THE YEAR IN COGNITIVE NEUROSCIENCE 2009 Genes, Cognition, and Communication Insights from Neurodevelopmental Disorders D.V.M. Bishop Department of Experimental Psychology, University of Oxford, Oxford, United Kingdom Twin and family studies have demonstrated that most cognitive traits are moderately to highly heritable. Neurodevelopmental disorders such as dyslexia, autism, and spe- ciﬁc language impairment (SLI) also show strong genetic inﬂuence. Nevertheless, it has proved difﬁcult for researchers to identify genes that would explain substantial amounts of variance in cognitive traits or disorders. Although this observation may seem paradoxical, it ﬁts with a multifactorial model of how complex human traits are inﬂuenced by numerous genes that interact with one another, and with the environment, to produce a speciﬁc phenotype. Such a model can also explain why genetic inﬂuences on cognition have not vanished in the course of human evolution. Recent linkage and association studies of SLI and dyslexia are reviewed to illustrate these points. The role of nonheritable genetic mutations (sporadic copy number variants) in causing autism is also discussed. Finally, research on phenotypic correlates of allelic variation in the genes ASPM and microcephalin is considered; initial interest in these as genes for brain size or intelligence has been dampened by a failure to ﬁnd phenotypic differences in people with different versions of these genes. There is a current vogue for investigators to include measures of allelic variants in studies of cognition and cognitive disorders. It is important to be aware that the effect sizes associated with these variants are typically small and hard to detect without extremely large sample sizes. Key words: genes; language; communication; speciﬁc language impairment (SLI); dyslexia; autism; microcephalin; FOXP2; ASPM; copy number variants; single nu- cleotide polymorphisms (SNPs); copy number variant (CNV)Most human genes have the same DNA se- Studies of twins provide a means of teas-quence, for all people. If a gene takes the same ing apart genetic and environmental inﬂu-form in virtually everyone it is said to be “ﬁxed” ences on a trait and have demonstratedin the population. Although mutations of ﬁxed moderate to high heritability for a host of cog-genes sometimes occur, when they do they are nitive traits and disorders, including verbal andoften either lethal or associated with disease nonverbal intelligence (Bouchard 1998), lan-or disability. When common individual differ- guage skill (Stromswold 2001), reading abilityences among people are shown to be heritable, (Harlaar et al. 2005), speciﬁc language impair-this points to a causal role for genes that are ment (SLI) (Bishop 2002), dyslexia (Grigorenkonot ﬁxed, but that show allelic variation from 2004), and autism (Rutter 2005). Such ﬁnd-person to person. ings led researchers to expect that once we de- veloped sufﬁciently sensitive molecular genetic tools we would readily identify genes that were implicated in cognition generally and commu- Address for correspondence: D.V.M. Bishop, Department of Experi- nication more speciﬁcally. Yet, after more thanmental Psychology, University of Oxford, South Parks Road, OX1 3UD,Oxford, United Kingdom. Voice: +44 1865 271369; fax +44 1865 a decade of research toward this goal, it is281255. email@example.com clear that the task is far harder than anyone The Year in Cognitive Neuroscience 2009: Ann. N.Y. Acad. Sci. 1156: 1–18 (2009). doi: 10.1111/j.1749-6632.2009.04419.x C 2009 New York Academy of Sciences. 1
2 Annals of the New York Academy of Sciencesanticipated. One way of reconciling the Language Development andmoderate-to-high heritability of cognitive abil- Disordersities with the lack of genes of large effect is topostulate that there are common allelic vari- Language is found in all human cultures. Re-ants that affect cognitive function, but they are gardless of whether they grow up in a Westernindividually of small effect size and do not have city or a remote Amazonian forest, childrentheir effects in isolation: They operate against a learn to talk. Furthermore, this remarkablebackground of genetic inﬂuences and their im- skill, unlike anything seen in another species,pact may also be affected by environmental fac- is mastered within around 4 years. Languagestors. Thus, an allelic variant may be associated vary substantially in the sounds and rules theywith lowered ability only when it co-occurs with use to convey meaning, and we are still a longdisadvantageous alleles on other genes or with way from understanding how language learn-speciﬁc environmental circumstances; against ing is achieved, but it is clear that most chil-a different background, it may be neutral, or dren acquire it rapidly and without expliciteven advantageous. instruction. Furthermore, language learning is Research conducted over the past decade remarkably robust in the face of neurobiologi-has increasingly supported a multifactorial cal insults, such as perinatal brain damage, andaccount of genetic inﬂuences on individual environmental adversities, such as limited lan-differences in human cognition, but when re- guage input (Bishop & Mogford 1988). Evensearchers consider speciﬁc neurodevelopmen- severe hearing loss need not handicap languagetal disorders, there is less agreement. Accord- acquisition, provided the child is exposed to aing to one view, the etiology of disorders is no sign language (Neville & Mills 1997). Never-different from the etiology of individual differ- theless, some children have problems with lan-ences in the normal range: Both are affected guage acquisition for no apparent reason. Inby the combined inﬂuence of many common most cases, these children learn to talk, butvariants, the effects of which are small, and they master language milestones much laterinterdependent with one another and the en- than normal, and they may continue to use sim-vironment (Plomin & Kovas 2005). However, pliﬁed syntax and vocabulary into adulthood.an alternative possibility is that we are dealing Where such a picture is seen in a child of other-with numerous but heterogeneous genetic vari- wise normal intelligence for no apparent causeants that are individually very rare (Keller & this is termed speciﬁc language impairment (SLI).Miller 2006). Thus, there might be genes that Although it is less well-known than develop-have a large effect size in affected individuals, mental dyslexia or autism, SLI is a relativelybut only a small effect at the level of the popu- common developmental disorder. The bound-lation, because they affect only a tiny minority ary between SLI and normality is not clearof cases. Rather than attempting to review the cut and prevalence rates depend on how it isenormous and growing literature on genetic in- deﬁned: Tomblin et al. (1997) gave estimatesﬂuences on normal and impaired cognition, I of 3 to 7%, depending on the cutoffs used.shall use examples from neurodevelopmental For many years it was assumed that SLI wasdisorders to illustrate these issues before brieﬂy caused by inadequate parental communication,reviewing studies on the role of copy number but a trio of twin studies in the 1990s providedvariants in autism, as well as on microcephalin evidence of high heritability (see Bishop 2002and ASPM, two putative genes for brain size. for review). One expects twins growing up to-Finally, I shall consider how we would inte- gether to resemble each other because they aregrate what we know about genetic inﬂuences subject to the same environmental inﬂuences.on human cognition and communication with If, however, monozygotic (MZ) twins, who areevolutionary considerations. genetically identical, are more concordant for
Bishop: Genes, Cognition, and Communication 3disorder than dizygotic (DZ) twins, who share neticists talking of complex multifactorial dis-on average only half their segregating genes, orders (Sing & Reilly 1993). There were sev-this is evidence for a genetic inﬂuence on disor- eral reasons for this shift in conceptualization.der. Findings from twin and family studies sug- First, many common disorders, including SLI,gested that it would only be a matter of time do not usually show family pedigrees indicativebefore a gene for SLI would be discovered. The of classic Mendelian inheritance: As Sing andexpectation seemed fulﬁlled when a mutation Reilly succinctly put it, these disorders aggre-of the FOXP2 gene was found to co-segregate gate but do not segregate in families. In thisperfectly with speech and language disorder regard, the KE family is the exception ratherin a three-generational British family, the KE than the rule, with 15 family members acrossfamily (see Fisher 2005, for review). Much de- three generations showing an autosomal domi-bate followed over whether this was a “gene nant pattern of inheritance (i.e., the probabilityfor language” or even a “gene for grammar,” of an affected parent having an affected childan oversimplistic and sensationalist view that is 50%). A second reason for favoring a multi-the researchers were at pains to dispel (Fisher factorial model is when a disorder is common.2006). Comparative studies indicated that this As argued by Keller and Miller (2006), mostgene was highly conserved (similar in DNA se- Mendelian disorders that affect reproductivequence) across mammalian species, with only ﬁtness have very low prevalence because of se-three amino acid substitutions distinguishing lectional pressures against the mutation. If webetween FOXP2 proteins in a human and a accept that language proﬁciency would havemouse. Two of these changes had occurred af- conferred a reproductive advantage for ances-ter divergence of the lineage between chim- tral humans (Pinker 2003), then it is hard topanzees and humans, and further analyses of explain the persistence of common heritablethe gene identiﬁed evidence that high survival language impairments in terms of a single genevalue had led to the changes rapidly spreading of large effect. A third line of evidence con-through human populations. As Fisher (2006) cerns the relationship between disorder and thenoted, the gene is a transcription factor that reg- distribution of abilities in the population as aulates other genes and had impact on many sys- whole. SLI, like many other disorders, does nottems, not just the brain; nevertheless, it is clear have pathognomic features; rather, it is deﬁnedthat disruption of its function affects develop- in terms of arbitrary cutoffs on a continuumment of brain regions important for language. of language ability. Plomin and Kovas (2005)However, FOXP2 is not a general explanation added a more technical argument, one basedfor SLI; investigations of other affected individ- on a speciﬁc analytic method that gives an es-uals seldom found any mutations of this gene timate of “group heritability”—the heritabil-(Newbury et al. 2002), except in a handful of ity for extreme scores on a dimension. Theycases with a similar complex phenotype involv- argued that if group heritability is signiﬁcant,ing both verbal dyspraxia and language deﬁcits then this is evidence of genetic continuity with(Macdermot et al. 2005). normality. However, neither of these latter two Other researchers conducting twin studies of lines of evidence is watertight. We know that theSLI argued that in the majority of cases it is not KE family’s problems are caused by a very rarea distinct disorder, but rather the extreme end of mutation, yet their difﬁculties can still be quan-a normal distribution of language ability, likely tiﬁed in terms of low language test scores. Moreto be inﬂuenced by multiple genetic and en- generally, there are some features of SLI thatvironmental inﬂuences of small effect (Plomin are not normally distributed in the population;& Kovas 2005). Similar conclusions have been rather most children acquire full competencereached about a host of physical disorders, such by around 4 years of age, leaving a tail of casesas heart disease, diabetes, and allergies, with ge- with persisting difﬁculties. Apparent continuity
4 Annals of the New York Academy of Scienceswith normal-range performance may simply the same DNA sequence—this means that onebe a consequence of the measuring scale. This can track how the DNA sequence in a givenis the case for problems using verb inﬂections chromosome region relates to the phenotype in(Bishop 2005) and difﬁculties in speech produc- multiple people from the same family. A com-tion (Bishop & Hayiou-Thomas 2008), both of mon misconception is that discovery of linkagewhich are highly heritable. Using simulations, equates to identiﬁcation of genes that cause dis-Bishop (2005) showed not only that signiﬁcant order. In fact, the highly variable genetic mark-group heritability could be found for a disor- ers used in linkage analysis are unlikely to beder caused by a rare mutation, but that the functional. However, sections of DNA that arepattern of results for verb inﬂection difﬁculties close together tend to be inherited together inwas more consistent with such a cause, rather blocks, so if we ﬁnd signiﬁcant linkage to a DNAthan that of multifactorial inheritance. Thus, marker then there is a good chance that a geneit seems likely that rare variants could account close to that marker may be involved. Thus,for at least some cases of SLI other than those genetic markers act as signposts to regions ofin the KE family. the genome that are likely to harbor risk genes. Nevertheless, it seems probable that for many The identiﬁed region may contain many dif-common forms of SLI we are unlikely to ﬁnd ferent genes, and further painstaking work isindividual genes of large effect; rather the eti- needed to characterize all of these and look forology will be complex and involve a constel- mutant DNA sequences within them. To illus-lation of inﬂuences, each of which is small in trate this point Newbury and Monaco (2008)magnitude. An important question for future noted that the linkage region initially identiﬁedresearch is whether we can distinguish pheno- in the KE family contained around 100 genes,typically between forms of SLI that are caused many of which were plausible candidates be-by rare genetic variants, those that are herita- cause they were known to affect neurologicalble but with complex polygenic etiology, and function. Progress in identifying FOXP2 mightthose that are more environmental in origin. have been much slower, had it not been for aTo resolve such questions we may need to move fortuitous discovery of a single case with a sim-from traditional clinical methods of assessment ilar phenotype having a chromosomal translo-and diagnosis and focus instead on theoretically cation that disrupted this gene.motivated measures of underlying processes of Linkage analysis is a useful method whenperception, memory, and linguistic representa- looking for genes of major effect, especially iftions (Bishop 2008). the same genes are involved in different fami- How, then, are we to discover the relevant lies. It is less good at detecting genes of smallgenes? The main approach available to molec- probabilistic effect, although the method can beular geneticists doing the ﬁrst studies in this made more powerful when quantitative traitsﬁeld was linkage analysis, which involves look- are considered, where one looks for a relation-ing for genetic markers that are shared at above ship between degree of genetic similarity andchance frequency in affected individuals from degree of phenotypic similarity between indi-the same family (see Newbury & Monaco 2008 viduals in a pair. Increasingly, with the adventfor overview). This method relies on the fact of fast automated genetic analysis, researchersthat there are many regions of the genome that are moving to an alternative method, associa-show a high degree of variation from one in- tion analysis. Association analysis is conceptu-dividual to another. Often these are in non- ally rather simpler than linkage analysis in thatcoding regions, thought to be unimportant for it involves categorizing individuals in terms ofcausing individual differences. However, their allelic status at a given locus, and looking for as-variability makes them useful because it is un- sociations with phenotypic measures. It is muchlikely that any two unrelated people will have more sensitive than linkage analysis to small
Bishop: Genes, Cognition, and Communication 5effects; however, the sensitivity of association doing this give the rider a clue as to location,analysis is counteracted by the fact that it is ef- but a sequence of four consecutive letters is farfective only if the marker is very close to the more informative as to where one is. Similarly,actual risk gene. Association analysis has tradi- haplotype analysis is more likely than analysistionally been used to home in on regions that of single SNPs to generate replicable ﬁndings.were identiﬁed by linkage analysis or to test for Ultimately, however, it has to be rememberedassociation with speciﬁc genes that were strong that the genes of interest do not directly codecandidates because their function was known. for behavior: They determine which proteinsIt has now become more common in molecular are produced by cells, thereby inﬂuencing braingenetic studies to perform association analysis structure and function. Proof of a causal rolecovering the whole genome by considering as- for a gene requires studies of its mode of ac-sociations with a large array of single nucleotide tion, with a demonstration that allelic variantspolymorphisms (SNPs) (i.e., variations at a sin- affect expression levels of proteins that servegle nucleotide site that differ between members key functions in the neurobiology of the traitof a species; McCarthy et al. 2008). However, in question (Newbury & Monaco 2008). Untilthe simplicity of this approach is deceptive, be- this is done we cannot know whether a SNPcause on the one hand the associations are likely or haplotype that is associated with a pheno-to be weak and probabilistic, and on the other type is a functional variant or merely a nearbyhand the number of loci that are screened is marker.potentially enormous—in current studies typi- Molecular genetic studies of SLI have beencally running into the hundreds of thousands. conducted by the SLI Consortium (SLIC), whoThis poses problems of interpretation when an assembled a large group of families affectedassociation is found, because adjustment has to by SLI from both epidemiological and clinicalbe made to p-values to take into account the samples. They focused on three main measuresinﬂated probability of chance ﬁndings. But, if of the phenotype: (1) scores from expressive andsuch correction is too stringent, one may end (2) receptive composites of a widely used clinicalup dismissing associations that are small but language assessment, and (3) a test of nonwordgenuine. repetition, previously shown to demonstrate One solution is to adopt a two-stage pro- particularly high heritability in twin studies bycedure in which the ﬁrst genome scan is used Bishop et al. (1996, 2006), and regarded as ato identify target markers that look promising, measure of phonological short-term memory.and the second to replicate in a new sample Linkage was found between a region on the(Thomas et al. 2005). However, since each sam- long arm (q) of chromosome 16 and the non-ple requires large numbers to detect weak as- word repetition phenotype, and between a re-sociations, such work is time-consuming and gion on the short arm (p) of chromosome 19expensive. Furthermore, failures to replicate and the expressive language score (SLI Consor-still occur even when a statistically conservative tium 2002). Both linkages have been replicatedmultistage approach is used. Another way to in additional samples, though the speciﬁc lan-improve reliability of ﬁndings is to capitalize on guage traits linked to chromosome 19 are notthe fact that contiguous SNPs tend to be inher- consistent from study to study (Falcaro et al.ited together in a “haplotype block” and to look 2008; SLI Consortium 2004). These two link-for associations between phenotypes and con- age sites were not, however, signiﬁcant in stud-stellation of alleles in a block (Daly et al. 2001). ies by a North American group, who insteadAn analogy of the difference between doing as- reported signiﬁcant linkage to chromosome 13sociation analysis using SNPs and using haplo- (Bartlett et al. 2002). Lack of agreement in re-types would be passing through a city on a train sults of genome scans is an all-too-commonand noting just one letter of the station name: ﬁnding and raises concerns about false
6 Annals of the New York Academy of Sciencespositives, despite statistical attempts to control quency of 35%, and was found in 40% of thosefor these. However, other explanations are plau- with nonword repetition deﬁcits (performancesible: Results can be inﬂuenced by different more than 2 SD below the population mean),methods of sampling, of phenotypic measure- as compared with 29% of those with good non-ment, or choice of statistical method. Further- word repetition (more than 0.5 SD above themore, where there is genuine but weak linkage, population mean). Children who had no copiesrandom sampling error will affect whether it is of this haplotype had a mean standard score ofdetected. 95.2 on nonword repetition, those carrying one Our ability to home in on genes relevant to copy had a mean score of 89.7, and those withSLI is hampered by our limited understand- two copies had a mean score of 89.4, consistenting of how genes build a brain that can learn with a dominant effect. It is noteworthy that thelanguage. Vernes et al. (2008) adopted a novel effect size of this haplotype, at just below d =approach by taking as a starting point the ﬁnd- 0.4, is large relative to many of the associationsings from the FOXP2 gene, which is known to described in this ﬁeld (see below); nevertheless,have a role in switching on and off other tar- presence of the risk haplotype does not guaran-get genes. Although mutations of FOXP2 are tee poor nonword repetition—indeed the ma-causal in only a small minority of cases of SLI, jority of those with poor nonword repetitionthese authors reasoned that FOXP2 regulates scores did not have this risk haplotype, andother genes that are important in the devel- many of those with good nonword repetitionopment of neural pathways implicated in lan- did have one or two copies of it. A further pointguage, and so they hypothesized that such genes to note is that replications of association stud-might be involved in cases of typical SLI. Vernes ies are always important, and they frequentlyet al. carried out a functional genomic screen ﬁnd smaller effect sizes than the original study.for FOXP2 targets and discovered that FOXP2 Also, the association does not seem speciﬁc tobinds to and downregulates a gene on the SLI: Genetic differences in CNTNAP2 havelong arm of chromosome 7 called “contactin- also been associated with autism (Alarc´ n et al. oassociated protein-like 2” (CNTNAP2). CNT- 2008) and schizophrenia (Friedman et al. 2008).NAP2 is a polymorphic gene known to be im- What can this tell us about genetic inﬂuencesportant in neural development. Around the on normal language development? One con-same time that Vernes et al. were conducting clusion is that language development is robusttheir study, Abrahams et al. (2007) had been in the face of genetic as well as environmen-looking for genes that show differential expres- tal risks; the fact that many people with risksion in the fetal brains of humans and rodents, alleles do not develop frank disorder suggestsand identiﬁed CNTNAP2 as one of two genes that it is unusual for a “single hit” to compro-that showed strikingly higher and more focal mise language development. This conclusionexpression in human prefrontal cortex com- from genetics is nicely consistent with behav-pared to other regions, and generally higher ioral data, suggesting that deﬁcits associatedcortical expression in humans than in rodents. with SLI, such as ones affecting memory or au-Using the sample from the SLIC Consortium, ditory perception, may be seen in family mem-Vernes et al. showed that in children with typi- bers who do not themselves show any severecal SLI, nonword repetition scores were signif- language difﬁculties (Barry et al. 2007, 2008).icantly associated with polymorphisms of this This makes sense if one assumes that languagegene. Using a cluster of nine SNPs that showed shows strong “canalization” (Waddingtonassociation with nonword repetition, the re- 1942), so that a range of genotypes can producesearchers identiﬁed four haplotypes that among the same phenotype. Only when there are twothem accounted for 94% of individuals. The or more factors disrupting language processesmost common of these haplotypes had a fre- will an overt deﬁcit be observed (Bishop 2006).
Bishop: Genes, Cognition, and Communication 7If this is the case then genetic studies might be they give no indication of how many genes aremost fruitful if they focus on component aspects involved. As with SLI, when researchers ﬁrstof the phenotype, which run in families but are started to investigate the genetics of dyslexiaonly probabilistically associated with clinical- there was an expectation that we might ﬁnd alevel impairment. single dominant or recessive gene that could explain the disorder. However, this did not turn out to be the case. Rather, probabilis- Reading and Developmental tic linkages were reported. In the ﬁrst link- Dyslexia age study conducted in this area, Smith et al. (1983) focused on a group of nine families in Language impairment and reading disorders whom dyslexia appeared to be inherited in anoften go hand in hand, yet from a genetic per- autosomal dominant manner. They tested 21spective there is an important difference be- markers and found linkage to a region of chro-tween them. Oral language is a universal hu- mosome 15, with one family contributing sub-man characteristic with obvious survival value. stantially to this result. Although the preciseWritten language, on the other hand, is a hu- location has varied from study to study, linkageman invention that is not found in all societies to the long arm of chromosome 15 was subse-and is of recent origin in the scale of human quently replicated both in an extended study byevolution. Although literacy has clear beneﬁts Smith and colleagues, and by other groups (seein enabling acquisition of knowledge across, as Fisher & DeFries 2002 for review of the earlywell as within, generations, it is not clear that work).illiteracy affects reproductive ﬁtness, and per- With the passage of time a wider range ofsistence of genetic variants that selectively im- markers and more sophisticated methods ofpair reading would pose far less of a paradox analysis became available, allowing further in-than would persistence of variants affecting oral vestigations of the same families. Using a quan-language. titative approach to linkage analysis, Cardon The diagnosis of developmental dyslexia is et al. (1994) found linkage to a region of themade when a child has unusual difﬁculty learn- short arm of chromosome 6 (6p22.2). This wasing to read for no apparent reason. The disor- subsequently replicated in several independentder is typically deﬁned in terms of a substantial samples, though there have also been some fail-mismatch between general intelligence and lit- ures to replicate (see Fisher & DeFries 2002;eracy skills, although the logic of this approach Fisher & Francks 2006).has been challenged (Lyon 2003). A genetic In the past few years, researchers have identi-basis for dyslexia was recognized in some of ﬁed speciﬁc genes that appear to be implicatedthe earliest work on this topic (see Schumacher in dyslexia. Findings do not always replicate,et al. 2007 for review), with Hallgren (1950) and it can be hard to know whether this is be-noting that dyslexia often ran in families and cause of type I error, lack of power to detectsuggesting that genes were implicated. Subse- small effects, or heterogeneity between popu-quent twin studies have largely conﬁrmed this lations. I shall focus on just on one region onimpression; two large-scale two studies, the Col- chromosome 6 where considerable progress hasorado Twin Study (Wadsworth et al. 2007) and been made in the past few years, but for detailedthe Twins Early Development Study (Harlaar critical review of other putative associations seeet al. 2005) both found signiﬁcantly higher con- Paracchini, Scerri, and Monaco (2007).cordance for reading disability in MZ than in Using samples from the UK and US, FrancksDZ twins. et al. (2004) reﬁned the linkage region on Twin studies provide evidence that genes 6p22.2 to a 77-kb region that spanned twoare implicated in developmental dyslexia, but genes, TTRAP and KIAA0319. Within this
8 Annals of the New York Academy of Sciencesregion they identiﬁed a risk haplotype, tagged Paracchini et al. (2006) noted that a causalby three SNPs, that had a frequency of around role for variation in KIAA0319 in dyslexia12% in these samples and had an average effect would be supported if it could be shown that thebetween −0.23 and −0.34 SD (depending on risk haplotype affected neural function. Theythe sample) on IQ-adjusted reading measures. conducted studies in human cell lines usingCope et al. (2005) focused on a region of chro- mass spectrometry to compare the level of tran-mosome 6p22.2 containing 7 candidate genes, scription generated from chromosomes withevaluating 137 SNPs in this region in an inde- high or low risk haplotypes. The experimentpendent UK sample. A multistage analysis was was carried out in different cell types (neu-used, ﬁrst identifying SNPs that showed asso- roblastoma and lymphoblastoids) using mul-ciation with dyslexia in pooled DNA samples, tiple genetic markers. Control cell lines werethen evaluating the association in individual also tested to guard against type I error. Thecases and controls, as well as looking at asso- results indicated that the risk haplotype wasciations within families. In addition, haplotype associated with reduced gene expression ofanalysis was conducted. Signiﬁcant association KIAA0319. This result requires replication,was found for a two-SNP haplotype within the as it was based on just six individuals, but itKIAA0319 gene: The most common form with ﬁts well with data from Harold et al. (2006)alleles 1–1 was equally frequent for affected indicating that the putative functional muta-versus unaffected cases, but two other common tion is likely to reside within the regulatoryforms, 1–2 and 2–1, showed contrasting effects. region of KIAA0319. Paracchini et al. (2007)(Conventionally, 1 is the more frequent allele, further reported that KIAA0319 affects neu-and 2 is the less frequent). The 1–2 haplotype ronal migration, providing a plausible link towas found in 35% of those with dyslexia and previous neuroanatomical work showing ab-27% of those without, whereas for 2–1, the ﬁg- normal neuronal migration in the brains ofures were 24% in those with dyslexia and 36% those with dyslexia (Galaburda et al. 2006).in unaffected controls. Note that while the re- They noted, however, a problem with this hy-port of this chapter is titled “Strong evidence pothesized causal route, which is that impair-that KIAA0319 is a susceptibility gene for de- ment of neuronal migration would be expectedvelopmental dyslexia,” this is not the same as to have a broad impact on many aspects ofsaying that KIAA0319 conveys strong suscepti- cognitive development, rather than a selec-bility. Extrapolating to the general population, tive effect on reading. Nevertheless, it is pos-one would expect that most individuals with the sible that a reading-speciﬁc vulnerability could1–2 risk haplotype would not be dyslexic, and be induced if the regional expression of themost dyslexic individuals would not have the 1– gene were moderated by the effect of other2 haplotype. In a further analysis of the samples genes. Another puzzle, though, concerns theof both Francks et al. and Cope et al., Harold extent to which a mechanism of abnormal neu-et al. (2006) analyzed all the markers previ- ronal migration ﬁts with a view of dyslexiaously identiﬁed by those researchers as well as as continuous with normality. Although oneadditional SNPs in this region and found fur- can undoubtedly have degrees of severity ofther evidence for association of dyslexia with migrational abnormalities, these abnormalitiesKIAA0319, with ﬁve SNPs showing associa- are usually regarded as a pathological phe-tion in both samples. Interestingly, they found nomenon. The fact that association betweenonly weak and inconsistent support for as- haplotypes and dyslexia is most striking whensociation with another gene located close to severe phenotypes are used is consistent withKIAA0319, namely DCDC2, which had pre- results from a sample studied by Deffenbacherviously been reported as also associated with et al. (2004) and is another pointer to the possi-dyslexia. bility that at least some forms of dyslexia may be
Bishop: Genes, Cognition, and Communication 9etiologically distinct from the low end of the of SNPs and haplotypes that had previouslynormal range. been identiﬁed as risk or protective factors for This contrasts with a multifactorial concep- dyslexia in over 5,000 children from a newtualization, which regards dyslexia as being in- population-based sample for whom readingﬂuenced by the same causal factors as operate measures were available. The three-SNP hap-across the whole continuum of reading abil- lotype previously identiﬁed as a risk factor byity. If this is correct, then it should be possi- Francks and colleagues again emerged as sig-ble to identify relevant genes not only in rare niﬁcantly associated with reading ability (anddyslexic samples, but also in general population in the same direction), with the association im-samples, where one would attempt to identify proving when IQ was controlled. Dependingallelic variants that were predictive of reading on the speciﬁc measure used, the regressionacross the range of ability. Two studies looked coefﬁcients (which directly reﬂect change in z-at haplotypes of KIAA0319 in general popula- score going from zero to one to two copies of thetion samples. Luciano et al. (2007) used a sam- 1−1−2 haplotype) ranged from around −0.03ple of 440 adolescent twins and their parents to −0.08. Though signiﬁcant, this is consider-who had been used to obtain estimates of ge- ably weaker than the association reported bynetic and environmental inﬂuences on literacy- Francks et al. They did not, furthermore, repli-related traits. Although they reported signiﬁ- cate the ﬁndings of Cope et al. for a two-SNPcant associations with 2 of 10 studied SNPs haplotype associated with good reading.in this gene, and with a three-SNP haplotype, Developmental dyslexia is often presentedaround half the associations were in the oppo- as a success story for the ﬁeld of genetics be-site direction to those previously reported in cause speciﬁc linkages on chromosomes 6 andstudies of dyslexic samples. This puzzling result 15 have now been replicated across a number ofcould just represent type I error, but it is not the samples. However, candidate gene associationsonly case of a “ﬂip-ﬂop” phenomenon, whereby account for only a small proportion of variance,association with a risk allele is replicated, but in contrast to the high heritability estimates ob-in the opposite direction. Lin et al. (2007) con- tained from twin studies. This, of course, is justducted simulations to show that the extent to what would be predicted by a model of complexwhich different markers were co-inherited (i.e., multifactorial etiology, but it emphasizes thatin linkage disequilibrium) can vary from one we are not ﬁnding genetic variants that are nec-population to another, and that where the phe- essary and sufﬁcient for causing dyslexia. Take,notype depends on a speciﬁc constellation of for instance, the SNP rs2038137 in KIAA0319,allelic variants, then ﬂip-ﬂops between popu- which gave the most signiﬁcant association inlations can occur. This further emphasizes the the chromosome 6p study of Harold et al.extent to which phenotypes depend on genetic (2006) when both samples were considered to-constellations rather than individual genes. An- gether. The risk allele had a frequency of 70%other point to note is that Luciano et al. did in cases of dyslexia and 62% in controls in thenot correct for IQ , and given that reading and Cardiff sample, a difference that would be farIQ tend to be correlated, the phenotype they too small to be of use in predicting outcomes.studied would be rather different from those Assuming a base rate of dyslexia of 10% inin samples of dyslexics, who are usually identi- the population, in a sample of 1000 people, weﬁed on the basis of a mismatch between poor would expect to ﬁnd 628 with the risk allele,reading and average or high IQ. Given the very of whom 11% would have dyslexia, and 372weak evidence for replication found by Luciano with a low-risk allele, of whom 8% would haveet al., it is of interest to ﬁnd more positive re- dyslexia. Clearly, even where signiﬁcant associ-sults in a general population sample studied by ations are replicated across several samples, theParacchini et al. (2008). They analyzed a set variants that have been found have only a small
10 Annals of the New York Academy of Sciencescontributory effect to the etiology of dyslexia. and found that the three domains of social im-There are two possible interpretations of this pairments, communication impairments, andresult: One possibility is that there is a func- restricted interests/repetitive behaviors showedtional variant that has a stronger causal rela- only weak phenotypic correlations and little ge-tion with disorder, which is close to, but not netic overlap. Of course, ﬁnding that compo-identical with, the region identiﬁed by associa- nents of autism can fractionate is not strongtion analysis. But, another very plausible inter- evidence against a distinctive etiology for thepretation is that these risk factors correspond syndrome: Consider, for instance, the case ofto genes of small effect and need to be com- Prader−Willi syndrome, caused by a deletionbined with other genetic and/or environmen- on chromosome 15 and characterized by ex-tal factors to exert their inﬂuence. When the cessive appetite, low muscle tone, and learningﬁrst behavioral genetic studies of dyslexia re- disabilities (Whittington & Holland 2004). Ifvealed high heritability, many of us in the ﬁeld one were to do an analogous study to that ofanticipated that sooner or later there would be Ronald and colleagues in the general popula-genetic tests that would make it possible to iden- tion, measuring these traits, it is unlikely theytify a child’s risk of dyslexia before the start of would have shared genetic variance, simply be-schooling. The much more complex picture re- cause Prader−Willi syndrome is a rare disor-vealed by molecular genetic studies makes that der that accounts for a tiny minority of cases.goal seem increasingly unattainable. Similarly, the causes of the triad of impair- ments in autism could be unitary, even though they can fractionate in the general population.Autism and Copy Number Variants Stronger supportive evidence for Ronald et al.’s model comes from studies of relatives indicat- When molecular geneticists ﬁrst began to ing that similar features, milder in kind andstudy autism, they anticipated that it would be sometimes occurring in isolation, can be seenrelatively straightforward to ﬁnd genes associ- in cases of the “broader phenotype” (Dawsonated with this disorder, because all the pheno- et al. 2002). Furthermore, the fact that thetypic data indicated extremely high heritability CNTNAP2 gene has been found to be asso-(Barnby & Monaco 2003). However, despite a ciated with autism as well as with SLI (Alarc´ n ohuge research effort from consortia using sam- et al. 2008) is consistent with the notion thatples gathered from all over the world, progress there may be common genetic risk factors forhas been slow. One possible reason could be both these disorders, which may be differen-that the phenotype is not appropriately speci- tiated only in terms of there being other riskﬁed. As we found for SLI, the clinical charac- alleles in those with autism that lead to addi-terization of a disorder may not be optimal for tional symptoms. It should be noted, however,deﬁning a genetically meaningful phenotype. that although this idea is currently popular, it isMore progress may be made if autism is recon- not fully supported by behavioral data on rela-ceptualized in a quantitative fashion, rather tives; the broad phenotype of autism does notthan as a syndrome. Furthermore, the possibil- appear to encompass the kind of nonword rep-ity has been made mooted that different com- etition deﬁcits associated with CNTNAP2 andponents of autism may have different genetic seen in individuals with SLI and their relativesorigins, with the full syndrome being observed (Bishop et al. 2004; Whitehouse et al. 2007).only when a speciﬁc constellation of deﬁcits is The possibility of etiological overlap betweenseen (Happ´ et al. 2006). Evidence came from e SLI and autism is currently a focus of con-Ronald et al. (2006), who gave a brief ques- siderable research interest, but the jury is stilltionnaire regarding autistic-like symptoms to out. One promising approach is the develop-parents of a general population sample of twins ment of instruments that allow one to quantify
Bishop: Genes, Cognition, and Communication 11underlying dimensions of autism; this allows us Two independent studies (Marshall et al.to assess subclinical features of this disorder in 2008; Sebat et al. 2007) reported increasedrelatives in the quest for genotype−phenotype rates of CNVs in individuals with autism. In-associations (Duvall et al. 2007). triguingly, most of the CNVs were not seen in One consequence of reconceptualizing the parents, indicating that they had arisen asautism as the result of a speciﬁc conjunction of sporadic mutations, rather than being inher-“risk” alleles is that it might explain why such ited. It is ironic that the massive push for ge-alleles persist in the population despite the fact netic studies of autism was prompted by twinthat individuals with autism are unlikely to re- and family studies that indicated high heritabil-produce (Keller & Miller 2006). The argument ity, yet these latest results have revealed geneticis sometimes made that features of autism that anomalies that arose de novo. This raises ques-are disadvantageous when they occur as part of tions as to whether the presence of CNVs maya syndrome could be beneﬁcial if they occur in provide an explanatory mechanism only for aisolation. For instance, Happ´ (1999) noted that e small subset of those with autism—as Beaudetthe detail-focused cognitive style seen in autism (2007) noted, sporadic CNVs are associatedcould be advantageous under certain circum- with cases of autism that are atypical in thatstances. Baron-Cohen (2000) made a similar they are associated with other syndromic fea-case, quoting Temple Grandin, who herself tures and have an equal sex ratio. It will also behas autism, as follows: “‘What would happen if of interest to know how far such de novo muta-you eliminated the autism genes from the gene tions are related to paternal age, which has beenpool? You would have a bunch of people stand- linked with risk of autism (Reichenberg et al.ing around in a cave, chatting, and socializing 2006).and not getting anything done!’” (p. 491). Another difﬁculty for studies of CNVs in In the past few years, there has been growing relation to disorder is the fact that CNVsinterest in an alternative line of explanation for are so common in the general population.the failure to ﬁnd genes for autism. The prob- Beckmann et al. (2007) noted that in thelem may have been not so much with pheno- HapMap project, when the focus was solely ontypic deﬁnition as with the nature of the molec- differences at the level of the single nucleotideular genetic investigations—investigations that polymorphism, it was estimated that the dif-have focused on looking for differences in ge- ference between any two randomly selectednetic markers that typically encompass 1 to genomes was only 0.1%, but this estimate has10 base pairs. In addition to these conven- now been estimated upwards to at least 1%,tional genetic markers, there is another kind of with most of the difference due to CNVs. Be-polymorphism, copy number variant (CNV), cause they can span regions of the chromosomewhich operates on a much larger scale, involv- containing many genes, CNVs can potentiallying deletions, insertions, duplications, and rear- affect a wide range of phenotypic characteris-rangements of sections of DNA of length from tics. The problem, then, is that if one ﬁnds that1000 base pairs up to several million base pairs an individual with autism has a CNV, it can-(Beckmann et al. 2007; Redon et al. 2006). not necessarily be assumed that this is a factorThese may arise as spontaneous mutations, or in causing the autism (Abrahams & Geschwindbe transmitted from parent to child. For years it 2008). It may be the case that research in thiswas thought that gene dosage was determined area will be less useful in identifying CNVs thatsolely by the alleles inherited from each parent, cause disorder than in identifying genes thatbut it is now evident that dosage can also be are duplicated or deleted by the CNV, whichaffected by the presence of two copies of a gene may be likely candidates for playing a role inon the same chromosome. autism.
12 Annals of the New York Academy of Sciences Microcephalin and Abnormal crocephalin and another microcephaly gene, Spindle-Like, ASPM (Abnormal Spindle-like, Microcephaly- Microcephaly-Associated (ASPM) associated), with both an ancestral form and a new derived form co-existing in human popu- The next example is something of a cau- lations. For both genes, the data were consistenttionary tale, showing how extrapolating from with positive selection pressure for the new de-disorder to gene function in the general popu- rived form. Such ﬁndings ﬁt with the idea thatlation is fraught with difﬁculties. The size and there were evolutionary pressures favoring thecomplexity of the brain is one of the most dis- cognitive advantages of a large brain that out-tinctive features differentiating humans from weighed the additional physiological costs ofother primates (Passingham 2008). The brains maintaining it. However, to conﬁrm that mi-of modern humans are more than 4 times larger crocephalin and/or ASPM were implicated inthan those of great apes. Genes that underlie this development, one would need to show thatthis difference were discovered in the course there were indeed differences in brain size andof studying individuals affected by primary mi- cognition associated with the derived and an-crocephaly. This condition is diagnosed when cestral forms of these two genes. Recent studieshead circumference is at least 3 SD below the have failed to support these predictions. Woodslevel expected for age and sex, in the absence et al. (2006) found no effect of genotype onof other syndromic features. Some forms of brain volume measured using magnetic reso-primary microcephaly are inherited as a reces- nance imaging with 120 participants. Rushtonsive autosomal condition, and to date, six genes et al. (2007) measured general mental ability,have been identiﬁed as important in the etiol- head circumference, and social intelligence inogy (Cox et al. 2006). Typically, microcephaly 644 Canadian adults of varied ethnic back-is associated with mental retardation, often ac- ground and again found no relationship be-companied by other signs of neurological im- tween these phenotypes and a person’s geno-pairment such as motor handicap and seizures. type. The largest study conducted to date by Microcephaly is a very rare condition, but Mekel-Bobrov et al. (2007) looked for associa-its genetic basis aroused considerable interest tion between the alleles of microcephalin andwhen it was discovered that there is wide varia- ASPM and intelligence in 2,393 individuals,tion (polymorphism) in one of the genes that and found no detectable effects of genotype onhad been implicated, microcephalin, which phenotype. These are sobering ﬁndings follow-when mutated leads to premature termina- ing the initial excitement regarding these twotion of synthesis of a protein involved in fe- genes, and they emphasize that one cannot al-tal brain development. Wang and Su (2004) ways predict what the correlates of commondocumented this variation and compared hu- genetic polymorphisms will be from knowledgeman forms of microcephalin with those seen in of the effects of a pathological mutation affect-12 species of nonhuman primate. They found ing the same genes. As Woods et al. (2006)surprisingly high allelic variation in humans, noted, both microcephalin and ASPM are ex-whereas there was much less within-species pressed in organs other than the brain, and itvariation in ape and monkey samples. If genetic may be that positive selection for the deriveddiversity is due to selective pressure rather than variants of these genes has nothing to do withrandom drift, then we expect to see more nu- cognition.cleotide replacements that alter a gene’s pro- A ﬁnal line of evidence about the possi-tein product than replacements that do not. ble effect of microcephalin and ASPM is in-Subsequent work by Mekel-Bobrov et al. (2005) direct: Dediu and Ladd (2007) noted that thereand Evans et al. (2005) produced such evidence were population differences in the frequencyfor continuing adaptive evolution of both mi- of derived and ancestral haplotypes for both
Bishop: Genes, Cognition, and Communication 13microcephalin and ASPM. For both genes the language conveys beneﬁts on the individual,new derived forms are relatively common in and would have an impact on survival valueEurope, North Africa, and parts of Asia, but and reproductive success in ancestral humansrarer in sub-Saharan Africa. In other popula- (Hurford 1991). Possession of oral languagetions, the frequencies of derived forms diverge makes it possible to transmit information overfor the two genes. The authors noticed a corre- time and space, to form social bonds, and tolation between the type of language spoken in a contemplate future as well as past events. Thepopulation and the frequency of allelic forms, fact that in the past few generations people havewith tone languages being most common in had access to birth control, altering the relation-populations where the ancestral alleles predom- ship between ability and reproductive success, isinate, and nontone languages being most com- irrelevant here, because evolution operates overmon in those populations where the derived al- a much longer time-scale. According to stan-lele is found with higher frequency. Dediu and dard evolutionary theory, early humans withLadd emphasized that they were not proposing a “high verbal” form of the gene would havethat genotype determined the type of language left more offspring than those without, so thatthat could be learned: in general a human child this optimal form gradually would become thecan learn any natural language it is exposed dominant one, even if the selective advantageto. However, they suggested that a genotype was relatively slight (Pinker 2003). How, then,that facilitated pitch discrimination might have can we explain the persistence of a “nonopti-played a role in determining which acoustic mal” version of the allele in the population?cues became linguistically salient when a lan- Keller and Miller (2006) addressed this ques-guage developed. Although the concordance tion in the context of mental disorders, but theirbetween population genotypes and type of lan- conclusions have relevance for individual differ-guage is intriguing, this work has to be seen ences in cognition. They noted that a commonas hypothesis-generating rather than deﬁnitive line of argument is in terms of “balancing se-(Nettle 2007). Nevertheless, it gives clear pre- lection,” whereby an allele is maintained in thedictions, not least of which is the idea that a population because disadvantageous effects areperson’s genotype will determine their ability counterbalanced by advantageous traits. Mostto learn to hear tone distinctions. Even more genes are pleiotropic, that is, have multiple ef-speculatively, we might predict that versions fects, and so one can envisage a situation whereof these genes will relate to language proﬁ- an allele might be maintained because, for in-ciency in opposite ways, depending on whether stance, it affects the balance of verbal and non-the language learned is a tone language. As verbal abilities, rather than the absolute level ofyet these remain hypotheses in need of formal either skill. Accounts of autism that stress thetest. advantages as well as the disadvantages of the “autistic cognitive style” could be regarded as instances of this theory. However, Keller and Why Do Genetic Inﬂuences on Miller (2006) are highly skeptical about this Cognition and Communicative type of explanation because it only works if the Skill Persist? balance between advantage and disadvantage of a trait is very close. A puzzling question for evolutionary biology Another possibility is that evolution of genesis why an optimal genotype has not reached ﬁx- for cognition is still underway. If these genesation in the species. Suppose, for instance, that emerged relatively recently they could still bewe have an allelic variant that is associated with subject to natural selection. We considered thisgood verbal skill. There is good reason to be- possibility in relation to microcephalin andlieve that possession of articulate and complex ASPM, both of which show all the hallmarks
14 Annals of the New York Academy of Sciencesof genes currently undergoing selection in hu- Future Directionsman populations. However, as we saw, therewas little evidence that they play a role in in- Over the past 25 years the role of genes inﬂuencing cognitive abilities within the normal cognition has changed from being a minorityrange. While it would be na¨ve to assume that ı interest to a hot topic. This trend has beenevolution had stopped with the emergence of largely driven by technological developmentsmodern humans, there is as yet no evidence in the ﬁeld of genetics that allow for rapid anal-for genetic variants that are adaptively evolv- ysis of an individual’s DNA. Nowadays any-ing and inﬂuencing cognitive abilities. one who has access to samples of DNA—from For common, heritable, psychiatric disor- blood or cheek-scrapings or other tissues—canders, Keller and Miller argued in favor of a send this material to a commercial labora-“common disease/rare variant” model (Wright tory which will categorize individuals accord-et al. 2003). According to this model there is ing to whatever aspect of the genotype the re-little consistency in genotype from one affected searcher requests. Over recent years the ﬁeldindividual to the next, and adverse mutations has changed to reﬂect these developments. Inaffecting any given locus are rare. Keller and the early days, research proposals focused onMiller noted that complex behaviors (which behavior and on deﬁning better ways of mea-would extend to language, intelligence, and so- suring phenotypes and building cognitive mod-cial behavior) involve integrating many com- els of underlying processes. With the advent ofplex pathways and so are potentially vulnera- brain imaging, proposals started to include able to mutations at many loci. Although some structural and/or functional imaging compo-geneticists ﬁnd this model “too depressing to nent, with the hope that identifying the braincontemplate” (Keller & Miller, p. 402), it seems regions and/or networks underlying the dis-increasingly plausible for autism (Abrahams & order might bring more order into a chaoticGeschwind 2008), and examples such as the ﬁeld. Within the past decade genetics has beenKE family indicate that it applies to at least bolted on as the latest weapon in the attacksome cases of speciﬁc language impairment. on complexity. Thus, whereas in the past we A ﬁnal model maintains that cognitive and assessed subtypes of reading disability or acti-communicative skills are affected by the com- vation of the frontal lobes, now we now canbined effect of many genetic and environmental categorize people according to allelic variantsinﬂuences, and individual alleles are unlikely to in the hope that this will reveal clearer patterns.account for substantial amounts of phenotypic So far, however, the result has not been greatervariance, even if heritability is high. When the clarity; on the contrary, as each new method-phenotype is a disorder, this becomes the “com- ology is incorporated into the study of disor-mon disease/common variant” model (Wright ders, greater degrees of complexity are encoun-et al. 2003), but advocates of this view argue tered. This is to be expected given what hasthat the term disease is inappropriate, because emerged about the multifactorial nature of thethere are no clear dividing lines between nor- inﬂuences on cognitive abilities and disabilities.mality and abnormality (Plomin 2000). Inso- We should not anticipate one-to-one relation-far as this model applies, it predicts that we ships between allelic variants and phenotypes,are likely to discover more genetic variants that be they traditional behavioral measures or neu-have small effects, that are common in the gen- robiological endophenotypes (Flint & Munaf` oeral population, and that affect skills in the nor- 2007). The amount of variance that will be ac-mal range as well as in disorders. As we have counted for by variations in a single gene willseen, research on both language and literacy usually be tiny and difﬁcult to detect exceptdisorders has been moving toward adoption of in very large samples. Plomin et al. (2006) putthis model for at least a proportion of cases. the ﬁeld in a sobering context by noting that
Bishop: Genes, Cognition, and Communication 15in a whole genome scan for variants associated Retherford & Sewell 1989), the average intel-with intelligence, the largest effect size found ligence of the population has increased overwas 0.1%, well below Cohen’s (1992) cutoff for the same period (Flynn 1984). This providesa “small” effect size (d = 0.2). Does this mean further evidence for the multiplicity and com-that the whole enterprise of molecular genetics plexity of inﬂuences on human cognition.of cognition and cognitive disorders should beabandoned? Acknowledgments My own view is less downbeat. Genetic stud-ies can help unravel the complex etiology of Thanks to Simon Fisher and Andrew White-cognition and disorders, provided a biologically house for comments on an early draft. Theinformed approach is adopted. The shotgun author is supported by a Principal Researchapplication of a whole genome scan to look for Fellowship from the Wellcome Trust.associations between allelic variants and cogni-tive traits is one way forward, but may not nec- Conﬂicts of Interestessarily be the best method to generate replica-ble ﬁndings. An alternative approach is to use The author declares no conﬂicts of interest.a rare monogenic disorder or an animal modelas an entry point, and, from an understand-ing of the biological pathways involved, iden- Referencestify other genes that are likely to be implicated,as was done by Vernes et al. (2008) in the case Abrahams, B. S., & Geschwind, D. H. (2008). Advances in autism genetics: On the threshold of a new neu-of developmental language disorders. Another robiology. Nature Rev. Genet., 9, 341–355.important issue is the extent to which gene ex- Abrahams, B. S., Tentler, D., Peredely, J. V., et al. (2007).pression varies according to the environment. Genome-wide analyses of human perisylvian cere-When the search for susceptibility genes is con- bral cortical patterning. Proc. Nat. Acad. Sci., 104,ducted using a design that includes relevant 17849–17854.measures of the environment, this opens up the Alarc´ n, M., Abrahams, B. S., Stone, J. L., et al. (2008). o Linkage, association, and gene-expression analysespossibility of exploring gene−environment in- identify CNTNAP2 as an autism-susceptibility gene.teraction (Rutter 2006). This may reveal genes Amer. J. of Human Genet., 82, 150–159.that have little or no effect in one context but Barnby, G., & Monaco, A. P. (2003). Strategies for autismexert a more powerful inﬂuence in another. candidate gene analysis. In G. Bock & J. Goode At one level, the complexity of genetic in- (Eds.), Autism: Neural Basis and Treatment Possibilities. Novartis Foundation Symposium 251 (pp. 48–63). Chich-ﬂuences on human traits may be a cause for ester: John Wiley.celebration. Certainly, it makes research in this Baron-Cohen, S. (2000). Is Asperger syndrome/high-area challenging, but it also keeps at bay the functioning autism necessarily a disability? Devel. andlooming specter of genetic selection for cogni- Psychopathol., 12, 489–500.tive traits. In the ﬁrst half of the 20th century Barry, J. G., Hardiman, M., White, K., et al. (2008). Dura-concerns were frequently voiced that, as hu- tion of auditory sensory memory in parents of chil- dren with SLI: A mismatch negativity study. Brainmans became able to control their fertility, the Lang., 104, 75–88.more intelligent would reproduce less, leading Barry, J. G., Yasin, I., & Bishop, D. V. M. (2007). Heritableto a decline in overall intelligence in the pop- risk factors associated with language impairments.ulation. This might indeed be expected if IQ Gene, Brain Behav., 6, 66–76.were determined by a few genes of major ef- Bartlett, C. W., Flax, J. F., Logue, M. W., et al. (2002). A major susceptibility locus for speciﬁc language im-fect, whose effects were independent of envi- pairment is located on 13q21. Amer. J. Human Genet.,ronment. However, although a negative associ- 71, 45–55.ation between intelligence and fertility has been Beaudet, A. L. (2007). Autism: Highly heritable but notreported in the US during the last century (e.g., inherited. Nat. Med., 13, 534–536.
16 Annals of the New York Academy of SciencesBeckmann, J. S., Estivill, X., & Antonarakis, S. E. (2007). Daly, M. J., Rioux, J. D., Schaffner, S. F., et al. (2001). Copy number variants and genetic traits: Closer to High-resolution haplotype structure in the human the resolution of phenotypic to genotypic variability. genome. Nat. Genet., 29, 229–232. Nature Rev. Genet., 8, 639–646. Dawson, G., Webb, S., Schellenberg, G. D., et al. (2002).Bishop, D. V. M. (2002). The role of genes in the etiology of Deﬁning the broader phenotype of autism: Ge- speciﬁc language impairment. J. Communic. Disorders, netic, brain, and behavioral perspectives. Devel. Psy- 35, 311–328. chopathol., 14, 581–611.Bishop, D. V. M. (2005). DeFries-Fulker analysis of twin Dediu, D., & Ladd, D. R. (2007). Linguistic tone is re- data with skewed distributions: Cautions and rec- lated to the population frequency of the adaptive ommendations from a study of children’s use of verb haplogroups of two brain size genes, ASPM and inﬂections. Behav. Genet., 35, 479–490. Microcephalin. Proc. Nat. Acad. Sci., 104, 10944–Bishop, D. V. M. (2006). Developmental cognitive genet- 10949. ics: How psychology can inform genetics and vice Deffenbacher, K. E., Kenyon, J. B., Hoover, D. M., et al. versa. Quar. J. Exp. Psych., 59, 1153–1168. (2004). Reﬁnement of the 6p21.3 quantitative traitBishop, D. V. M. (2008). Speciﬁc language impairment, locus inﬂuencing dyslexia: Linkage and association dyslexia, and autism: Using genetics to unravel their analyses. Human Genet., 115, 128–139. relationship. In C. F. Norbury, J. B. Tomblin & D. V. Duvall, J. A., Lu, A., Cantor, R. M., Todd, R. D., et al. M. Bishop (Eds.), Understanding Developmental Language (2007). A quantitative trait locus analysis of social Disorders: From Theory to Practice (pp. 67–78). Hove: responsiveness in multiplex autism families. Amer. J. Psychology Press. Psych., 164, 656–662.Bishop, D. V. M., Adams, C. V., & Norbury, C. F. (2006). Evans, P. D., Gilbert, S. L., Mekel-Bobrov, N., Vallender, Distinct genetic inﬂuences on grammar and phono- E. J., et al. (2005). Microcephalin, a gene regulating logical short-term memory deﬁcits: Evidence from brain size, continues to evolve adaptively in humans. 6-year-old twins. Genes Brain Behav., 5, 158–169. Science, 309, 1717–1720.Bishop, D. V. M., & Hayiou-Thomas, M. E. (2008). Heri- Falcaro, M., Pickles, A., Newbury, D. F., et al., SLI Con- tability of speciﬁc language impairment depends on sortium. (2008). Genetic and phenotypic effects of diagnostic criteria. Genes Brain Behav., 7, 365–372. phonological short-term memory and grammaticalBishop, D. V. M., Maybery, M., Wong, D., Maley, A., Hill, morphology in speciﬁc language impairment. Genes W., & Hallmayer, J. (2004). Are phonological process- Brain Behav., 7, 393–402. ing deﬁcits part of the broad autism phenotype? Amer. Fisher, S. E. (2005). Dissection of molecular mechanisms J. Med. Genet.: Neuropsychiatr. Genet., 128B(54–60). underlying speech and language disorders. Appl. Psy-Bishop, D. V. M., & Mogford, K. (Eds.). (1988). Language choling., 26, 111–128. Development in Exceptional Circumstances. Edinburgh: Fisher, S. E. (2006). Tangled webs: Tracing the connec- Churchill Livingstone. tions between genes and cognition. Cognition, 10,Bishop, D. V. M., North, T., & Donlan, C. (1996). Non- 270–297. word repetition as a behavioural marker for inherited Fisher, S. E., & DeFries, J. C. (2002). Developmental language impairment: Evidence from a twin study. dyslexia: Genetic dissection of a complex cognitive J. of Child Psych. & Psych., 37, 391–403. trait. Nat. Rev., Neurosci., 3, 767–780.Bouchard, T. (1998). Genetic and environmental inﬂu- Fisher, S. E., & Francks, C. (2006). Genes, cognition and ences on adult intelligence and special mental abili- dyslexia: learning to read the genome. Trends Cog. ties. Human Biol., 70, 257–279. Sci., 10, 250–257.Cardon, L. R., DeFries, J. C., Fulker, D. W., Kimberling, Flint, J., & Munaf` , M. R. (2007). The endophenotype o W. J., Pennington, B. F., & Smith, S. D. (1994). Quan- concept in psychiatric genetics. Psychol. Med., 37, titative trait locus for reading disability on chromo- 163–180. some 6. Science, 265, 276–279. Flynn, J. R. (1984). The mean IQ of Americans: massiveCohen, J. (1992). A power primer. Psych. Bull., 112, 155– gains 1932 to 1978. Psychol. Bull., 95, 29–51. 159. Francks, C., Paracchini, S., Smith, S. D., et al. (2004).Cope, N., Harold, D., Hill, G., Moskvina, V., Steven- A 77-kilobase region of chromosome 6p22.2 is as- son, J., Holmans, P., et al. (2005). Strong evidence sociated with dyslexia in families from the United that KIAA0319 on chromosome 6p is a suscepti- Kingdom and from the United States. Amer. J. Hum. bility gene for developmental dyslexia. Amer. J. Med. Genet., 75, 1046–1058. Genet.: Neuropsychiatr. Genet., 76, 581–591. Friedman, J. I., Vrijenhoek, T., Markx, S., et al. (2008).Cox, J., Jackson, A. P., Bond, J., & Woods, C. G. (2006). CNTNAP2 gene dosage variation is associated with What primary microcephaly can tell us about brain schizophrenia and epilepsy. Mol. Psychiatry, 13, 261– growth. Trends Mol. Med., 12, 358–366. 266.
Bishop: Genes, Cognition, and Communication 17Galaburda, A. M., LoTurco, J. J., Ramus, F., et al. (2006). Nettle, D. (2007). Language and genes: A new perspective From genes to behavior in developmental dyslexia. on the origins of human cultural diversity. Proc. Nat. Nat. Neurosci., 9, 1213–1217. Acad. Sci., 104, 10755–10756.Grigorenko, E. L. (2004). Genetic bases of developmental Neville, H. J., & Mills, D. (1997). Epigenesis of language. dyslexia: A capsule review of heritability estimates. Mental Retardation and Devel. Disabilities Res. Rev., 3, Enfance, 56, 273–288. 282–292.Hallgren, B. (1950). Speciﬁc dyslexia. Acta Neurologica and Newbury, D. F., Bonora, E., Lamb, J. A., et al., In- Psychiatrica Scandinavica Supplement, 65. ternational Molecular Genetic Study of AutismHapp´ , F. (1999). Autism: cognitive deﬁcit or cognitive e Consortium. (2002). FOXP2 is not a major sus- style? Trends Cog. Sci., 3, 216–222. ceptibility gene for autism or speciﬁc languageHapp´ , F., Ronald, A., & Plomin, R. (2006). Time to give e impairment. Amer. J. Hum. Genet., 70, 1318– up on a single explanation for autism. Nat. Neurosci., 1327. 9, 1218–1220. Newbury, D. F., & Monaco, A. P. (2008). The applicationHarlaar, N., Spinath, F. M., Dale, P. S., et al. (2005). of molecular genetics to the study of language im- Genetic inﬂuences on early word recognition abilities pairments. In C. F. Norbury, J. B. Tomblin & D. V. and disabilities: a study of 7-year-old twins. J. Child M. Bishop (Eds.), Understanding Developmental Language Psych. and Psych., 46, 373–384. Disorders. Hove: Psychology Press.Harold, D., Paracchini, S., Scerri, T., et al. (2006). Fur- Paracchini, S., Scerri, T., & Monaco, A. P. (2007). The ther evidence that the KIAA0319 gene confers sus- genetic lexicon of dyslexia. Ann. Rev. Genomics and ceptibility to developmental dyslexia. Mol. Psych., 11, Human Genet., 8, 57–79. 1085–1091. Paracchini, S., Steer, C. D., Buckingham, L., et al.Hurford, J. R. (1991). The evolution of the critical period (2008). The KIAA0319 dyslexia susceptibility gene for language acquisition. Cognition, 40, 159–201. is associated with reading skills in the gen-Keller, M. C., & Miller, G. (2006). Resolving the para- eral population. Amer. J. Psych., doi: 10.1176/ dox of common, harmful, heritable mental disorders: appi.ajp.2008.07121872. Which evolutionary genetic models work best? Behav. Paracchini, S., Thomas, A., Castro, S., et al. (2006). Brain Sci., 29, 385–452. The chromosome 6p22 haplotype associated withLin, P. I., Vance, J. M., Pericak-Vance, M. A., et al. (2007). dyslexia reduces the expression of KIAA0319, a No gene is an island: The ﬂip-ﬂop phenomenon. novel gene involved in neuronal migration. Hum. Amer. J. Hum. Genet., 80, 531–538. Mol. Genet., 15, 1659–1666.Luciano, M., Lind, P. A., Duffy, D. L., et al. (2007). A Passingham, R. (2008). What Is Special About the Human haplotype spanning KIAA0319 and TTRAP is asso- Brain? Oxford: Oxford University Press. ciated with normal variation in reading and spelling Pinker, S. (2003). Language as an adaptation to the cog- ability. Biol. Psych., 62, 811–817. nitive niche. In M. Christiansen & S. Kirby (Eds.),Lyon, G. R. (2003). A deﬁnition of dyslexia. Ann. Dyslexia, Language Evolution: States of the Art (pp. 16–37). New 53, 1–14. York: Oxford University Press.Macdermot, K. D., Bonora, E., Sykes, N., et al. (2005). Plomin, R. (2000). Behavioural genetics in the 21st cen- Identiﬁcation of FOXP2 truncation as a novel cause tury. Int. J. Behav. Develop., 24, 30–34. of developmental speech and language deﬁcits. Amer. Plomin, R., Kennedy, J. K. J., & Craig, I. W. (2006). J. Hum. Genet., 76, 1074–1080. The quest for quantitative trait loci associated withMarshall, C. R., Noor, A., Vincent, J. B., et al. (2008). intelligence. Intelligence, 34, 513–526. Structural variation of chromosomes in autism spec- Plomin, R., & Kovas, Y. (2005). Generalist genes and trum disorder. Amer. J. Hum. Genet., 82, 477–488. learning disabilities. Psych. Bull., 131, 592–617.McCarthy, M. I., Abecasis, G. R., Cardon, L. R., et al. Redon, R., Ishikawa, S., Fitch, K. R., et al. (2006). Global (2008). Genome-wide association studies for com- variation in copy number in the human genome. plex traits: consensus, uncertainty and challenges. Nature, 444(7118), 444–454. Nat. Rev. Genet., 9(5), 356–369. Reichenberg, A., Gross, R., Weiser, M., et al. (2006). Ad-Mekel-Bobrov, N., Gilbert, S. L., Evans, P. D., et al. (2005). vancing paternal age and autism. Arch. General Psych., Ongoing adaptive evolution of ASPM, a brain size 63, 1026–1032. determinant in Homo sapiens. Science, 309, 1720– Retherford, R. D., & Sewell, W. H. (1989). How intelli- 1722. gence affects fertility. Intelligence, 13, 169–185.Mekel-Bobrov, N., Posthuma, D., Gilbert, S. L., et al. Ronald, A., Happ´ , F., Price, T. S., et al. (2006). Pheno- e (2007). The ongoing adaptive evolution of ASPM typic and genetic overlap between autistic traits at and Microcephalin is not explained by increased in- the extremes of the general population. J. Amer. Acad. telligence. Hum. Mol. Genet., 16, 600–608. Child & Adol. Psych., 45, 1206–1214.
18 Annals of the New York Academy of SciencesRushton, J. P., Vernon, P. A., & Bons, T. A. (2007). No ev- view and meta analyis of twin, adoption, and linkage idence that polymorphisms of brain regulator genes studies. Language, 77, 647–723. Microcephalin and ASPM are associated with gen- Thomas, D. C., Haile, R. W., & Duggan, D. (2005). Recent eral mental ability, head circumference or altruism. developments in genomewide association scans: A Biol. Letters, 3, 157–160. workshop summary and review. Amer. J. Hum. Genet.,Rutter, M. (2005). Genetic inﬂuences and autism. In F. 77, 337–345. R. Volkmar, R. Paul, A. Klin & D. Cohen (Eds.), Tomblin, J. B., Records, N. L., Buckwalter, P., et al. Handbook of Autism and Pervasive Developmental Disorders (1997). Prevalence of speciﬁc language impairment (3rd ed., Vol. 1, pp. 425–452). Hoboken, NJ: John in kindergarten children. J. Speech Hearing Res., 40, Wiley and Sons. 1245–1260.Rutter, M. (2006). Genes and Behavior: Nature-Nurture Interplay Vernes, S. C., Newbury, D. F., Abrahams, B. S., et al. Explained. Oxford: Blackwell. (2008). A functional genetic link between distinctSchumacher, J., Hoffmann, P., Schmal, C., et al. (2007). developmental language disorders. New Engl. J. Med., Genetics of dyslexia: The evolving landscape. J. Med. 359, 2337–2345. Genet., 44, 289–297. Waddington, C. H. (1942). Canalisation of developmentSebat, J., Lakshmi, B., Malhotra, D., et al. (2007). Strong and the inheritance of acquired characters. Nature, association of de novo copy number mutations with 150, 563–565. autism. Science, 316, 445–449. Wadsworth, S. J., DeFries, J. C., Olson, R. K., et al. (2007).Sing, C. F., & Reilly, S. L. (1993). Genetics of com- Colorado longitudinal twin study of reading disabil- mon diseases that aggregate, but do not segregate ity. Ann. Dyslexia, 57, 139–160. in families. In C. F. Sing & C. L. Hanis (Eds.), Ge- Wang, Y. Q., & Su, B. (2004). Molecular evolution of mi- netics of Cellular, Individual, Family and Population Vari- crocephalin, a gene determining human brain size. ability (pp. 140–161). New York: Oxford University Hum. Mol. Genet., 13, 1131–1137. Press. Whitehouse, A., Barry, J. G., & Bishop, D. V. M. (2007).SLI Consortium. (2002). A genome-wide scan identi- The broader language phenotype of autism: A com- ﬁes two novel loci involved in speciﬁc language parison with Speciﬁc Language Impairment. J. Child impairment (SLI). Amer. J. Hum. Genet, 70, 384– Psych. & Psych., 48, 822–830. 398. Whittington, J., & Holland, T. (2004). Prader-Willi Syn-SLI Consortium. (2004). Highly signiﬁcant linkage to SLI1 drome: Development and Manifestations. Cambridge: locus in an expanded sample of individuals affected Cambridge University Press. by Speciﬁc Language Impairment (SLI). Amer. J. Woods, R. P., Freimer, N. B., De Young, J. A., et al. (2006). Hum. Genet., 94, 1225–1238. Normal variants of Microcephalin and ASPM do notSmith, S. D., Kimberling, W. J., Pennington, B. F., et al. account for brain size variability. Hum. Mol. Genet., (1983). Speciﬁc reading disability: Identiﬁcation of 15, 2025–2029. an inherited form through linkage analysis. Science, Wright, A., Charlesworth, B., Rudan, I., et al. (2003). A 219(4590), 1345–1347. polygenic basis for late-onset disease. Trends Genet.,Stromswold, K. (2001). The heritability of language: A re- 19, 97–106.