The Genetic Basis of Complex Human Behaviors Robert Plomin; Michael J. Owen; Peter McGuffin Science, New Series, Vol. 264, No, 5166 (Jun. 17, 1994), 1733-1739. Stable URL: bttp://links.jstor.org/sici?sici=0036-8075% 28199406179 293%3A264%3A5166%3C1733%3A TGBOCH%3E2,0,CO%3B2-V Science is currently published by American Association for the Advancement of Science, ‘Your use of the ISTOR archive indicates your acceptance of JSTOR’s Terms and Conditions of Use, available at hhup:/www.jstororg/about/terms.hml. JSTOR’s Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at hup:/www jstor.org/journalsaaas.huml Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the sereen or printed page of such transmission, STOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals, For more information regarding JSTOR, please contact support @jstor.org. upswww jstor.org/ ‘op 17 07:41:38 2008137. 138. 138.8 ic 10.5 1. Be ray Nadeau, Temas Gone, 5, 82 (1980 and AH. Rater (120), pp 500-598. Batre oat, Nats Sm, 751 (1008). ‘iar, Manimat Genome. sees yer 5. Wasser, G- Kingsbury, 2363, 375 (10), {45, D-Baneon 0. Lipman Otel, Alle Actas 989) f ‘ws. PeAmar, L-Ama, L,Dancolo, JL. Guinat 1M t 18M. Dareipor, 0. &. Farber. M. Poysor, A O The Genetic Basis of Complex Human Behaviors Robert Plomin, Michael J. Owen, Peter McGuffin ‘Quantitative genetic research has built a strong case forthe importance of genetic factors inmany complex behavioral disorders and dimensions in the domains of psychopathology, Personality, and cognitive abilties. Quantitative genetics can also provide an empirical ‘guide and a conceptual framework for the application of molecular genetics. The success. ‘f molecular genetics in elucidating the genetic basis of behavioral disorders has largely relied on a reductionistic one gene, one disorder (OGOD) approach in which a single gene is necessary and sufficient to develop a disorder. In contrast, a quantitative trait loci (QTL) approach involves the search for multiple genes, each of which is neither necessary nor sufficient for the development of a trait. The OGOD and QTL approaches have both advantages and disadvantages for identifying genes that affect complex human behaviors. ‘The received wisdom of the behavioral sciences conceming the importance of “na- ture” (genetics) and “nurture” (environ- ment) in the origins of behavioral difer- ‘ences among people has changed dramati- cally during the past few decades. Environ- smentalism, which attributes all that we are to nurture, peaked in the 1950s. A more balanced view that considers both nature and nurture swept into psychiatry in the 19605 and 1970s. Although this balanced view has been slower to reach some realms of poychology, there are signs that it has anived. For example, at its centennial meet- ing in 1992, the American Psychological ‘Assocation identified genetics as one of the themes that best represent the present and especially the future of psychology (I). Behavioral genetic research began in the 1920s with inbred strain and selection stud- {es of animal behavior and family, twin, and Wales OF UK. ‘adoption studies of human behavior (2). ‘These quantitative genetic designs ases the “bottom line” of transmissible genetic eects on behavior, regardless of the number of genes involved, the complexity of their in- teractions, or the influence of nongenetic factors. As discussed in the first part of this atticle, quantitative genetic research has bull a strong case for the importance of genetic factors in many complex dimensions find disorders of human behavior. ‘Although more quantitative genetic re- search is needed, the future of behavioral genetics les in hamessing the power of molecular genetis to idenily specific genes for complex behaviors. Inthe second par of this paper, inital suoceses are described and research strategies are discused. Although ‘more powerful methods and results are avail- able forthe investigation of animal than Jhuman behavior, animal works discussed in accompanying atiles inthis isue. Quantitative Genetics ‘The change from antipathy to acceptance of genetic factors in the behavioral sciences has occured so rapidly and thoroughly, SCIENCE + VOL. 264 + 17 JUNE 1994 160. 0. P. Hal and W. Ware, Mato Enayel 225 684 (1085) 11. G Predren and. Soran, ip et 162, Speci tars are duo to W-F- Dove ese wiodom ar Alsons se, rfeted| Sroghou the papa We ark &-R Kanda 8, ‘Tonagawa £8 Lda 8 Brow NA. Joins, NOG Copelae, Mri, M Baer Low, Eppig ane FW. Wir for caco and or ‘rena and cur colleagues inthe NSF Coda fo tga Tang rar upport Supported by 8 gant fom tne Macau Pounaon Neko (1 Dopresaon, tha NSF Corer for Bclegea! Siming a Stoo ioe HECA award (SF Wu ard LHP), NI rain MHS0S0, Helen andereeh IS a fl waning avards Tee NSOTIWO ond TS2 Boooots wait especially in psychiatry, that a reminder is warranted about how environmentalistic the behavioral sciences were, even in the 19608. For example, the major explanation for schizophrenia was abnormal parenting. ‘Adoption studies were pivotal in leading psychiatrists to consider nature as well as ‘nurture. Schizophrenia was known to fun in families, with a risk of 13% for offspring of schizophrenic parents, 13 times the pop- ulation rate of about 19% (3). Adoption experiments allow a determination of whether schizophrenia runs in families for reasons of nature or of nurture. In.a clasic study, Heston (4) examined the offspring of schizophrenic mothers who had been adopted at birth and compared their rate of schizophrenia toa control group of adopted offspring. Of the 47 adopted-away offspring of schizophrenic mothers, 5 were diagnosed as schizophrenic, as compared to none of the 50 control adoptees. Indeed, the risk of schizophrenia for the adopted-away off- spring of schizophrenic mothers is the same as the risk for individuals reared by a schizo- phrenic parent. ‘These findings implicating substantial genetic influence in schizophrenia have been replicated and extended in other adoption studies, and they confirm the re- sults of ewin studies that show greater con- cordance for identical twins (about 45%) than fraternal ewins (about 15%) (3). This ‘win method is a natural experiment in which the phenotypic resemblance for pairs of genetically identical individuals lidenti- cal, monozygotic (MZ) twins] is compared to the resemblance for pairs of individuals whose coefficient of genetic relationship is only 0.50 fraternal, dizygotic (DZ) twins) ‘The convergence of evidence from fam- ily, ewin, and adoption designs—each with distinct ‘assumptions—provides the most convincing argument forthe importance of genetic factors in behavioral traits. Behavioral disondes. Evidence for genetic iuence has been found for nearly all bbchavioral disorders that have been inves- tigated (5). Figure I summarizes the results 1738RSTO TN SAS SU ET ECE TSE ‘of twin studies for some of the best studied disorders. Genetic influence is substantial for schizophrenia, Alzheimer’s disease, au- tism, major affective disorder, and reading disability (6). Not all behavioral disorders are influenced tothe same degree by genetic factors. For example, diagnoted alcoholism hasbeen assumed to be highly heritable, Bur new twin studies show only modest ‘renetic influence for males and negligible ‘enetic influence for females. Interestingly, the amount of alcohol consumed shows areater genetic influence than diagnosed alcoholism (7). In contrast to diagnosed alcoholism, autism, which until the 1970s was assumed to be environmental in origin, appears to be among the most heritable psychiatric disorders. Tn addition to the examples in Fig. 1, the following disorders have also shown some evidence of genetic influence: specific language disorder, panic disorder, eating disorders, antisocial personality disorder, and Tourette's syndrome. Some behavioral disorders such as mild mental retardation hhave mot yet been analyzed by genetic research. Figure 2 summarizes results ffom twin studies for some of the best studied common medical disorders. Like behavioral disor- ders, some medical disorders show substan- tial genetic influence—theumatoid arthri- tis, peptic uleers, and idiopathic epilepsy. Others show more modest genetic inva. ence, such as hypertension and ischemic heart disease. Several common medical ds- corders show negligible genetic influence. For example, twin studies suggest negligible heritability for breast cancer as a whole in the general population, even though a rare carly onset familial eype is linked to markers on chromosome 17 (8). By comparing Figs- Land 2, it appears that behavioral disorders on average show greater genetic influence than common medical disorders. Behavioral dimensions. Data on behavior- al variability within the normal range also indicate widespread genetic influence. Fig- ure 3 summaries results of ewin studies for personality (neuroticism and extraverson), vocational interests, scholastic achieve- ment, and cognitive abilities (memory, spa- tial reasoning, processing speed, verbal rea- soning, and’ general intelligence). For quantitative dimensions, the size of the genetic effect can be estimated roughly by doubling the difference between MZ and DZ correlations. This estimate is called heritability, which is a statistic that de- scribes the proportion of phenotypic vari ance in a population that can be attributed to genetic influences. Heritabilities range from about 40 to 50% for personality, vor cational interests, scholastic achievement, and general intelligence. For specific cogni- tive abilities, heritabilities are aso in this range for spatial reasoning and verbal rea- soning but lower for memory and processing speed. Recent research also suggests genetic influence for other cognitive measures such ax information processing, electroencepha- lographic evoked potentials, and cerebral elucose metabolism (9). Examples of re- ently studied noncognitive behaviors that show genetic influence are self-esteem (10), socal attitudes (11), and sexual orientation (12). Lite is known about genetic eflects for perception and learning and for many health-related behaviors (for example, re- sponses to stress, exercise, and diet). Beyond heritability. Quantitative genetic research has gone beyond merely demon- strating the importance of genetics for com- plex human behaviors. Three new tech- niques ae especially wel for this advance: tment, as can be seen most clearly in re- search ‘on cognitive abilities, the most studied domain of behavior. Fast, develop- mental genetic analysis monitors change in genetic effects during development. For cognitive abiley, genetic factors become increasingly important for general intelli- gence throughout the lifespan, reaching heriabilities as high as 80% later in life (13). This is the highest heritability report ced for any behavioral dimension. In addi- tion, with longitudinal genetic designs, its posible to investigate the etiology of age-to- age change—that is, to what extent do genetic efects at one age overlap with ge- netic effects at another age? For general cognitive ability, longitudinal genetic anal- Yes during childhood suggest tha genetic ‘fects donot completely overlap from age age, indicating changes in genetic effects, especially at the early school years (14). ‘A second advance is multivariate renet- ‘ic analysis, which assesses genetic contsibu- tions to covariance among traits rather than to the variance of each trait considered separately. Multivariate analyses of specific cognitive abilities suggest that genetic in- fivences on all specific cognitive abilities overlap to a surprising degree, although some genetic eficts are unique to each ability (15). This finding implies that genes associated with one cognitive ability are likely to be asociated with other cognitive abilities as well. Multivariate analyses also indicate that genetic effects on scholastic achievement overlap completely with ge- netic effects on general cognitive ability (16). Such techniques ean also be wed t0 ankles the fundamental issues of heteroge- neity and comorbidity fr poychintri disor. ders, contributing to a nosology atthe level 1 f mwa coz Fig. 1 (left) Identical twin {monazygotic (MZ)] and fraternal twin [dizy- ‘tic (02)] probandise concordances for bohavoral disorders. Average weighted concordances were derived from the references in (60). 1134 ‘he references in (61). SCIENCE + VOL. 264 + 17 JUNE 1994 Fig. 2 (fight). MZ and 02 probandaise concordances for common medical disorders. Average weighted concocdances were derived itm(6 Rea. a (a oe Rm RC RE a of genetic effects rather than at the level of symptoms. Longitudinal and multivariate approaches have been facilitated by ad- vances in analysis that test the fit between a ‘model and observed data (17). ‘The third example, called extremes analysis, addreses genetic links between rormal and abnormal behavior. If as seems likely, multiple genes are responsible for genetic influences on behavioral dimen- sions and disorders, a continuum of genetic tisk, is likely to extend from normal to abnormal behavior. For example, is major depressive disorder merely the extreme of a continuous dimension of genetic and envi- ronmental variability? A’ quantitative ge- netic technique developed during the past decade investigates the extent t0 which a disorder is the etiological extreme of a continuous dimension (18). Preliminary search with this approach suggests. that some common behavioral disorders such as depresive symptoms (19), phobias (20), and reading disability (21) represent the fenetic extremes of continuous dimensions. ‘Nurture aswell as nature. Another way which genetic research has gone beyond merely documenting genetic influence i 0 focus on the implications of genetic re- search for understanding environmental fluences. Genetics research provides the best available evidence for the importance cof nonhertabe factors. Usually genetic fac- tors do not account for more than about half ofthe variance for behavioral disorders and dimensions. Most of the disorders and dimensions summarized in Figs. 1 and 3 10 eg sg ¢ “win rains atts ertabites: 022 a a Fig. 3. MZ and DZ twin intraclass correlations for personally (neuroticism and extraversion), vocational interests in adolescence, scholastic actievement in adolescence (combined across ‘similar esuts for English usage, mathematics, social studies, and natural science), specie copative bites in adolescence (memory, spatial reasoning, processing speed, verbal reasoning) and ‘general inteligence. Average weighted correlations were derive trom the relerences in (62), ast show as much nonheritable as heritable influence. The current enthusiasm for ge- netics should not obscure the important contribution of nonheritable factors, even though these are more dificult to investi- gate. For environmental transmission, there is nothing comparable to the laws of hereditary transmission or to the gene a8 a basic unit of transmission. It should be noted that the “environmental” in quanti- tative genetics denotes all nonheritablefac- tor, including nontransmissible stochastic DNA events such a somatic mutation, imprinting, and unstable DNA sequences 2). Two specific discoveries from genetic research are important for understanding environmental influences. First, the way in which the environment influences behav- ioral development contradicts socialization theories from Freud onward. For example, the fact that psychopathology runs in fam- ilies has reasonably, but wrongly, been in- terpreted to indicate that psychopathology is under environmental control. Research shows that genetics generally accounts for this familial resemblance. Environmental influences on most behavioral disorders and dimensions serve to make children growing up in the same family diferent, not similar (23). Thi effect, called nonshared environ- ment, leads to the question of how children in the same family experience such different environments. For example, what are the nonshared experiences that make identical twins growing up in the same family s0 ‘often discordant for schizophrenia? ow ox 07 SCIENCE * VOL.264 * IT JUNE 1994 The second genetic discovery about the environment concerns ‘what has been called the nature cof nurture (24). Many widely used measures of the environment show genetic influence in dozens of twin and adoption studies. Research with diverse ‘win and adoption experimental designs has found genetic influence on parenting, childhood accidents, television viewing, classtoom environments, peer gOUPS, 50. cial support, work environments, life events, divorce, exposute to drugs, educa- tion, and socioeconomic status (25). Al thouigh these results might seem paradoxi- cal, what they mean is that ostensible measures of the environment appear to asses genetically influenced characteristics of individuals. To some extent, individuals create their own experiences’ for genetic reasons (26). In addition, genetic factors contribute 10 the prediction of develop- mental outcomes from environmental mes- sures (25). For example, genetics is part of the reason why parenting behavior predicts children’s cognitive development and why negative life events predict depression. ‘Quantitative genetics and molecular genet- ies. Quantitative genetic research is need- ced to inform molecular genetic research. Most fundamentally, quantitative genetic research can steer molecular genetic re- search toward the most heritable syn- dromes and combinations of symptoms. Genes are less likely to be identified for complex behaviors that show litle genetic influence in the population unless some aspect of the trait can be found that is mote highly heritable, as in the case of breast cancer. Although genetic influence hhas been detected for many behavioral disorders and dimensions (Figs. | and 3), litle is known about the most heritable aspects within these domains. Even more usefl is quantitative genetic research that goes beyond heritability to take advantage of new techniques men- tioned above. For example, developmental senetic research shows that genetic influ- ence increasingly afects cognitive abilities ‘throughout the life-span. This suggests that molecular genetic research on cognitive abilities is mos likely to be successful later in life when phenotype better represents genotype. Multivariate genetic research in- dicates that genes associated with one cog- nitive ability are likely to be associated with other cognitive abilities. The clue here is that molecular genetic research will profit from focusing on what cognitive abilities hhave in common. Quantitative genetic re- search suggests that common disorders rep- resent the quantitative extremes of contin- uous dimensions. This suggests that genes associated with disorders can be found by investigating continuous dimensions and 1738FEV FLEE RL ADS PAT SATS SEEN SEES OS RINE TE A TILTON TRL, vice versa. Finally, quantitative genetic research suggests that nongenetic factors ‘generally account for as much variance a5 ‘enetic factors, that behavior-relevant en- ‘vironmental factors generally operate in a ‘onshared manner t0 make children grow- ing up in the same family diferent, not similar, and that genetic factors play a role in individuals actively creating their own experience. Molecular genetic research will benefit from incorporating environmental ‘measures, especially measures of nonshared Molecular Genetics ‘Quantitative genetic esearch leaves litle room for doubt about the importance of genetic influence in behavior. The next sep isto begin to identify some of these genes. This is obviously & more dificult Zep, especially in the case of complex traits and some of the initial steps in this direction have faltered. However, the dif ficulty of identifying specific genes under- lying complex eraits should nt obscure the evidence forthe importance of genetic influence. Many rare disorders such as Hunting- ton's disease show simple Mendelian pat- tems of inheritance for which defects in a single gene are the necesary and suficient cause ofthe disorder. Linkage analysis and the rapidly expanding map of the human genome guarantee that the underlying genes will be mapped and eventually cloned, as has already happened for scores of singlegene disorder. ‘The new frontier for molecular genetics lies with common and complex dimensions, disorders, and diseases. The challenge isto use molecular genetic techniques to idenefy genes in- ‘olved in such complex system influenced by muleiple genes at well mliple non- genetic factors, especially when any single gene is neither necessary nor sufficient. Because this challenge isthe same for com plex behaviors as for common medical dis ‘orders, ther futures will be inertwined. ‘One gene, one disorder? Complex its that show no simple Mendelian pattern of inheritance are unlikely to yield simple genetic answers. For this reason, it has foften been assumed that complex disorders consist of concatenation of several disor- ders, each caused by a single gene, or at least a gene of major effect that largely accounts for genetic influence. Indeed, one definition ofthe word “complex” is a com paaite of distinguishable constituents. This ould be called the one gene, one disorder {OGOD) hypothesis. The OGOD hypoth: ‘esis more than a simple single-gene by pothesis. It does not look fora single gene for complex tats, but rather assumes that complex traits comprise several subvraits 1736 cach influenced by a single gene. Even if Single genes corresponding to subtypes of a disorder cannot be found throughout the population, the hope is that by analysing Tinkage in large pedigrees, it may be post be to find a single gene responsible for a family’s particular version of the disorder. ‘The OGOD strategy, has already been succes for some complex behavioral dis- orders, expecially severe mental retarda- tion, A classic example isthe distinct type of mental retardation, phenylketonuria (PKU), caused by recesive mutations in the phenylalanine hydroxylase (PAH) gene on chromosome 12 (27). Although its in- cidence is low (fewer than 1 in 10,000 bisths), PKU accounted for about 1% of instituionalized mentally retarded individ tals before diets low in phenylalanine were {implemented Recently, another distinct type of men- tal retardation was discovered, fragile X, whichis caused by an unstable expansion of &.CGG repeat in the FMR-I gene on the X chromosome (28). Its incidence is 1 in 1250 males and 1 in 2500 females, making ‘tthe single most important eause of mental retardation ater Down syrome. Another fragile sive on the X chromosome has been linked to a less common form of mental retardation (29). In addition to these de- sin single genes necessary and sulficient to develop distinct forms of mental rtarda- tion, more than 100 other rare sngle-gene disorders include mental retardation among their symptoms G0). ‘Another example of the success of the OGOD approach for behavior involves a common, syndrome, dementia, which i marked by progressive memory loss and confusion. Dementia of the Alsheimer's disease (AD) type includes a rare, familial dementia that appears in middle age, shows dominant Mendelian pattern of inher tance, but which accounts for fewer than 1%6 of AD cases. Mutations in the amyloid precursor protein gene on chromosome 21 Ssegrezae with the disease in some families ‘with autosomal dominant AD (31). The majority of such cass are linked to chro- rmosome 14 (32), although the gene is not yet identified. ‘An example of a successful OGOD ap- proach outside the cognitive realm of retar dation and dementia involves a particulae type of violence. A point mutation in the monoamine oxidase’ A (MAOA) gene, which distupts MAOA activity, has been linked eo impulsive violence in one Dutch family (33). The wellknown false positive linkage results for bipolar afecive disorder and schizophrenia (34) and the mote recent failure to replicate reported X-linkage for bipolar affective disorder (35) were caused by procedural and interpretative problems SCIENCE + VOL. 264 + 17 JUNE 1994 rather than by faults with the analytic technology itself. Ifa single gene is respon- sible for genetic influence on a tat, linkage can detect it. Although the entire genome hhas not yet been screened for linkage with these disorder, itis posible that there are ro genes of major effect tobe found despite clear twin and adoption evidence for genet- ic influence. Conventional linkage analysis of extended pedigrees is unlikely to have sufficient power to detect a gene unless the gene accounts for most of the genetic vat ance. Newer linkage methods such as “af- fected-elative-poit” linkage designs are ‘more robust than traditional pedigree stud- ies because they do not depend on assump- tions about _mode of inheritance (36) ‘These newer methods may be able to detect genes of somewhat smaller effect site iflarge samples (for example, several hundred sib- ling pairs) are used. They ean also incorpo- rate quantitative measures (37). Nonethe- less, linkages found with these methods {imply that a single gene explains most of the genetic effects on the trait, especially if sample sizes are not large. ‘Two behavioral examples involve link- ages reported for sexual orientation and reading disebility. For sexual orientation, linkage has been reporeed with markers on the X chromosome in a study of 40 homo- sexual brothers selected for pedigrees con- sistent with maternal transmission (38). Reading disability has been linked to mark- ers on chromosome 15 and possibly to chromosome 6 in a family pedigree linkage analysis (39) as well as sib-pair linkage analyses of sbling pairs in these same fam- ilies (40), although later reports show less evidence for chromosome 15 linkage. Quantitative tat loci, Quanticative ge- neticsts assume that genetic influences on complex, common behavioral disorders are the result of multiple genes of varying effect size. These multiple-gene effects can con- tribute additively and interchangeably, like risk factors, to vulnerability to a disorder. In this case, the word “complex” means “complicated” in the sense of multigenic and multifactorial rather than a composite ‘of OGOD constituents. Any single gene in a multigene system is neither necessary nor suificient to cause a disorder. In other words, genetic effects involve probabilistic propensities rather than predetermined pro- ‘gramming. ‘Genes that contribute to genetic vari- ance in quantitative traits are called quan- titative trait lock (QTL) (41). One impli- cation of a multigene system is that geno- types are distributed quantitatively (dimen- sionally) even when traits are assessed phenotypically by dichotomous diagnoses. For this reason, the term QTL is apropos for the liability to diagnosed disorders, not juse quantitative traits. The term QTL re-(Rb ae he ee NO le A places the word “polygenic,” which literally tneans “multiple genes” but has come t0 onnote many genes of such infinitesimal ‘effect size that they are unidentifiable. QTL denote multiple genes of varying effect size. ‘The hope is to be able to detect QTL of modest effect size. “Oligogenic” is another word that has been used as a substitute for ppolygenie, bue i¢ presupposes that only a few (oligo”) genes are involved. {QTL examples have been detected by allelic asocation, often called linkage dis esuilibrum, Allelic association refers to a correlation in the population between a Phenotype and a particular allele, usually fssesed as an allelic or genotypic Requency difference between cases and controls. Al Ielie association has often been used to pin down a single-gene effect, but it also. pro- vides the statistical power to detect mall QTL effects, as discussed below. Allelic ‘sociations involving small genetic effects in multiple gene systems could be called QTL associations. The best QTL example for a common medical disorder is provided by the asoctations between apolipoprotein genes and risk for cardiovascular disease, Accounting for as mach as a quarter of the genetic variance (42). "Two recent QTI associations from med- ical research are especially noteworthy in relation to complex behaviors. A deletion polymorphism in the angiotensin-conver ing enzyme (ACE) gene is associated with cardiovascular disease independent of ef- fects on lipid metabolism (43). The fre- quency of individuals homozygous for the AACE deletion was 32% for patients with ‘myocardial infarction and 27% for controls ‘This slightly increased relative risk of 1. which accounts for lest than 1% of the Tibility forthe donde, significant stax tistically because the sample was extremely, ow——_] 81>] e2——o—] large (610 cases and 733 controls). The second example involves longevity, which is only modesty heritable. Significant aso- ciations with longevity have recently been reported for both the ACE deletion and allele 4 of the apolipoprotein E (Apo-E) gene (44). Allelic frequencies for 325 cen- tenarians difered from 20- to 70-year-old controls for the ACE deletion (62% as compared with 53%) and for Apo-E# (5% as compared with 119). Again, these mod estaleic frequency ferences are statsti- cally signiicane because the sample size was so large. In addition, Apo-E2 was associated with increased longevity. The best QTL exampie fora behavioral disorder is the recently discovered associa- tion between late onset AD and Apo-E4 (45). Unlike the rae, arly onset, atoso- smal dominant form of dementia discussed ahove, the prevalence of AD increases steeply with age from less than 1% at age 65 Yeats to 15% in the ninth decade (46). The Frequency of che Apo-E¢ allele i about 0.40 {n individuals with AD as compared with 0.15 in control populations. The odds ra- tio, oF approximate relative risk, is 6.4 for individuals with one of two Apo-E4 alleles (47). The Apo-E4 allele is neither necessary ‘or sufficient to develop the disorder: Many individuals with AD'do not posses: an ‘Apo-E4 allele, and many individuals with an Apo-E¢ allele donot develop AD. Iehas been estimated that Apo-E#" contbutes approximately 17% to the population vari- ance in liability to develop the disorder G2). Although this is & lage effect for a QTL, ie is much too small to qualify as a single-gene effect. A linkage study of 32 pedigrees found only relatively modest evi- dence of linkage for the Apo-E4 region of chromosome 19 (48). A QTL ofthis mag- nitude may be near the lower limit of detection by linkage analysis with realistic sample sizes, as discussed below. We predict that QTL associations will soon. be found for other complex human behaviors. For example, a weak sociation has been sugiested for paranoid schizophrenia in seven fof nine studies with the A9 allele of human leukocyte antigen (HLA), yield- ing a combined relative risk of 1.6, which accounts for about 1% of the liability to the disorder (49). Severe alcoholism (50) and other forms of drug abuse (51) have been in several studies to be associated with the AI allele of dopamine receptor Da, but the association remains controversial 652). A QTL association study of general cognitive ability has found ewo suggestive but as yet unreplicated associations for DNA markers in or near neurally relevant genes (53). Thyroid receptor-B gene has been associated with symptoms of attention deficit-hyperactivity disorder (54). Howev- fet, because this allelic association was found in individuals hospitalized for resis- tance to thyroid hormone, its posible that symptoms of hyperactivity were due to the disease itself As illustrated in Fig. 4 for mental retar~ dation, both the OGOD and QTL ap- proaches are likely to contribute to the elucidation of the genetic basis of complex behaviors. Although we have emphasized the distinction between the OGOD and QTL approaches, we recognize that in fact there isa continuum of varying effect sizes. The relative contributions of single-gene effects at one end of the continuum and undetectably small effects atthe other end are unknown. If genetic efects on complex behaviors are single-gene effects, traditional linkage approaches will detect them. If effects are infinitesimal (for example, ac- ‘counting for less than 0.1% of the vari- ance), they will never be detected. In the riddle of the continuum, QTL of large effect size (for example, genes accounting for 10% of the variance) might be detected by the newer linkage strategies. In the ‘example of Apo-E and AD, linkage analysis suggested the possibility of a gene in this Table 1. Reported linkages and associations with complex behaviors. eat Fig. 4. Complex behaviors such as mental ‘retardation are ikely to Involve single genes, feach responsibie fora distinct subtype of the ‘disorder, as well as QTL that contribute proba bilscaly and interchangeably to genetic risk Behavior ‘Gene, Gromosome Reterence ‘Mental retardation Pherytetonutia aH. 12 (27) Fragile 1 FMB X (es) Fragile XE France x (29) Alzheimer’ disease Early onset, dominant Are, 21 () oa @2) Late onset Avot, 19 3) Violence MAOA, x Gs Hyperactivity Thyroid receptor-8, 3 Gs) Paranoid schizophrenia HMA 6 (49) ‘Alcoholism, crug abuse Dopamine receptor-D,, 11 (60.51) Sexual eientation 3 (38) reading disabilty 218 (63.40) SCIENCE * VOL. 264 + 17JUNE 1994 37SERS EE A Ne TTT HT ST TS region of chromosome 19, and association analysis identified the gene. QTL of small effect size (for example, genes accounting for 1% of the variance) cannot be detected by linkage. Allelic association can detect such QTL, asin the example of ACE and ‘myocardial infarction. ‘Of the few loci that have been implicat- ‘ed to date for complex behaviors (Table 1), ‘most are genes of major effect rather than QTL, especially the indisputable linkages for PKU, fragile X, and early onset, domi- nant dementia. This may be the result of reliance on linkage approaches that are only able to detect genes of major effect. ‘The replicated association between Apo-E4 and AD makes it likely that more system- atic asociation studies will be undertaken to identify QTL with modest effects on complex behaviors. ‘Alli asciton, The advantage fk tage approaches is that they can genes without a priori knowledge of patho- logical processes in a systematic search of the genome by using afew hundred highly polymorphic DNA markers. Such systemat- fe screens of the genome can also exclude the presence of genes of major effect. How- ever, they eannot exclude small QTL ef- fects, atleast when realistic sample sizes are used. We predict that failure to find major gene effects by exclusion mapping for com- plex behaviors will by default provide the best evidence for QTL. The disadvantage of traditional linkage designs is that they are only able co detect single genes or genes largely responsible for the trait. “Although linkage remains the strategy of choice for detecting single-gene eects and for identifying the largest QTL effects, oth- er strategies are needed to detect QTL of smaller effect size. Most likely, new tech- ‘niques will soon be developed to reach this goal. For the present, allelic association represents an increasingly used strategy that is complementary to linkage (55). Allelic association can provide the statistical power needed to detect QTL of small effect size. ‘As in the examples of allelic associations between myocardial infarction and the [ACE deletion polymorphism and between longevity and ACE and Apo-E, statistical power can be increased to detect small QTL associations by increasing sample sizes of relatively easy-to-obtain unrelated subjects. Such small QTL effects could not be detect- ed by linkage analysis with realistic sample sizes. ‘As noted above, allelic association refers toa conelation between a phenotype and a particular allele in the population. Loose linkage between two loci does not result in allelic associations in the population. be- ‘ause alleles on the same chromosome at all ‘but the tightest linked loci are separated by recombination with sufcient frequency 1738 that both sets of alleles quickly return to Tinkage equilibrium in the population. ‘When allelic asociation depends on link- age disequilibrium between a DNA marker land the trait locus, the marker must be very close to the trait locus and both must have low rates of mutation. For example, when the marker and trait Locus are separated by about one million base pairs (that is, a recombination fraction of 0.01), an allelic association would return haliay to equilib- rium in about 70 generations or about 2000 years (56). For this reason, allelic associa- tion research on complex traits can use smatkers in of near relevant genes, because the markers are likely to be in linkage disequilibrium with any functional poly- ‘morphism in the gene. Linkage disequilibrium is not the only ‘cause of allelic association between a marl cer and a trait. Allelic asociation can also ‘occur because the marker itself codes for a functional polymorphism that directly af- fects the phenotype (pleiotropy). Use of such functional polymorphisms greatly en- hanees the power of the allelic association approach to detect QTL (57). It is note- worthy that both the Apo-E4 and ACE deletion markers show direct physiological ceficts. The new generation of complemen- tary DNA markers and techniques to detect Point mutations in coding sequences are rapidly produeing markers ofthis type. ‘Another distinction between linkage and allelic association involves the issue of dimensions and disorders. As mentioned earlier, complex behaviors in multigene systems are likely to be distributed as con- tinuous quantitative dimensions rather than as qualitative dichotomies. Quantita- tive dimensions cannot be easily analyzed by linkage, which is based on cosegregation between @ DNA marker and a disorder, alchough a newly developed sib-pair linkage technique for use with quantitative mea- sures employing interval mapping is prom- ising (37). In contrast, allelic association is as easily applied to quantitative dimensions as to qualitative disorders. Limitations of allelic association analysis include ethnic stratfcation and chance positive results when many markers a examined. The possibility that an allelic association might be the result of ethnic diferences can be investigated by using within-family controls (58). False positives can best be addressed by replication (59). ‘The major limitation to the use of allelic sociation analysis is that a systematic search of the genome would require thou- sands of DNA markers separated by about 500 kb or less and would detect only QTL with low mutation rates. ‘Until such massive genotyping is feasi- ble, allelic association will be limited to screening. functional polymorphisms or VOL. 264 + 17 JUNE 1994 SCIENCE + DNA markers in or near possible candi- date genes. For complex behaviors, the problem is that few candidate genes are known that are as specific as the apolipo- protein genes associated with cardiovascu- lar disease. Nonetheless, many genes ex- pressed in the brain are likely co make very mall contributions to the genetic vari- ‘ance for complex behaviors, which can be detected wih large samples. A single very large representative sample could be used to sereen functional polymorphisms for a ‘multitude of common behavioral as well as ‘medical dimensions and disorders. Inclu- sion of sib pairs would permit sib-pair linkage analyses as well as provide within- family control groups for allelic assoc tion analyses. Such a sample could serve as ‘a cumulative and integrative resource for QTL allelic assoctation research. The goal of the genome project is t0 sequence the entire human genome. How- ever, there is no single human genome. We need to determine the variability of genes ‘between individuals and then to determine how this variation contributes to phenotyp- ic differences between individuals. For com- plex traits (including behavior), this wll be facilitated by a merger between quantitative zenetics and molecular genetics. Conclusions Most of what is currently known about the genetics of complex human behavior comes from quantitative genetic research. Twin and adoption studies have docu: mented ubiquitous genetic influence for ‘most reliably measured behavioral dimen- sions and disorders. More quantitative ge- netic research is now needed that goes beyond merely documenting the presence of genetic influence. This will guide mo- lecular genetic research by identifying the most heritable domains of behavior and the most heritable dimensions and disor- ders within domains. New quantitative genetic techniques can also track the de- velopmental course of genetic contribu- tions to behavior, identify genetic hetero- geneity, and explore genetic links between the normal and abnormal. The same quan- titative genetic data that document signif- icant and substantial genetic influence for complex behavior also provide the best available evidence for the importance of rnongenetic factors. Possible environmen- tal factors need to be investigated in the context of genetically sensitive designs to follow up on the far-reaching findings of, nonshared environment and genetic influ fences on experience and to explore the developmental processes of genotype-en- vironment correlation and interaction by which genotypes become phenotypes. ‘This research will in turn facilitate molecular genetic attempts to identify specific jgenes that contribute to genetic variance in complex behaviors. The confluence of ‘quantitative genetics and molecular genet- is will be synergistic for the elucidation of complex human behaviors. REFERENCES AND NOTES 1. Pomin and GE McClean, Es, Natwo, Future. and Psychology (ketcan Psyeoog- ‘al Associaton, Washington, DC, 1965), 2. R Plann, J.©. Derren, GE. MeCloam, Bohaw- feral Gancics: A Primer reoman, Now Yor, 00 2000, 8. FU Gatesman, Sanzoptvene Ganetcs: The O- lain of Macnose Feaman, Now York, 1991) Fsenblanes for tera na 6 fon greta ‘hn for rant slngs. The suggests reson hated endronment for tn, wn may Be ‘eau of shred penal erarormert or grown Up nthe sare famty at exacly the samo 292 he’ dose: not alc! the in Geng eet of anole iusnce, which compares reser Banos of ateral wine tothe esembance ot ‘cents we, 4, CL Reston fu Payhiany 112, 819 (1968. 15 B McGufin, M3. Owen, N.C. ODanovan, A “Thapar Gotisran. Soars Payette Gano Gasko Landon, 1954) 6, Athough tho szo of fe diererco between MZ ‘and DZ wer coneardancos sugges the magn {toe of th goneteofoc, dchomous core ‘ance daa presrt satsical redken prod Ing osteo reriabity, he tats roresen- ing the oer sz of goneto nave 8 popes inion Forti reson, concordance ae way ‘cowerod To habit ebachoss) corelatons, ‘sion ascume nat corinuass Stun of Tiy undereste lohoomeavs dala (6) He tebity estates boeed on Habiy ecreltons ‘estate genet intuence ana hypohetical c Struct of cortruous taba hate carved Kom a ‘chetorcus Gages tis nol to hontai of the deorcar 22 ‘on atmate foe erably olay are fen ghar han ho wn concorcancs ata would suggest For exam, {erocrzoorrni, some estrtes fhe hea iy of babi excood 0.80, aver ough the MZ ‘concordance lass than 80" Such gh Baby hortabltes might bo mseaang to molouar ‘goretciss coaching fo epoote gones whon Thar osoar 6 bod onthe cro as day ‘oaod As iscssod tr ores good raason > Bote tat rule gone intueness on complex ‘Secret mn coninunis dimension ator tran Genotemas Goordere An imprint e- ‘Search recto that wil erase the pow ‘done eaaarch 10 atop o assacs cman Sons reey rather an to assume en am ‘ehotomens Gagposes, 7.1K epi al, Alana. in Exp As. 11, 249 ‘iosn, 8. JM Hal eal, Science 280, 1694 (190) 8. B.A Veron, Eo, Boies! Azoroace fo the Sucy of umn’'ntligence (Aces, Norwood. Nisto39, 10. R Pomin, Social Dev. 9, 37 (1998, 11, LJ. Eaves HJ. Eyeanck, NG. Marin, Genes, 12, 2M Batley ara FC. Pilar, Ach. Gon. Peet ‘ay 48. 1060 (1901; MC. Neal, ‘age 80, 217 (30 0 16 w. 8 19. 8 sake SCIENCE + VOL. 264 + M, McGuo 7, Bouchag WG, acon, .T Un, (bp 59-76 BIW uor 88, Chey, LR. Caron, (0). ep. 7789 ROL Pederson, A Plain, GE. MeClan, It fgonce in pros. wn D. K. Deteman. R. Pm, Pojcnal Sa! 2,158 (1801) SJ. Waceworh. Mh Natura ard Narre During Medcle Chiood Ciberiee,R Poni, 0, Fler, Eos, lage el, Cambridge, MA, 1904), pp. 66-10% MG. Neale and Le Cardo, Molo fr Gane Stuioe of Twine and Famies (lower ‘acetic, Nowe, MA, 198), °C" Dares ancl OW. Fukar, Bohaw. Gon. 1, Paycot Peyniry 38, 97 (1982). TC Sernee ands lla (1), pp 121-145, P. Neg Acheson, MJ. Over. A. Famer, But Payor Tot 99 (1098). FBemin nO. Dares, Bora Bran Sol 10,1 (oan FR. Pemin and . 8. Berguman, ibis 14, 373 (Goan F. Plain, Gonetios and Experience: The inter ‘py Botwoun Nato ord rte (Sago, New fey Park CA, 1898) ewe, Tne Lite of Fam infuence (Ca fer New York, 190. Seat, Cd Dow. 681 (1980, 5.(-C. Woo, a Gone, fai, and Bahavi PP. Mefiugh and VA MeKustk. Eos, aver, Now York, 1881, pp. 188-208, ALL Vera ta, Gal 6, 905 (1201), SI Lokmght of a, i 4,127 (1099) ston Men. Dati. ls 98,1 (1900) Trends Newt ‘smh, ‘Adblese. Psychiatry 28,368 a Human isso, G Bock ‘fa GM Colina, Eas, (ly, Cachet, UK 1987. pp 90-120 08,672 (1098), 17 JUNE 1994 eee a se2 8 8 61 M.A Poicak-Vance et at, Am. J Yum: Ganet 48, 1 (1094. P-McGutin arc, Sur Hum Heed Fubintin, Ann Hum, Ganet ‘warhHeineman, Landon, 199), pp 165-18 Feaaing daabiy JC. Dots and JG, (pp. 21-148, Pardons diese: Ekidge_ and We A, Foon The Goneue Baas of Common is ‘esos, FAK, 1 Roe, A G. Moho, Eos (Gutord Uni. Prose, Now York, 1999), pp. 775 "794; boast cancer NV. Holm: M Hauge, ©. M Jeroen: Cancer Suv. 3,17 (1962) vere Euctve pulmonary dsbase, hypertension, an ‘Scheme heat disease: K. 8. Kendlr and ©. 0. Fotinate, Am J. Poynity 140, 1381 (1989) oumatog avai: RA. King. The Gone fete of Common Diseases R.A. Keg, 3 Fate, A. Motul, Ee, (Odor Un Press, Now York, 1992), p. 506-628 pepe cor (= ‘agreed by say) | Rate. Shoal. .M. Petree, lid, pp. 240-278: play: T.O. Bint. pp. 732-782, NNowrtcion and araversion: J. ©. Losin, (Genes and Envanmentin Personally Develop ‘ment (Sage, Noweury Park, CA 1902) vocation ‘meres J.C. Loshin and Rieho, erety, Enironmort and Personality (Uni. Texas Pose, stn, 1976) C. A Robots and 18 dohanscon, I" Vocatonal Behav. 4, 237 (974) seelasic achievement 3 C. Loon na FG. Ness, yEnircrment and Porsonaty (Un. of Texas Press, Asi, 1976), Specie cogntive abies: LF. Schoanol, ‘Mas. Eval Gu. 1. 130 (1968); M. Worn, Brag if akretarghatensPsykotogi, en Tu Ingurcersolring (Canoe, Goiebors, Sweden, 1882) goneralWeoligonce’ 4. Bouchard hr. ane M NeSuo, Scionce 212, 1055 (186). Suppoted by grant HO-2760¢ tom the Nona Incbtes of Onid Hoath and Hunan Devacp- ‘ert. The arte benefted Rom sungeston by CrCrabbo, dC. Does, Roagues, and Kh Savdine 1739