (SNP) sets were predictive of case-control status [42] and, most importantly, that genotyped SNPs accounted for about a third of the variance in liability. In other words, the genetic architecture of schizophrenia consists of the combined effect of many hundreds, if not thousands, of small effect loci – a model sometimes referred to as quasi-inﬁnitesimal. Small effects dominate, and are consistent with available data on the genetic architecture of complex traits [44]. This ﬁnding suggests that the genetic architecture of schizophrenia is, in large part, simply a very long tail of inﬁnitesimal effects, which together nevertheless produce substantial heritability. The same is likely to be true for personality. If this insight is correct, then the successful genetic dissection of behavioural and cognitive phenotypes will need much larger sample sizes than currently considered – perhaps consisting of hundreds of thousands of individuals. This will obviously be an onerous task, and one that any single research group working in isolation will be unable to achieve. While current genotyping costs make

Review it difﬁcult to see how such studies would be funded, molecular technologies continue to improve and costs fall. The ﬁrst population-scale genome sequencing has recently been achieved [45], and it is not inconceivable that within the next ﬁve years case-control samples of tens of thousands will routinely be subject to genome sequencing, with even larger samples to follow soon after. At that point we may, at last, begin to understand the molecular basis of personality. Acknowledgements MRM is a member of the UK Centre for Tobacco Control Studies, a UKCRC Public Health Research Centre of Excellence. Funding from the Economic and Social Research Council, the British Heart Foundation, Cancer Research UK, the Department of Health and the Medical Research Council, under the auspices of the UK Clinical Research Collaboration, is gratefully acknowledged. JF is supported by the Wellcome Trust.

References 1 Goldberg, L.R. and Werts, C.E. (1966) The reliability of clinician’s judgments: a multitrait-multimethod approach. J. Consult. Psychol. 30, 199–206 2 Kenrick, D.T. and Stringﬁeld, D.O. (1980) Personality-traits and the eye of the beholder - crossing some traditional philosophical boundaries in the search for consistency in all of the people. Psychol. Rev. 87, 88–104 3 Funder, D.C. (2001) Personality. Annu. Rev. Psychol. 52, 197–221 4 Floderus-Myrhed, B. et al. (1980) Assessment of heritability for personality, based on a short-form of the Eysenck personality inventory: A study of 12,898 twin pairs. Behav. Genet. 10, 153–162 5 Benjamin, J. et al. (1996) Population and familial association between the D4 dopamine receptor gene and measures of novelty seeking. Nat. Genet. 12, 81–84 6 Lesch, K.P. et al. (1996) Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science 274, 1527–1531 7 Malhotra, A.K. et al. (1996) The association between the dopamine D-4 receptor (D4DR) 16 amino acid repeat polymorphism and Novelty Seeking. Mol. Psychiatry 1, 388–391 8 Munafo`, M.R. et al. (2005) Does measurement instrument moderate the association between the serotonin transporter gene and anxietyrelated personality traits?. A meta-analysis. Mol. Psychiatry 10, 415–419 9 Munafo`, M.R. et al. (2003) Genetic polymorphisms and personality in healthy adults: A systematic review and meta-analysis. Mol. Psychiatry 8, 471–484 10 Munafo`, M.R. and Flint, J. (2004) Meta-analysis of genetic association studies. Trends Genet. 20, 439–444 11 Munafo`, M.R. et al. (2009) 5-HTTLPR genotype and anxiety-related personality traits: A meta-analysis and new data. Am. J. Med. Genet. B: Neuropsychiatr. Genet. 150B, 271–281 12 Munafo`, M.R. et al. (2008) Association of the dopamine D4 receptor (DRD4) gene and approach-related personality traits: Meta-analysis and new data. Biol. Psychiatry 63, 197–206 13 Chen, C. et al. (2011) Sex modulates the associations between the COMT gene and personality traits. Neuropsychopharmacology 36, 1593–1598 14 Calati, R. et al. (2011) Catechol-o-methyltransferase gene modulation on suicidal behavior and personality traits: review, meta-analysis and association study. J. Psychiatr. Res. 45, 309–321 15 Jinks, J.L. and Fulker, D.W. (1970) Comparison of the biometrical genetical, MAVA, and classical approaches to the analysis of human behavior. Psychol. Bull. 73, 311–349 16 Bronfenbrenner, U. and Ceci, S.J. (1994) Nature-nurture reconceptualized in developmental perspective: a bioecological model. Psychol. Rev. 101, 568–586 17 Lahti, J. et al. (2005) Novelty seeking: interaction between parental alcohol use and dopamine D4 receptor gene exon III polymorphism over 17 years. Psychiatr. Genet. 15, 133–139

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 18 Pluess, M. et al. (2010) 5-HTTLPR moderates effects of current life events on neuroticism: differential susceptibility to environmental inﬂuences. Prog. Neuropsychopharmacol. Biol. Psychiatry 34, 1070– 1074 19 Munafo`, M.R. et al. (2008) Gene environment interactions in psychiatric genetics. Behav. Pharmacol. 19, 651 20 Munafo`, M.R. et al. (2009) Gene environment interactions at the serotonin transporter locus. Biol. Psychiatry 65, 211–219 21 Munafo`, M.R. and Flint, J. (2009) Replication and heterogeneity in gene environment interaction studies. Int. J. Neuropsychopharmacol. 12, 727–729 22 Munafo`, M.R. (2009) Reliability and replicability of genetic association studies. Addiction 104, 1439–1440 23 Brookes, S.T. et al. (2001) Subgroup analyses in randomised controlled trials: quantifying the risks of false-positives and false-negatives. Health Technol. Assess. 5, 1–56 24 Stefansson, K. et al. (2008) Many sequence variants affecting diversity of adult human height. Nat. Genet. 40, 609–615 25 Wellcome-Trust-Case-Control-Consortium (2007) Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature 447, 661–678 26 Abecasis, G.R. et al. (2009) Genome-wide association study identiﬁes eight loci associated with blood pressure. Nat. Genet. 41, 666–676 27 Levy, D. et al. (2009) Genome-wide association study of blood pressure and hypertension. Nat. Genet. 41, 677–687 28 Flint, J. et al. (2008) A whole genome association study of neuroticism using DNA pooling. Mol. Psychiatry 13, 302–312 29 Terracciano, A. et al. (2010) Genome-wide association scan for ﬁve major dimensions of personality. Mol. Psychiatry 15, 647–656 30 van den Oord, E.J.C.G. et al. (2008) Genomewide association analysis followed by a replication study implicates a novel candidate gene for neuroticism. Arch. Gen. Psychiatry 65, 1062–1071 31 de Moor, M.H.M. et al. (2009) Meta-analysis of genome-wide association results in >10.000 individuals for the big ﬁve personality traits. Behav. Genet. 39, 643 32 Fullerton, J. et al. (2003) Linkage analysis of extremely discordant and concordant sibling pairs identiﬁes quantitative-trait loci that inﬂuence variation in the human personality trait neuroticism. Am. J. Hum. Genet. 72, 879–890 33 Gillespie, N.A. et al. (2008) A genome-wide scan for Eysenckian personality dimensions in adolescent twin sibships: Psychoticism, Extraversion, Neuroticism, and Lie. J. Pers. 76, 1415–1445 34 Kuo, P.H. et al. (2007) A genome-wide linkage analysis for the personality trait neuroticism in the Irish affected sib-pair study of alcohol dependence. Am. J. Med. Genet. B: Neuropsychiatr. Genet. 144B, 463–468 35 Nash, M.W. et al. (2004) Genome-wide linkage analysis of a composite index of neuroticism and mood-related scales in extreme selected sibships. Hum. Mol. Genet. 13, 2173–2182 36 Neale, B.M. et al. (2005) A genome scan of neuroticism in nicotine dependent smokers. Am. J. Med. Genet. B Neuropsychiatr. Genet. 132B, 65–69 37 Wray, N.R. et al. (2008) Genome-wide linkage analysis of multiple measures of neuroticism of 2 large cohorts from Australia and the Netherlands. Arch. Gen. Psychiatry 65, 649–658 38 Willis-Owen, S.A.G. et al. (2005) The serotonin transporter length polymorphism, neuroticism, and depression: A comprehensive assessment of association. Biol. Psychiatry 58, 451–456 39 Sterne, J.A.C. and Davey-Smith, G. (2001) Sifting the evidence - what’s wrong with signiﬁcance tests? Br. Med. J. 322, 226–231 40 Kavvoura, F.K. et al. (2008) Evaluation of the potential excess of statistically signiﬁcant ﬁndings in published genetic association studies: application to Alzheimer’s disease. Am. J. Epidemiol. 168, 855–865 41 Duncan, L.E., and Keller, M.C. (2011) A critical review of the ﬁrst ten years of candidate gene-by-environment interaction research in psychiatry. Am. J. Psychiatry (in press) 42 Purcell, S.M. et al. (2009) Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature 460, 748–752 43 Wray, N.R. and Visscher, P.M. (2010) Narrowing the boundaries of the genetic architecture of schizophrenia. Schizophr. Bull. 36, 14–23 44 Wray, N.R. et al. (2011) Synthetic associations created by rare variants do not explain most GWAS results. PLoS Biol. 9, e1000579

399

Review 45 Durbin, R.M. et al. (2010) A map of human genome variation from population-scale sequencing. Nature 467, 1061–1073 46 Tupes, E.C. and Christal, R.E. (1992) Recurrent personality-factors based on trait ratings. J. Pers. 60, 225–251 47 McCrae, R.R. and Costa, P.T. (1985) Comparison of EPI and Psychoticism scales with measures of the 5-Factor Model of personality. Pers. Indiv. Diff. 6, 587–597 48 Zuckerman, M. (1992) What is a basic factor and which factors are basic - turtles all the way down. Pers. Indiv. Diff. 13, 675–681 49 Cloninger, C.R. (1994) Temperament and personality. Curr. Opin. Neurobiol. 4, 266–273 50 Eysenck, H.J. (1991) Dimensions of personality - 16, 5 or 3 - criteria for a taxonomic paradigm. Pers. Indiv. Diff. 12, 773–790 51 Eysenck, H.J. (1992) 4 ways 5 factors are not basic. Pers. Indiv. Diff. 13, 667–673

400

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 52 Zuckerman, M. and Cloninger, C.R. (1996) Relationships between Cloninger’s, Zuckerman’s, and Eysenck’s dimensions of personality. Pers. Indiv. Diff. 21, 283–285 53 Heath, A.C. et al. (1994) Testing a model for the genetic-structure of personality - a comparison of the personality systems of Cloninger and Eysenck. J. Pers. Soc. Psychol. 66, 762–775 54 Dickson, S.P. et al. (2010) Rare variants create synthetic genome-wide associations. PLoS Biol. 8, e1000294 55 1000-Genomes-Project-Consortium (2010) A map of human genome variation from population-scale sequencing. Nature 467, 1061–1073 56 Mills, R.E. et al. (2011) Mapping copy number variation by populationscale genome sequencing. Nature 470, 59–65 57 Li, Y. et al. (2011) Low-coverage sequencing: Implications for design of complex trait association studies. Genome Res. 21, 940–951

Review

Special Issue: The Genetics of Cognition

Genetics of emotion Laura Bevilacqua and David Goldman Laboratory of Neurogenetics, National Institute on Alcohol Abuse and Alcoholism, NIH, Rockville, MD 20852, USA

Emotion is critical to most aspects of human behavior, and individual differences in systems recruited to process emotional stimuli, expressed as variation in emotionality, are characteristic of several neuropsychiatric disorders. We examine the genetic origins of individual differences in emotion processing by focusing on functional variants at five genes: catechol-O-methyltransferase (COMT), serotonin transporter (SLC6A4), neuropeptide Y (NPY), a glucocorticoid receptor-regulating co-chaperone of stress proteins (FKBP5) and pituitary adenylate cyclase-activating polypeptide receptor (ADCYAP1R1). These represent a range of effects of genes on emotion as well as the variety of mechanisms and factors, such as stress, that modify these effects. The new genomic era of genome-wide association studies (GWAS) and deep sequencing may yield a wealth of new loci modulating emotion. The effects of these genes can be validated by neuroimaging, neuroendocrine and other studies accessing intermediate phenotypes, deepening our understanding of mechanisms of emotion and variation in emotionality. Candidate gene studies of emotion Emotion, or the assignment of affective valence to objects, states and situations (see Glossary), is critical to most aspects of human behavior: all our actions and decisions occur in emotional contexts, and cognitive functions are colored by emotional states. Neurobiologically, emotion is linked to the response of neural circuits by which individuals assign intensity and valence to objects and situations in the environment, and to internal states. Emotionality represents variation in emotion: it varies between individuals and also varies within an individual, depending on context and life history. Variation in systems recruited to process stimuli that trigger emotion and regulate emotional outputs leads to differences in emotionality. Crucially, emotionality is a factor in a wide spectrum of psychiatric diseases including mood [1,2] and anxiety disorders [3]. Signiﬁcant progress has been made in identifying the brain structures that underlie affective processing, including the prefrontal cortex (PFC), anterior cingulate cortex, hippocampus and amygdala [4,5], in visualizing and modifying the activity of the circuitry of emotion [6] and in identifying genetic variants that modulate inter-individual differences in emotionality. In this review, we focus on the genetic origins of individual differences in emotion processing. Heritable individual differences in affect, temperament and personality shape other complex behaviors, as well as responses to an ever-changing environment. These

differences can also be important predictors of vulnerability to neuropsychiatric disorders that are themselves genetically inﬂuenced. Functional variants at ﬁve genes (catecholO-methyltransferase (COMT), serotonin transporter (SLC6A4), neuropeptide Y (NPY), a glucocorticoid receptor-regulating co-chaperone of stress proteins (FKBP5) and pituitary adenylate cyclase-activating polypeptide receptor (ADCYAP1R1)) will be used to illustrate a range of effects of ‘emotion genes’ and factors that alter or confound these effects. These genes have been selected on the basis of convergent neurobiological evidence in modulation of emotional processes. Notably, each of these represents a discovery made prior to a new era in which genome-wide association studies (GWAS) and deep sequencing are expected to yield a wealth of new loci modulating emotion. As will be described, the effects of these functional alleles has to some extent been explored, and validated, both via human brain imaging and in animal models, in which the genes have been genetically manipulated, although space considerations prevent full review of the functional connections to the neurobiology of emotion for any of these genes. Emotionality is moderately heritable (40-60%) [7] but is also strongly inﬂuenced by exposure to stress in a pattern consistent with gene environment interaction. Identiﬁcation of the mechanisms that give rise to inter-individual differences in emotional stability and vulnerability to stress and anxiety will deepen our understanding of human behavior and predisposition to disease. The effect sizes of genetic variants implicated in emotional responses measured as psychological traits have been small. For example, in a meta-analytical study, the serotonin transporter promoter polymorphism was found to have only a 0.106 standard deviation effect on neuroticism per copy of the low expression allele [8]. However, Glossary Allele: any of the alternative variants at a given locus. Candidate gene study: a study focused on genes thought to have a causal role. Diplotype: a genotype composed of haplotypes. Emotion: the assignment of affective valence to objects, states and situations. Emotionality: the tendency to assign strong emotional valence to objects, states or situations. Endophenotype: a heritable, disease associated, intermediate phenotype. Epistasis: non-additive interaction between alleles at different loci affecting a phenotype. Genome-wide association study: allele-based linkage using markers genotyped across the entire genome. Haplotype: a combination of alleles at loci located on the same chromosome. Intermediate phenotype: mechanism-related manifestation of a complex phenotype. Pleiotropy: different phenotypic effects of the same allele. Polymorphism: inter-individual genetic variation occurring at a frequency of >1%.

Corresponding author: Goldman, D. ([email protected]). 1364-6613/$ – see front matter . Published by Elsevier Ltd. doi:10.1016/j.tics.2011.07.009 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

401

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Box 1. Heritability of emotionality and emotion-induced fMRI activations Emotionality, measured as neuroticism, introversion or harm avoidance, is moderately to highly heritable in both sexes: 40-60%, with little effect of shared environment [7,75]. Behavioral inhibition is manifested early in life and predicts anxiety disorders and depression later in life [76–78]. In humans, trait emotionality has been tied to brain metabolic measures obtained by neuroimaging. However, while these measures show long-term stability and are thus trait-like [10], they have not been shown to be heritable [79]. Non-human primates represent an alternative model for the study of the relationship of brain metabolic activity to emotionality and direct evaluation of heritability. In non-human primates, emotionality and correlated brain activations are also stable and persistent across environments [80]. As shown in Figure I, in the rhesus macaque, the central nucleus of the amygdala and anterior hippocampus are components of a neural

[(Box_1)TD$FIG]

(b)

Dorsal amygdala metabolism and AT 4

Peak FDG uptake (standardized residuals)

Peak FDG uptake (standardized residuals)

(a)

circuit predictive of anxious temperament [81]. In these monkeys, brain metabolic activity was measured by fludeoxyglucose PET (FDG PET) and anxious behavior was measured by observation under a threatening condition. Using a multigenerational pedigree, the heritability of anxious temperament was 0.36 [81,82]. Heritability of threat-provoked hippocampal metabolic activity, but surprisingly not for the amygdala, was even higher (0.65 and 0.76 in right and left hippocampus, respectively). Concerning the lack of heritability of amygdala FDG, a key question would be whether it reflects measurement properties of FDG in hippocampus vs. amygdala. These findings strongly suggest that brain metabolic responses measured by neuroimaging represent an endophenotype—an intermediate phenotype that is both heritable and disease-associated [11]—for emotionality and related disorders.

3 2 1 0 –1 –2 –3 –4 –3

–2

–1 0 1 AT (standardized residuals)

2

3

Anterior hippocampus metabolism and AT 4 3 2 1 0 –1 –2 –3 –4 –3

–2

–1 0 1 AT (standardized residuals)

2

3

TRENDS in Cognitive Sciences

Figure I. Correlation between anxious temperament and glucose metabolism in (a) the amygdala and (b) the hippocampus in rhesus macaques. Reproduced, with permission, from [81].

several of the same variants, including this serotonin transporter allele, have larger effects on metabolic responses of the brain to emotional stimuli accessed in real time by brain imaging. These studies using intermediate phenotypes have not only validated the effects of these genes in emotion, but deepened our understanding of mechanisms by which neural circuits alter emotional responses and cognition. As described in Boxes 1 and 2, variation in BOLD fMRI measures of emotional response is both heritable and associated with emotion in the rhesus macaque. In humans, these fMRI measures of emotion have not been shown to be heritable, but they are a temporally stable and reliable index of brain function [9,10], and correlate with emotionality traits including diseases whose heritability has been demonstrated. Thus, they are deﬁned as endophenotypes [11]. COMT Catechol-O-methytransferase (COMT) metabolizes dopamine, noradrenaline and other catecholamines. This enzyme plays an important role in the prefrontal cortex where the dopamine transporter is not abundant [12]. Val158Met is a common functional polymorphism that alters enzyme stability [13] and is predicted to lower dopamine levels in the prefrontal cortex. Under normal 402

conditions the COMT knockout mouse may not have altered catecholamine levels but under others, for example when challenged with dihydroxyphenylalanine, region and sex-speciﬁc changes have been observed [14]. Individuals with the Val158 allele perform less well on tests of working memory and executive cognition assessed by several methods [15–18]. On the other hand, the Met158 allele, although associated with better cognitive performance, leads to higher anxiety [19] and emotionality, as summarized in the ‘warrior’ (Val158) versus ‘worrier’ (Met158) model [20]. The two alleles potentially represent an example of balanced selection: counterbalancing effects of the Val and Met alleles in stress resiliency and cognition, respectively, may account for their conservation and high abundance across populations. This serves as a reminder that many of the common alleles with roles in diseases are not disease alleles per se. Met158 leads to enhanced emotionality as visualized by emotional responses assessed by Positron Emission Tomography (PET) and functional Magnetic Resonance Imaging (fMRI). Met158 leads to reduced pain thresholds and increased emotional response to pain [21,22], apparently via diminished regional opioid responses [21]. Amygdala fMRI response to emotional challenge is stronger in Met158 carriers [23], and the effect also appears to be

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Box 2. The power of endophenotypes for studying the effects of emotionality genes A functional promoter polymorphism of NPY, a gene that encodes an anxiolytic neuropeptide, appears to predict emotionality and stress response [57]. This study evaluated the effect of common functional NPY haplotypes on molecular, neuroimaging and behavioral phenotypes. Illustrated in Figure I, in one of the imaging genetic studies in this report, amygdala and hippocampal activations were measured in response to threat-related facial expressions. NPY diplotypes were grouped into low (LL), intermediate (LH) and high (HH) expression groups, based on the in vitro and in vivo RNA levels associated with different haplotypes, and the identification of a functional locus within the haplotype, and located in the NPY promoter region. Task-related reactivity was predicted in an allele-

[(Box_2)TD$FIG]

dosage fashion in both right dorsal amygdala and hippocampus, accounting for 6-9% of the variance in fMRI response to emotional challenge. Emotion-induced activations were highest in individuals with low NPY expression diplotypes (LL) and lowest in those with high-expression (HH) diplotypes. In contrast with large effects of NPY on brain functional responses and measures of NPY mRNA and protein levels, the effect on trait anxiety, measured as Harm Avoidance with the Tridimensional Personality Questionnaire, was directionally congruent in an independent dataset but accounted for only 3.3-3.4% of the variance. Intermediate phenotypes, if they are heritable endophenotypes, may, therefore, more closely reflect the effects of genes.

T value 3.5 3 2.5 2 1.5 1 0.5 0 Amygdala

Hippocampus

fMRI activation (Arbitrary units)

2.5

Key: Amygdala Hippocampus

2

1.5

P = 0.003

1

0.5 P = 0.006 0 LL (n = 11)

LH (n = 42)

HH (n = 18)

NPY diplotype TRENDS in Cognitive Sciences

Figure I. Effect of diplotype-predicted NPY mRNA expression on fMRI measured amygdala and hippocampal activation in response to threat related facial expressions. Top: Genotype-dependent NPY expression predicts the fMRI activation of amygdala and hippocampus in response to threat-related facial expressions. Bottom: Right amygdala (blue diamonds) and hippocampus (orange squares) activation predicted by NPY genotype. The lower expression NPY haplotype predicted stronger threatrelated activations. Reproduced, with permission, from [57].

additive with the low expression allele of the serotonin transporter promoter polymorphism [24]. These fMRI measures of metabolic activity of regions, such as the amygdala and hippocampus, appear to have validity for understanding emotion [25], leading to their intensive use, and creating the opportunity for imaging genetic studies. Interestingly, COMT Met/Met individuals present increased reactivity and stronger connectivity of brain circuitry implicated in generating and regulating emotional

responses [23]. This circuitry includes the amygdala, orbito-frontal and ventro-lateral prefrontal cortex and the hippocampus. It is clear that speciﬁc brain regions play an important role in emotion. However, they do not work in isolation, but are part of networks that are responsible for perception, emotion and cognitive processes [26]. Maladaptation to stress is critical in the development of many psychiatric disorders, including mood and anxiety disorders [27]. For COMT, the context-dependent effect of 403

Review the Met158 allele on emotionality was demonstrated in a study conducted with Rwandan refugees [28]. Post-traumatic stress disorder (PTSD) was more common in carriers of the Met158 alleles, except that the highest levels of stress exposure overwhelmed the resiliency effect of the Val158 allele, as might have been expected. The effects of genes on emotion is strongly inﬂuenced by prior exposure to environmental stress [29], but this study of Kolassa and colleagues [28] provides evidence that the interactions between genes and environmental stress are furthermore not necessarily linear or monophasic. Of interest is recent evidence for epigenetic regulation of the COMT Val158 allele by lifetime stress exposure [30]. Greater stress correlates with reduced methylation at the CpG dinucleotide created by the Val158 allele, possibly altering the expression of COMT and traits associated with it. 5-HTTLPR The serotonin transporter gene is responsible for the reuptake of serotonin from the synaptic cleft and is a major target in the pharmacologic treatment of depression and anxiety. A common polymorphism (5-HTTLPR) is located upstream from the gene, SLC6A4, also known as 5-HTT [31]. A 14-repeat allele (S) has reduced transcriptional efﬁciency compared to the 16 repeat allele (L). Moreover, the L allele frequently contains a relatively common, A [TD$INLE] G substitution that makes it functionally equivalent to the low-expression S allele via binding of a deﬁned transcription factor [32]. Subsequent to initial association studies to anxiety, Hariri and colleagues [33] observed that the low transcription allele (S) increased activation of the amygdala after passively viewed emotional stimuli. As mentioned above, the metabolic response of the amygdala to emotional challenge is contingent on modulation by other regions [23]. Consistent with this, connectivity studies revealed that the low expression allele altered the functional coupling between the amygdala and the ventromedial prefrontal cortex (vmPFC) [34] and the perigenual cingulate [35], potentially impairing fear extinction by the medial prefrontal cortex. S allele carriers showed increased coupling between the amygdala and vmPFC and relative uncoupling between the amygdala and the perigenual region. At the neural level, multiple lines of evidence link the S allele to stronger emotional arousal (see [36], for a review). These include structural alterations in the uncinate fasciculus, which connects the amygdala and PFC [37], in the pulvinar [38] and in the amygdala itself [35]. As with COMT, there is accumulating evidence that 5HTTLPR and stress interact to determine susceptibility to disorders later in life [29]. In the case of the serotonin transporter, however, the evidence is both more extensive and more conﬂicting, with contradictory meta-analyses [39,40]. Carriers of the low transcribing allele appear to be more likely to be depressed and suicidal following stressful life events as compared to individuals with two copies of the high expression allele [29]. The same effect was observed in other populations, including substance abusers, who overall have high exposure to stress and a high risk of suicidality [41]. As reviewed in [36], functional ﬁndings providing clues to the mechanisms of effects of 404

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

5-HTT reduction-of-function genotypes on emotion include increased acquisition of conditioned fear responses [42], increased auditory startle [43,44] and increased hypothalamic pituitary adrenal (HPA) axis activation to aversive stimuli [45–48]. The ﬁndings on human 5-HTTLPR are paralleled by studies in the rhesus macaque, mice and rats. To some extent, these studies are important as validating the sometimes conﬂicting human ﬁndings, but they have also enabled further exploration of the effects of low and high expression serotonin transporter genotypes. The rhesus macaque has an orthologous rh-HTTLPR polymorphism that leads to increased stress responses including cortisol release, and again only in the context of early life stress exposure, namely maternal separation [49]. In the rhesus macaque, the lower expression allele is associated with lower volumes of several of the same brain regions that are affected in humans, including the amygdala, PFC and pulvinar [50], and increased metabolic activity was observed in these regions in a stress paradigm involving relocation [51]. There is no rodent orthologue of 5HTTLPR, but in both mice and rats silencing of 5-HTT heightens anxiety-like behaviors, impairs fear extinction and exaggerates HPA axis response [52,53]. The ﬁnding that 5-HTT null mice have higher excitatory dendritic spine density on amygdala neurons and increased PFC dendritic arborization [54] again points to a developmental effect of this gene and its functional variants. The plasticity of this interaction is underlined by the ability of SSRIs administered early in development to mimic the 5-HTT knockout mouse, at least in terms of anxiety behaviors [55]. NPY Neuropeptide Y (NPY) is an anxiolytic neuropeptide whose release is induced by stress. The release of NPY inﬂuences stress response and vulnerability to PTSD, for example in combat veterans [56]. At the molecular level, NPY RNA and neuropeptide expression is modulated by variation in the NPY promoter region, as shown in vitro and in vivo [57]. Individuals with low expression haplotypes exhibit stronger hemodynamic responses in the amygdala when presented with threat-related stimuli (facial expressions), lower endogenous opioid release during a pain stressor, and greater trait anxiety [57]. Also, amygdala activation to angry faces is greater in normal [57] and depressed patients [58] with low expression NPY genotypes. The dilution of effect of this functional locus from molecule to complex behavior is described in Box 2. FKBP5 Acute stress activates hypothalamic release of corticotropin-releasing hormone (CRH) from the paraventricular nulecus to the pituitary, where it stimulates the secretion of the adrenocorticotropic hormone (ACTH). CRH directly, and through the action of ACTH, regulates adrenal cortisol release, steroidogenesis and catecholamine synthesis and release by the adrenal gland. Negative feedback to block release of CRH is necessary for normal function of the HPA axis, preventing prolonged or excessive activation [59]. Glucocorticoids mediate physiologic responses to stress but also reduce release of these neuropeptides by activa-

Review tion of glucocorticoid receptors (GR) in the paraventricular nucleus [59]. GR is a ligand-activated transcription factor that translocates from the cytosol to the nucleus after binding cortisol. GR function is regulated by a large molecular complex [60]. This molecular machinery includes hsp90/hsp70 chaperones and a number of co-chaperones, including FKBP5, a co-chaperone of hsp90 [61]. When FKBP5 is bound to the GR complex, the receptor has lower afﬁnity for cortisol. Increased expression of FKBP5 therefore leads to cellular glucocorticoid resistance [62]. Although the precise functional locus has not been mapped, functional variation in the FKBP5 gene has been associated with response to antidepressants, recurrence of depressive episodes [63], suicide attempts in bipolar patients [64] and incomplete normalization of stress-elicited cortisol secretion [65]. Moreover, it has been shown that FKBP5 interacts with childhood trauma to predict PTSD [66] and suicidal behavior [67]. In controls, several FKBP5 loci are associated with high protein expression, increased glucocorticoid resistance, and thus reduced dexamethasone suppression [63]. However, in the presence of PTSD this functional association is altered. An interaction between high expression FKBP5 alleles and childhood trauma increases risk for PTSD and these alleles are associated with increased glucocorticoid sensitivity [66]. A similar relationship is observed in depressed patients [63]. Thus, genetic variation in FKBP5 may modulate effects of childhood trauma on cortisol release and abnormal protein expression may lead to altered GR responsiveness in target organs and long-lasting alterations in HPA axis reactivity. ADCYAP1R1 Among the genes that modulate stress response, the pituitary adenylate cyclase-activating polypeptide (PACAP) and its selective PAC1 receptor gene have recently been demonstrated to play a role in abnormal stress response underlying PTSD in females [68]. In this study, PACAP blood levels were correlated with the diagnosis and symptoms of PTSD. Ressler et al. [68] found that a single nucleotide polymorphism (SNP) within the PAC1 receptor gene (ADCYAP1R1) was associated with PTSD in females but not males, in two distinct samples of highly traumatized urban subjects. Expression of this gene is induced in the amygdala by fear [68]. The SNP, which alters mRNA expression, is located in an oestrogen response element, which could explain the striking gender difference. In female rats, oestradiol was shown to increase both Adcyap1 and Adcyap1r1 transcript levels in the bed nucleus of the stria terminalis, a component of the extended amygdala. Finally, methylation of ADCYAP1R1 was associated with PTSD. Concluding remarks Emotional differences arise because of the action of hundreds of genes, complex circuitries and environmental exposures. One can reasonably argue that it will not be possible to disentangle these factors or that, if it is possible, this will require far better measurement of behavior, intermediate processes and environmental exposures. However, the fact that emotionality is genetically driven—in

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Box 3. COMT and pleiotropy The COMT Val158Met polymorphism has pleiotropic effects on both emotion and cognition and it represents an example of a multilevel confirmation of the effect of a functional polymorphism. COMT was not genome-wide significant on any GWAS conducted so far, but has a functional locus that coherently influences several aspects of emotion and cognition. The understanding of this polymorphism’s effects was extended via pharmacological challenge studies and use of transgenic animal models. The effects of COMT Val158Met are pleiotropic, consistent with the multiple functions of the enzyme and its widespread distribution in the brain and body. In numerous studies and different populations including normal controls [15–17], head-injured patients [83], schizophrenic patients and their healthy siblings [16], the Met158 allele is associated with better performance on neuropsychological tests of executive cognition. The role of COMT as a cognitive predictor has been comprehensively reviewed [84] and meta-analyzed with variable results, as in Barnett et al [85], where the genotype was associated with IQ but not with other cognitive measures. Subsequently, Met158 was again associated with higher IQ in a large longitudinal dataset [86]. However, all observations made on the effects of the Val158Met polymorphism are consistent with the role played by dopamine in regulating frontal cortical function. COMT inhibitors improve working memory and attention in animals [87] and in humans [88], but the improvement is lost in Met158/Met158 homozygotes, who tend to have superior baseline prefrontal function assayed by fMRI [89]. Also, under easy task conditions, where cognitive performance of individuals with different genotypes is equivalent, there is reduced cortical efficiency in Val158/ Val158 homozygotes, consistent with lower dopamine levels.

other words, heritable—has enabled genetic approaches to be successfully applied against that large portion of the variance in emotionality which is driven by allelic differences. The ﬁve genes discussed here represent instances in which common functional variants alter pathways of emotion and stress response. Their naturally occurring genetic variants in humans, and parallel genetic models in other animals, represent a unique opportunity to disentangle the complexity of emotion and identify speciﬁc origins of individual differences in emotionality. Multiple factors complicate the task of relating genes to emotion and suggest that future efforts to relate genes to emotion should be better focused, larger in scale, as well as deeper in genomic coverage. Gene environment interactions play a critical role, suggesting that it is vital to study exposed populations, as exempliﬁed here by the studies on Rwandan refugees [28], and that at other times environmental exposures may mask genetic effects. Most genes and environmental factors that alter brain function, alter the function of multiple neurons and circuits and may have effects elsewhere in the body, a phenomenon known as pleiotropy (see Box 3), suggesting that the genetics of emotionality is likely to be tied to several other problems. Gene gene interaction is a territory largely unexplored (see also Box 4). The complexity of the brain and its cellular molecular networks seems to suggest that variation in emotionality should arise due to the simultaneous action of variants at many genes. Under a heterogeneity model, different variants play a role in different individuals. Under a polygenic model, inheritance of many genetic variants is necessary and the effects may or may not be additive. Under the epistatic model, effects are nonadditive—it is unique combinations of alleles at different 405

Review Box 4. Priorities for future research Achieve deep (i.e., rich) phenotyping that includes the use of intermediate phenotypes or endophenotypes. Establish consistency in definition of phenotypes across studies. Carry out GWAS using intermediate phenotypes or endophenotypes. Perform genome sequencing of individuals with extreme phenotypes, including intermediate phenotypes. Identify, evaluate, select and prioritize rare and uncommon variants. Establish the function of variants. Perform gene-wide studies that incorporate multiple rare variants at the same gene. Use population isolates with founder characteristics and families for the identification of rare and uncommon genetic variants with large effects on disease. Study exposed populations and develop new measures of exposure, including epigenetic signatures.

loci that lead to the phenotype. Based on ratios of trait concordances between individuals at different degrees of relationship, emotionality, other personality traits and various psychiatric diseases may be predominantly additive in their inheritance (see, for example, [69] for a review of the MZ:DZ ratios for ten different addictive disorders). As mentioned above, the effect of COMT and 5-HTTLPR on fMRI emotion response appears to be additive [24]. However, it would be unsurprising if instances of epistatic interaction were identiﬁed: for example such instance has been reported between COMT Val158Met and various other genes [70], although the functional variants have not been identiﬁed at these other genes. Replication will be particularly important for gene gene interactions, as will care in correcting for the multiple comparisons necessitated by the search for epistatic interactions. GWAS is a hypothesis-free search strategy with great power to detect effects of relatively common alleles of moderate effect. However, and as may be a particular problem for genetic studies on emotionality, GWAS is unlikely to detect the effects of rare and uncommon alleles because, as sample size is increased, various types of within-sample heterogeneity, including consistency of phenotyping, ascertainment, environmental exposure and genetic background, are likely to detract from power. Different rare alleles at the same gene are likely to reside on different genetic backgrounds (haplotypes). A GWAS of neuroticism, a personality trait that increases the likelihood of internalizing disorders, which include depression and anxiety disorders, failed to identify any genome-wide signiﬁcant loci, despite the fact that it was based on the genotyping of individuals with the most extreme phenotypes from a total sample approaching 88,000 [71]. Furthermore, this study failed to detect—at a genome-wide level—the inﬂuence on emotion of several genes that have been discovered and otherwise validated, as the genes discussed in this review. The genetic factors underlying neuroticism are largely shared with those that inﬂuence liability to internalizing disorders [72]. A powerful implication of these heritability ﬁndings is that it should be feasible to map the locations of genes that lead to neuroticism, to identify the relevant functional loci and deﬁne a pathway of causation from gene to behavior. The lack of genome-wide signiﬁcant loci for this heritable trait could be 406

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

due to the fact that heritability of neuroticism arises from multiple common alleles explaining a small percentage of the variance, an even larger number of rare alleles, the need for a larger sample, the strategy or implementation of this study, or as is most likely, a combination of factors. It is likely that increasing the sample size will result in the genome-wide capture of some genes inﬂuencing emotion, but the question is how many, and how much of the heritable trait variance. Instead of performing new genome scans of even larger samples, it would perhaps be more important to shift attention to the use of intermediate phenotypes, or to endophenotypes, for the identiﬁcation of functional common variation as of rare and uncommon variation. The search for rare variation will also require the use of massively parallel sequencing technologies, the selection of appropriate phenotypes and the insightful evaluation, prioritization and follow-up of the great number of rare variants that will be identiﬁed [73]. The use of isolated populations with founder characteristics, as well as individual families, is likely to be highly beneﬁcial to reduce genetic heterogeneity, as with an HTR2B stop codon predictive of severe impulsivity, but only common in individuals of Finnish descent [74]. These observations suggest that emotion and diseases associated with emotional dysregulation will not be ‘‘solved’’ by a one-to-one correspondence of gene to behavior as currently deﬁned. This encourages a multilevel approach to the genetic analysis of emotion, and a stepwise deconstruction of current typology based on predictive genetic and psychophysiologic measures, together with observed clinical state and history of exposure. References 1 Phillips, M.L. et al. (2003) Neurobiology of emotion perception II: Implications for major psychiatric disorders. Biol. Psychiatry 54, 515–528 2 Whalen, P.J. et al. (2002) Functional neuroimaging studies of the amygdala in depression. Semin. Clin. Neuropsychiatry 7, 234–242 3 Etkin, A. et al. (2004) Individual differences in trait anxiety predict the response of the basolateral amygdala to unconsciously processed fearful faces. Neuron 44, 1043–1055 4 Etkin, A. et al. (2011) Emotional processing in anterior cingulate and medial prefrontal cortex. Trends Cogn. Sci. 15, 85–93 5 Elliott, R. et al. (2011) Affective cognition and its disruption in mood disorders. Neuropsychopharmacology 36, 153–182 6 Hariri, A. (2009) The neurobiology of individual differences in complex behavioral traits. Annu. Rev. Neurosci. 32, 225–247 7 Bouchard, T.J., Jr and Loehlin, J.C. (2001) Genes, evolution, and personality. Behav. Genet. 31, 243–273 8 Sen, S. et al. (2004) Meta-analysis of the association between a serotonin transporter promoter polymorphism (5-HTTLPR) and anxiety-related personality traits. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 127, 85–89 9 Johnstone, T. et al. (2005) Stability of amygdala BOLD response to fearful faces over multiple scan sessions. Neuroimage 25, 1112–1123 10 Manuck, S.B. et al. (2007) Temporal stability of individual differences in amygdala reactivity. Am. J. Psychiatry 164, 1613–1614 11 Gottesman, I.I. and Gould, T.D. (2003) The endophenotype concept in psychiatry: etymology and strategic intentions. Am. J. Psychiatry 160, 636–645 12 Yavich, L. et al. (2007) Site-speciﬁc role of catechol-Omethyltransferase in dopamine overﬂow within prefrontal cortex and dorsal striatum. J. Neurosci. 27, 10196–10209 13 Scanlon, P.D. et al. (1979) Catechol-O-methyltransferase: thermolabile enzyme in erythrocytes of subjects homozygous for allele for low activity. Science 203, 63–65

Review 14 Huotari, M. et al. (2002) Brain catecholamine metabolism in catechol-Omethyltransferase (COMT)-deﬁcient mice. Eur. J. Neurosci. 15, 246–256 15 Barnett, J.H. et al. (2007) Effects of the catechol-O-methyltransferase Val158Met polymorphism on executive function: a meta-analysis of the Wisconsin Card Sort Test in schizophrenia and healthy controls. Mol. Psychiatry 12, 502–509 16 Egan, M.F. et al. (2001) Effect of COMT Val108/158 Met genotype on frontal lobe function and risk for schizophrenia. Proc. Natl. Acad. Sci. U.S.A. 98, 6917–6922 17 Malhotra, A.K. et al. (2002) A functional polymorphism in the COMT gene and performance on a test of prefrontal cognition. Am. J. Psychiatry 159, 652–654 18 Goldberg, T.E. et al. (2003) Executive subprocesses in working memory: relationship to catechol-O-methyltransferase Val158Met genotype and schizophrenia. Arch. Gen. Psychiatry 60, 889–896 19 Enoch, M.A. et al. (2003) Genetic origins of anxiety in women: a role for a functional catechol-O-methyltransferase polymorphism. Psychiatr. Genet. 13, 33–41 20 Ducci, F. and Goldman, D. (2008) Genetic approaches to addiction: genes and alcohol. Addiction 103, 1414–1428 21 Zubieta, J.K. et al. (2003) COMT val158met genotype affects mu-opioid neurotransmitter responses to a pain stressor. Science 299, 1240–1243 22 Diatchenko, L. et al. (2005) Genetic basis for individual variations in pain perception and the development of a chronic pain condition. Hum. Mol. Genet. 14, 135–143 23 Drabant, E.M. et al. (2006) Catechol-O-methyltransferase val158met genotype and neural mechanisms related to affective arousal and regulation. Arch. Gen. Psychiatry 63, 1396–1406 24 Smolka, M.N. et al. (2007) Gene-gene effects on central processing of aversive stimuli. Mol. Psychiatry 12, 307–317 25 Phelps, E.A. and LeDoux, J.E. (2005) Contributions of the amygdala to emotion processing: from animal models to human behavior. Neuron 48, 175–187 26 Pessoa, L. (2010) Emergent processes in cognitive-emotional interactions. Dialogues Clin. Neurosci. 12, 433–448 27 McEwen, B.S. (2000) Allostasis and allostatic load: implications for neuropsychopharmacology. Neuropsychopharmacology 22, 108–124 28 Kolassa, I.T. et al. (2010) The risk of posttraumatic stress disorder after trauma depends on traumatic load and the catechol-o-methyltransferase Val(158)Met polymorphism. Biol. Psychiatry 67, 304–308 29 Caspi, A. et al. (2003) Inﬂuence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 301, 386–389 30 Ursini, G. et al. (2011) Stress-related methylation of the catechol-Omethyltransferase Val 158 allele predicts human prefrontal cognition and activity. J. Neurosci. 31, 6692–6698 31 Lesch, K.P. et al. (1996) Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science 274, 1527–1531 32 Hu, X. et al. (2006) Serotonin transporter promoter gain-of-function genotypes are linked to obsessive-compulsive disorder. Am. J. Hum. Genet. 78, 815–826 33 Hariri, A.R. et al. (2002) Serotonin transporter genetic variation and the response of the human amygdala. Science 297, 400–403 34 Heinz, A. et al. (2005) Amygdala-prefrontal coupling depends on a genetic variation of the serotonin transporter. Nat. Neurosci. 8, 20–21 35 Pezawas, L. et al. (2005) 5-HTTLPR polymorphism impacts human cingulate-amygdala interactions: a genetic susceptibility mechanism for depression. Nat. Neurosci. 8, 828–834 36 Caspi, A. et al. (2010) Genetic sensitivity to the environment: the case of the serotonin transporter gene and its implications for studying complex diseases and traits. Am. J. Psychiatry 167, 509–527 37 Kim, M.J. and Whalen, P.J. (2009) The structural integrity of an amygdala-prefrontal pathway predicts trait anxiety. J. Neurosci. 29, 11614–11618 38 Young, K.A. et al. (2007) 5HTTLPR polymorphism and enlargement of the pulvinar: unlocking the backdoor to the limbic system. Biol. Psychiatry 61, 813–818 39 Risch, N. et al. (2009) Interaction between the serotonin transporter gene (5-HTTLPR), stressful life events, and risk of depression: a metaanalysis. JAMA 301, 2462–2471 40 Karg, K. et al. (2011) The Serotonin Transporter Promoter Variant (5HTTLPR), Stress, and Depression Meta-analysis Revisited: Evidence of Genetic Moderation. Arch. Gen. Psychiatry 68, 444–454

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 41 Roy, A. et al. (2007) Interaction between childhood trauma and serotonin transporter gene variation in suicide. Neuropsychopharmacology 32, 2046–2052 42 Lonsdorf, T.B. et al. (2009) Genetic gating of human fear learning and extinction: possible implications for gene-environment interaction in anxiety disorder. Psychol. Sci. 20, 198–206 43 Armbruster, D. et al. (2009) Serotonin transporter gene variation and stressful life events impact processing of fear and anxiety. Int. J. Neuropsychopharmacol. 12, 393–401 44 Brocke, B. (2006) Serotonin transporter gene variation impacts innate fear processing: acoustic startle response and emotional startle. Mol. Psychiatry 11, 1106–1112 45 Way, B. and Taylor, S. (2010) The serotonin transporter promoter polymorphism (5-HTTLPR) is associated with cortisol response to psychosocial stress. Biol. Psychiatry 67, 487–492 46 Gotlib, I.H. et al. (2008) HPA axis reactivity: a mechanism underlying the associations among 5-HTTLPR, stress, and depression. Biol. Psychiatry 63, 847–851 47 Alexander, N. et al. (2009) Gene-environment interactions predict cortisol responses after acute stress: implications for the etiology of depression. Psychoneuroendocrinology 34, 1294–1303 48 Mueller, A. et al. (2010) The role of the serotonin transporter polymorphism for the endocrine stress response in newborns. Psychoneuroendocrinology 35, 289–296 49 Spinelli, S. et al. (2007) Association between the recombinant human serotonin transporter linked promoter region polymorphism and behavior in rhesus macaques during a separation paradigm. Dev. Psychopathol. 19, 977–987 50 Jedema, H.P. et al. (2010) Cognitive impact of genetic variation of the serotonin transporter in primates is associated with differences in brain morphology rather than serotonin neurotransmission. Mol. Psychiatry 15, 512–522 51 Kalin, N.H. et al. (2008) The serotonin transporter genotype is associated with intermediate brain phenotypes that depend on the context of eliciting stressor. Mol. Psychiatry 13, 1021–1027 52 Murphy, D.L. and Lesch, K.P. (2008) Targeting the murine serotonin transporter: insights into human neurobiology. Nat. Rev. Neurosci. 9, 85–96 53 Homberg, J. et al. (2007) Characterization of the serotonin transporter knockout rat: a selective change in the functioning of the serotonergic system. Neuroscience 146, 1662–1672 54 Hariri, A.R. and Holmes, A. (2006) Genetics of emotional regulation: the role of the serotonin transporter in neural function. Trends Cogn. Sci. 10, 182–191 55 Ansorge, M. et al. (2004) Early-life blockade of 5-HT transporter alters emotional behavior in adult mice. Science 306, 879–881 56 Morgan, C.A., 3rd et al. (2000) Plasma neuropeptide-Y concentrations in humans exposed to military survival training. Biol. Psychiatry 47, 902–909 57 Zhou, Z. et al. (2008) Genetic variation in human NPY expression affects stress response and emotion. Nature 425, 997–1001 58 Mickey, B.J. et al. (2011) Emotion Processing, Major Depression, and Functional Genetic Variation of Neuropeptide Y. Arch. Gen. Psychiatry 68, 158–166 59 Kovacs, J.K. et al. (2000) Glucocorticoid negative feedback selectively targets vasopressin transcription in parovocellular neurosecretory neurons. J. Neurosci. 20, 3843–3852 60 Pratt, W.B. et al. (2006) Chaperoning of glucocorticoid receptors. Handb. Exp. Pharmacol. 172, 111–138 61 Grad, I. and Picard, D. (2007) The glucocorticoid responses are shaped by molecular chaperones. Mol. Cell Endocrinol. 275, 2–12 62 Binder, E.B. (2009) The role of FKBP5, a co-chaperone of the glucocorticoid receptor in the pathogenesis and therapy of affective and anxiety disorders. Psychoneuroendocrinology 34S, S186–S195 63 Binder, E.B. et al. (2004) Polymorphisms in FKBP5 are associated with increased recurrence of depressive episodes and rapid response to antidepressant treatment. Nat. Genet. 36, 1319–1325 64 Willour, V.L. et al. (2008) Family-based association of FKBP5 in bipolar disorder. Mol. Psychiatry 14, 261–268 65 Ising, M. et al. (2008) Polymorphisms in the FKBP5 gene region modulate recovery from psychosocial stress in healthy controls. Eur. J. Neurosci. 28, 389–398

407

Review 66 Binder, E.B. et al. (2008) Association of FKBP5 polymorphisms and childhood abuse with risk of posttraumatic stress disorder symptoms in adults. JAMA 299, 1–15 67 Roy, A. et al. (2010) Interaction of FKBP5, a Stress-Related Gene, with Childhood Trauma Increases the Risk for Attempting Suicide. Neuropsychopharmacology 35, 1674–1683 68 Ressler, K.J. et al. (2011) Post-traumatic stress disorder is associated with PACAP and the PAC1 receptor. Nature 470, 492–497 69 Goldman, D. et al. (2005) The genetics of addictions: uncovering the genes. Nat. Rev. Genet. 6, 521–532 70 Nicodemus, K.K. et al. (2007) Evidence for statistical epistasis between catechol-O-methyltransferase (COMT) and polymorphisms in RGS4, G72 (DAOA), GRM3, and DISC1: inﬂuence on risk of schizophrenia. Hum. Genet. 120, 889–906 71 Shifman, S. et al. (2008) A whole genome association study of neuroticism using DNA pooling. Mol. Psychiatry 13, 302–312 72 Hettema, J.M. et al. (2006) A population-based twin study of the relationship between neuroticism and internalizing disorders. Am. J. Psychiatry 163, 857–864 73 Goldstein, D.B. (2009) Common genetic variation and human traits. N. Eng. J. Med. 360, 1696–1698 74 Bevilacqua, L. et al. (2010) A population-speciﬁc HTR2B stop codon predisposes to severe impulsivity. Nature 468, 1061–1066 75 Eaves, L.J. et al. (1989) Genes, Culture and Personality: An Empirical Approach, Academic Press, (New York) 76 Biederman, J. et al. (2001) Further evidence of association between behavioral inhibition and social anxiety in children. Am. J. Psychiatry 158, 1673–1679 77 Caspi, A. and Silva, P.A. (1995) Temperamental qualities at age three predict personality traits in young adulthood: Longitudinal evidence from a birth cohort. Child. Dev. 66, 486–498 78 Fox, N.A. et al. (2005) Behavioral inhibition: Linking biology and behavior within a developmental framework. Annu. Rev. Psychol. 56, 235–262

408

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 79 Glahn, D.C. et al. (2007) Neuroimaging endophenotypes: strategies for ﬁnding genes inﬂuencing brain structure and function. Hum. Brain. Mapp. 28, 488–501 80 Fox, A.S. et al. (2008) Trait-like brain activity during adolescence predicts anxious temperament in primates. PLoS ONE 3, e2570 81 Oler, J.A. et al. (2010) Amygdalar and hippocampal substrates of anxious temperament differ in their heritability. Nature 466, 864–867 82 Rogers, J. et al. (2008) Genetic inﬂuences on behavioral inhibition and anxiety in juvenile rhesus macaques. Genes Brain Behav. 7, 463–469 83 Lipsky, R.H. et al. (2005) Association of COMT Val158Met genotype with executive functioning following traumatic brain injury. J. Neuropsychiatry Clin. Neurosci. 17, 465–471 84 Tunbridge, E.M. et al. (2006) Catechol-O-methyltransferase, cognition, and psychosis: Val158Met and beyond. Biol. Psychiatry 60, 141–151 85 Barnett, J.H. et al. (2008) Meta-analysis of the cognitive effects of the Catechol-O-Methyltransferase gene Val158/108Met polymorphism. Biol. Psychiatry 64, 137–144 86 Barnett, J.H. et al. (2011) Cognitive effects of genetic variation in monoamine neurotransmitter systems: a population-based study of COMT, MAOA, and 5HTTLPR. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 156, 158–167 87 Tunbridge, E.M. et al. (2004) Catechol-o-methyltransferase inhibition improves set-shifting performance and elevates stimulated dopamine release in the rat prefrontal cortex. J. Neurosci. 24, 5331–5335 88 Apud, J.A. et al. (2007) Tolcapone improves cognition and cortical information processing in normal human subjects. Neuropsychopharmacology 32, 1011–1020 89 Mattay, V.S. et al. (2003) Catechol-O-methyltransferase val158-met genotype and individual variation in the brain response to amphetamine. Proc. Natl. Acad. Sci., U.S.A. 100, 6186–6191

Review

Special Issue: The Genetics of Cognition

Genetics of autism spectrum disorders Daniel H. Geschwind1,2,3 1

Program in Neurogenetics, Department of Neurology, David Geffen School of Medicine, University of California, Los Angeles, CA 90095, USA 2 Center for Autism Research and Treatment, Semel Institute for Neuroscience and Human Behavior, David Geffen School of Medicine, University of California, Los Angeles, CA 90095, USA 3 Departments of Neurology, Psychiatry, and Human Genetics, David Geffen School of Medicine, University of California, Los Angeles, CA 90095, USA

Characterized by a combination of abnormalities in language, social cognition and mental flexibility, autism is not a single disorder but a neurodevelopmental syndrome commonly referred to as autism spectrum disorder (ASD). Several dozen ASD susceptibility genes have been identified in the past decade, collectively accounting for 10–20% of ASD cases. These findings, although demonstrating that ASD is etiologically heterogeneous, provide important clues about its pathophysiology. Diverse genetic and genomic approaches provide evidence converging on disruption of key biological pathways, many of which are also implicated in other allied neurodevelopmental disorders. Knowing the genes involved in ASD provides us with a crucial tool to probe both the specificity of ASD and the shared neurobiological and cognitive features across what are considered clinically distinct disorders, with the goal of linking gene to brain circuits to cognitive function. Autism genetics: a decade of progress In many ways, autism is a mysterious disorder because it involves core abnormalities in social cognition and language, both of which are central to what makes us human. Because of this, understanding autism will have a signiﬁcant impact on our basic knowledge of these fundamental cognitive processes, in addition to the obvious crucial role that such mechanistic understanding has on therapeutic development. Until the past decade, virtually nothing was known about the neurobiological basis of autism spectrum disorder (ASD). There was no consistent neuropathology and only scant knowledge of a few causal genetic factors. The past decade has brought an explosion of genetic ﬁndings in ASD, surpassing most other common neuropsychiatric disorders, so that we now have knowledge of the etiology of ASD for between 10 and 20% of cases. Here, I review these exciting ﬁndings with respect to what they tell us about the genetic architecture of ASD and how this informs our mechanistic understanding of the disorder and in a broader sense, human brain function. Definition and evolution of ASD Autism is a developmental neuropsychiatric syndrome with onset before the age of three. The fundamental conceptualization of the disorder is based on the initial observation of Corresponding author: Geschwind, D.H. ([email protected]).

Kanner in 1943 [1], where he described 11 children with autism, mostly boys with a combination of severe social and variable language dysfunction and the presence of repetitive restrictive behaviors. Kanner made numerous interesting observations based on these case studies, including the identiﬁcation of large head size in about half of the subjects and postulated a biological genetic basis for the disorder. However, until the 1980s autism was not considered a distinct disorder in the manuals of psychiatric diagnosis, nor was it considered by most to be biologically based. A major change in perspective came with the pioneering twin studies of Rutter and Folstein that demonstrated a Glossary Complex genetics: a term used in contrast to Mendelian patterns of inheritance, whereby instead of being dominantly or recessively inherited a phenotype is caused by multiple genes that could interact with each other and/ or the environment. Concordance: the rate at which a second individual has the same phenotype as the first, for example in the second sibling of a pair, or the second twin. Copy number variation: a greater than one Kb change in chromosomal complement so that a particular region or chromosome diverges from the normal diploid human copy number (aneuploidy) over that region. A deletion would yield one copy, and a duplication of the region would yield 3 copies. Copy number changes can be inherited or de novo. De novo: a genetic alteration that arises either in the gamete or in the fertilized egg during its early development but is not observed in the somatic cells of the parent. Endophenotype: a heritable phenotypic feature that is closely related to a disorder but represents a more narrow or simple component than the disorder itself. An endophenotype should be observed in first-degree relatives more frequently than in the general population and might be quantitative rather than binary. Epigenetic: non-sequence based changes to DNA that affect transcription, leading to changes in the expression of the gene product. Known major epigenetic modifications include DNA methylation and histone modifications, such as acetylation or methylation. Heritability: in general this refers to the extent to which a particular trait is due to inherited DNA sequence variation. A heritability of 1.0 indicates that a trait is 100% heritable and no environmental component contributes. However, heritability is an estimate based on the specific measurements taken in a particular population, and therefore is only a rough guide. Mendelian inheritance: a dominant or recessive pattern of inheritance of a trait first defined by Gregor Mendel. Penetrance: the frequency at which a particular phenotype is observed, given a specific genotype. Penetrance is always relative to the phenotype in question. Pleiotropy: occurs when a gene influences more than one phenotype. Polymorphism: a particular variant that is observed in more than 1% (common) in the population. A SNP is a single DNA base pair change. Proband: the index case being studied usually based on it having a particular phenotype or disorder. Whole genome or whole exome sequencing: the use of new high-throughput Next Generation Sequencing methods to sequence whole genomes or the protein coding portion of the genome (exome) in patients and controls.

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.003 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

409

Review genetic susceptibility to the disorder, providing incontrovertible evidence as to its biological origins [2,3]. Over the past two decades, the concept of autism has broadened from the strict diagnosis of autistic disorder to include those with normal intelligence and language (Asperger syndrome) and those almost meeting the strict diagnostic criteria in all three domains (pervasive development disorder – not otherwise speciﬁed, PDD-NOS). These diagnoses are not based on etiology but on expert observation and assessment of behavior and cognition; they can be considered arbitrary: which category one ﬁts into depends partially on subjective criteria. This is further emphasized by the current trend in the DSM-V (http://www.dsm5.org/proposedrevision/Pages/ NeurodevelopmentalDisorders.aspx), in which the category of Asperger syndrome is removed and the diagnostic criteria for autism are modiﬁed under the new heading of ASD. This change in diagnostic criteria is not based on known similarities or differences in causation between these clinically deﬁned categories but rather on the consensus of opinions of expert clinicians. Concurrently, over the past two decades, the notion that autism represents a quantitative spectrum of impairments, rather than representing discrete disorders, has gained increasing traction for researchers [4,5]. From this perspective, the term ASD attempts to crystalize the notion that patients represent a clinically variable population that suffers from pathologic levels of quantitative variation in the major cognitive and behavioral domains that are disrupted, rather than a distinct clinical disorder. How these two varying conceptualizations, autism as a unitary disorder versus a spectrum of dysfunction, relate to underlying etiologies is a key question facing the ﬁeld. Furthermore, how the clinical domains relate to underlying dysfunction in speciﬁc cognitive domains is essentially unknown, although some clues are starting to emerge [6,7]. Autism is a complex genetic disorder Perhaps the biggest advance in understanding autism pathophysiology has been the appreciation of a signiﬁcant genetic contribution to the etiology of ASD. Three main areas of evidence support a genetic etiology in ASD: twin studies, comparing monozygotic twins (MZ) and dizogotic twins (DZ), family studies comparing the rate of autism in ﬁrst degree relatives of affected probands (see Glossary) versus the population, and studies of rare genetic syndromes with a comorbid autism diagnosis. Because MZ twins share 100% of their genetic material and DZ twins share 50% (similar to nontwin siblings) and both share the in utero environment with their twin, higher disease cooccurrence in MZ twins than DZ twins supports a genetic etiology. This is what is observed in every twin study in ASD; overall consistent with heritability estimates of about 70–80% [8,9]. One exception is a very recent study with a large sample of twins, which, despite showing a concordance of about 0.6 for MZ twins and 0.25 for DZ twins (consistent with a Falconer heritability of 0.7), comes to the conclusion that shared environment plays a larger role than genetic factors [10]. Given that the concordance for strict autism in DZ twins (who can be considered genetic siblings who share the same in utero environment) is not clearly different from modern infant sibling recur410

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

rence rates [11], how shared environment would have a more major role than genetics is not clear. Moreover, studies in families show that ﬁrst-degree relatives of autistic probands have a markedly increased risk for autism relative to the population, consistent with a strong familial, or genetic, effect observed in twins [12]. This is not to dispute the role of the environment but to emphasize that genes play an important role. First-degree relatives of ASD probands have an increase in behavioral or cognitive features associated with autism, such as social or language dysfunction, albeit in lesser forms, when compared with the population prevalence [13]. This has been called ‘the broader phenotype’ and includes restrictive repetitive behaviors and subthreshold deﬁcits in social cognition, as well as language dysfunction [14–17]. For example, language delay (LD) is observed in a signiﬁcant proportion of nonautistic siblings of autistic probands [18,19]. Similarly, autistic-like social impairment clearly is heritable [20] and increased in unaffected parents and children of autistic probands [21]. Studies using multiple measures of subthreshold autistic traits in population cohorts suggest that different components, separately representing language, social function and repetitive or stereotyped behaviors contribute to ASD [22–24]. On aggregate, these data suggest that different features of autism represent a quantitative continuum of function that could be inherited in distinct patterns. This is consistent with the knowledge that speciﬁc genetic factors contribute to the development and function of speciﬁc brain structures, and the hypothesis that distinct brain circuits might underlie different components of autism [25]. It has also been known for several decades that many rare medical or genetic conditions are associated with autism. Dozens of genetic syndromes including Joubert Syndrome, Smith–Lemli–Opitz syndrome, Tuberous Sclerosis and Fragile X are known to cause autism, although many with less than 50% penetrance [26,27]. Although this provided strong evidence in favor of a genetic cause for ASD, these syndromic forms were considered exceptional cases, and not relevant for common forms of idiopathic autism because each was rare; none account for more than 1% of ASD cases, and most are far less common [27]. Instead, similar to other common diseases with genetic contributions, autism was thought to ﬁt a model in which multiple common variants, each with small to moderate effect sizes, interact with each other and perhaps in some cases, environmental factors, to lead to autism; a situation referred to as complex genetics [27]. In contrast to Mendelian genetics, under the complex genetic model, speciﬁc common variants in genes increase susceptibility to ASD but each is not on its own sufﬁcient to be causal (Figure 1). These same variants contribute to normal variation in cognition and behavior in unaffected individuals but when combined unfortunately as they are in the autistic proband, they cause ASD. This model would explain the presence of subthreshold traits in nonautistic ﬁrst-degree relatives, such as the siblings or parents of autistic individuals, because they would be expected to harbor a subset of the genetic variants that cause autism in the affected individual.

(Figure_1)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Effect size Mutation screens Frequency

Association studies TRENDS in Cognitive Sciences

Figure 1. Common and rare variants. We usually think of common and rare variants representing two extremes on the allelic spectrum. SNPs with modest effects sizes (gray oval) have been identified through association studies. Rare events that segregate with disease have also been identified through resequencing efforts and CNV studies. Less attention has been directed at events not at either extreme but also likely to modulate risk. Larger studies will permit consideration of lower frequency intermediate effect alleles (ovals with dotted lines). Current data do not support the presence of common alleles of large effect for affection status (bottom right corner). However, it is possible that such associations might be observed in the future when appropriate quantitative endophenotypes are used. Adapted with permission from [87].

The role of rare mutations versus common polymorphisms (CPs) in ASD A series of important ﬁndings over the past four years clearly challenges the notion that autism is mainly caused by combinations of common variants by identifying a large number of rare, recurrent and nonrecurrent mutations that lead to ASD. At the same time, whole genome association studies with common variants, although identifying a few loci with very small effect sizes, have not yielded independently replicated results [28,29]. These rare mutations, mostly in the form of submicroscopic chromosomal structural variation, called copy number variants (CNV), are now known to account for up to 10% of cases of idiopathic autism (those with no obvious clinical syndrome) [30–34]. Because many of these CNV have large effect sizes and thus are thought sufﬁcient to cause ASD, they are predicted to signiﬁcantly reduce reproductive ﬁtness. Consistent with this, these causal CNV are often not transmitted from the parent but instead occur de novo in the germline [33,35]. However, in some cases, such as CNV at 16p11 and 15q11–13, the CNV are transmitted from an unaffected parent to cause the disorder in an offspring [33]. The genetic or epigenetic mechanism of reduced penetrance for ASD in the mutation-carrying parent is not known. However, it is also probable that the parent carriers of such CNV have more subtle neuropsychiatric or cognitive phenotypes that have not yet been systematically identiﬁed. The ﬁrst key issue in genotype–phenotype mapping in ASD is reconciling the role of causal rare de novo variants that occur in probands and are not inherited by the unaffected siblings with the observations of broader subthreshold traits in siblings. The latter suggests the role of common variation but as mentioned above, the genetic evidence for speciﬁc common variants in ASD so far has been sparse. De novo events are major-effect genetic mutations that are mainly observed in probands, and are therefore not expected to contribute to the broader phenotype. In

this regard, two observations could be informative. First, de novo CNV are observed in 5–10% of ASD probands but they are also observed in 1–2% of unaffected controls; so all such CNV are not necessarily causal or fully penetrant [36]. From this perspective, some CNV could be acting as complex genetic risk factors, with intermediate effect sizes, and variable penetrance and expressivity. Such is certainly the case with 16p11 duplications, which are observed in ASD, schizophrenia (SZ) and various forms of developmental delay (DD) [37–39], as well as Neurexin 1 CNV, which are observed in about 50% of unaffected carriers [36], and observed across several neurodevelopmental disorders (Table 1). Such pleiotropy and variable penetrance and expressivity are the rule, rather than the exception (Table 1). Second, the frequency of de novo mutations is signiﬁcantly lower in multiplex (familial) cases of autism versus those with only one affected proband, so called simplex families [30,31,34]. This difference in the frequency of rare de novo CNV in families with different structures (simplex vs multiplex) suggests that the contribution from different types of genetic variation between simplex and multiplex autism might differ. Recent family data based on the social responsiveness scale, which is a quantitative measure of autism-related features heavily weighted on social factors, suggests that this is indeed the case [40]. This discussion of rare and common genetic variants highlights an area of major tension in autism genetic research. The causal nature of Mendelian (recessive or dominantly acting) mutations is fairly easy to establish because there is essentially a strong correspondence between the presence of a mutation and a disease phenotype. Certainly, dominant or recessive mutations are easier to model in vitro and in vivo. However, few such variants cause autism alone but instead could cause a wide variety of phenotypes (variable expressivity) with or without autism, ranging from intellectual disability (ID), epilepsy, psychosis and global DD to neurotypical [27,41]. Thus, even for the simpler Mendelian mutations, understanding which aspects of the phenotype observed in a model system or a human patient relate to autism, and which might relate to more global aspects of cognitive dysfunction, needs to be carefully established. This provides a great opportunity for detailed phenotypic study of patients with and without autism who harbor such rare causal mutations, so as to connect speciﬁc aspects of brain structure and function with cognitive and behavioral phenotypes that underlie the clinical features of the disease. Common variants, although not causal for the disease, seem to have more subtle effects, making them harder to model in vitro or in animal models. However, perhaps because they are common in human populations and could have more subtle or speciﬁc effects on particular aspects of cognition and behavior, common risk variants might be easier to relate to particular brain phenotypes or circuits in humans. This issue will be discussed below in more detail with respect to imaging-genetic ﬁndings. Last, it should also be emphasized that although they are often presented as dichotomous models, the contributions of rare and common variants to ASD in an individual are not mutually exclusive [25]. For example, although rare de novo CNV are considered causal mutations, it is 411

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Table 1. Pleiotropic effects of major genes/mutations associated with ASD and allied neurodevelopmental disorders Gene/region NRXN1 CNTNAP2 16pdel 16pdup 15q13.3del 17q12del 15q11–13dup 22q11 1q21

Mechanism CNV, PM CNV, PM, CP CNV CNV CNV CNV CNV CNV CNV

Disorders ASD, SZ ASD, ID, epilepsy, LD/SLI, TS ASD, SZ, DD, LD, normal carrier SZ, ID, DD, LD, ADHD, normal carrier SZ, epilepsy, ASD, normal carrier SZ, ASD, ID ASD, SZ/psychosis ASD, ADHD, SZ, ID, epilepsy ASD, SZ, ID, epilepsy

References [36], [50,72,73] [59,61,74,75] [33,37,42,76–78] [33,39,42,78] [42,73,79] [80] [33,81] [73,82–84] [42,85,86]

Here we list genes, form of genetic risk variant, clinical disorders where the mutation has been observed, and some representative references. This table is not meant to be exhaustive but illustrative of the pleiotropic effects of known ASD genes or loci with relatively large effect sizes (OR > 5–10 for ASD). New abbreviations are as follows: TS, Tourette syndrome and PM, point mutation.

clear that mutations could have intermediate effect sizes and variable penetrance, as mentioned above [36,42], so such mutations must be interacting with other factors. Penetrance less than 50% for ASD is also observed in many Mendelian forms of ASD, such as Fragile X, Jouberts Syndrome and Tuberous Sclerosis Complex (TSC), so other genetic, epigenetic or environmental factors must be contributing to autism causation in such cases as well [27]. Thus, different rare variants could act as either susceptibility or strongly causal alleles, and, in conjunction with common variation, might contribute in differing proportions to ASD risk in individual patients [25]. Fortunately, over the next several years, whole genome or exome sequencing, if performed in large enough populations, will settle this issue by providing an empiric assessment of the relative contributions of both rare and common alleles of different effect sizes [43,44]. The molecular diversity of ASD Several recent reviews summarize the growing list of dozens of common and rare genetic variants that have been associated with ASD at varying levels of statistical evidence [27,41,45–47], so I refer the reader to these reviews for comprehensive gene lists. Here, instead, I will try to synthesize what these ﬁndings might be telling us about ASD pathophysiology at what is still an early juncture in the search for ASD susceptibility genes. The most obvious general conclusion from all of the published genetic studies is the extraordinary etiological heterogeneity of ASD. No speciﬁc gene accounts for the majority of ASD; rather, even the most common genetic forms account for not more than 1–2% of cases [27]. Further, these genes, including those mentioned earlier, represent a diversity of molecular mechanisms, ranging from cell adhesion, synaptic vesicle release and neurotransmission, synaptic structure, RNA processing/splicing, and activity-dependent protein translation. On one hand, this should not be surprising because autism is deﬁned based on observation of cognition and behavior, not etiology. On the other hand, the diversity of potential mechanisms and the apparent lack of speciﬁcity of mutations for ASD begs the question as to whether ASD should even be viewed as a unitary disorder. Therefore, asking whether the diverse genes implicated in autism might converge on common pathways becomes an important question for understanding autism and developing new therapeutics. 412

Converging circuits and common pathways Convergence of pathways could be assessed at many levels, ranging from molecular mechanisms to brain circuitry. The earliest synthetic molecular model was based on the notion that the primary area of convergence in ASD was the postnatal, experience-dependent development of the synapse [48]. This was a highly productive model based on several forms of Mendelian mutations in ASD, which has led to successful exploration of synaptic scaffolding molecules and other synaptic genes as ASD susceptibility genes [27,46]. Interestingly, because many mRNA for synaptic genes are alternatively spliced and the proteins regulated by ubiquitination, this paper was prescient in predicting alterations in mRNA splicing and ubiquitination pathways that have since been identiﬁed [49,50]. Nevertheless, because the synapse is integral to the function of all neurons, how the cognitive and behavioral speciﬁcity of ASD emerges distinctly from ID or other synaptic disorders, is not yet explained by this molecularly focused model. Belmonte and colleagues have articulated a cogent unifying model based on imaging, electrophysiological and anatomical data, which implicates abnormal neural connectivity in disrupting timing and information processing [51,52]. This general model was subsequently anatomically reﬁned and modiﬁed to focus primarily on reduced long-range, and increased short-range frontal lobe connectivity [53]. Since the general connectivity model was proposed, many subsequent studies provide strong supporting data, including a recent study showing weak interhemispheric functional connectivity in young toddlers with ASD [54]. This study is particularly important because it shows altered connectivity in developing language circuits to be an early (and thus potentially causal) feature of the disorder, which could not be inferred from studies of older children or adults. A more recent model [55] synthesizes several key features of the synaptic and connectivity models above in conjunction with genetic ﬁndings, and is based on developmental disconnection of frontal and temporal cortical regions. This model strives to explain the diversity of genetic ﬁndings without necessitating a particular molecular mechanism. Instead, it relies on a behavioral neurologic framework to integrate many potential levels of molecular dysfunction, from synapse formation and maturation, vesicle trafﬁcking and signaling, to defective neuronal migration, interneuron development, dendritic

(Figure_2)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Stochastic or Environmental processes

(dup) 15q11-13 •Abnormal neuronal migration TSC1

•Abnormalities in patterning (regionalization or connectivity)

FMR1

•Faulty wiring/axon pathfinding •Aberrant synaptogenesis or

(del) 22q

synaptic function

Developmental disconnection

of critical brain circuits: •Fronto-striatal •Fronto-temporal

•Dendritic abnormalities (del) 16p

•Fronto-parietal

•Dysfunctional neural transmission

CACNA1C

ASD Genetic modifier loci (CNTNAP2, MET, rs4307059, etc…) TRENDS in Cognitive Sciences

Figure 2. Convergence of genes on neural systems. Here we illustrate a working model for how autism risk alleles with major effects, such as those shown at the left, might act to lead to autism. Because these Mendelian conditions are not specific for ASD and typically lead to ID, either environmental or genetic factors must modify their effects on brain development. The biological pathways through which such genes are known to act (shown in the center box) are myriad. Although gene expression and proteomic studies identify molecular and biological pathways that provide a source of convergence, the ultimate convergence must lie in neural systems. At a neural systems level, the convergent process will likely be disconnection of the circuits outlined in the far right box because these systems are thought to underlie the core deficits of ASD. Adapted with permission from [88].

maturation or axon path ﬁnding, all of which could contribute to functional disconnection [55] (Figure 2). Indeed several of the monogenic causes of ASD, including TSC [56] and Joubert Syndrome [57,58] show manifestations of abnormal axon path ﬁnding or migration abnormalities that are known to lead to alterations in brain structural connectivity. The study of common variants, for example in the gene CNTNAP2, also supports disconnection as a causal phenomenon based on initial imaging-genetic evidence. Two SNPs in a relatively circumscribed area of CNTNAP2 have been shown to modulate language function in ASD and other conditions [59–61]. Scott-Van Zeeland and colleagues [62] showed long-range disconnection of medial prefrontal cortex and the precuneus, and shortrange overconnectivity of the prefrontal region in carriers of the CNTNAP2 risk allele, regardless of whether they were autistic or neurotypical. This ﬁnding is particularly important because the genetic mediation of this disconnection supports its causal nature. Endophenotypes, common variants and domain specificity in ASD The relations between speciﬁc genetic variants and speciﬁc cognitive processes, such as language, highlight the notion

that the broad syndrome of ASD can be broken down into many component or intermediate phenotypes, referred to as endophenotypes. The familial segregation of endophenotypes provides a genetic basis for the broader phenotype described earlier. A logical extension of this concept is that these endophenotypes represent one end of the continuum of the normal spectrum of behavior and cognition [25]. Several groups have demonstrated that this is indeed the case with respect to CNTNAP2 [60,61,63] because variants in the same region of the gene that increase risk for LD in ASD, increase risk for speciﬁc language impairment (SLI) and modify language ability in the general population. Similarly, CNTNAP2 single nucleotide polymorphism (SNP) [64] has been shown to modulate brain morphology in several ASD-related cortical regions in normal controls [65]. However, speciﬁcity for ASD is not only an issue for rare Mendelian variants but also for common variation [66]. Because the effects of genes on behavior are mediated through neural circuits that mediate a multiplicity of functions (e.g. frontal-striatal circuits), their pleiotropic effects on behavior and cognition should not be surprising. In fact, pleiotropy should be the rule, rather than the exception. Recently, CNTNAP2 variants have been associ413

Review ated with mutism and social anxiety [67]. This complex relation between genetic variation and phenotype highlights the importance of in-depth exploration of cognitive and neurobehavioral phenotypes in ASD patients, their family members and in the general population. Phenotype discovery and the relation of core ASD-related endophenotypes to each other and speciﬁc brain circuits in patients with ASD and related neurodevelopmental disorders remains a key but underexplored area. Evidence for converging molecular pathways Several recent studies have suggested that in addition to convergent brain pathways, there could as well be convergence at the level of molecular mechanisms in ASD. One class of such studies has asked whether putative ASD susceptibility genes are enriched in members for speciﬁc molecular or biological processes more than expected by chance. The value of this approach depends on the level of experimental support for the speciﬁc genes tested and the degree to which current pathway annotations represent reality [25,68]. For genes identiﬁed within CNV this can be particularly problematic because most known pathological CNV contain more than one gene and it is not expected that all genes within the CNV contribute to ASD, potentially increasing noise in this analysis. One recent study [69] reduced such background by using a new phenotype-driven method to group genes within high conﬁdence de novo CNV [34,69], identifying signiﬁcant enrichment for several categories of genes, including axon outgrowth, synaptogenesis, cell–cell adhesion, GTPase signaling and the actin cytoskeleton. These results replicate and extend earlier composite pathway analysis of putative ASD susceptibility genes compiled from the literature [68], and CNV pathway analysis in the Autism Genetic Resource Exchange and other cohorts [32,36]. However, these studies place ASD genes within a multiplicity of pathways, several of which are broad and do not necessarily demonstrate convergence on ﬁnal common molecular processes in individuals. In this regard, two recent studies use different systems biology approaches to provide a new perspective on the concept of molecular convergence. The ﬁrst, an analysis of gene expression in postmortem autism brain, provides strong evidence for a shared set of molecular alterations in a majority of cases of ASD. This included disruption of the normal gene expression pattern that differentiates frontal and temporal lobes (consistent with an early developmental patterning defect), and two groups of genes dysregulated in ASD brains: one related to neuronal function, and the other to immune/inﬂammatory responses [49]. The neuronal function genes were enriched in genetic association signals, providing evidence that these changes were causal, rather than the consequence of the disease [49]; in contrast, the immune/inﬂammatory changes did not show a strong genetic signal, implicating environmental or epigenetic factors instead. It is also notable that the several of the same biological pathways identiﬁed in this gene expression study overlapped with the pathway analysis of CNV described above. This analysis of postmortem autism brain also showed downregulation of several markers of GABAergic interneurons, suggesting potential inhibitory interneuron dysfunction. These results provide 414

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

the ﬁrst strong evidence for both a shared genetic and an environmental/epigenetic basis for ASD and the presence of an early developmental patterning defect. It is tempting to speculate that the abnormalities in cortical patterning and interneuron function provide a molecular basis for cortical–cortical and cortical–fugal disconnection, linking molecular abnormalities with anatomical, physiological and imaging ﬁndings. The second study on molecular convergence in ASD identiﬁed protein interactors of known ASD or ASD-associated genes [70]. This interactome revealed several novel interactions, including between two ASD candidate genes, Shank3 and TSC1. The biological pathways identiﬁed in this study include synapse, cytoskeleton and GTPase signaling, demonstrating a remarkable overlap with those identiﬁed by the gene expression and CNV pathway studies discussed above. This study differs from other genomewide studies (such as mRNA expression) because it begins with known ASD genes and asks about their relations. So although it is ‘–-omic’ in nature [71], it is not as unsupervised as methods that survey the genome more agnostically. Despite the signiﬁcant heterogeneity in ASD, these studies using diverse methods identify several common areas of molecular convergence in ASD. Understanding how these pathways relate to individual differences now becomes an important research priority. Concluding remarks Many genes have been identiﬁed for ASD and some biologically coherent functional pathways that link these genes are emerging. Few of the genes, whether contributing by common or rare variation, are speciﬁc to ASD but instead variably contribute to genetic risk for ASD, ID, SZ, SLI, epilepsy and even attention deﬁcit hyperactivity disorder (ADHD). Therefore, understanding the speciﬁcity of individual variants to ASD is a crucial challenge that requires several advances to be made (Box 1). Diseases of cognition and behavior have their basis in brain circuit dysfunction. Therefore, we need to understand how speciﬁc genetic risk variants lead to changes in neural circuitry and function in those with and without a speciﬁc diagnosis. This will probably involve understanding gene–gene and gene–environment interactions, as well as epigenetics, areas that have not yet been explored in ASD. We need to understand how Box 1. Priorities for future genetic research in ASD Delineate areas of molecular convergence between diverse mutational mechanisms underlying ASD susceptibility using unbiased systems biology methods. Perform whole genome sequencing in large numbers of patients with a variety of neurodevelopmental and psychiatric disorders to clearly define genetic overlap and differences in susceptibility. Define epigenetic contributions to ASD. Define appropriate endophenotypes through family case-control designs, so as to have a sound basis for understanding cross disorder genotype–phenotype relations. Translate the knowledge of mutational basis of ASD into mechanistic understanding at the cellular and circuit levels. Characterize subtypes of ASD based on molecular or genetic signatures and relate them to trajectory and treatment response. Identify environmental modifiers and gene- by- environment interactions.

Review the implicated neural circuitry relates to speciﬁc cognitive and behavioral elements, and we need to link the neural circuitry and behaviors to disease. Additionally, such circuits are related to speciﬁc aspects of human higher cognition and a more detailed notion of the autism cognitive phenotype is required. Components of this understanding have been achieved in a preliminary manner for only a few elements but to understand ASD from a systems level mechanistic perspective, we will need to integrate all of these levels of understanding [71]. This will no doubt involve the use of model systems, including mouse and invertebrates, so it will be important to be cognizant of evolutionary similarity and divergence in the interpretation of such data with respect to a disorder of human higher cognition. Regardless of the challenges, the identiﬁcation of causal genetic variants provides a starting point for a new mechanistic understanding of ASD, enabling connection of genes to brain to cognition and behavior. Acknowledgments I thank Lauren Kawaguchi for her editorial assistance and Brett Abrahams for permission to use a previously published ﬁgure. Our laboratory’s work in autism and the molecular basis of human higher cognition is supported by two Autism Center of Excellence grants from NIMH: ACE Network 5R01MH081754-03 to D.H.G and ACE Center 5P50HD055784-03 (S. Bookheimer PI, D.H.G. co-PI), a Merit award from NIMH to D.H.G., 4R37MH060233-10, as well as funding from Autism Speaks and The Simons Foundation.

References 1 Kanner, L. (1943) Autistic disturbances of affective contact. Nerv. Child 2, 217–250 2 Folstein, S. and Rutter, M. (1977) Infantile autism: a genetic study of 21 twin pairs. J. Child Psychol. Psychiatry 18, 297–321 3 Folstein, S. and Rutter, M. (1977) Genetic inﬂuences and infantile autism. Nature 265, 726–728 4 Wing, L. (1988) The continuum of autistic characteristics. In Diagnosis and Assessment in Autism (Schopler, E.M.G., ed.), pp. 91–110, Plenum 5 Constantino, J.N. and Todd, R.D. (2003) Autistic traits in the general population: a twin study. Arch. Gen. Psychiatry 60, 524–530 6 Mundy, P. et al. (2010) Self-referenced processing, neurodevelopment and joint attention in autism. Autism 14, 408–429 7 Charman, T. et al. (2011) Deﬁning the cognitive phenotype of autism. Brain Res. 1380, 10–21 8 Bailey, A. et al. (1995) Autism as a strongly genetic disorder: evidence from a British twin study. Psychol. Med. 25, 63–77 9 Rosenberg, R.E. et al. (2009) Characteristics and concordance of autism spectrum disorders among 277 twin pairs. Arch. Pediatr. Adolesc. Med. 163, 907–914 10 Hallmayer, J. et al. (2011) Genetic heritability and shared environmental factors among twin pairs with autism. Arch. Gen. Psychiatry DOI: 10.1001/archgenpsychiatry.2011.76 11 Brian, J. et al. (2008) Clinical assessment of autism in high-risk 18month-olds. Autism 12, 433–456 12 Bolton, P. et al. (1994) A case-control family history study of autism. J. Child Psychol. Psychiatry 35, 877–900 13 Losh, M. et al. (2009) Neuropsychological proﬁle of autism and the broad autism phenotype. Arch. Gen. Psychiatry 66, 518–526 14 Warren, Z.E. et al. (2011) Neurocognitive and behavioral outcomes of younger siblings of children with Autism Spectrum Disorder at age ﬁve. J. Autism Dev. Disord. DOI: 10.1007/s10803-011-1263-4 15 Constantino, J.N. (2011) The quantitative nature of autistic social impairment. Pediatr. Res. 69, 55R–62R 16 Gamliel, I. et al. (2009) Developmental trajectories in siblings of children with autism: cognition and language from 4 months to 7 years. J. Autism Dev. Disord. 39, 1131–1144 17 Pickles, A. et al. (2000) Variable expression of the autism broader phenotype: ﬁndings from extended pedigrees. J. Child Psychol. Psychiatry 41, 491–502

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 18 Gamliel, I. et al. (2007) The development of young siblings of children with autism from 4 to 54 months. J. Autism Dev. Disord. 37, 171–183 19 Ben-Yizhak, N. et al. (2011) Pragmatic language and school related linguistic abilities in siblings of children with autism. J. Autism Dev. Disord. 41, 750–760 20 Constantino, J.N. and Todd, R.D. (2005) Intergenerational transmission of subthreshold autistic traits in the general population. Biol. Psychiatry 57, 655–660 21 Constantino, J.N. et al. (2006) Autistic social impairment in the siblings of children with pervasive developmental disorders. Am. J. Psychiatry 163, 294–296 22 Steer, C.D. et al. (2010) Traits contributing to the autistic spectrum. PLoS ONE 5, e12633 23 Ronald, A. et al. (2006) Genetic heterogeneity between the three components of the autism spectrum: a twin study. J. Am. Acad. Child Adolesc. Psychiatry 45, 691–699 24 Ronald, A. et al. (2010) A twin study of autism symptoms in Sweden. Mol. Psychiatry DOI: 10.1038/mp.2010.82 25 Geschwind, D.H. (2008) Autism: many genes, common pathways? Cell 135, 391–395 26 Gillberg, C. and Coleman, M. (1985) The Biology of the Autistic Syndromes, Greenwood Publishing Group 27 Abrahams, B.S. and Geschwind, D.H. (2008) Advances in autism genetics: on the threshold of a new neurobiology. Nat. Rev. Genet. 9, 341–355 28 Wang, K. et al. (2009) Common genetic variants on 5p14.1 associate with autism spectrum disorders. Nature 459, 528–533 29 Anney, R. et al. (2010) A genome-wide scan for common alleles affecting risk for autism. Hum. Mol. Genet. 19, 4072–4082 30 Sebat, J. et al. (2007) Strong association of de novo copy number mutations with autism. Science 316, 445–449 31 Marshall, C.R. et al. (2008) Structural variation of chromosomes in autism spectrum disorder. Am. J. Hum. Genet. 82, 477–488 32 Pinto, D. et al. (2010) Functional impact of global rare copy number variation in autism spectrum disorders. Nature 466, 368–372 33 Sanders, S.J. et al. (2011) Multiple recurrent de novo CNVs, including duplications of the 7q11.23 Williams Syndrome region, are strongly associated with autism. Neuron 70, 863–885 34 Levy, D. et al. (2011) Rare de novo and transmitted copy-number variation in autistic spectrum disorders. Neuron 70, 886–897 35 Beaudet, A.L. (2007) Autism: highly heritable but not inherited. Nat. Med. 13, 534–536 36 Bucan, M. et al. (2009) Genome-wide analyses of exonic copy number variants in a family-based study point to novel autism susceptibility genes. PLoS Genet. 5, e1000536 37 Fernandez, B.A. et al. (2010) Phenotypic spectrum associated with de novo and inherited deletions and duplications at 16p11.2 in individuals ascertained for diagnosis of autism spectrum disorder. J. Med. Genet. 47, 195–203 38 Bijlsma, E.K. et al. (2009) Extending the phenotype of recurrent rearrangements of 16p11.2: deletions in mentally retarded patients without autism and in normal individuals. Eur. J. Med. Genet. 52, 77–87 39 McCarthy, S.E. et al. (2009) Microduplications of 16p11.2 are associated with schizophrenia. Nat. Genet. 41, 1223–1227 40 Virkud, Y.V. et al. (2009) Familial aggregation of quantitative autistic traits in multiplex versus simplex autism. Am. J. Med. Genet. B: Neuropsychiatr. Genet. 150B, 328–334 41 Betancur, C. (2011) Etiological heterogeneity in autism spectrum disorders: more than 100 genetic and genomic disorders and still counting. Brain Res. 1380, 42–77 42 Vassos, E. et al. (2010) Penetrance for copy number variants associated with schizophrenia. Hum. Mol. Genet. 19, 3477–3481 43 Gibb, B.C. (2009) Teetering towards chaos and complexity. Nat. Chem. 1, 17–18 44 Zeggini, E. (2011) Next-generation association studies for complex traits. Nat. Genet. 43, 287–288 45 Speicher, M.R. et al., eds (2009) Vogel and Motulsky’s Human Genetics: Problems and Approaches, Springer 46 State, M.W. (2010) The genetics of child psychiatric disorders: focus on autism and Tourette syndrome. Neuron 68, 254–269 47 Cook, E.H., Jr and Scherer, S.W. (2008) Copy-number variations associated with neuropsychiatric conditions. Nature 455, 919–923

415

Review 48 Zoghbi, H.Y. (2003) Postnatal neurodevelopmental disorders: meeting at the synapse? Science 302, 826–830 49 Voineagu, I. et al. (2011) Transcriptomic analysis of autistic brain reveals convergent molecular pathology. Nature 474, 380–384 50 Glessner, J.T. et al. (2009) Autism genome-wide copy number variation reveals ubiquitin and neuronal genes. Nature 459, 569–573 51 Belmonte, M.K. et al. (2004) Autism and abnormal development of brain connectivity. J. Neurosci. 24, 9228–9231 52 Belmonte, M.K. et al. (2004) Autism as a disorder of neural information processing: directions for research and targets for therapy. Mol. Psychiatry 9, 646–663 53 Courchesne, E. and Pierce, K. (2005) Why the frontal cortex in autism might be talking only to itself: local over-connectivity but long-distance disconnection. Curr. Opin. Neurobiol. 15, 225–230 54 Dinstein, I. et al. (2011) Disrupted neural synchronization in toddlers with autism. Neuron 70, 1218–1225 55 Geschwind, D.H. and Levitt, P. (2007) Autism spectrum disorders: developmental disconnection syndromes. Curr. Opin. Neurobiol. 17, 103–111 56 Tsai, P. and Sahin, M. (2011) Mechanisms of neurocognitive dysfunction and therapeutic considerations in tuberous sclerosis complex. Curr. Opin. Neurol. 24, 106–113 57 Poretti, A. et al. (2011) Joubert Syndrome and related disorders: spectrum of neuroimaging ﬁndings in 75 patients. AJNR Am. J. Neuroradiol. DOI: 10.3174/ajnr.A2517 58 Lee, J.E. and Gleeson, J.G. (2011) Cilia in the nervous system: linking cilia function and neurodevelopmental disorders. Curr. Opin. Neurol. 24, 98–105 59 Alarcon, M. et al. (2008) Linkage, association, and gene-expression analyses identify CNTNAP2 as an autism-susceptibility gene. Am. J. Hum. Genet. 82, 150–159 60 Whitehouse, A.J. et al. (2011) CNTNAP2 variants affect early language development in the general population. Genes Brain Behav. 10, 451–456 61 Vernes, S.C. et al. (2008) A functional genetic link between distinct developmental language disorders. N. Engl. J. Med. 359, 2337–2345 62 Scott-Van Zeeland, A.A. et al. (2010) Altered functional connectivity in frontal lobe circuits is associated with variation in the autism risk gene CNTNAP2. Sci. Transl. Med. 2, 56–80 63 Peter, B. et al. (2011) Replication of CNTNAP2 association with nonword repetition and support for FOXP2 association with timed reading and motor activities in a dyslexia family sample. J. Neurodev. Disord. 3, 39–49 64 Arking, D.E. et al. (2008) A common genetic variant in the neurexin superfamily member CNTNAP2 increases familial risk of autism. Am. J. Hum. Genet. 82, 160–164 65 Tan, G.C. et al. (2010) Normal variation in fronto-occipital circuitry and cerebellar structure with an autism-associated polymorphism of CNTNAP2. Neuroimage 53, 1030–1042 66 State, M.W. (2011) The erosion of phenotypic speciﬁcity in psychiatric genetics: emerging lessons from CNTNAP2. Biol. Psychiatry 69, 816–817 67 Stein, M.B. et al. (2011) A common genetic variant in the neurexin superfamily member CNTNAP2 is associated with increased risk for selective mutism and social anxiety-related traits. Biol. Psychiatry 69, 825–831 68 Bill, B.R. and Geschwind, D.H. (2009) Genetic advances in autism: heterogeneity and convergence on shared pathways. Curr. Opin. Genet. Dev. 19, 271–278

416

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 69 Gilman, S.R. et al. (2011) Rare de novo variants associated with autism implicate a large functional network of genes involved in formation and function of synapses. Neuron 70, 898–907 70 Sakai, Y. et al. (2011) Protein interactome reveals converging molecular pathways among autism disorders. Sci. Transl. Med. 3, 86ra49 71 Geschwind, D.H. and Konopka, G. (2009) Neuroscience in the era of functional genomics and systems biology. Nature 461, 908–915 72 Gauthier, J. et al. (2011) Truncating mutations in NRXN2 and NRXN1 in autism spectrum disorders and schizophrenia. Hum. Genet. DOI: 10.1007/s00439-011-0975-z 73 Levinson, D.F. et al. (2011) Copy number variants in schizophrenia: conﬁrmation of ﬁve previous ﬁndings and new evidence for 3q29 microdeletions and VIPR2 duplications. Am. J. Psychiatry 168, 302–316 74 Strauss, K.A. et al. (2006) Recessive symptomatic focal epilepsy and mutant contactin-associated protein-like 2. N. Engl. J. Med. 354, 1370–1377 75 Verkerk, A.J. et al. (2003) CNTNAP2 is disrupted in a family with Gilles de la Tourette syndrome and obsessive compulsive disorder. Genomics 82, 1–9 76 Kumar, R.A. et al. (2008) Recurrent 16p11.2 microdeletions in autism. Hum. Mol. Genet. 17, 628–638 77 Weiss, L.A. et al. (2008) Association between microdeletion and microduplication at 16p11.2 and autism. N. Engl. J. Med. 358, 667–675 78 Shinawi, M. et al. (2010) Recurrent reciprocal 16p11.2 rearrangements associated with global developmental delay, behavioural problems, dysmorphism, epilepsy, and abnormal head size. J. Med. Genet. 47, 332–341 79 van Bon, B.W.M., et al. (1993) 15q13.3 Microdeletion. (Pagon RA, B.T., Dolan CR, Stephens K, ed), University of Washington, Seattle 80 Moreno-De-Luca, D. et al. (2010) Deletion 17q12 is a recurrent copy number variant that confers high risk of autism and schizophrenia. Am. J. Hum. Genet. 87, 618–630 81 Mefford, H.C. et al. (2010) Genome-wide copy number variation in epilepsy: novel susceptibility loci in idiopathic generalized and focal epilepsies. PLoS Genet. 6, e1000962 82 Fine, S.E. et al. (2005) Autism spectrum disorders and symptoms in children with molecularly conﬁrmed 22q11.2 deletion syndrome. J. Autism Dev. Disord. 35, 461–470 83 Niklasson, L. et al. (2009) Autism, ADHD, mental retardation and behavior problems in 100 individuals with 22q11 deletion syndrome. Res. Dev. Disabil. 30, 763–773 84 Vorstman, J.A. et al. (2006) The 22q11.2 deletion in children: high rate of autistic disorders and early onset of psychotic symptoms. J. Am. Acad. Child Adolesc. Psychiatry 45, 1104–1113 85 Mefford, H.C. et al. (2008) Recurrent rearrangements of chromosome 1q21.1 and variable pediatric phenotypes. N. Engl. J. Med. 359, 1685–1699 86 Stefansson, H. et al. (2008) Large recurrent microdeletions associated with schizophrenia. Nature 455, 232–236 87 Abrahams, B.S. and Geschwind, D.H. (2010) Genetics of autism. In Vogel and Motulsky’s Human Genetics (Speicher, M.R. et al., eds), pp. 699–714, Springer 88 Geschwind, D.H. (2011) Autism genetics and genomics: a brief overview and synthesis. In Autism Spectrum Disorders (Amaral, D.G. et al., eds), pp. 812–824, Oxford University Press

Review

Special Issue: The Genetics of Cognition

Understanding risk for psychopathology through imaging gene–environment interactions Luke W. Hyde1, Ryan Bogdan2 and Ahmad R. Hariri2,3 1

Department of Psychology and Center for the Neural Basis of Cognition, University of Pittsburgh, 210 South Bouquet St, Pittsburgh, PA 15260, USA 2 Laboratory of NeuroGenetics, Department of Psychology and Neuroscience, Duke University, Box 90086, 417 Chapel Drive, Durham, NC 27708, USA 3 Institute for Genome Sciences and Policy, Duke University, 417 Chapel Drive, Durham, NC 27708, USA

Examining the interplay of genes, experience and the brain is crucial to understanding psychopathology. We review the recent gene–environment interaction (G E) and imaging genetics literature with the goal of developing models to bridge these approaches within single imaging gene–environment interaction (IG E) studies. We explore challenges inherent in both G E and imaging genetics and highlight studies that address these limitations. In specifying IG E models, we examine statistical methods for combining these approaches, and explore plausible biological mechanisms (e.g. epigenetics) through which these conditional mechanisms can be understood. Finally, we discuss the potential contribution that IG E studies can make to understanding psychopathology and developing more personalized and effective prevention and treatment. Genes, experience and the brain A burgeoning synergy of disciplines and technologies are providing unique insights into how the dynamic interplay between genes, brain and experience shapes individual risk for psychopathology. This interplay is being articulated at multiple levels of analysis from molecules to cells to neural circuits, from emotional responses to cognitive functions to personality, and from populations to families to individuals [1–4]. Here, we brieﬂy review recent endeavors that highlight the potential value of such interdisciplinary research. We then provide perspectives on how existing approaches and methods could be leveraged further to advance understanding of the etiology, pathophysiology and, ultimately, treatment and prevention of psychopathology. Gene–environment interaction (G E) [5] and imaging genetics [6] studies have both been very useful approaches to studying psychopathology. G E studies have emphasized the transactional nature of experience and the genome in the development of behavior, and imaging genetics studies have provided more proximal phenotypes and plausible mechanisms through which genes affect Corresponding author: Hyde, L.W. ([email protected]).

Glossary 5-HTTLPR: serotonin (5-HT) transporter gene linked polymorphic region. The 5HTTLPR is a variable number of tandem repeats polymorphism in the promoter region of the serotonin transporter gene. The serotonin transporter mediates active reuptake of synaptic serotonin and is thus crucial to regulating the duration and magnitude of serotonin signaling. Candidate gene: a gene whose protein product suggests that it could be involved in a phenotype of interest or a construct relevant to the phenotype or a gene that has been linked to a phenotype through GWAS. Epistasis: interaction between two or more polymorphisms so that the observed phenotype differs from what would be expected by either polymorphism independently. Gene–environment correlation (rGE): occurs when exposure to environmental conditions is dependent on one’s genotype. For example, the correlation between an ‘environmental’ risk factor such as harsh parenting and aggression could actually reflect a genetic pathway (mothers who are harsh could pass on genes to their children that increase the likelihood that they are aggressive). Genetic polymorphism: a variation in DNA with a frequency of at least 1% in the population. Functional genetic polymorphisms could reflect changes in a single (or multiple) base pair that can affect subsequent transcription of a gene or the structure of the resulting translated protein. Genome-wide association study (GWAS): an examination of genetic variation across the entire genome. Heritability: extent to which individual genetic differences contribute to phenotypic individual differences. Statistically, heritability represents the relative contribution of ‘genetics’ as compared to ‘environment’ when conceptualized as independent forces in shaping behavior and thus is a measure of the reliability estimate of the passage of traits from parent to offspring [4]. Hidden heritability: variance accounted for in twin studies of phenotypes that is unaccounted for by molecular genetic studies. Latent variables: mathematically inferred variables that represent the underlying commonality between directly measured variables. In practice, latent variables are variables that model the shared variance of similar predictor variables and thus decrease the error inherent in any one individual measure. For example, a measure of harsh parenting that includes observations of parenting, self-reports of parenting and reports of parenting by a significant other would more precisely model the underlying harsh parenting construct. Minor allele: less common allele at a polymorphic locus. Penetrance: likelihood that a genotype will result in a phenotype. Statistical mediation/Indirect effects: occurs when the link between a predictor and dependent variable is dependent on the effects of the predictor variable on an intermediate variable. This intermediate variable could serve as the mechanism linking the independent and dependent variable. Similarly, indirect effects denote the extent to which the independent variable affects the dependent variable through the independent variable’s effect on the mediator (and the mediator’s effect on the dependent variable). Note that consistent with others [58], we use the terms mediation and indirect effects interchangeably in this paper and thus do not imply that direct effects must be present between independent and dependent variables in order to find indirect effects. Statistical moderation: occurs when a ‘moderator’ variable affects the direction and/or strength of the relationship between a predictor variable and a dependent variable. A moderator variable is thus one that qualifies a relationship between two other variables. In other words, the relationship between variable A and B differs depending on the level of variable C.

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.001 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

417

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

behavior. However, these approaches are not yet well integrated even though they have great potential to inform each other. In designing and carrying out studies that combine these methods, it is crucially important for researchers to address and understand challenges to progress inherent in each approach and to consider approaches that address these challenges. Moreover, in order to fruitfully combine these approaches, it is also important to consider statistical methods for analyzing these studies and to have an appreciation for biological mechanisms (e.g. epigenetics) through which genes and experience affect subsequent brain function and behavior. With careful consideration of all of these points, future research that combines G E and imaging genetics approaches has the potential to greatly inform our understanding of psychopathology and delineate more personalized and successful prevention and interventions. Gene-Environment Interactions G E occurs when the relationship between an environmental experience (e.g. exposure to toxins, trauma, stress) and the emergence of altered physiological or behavioral responses (e.g. compromised immune function, psychopathology) is contingent on individual differences in genetic make-up (i.e. genetic polymorphisms) (see Glossary) [5]. With G E, the effect of an environmental experience on outcome is conditional on genetic background (i.e. genotype) or, conversely, the effect of individual genotype on behavior or health is conditional on an environmental experience (Figure 1). For example, in key early work, Caspi and colleagues demonstrated longitudinally that well-established links between life stress and subsequent depression were contingent on serotonin transporter linked polymorphic region (5-HTTLPR) genotype [7]. Speciﬁcally, individuals with the transcriptionally less efﬁcient short allele (fewer transporter molecules available to remove serotonin from the synapse) had a strong and positive relationship between life stress and depressive phenotypes, whereas those with the long allele had little or no relationship between life stress and depression. These ﬁndings are supported by meta-analysis [8] and animal

[(Figure_1)TD$IG]

(a) G × E

models [3], and a wealth of other G E studies have demonstrated similar associations across other genes, environments and phenotypes; for example monoamine oxidase A (MAOA) genotype moderates the relationship between maltreatment and antisocial behavior [9], and catechol O-methyltransferase (COMT) genotype moderates the relationship between cannabis use and psychosis [10]. Theoretical reviews have revealed several key principles for conducting G E research and evaluating resulting patterns [2,5]: researchers should consider the broad heritability of the target behavior, and then leverage knowledge generated in physiology and neuroscience to select polymorphisms in candidate genes that are of functional signiﬁcance in the biological response to the environmental experience. Moreover, there should be evidence of variability in the response to the selected environmental experience, for which accurate measurement and, ideally, quantiﬁcation should be available. Finally, there should be causal evidence linking the environmental experience with psychopathology. Because this approach does not presuppose a large main effect of single genetic variants (or experiences) on behavior but rather emphasizes an interaction with experience, carefully conducted studies of G E are instrumental in addressing issues of ‘hidden heritability’ (see glossary [11]), the generally weak penetrance of polymorphisms in candidate genes [11], and the lack of consistent replication in genetic association studies of psychopathology [2,12]. G E research often represents a more plausible model of disease in which individual experiences and genetic make-up interact across development to inﬂuence relative risk rather than more simplistic models hypothesizing independent effects of particular genes or experiences. Challenges to progress and possible advances in the field Although G E research has already advanced our understanding of the etiology of psychopathology, there are outstanding issues that deserve further consideration.

(b) Imaging genetics 2D

Genetic variation Neural functioning 1A

Gene × environment

1B

1C

Environments

Behavior & psychopathology

2A

Genetic variation

2B

2C

Behavior & psychopathology TRENDS in Cognitive Sciences

Figure 1. A conceptual and statistical diagram of G E and imaging genetics studies. (a) G E framework. Genes and environments might each have a ‘main effect’ on behavior (paths 1A and 1C) but the focus of these studies is on the interaction term which is modeled as a product of the two variables. (b) An ideal imaging genetics framework. Genetic variation in individuals leads to individual variability in neural functioning (path 2A), individual variability in neural functioning leads to differences in behavior or psychopathology (path 2B). Genetic variation might or might not have a direct impact on distal complex behavior (path 2C). Genetic variation has an indirect or mediated effect on behavior via its effect on neural functioning (arrow 2D: note that this path is not actually modeled statistically but is provided for conceptual clarity; this effect can be modeled as the product of the 2A and 2B paths).

418

Review First, it is unclear if G E pertains only to harsh environments and undesirable outcomes [13]. Some authors have argued that ‘crossover’ effects (in which a speciﬁc polymorphism is disadvantageous in some environments but advantageous in others) suggest that some polymorphisms cannot be cast as simply conferring relative ‘risk’ but rather as shaping the range or ‘plasticity’ or ‘differential susceptibility’ to environmental triggers or contexts [14,15]. Although a ‘plasticity’ model is appealing, others have argued that the limited empirical data thus far suggest that the hypothesized ‘plasticity’ effects might not fall within a meaningful range of the data (i.e. actually observed in the real world) [16]. Fortunately, this question can be addressed through continued research, especially research that addresses enriching environments and positive outcomes. A second outstanding issue reﬂects controversy in the use and deﬁnition of ‘environment’ in G E research [17]. Typically, environment refers to both experiential phenomena, including childhood abuse or adult stressors such as divorce or unemployment [7], and exposure to physical forces such as toxins, natural disasters (e.g. hurricanes, tornadoes) and acts of violence (e.g. war, terrorism) [18]. However, experiential phenomena and physical forces differ crucially in the degree to which the affected individual contributes to the environmental trigger: little to none with physical forces but possibly a signiﬁcant amount with experiential factors. Reﬂected in the latter is gene–environment correlation (rGE), which captures the inﬂuence of genetically driven variability in behavior as a precipitator or correlate of speciﬁc experiential triggers (e.g. difﬁcult temperament resulting in harsh parenting). Thus, some G E studies might be biased by rGE [19]. This issue of rGE has been addressed through using designs including behavior genetic approaches [20] (e.g. adoption [21], twin studies [22]), natural disaster [23] and natural experiments [24], experimental manipulation in humans [25] and non-human primates [26], and treatment designs [27]. Third, although G E research alone has increased the depth and complexity of our understanding of factors inﬂuencing the etiology of psychopathology, it is certain that even greater complexity exists in the form of G E E and G G E [28–30]. For example, in a G E E study, the authors report that the 5-HTTLPR genotype maltreatment interaction predicting depressive symptoms originally reported by Caspi and colleagues [7] was further moderated by social support wherein only short homozygotes with a history of childhood maltreatment and low social support showed increased depressive symptoms [28]. These results emphasize the complex and multifaceted nature of these systems in which some experiences exacerbate risk (maltreatment), whereas others are protective (high social support). Consistently replicating such increasingly complex interactions requires sample sizes and statistical power not present in even the largest datasets published, particularly when analyzing interactions using canonical approaches that involve identifying ﬁrst the main effects of each variable (e.g. genotype 1, genotype 2, environment 1, environment 2) [31]. In this approach, the interaction is limited in power by inherent distributional properties of the interaction term in nonexperimental

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

studies and by the need to account for main effects before examining interactions. Moreover, this limitation in power is often compounded by the frequency of the minor allele of the polymorphism, the rate at which individuals are exposed to a given trigger (and severity of the exposure [32]) and the frequency (and error of measurement) of possible dichotomous psychiatric diagnosis ([3,33] although see [34]) (see [31] for a discussion of approaches that might yield more power and note that experimental studies have much greater power to detect interactions). Fourth, it is also important to be cautious of spurious G E ﬁndings, which could arise due to selected sampling as well as scaling artifacts within logistic regression. Ideally, Eaves [35] suggests that these issues can be addressed by evaluating transformed data to examine if the interaction remains and using continuous variables and random sampling when possible (for more details see [33,36]). Fifth, it is important for G E ﬁndings to be replicated and these ﬁndings to be supported by meta-analysis. Conﬂicting reports on the interaction of the 5-HTTLPR and life stress predicting depression underscore this point. After initial ﬁndings (e.g. [7]), a meta-analysis suggested no reliable effect of this interaction on depression diagnosis [37]. However, this meta-analysis has been criticized for a biased selection of included studies. Speciﬁcally, authors have noted that included studies were characterized by relatively poor stress measurement [38], and an emphasis on dichotomous outcomes [33]. In line with these concerns, and in contrast to the conclusions of this meta-analysis [37], more thorough and inclusive meta-analyses support the reliability of the 5-HTTLPR stress interaction predicting depression [8,38]. Moreover, recent reviews have documented this interaction effect across model species (e.g. rhesus macaque and transgenic mice) and methodologies [3]. Nevertheless, this ongoing debate clearly highlights the importance of good construct measurement (of both environment and outcome). Finally, beyond issues of measurement, demographic variables such as age [39] and gender [40], as well as race/ ethnicity [41–43] and possible genetic substructure [44] are all likely to inﬂuence ﬁndings and require careful control and examination as additional moderators. G E research has provided a more nuanced understanding of the interplay between biology and environment in shaping risk for psychopathology. However, G E alone has not revealed the speciﬁc biological mechanisms for this risk [45]. Ultimately, for a genetic or environmental variable to affect behavior, it must ‘get under the skin’ [29,36,46]. G E must be instantiated in the brain if it is to affect behavior and the etiology of psychopathology. Imaging genetics Linking common genetic polymorphisms to variability in brain structure, function and connectivity is the foundation of imaging genetics [6,47,48]. This foundation is important for several reasons: ﬁrst, by connecting genetic variation to an intermediate biological phenotype (the brain), a plausible mechanism is provided through which genes affect behavior (Figure 1). For example, several studies have demonstrated a link between the short allele of the 5-HTTLPR and increased amygdala reactivity to threat 419

Review [6,47], as well as altered functional connectivity between the amygdala and prefrontal regions [48]. Given links between increased amygdala reactivity and anxiety and depression [49,50], these studies address how and why variation in the 5-HTTLPR might affect risk for these psychopathologies. Second, when the target polymorphism is of known functionality (e.g. altered gene transcription), the genetic variant serves as a proxy for individual differences in brain chemistry and thus offers clues into the molecular mechanisms through which differences in brain arise at the genetic and molecular (e.g. neurotransmitter) level. For example, in the case of the 5-HTTLPR, the short allele has been linked to decreased transcription of the serotonin transporter [51] which affects serotonin signaling. Third, the neural and genetic variables of interest allow for more effective synergy with animal models (e.g. transgenic mouse models, optogenetics), which in turn can advance the detailed understanding of molecular and cellular events ultimately linking genetic variation to brain to behavior [3,45,52]. In addition, imaging genetics using multimodal positron emission tomography (PET)/functional magnetic resonance imaging (fMRI) [53] and pharmacological fMRI designs [54] has the potential to further illuminate speciﬁc molecular pathways mediating genetic effects on brain [1,3]. Fourth, by focusing on dimensional and relatively objective intermediate phenotypes (e.g. regional brain activation to speciﬁc stimuli), analyses are not limited by broad nosological deﬁnitions (e.g. DMS-IV diagnoses) that are often plagued by heterogeneity in symptoms/behaviors or inherent biases in self-report (e.g. [55]). Moreover, by using a biological phenotype (i.e. behaviorally relevant brain structure and function) more proximal to the functional effects of genetic variants, imaging genetics gains power relative to research with more distal behavioral phenotypes, and is poised to uncover novel candidate genetic variants (possibly through GWAS). As these novel candidates identiﬁed through imaging genetics will necessarily provide demonstrated effects on speciﬁc neurobiological pathways, they can in turn be targeted in association studies with behavioral and/or clinical phenotypes [56]. In sum, imaging genetics offers new insight into psychopathology by mapping predictive links between genes, brain and behavior, furthering our understanding of the etiology of disorders at the genetic and molecular level. Challenges to progress and possible advances in the field As in G E, imaging genetics studies have contributed to our understanding of psychopathology but some major issues are worth noting. For example, a majority of imaging genetics studies, especially early research, established links between genetic polymorphisms and brain but failed to link either directly to meaningful differences in behavior [47,48]. Recently, imaging genetics studies have begun to establish such meaningful links by modeling indirect pathways from genes to behavior via the brain [49,57]. Studies that draw indirect pathways between gene and behavior through the brain, when no direct gene–behavior link exists [49], emphasize the importance of using statistical approaches that can model indirect (mediated) pathways 420

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

[58]. Moreover, similar to G E studies, imaging genetics studies demonstrate that there are important relationships between genes and behavior even when large direct relationships are not evident. Another crucial challenge is to model even greater complexity of genetic effects on the brain. G E studies clearly demonstrate the importance of environmental experience in understanding the ultimate effects of genetic variation on behavior and thus the environment should be modeled in future studies (see description of imaging gene– environment interactions (IG E) below). Beyond issues of the environment, as in G E, the issue of epistasis and the probable small effect of any single polymorphism highlights the need for novel analytic approaches such as investigating G G interactions [59,60], constructing cumulative genetic proﬁles [61], attempting hypothesis-free imaging GWAS [62] as has been done with G E [63] (although greater application of GWAS to G E and neuroimaging are both needed), examination of rare gene or copy number variants [45], and novel statistical approaches to integrate multiple genes into models [64,65]. Furthermore, beyond interaction effects (G E, G G), future studies that incorporate complementary techniques (e.g. neuroreceptor PET, pharmacologic challenge, animal models) or approach the modeling of neural reactivity in novel ways (e.g. machine learning [66], graph theory [67]) will better capture the molecular mechanisms mediating genetic effects on brain [1,68]. Imaging G E Both G E and imaging genetics research examine potential relationships between genetic variation and individual differences in behavior and risk for psychopathology. In G E, the relationship is conditional (statistical moderation) on experiences that are necessary to unmask genetic effects (or vice versa). In imaging genetics, a biological mechanism can be speciﬁed (statistical mediation/indirect effects) in which variability in brain links genes and behavior. Here, we advocate for an integration of these approaches to help understand conditional mechanisms through which genes, environments and the brain interact to predict behavior and risk for psychopathology. We term this integrative strategy: imaging gene-environment interactions (IG E) (Figure 2). Several recent reviews [3,46,69] have demonstrated possible IG E by combining ﬁndings from research in animal models, G E and imaging genetics to explain the interactions of genetic variants with environmental variables to predict learning, memory and psychopathology. Although these reviews are exciting, empirical studies are only beginning to test components of IG E directly [70,71] and thus we explore how these conditional mechanisms can be speciﬁed statistically and conceptually in a human neuroimaging study. Conceptual models Statistically, the concept of IG E can be modeled by a moderated mediation framework (also called conditional indirect effects) [58] in which mediated/indirect effects are moderated by a third variable. In this framework, any or all paths within a mediation framework (gene to brain, brain to behavior, gene to behavior via brain) could differ

(Figure_2)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

IG × E Neural functioning

b Genetic variation

d c

b

Gene × environment

e

Genes or environments

a a

rGE

a Environments

Behavior & psychopathology TRENDS in Cognitive Sciences

Figure 2. A conceptual model of IG E. To understand how IG E might be modeled conceptually and statistically, we demonstrate the relationship of the variables through highlighting traditional G E and imaging genetics paths as well as new paths possible in IG E studies. The a paths (green) model typical G E relationship. The b paths (purple) model the paths from an ideal imaging genetics study. The c path (blue) models the direct effect of the environment on neural functioning demonstrated in epigenetic studies. The d path (gold) models a G E predicting neural functioning (IG E effect). In this interaction, a gene would be more predictive or have a greater effect on a neural phenotype in some environments but not others (or the reverse: the environment would be predictive of neural functioning for those with one genetic variant but not another). The e path (red) represents another interaction: the possibility that genetic variation or an environmental variable could interact with neural functioning to predict behavior. For example, those with a variant in a gene affecting endocannabinoid signaling show greater correlation between reward-related brain reactivity and a measure of impulsivity [93]. Additionally, those with low social support have a greater relationship between their threat-related neural reactivity and trait anxiety [94]. Interactions involving the environment could be between gene and environment predicting neural function (d path) or between gene and neural functioning predicting behavior (e path) but in typical G E studies both of these interactions would be equivalent even though these interactions are likely to be due to very different mechanisms. Note that indirect and mediated pathways can be connected between many of the variables (e.g. G E to behavior through neural functioning) and thus an ideal IG E finding would be that the G E interaction term predicts behavior through neural functioning. Finally, within an SEM model, modeling a continuous interaction, the covariance between a genetic variant and an environment can be modeled which reflects the rGE between the specific genetic variant and specific environment.

depending on the level of a moderator variable (e.g. presence or absence of childhood abuse). As seen in Figure 2, there are multiple ways in which genetic, neural, environmental and behavioral variables could interact, and each model yields answers to slightly different questions (see also [58]). A particularly intuitive IG E model is a G E in which the interaction term predicts behavior through its effect on brain function (Figure 2, path D). In this case there are direct effects of both genetic and environmental variables on brain function but their interaction predicts nonadditive unique variance, which in turn predicts behavior. For example, genetic variation in serotonin signaling predicts increased amygdala reactivity [47], as do experiences of extreme early environmental deprivation [72], and individuals with both this genetic variation and environmental experience could show a synergistic increase in amygdala reactivity which could then predict increased anxiety symptoms. Alternatively, a positive environment such as parental warmth could negate any relationship between genetic variation in serotonin signaling and amygdala reactivity, and this lowered amygdala reactivity could then predict average levels of anxiety symptoms (Figure 3). This particular interaction (G E predicting brain function) also underlies much of the potential of IG E approaches. By combining the power of proximal intermediate phenotypes and the potential of G E to clarify such relationships, IG E can provide further insight into the conundrum of hidden heritability. For example, if a genetic variant has no association with a neural or behavioral phenotype in most

circumstances but has a robust association in relatively rare environments (e.g. physical abuse), IG E might be able to detect this association particularly with more proximal neural phenotypes. Such relationships can be tested using path or Structural Equation Modeling (SEM). As in G E studies, the way these relationships are tested (and graphed) can affect the interpretation of the results. For example, following from imaging genetics models, the environment could be seen in IG E studies as the moderator of the paths in an imaging genetics analysis (Figure 4). However, this approach privileges genetic factors as the ‘direct’ predictors of neural activity even though there is evidence that experience can affect the brain in direct and causal ways (see epigenetics section below) and G E studies often model genes as the moderator. Beyond this conceptual distinction, the analysis method can also affect results and their interpretation. For example, moderation in SEM is often tested in a multigroup model in which the path between genetic variation and brain function could be compared across two groups of subjects (e.g. those with or without a history of abuse). This multigroup model is best for dichotomous moderators (e.g. two different alleles of a polymorphism or an environmental extreme) and could be more easily and intuitively understood by readers. However, multigroup approaches have less value when otherwise continuous variables (e.g. continuously measured parenting) are dichotomized (e.g. harsh versus warm parenting). The alternative is to specify continuous interactions between variables of interest, even 421

(Figure_3)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

(a)

Genetic variant (e.g., 5-HTTLPR)

Environmental experience (e.g., abuse)

Short allele carriers have less 5-HT synaptic clearance which leads to increased amygdala reactivity to threat

Two separate mechanisms cause very high chronic levels of amygdala reactivity and broader cortico-limbic reactivity Abuse increases amygdala reactivity by sensitizing amygdala to environmental threat

Above average level of anxiety symptoms

(b)

Genetic variant (e.g., 5-HTTLPR)

Environmental experience (e.g., high levels of social support)

Short allele carriers have less 5-HT synaptic clearance which leads to increased amygdala reactivity to threat

Amygdala reactivity initially increased through greater 5-HT signaling

Social support helps child learn coping mechanisms to deal with anxiety thoughts, helps child learn stimuli are not threatening

Prefrontal cortex is strengthened through learning coping strategies and down-regulates amygdala reactivity

Average level of anxiety symptoms

TRENDS in Cognitive Sciences

Figure 3. Possible biological models of IG E interactions within the brain. As research suggests there are plausible causal biological mechanisms through which experience affects transcriptional effects of genes on neural functioning, it is helpful to specify how genes and environments might interact at a conceptual level to bring out the statistical relationships that could be found in IG E studies. (a) A synergistic model. Both genes and environments directly act on one brain area on similar mechanisms at the synapse. For example, short carriers of 5-HTTLPR could have increased 5-HT signaling in the amygdala [47], leading to greater reactivity to threat, and abuse or extreme neglect could increase the transcription of nonindividually varying sequences in genes that affect amygdala function [72] causing parallel increases in amygdala reactivity to threat. Thus the amygdala could have two pushes towards being more reactive to threat and show a multiplicatively exaggerated response. (b) A buffering model. Although a carrier of the short allele of the 5-HTTLPR has increased amygdala 5-HT signaling, high levels of social support cause changes in areas of the prefrontal cortex which are able to downregulate amygdala reactivity leading to normal reactivity to threat (alternatively abuse could affect the prefrontal cortex diminishing its ability to regulate the amygdala [84]). In both (a) and (b) interactions could occur within the same brain area or across multiple brain areas within a related circuitry (e.g. a corticolimbic circuitry). Note: it is also important to keep in mind that all of these relationships are ‘probabilistic’, not deterministic, and thus these models offer possibilities as a way of understanding IG E.

latent variables, and common statistical packages have recently made estimation of continuous latent interactions possible [58]. Continuous interactions are likely to provide more power and reﬂect the dimensional nature of many of 422

the variables (e.g. environmental experiences, neural function), as well as allow for the modeling and evaluation of the rGE between the speciﬁc genetic and environmental variables in the model (Figure 2).

(Figure_4)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Neural reactivity Targets: Brain function, structure and connectivity Novel approaches: Factor analysis of resting BOLD data to identify distributed functional circuitries; Graph theory and machine learning approaches

Genetic variability

Behavior

Targets: Genotypes with demonstrated functional effects Novel approaches: Cumulative genetic profiles; Modeling cascades of multiple interacting genes and molecular patthways; Regression trees and recursive partitioning

Targets: Personality, mood and affect, psychopathology Novel approaches: Latent factors to identify underlying constructs across measures; Observational methods

rGE Environmental experiences Targets: Experiences that precipitate individual variability in response; Experiences linked to psychopathology; Experiences linked to changes in biology or gene expression; Extreme environments likely to have larger effects on brain and behavior Novel approaches: Observational methods, longitudinal measurement of environments, experimental manipulation or intervention studies TRENDS in Cognitive Sciences

Figure 4. Specific targets for G E, imaging genetics and IG E studies. Importantly, novel approaches across each domain are needed to help progress understanding across all models. Moreover, similar to Figure 2, this model emphasizes the interaction between the environment and biology (genes, neural reactivity) as these variables predict behavior. More transparent arrows signify links made in traditional research. Bolded arrows represent newly proposed paths specific to IG E models. BOLD: blood oxygen level-dependent.

Considerations The above promise of IG E, like that of its parent strategies, is not without challenges. First, the challenges noted in the G E and imaging genetics sections generally apply to IG E models (Box 1; Figure 4). Second, IG E models test statistical correlations in humans specifying possible relationships and thus need to be paralleled by work in animal models or with experimental designs (e.g. drug treatment protocols, adoption studies) that can infer causality [4]. Moreover, as we discuss below, these models should be guided by biologically plausible relationships between variables. Third, these complex models require signiﬁcantly larger samples than those currently available to have acceptable levels of power. Moderated mediation models require starting sample sizes in the range of 500– 1000 subjects to examine the expected small to moderate effects of each variable [58]. Moreover, this estimate does not include issues such as low minor allele frequencies and environmental exposure rates, which could necessitate even larger samples. Although samples of this size might sound untenable in neuroimaging, there are already pub-

lished studies with samples of this size (e.g. [73]) and consortium projects are addressing this issue by pooling data across sites/studies (e.g. [74]). Fourth, it is important to understand that development plays a large role in the unfolding of gene–environment–brain–behavior relationships. For example, many studied genetic variants (e.g. MAOA, 5-HTTLPR) likely function in utero or very early in development [75–77]. Moreover, environmental experiences differ in their impact depending on the developmental stage of the individual (e.g. types of stressors for a child might differ from those for an adult) [78,79] and epigenetic studies demonstrate that certain experiences have a greater biological impact during ‘sensitive periods’ of development [4]. Finally, just as G E and imaging genetics studies required researchers to bridge several areas and/ or work in multidisciplinary teams, IG E studies require even greater knowledge and collaboration. We hope that the conceptual models introduced in IG E will garner even greater appreciation for the work of colleagues in disparate ﬁelds (e.g. animal neurophysiology, biostatistics, epidemiology, experimental psychology). 423

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Box 1. Challenges to progress and outstanding questions Genes Challenges: single polymorphisms are of small effect. Issues such as epistasis and developmental regulation of genes have not been addressed in most studies. Solutions: genetic risk profiles representing the cumulative impact of multiple functional polymorphisms within a system (e.g. dopamine) and statistical models combining polymorphisms within and between systems (recursive partitioning, regression trees) can identify small genetic effects and their interactions. Longitudinal studies of genetic effects in animals and humans can inform when and how each genetic variant might affect brain and behavior. Outstanding questions: when and how do most genes of interest have their effect on brain and behavior? Are there more complex mechanisms or organized ways in which genes interact across development? Environments Challenges: many G E studies have relied on self-report or other measures with substantial error (e.g. retrospective reports). For many environmental variables it is not clear when certain experiences might have their effect on brain or behavior. For many experiences, G E studies have not paid attention to whether it is the objective account or the subjective report that matters. Solutions: observational measures and multiple well-validated measures of the same construct can help decrease error of measurement, as can modeling latent constructs of these variables. Prospective longitudinal studies can address developmental cascades and determine ‘sensitive periods’ during which certain experiences might have the greatest impact. Studies with multiple informants and methods can compare the impact of subjective versus objective accounts of experiences. Outstanding questions: are there certain experiences that have an impact no matter when they occur? Are there experiences that interact differently with genetic polymorphisms depending on when they occur? Are there experiences for which objective or subjective reporting is more important? Brain Challenges: much imaging genetics research focuses on a single brain region or the simple relationships between two regions whereas behavior reflects complex interactions within and across

Plausible biological mechanisms Imaging genetics studies in humans and non-human primates (e.g. [26,77]) as well as studies of strain differences in laboratory mice (e.g. [52]) convincingly link interindividual genetic variability to differences in brain and behavior. What about the environment: does it alter biology in ways that affect brain and behavior? For many biologists, including neuroscientists, the obvious answer might be ‘yes’, but given the ‘nature–nurture’ debates in some areas of psychology [4] it is important to specify models whereby experiences are transduced into functional biological signals that affect brain function and subsequent behavior. A fundamental example of such transduction comes from molecular studies demonstrating that learning is supported by long-term changes (i.e. long-term potentiation and depression) in synaptic physiology, which are mediated by changes in gene expression [80,81]. Thus, activity-dependent gene regulation drives changes in protein expression and adaptations in the molecular machinery for neurons and neuronal circuits supporting behavior. Importantly, such environmentally induced changes ultimately manifest in the 424

multiple brain regions. fMRI studies are relatively indirect measures of cellular activity. Solutions: exploratory statistical techniques such as machine learning, factor analysis and graph theory analysis use a data-driven approach to identify complex circuit function and whole-brain network organization. Multimodal human and animal studies can help address cellular and molecular mechanisms underlying brain activity. Mediation analyses can be used in multimodal studies to provide plausible pathways (e.g. does brain structure mediate gene– brain function relationship? Do receptor levels, assayed with PET, mediate the gene–brain function relationship?). In addition, experimental studies (e.g. experiments that manipulate the environment, pharmacological studies to manipulate neural chemistry) and animal studies can address mechanisms from a causal perspective. Outstanding questions: are studies finding relationships between single brain areas (e.g. the amygdala) and behavior the result of more complex interactions between multiple brain structures we do not yet understand? How does the interaction between brain regions map onto behavior? To what extent do genetic variants affect behavior through their influence on function versus structure versus connectivity in the brain? Outcome behavior/psychopathology Challenges: our conceptualization and resulting measurement of psychopathology is still rudimentary and based on observable behavior, which can lead to increased error in diagnosis. Dichotomous diagnoses limit statistical and inferential power, and miss the likely dimensional nature of most psychopathology. Solutions: imaging genetics and IG E provide intermediate continuous phenotypes, which might be more objectively measured and more powerful. Continuous and hierarchical models of broad psychopathology can increase power and model the high comorbidity found in studies of psychopathology. Observational methods of behavior can provide more reliable measures particularly when combined in latent constructs with multiple converging self-report measures. Outstanding questions: how do we account for the high levels of comorbidity across most psychopathology? Can intermediate phenotypes and, ultimately, the genetic polymorphisms by which they are predicted, usefully inform diagnosis and treatment? How can we define and delineate subgroups within broad diagnostic categories (e.g. Antisocial Personality Disorder) that express more homogeneous alterations in behavior and, by extension, brain dysfunction?

reorganization of brain circuits and their functional responses [80,81]. Another fundamental mechanism governing the transduction of experience into changes in biology and behavior is epigenetics [4,82,83]. Reviews of epigenetic regulation of brain and behavior are available elsewhere [e.g. 4,82–84]. Brieﬂy, epigenetic regulation refers broadly to the local (i.e. cell speciﬁc) modiﬁcation of gene expression of the DNAhistone complex and resulting accessibility of speciﬁc genes for transcription. Studies have demonstrated that early experience can alter epigenetic markers and subsequent patterns of transcription in a way that affects brain structure and function as well as behavior [4]. Chief among studies of epigenetic regulation of behavior are those conducted by Meaney and colleagues demonstrating that in rats, maternal care of offspring affects later adult behavior through epigenetic regulation of hypothalamic– pituitary–adrenal (HPA) axis reactivity to stress. Speciﬁcally, higher levels of maternal licking and grooming and arched-back nursing (LG-ABN) of rat pups during the ﬁrst week of life leads to increased serotonin levels, which drive the expression of nerve growth factor-inducible protein A

Review (NGFI-A). Increased NGFI-A, in turn, leads to decreased methylation and increased acetylation of the promoter region of the glucocorticoid receptor (GR) gene in hippocampal neurons. This pattern of decreased methylation and increased acetylation results in increased gene expression and higher GR numbers in the hippocampi, which mediate negative feedback regulation of the HPA axis response to stress. These changes persist throughout the lifespan and promote adult behavior that is characterized by relative stress resilience and increased subsequent maternal care. Thus, through this epigenetic mechanism, high LG-ABN mothers beget relatively stress-resilient pups that become high LG-ABN mothers by experience-dependent mechanisms [4,85]. In these and similar studies, early experience affects epigenetic modiﬁcations triggering a cascade of changes in cellular signaling (particularly in the brain), which shape adult behaviors. In a compelling extension of this research to humans, a study of postmortem hippocampal tissues from individuals who committed suicide (compared to others who had accidental deaths) found increased methylation of the human GR promoter and decreased GR mRNA. However, this difference was only observed in a subset of suicide completers who had been abused as children and not in completers without history of abuse. Thus, there could be a remarkable conservation of epigenetic mechanisms regulating brain and behavior across species, which gives us conﬁdence in developing plausible biological models of IG E in humans based on ﬁndings in animal models [86]. Similar epigenetic effects have been documented in other genes and brain regions associated with psychopathology [84,87]. Collectively these studies suggest that the environment has a very direct and long-lasting effect on biology at the epigenetic and neural level and that these effects translate into differences in behaviors, thus emphasizing that G E is the rule rather than the exception when understanding variability in behavior [4]. Trying to parse main effects of genetic versus environmental variables is to ignore that the genome and environment are in constant interaction [4]: the biological primacy of G E is apparent from the realization that transcription factors can be and often are controlled by environmental signals [82]. Thus, these biological mechanisms indicate that the impact of genetic variation on relative risk and resilience for psychopathology will be experience and context dependent [88]. It is unclear, however, if such changes can be examined in the context of human IG E research because data demonstrating that epigenetic markers in peripheral human tissue (e.g. blood cells) are faithful proxies for changes in the brain is lacking ([83,84], although see [87,89]). Moreover, future studies are needed that examine the impact of epigenetic mechanisms on genetic polymorphisms, especially promoter variants, to test true epigenetic G E relationships [84,90]. Looking forward With the emergence of detailed measures for both genes and brain, IG E research is poised to accelerate the pace of scientiﬁc discovery by fueling novel exchanges between studies in humans and those in animals. Speciﬁc brain

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

substrates (e.g. amygdala reactivity), environments (e.g. childhood neglect) and genes (e.g. 5-HTT) identiﬁed through human IG E research can generate the next set of targets in animal models that can delve into the detailed molecular mechanisms that link these larger elements. Likewise, research in animals, especially studies that identify novel molecular and genetic factors in the regulation of brain and behavior, can generate targets for human research, which can model these factors through common polymorphisms in the genes of interest and fMRI probes of the relevant brain circuits. Dynamic exchanges across human studies and animal models promise to elucidate tractable biological mechanisms that can inform the etiology and pathophysiology of psychopathology. Within human studies, an IG E approach connects the pieces of the puzzle: whereas G E studies of the past have implied that the mechanism through which G E affects behavior is the brain, and whereas imaging genetics studies have missed the interaction of biology with experience, IG E studies can elucidate conditional mechanisms through which genes and experience interact to affect neural structure and function and ultimately behavior and psychopathology. Specifying these models through careful statistical and methodological approaches in well-characterized samples is crucial for the ability of IG E to inform our understanding of psychopathology. The treatment implications of such work are critically important as medicine moves towards greater personalization [91]. For example, IG E studies could lead to intervention and prevention trials that target those at speciﬁc genetic and/or environmental risk [4,27] by identifying more homogenous subgroups of individuals within the same diagnosis [92]. Thus, future IG E research might inform the development of frameworks for determining when and for whom certain treatments will work (e.g. which environments could sabotage the treatment process, which genes could predict treatment success, which combinations of genes and environments could be the targets of early preventative intervention projects) and might help to reﬁne diagnostic criteria. Overall, IG E can provide a more nuanced and complex model of human nature in health and disease by extending beyond nature–nurture debates and revealing speciﬁc mechanisms through which the constantly interacting environment and genome can be understood at the level of brain function and behavior. Acknowledgments We would like to thank funding sources (NIH grants 5R01-DA026222, T32-GM081760 and P30-DA023026), as well as Stephen Manuck and the Pitt Genetics Journal Club for thoughtful comments on ideas in this article.

References 1 Hariri, A.R. (2009) The neurobiology of individual differences in complex behavioral traits. Annu. Rev. Neurosci. 32, 225–247 2 Caspi, A. and Mofﬁtt, T.E. (2006) Gene–environment interactions in psychiatry: joining forces with neuroscience. Nat. Rev. Neurosci. 7, 583– 590 3 Caspi, A. et al. (2010) Genetic sensitivity to the environment: the case of the serotonin transporter gene and its implications for studying complex diseases and traits. Am. J. Psychiatry 167, 509–527 4 Meaney, M.J. (2010) Epigenetics and the biological deﬁnition of gene environment interactions. Child Dev. 81, 41–79 425

Review 5 Mofﬁtt, T.E. et al. (2005) Strategy for investigating interactions between measured genes and measured environments. Arch. Gen. Psychiatry 62, 473–481 6 Hariri, A.R. et al. (2006) Imaging genetics: perspectives from studies of genetically driven variation in serotonin function and corticolimbic affective processing. Biol. Psychiatry 59, 888–897 7 Caspi, A. et al. (2003) Inﬂuence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 301, 386–389 8 Karg, K. et al. (2011) The serotonin transporter promoter variant (5HTTLPR), stress, and depression meta-analysis revisited: evidence of genetic moderation. Arch. Gen. Psychiatry 68, 444–454 9 Caspi, A. et al. (2002) Role of genotype in the cycle of violence in maltreated children. Science 297, 851–854 10 Caspi, A. et al. (2005) Moderation of the effect of adolescent-onset cannabis use on adult psychosis by a functional polymorphism in the catechol-O-methyltransferase gene: longitudinal evidence of a gene environment interaction. Biol. Psychiatry 57, 1117–1127 11 Maher, B. (2008) Personal genomes: the case of the missing heritability. Nature 456, 18–21 12 Plomin, R. (2005) Finding genes in child psychology and psychiatry: when are we going to be there? J. Child Psychol. Psychiatry 46, 1030– 1038 13 Heiming, R.S. and Sachser, N. (2010) Consequences of serotonin transporter genotype and early adversity on behavioral proﬁle – pathology or adaptation? Front. Neurosci. 4, 187 14 Belsky, J. and Pluess, M. (2009) Beyond diathesis stress: differential susceptibility to environmental inﬂuences. Psychol. Bull. 135, 885–908 15 Ellis, B.J. and Boyce, W.T. (2011) Differential susceptibility to the environment: toward an understanding of sensitivity to developmental experiences and context. Dev. Psychopathol. 23, 1–5 16 Manuck, S.B. (2009) The reaction norm in gene environment interaction. Mol. Psychiatry 15, 881–882 17 Manuck, S.B. and McCaffery, J.M. (2010) Genetics of stress: genestress correlation and interaction. In Handbook of Behavioral Medicine: Methods and Applications (Steptoe, A. et al., eds), pp. 454–478, Springer 18 Franklin, T.B. and Mansuy, I.M. (2010) Epigenetic inheritance in mammals: evidence for the impact of adverse environmental effects. Neurobiol. Dis. 39, 61–65 19 Jaffee, S.R. (2011) Genotype-environment correlations: deﬁnitions, methods of measurement and implications for research on adolescent pscyhopathology. In The Dynamic Genome and Mental Health (Kendler, K.S. et al., eds), pp. 79–102, Oxford 20 Reiss, D. and Leve, L.D. (2007) Genetic expression outside the skin: clues to mechanisms of Genotype Environment interaction. Dev. Psychopathol. 19, 1005–1027 21 Leve, L.D. et al. (2010) Infant pathways to externalizing behavior: evidence of Genotype Environment interaction. Child Dev. 81, 340– 356 22 Jaffee, S.R. et al. (2005) Nature nurture: genetic vulnerabilities interact with physical maltreatment to promote conduct problems. Dev. Psychopathol. 17, 67–84 23 Kilpatrick, D.G. et al. (2007) The serotonin transporter genotype and social support and moderation of posttraumatic stress disorder and depression in hurricane-exposed adults. Am. J. Psychiatry 164, 1693– 1699 24 Costello, E.J. et al. (2003) Relationships between poverty and psychopathology. JAMA 290, 2023–2029 25 King, A.P. and Liberzon, I. (2009) Assessing the neuroendocrine stress response in the functional neuroimaging context. Neuroimage 47, 1116–1124 26 Bennett, A. et al. (2002) Early experience and serotonin transporter gene variation interact to inﬂuence primate CNS function. Mol. Psychiatry 7, 118–122 27 Brody, G.H. et al. (2009) Prevention effects moderate the association of 5 HTTLPR and youth risk behavior initiation: gene environment hypotheses tested via a randomized prevention design. Child Dev. 80, 645–661 28 Kaufman, J. et al. (2004) Social supports and serotonin transporter gene moderate depression in maltreated children. Proc. Natl. Acad. Sci. U.S.A. 101, 17316–17321 29 Rutter, M. and Dodge, K.A. (2011) Gene–environment interactions: the state of science. In Gene–Environment Interaction in Developmental

426

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

30

31

32

33

34 35

36

37

38

39

40

41

42

43

44 45

46

47 48

49 50 51

52

53

Psychopathology (Dodge, K.A. and Rutter, M., eds), pp. 87–101, Guilford Wenten, M. et al. (2009) Functional variants in the catalase and myeloperoxidase genes, ambient air pollution, and respiratoryrelated school absences: an example of epistasis in gene– environment interactions. Am. J. Epidemiol. 170, 1494–1501 McClelland, G.H. and Judd, C.M. (1993) Statistical difﬁculties of detecting interactions and moderator effects. Psychol. Bull. 114, 376–390 Weder, N. et al. (2009) MAOA genotype, maltreatment, and aggressive behavior: the changing impact of genotype at varying levels of trauma. Biol. Psychiatry 65, 417–424 Uher, R. (2011) Gene-environment interactions. In The Dynamic Genome and Mental Health (Kendler, K.S. et al., eds), pp. 29–58, Oxford Eaves, L. et al. (2003) Resolving multiple epigenetic pathways to adolescent depression. J. Child Psychol. Psychiatry 44, 1006–1014 Eaves, L. (2006) Genotype environment interaction in psychopathology: factor or artifact? Twin Res. Hum. Genet. 36, 74–87 Kendler, K.S. (2011) A conceptual overview of gene-environment interaction and correlation in a developmental context. In The Dynamic Genome and Mental Health (Kendler, K.S. et al., eds), Oxford, pp. 5–28 Risch, N. et al. (2009) Interaction between the serotonin transporter gene (5-HTTLPR), stressful life events, and risk of depression. J. Am. Med. Assoc. 301, 2462–2471 Uher, R. and McGufﬁn, P. (2010) The moderation by the serotonin transporter gene of environmental adversity in the etiology of depression: 2009 update. Mol. Psychiatry 15, 18–22 Lenroot, R.K. and Giedd, J.N. (2011) Annual Research Review: developmental considerations of gene by environment interactions. J. Child Psychol. Psychiatry 52, 429–441 Sjo¨berg, R.L. et al. (2006) Adolescent girls and criminal activity: role of MAOA-LPR genotype and psychosocial factors. Am. J. Med. Genet. B 144, 159–164 Serretti, A. et al. (2006) Meta-analysis of serotonin transporter gene promoter polymorphism (5-HTTLPR) association with selective serotonin reuptake inhibitor efﬁcacy in depressed patients. Mol. Psychiatry 12, 247–257 Munafo`, M.R. et al. (2009) 5-HTTLPR genotype and anxiety-related personality traits: a meta-analysis and new data. Am. J. Med. Genet. B 150, 271–281 Widom, C.S. and Brzustowicz, L.M. (2006) MAOA and the ‘‘cycle of violence:’’ childhood abuse and neglect, MAOA genotype, and risk for violent and antisocial behavior. Biol. Psychiatry 60, 684–689 Cardon, L.R. and Palmer, L.J. (2003) Population stratiﬁcation and spurious allelic association. Lancet 361, 598–604 Blakely, R. and Veenstra-Vanderweele, J. (2011) Genetic indeterminism, the 5-HTTLPR, and the paths forward in neuropsychiatric genetics. Arch. Gen. Psychiatry 68, 457–458 Meyer-Lindenberg, A. (2011) Neurogenetic mechanisms of geneenvironment interactions. In Gene-Environment Interactions in Developmental Psychopathology (Dodge, K.A. and Rutter, M., eds), pp. 71–87, Guilford Hariri, A.R. et al. (2002) Serotonin transporter genetic variation and the response of the human amygdala. Science 297, 400–403 Pezawas, L. et al. (2005) 5-HTTLPR polymorphism impacts human cingulate-amygdala interactions: a genetic susceptibility mechanism for depression. Nat. Neurosci. 8, 828–834 Fakra, E. et al. (2009) Effects of HTR1A C (-1019) G on amygdala reactivity and trait anxiety. Arch. Gen. Psychiatry 66, 33–40 Price, J.L. and Drevets, W.C. (2010) Neurocircuitry of mood disorders. Neuropsychopharmacology 35, 192–216 Lesch, K.P. et al. (1996) Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science 274, 1527–1531 Holmes, A. (2008) Genetic variation in cortico-amygdala serotonin function and risk for stress-related disease. Neurosci. Biobehav. Rev. 32, 1293–1314 Fisher, P.M. et al. (2006) Capacity for 5-HT1A–mediated autoregulation predicts amygdala reactivity. Nat. Neurosci. 9, 1362– 1363

Review 54 Bigos, K.L. et al. (2008) Acute 5-HT reuptake blockade potentiates human amygdala reactivity. Neuropsychopharmacology 33, 3221–3225 55 Andreasen, N.C. (2000) Schizophrenia: the fundamental questions. Brain Res. Rev. 31, 106–112 56 Hasler, G. and Northoff, G. (2011) Discovering imaging endophenotypes for major depression. Mol. Psychiatry 1, 1–16 57 Furmark, T. et al. (2008) A link between serotonin-related gene polymorphisms, amygdala activity, and placebo-induced relief from social anxiety. J. Neurosci. 28, 13066–13074 58 Preacher, K.J. et al. (2007) Addressing moderated mediation hypotheses: theory, methods, and prescriptions. Multivariate Behav. Res. 42, 185–227 59 Talkowski, M.E. et al. (2008) A network of dopaminergic gene variations implicated as risk factors for schizophrenia. Hum. Mol. Genet. 17, 747–758 60 Buckholtz, J. et al. (2007) fMRI evidence for functional epistasis between COMT and RGS4. Mol. Psychiatry 12, 893–895 61 Nikolova, Y.S. et al. (2011) Multilocus genetic proﬁle for dopamine signaling predicts ventral striatum reactivity. Neuropsychopharmacology 36, 1940–1947 62 Liu, X. et al. (2010) A genome-wide association study of amygdala activation in youths with and without bipolar disorder. J. Am. Acad. Child Adolesc. Psychiatry 49, 33–41 63 Ege, M.J. et al. (2011) Gene-environment interaction for childhood asthma and exposure to farming in Central Europe. J. Allergy Clin. Immunol. 127, 138–144 64 Gruenewald, T.L. et al. (2006) Combinations of biomarkers predictive of later life mortality. Proc. Natl. Acad. Sci. U.S.A. 103, 14158–14163 65 Hizer, S.E. et al. (2004) Genetic markers applied in regression tree prediction models. Anim. Genet. 35, 50–52 66 Pereira, F. et al. (2009) Machine learning classiﬁers and fMRI: a tutorial overview. Neuroimage 45, S199–S209 67 Astolﬁ, L. et al. (2007) Imaging functional brain connectivity patterns from high resolution EEG and fMRI via graph theory. Psychophysiology 44, 880–893 68 Ressler, K.J. et al. (2011) Post-traumatic stress disorder is associated with PACAP and the PAC1 receptor. Nature 470, 492–497 69 Casey, B.J. et al. (2009) Brain-derived neurotrophic factor as a model system for examining gene by environment interactions across development. Neuroscience 164, 108–120 70 Kohli, M.A. et al. (2011) The neuronal transporter gene SLC6A15 confers risk to major depression. Neuron 70, 252–265 71 Gerritsen, L. et al. (2011) BDNF Val66Met genotype modulates the effect of childhood adversity on subgenual anterior cingulate cortex volume in healthy subjects. Mol. Psychiatry DOI: 10.1038/mp.2011.51 72 Tottenham, N. et al. (2011) Elevated amygdala response to faces following early deprivation. Dev Sci. 14, 190–204 73 Butterworth, P. et al. (2011) The association between ﬁnancial hardship and amygdala and hippocampal volumes: results from the PATH through life project. Soc. Cogn. Affect Neurosci. DOI: 10.1093/ scan/nsr027 74 Schumann, G. et al. (2010) The IMAGEN study: reinforcement-related behaviour in normal brain function and psychopathology. Mol. Psychiatry 15, 1128–1139

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 75 Fowler, J.S. et al. (2007) Evidence that brain MAO A activity does not correspond to MAO A genotype in healthy male subjects. Biol. Psychiatry 62, 355–358 76 Sibille, E. and Lewis, D.A. (2006) SERT-ainly involved in depression, but when? Am. J. Psychiatry 163, 8–11 77 Jedema, H.P. et al. (2009) Cognitive impact of genetic variation of the serotonin transporter in primates is associated with differences in brain morphology rather than serotonin neurotransmission. Mol. Psychiatry 15, 512–522 78 Sroufe, L.A. and Rutter, M. (1984) The domain of developmental psychopathology. Child Dev. 55, 17–29 79 Kendler, K.S. and Myers, J. (2009) A developmental twin study of church attendance and alcohol and nicotine consumption: a model for analyzing the changing impact of genes and environment. Am. J. Psychiatry 166, 1150–1155 80 Tada, T. and Sheng, M. (2006) Molecular mechanisms of dendritic spine morphogenesis. Curr. Opin. Neurobiol. 16, 95–101 81 Kandel, E.R. (1991) Cellular mechanisms of learning and the biological basis of individuality. In Principles of Neural Science 4th edn (Kandel, E.R. et al., eds), pp. 1247–1279, McGraw-Hill 82 Zhang, T.Y. and Meaney, M.J. (2010) Epigenetics and the environmental regulation of the genome and its function. Annu. Rev. Psychol. 61, 439– 466 83 Mill, J. (2011) Epigenetic effectson gene function and their role in mediating gene-environment interactions. In The Dynamic Genome and Mental Health (Kendler, K.S. et al., eds), pp. 145–171, Oxford 84 Roth, T.L. and Sweatt, J.D. (2011) Annual Research Review: epigenetic mechanisms and environmental shaping of the brain during sensitive periods of development. J. Child Psychol. Psychiatry 52, 398–408 85 Weaver, I.C.G. et al. (2004) Epigenetic programming by maternal behavior. Nat. Neurosci. 7, 847–854 86 McGowan, P.O. et al. (2009) Epigenetic regulation of the glucocorticoid receptor in human brain associates with childhood abuse. Nat. Neurosci. 12, 342–348 87 Tsankova, N. et al. (2007) Epigenetic regulation in psychiatric disorders. Nat. Rev. Neurosci. 8, 355–367 88 Masten, A.S. (2001) Ordinary magic. Resilience processes in development. Am. Psychol. 56, 227–238 89 Fraga, M.F. et al. (2005) Epigenetic differences arise during the lifetime of monozygotic twins. Proc. Natl. Acad. Sci. U.S.A. 102, 10604–10609 90 Ursini, G. et al. (2011) Stress-related methylation of the catechol-Omethyltransferase Val158 allele predicts human prefrontal cognition and activity. J. Neurosci. 31, 6692–6698 91 Willard, H.F. and Ginsburg, G.S. (2009) Essentials of Genomic and Personalized Medicine, Academic Press 92 Mehta, D. et al. (2011) Using polymorphisms in FKBP5 to deﬁne biologically distinct subtypes of posttraumatic stress disorder: evidence from endocrine and gene expression studies. Arch. Gen. Psychiatry DOI: 10.1001/archgenpsychiatry.2011.50 93 Hariri, A.R. et al. (2009) Divergent effects of genetic variation in endocannabinoid signaling on human threat- and reward-related brain function. Biol. Psychiatry 66, 9–16 94 Hyde, L.W. et al. (2011) Social support moderates the link between amygdala reactivity and trait anxiety. Neuropsychologia 49, 651–656

427

Review

Special Issue: The Genetics of Cognition

The genetics of cognitive impairment in schizophrenia: a phenomic perspective Robert M. Bilder1, Andrew Howe2, Nic Novak3, Fred W. Sabb1 and D. Stott Parker2 1

Department of Psychiatry and Biobehavioral Sciences, David Geffen School of Medicine, and Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles, CA 90095, USA 2 Department of Computer Science, Samueli School of Engineering, University of California, Los Angeles, CA 90095, USA 3 Interdepartmental Program in Neuroscience, University of California, Los Angeles, CA 90095, USA

Cognitive impairments are central to schizophrenia and could mark underlying biological dysfunction but efforts to detect genetic associations for schizophrenia or cognitive phenotypes have been disappointing. Phenomics strategies emphasizing simultaneous study of multiple phenotypes across biological scales might help, particularly if the high heritabilities of schizophrenia and cognitive impairments are due to large numbers of genetic variants with small effect. Convergent evidence is reviewed, and a new collaborative knowledgebase – CogGene – is introduced to share data about genetic associations with cognitive phenotypes, and enable users to meta-analyze results interactively. CogGene data demonstrate the need for larger studies with broader representation of cognitive phenotypes. Given that meta-analyses will probably be necessary to detect the small association signals linking the genome and cognitive phenotypes, CogGene or similar applications will be needed to enable collaborative knowledge aggregation and specify true effects. Cognitive impairment in schizophrenia: in search of a genomic basis Cognitive impairment has been seen as a hallmark of schizophrenia at least since Emil Kraepelin described the syndrome of dementia praecox in 1893 [1] but in the past few decades it has assumed new importance at least in part due to hopes that the cognitive functions might prove more tractable targets for genetic study than are the characteristic symptoms used to diagnose schizophrenia (Box 1). Despite heritability of the schizophrenia phenotype estimated at near 80%, initial family-based association studies, and then case-control genome-wide association studies (GWAS) (see Glossary) have failed to identify any common genetic variants with large effects. A handful of reported associations surpass conventional levels of genome-wide signiﬁcance ( p < 5 10 8) but none accounts for much of the morbidity associated with schizophrenia; see SZGene (SchizophreniaGene) Field Synopsis of Genetic Corresponding author: Bilder, R.M. ([email protected]).

428

Association Studies (http://www.szgene.org) for regularly updated meta-analytic ﬁndings. It seems probable that many genetic variants (perhaps thousands) interact with each other and the environment to account for schizophrenia risk, and indeed this risk is likely to be shared at least with bipolar disorder and probably autism and other complex neurodevelopmental disorders [2,3]. Glossary Candidate gene study: a study of association between genotype and phenotype(s) in which the genetic variation is selected based on prior evidence of association. Copy number variation (CNV): a variation between individuals of the same species in the structural sequence of nucleotides due to repetition or deletion of part of the sequence; for instance, a ‘trinucleotide repeat’ is the repetition of three nucleotides; ‘CAG repeats’ are one example already identified as responsible for multiple neurological disorders including Huntington’s disease [34]. Endophenotype/Intermediate phenotype: the term ‘endophenotype’ as it is now widely used in psychiatry stems from original usage by Gottesman and Shields who referred to ‘internal features’, ‘only knowable after aid to the naked eye’, such as the help provided by ‘a biochemical test or a microscopic examination’ ([35], p. 19). More recently, Gottesman and Gould defined endophenotypes as ‘measurable components unseen by the unaided eye along the pathway between disease and distal genotype’ ([36], p. 636). We believe this added specification is important because it implicates endophenotypes as being the product of genetic variation that is part of the ‘causal path’ between genotype and some other phenotype of interest (usually one that is either the diagnosis or part of the diagnosis). The terms ‘endophenotype’ and ‘intermediate phenotype’ are now widely used interchangeably to refer to any phenotype that is presumed to be ‘closer’ to the level of gene action than is the observable phenotype of interest (for example, ‘molecular expression’ is considered closer to gene action than behavioral ‘symptoms’ of schizophrenia). We believe the term ‘intermediate phenotype’ better reflects the putative causal attributes of phenotypes that actually are part of the causal path from genotype to some higher-level phenotype; but so far few such intermediate phenotypes have been proven. Genome-wide association study (GWAS): a study of association between genotypes and phenotype(s) in which genetic variations are sampled widely over the entire genome, theoretically in a manner that provides ‘coverage’ of the entire genome. Paraphenotype: a phenotype at the same level of expression with respect to causal models of the relation between a genotype and some other phenotype of interest; a phenotype that shares causal precedents with another phenotype (Figure 1). Phenomics: the systematic study of phenotypes on a genome-wide scale, emphasizing the simultaneous analysis of multiple phenotypes. Single-nucleotide polymorphism (SNP): a variation between individuals of the same species in the structural sequence of nucleotides in the DNA molecule at a single location involving the substitution of one nucleotide for another. Given that there are four nucleotides: adenosine, thymine, cytosine and guanine; a SNP is said to exist when one of these nucleotides is replaced by another.

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.002 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

Review Box 1. Cognitive impairment in schizophrenia: why is it important? The cognitive impairment associated with schizophrenia is severe, widespread and apparent long before overt signs of psychosis emerge [37–39]. There were initially hopes that new antipsychotic drug treatments might ameliorate these deficits but large-scale trials with these agents have shown limited success. Large-scale effectiveness studies suggest no significant cognitive benefit for new antipsychotic agents [40,41], and so far no agent has been approved by the US Food and Drug Administration for the indication of cognitive impairment associated with schizophrenia despite wide interest. There remains hope that in the future pharmacogenomic strategies could yield treatment benefits for schizophrenia [42], and innovative strategies are being advanced particularly to identify new molecular targets linked directly to cognitive dimensions rather than the traditional symptomatic dimensions of schizophrenia [43]. These findings highlight the importance of identifying genetic bases of cognitive dysfunction in schizophrenia both for increasing understanding of pathophysiology and for developing more effective treatments.

Investigations focused on ‘endophenotypes’ or ‘intermediate phenotypes’, including cognitive phenotypes, are now emerging [4], with the hope that these might offer more traction in deciphering the complex genetics of brain dysfunction in schizophrenia and perhaps other neurodevelopmental syndromes. Meanwhile, most investigators are likely to agree that unraveling the genetic bases of schizophrenia and its associated cognitive impairment is proving more difﬁcult than was hoped earlier [5]. Major questions remain as to how we might best gain traction on the elusive biological roots of schizophrenia. The new transdiscipline referred to as ‘phenomics’ (the systematic study of phenotypes on a genome-wide scale) could offer one perspective [6]. Most efforts so far have targeted the syndromal phenotype of schizophrenia in case-control studies in hopes of ﬁnding genetic association. By contrast, phenomics approaches consider multiple phenotypes, including those that could be measured on different biological scales, to better deﬁne biologically plausible traits. It is assumed that by combining information from multiple levels (from the level of gene expression through proteomic, metabolomic, cellular and systems levels) we could better characterize the biological contributions of speciﬁc genetic variants and their interactions in a way that ultimately permits personalization of diagnosis and rational treatment. In research on schizophrenia, for example, it has been hoped that we might gain clearer insights from the simultaneous focus on the symptoms that mark the syndrome and the cognitive deﬁcits that consistently accompany the syndrome. In this article we aim to: (i) provide a brief synopsis of current knowledge about the genetic basis of cognitive deﬁcits in schizophrenia; (ii) highlight selected conceptual issues that we believe will be important to make further progress in ﬁnding the genetic bases for cognitive phenotypes in schizophrenia and (iii) introduce a new freely available resource – CogGene – that we hope can serve the ﬁeld by helping to aggregate, visualize and analyze relevant evidence for those interested in the genetic bases of cognitive function in schizophrenia and other disorders. An important point about a knowledgebase such as CogGene is that it might help advance understanding of the

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

genetic bases of cognitive deﬁcits in schizophrenia even if the knowledgebase is focused selectively on ﬁndings in healthy people. In brief, we hope a tool such as CogGene can help researchers ‘triangulate’ genetic association ﬁndings: if a speciﬁc genetic variation is associated with the diagnosis of schizophrenia AND the same genetic variation is associated with cognitive impairment in otherwise healthy people, then it increases the likelihood that this variant could be related to BOTH schizophrenia and cognitive impairment through a common mechanism. Indeed this strategy might be more informative than examining association of a genetic variant with cognitive deﬁcits within schizophrenia samples because cognitive impairment is ‘confounded with’ the diagnosis of schizophrenia. What do we know so far about the genetics of cognitive impairment in schizophrenia? It has long been known that both schizophrenia and cognitive impairment are highly heritable and it has long been assumed that some genetically mediated anomaly, probably a neurodevelopmental anomaly, underlies the vulnerability to both schizophrenia and the cognitive impairment that invariably accompanies the syndrome. The heritability of schizophrenia is estimated at greater than 0.8, whereas the heritability of cognitive phenotypes is most often found to be near 0.5, regardless of whether the estimate is derived from healthy or ill groups [7,8]. Compelling evidence has also been provided to show that many different cognitive abilities could be linked, not only by their covariation within individuals but further by their shared genetic correlations; indeed this has led to the ‘generalist gene’ hypothesis that many presumably diverse cognitive functions are likely to be associated with a common set of genetic variations [9–11]. Recent work has begun to identify the shared genetic components of cognitive phenotypes and syndromal phenotypes such as schizophrenia; in brief, the cognitive phenotypes and schizophrenia are signiﬁcantly correlated, and the lion’s share of this covariance (72% to 92%) is due to shared genetic effects [12–14]. Despite the relatively high heritability of these phenotypes and their high genetic correlation, we so far have in hand no well-validated candidate genes that explain much of the variance in either schizophrenia or cognitive phenotypes (with the exception of selected rare genes in which mutations cause large effects on cognition, as described below). Meanwhile it increasingly looks as though the shared liability for schizophrenia and cognitive impairment is most likely to be identiﬁed through a relatively large number of genetic inﬂuences, some coming from larger impacts of rare variants, and some coming from larger numbers of more common variants with very small effects working in combination to undermine healthy brain development and signaling. The search for cognitive phenotypes Cognitive phenotypes as intermediate phenotypes or as paraphenotypes One frequently asked question is whether cognitive deﬁcits or symptoms are epiphenomenal: that is, do cognitive deﬁcits cause schizophrenia or are cognitive deﬁcits caused by schizophrenia? Although this distinction might be seen 429

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

[(Figure_1)TD$IG]

as an irrelevant exercise in semantics, it is important for modeling. We see both symptomatic and cognitive measures as similar in level of explanation, given that both are behavioral manifestations of neural systems activity. This point of view casts skepticism on the likelihood that cognitive phenotypes will serve well as intermediate phenotypes because they are not really intermediate in the sense that this term is used in causal models (i.e. they are not likely to be mediating variables for symptoms, at least in schizophrenia). Cognitive phenotypes might nevertheless be of value as ‘paraphenotypes’ (i.e. phenotypes that are at the same hierarchic level within a causal model, and are alongside each other) because they are better validated with respect to neural systems phenotypes (Figure 1b: if path coefﬁcient x>y). Adding cognitive measures to a multivariable phenotype could thus help constrain the neural system phenotypes to a subset of all neural systems that might be part of the mechanistic path from genome to syndrome, and thereby help increase statistical power for detecting associations with ‘lower’ level biological processes including genetic variation. It should be recognized, however, that such arguments are at this point largely theoretical and there are few conﬁrmatory or disconﬁrmatory examples in practice. So far we can say only that cognitive phenotypes have not shown ‘simpler’ genetic architecture than complex syndromal phenotypes [15]. The genetic bases of cognitive phenotypes There are multiple potential windows on the genetic bases of cognitive phenotypes. Some of the earliest and most successful approaches found genetic associations with cognitive impairment syndromes, particularly mental retardation. Indeed mental retardation could be seen as a phenotype for which genetic studies have been particularly successful, with approximately 300 identiﬁed monogenic causes; however, it should be recognized that these are rare (i.e. most account for only 0.01% of all cases) [16].

(a) Intermediate phenotype or endophenotype model Neural system phenotype

Cognitive phenotype

Symptom phenotype

(b) Parallel, common parent, or paraphenotype model

x

Cognitive phenotype

Neural system phenotype

y

Symptom phenotype TRENDS in Cognitive Sciences

Figure 1. Relations of cognitive phenotypes to neural systems and diagnosis. Path diagrams schematizing different points of view regarding the role of cognitive phenotypes as: (a) intermediate phenotypes or endophenotypes or (b) as different behavioral effects – paraphenotypes – that differ primarily in the strength of relations to neural system activity.

430

Despite the low frequency of these conditions, they could be informative about mechanisms important to brain development and cognition. For example, the study of Fragile X syndrome (a genetic condition involving changes in part of the X chromosome) has led to multiple insights about the genetics of trinucleotide repeats, X-linked genetic disorders, and the enormous pleiotropy of single-gene deﬁcits on neural and other systems [17]. Similarly the study of neuroﬁbromatosis (a genetic disorder of the nervous system, which mainly affects how nerve cells form and grow), and the NF1 gene, has yielded major insights into the molecular basis of these syndromes, yielded novel transgenic rodent models in which mutants have superior abilities, and might stimulate novel treatment development [18,19]. It should be recognized that even when genetic studies reveal compelling associations that are considered signiﬁcant at genome-wide levels and replicated, the identiﬁed variants might still account for only a small amount of the known heritability. Human height is a good example phenotype: despite a heritability near 80%, only about 5% of phenotypic variance is explained by more than 40 known loci [20]. This has been referred to as the problem of ‘missing heritability’ or the ‘dark matter’ of heritability, and could be due to many reasons, including: (i) variants that the GWAS arrays are missing, that is the single nucleotide polymorphisms (SNPs) that have yielded association ﬁndings might not be the causative SNPs, and the true causative SNPs might have larger effects; (ii) gene–gene interactions (epistasis) and/or gene–environment interaction effects too complicated to assess given current sample sizes and analytic strategies; (iii) epigenetic effects; (iv) much larger numbers of genetic variants with even smaller effects remaining to be found; (v) inadequate accounting for shared environmental variance among relatives [21]. Although work so far using GWAS to detect associations with cognitive phenotypes has been unsuccessful at replicating results from some prior ‘candidate gene’ studies, it is worth noting that the results remain consistent with the hypothesis that cognitive impairment could be associated with an increase in genetic variants each with small effect [22]. Sabb and colleagues summarized prior work on candidate genes for which investigators reported associations with cognitive phenotypes comprising ‘memory’ (51 effects) and ‘intelligence’ (42 effects) [23]. They found generally modest associations of candidate genes with varying cognitive phenotypes, with most effect sizes (Cohen’s d for the effect distinguishing alleles) ranging from 0.09 to 0.23. An interesting result of this survey was that among genes investigated, two had relations speciﬁcally with intelligence (CHRM2, DRD2), two had relations speciﬁcally with memory phenotypes (5-HTT, KIBRA), and four had reported links to both intelligence and memory phenotypes (DTNB1, COMT, BDNF, APOE). Others have highlighted the replication of selected ﬁndings related to rare variants in key genetic regions (such as PDE10A, CYSIP1, KCNE1/ KCNE2, CHRNA7) and their possible connection to both

Review schizophrenia and cognitive impairment phenotypes [24]. It must be recognized, however, that these ﬁndings could still reﬂect false positive reports. Sources of bias include the targeting of certain genes as candidates without very strong a priori evidence, considering different SNPs within a gene as replications of a speciﬁc gene ﬁnding, and as highlighted by Sabb and colleagues, dubious measurement of cognitive phenotypes (for example, one of the measures of ‘memory’ was the Mini Mental State Exam, and another was ‘Fluency’, despite the fact that these would be questioned by most investigators). Sabb and colleagues established an open database to share knowledge about these associations; see Phenowiki: an Online Collaborative Database for Phenotype Annotation (http://www.Phenowiki. org) [7]. Hopefully, continued development of such resources will ultimately enable convergence on the meaningful associations and reﬁnement of phenotype deﬁnitions perhaps narrowing down to those that have the clearest relations for broader population studies. CogGene: a collaborative knowledgebase for documenting genetic associations with cognitive phenotypes With this background, we aimed to determine to what extent data regarding genetic associations with the schizophrenia phenotype might be enriched by examining correlations of the same genetic targets with cognitive phenotypes. This approach could be considered a triangulation of genes at the intersection of schizophrenia and cognitive impairment. To gather data relevant to this we elaborated on the Phenowiki database architecture and created a new knowledgebase and web service speciﬁcally to represent genetic association ﬁndings for cognitive phenotypes: UCLA CogGene (http://www.CogGene.org). We aimed to have an interface similar in some ways to those available in SZGene and AlzGene, which provide forest plots of effect sizes, but our data differ insofar as the phenotypes to be represented are quantitative trait scores (rather than categorical diagnoses). Further, due to the richness of the cognitive phenotype data (which includes both test names and then speciﬁc measurement variables or ‘indicators’ within each test), we created features in CogGene to enable dynamic sorting and computation of weighted effect size statistics over groups of results that can be selected simply by clicking on the effect labels. The data discussed in this paper and the CogGene system are now viewable at UCLA CogGene, and information is posted on the site regarding how to submit additional contributions. The ﬁndings described here were culled from publications identiﬁed through literature mining if they cited the names of at least one gene and related polymorphism, and at least one cognitive test: the names are from a lexicon developed in the Consortium for Neuropsychiatric Phenomics at UCLA (http://www.phenomics. ucla.edu). From these publications we selected those with usable data (i.e. with data specifying at least one statistical association between a speciﬁc SNP and a speciﬁc cognitive test indicator), and extracted quantitative effect sizes for associations between SNPs and cognitive test indicators. We highlight that these data were selected to represent results from healthy samples, in order to maximize the

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

independence of ﬁndings from those in SZGene (and thus enabling us to inspect possible overlaps in the ‘top hits’ free of the potential confounds between schizophrenia and cognitive impairment phenotypes). The results can now be browsed, sorted and reanalyzed using custom-designed software tools that permit visualization and execution of ‘meta-analysis’ (sample-size-weighted averaging) over selected effects under user control. In brief, the CogGene system permits visualization of the effect sizes for speciﬁc allelic variants on the cognitive trait scores (expressed as Cohen’s d statistic, which is the standardized difference between group means), and the 95% conﬁdence intervals around these difference scores, in an interactive forest plot. This is similar to the representation of genetic association data in other widely used resources (SZGene, AlzGene) which use similar (but static) forest plots to show allelic associations with case-control differences (but in these examples, the effect sizes are expressed as odds ratios rather than group differences). A typical screenshot of CogGene is shown in Figure 2. The SZGene database contains information from 1727 studies, reporting data on 1008 genes, and 8788 polymorphisms; this database has 287 meta-analyses [25]. SZGene ranks its ‘top results’ using the HuGENet interim guidelines published by Ioannidis and colleagues [26], which consider the ‘amount’ of evidence (i.e. Grade ‘A’ is given to studies in which the total number of minor alleles exceeds 1000), ‘consistency’ of evidence (i.e. Grade ‘A’ is given only when inconsistency is modest, for example I2<25) and ‘bias’ (with Grade ‘A’ given when there is probably no bias). Inspecting initial entries into the CogGene database, we note that the quality of genetic association studies for cognitive phenotypes so far is relatively low. For example, none of the studies meets the criteria to be considered Grade ‘A’ following HuGENet criteria for amount of evidence, and only 3 studies would receive a Grade of ‘B’ (i.e. with minor allele frequencies greater than 100; the rest would all be considered Grade ‘C’). Analysis of existing studies is further complicated by the lack of uniformity in phenotype deﬁnition, rendering replication of results difﬁcult to determine because few studies use exactly the same indicators. Finally, the degree of bias in the cognitive studies could be considered relatively high, given the paucity of large effect sizes. Examining the 45 top results of SZGene, we ﬁnd that 10 of the same genes are listed in CogGene. Among these 10 genes, we ﬁnd that only evidence supporting association for two of these genes (APOE, HTR2A) is considered Grade ‘A’ in SZGene. APOE (e2/3/4; contrasting 4 versus 3 allele) is signiﬁcantly associated with schizophrenia among Caucasians; and HTR2A (rs6311; contrasting A versus G allele) is associated with schizophrenia, also selectively in Caucasian samples. Figure 3 provides a graphical summary of the top results from CogGene, considering only those individual SNP effects that had 95% conﬁdence intervals not including zero difference between allelic variants. Among these, only the APOE genotype overlaps with those identiﬁed as a top result in the SZGene database, and as Figure 3 shows, the average effect size for APOE is small (d = 0.069, 95% conﬁdence interval = 0.014 to 0.124). It should 431

(Figure_2)TD$IG][ Review

pI D

ca

0.31

0.47

0.02

0.45 17892388

S

eq By h... Bx h... Ay h... Ax h... Avg ... Ind i

re f rs4570625

TPH2

Ge n

e

to

r

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

0.31

0.31

-0.41

0

0.9 TRENDS in Cognitive Sciences

Figure 2. Screenshot illustrating features of the CogGene web service. Users can filter records in the CogGene database either by ‘gene’ or by ‘task’, and the system will then show all relevant effects in a dynamic forest plot. The example shows results selected for the task ‘AX-CPT’ (the ‘AX’ version of the Continuous Performance Test), filtered to show results only for the gene TPH2 from a single publication (see green bar on right side of figure, which gives the PubMedID (pID), and a single SNP (see purple bar on left side of figure, labeled ‘refSeq’). In this example, the publication had examined 5 different Indicators (see 5 bars stacked on the left side of the figure). If the user clicks on any of these bars the CogGene system sorts all results based on that column (i.e. by gene, task, indicator, refSeq or PubMedID). Note also that each bar has a checkbox in its upper left corner; users can check these boxes and then the meta-analysis calculations will be executed over the checked results. The actual entries in the forest plot show the effect size d for the specific indicator and refSeq contrast for that study (orange circles) (not shown is that the mouse flyover function reveals exactly which allelic contrast is being represented). At the bottom of the figure, the blue diamond represents the meta-analytic result (sample-size-weighted average of all selected effects). The gray bars represent 95% confidence intervals around the individual or meta-analytic effect size values. Further details are available at CogGene (http:// www.CogGene.org).

be recognized that this effect for APOE genotype is small in part because it is averaging together effects on different cognitive indicators. We have for APOE two studies with the same cognitive indicator (Buschke Selective Reminding Test, Long Term Recall), and the same contrast among alleles; these two studies [27,28] have overlapping samples so probably the larger of the two studies should be relied on by itself. On a positive note, this single study [28] showed a medium effect (i.e. comparing the e2/2 and e2/3 to e3/3 had d = 0.29 and comparing e2/2 and e2/3 to e4 allele carries had d = 0.39), with total sample size of 912 and minor allele frequency of 76; this more detailed inspection of the ﬁndings indicates that this result stands as an isolated ﬁnding without replication. Among the top results in CogGene, none of the effects so far would be considered signiﬁcant at conventional genome-wide levels, at least in part because the sample sizes are so low. For example, only the APOE and DTNB1 ﬁndings are supported by a study with a sample size exceeding 500 cases, and the largest effect (CACNA1C) is supported by a study of only 80 people, only 10 of who possessed the minor allele at the investigated locus (rs1006737). These results are consistent with those reported by Sabb and colleagues [23], where as noted above, all effect sizes were in the range of d = 0.09 to d = 0.23, with the single exception (d = 0.44) being a study for which total sample size was only 201. This highlights the possibility of publication bias and, so far, small sample studies, which poses a major challenge to cognitive genomics, and the likelihood that many of the reported associations will turn out to be false positive results. Currently, the literature remains rife with ﬁndings that center on 432

selected ‘candidate’ genes that have been investigated at least in part due to inertia from earlier positive reports (for example, the study of APOE genotype in schizophrenia reﬂects more the ‘smoke’ from positive ﬁndings in Alzheimer’s disease than the likely ‘ﬁre’ in schizophrenia). This bias might soon be overcome as more GWAS results and then genome sequencing ﬁndings are disseminated. At that point the biggest priorities will be to obtain unbiased sampling of the ‘cognitive phenome,’ else we will run similar risk of biases from studying the wrong candidate phenotypes that we currently face in studying false positive candidate genotypes [29]. This will be an interesting challenge for future investigations, which will need to balance consistency and standardization of phenotyping that are crucial for replication, with sufﬁciently broad sampling to help reduce phenotyping bias. We hope that further development of CogGene will help aggregate ﬁndings across investigations, increase our understanding of where relevant signals might lie, and shed light on the design of future studies and collaborative research programs. The most recent ﬁndings regarding the genetics of schizophrenia and cognitive impairment phenotypes suggest we are likely to face a deluge of associations with very small effects, and a smaller number of rare variants possibly with larger effects, along with probable complex gene– gene and gene–environment interactions. These observations make the availability of a collaborative knowledgebuilding tool such as CogGene particularly valuable because sifting through the ﬁndings, and aggregation of results across diverse studies, could ultimately be more important than results from any single study. By structuring knowledge in CogGene we hope also to facilitate links

(Figure_3)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Effect size (d) .0

.1

.2

.3

.4

.5

.6

.7

.8

.9

APOE CHRM2 DTNBP1 DRD2 (a) DRD2 (b) CHRM2,CHRNA4 IL1B KIBRA SNAP25 (a) SNAP25 (b) IL1RAPL1 CACNA1C

TRENDS in Cognitive Sciences

Figure 3. Top results from CogGene. Effect size statistics from CogGene (expressed in terms of Cohen’s d statistic) for genetic associations with cognitive indicators, based on sample-size-weighted mean effect sizes for each SNP. Error bars provide 95% confidence intervals around each mean effect, only effects not including zero are shown. DRD2 and SNAP25 each had two SNPs satisfying these criteria.

to other knowledgebases (such as the Entrez systems supported by the National Library of Medicine) to promote biological discovery and better constrain our models of the causal paths that connect the human genome to complex disorders of brain and behavior. A related challenge pertains to developing standards for cognitive phenotyping and reﬁnement of ontologies that can help formalize knowledge within this scientiﬁc domain. Sabb and colleagues showed how ﬁckle investigators can be, introducing new concept labels despite lack of change in the actual measurement methods [7]. We have suggested frameworks for developing cognitive ontologies elsewhere [6,30,31], and the Cognitive Atlas project (http:// www.CognitiveAtlas.org) is dedicated speciﬁcally to development of a consensus ontology about cognitive concepts and their measurement. This work will be essential to help determine which speciﬁc ﬁndings can be meaningfully averaged in meta-analytic studies that will ultimately help us identify and understand what are likely to be myriad small signals relating cognitive phenotypes to the genome. Finally, the development of tools such as CogGene can help represent quantitative trait data for genetic associations and thus offer a means for collaboration, storage and reuse of knowledge that is important to the dimensional

representation of phenotypes. This is compatible with the National Institute of Mental Health Strategic Plan, and speciﬁcally of potential value to the new Research Domain Criteria initiative [32,33], which aims to support research on phenotypic dimensions that could be more informative than traditional diagnostic phenotypes. Concluding remarks In summary, we considered the existing literature on genetic associations with the syndromal phenotype of schizophrenia and its conjunction with ﬁndings from the study of genetic associations with cognitive phenotypes in healthy people. The work on schizophrenia is more advanced and contains a few leads, albeit we are seeing at most the tip of the iceberg in understanding the contributions to this genetic risk, and much ‘dark matter’ (missing heritability) remains to be deﬁned. The work on cognitive genomics requires signiﬁcant advances in methods and study quality to yield more credible ﬁndings, and to add substantively to understanding genetic risks for cognitive impairments in schizophrenia and other neurodevelopmental disorders. Among the challenges are: (i) increasing sample sizes, probably at least by an order of magnitude relative to the published work available now; (ii) increasing 433

Review Box 2. Questions for future research How can we best investigate those cognitive phenotypes that reflect the most discrete and biologically meaningful targets (i.e. what are the best measures to increase understanding of the biological bases of cognitive phenotypes and how do we validate these)? How can we best aggregate information from different measures that we think might be measuring the same basic dimension of cognitive function (i.e. when can we successfully ‘mate’ findings from different studies in meta-analyses)? How can we best translate cognitive phenotype studies across species, and particularly how can we best leverage the ability to develop strong genetic models in the mouse that will be informative about cognitive phenotypes in humans? How can we best target genetic analyses based on prior knowledge about cognitively-relevant biology (i.e. can we effectively constrain analysis models based on prior knowledge of selected neurodevelopmental processes, or signaling pathways, that we believe are crucial for cognitive function)? Assuming we can identify some genetic variants that confer shared vulnerability to schizophrenia and cognitive impairment, can we leverage this knowledge to better understand those genetic features that are unique to schizophrenia and not associated with cognitive impairment (and those that are associated with cognitive impairment but not schizophrenia)?

standardization in cognitive phenotyping to enable more direct replication; (iii) increasing coverage of cognitive domains within each study to help attenuate bias in sampling from the cognitive phenome (Box 2). These goals are particularly daunting to achieve given constraints on time and budgets. Major progress could be fostered by routine aggregation of genome-wide sequencing data (which we consider probable within a decade) together with widely distributed (internet- and mobile application-based) cognitive phenotyping, and the reﬁnement of methods to represent cognitive concepts and the speciﬁc measurements used to deﬁne these. We introduced here the CogGene knowledgebase, a freely available online collaborative tool to help aggregate and meta-analyze relevant evidence, which we hope will advance understanding of the genetic bases of complex syndromes involving brain and behavior. Acknowledgements This work was supported by the Consortium for Neuropsychiatric Phenomics (NIH Roadmap for Medical Research grants UL1-DE019580, RL1LM009833, PL1MH083271) and the Tennenbaum Family Center for the Biology of Creativity.

References 1 Kraepelin, E. (1919) Dementia Praecox and Paraphrenia, E and S Livingstone 2 Purcell, S.M. et al. (2009) Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature 460, 748–752 3 Rzhetsky, A. et al. (2007) Probing genetic overlap among complex human phenotypes. Proc. Natl. Acad. Sci. U.S.A. 104, 11694– 11699 4 Greenwood, T.A. et al. (2011) Analysis of 94 candidate genes and 12 endophenotypes for schizophrenia from the Consortium on the Genetics of Schizophrenia. Am. J. Psychiatry DOI: 10.1176/ appi.ajp.2011.10050723 5 Bilder, R.M. (2009) The neuropsychology of schizophrenia circa 2009. Neuropsychol. Rev. 19, 277–279 6 Bilder, R.M. et al. (2009) Phenomics: the systematic study of phenotypes on a genome-wide scale. Neuroscience 164, 30–42 7 Sabb, F.W. et al. (2008) A collaborative knowledge base for cognitive phenomics. Mol. Psychiatry 13, 350–360

434

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 8 Greenwood, T.A. et al. (2007) Initial heritability analyses of endophenotypic measures for schizophrenia: the consortium on the genetics of schizophrenia. Arch. Gen. Psychiatry 64, 1242–1250 9 Plomin, R. et al. (2007) Generalist genes: genetic links between brain, mind, and education. Mind Brain Educ. 1, 11–19 10 Haworth, C.M. et al. (2009) Generalist genes and high cognitive abilities. Behav. Genet. 39, 437–445 11 Plomin, R. and Haworth, C.M. (2009) Genetics of high cognitive abilities. Behav. Genet. 39, 347–349 12 Toulopoulou, T. et al. (2007) Substantial genetic overlap between neurocognition and schizophrenia: genetic modeling in twin samples. Arch. Gen. Psychiatry 64, 1348–1355 13 Toulopoulou, T. et al. (2010) Impaired intellect and memory: a missing link between genetic risk and schizophrenia? Arch. Gen. Psychiatry 67, 905–913 14 Owens, S.F. et al. (2011) Genetic overlap between episodic memory deﬁcits and schizophrenia: results from the Maudsley Twin Study. Psychol. Med. 41, 521–532 15 Flint, J. and Munafo, M.R. (2007) The endophenotype concept in psychiatric genetics. Psychol. Med. 37, 163–180 16 Butcher, L.M. et al. (2006) Generalist genes and cognitive neuroscience. Curr. Opin. Neurobiol. 16, 145–151 17 Heulens, I. and Kooy, F. (2011) Fragile X syndrome: from gene discovery to therapy. Front. Biosci. 16, 1211–1232 18 Silva, A.J. et al. (2009) Molecular and cellular approaches to memory allocation in neural circuits. Science 326, 391–395 19 Lee, Y.S. and Silva, A.J. (2009) The molecular and cellular biology of enhanced cognition. Nat. Rev. 10, 126–140 20 Visscher, P.M. (2008) Sizing up human height variation. Nat. Genet. 40, 489–490 21 Manolio, T.A. et al. (2009) Finding the missing heritability of complex diseases. Nature 461, 747–753 22 Need, A.C. et al. (2009) A genome-wide study of common SNPs and CNVs in cognitive performance in the CANTAB. Hum. Mol. Genet. 18, 4650–4661 23 Sabb, F.W. et al. (2009) Challenges in phenotype deﬁnition in the whole-genome era: multivariate models of memory and intelligence. Neuroscience 164, 88–107 24 Tam, G.W. et al. (2010) Conﬁrmed rare copy number variants implicate novel genes in schizophrenia. Biochem. Soc. Trans. 38, 445–451 25 Allen, N.C. et al. (2008) Systematic meta-analyses and ﬁeld synopsis of genetic association studies in schizophrenia: the SzGene database. Nat. Genet. 40, 827–834 26 Ioannidis, J.P. et al. (2008) Assessment of cumulative evidence on genetic associations: interim guidelines. Int. J. Epidemiol. 37, 120–132 27 Helkala, E.L. et al. (1996) Memory functions in human subjects with different apolipoprotein E phenotypes during a 3-year populationbased follow-up study. Neurosci. Lett. 204, 177–180 28 Helkala, E.L. et al. (1995) The association of apolipoprotein E polymorphism with memory: a population based study. Neurosci. Lett. 191, 141–144 29 Yarkoni, T. et al. (2010) Cognitive neuroscience 2.0: building a cumulative science of human brain function. Trends Cogn. Sci. 14, 489–496 30 Bilder, R.M. (2011) Neuropsychology 3.0: evidence-based science and practice. J. Int. Neuropsychol. Soc. 17, 7–13 31 Bilder, R.M. et al. (2009) Cognitive ontologies for neuropsychiatric phenomics research. Cogn. Neuropsychiatry 14, 419–450 32 Insel, T.R. and Cuthbert, B.N. (2009) Endophenotypes: bridging genomic complexity and disorder heterogeneity. Biol. Psychiatry 66, 988–989 33 Insel, T. et al. (2010) Research domain criteria (RDoC): toward a new classiﬁcation framework for research on mental disorders. Am. J. Psychiatry 167, 748–751 34 Stranger, B.E. et al. (2007) Relative impact of nucleotide and copy number variation on gene expression phenotypes. Science 315, 848 35 Gottesman, I.I. and Shields, J. (1973) Genetic theorizing and schizophrenia. Br. J. Psychiatry 122, 15–30 36 Gottesman, I.I. and Gould, T.D. (2003) The endophenotype concept in psychiatry: etymology and strategic intentions. Am. J. Psychiatry 160, 636–645 37 Allott, K. et al. (2011) Cognition at illness onset as a predictor of later functional outcome in early psychosis: systematic review and methodological critique. Schizophr. Res. 125, 221–235

Review 38 Bora, E. et al. (2009) Cognitive functioning in schizophrenia, schizoaffective disorder and affective psychoses: meta-analytic study. Br. J. Psychiatry 195, 475–482 39 Mesholam-Gately, R.I. et al. (2009) Neurocognition in ﬁrst-episode schizophrenia: a meta-analytic review. Neuropsychology 23, 315–336 40 Keefe, R.S. et al. (2007) Neurocognitive effects of antipsychotic medications in patients with chronic schizophrenia in the CATIE trial. Arch. Gen. Psychiatry 64, 633–647

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 41 Hill, S.K. et al. (2010) Effect of second-generation antipsychotics on cognition: current issues and future challenges. Expert Rev. Neurotherapeutics 10, 43–57 42 Cacabelos, R. et al. (2011) Pharmacogenomics of antipsychotics efﬁcacy for schizophrenia. Psychiatry Clin. Neurosci. 65, 3–19 43 Krystal, J.H. et al. (2009) Neuroplasticity as a target for the pharmacotherapy of anxiety disorders, mood disorders, and schizophrenia. Drug Discov. Today 14, 690–697

435

Review

Special Issue: The Genetics of Cognition

The contribution of imaging genetics to the development of predictive markers for addictions Eva Loth*, Fabiana Carvalho* and Gunter Schumann MRC-SGDP-Centre, Institute of Psychiatry, King’s College, London SE5 8AF, UK

A key challenge for intervention and prevention of addictions is the identification of genetic, neurobiological and cognitive risk profiles that can predict which adolescents are at risk for addiction. Abnormalities in reinforcement behaviour have been linked to addiction vulnerability and imaging genetic studies have begun to elucidate the mechanisms by which genetic and environmental factors influence brain function underlying individual variability in reinforcement behaviour. Most studies have examined associations between a few well-characterised candidate polymorphisms and task-related brain activation differences in individual regions of interest. Here we propose that integrating the imaging genetic strategy with biological network approaches and longitudinal adolescent designs in large multi-centre samples may offer promising opportunities to identify risk markers for early diagnosis, progression and prediction of addictions. The role of reinforcement behaviour – reward processing, inhibitory control and emotional responses – in addiction vulnerability Addictions are moderately to highly heritable disorders [1]. Although most people drink alcohol or try one or more drugs at some point in their lives, only a subset of users progress to abuse and dependence. Age of ﬁrst experimentation with drugs is currently one of the best predictors for addiction development and severity [2]. For example, most adult daily smokers started smoking before the age of 14 [3], and adolescents who started using cocaine at 13-14 years are four times more likely to become dependent than those starting at 18-19 years [4]. Identiﬁcation of risk proﬁles at the genetic, neurobiological and cognitive levels in adolescents, which predict future addictions, will make it possible to maximize the efﬁcacy and efﬁciency of early interventions and reduce individual suffering as well as the enormous public health burden caused by addictions (World Health Organization, 2004). One way of conceptualizing risk for drug use is by investigating individual variability in reinforcement processes. Reinforcement behaviour includes sensitivity to rewards and punishment, inhibitory control, and

*

436

Corresponding author: Schumann, G. ([email protected]) These authors contributed equally to this work.

emotional reactivity [5]. Longitudinal studies have shown that individual variability in personality traits involving positive (impulsivity, sensation seeking, reward dependence) or negative reinforcement processes (i.e., relief of negative states, such as anxiety, stress) [6] predict different motivations for drug use behaviours. However, candidate gene, genetic linkage and genomewide association studies with quantitative personality traits have produced mixed results with small effect sizes and are often difﬁcult to replicate [7]. Furthermore, these associations do not explain the complexity of underlying neurobiological mechanisms by which genetic variants and environmental factors affect temperament and behaviour. To address these issues, over the past few years, the Glossary Endophenotypes (or intermediate phenotypes): stable, heritable traits at the personality, cognitive, or neurobiological levels, which are related to a particular disorder or symptoms, detectable in ‘at-risk’ but behaviourally unaffected individuals and are thought to be more homogeneous and ‘closer’ to genetic variants than the clinical phenotype. Endophenotype approaches have become influential concepts in genetic research aimed at reducing the phenotypic heterogeneity characteristic of many complex psychiatric disorders. Imaging genetic (or gene-neuroimaging) approaches: association studies that link genetics to individual variability in brain morphology or brain function. It rests on the assumption that anatomical differences or blood-oxygenated dependent level (BOLD] responses during particular cognitive or emotional processes may provide a temporally stable intermediate phenotype. Systems biology: an inter-disciplinary approach in bioscience research, which studies the interactions between the components of biological systems, and how these interactions give rise to the function and behaviour of that system, including so-called ‘emergent properties’ (for example, the enzymes and metabolites in a metabolic pathway). Reinforcement: the term ‘reinforcement’ stems from the operant conditioning literature to describe the process by which the probability or rate of a behaviour or response is increased by the delivery of a stimulus immediately or shortly after performing the behaviour. Positive reinforcement refers to the delivery of an appetitive stimulus to increase a certain behaviour or response. Negative reinforcement refers to the alleviation or omission of an aversive stimulus to increase a certain behaviour or response. Neurotransmitters: endogeneous chemicals, which transmit signals from one neuron to another neuron across a chemical synapse. Neurotransmitters are packaged into synaptic vesicles beneath the pre-synaptic membrane and released into the synaptic cleft following an action potential, where they bind to specific receptors in the post-synaptic membrane. Neurotransmitters are commonly classified into amino acids (for example, glutamate is the most prevalent excitatory neurotransmitter, y-aminobutyric acid (GABA), the most prevalent inhibitory neurotransmitter), monoamines, (which include the catecholamines dopamine (DA), norephinphrine (also known as noradrenaline, NE/ NA), and epinephrine/ adrenaline), as well as histamine and serotonin, and neuropeptides (for example, neuropeptide Y, b-endorphin, vasopressin, oxytocin, among others).

1364-6613/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2011.07.008 Trends in Cognitive Sciences, September 2011, Vol. 15, No. 9

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Table 1. Gene variants associated with activation in brain areas related to cognitive systems of reinforcement Gene

Function/System

Polymorphism

Brain Regions

DRD4 DRD2 DAT1 (SLC6A3) COMT

receptor/dopaminergic receptor/dopaminergic transporter/dopaminergic

VNTR SNP VNTR

IFG, VS VS ACC, VS, PMC

metabolism/catecholaminergic

SNP

SERT (SLC6A4) MAOA

transporter/serotonergic

5HTTLPR

metabolism/catecholaminergic (primarily serotonergic) ligand/neuromodulatory ligand/neuromodulatory

VNTR

BDNF NPY

SNP SNP, CNV

Cognitive System

Personality Trait

Addiction

Refs.

R U U U

IC U – U

ER – U –

Imp Imp/Anx Imp

U U U

[30] [30,56] [55,57]

ACC, VS, IFG, Pre-SMA, Amy ACC, Amy, vmPFC

U

U

U

Imp/Anx

U

[33,36,81]

–

U

U

Imp/Anx

U

[78,79,81]

ACC, Amy, vlPFC, mPFC Amy Amy, VS, insula

U

U

U

Imp/Anx

U

[53,54]

– U

– –

U U

Anx Imp/Anx

U U

[84] [86]

R: reward; IC: inhibitory control; ER: emotional responses; VNTR: variable number of tandem repeats; SNP: single nucleotide polymorphism; 5HTTLPR: serotonin transporter linked polymorphic region; CNV: copy number variation; IFG: inferior frontal gyrus; VS: ventral striatum; ACC: anterior cingulated cortex; PMC: premotor cortex; Pre-SMA: presupplementary motor area; Amy: amygdala; mPFC: medial prefrontal cortex; vlPFC: ventrolateral prefrontal cortex; Imp: impulsivity; Anx: anxiety.

endophenotype approach and in particular imaging genetic approaches [8] have gained prominence. In a ﬁrst step, this strategy aims to link individual differences in personality traits and cognitive processes to variability in task-related brain function and/or structural differences. In a second step, relationships between brain function or structure and genetic markers, frequently selected for their known or putative function in biochemical signaling pathways, are investigated. In this review, we integrate recent evidence from imaging genetic studies of reward processes, inhibitory control and emotional reactivity with trends in a further research area, which is increasingly appreciated as relevant for the pathophysiology of addictions: brain development in adolescence. We explore current progress and limitations of imaging genetic studies and propose that integrating the imaging genetic strategy with biological network approaches and longitudinal adolescent designs in large multi-centre samples may offer promising opportunities to identify risk markers for early diagnosis, progression and prediction of addictions. Maturational factors conferring addiction risk Compared to both children and adults, adolescents commonly engage in more impulsive and risk-taking behaviour, are subject to intense emotional experiences, anxiety, and low self-esteem, and are sensitive to social reinforcement, such as acceptance/rejection from peers [9]. On the one hand, these cognitive, motivational and affective vulnerabilities have been suggested to be evolutionarily promoted [10]. On the other hand, they appear to reﬂect an imbalance in brain maturation between subcortical limbic structures (amygdala, nucleus accumbens/ventral striatum), which peak in early adolescence, and the more protracted, gradual development of prefrontal regions [11,12]. For example, impulsivity is thought to be a multidimensional trait, including ‘cold’ attentional as well as negative urgency-related components. The ‘opponent process hypothesis’ [13] suggests that impulsivity may stem from the combination of an overactive reward/ motivational system (the ventral striatum, and in particular nucleus

accumbens) and an underactive or as yet immature ‘topdown’ inhibitory control system (right lateral and inferior frontal regions), which in turn creates vulnerability for externalizing behaviour, drug use and addictions. In line with this hypothesis, several neuroimaging studies investigating the developmental trajectory of reward processing have shown that adolescents had higher ventral striatal activity during the anticipation [14] and receipt of rewards than adults [15]. However, lower ventral striatal activation during reward anticipation in adolescents compared to adults has also been found [16]. More consistent are studies showing that adolescents have greater difﬁculties in response inhibition than adults, and that this relates to reduced fronto-striatal connectivity and fronto-striatal under-maturation [17], as well as diminished control over drug use [18]. However, clearly, not all teenagers take drugs or exhibit these risk behaviours in equal measure. A large-scale longitudinal adolescent structural brain imaging study has shown that teenagers with higher impulsivity scores had slower rates of cortical thinning, predominantly in prefrontal regions [19]. Cortical thinning rates were in turn inﬂuenced by common polymorphisms in the dopamine D4 receptor gene (DRD4) and catechol-O-methyltransferase gene (COMT), which degrades dopamine and other monoaminergic neurotransmitters [20]. Together, this suggests that adolescence-speciﬁc maturational ‘growth curves’, rather than adult norms, need to be considered when assessing individual differences, and that some genes may exert their effects in a developmentallyspeciﬁc way. The following sections consider genetic and neurobiological factors in individual differences in each component of reinforcement behavior, with a focus on the currently available literature on adolescents (see also Table 1). Reward processing Individual differences in reward processing conferring addiction risk Virtually all addictive drugs affect dopamine activity in reward signalling. The brain reward system comprises the 437

Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Box 1. Commonly used neuroimaging tasks to test behavioural and fMRI BOLD responses during reward processing and inhibitory control Monetary Incentive Delay task [19] (Figure I) The MID task was designed to separately examine brain responses during anticipation and receipt of a monetary reward. The paradigm involves three parts: i) Anticipation: The volunteer sees a cue indicating what is at stake in this trial (for example, win, loss, magnitude of win). ii) Response: A target is presented and the participant needs to respond to it by button press within a time limit to gain a reward (or in some versions, to avoid losses). iii) Outcome: The participant is notified whether or not s/he obtained a reward on this trial, and the overall amount of points/ money won so far in the experiment is displayed. Different versions of the MID task differ in terms of manipulations of reward magnitudes, reward probabilities or whether rewards are framed in the context of ‘no-win’ trials or losses. Manipulation of duration of the target presentation determines how often the participant is successful. Using a tracking algorithm, in some versions duration of target presentation is adjusted based on individual performances to keep gains and losses approximately constant across participants. Striatal activation is sensitive to both reward probability and reward magnitude and the resulting expected value (probability magnitude), with greater activity in response to higher and more probable rewards.

Stop-Signal Reaction Time Task [33] and Go-No/Go Task [35] (Figure II) In the Stop Signal Reaction Time (SSRT) task, participants must respond to the go-signal as quickly as possible (GO trial). However, on a small proportion of trials (20%), a stop-signal is presented, which indicates that participants must cancel responding (STOP trial). Manipulation of the time (ms) between the onset of the go-signal and the onset of the stop-signal – the stop-signal delay (SSD) – changes the likelihood of the response inhibition to occur (response cancellation becomes more difficult with increasing SSD). In the Go/No-Go task, subjects must respond as quickly as possible to the target (go-signal) and withhold the response to the less frequent non-target (no-go-signal). In the SSRT, every trial starts off as a GO trial; if the STOP-signal occurs, the motor response must be cancelled during its execution. The Go/No-Go task contains an extra decision-making component since subjects have to choose between ‘go’ or ‘no-go’ before initiating the response.

[(Box_1)TD$FIG]

Go trial (Go/No-Go and SSRT)

[(Box_1)TD$FIG]

Cue

Delay

No-Go trial (Go/No-Go) Target

Outcome +$ [total $]

Stop trial (SSRT) Time SSD Reward magnitude

Reward probability

No win / loss

Low probability win

Small win

Small loss

Large win

Large loss

Response process

High probability win High probability loss TRENDS in Cognitive Sciences

Figure I. The Monetary Incentive Delay (MID) task.

ventral tegmental area, the nucleus accumbens (NAcc) in the ventral striatum and the mesial and orbital frontal cortices [21]. In addition to the ‘opponent process’ hypothesis discussed above, two other motivational theories of addiction vulnerability have been proposed. The ‘reward deﬁciency hypothesis’ suggests that the failure of non-drug rewards to activate the reward-motivational circuit predisposes risk for substance abuse [22]. This hypothesis was built on ﬁndings from positron emission tomography (PET) studies, which have linked decreased dopamine D2-receptor density in the striatum [23] in drug abusers to lower 438

Time

Low probability loss

Respond (Go) or Withhold (No-Go) response

Stop responding

(SSRT)

TRENDS in Cognitive Sciences

Figure II. The Stop-Signal Reaction Time and Go-No/Go tasks.

sensitivity to natural reinforcers. Although impairments seen in cross-sectional case-control studies do not allow us to disentangle preceding vulnerability factors from the consequences of long-term drug abuse, the potential causal role was suggested by persistently lower DA-receptor availability after prolonged drug-abstinence [24]. Moreover, in rodents selectively bred for high trait-impulsivity, reduced D2/3 receptor availability in the nucleus accumbens before drug exposure has recently been shown to predict cocaine self-administration [25]. The ‘incentive salience’ hypothesis proposes that a critical factor in the transition from drug use to abuse is

Review heightened motivational salience speciﬁcally of drug-related cues [26]. This hypothesis, however, makes no speciﬁc predictions for individual differences in reward function before the onset of drug use. Functional magnetic resonance imaging (fMRI) studies investigating the link between reward-related brain function and personality risk-traits and case-control studies with addicted groups have produced mixed results. A commonly used neuroimaging task taken to separate reward anticipation (i.e., motivational sensitivity to rewardrelated cues) and reward outcome is the Monetary Incentive Delay (MID) task [27] (see Box 1). On the one hand, two adolescent studies reported an association between higher ventral striatal activity during reward anticipation or reward outcome and higher sensation seeking scores [28], and externalizing symptoms [29]. In adults, higher ventral striatal activity during probabilistic guessing games was associated with higher trait impulsivity [30] and steeper delay discounting, an index of greater reward impulsivity [31]. This, as well as the ﬁnding of higher ventral striatal sensitivity to reward and loss deliveries in substance dependent patients [32], would be consistent with the ‘opponent process hypothesis’. On the other hand, a recent large study of 266 healthy 13-14 year old adolescents (Schneider et al., unpublished data) and an adult study [33] found that lower ventral striatal activity during reward anticipation was related to higher risk-taking and sensation seeking scores, respectively. Reduced ventral striatal activation during reward anticipation in the MID task was also observed in detoxiﬁed alcoholics [34], which together would support the reward deﬁciency hypothesis. Several factors may contribute to these inconsistencies, including often small sample sizes, differences in participant characteristics, and, in studies with addicted groups, differences in severity, duration, nature of abused substances, and comorbidities. The impact of even apparently subtle differences between neuroimaging tasks, including design, acquisition parameters, and task context, is also increasingly recognized (see Box 1). Moreover, no study has yet prospectively investigated in humans the effect of differential activation of the reward system on development of substance abuse and addictions. Candidate gene differences in reward processing Recent imaging genetic studies of reward sensitivity have focused on common polymorphisms encoding different stages of dopamine neurotransmission. Dopamine activity is inﬂuenced by dopamine synthesis and release, re-uptake, receptor activity and metabolism, as well as modulating signals of other neurotransmitter systems, including the opioidergic system, brain-derived neurotrophic factor (BDNF), serotonin, gamma-aminobutyric acid (GABA), acetylcholine and endocannabinoids [35]. Adult fMRI studies using probabilistic reward-related guessing tasks [30,33] and an MID task [36] reported increased ventral striatal activity associated with a DAT1 9-repeat Variable Number Tandem Repeat (VNTR), which is linked to decreased synaptic clearance of dopamine. These studies also found increased striatal activity to be associated with a DRD4 VNTR genotype linked to

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

decreased postsynaptic inhibition and a DRD2 141C deletion linked to reduced expression of the inhibitory DRD2 receptors. In addition, higher prefrontal cortex activity during reward anticipation in Met/Met carriers of the catechol-O-methyltransferase (COMT) Val/Met-polymorphism was linked to higher tonic DA-levels [33,36]. Together, these single gene results appear to indicate a positive correlation of dopaminergic activity and ventral striatal activation. However, when gene gene interactions of DAT and COMT on ventral striatal activation during reward anticipation were studied, an epistatic effect was observed. An inverted U-shaped relationship between genotype combinations and ventral striatal BOLD response showed that intermediate dopaminergic tone was associated with maximal ventral striatal activation [33]. Genotype groups with lower striatal reward encoding also had the highest sensation seeking scores in adults [33]. In sum, imaging genetic studies showed effects of different genetic polymorphisms inﬂuencing dopamine neurotransmitter systems on fronto-striatal activity during reward processing. However, considerable inconsistencies between studies in terms of gene-brain and brain-behaviour relationships remain. Although it is possible that both hypo- and hyper-sensitivity to reward cues or outcomes may provide alternative routes to addiction vulnerability, the underlying mechanisms, including the potential inﬂuence of genetic factors on functional and behavioural inconsistencies, remain to be better understood. Furthermore, the combination of hyper-sensitivity to rewards and reduced inhibitory control in addiction vulnerability, as predicted by the ‘opponent-process hypothesis’, and the relationship between functional responses to drug-related vs. drug-unrelated reward cues, as predicted by the ‘incentive salience hypothesis’, are yet to be tested. Findings of gene gene interactions indicate that the genetic inﬂuences governing dopaminergic activity are complex and suggest that systems-based approaches, such as the construction of gene co-expression networks [37], which also account for modulating neurotransmitter systems, may be a useful tool to attain more comprehensive genetic risk proﬁles in reward-related brain function and cognitive processes. Inhibitory control Individual differences in inhibitory control conferring addiction risk Inhibitory control includes the ability to deliberately stop a pre-potent response in a given situation due to rule shifting, or to suppress actions that are risky, inappropriate, or no longer required. One relevant facet of inhibitory control – action inhibition – is commonly tested using rapid stimulusresponse paradigms, such as the Go/No-Go and Stop Signal Reaction Time tasks [38,39] (see Box 1). The ‘stop circuit’ relies on the right inferior frontal gyrus (rIFG), anterior cingulate cortex (ACC), pre-supplementary motor cortex (pre-SMA), and ventral and dorsal regions of the striatum [40]. Poor performance on fast information processing tasks has been related to increased difﬁculty in action planning as well as drug taking [41]. These behaviours with longer time courses rely on intact top-down control to select appropriate responses over prepotent ones. 439

Review Neuropharmacological studies have established a central role of monoamine transmission in the two major dissociable forms of action inhibition: ‘action restraint’ (Go/NoGo) and ‘action cancellation’ (SSRT) [42]. The dopaminergic mesolimbic projection to the VS is directly modulated by serotonin and indirectly by noradrenergic inputs (possibly via the PFC), both facilitating dopamine release [43]. Dopamine has been implicated in right inferior frontal gyrus activation during both types of action inhibition, whereas serotonin seems to be predominant in ‘action restraint’ where a higher-level, cortical decision-making process is involved [44,45]. This process appears to involve the orbitofrontal cortex (OFC), which has been implicated as the locus for 5-HT action in Go/No-Go task [46]. Animal and human studies have linked aggressive behaviour to low serotonin levels and decreased ‘action restraint’ (Go/No-Go) [47], while no link has been found between aggression and poor performance on the SSRT [48]. These observations suggest that performance on the Go/No-Go task might be more predictive of antisocial and criminal behaviour than performance on the SSRT. Candidate gene differences in inhibitory control In a human genetic study investigating criminals, genetic polymorphisms in serotonin receptor 2B (HTR2B) and mono-amino-oxidase A (MAOA), which is involved in the metabolism of serotonin and other monoaminergic neurotransmitters, have been associated with severe impulsivity as well as vulnerability to alcoholism [49]. Polymorphisms in dopaminergic genes, including VNTR polymorphisms in DRD4 and DAT1, have been associated with poor ‘action restraint’ and ‘action cancellation’ [50,51], and have been implicated in alcohol abuse in adolescence [52]. Imaging genetic studies have revealed dissociable effects of polymorphisms in monoaminergic genes on the response of neural circuits underlying action inhibition. Genetic variants in both dopaminergic and serotonergic systems were shown to inﬂuence brain activation during a Go/No-Go task. For instance, the low MAOA VNTR expression variant is associated with diminished activation in the dorsal ACC [53] and right ventrolateral prefrontal cortex and is correlated with impulsive behaviour [54]. Also in a Go/No-Go task, DRD4 and DAT1 VNTR polymorphisms, as well as a DRD2 SNP, were associated with increased activation in the IFG, striatum and cingulate cortex in adult heavy drinkers and adolescents with ADHD [55,56]. fMRI studies using SSRT tasks showed association between dopaminergic gene polymorphisms including DAT1 9-repeat and COMT Met allele and greater activation in the right IFG and pre-SMA [57]. While no geneneuroimaging data associating polymorphisms in serotonergic genes and brain activation during SSRT have been published, gene-behaviour studies found that the 5HTTLPR polymorphism had no effect on SSRT, even under acute tryptophan depletion [58]. Taken together, these results suggest that ‘action restraint’ may be more sensitive to variation in serotonergic function than ‘action cancellation’. The imaging genetic ﬁndings reviewed above demonstrate that the effect of genetic variants in the monoaminergic system on deﬁcits in action inhibition is modulated by 440

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

fronto-striatal activity. Dysfunctional fronto-striatal networks, as observed in adolescents, have been proposed as vulnerability markers for addictions and conduct disorder [59]. To test whether gene neuroimaging proﬁles of different forms of inhibitory control can be useful predictors of speciﬁc behavioural problems (for example addictive behaviour vs. conduct disorder), longitudinal studies that recognize the different neuropsychological components of behavioural inhibition in standardised functional neuroimaging paradigms will be needed. Stress responses Individual differences in stress response conferring addiction risk Experience of severe or chronic life stress is another wellestablished risk factor for addictions. Stress increases vulnerability to the negative reinforcement properties of drug and alcohol use, both in adolescents and in relapsing addicts [60]. Moreover, several studies have shown striking individual variability in behavioural and physiological responses to stressful experiences [61]. The development of negative emotional states following trauma and chronic stress (as well as prolonged alcohol abuse) has been proposed to include the recruitment and subsequent deregulation of brain stress systems, such as the hypothalamic–pituitary–adrenal (HPA) axis, extrahypothalamic corticotrophin releasing factor (CRF), noradrenaline, neuropeptide Y, tachykinins and the dynorphins amongst others [62]. Activation of the HPA-axis triggers a cascade of neuroendocrine factors, including CRF and adrenocorticotropic hormone (ACTH), and results in secretion of cortisol. Cortisol binds to glucocorticoid receptors (GR), which are expressed in both brain and peripheral tissue. GRs form homodimers and bind to glucocorticoid response elements (GREs) in multiple target genes, including serotonergic genes, thus regulating gene transcription [63]. As part of its extrahypothalamic activity, CRF modulates neuronal activity in the basolateral amgydala, a region involved in the processing of fear and threat [64]. Neuroimaging studies have consistently shown that the processing of threat/stress related stimuli, as well as fear conditioning, mainly rely on the amygdalaprefrontal circuitry. Adolescents [65] and adults [66] with high state/trait anxiety and anxiety disorders typically show heightened amygdala responses to threat-related stimuli and aberrant amygdala–prefrontal coupling. Candidate gene differences in stress response The striking individual variability in response to stress [61] has generated interest in the study of risk-resilience factors, including the effect of genes that modulate the impact of stress on addictive behaviour. Associations, including gene environment interactions have been identiﬁed for genes, which encode crucial components of stress response systems, including CRHR1 [67,68], the ACTH– precursor POMC [69], the GC–receptor gene NR3C1 [70], as well as the norepinephrine transporter gene SLC6A2 and the adrenergic receptor gene ADRA2A [71]. Interestingly, evidence for gene environment interactions of risky alcohol drinking in adolescence with genes linking stress response to dopaminergic signalling, including the

Review potassium channel gene KCNJ6 [72] and the circadian rhythm gene Period1 [73] is also emerging. Despite the central role of the HPA axis, CRF and noradrenergic systems in stress regulation, imaging genetic studies of threat-related emotional reactivity have predominantly examined a repeat-length polymorphism (5HTTLPR) of the serotonin transporter (5-HTT) gene (SLC6A4). 5-HTT regulates synaptic serotonin [74] and is sensitive to stress due to a glucocorticoid response element (GRE) in the promoter region. Numerous epidemiological and clinical studies have demonstrated a moderating effect of 5-HTTLPR on the relationship of stress and depression [75,76]. Adolescents carrying the short allele (S-carriers), which is associated with lower gene expression and higher synaptic serotonin concentration, are also more prone to engage in earlier alcohol [77] use after having experienced negative life events. Imaging genetic studies have consistently shown that Scarriers have signiﬁcantly higher amygdala responses to fearful faces [78,79] compared to L-carriers. Aberrant amgydala-prefrontal coupling [79] and reduced amygdala structural volumes [80] were also associated with the Sallele. When gene gene interactions of 5-HTTLPR and COMT genotypes were analysed, an additive effect was observed, which explained about 40% of individual variance in amygdala response to aversive stimuli. This amounts to a tenfold increase in effect size compared to 4% variability by 5-HTTLPR effects on anxiety [81]. Together, these ﬁndings are amongst the strongest support for the endophenotype strategy and suggest that 5HTTLPR and COMT genotype effects on amygdala-prefrontal dysfunctions might be tested in longitudinal studies for their suitability as a pre-clinical vulnerability marker for anxiety disorders, and possibly addictions. Recently, genetic variants affecting the brain-derived neurotrophic factor (BDNF), which mediates long-term neural plasticity in response to adverse social experiences, have been identiﬁed [82]. The Met allele of the BDNF Val66Met polymorphism is associated with decreased secretion of BDNF, higher anxiety and cortisol levels in response to stress, and greater alcohol consumption in healthy adults [83]. A recent study showed a diagnosis (anxiety or major depressive disorder vs. control) genotype interaction, suggesting that, only in adolescents with anxiety or depression, the Met allele was associated with higher amygdala-hippocampal responses to emotional stimuli than the Val/Val genotype [84]. BDNF in turn increases the expression of NPY [85]. Haplotype-driven NPY expression predicted higher threat-related amygdala activity in low-expression genotypes, lower resilience to pain/ stress-induced activation of endogeneous opioid neurotransmission, and higher trait anxiety [86]. In sum, until recently, imaging genetic studies of stressrelated emotional responses concentrated on relatively few well-characterised polymorphisms affecting monoaminergic, in particular serotonergic neurotransmission. While these studies yielded robust ﬁndings indicating the relevance of serotonergic genes for fronto-limbic function, they proceeded largely separately from neurobiological research on stress response. More recent imaging genetic analyses

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

of NPY and BDNF variants are tapping into the potential provided by state of the art research on emotional reactivity. This neurobiologically informed approach, together with the consideration of environmental factors has great potential in further delineating risk-resilience mechanisms for early onset drug or alcohol use in adolescents exposed to adverse experiences. Social reinforcement Developmental and individual differences in social reinforcement conferring addiction risk Peer group characteristics and dynamics (for example, the desire to be accepted, fear of rejection) constitute a further risk factor for drug–use behaviour during adolescence [87]. A study that divided adolescents into groups with high vs. low vulnerability to peer inﬂuence showed that resistance to peer pressure was associated with stronger functional connectivity between prefrontal and pre-motor regions involved in decision-making and cognitive control when watching video clips of angry hand and face expressions [88]. Another important factor in peer group interactions is sensitivity to social reinforcement, which involves linking social signals (for instance, smile, anger) to reward and reinforcement processes [89]. For example, recent neuroimaging studies have shown that the presence of peers increases risk-taking behavior and ventral striatal activity during reward anticipation in adolescents (but not adults) [90] and that anticipated social evaluation from peers triggers a similar functional response as threat-cues (increased amygdala and PFC activity) in adolescents with social anxiety disorder [91]. The nanopeptides oxytocin and vasopressin have been implicated in the regulation of stress physiology and the development of social attachment and afﬁliation. Vasopressin V1b receptors have also been shown to mediate the transition to excessive drinking in alcohol-dependent rats [92]. Several common polymorphisms in the oxytocin receptor gene (OXTR) have been shown to modulate anxiety, empathy, and trust. One mechanism by which OXTR appears to increase social behaviour is by reducing fear-related amygdala reactivity [93]. However, comparative research also suggests a more direct effect of OXTR on sensitivity in brain reward regions, in particular social reinforcement [89]. In future work, gene environment interactions in sensitivity to social reinforcement may deserve closer consideration, as the nature of peer groups and family environment can both initiate and/or exacerbate alcohol or drug use, or act as protective factors. Concluding remarks In this paper we review imaging genetic studies, which investigate individual variability in positive and negative reinforcement, with a view towards current progress and future avenues for the identiﬁcation of predictive biological markers for addictions. To date, imaging genetic studies have mainly analysed the association of single regions of interest with few, wellvalidated genetic variations in selected genes. They have aimed to explore and evaluate the functional impact of brain-relevant genetic polymorphisms with the potential 441

(Figure_1)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

(a) ROI 1 SNP1

Trait 1

Addiction

Trait 2

Addiction

Trait 3

Addiction

ROI 2

ROI 2 SNP2 ROI 3

ROI 4 SNP3 ROI 2

(b)

Trait 1 Trait 3

Addiction

Trait 2

TRENDS in Cognitive Sciences

Figure 1. The candidate gene-neuroimaging and the systems-level gene-neuroimaging approaches. (a) Candidate gene-neuroimaging studies typically select one welldescribed candidate SNP (for example, DAT) and correlate it with task-related fMRI BOLD response in one or a few selected regions of interest (ROIs, for example, ventral striatum, orbital frontal cortex). Individual variability in fMRI BOLD responses is in turn mapped to dimensional personality trait measures (for example, impulsivity) thought to relate to the clinical phenotype (for example, addiction). Although most brain regions participate in different cognitive processes (for example, PFC is involved in reward processing, inhibitory control, attentional control of emotions, among many others), and different cognitive processes may be implicated in one personality trait (for example, both reward sensitivity and inhibitory control are thought to be components of impulsivity), studies typically only examine one construct at a time. Whole-brain voxel-wise regression analyses with one candidate SNP (not illustrated), which do not require a priori predictions of genetic effects on activity in a particular brain region, are also relatively common. Conversely, genome-wide analyses of individual ROIs allow the discovery of ‘new’ genes associated with variability in brain-structure or function. Recently developed voxel-wise genome-wide association methods have overcome computational and multiple comparison problems posed by the integration of high-dimensional data sets [94]. However, each of these approaches focuses on the identification of links between isolated components on the gene and brain levels, and does not consider interactions and resulting emerging properties between elements of a complex system. (b) The aim of a systems-level gene-neuroimaging approach is to map dynamic networks of brain function, structure and genetics. Several multivariate statistical methods, including ‘seed-based’ effective connectivity models, such as Dynamic Causal Modelling or Granger Causality, or data-driven clustering techniques, such as principal component analysis, partial least squares (PLS) or independent component analysis (ICA), have been developed to examine functional or structural connectivity underlying particular cognitive systems. For example, PLS or ICA can be used to simultaneously analyse brain images and genetic information to uncover hidden cross-information from a larger number of genes and to relate this to functionally connected activation patterns [95]. Multi-modal imaging (relating, for example, structural and functional connectivity to one another or EEG/ MEG and/ or PET data) aims to increase sensitivity and specificity of the neuroimaging ‘endophenotype’ [99], Graph-theoretical approaches define networks as a set of nodes (for example, brain regions or genes) and edges or lines (connections) between them. The description of topological properties of complex networks, such as modularity, centrality, small-worldness, and the distribution of network ‘‘hubs’’, has been applied to functional (using primarily ‘resting-state’ data), structural brain networks [96] and gene-expression networks [37] to quantify differences in health and disease. The integration of these different levels of analysis may significantly contribute to identifying disease mechanisms and pathways [98].

to understand their impact on behaviour [8]; however, they have not been designed to reﬂect the complexity of neurobiological mechanisms underlying reinforcement-related behaviour (see Figure 1a). Given the diversity of entry 442

mechanisms into addictions and the heterogeneity of addictions themselves, the identiﬁcation of stratiﬁcation markers might be contingent on the identiﬁcation of speciﬁc neurobiological clusters. These cannot be described on

(Figure_2)TD$IG][ Review

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9

Addictions

Drug use behaviour

Inhibitory control Reward Emotion

H2

H1

H3

nt

nme

ro Envi

13 yrs

14 yrs

15yrs

16 yrs

17 yrs

18 yrs

19 yrs

20 yrs

TRENDS in Cognitive Sciences

Figure 2. Integrating systems biology and developmental approaches to test the predictive value of genetic and neurobiological markers for early onset drug-use and transition to abuse. A better understanding of genetic-neurobiological risk profiles of early-onset drug use might be a precondition for studying the mechanisms by which early-onset drug use increases risk for the transition to addictions. Longitudinal prospective adolescent designs enable testing, for example, whether (i) some risk factors for early-onset drug use and for addiction vulnerability are shared; (ii) examining brain development drug-consumption interactions (i.e., whether drugs have a differential effect on the premature adolescent vs. more mature adult brain); and (iii) testing the interaction between predispositional factors (for example, neurobiological clusters implicated in impulsivity or chronic stress) and the effect that drugs exert on brain development and function at different developmental stages. This design also allows identification (in an exploratory manner) and testing (in a hypothesis-driven manner) of potential ‘protective genetic and/or environmental factors’ for high-risk groups (for example, high impulsive adolescents who do not take drugs), such as genes involved in the metabolism of specific substances (for example, ALDH*2), or family and social environment.

the basis of few polymorphisms and the analysis of isolated regions of interest alone, or the opposite approach, voxelwise genome-wide analyses [94]. Instead, this might be achieved by pursuing a systems approach, which reﬂects the interrelatedness of genetic function, neurophysiological activation and behaviour (see Figure 1b). A number of recent methodological advances in neuroimaging research provide novel tools to examine individual differences in structural and functional connectivity pat-

Box 2. Questions for future research To what extent are supra-regional functional connectivity patterns across tasks and domains domains or between brain structure and function more sensitive and/or specific for addiction-related personality traits than analyses with single regions within any one domain? How can we currently address the need for standardization of neuroimaging tasks required for comparison between studies and large-scale gene-neuroimaging studies with the concurrent need for detailed comparisons of genetic effects on brain function between different tasks or parameters in a cost-efficient manner? How do different drug-use risk profiles interact with acute or long term effects of drug intake at different developmental stages in modulating vulnerability for drug dependence and addiction?

terns [95,96]. Imaging genetic studies might also beneﬁt from genetic approaches informed by co-expression network analyses and GWAS-based gene enrichment studies [97,98]. To test the potential predictive value of multimodal risk proﬁles drawing upon neuroimaging, neuropsychological, behavioural and genomic information, prospective longitudinal adolescent studies are required, which compare pre-dispositional and environmental risk factors before the onset of signiﬁcant drug use in the context of ‘normal’ brain maturational changes with patterns of early-onset alcohol or drug use behaviour (see Figures 1b and 2, and Box 2). As proof-of-concept, in psychiatric research, the combination of genetics and longitudinal neuroimaging used to identify early prospective biological markers is currently most advanced in relation to Alzheimer’s disease. For example, the combination of brain atrophy and cortical thickness changes, as measured by structural MRI, hypometabolism quantiﬁcation as assessed with ﬂuorodoxyglucose-PET, and detection of amyloid beta in cerebrospinal ﬂuid achieved a classiﬁcation accuracy of over 93% [99]. Small sample sizes, which limit the number of multiple comparisons in each individual study, as well as the use of non-standardised tasks, which render comparisons across studies, including meta-analyses, difﬁcult have precluded 443

Review a systems-based gene neuroimaging approach in the past. However, the creation of open-access data repositories to increase data sharing and pooling across different centres, such as the 1000 Functional Connectomes Project [100] or the International Neuroinformatics Coordinating Facility (INCF), on-line data bases, tools, and atlases, such as the ﬁrst whole human brain gene expression map currently developed at the Allen Brain Institute, as well as several multi-disciplinary multi-centre consortia respond to these limitations. For example, the European multi-centre study IMAGEN is a gene-neuroimaging study of 2000 14-year-old adolescents, with follow-up measures at age 16 and 18/19 years. The aim is to identify the genetic and neurobiological basis of individual variability in reward sensitivity, inhibitory control and emotional reactivity, and to determine their predictive value for the development of common psychiatric disorders [101], including addictions. Analysing this and other emerging large gene-neuroimaging datasets holds the promise to identify markers for neurobiological clusters, which represent speciﬁc pathological processes for targeted treatment. Acknowledgements This work was supported by the European Union-funded FP6 Integrated Project IMAGEN (Reinforcement-related behavior in normal brain function and psychopathology) (LSHM-CT-2007-037286), the FP7 project ADAMS (Genomic variations underlying common neuropsychiatric diseases and disease related cognitive traits in different human populations) (242257) and the FP7 IMI-Project ‘‘AIMS’’, as well as the United Kingdom National Institute for Health Research (NIHR) Biomedical Research Centre Mental Health, the Medical Research Council Programme Grant ‘‘Developmental pathways into adolescent substance abuse’’ (93558).

References 1 Goldman, D. et al. (2005) The genetics of addictions: uncovering the genes. Nat. Rev. Genet. 6, 521–532 2 Chambers, R.A. et al. (2003) Developmental neurocircuitry of motivation in adolescence: a critical period of addiction vulnerability. Am. J. Psychiatry 160, 1041–1052 3 Taioli, E. and Wynder, E.L. (1991) Effect of the age at which smoking begins on frequency of smoking in adulthood. N. Engl. J. Med. 325, 968–969 4 O’Brien, M.S. and Anthony, J.C. (2005) Risk of becoming cocaine dependent: epidemiological estimates for the United States, 2000– 2001. Neuropsychopharmacology 30, 1006–1018 5 Goodman, A. (2008) Neurobiology of addiction: An integrative review. Biochemical. Pharmacol. 75, 266–322 6 Caspi, A. et al. (1996) Behavioral observations at age 3 years predict adult psychiatric disorders: Longitudinal evidence from a birth cohort. Arch. Gen. Psychiatry 53, 1033–1039 7 Hamer, D. (2002) Rethinking behavior genetics. Science 298, 71 8 Meyer-Lindenberg, A. and Weinberger, D.R. (2006) Intermediate phenotypes and genetic mechanisms of psychiatric disorders. Nat. Rev. Neurosci. 7, 818–827 9 Somerville, L.H. et al. (2010) A time of change: Behavioural and neural correlates of adolescent sensitivity to appetitive and aversive environmental cues. Brain Cogn. 72, 124–133 10 Barkley, R.A. (2001) The executive functions and self-regulation: an evolutionary neuropsychological perspective. Neuropsychol. Rev. 11, 1–29 11 Paus, T. et al. (2008) Why do many psychiatric disorders emerge during adolescence? Nat. Rev. Neurosci. 9, 947–957 12 Giedd, J.N. et al. (2010) Anatomic Magnetic Resonance Imaging of the Developing Child and Adolescent Brain and Effects of Genetic Variation. Neuropsychol. Rev. 20, 349–361 13 Bechara, A. (2005) Decision making, impulse control and loss of willpower to resist drugs: a neurocognitive perspective. Nat. Neurosci. 8, 1458–1463 444

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 14 Van Leijenhorst, L. et al. (2010) What motivates the adolescent? Brain regions mediating reward sensitivity across adolescence. Cerebr. Cortex 20, 61–69 15 Galvan, A. et al. (2006) Earlier development of the accumbens relative to orbitofrontal cortex might underlie risk-taking behavior in adolescents. J. Neurosci. 26, 6885–6892 16 Bjork, J.M. et al. (2010) Adolescents, adults and rewards: comparing motivational neurocircuitry recruitment using fMRI. PloS ONE 5, e11440 17 Rubia, K. et al. (2006) Progressive increase of frontostriatal brain activation from childhood to adulthood during event-related tasks of cognitive control. Hum. Brain Mapp. 27, 973–993 18 Nigg, J.T. et al. (2006) Poor response inhibition as a predictor of problem drinking and illicit drug use in adolescents at risk for alcoholism and other substance use disorders. J. Am. Acad. Child Adolesc. Psychiatry 45, 468–475 19 Shaw, P. et al. (2011) Cortical development in typically developing children with symptoms of hyperactivity and impulsivity: support for a dimensional view of attention deﬁcit hyperactivity disorder. Am. J. Psychiatry 168, 143 20 Shaw, P. et al. (2007) Polymorphisms of the dopamine D4 receptor, clinical outcome, and cortical structure in attention-deﬁcit/ hyperactivity disorder. Arch. Gen. Psychiatry 64, 921–931 21 Schultz, W. (2011) Potential Vulnerabilities of Neuronal Reward, Risk, and Decision Mechanisms to Addictive Drugs. Neuron 69, 603–617 22 Robbins, T.W. and Everitt, B.J. (1999) Drug addiction: bad habits add up. Nature 398, 567–570 23 Volkow, N.D. et al. (1997) Decreased striatal dopaminergic responsiveness in detoxiﬁed cocaine-dependent subjects. Nature 386, 830–833 24 Volkow, N.D. et al. (2002) Effects of alcohol detoxiﬁcation on dopamine D2 receptors in alcoholics: a preliminary study. Psychiatry Res. 116, 163–172 25 Dalley, J.W. et al. (2007) Nucleus accumens D2/3 receptors predict trait impulsivity and cocaine reinforcement. Science 315, 1267–1270 26 Robinson, T.E. and Berridge, K.C. (2001) Incentive-sensitization and addiction. Addiction 2001, 96 27 Knutson, B. et al. (2001) Anticipation of increasing monetary reward selectively recruits nucleus accumbens. J. Neurosci. 21, 1–5 28 Bjork, J.M. et al. (2008) Incentive elicited striatal activation in adolescent children of alcoholics. Addiction 103, 1308–1319 29 Bjork, J.M. et al. (2010) Incentive-elicited mesolimbic activation and externalizing symptomatology in adolescents. J. Child Psychol. Psychiatry 51, 827–837 30 Forbes, E.E. et al. (2009) Genetic variation in components of dopamine neurotransmission impacts ventral striatal reactivity associated with impulsivity. Mol. Psychiatry 14, 60–70 31 Hariri, A.R. et al. (2006) Preference for immediate over delayed rewards is associated with magnitude of ventral striatal activity. J. Neurosci. 26, 13213–13217 32 Bjork, J.M., Smith, A.R. and Hommer, D.W. (2008) Striatal sensitivity to reward deliveries and omissions in substance dependent patients. Neuoimage 42, 1609–1621 33 Yacubian, J. et al. (2007) Gene-gene interaction associated with neural reward sensitivity. Proc. Natl. Acad. Sci. U.S.A. 104, 8125– 8130 34 Wrase, J. et al. (2007) Dysfunction of reward processing correlates with alcohol craving in detoxiﬁed alcoholics. Neuroimage 35, 787–794 35 Koob, G.F. and Volkow, N.D. (2009) Neurocircuitry of addiction. Neuropsychopharmacology 35, 217–238 36 Dreher, J.C. et al. (2009) Variation in dopamine genes inﬂuences responsivity of the human reward system. Proc. Natl. Acad. Sci. U.S.A. 106, 617–622 37 Schadt, E.E. (2009) Molecular networks as sensors and drivers of common human diseases. Nature 461, 218–223 38 Chambers, C.D. et al. (2006) Executive ‘‘brake failure’’ following deactivation of human frontal lobe. J. Cogn. Neurosci. 18, 444– 455 39 Menon, V. et al. (2001) Error-related brain activation during a Go/ NoGo response inhibition task. Hum. Brain Mapp. 12, 131–143 40 Aron, A.R. et al. (2007) Triangulating a cognitive control network using diffusion-weighted magnetic resonance imaging (MRI) and functional MRI. J. Neurosci. 27, 3743–3752

Review 41 Li, C.R. et al. (2009) Altered impulse control in alcohol dependence: neural measures of stop signal performance. Alcohol. Clin. Exp. Res. 33, 740–750 42 Schachar, R. et al. (2007) Restraint and cancellation: multiple inhibition deﬁcits in attention deﬁcit hyperactivity disorder. J. Abnorm. Child Psychol. 35, 229–238 43 Horne, M. et al. (2008) Long-term administration of cocaine or serotonin reuptake inhibitors results in anatomical ad neurochemical changes in noradrenergic, dopaminergic, and serotonin pathways. J. Neurochem. 106, 1731–1744 44 Eagle, D.M. et al. (2008) The neuropsychopharmacology of action inhibition: cross-species translation of the stop-signal and go/no-go tasks. Psychopharmacology (Berl.) 199, 439–456 45 Dalley, J.W. et al. (2011) Impulsivity, compulsivity, and top-down cognitive control. Neuron 69, 680–694 46 Rubia, K. et al. (2005) Abnormal brain activation during inhibition and error detection in medication-naive adolescents with ADHD. Am. J. Psychiatry 162, 1067–1075 47 Leyton, M. et al. (2001) Brain regional a-[11C] Methyl-L-Tryptophan trapping in impulsive subjects with borderline personality disorder. Am. J. Psychiatry 158, 775 48 Van den Bergh, F. et al. (2006) Relationship of delay aversion and response inhibition to extinction learning, aggression, and sexual behaviour. Behav. Brain Res. 175, 75–81 49 Bevilacqua, L. et al. (2010) A population-speciﬁc HTR2B stop codon predisposes to severe impulsivity. Nature 468, 1061–1066 50 Munafo`, M. et al. (2008) Association of the dopamine D4 receptor (DRD4) gene and approach-related personality traits: meta-analysis and new data. Biol. Psychiatry 63, 197–206 51 Congdon, E. et al. (2008) Analysis of DRD4 and DAT polymorphisms and behavioral inhibition in healthy adults: implications for impulsivity. Am. J. Med. Genet. B Neuropsychiatr. Genet. 147B, 27–32 (in eng) 52 Laucht, M. et al. (2007) Novelty seeking involved in mediating the association between the dopamine D4 receptor gene exon III polymorphism and heavy drinking in male adolescents: results from a high-risk community sample. Biol. Psychiatry 61, 87–92 53 Meyer-Lindenberg, A. et al. (2006) Neural mechanisms of genetic risk for impulsivity and violence in humans. Proc. Natl. Acad. Sci. U.S.A. 103, 6269–6274 54 Passamonti, L. et al. (2006) Monoamine oxidase-a genetic variations inﬂuence brain activity associated with inhibitory control: new insight into the neural correlates of impulsivity. Biol. Psychiatry 59, 334–340 55 Braet, W. et al. (2011) fMRI activation during response inhibition and error processing: the role of the DAT1 gene in typically developing adolescents and those diagnosed with ADHD. Neuropsychologia 49, 1641–1650 56 Filbey, F.M. et al. (2011) Dopaminergic genes modulate response inhibition in alcohol abusing adults. Addiction Biol. DOI: 10.1111/ j.1369-1600.2011.00328 57 Congdon, E. et al. (2009) Inﬂuence of SLC6A3 and COMT variation on neural activation during response inhibition. Biological. Psychol. 81, 144–152 58 Clark, L. et al. (2005) Stop signal response inhibition is not modulated by tryptophan depletion or the serotonin transporter polymorphism in healthy volunteers: implications for the 5-HT theory of impulsivity. Psychopharmacology 182, 570–578 59 Sterzer, P. and Stadler, C. (2009) Neuroimaging of aggressive and violent behaviour in children and adolescents. Front Behav. Neurosci. 3, 35 60 Pilowsky, D.J. et al. (2009) Adverse childhood events and lifetime alcohol dependence. Am. J. Pub. Health 99, 258 61 Rutter M (2007) Gene–environment interdependence. Dev. Sci. 10, 12–18 62 Koob, G.F. (2008) A role for brain stress systems in addiction. Neuron 59, 11–34 63 Karst, H. et al. (2000) Corticosteroid actions in hippocampus require DNA binding of glucocorticoid receptor homodimers. Nat. Neurosci. 3, 977–978 64 LeDoux, J. (2003) The emotional brain, fear, and the amygdala. Cell Mol. Neurobiol. 23, 727–738 65 Beesdo, K. et al. (2009) Common and distinct amygdala-function perturbations in depressed vs anxious adolescents. Arch. Gen. Psychiatry 66, 275–285 (in eng)

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 66 Bishop, S. et al. (2004) Prefrontal cortical function and anxiety: controlling attention to threat-related stimuli. Nat. Neurosci. 7, 184–188 67 Treutlein, J. et al. (2006) Genetic association of the human corticotropin releasing hormone receptor 1 (CRHR1) with binge drinking and alcohol intake patterns in two independent samples. Mol. Psychiatry 11, 594–602 68 Blomeyer, D. et al. (2008) Interaction between CRHR1 gene and stressful life events predicts adolescent heavy alcohol use. Biol. Psychiatry 63, 146–151 69 Zhang, H. et al. (2009) Pro-opiomelanocortin gene variation related to alcohol or drug dependence: evidence and replications across family-and population-based studies. Biol. Psychiatry 66, 128–136 70 Desrivie`res, S. et al. (2011) Glucocorticoid receptor (NR3C1) gene polymorphisms and onset of alcohol abuse in adolescents. Addict. Biol. 16, 510–513 71 Clarke, T.K. et al. (2010) Multiple polymorphisms in genes of the adrenergic stress system confer vulnerability to alcohol abuse. Addict. Biol. DOI: 10.1111/j.1369-1600.2010.00263 72 Clarke, T.K. et al. (2011) KCNJ6 is associated with adult alcohol dependence and involved in gene early life stress interactions in adolescent alcohol drinking. Neuropsychopharmacology 36, 1142– 1148 73 Dong, L, et al. (2011) The circadian rhythm gene Period1 is associated with psychosocial stress-induced alcohol drinking Am. J. Psychiatry (in press) 74 Lesch, K.P. et al. (1996) Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science 274, 1527–1531 75 Caspi, A. et al. (2003) Inﬂuence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 301, 386–389 76 Karg, K. et al. (2011) The serotonin transporter promoter variant (5HTTLPR), stress and depression meta-analysis revisited. Arch. Gen. Psychiatry 68, 444–454 77 Kaufman, J. et al. (2007) Genetic and environmental predictors of early alcohol use. Biol. Psychiatry 61, 1228–1234 78 Hariri, A.R. et al. (2002) Serotonin transporter genetic variation and the response of the human amygdala. Science 297, 400–403 79 Heinz, A. et al. (2004) Amygdala-prefrontal coupling depends on a genetic variation of the serotonin transporter. Nat. Neurosci. 8, 20– 21 80 Pezawas, L. et al. (2005) 5-HTTLPR polymorphism impacts human cingulate-amygdala interactions: a genetic susceptibility mechanism for depression. Nat. Neurosci. 8, 828–834 81 Smolka, M. et al. (2007) Gene–gene effects on central processing of aversive stimuli. Mol. Psychiatry 12, 307–317 82 Berton, O. et al. (2006) Essential role of BDNF in the mesolimbic dopamine pathway in social defeat stress. Science 311, 864–868 83 Colzato, L. et al. (2011) BDNF Val (66) Met polymorphism is associated with higher anticipatory cortisol stress response, anxiety, and alcohol consumption in healthy adults. Psychoneuroendocrinology DOI: 10.1016/j.psyneuen.2011.04.010 84 Lau, J.Y.F. et al. (2010) BDNF gene polymorphism (Val66Met) predicts amygdala and anterior hippocampus responses to emotional faces in anxious and depressed adolescents. Neuroimage 53, 952–961 85 Pandey, S.C. (2003) Anxiety and alcohol abuse disorders: a common role for CREB and its target, the neuropeptide Y gene. Trends Pharmacol. Sci. 24, 456–460 86 Zhou, Z. et al. (2008) Genetic variation in human NPY expression affects stress response and emotion. Nature 452, 997–1001 87 Burnett, S. et al. (2011) The social brain in adolescence: Evidence from functional magnetic resonance imaging and behavioural studies. Neurosci. Biobeh. Rev. 35, 1654–1664 88 Grosbras, M-H. et al. (2007) Neural mechanisms of resistance to peer inﬂuence in early adolescence. J. Neurosci. 27, 8040–8045 89 Insel, T.R. (2010) The challenge of translation in social neuroscience: a review of oxytocin, vasopressin, and afﬁliative behavior. Neuron 65, 768–779 90 Chein, J. et al. (2010) Peers increase adolescent risk taking by enhancing activity in the brain’s reward circuitry. Dev. Sci. 14, F1– F10

445

Review 91 Guyer, A.E. et al. (2008) Amygdala and ventrolateral prefrontal cortex function during anticipated peer evaluation in pediatric social anxiety. Arch. Gen. Psychiatry 65, 1303–1312 92 Edwards, S. et al. (2011) Evidence that vasopressin V1b receptors mediate the transition to excessive drinking in ethanol dependent rats. Addiction Biol. DOI: 10.1111/j.1369-1600.2010.00291 93 Tost, H. et al. (2010) A common allele in the oxytocin receptor gene (OXTR) impacts prosocial temperament and human hypothalamiclimbic structure and function. Proc. Natl. Acad. Sci. 107, 13936–13941 94 Stein, L. et al. (2010) Voxelwise genome-wide association study (vGWAS). Neuroimage 53, 1160–1174 95 Calhoun, V.D. et al. (2009) A review of group ICA for fMRI data and ICA for joint inference of imaging, genetic, and ERP data. Neuroimage 45, S163–S172

446

Trends in Cognitive Sciences September 2011, Vol. 15, No. 9 96 Bullmore, E. and Sporns, O. (2009) Complex brain networks: graph theoretical analysis of structural and functional systems. Nat. Rev. Neurosci. 10, 186–198 97 Voineagu, I. et al. (2011) Transcriptomic analysis of autistic brain reveals convergent molecular pathology. Nature 474, 380–384 98 Geschwind, D.H. and Konopka, G. (2009) Neuroscience in the era of functional genomics and systems biology. Nature 461, 908–915 99 Zhang, D. et al. (2011) Multimodal classiﬁcation of Alzheimer’s disease and mild cognitive impairment. Neuroimage 55, 856–867 100 Biswal, B.B. et al. (2010) Toward discovery science of human brain function. Proc. Natl. Acad. Sci. U.S.A. 107, 4735–4739 101 Schumann, G. et al. (2010) The IMAGEN study: reinforcementrelated behaviour in normal brain function and psychopathology. Mol. Psychiatry 15, 1128–1139

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