RESEARCH ARTICLE
Genes Related to Sex Steroids, Neural Growth, and Social–Emotional
Behavior are Associated with Autistic Traits, Empathy, and Asperger
Syndrome
B. Chakrabarti, F. Dudbridge, L. Kent, S. Wheelwright, G. Hill-Cawthorne, C. Allison, S. Banerjee-Basu, and
S. Baron-Cohen
Genetic studies of autism spectrum conditions (ASC) havemostly focused on the ‘‘low functioning’’ severe clinical subgroup,
treating it as a rare disorder. However, ASC is now thought to be relatively common (!1%), and representing one end of a
quasi-normal distribution of autistic traits in the general population. Here we report a study of common genetic variation in
candidate genes associated with autistic traits and Asperger syndrome (AS). We tested single nucleotide polymorphisms in 68
candidate genes in three functional groups (sex steroid synthesis/transport, neural connectivity, and social–emotional
responsivity) in two experiments. These were (a) an association study of relevant behavioral traits (the Empathy Quotient
(EQ), the Autism Spectrum Quotient (AQ)) in a population sample (n5349); and (b) a case–control association study on a
sample of people with AS, a ‘‘high-functioning’’ subgroup of ASC (n5174). 27 genes showed a nominally significant
association with autistic traits and/or ASC diagnosis. Of these, 19 genes showed nominally significant association with AQ/
EQ. In the sex steroid group, this included ESR2 and CYP11B1. In the neural connectivity group, this included HOXA1,
NTRK1, and NLGN4X. In the socio-responsivity behavior group, this included MAOB, AVPR1B, and WFS1. Fourteen genes
showed nominally significant association with AS. In the sex steroid group, this included CYP17A1 and CYP19A1. In the
socio-emotional behavior group, this included OXT. Six genes were nominally associated in both experiments, providing a
partial replication. Eleven genes survived family wise error rate (FWER) correction using permutations across both
experiments, which is greater than would be expected by chance. CYP11B1 and NTRK1 emerged as significantly associated
genes in both experiments, after FWER correction (Po0.05). This is the first candidate-gene association study of AS and of
autistic traits. The most promising candidate genes require independent replication and fine mapping.
Keywords: genetics; Asperger syndrome; autism; empathy; autistic traits; visual search; emotion recognition; SNP;
broader autism phenotype
Introduction
Autism spectrum conditions (ASC) entail a disability
in social and communication development, alongside
unusually narrow interests (‘‘obsessions’’) and repetitive
behavior [APA, 1994; ICD-10, 1994]. ASC have a genetic
basis, indicated by concordance rates fromMZ and DZ twins
[Bailey et al., 1995], and with heritability estimates of over
90% [Bailey et al., 1995; LaBuda, Gottesman, & Pauls, 1993].
Multiple common susceptibility alleles are implicated, along
with environmental and epigenetic factors. Mixed evidence
from genome-wide linkage studies of samples that do not
differentiate between classic (low-functioning) autism and
autism spectrum disorder have implicated nearly all
chromosomes [Abrahams & Geschwind, 2008]. This could
be due to genetic heterogeneity, or potential confounds
from comorbid conditions (e.g. epilepsy, language delay,
below average IQ, or hyperactivity).
While most neuroimaging and behavioral studies of
ASC focus on the higher-functioning end of the autism
spectrum (high-functioning autism and/or Asperger
syndrome (AS)), the large-scale genetic studies have
primarily investigated the lower-functioning end, focus-
ing on classic autism. In this study, we address this
important gap in literature, by reporting two genetic
association studies. The first is of AS, and the second is of
autistic traits in the general population. AS is marked by
social and behavioral impairments, but is not associated
with language delay during development. We chose 68
candidate genes for these two experiments, derived from
INSAR Autism Research, 2009 1
Received September 2, 2008; revised May 20, 2009; accepted for publication May 26, 2009
Published online in Wiley InterScience (www. interscience.wiley.com)
DOI: 10.1002/aur.80
& 2009 International Society for Autism Research, Wiley Periodicals, Inc.
Additional Supporting Information may be found in the online version of this article.
From the Autism Research Centre, Department of Psychiatry, Cambridge University, Cambridge, UK (B.C., S.W., G.H.-C., C.A., S.B.-B., S.B.-C.); MRC
Biostatistics Unit, Institute of Public Health, University Forvie Site Robinson Way, Cambridge, UK (F.D.); Department of Psychiatry, Cambridge
University, Cambridge, UK (L.K.); Department of Psychology, University of Reading, Reading, UK (B.C.)
L. Kent’s present address is Bute Medical School, University of St.Andrews, Fife KY16 9TS, UK.
Address for correspondence and reprints: B. Chakrabarti and S. Baron-Cohen, Autism Research Centre, Department of Psychiatry, Cambridge
University, Douglas House, 18B Trumpington Rd., Cambridge CB2 8AH, UK. E-mails: bhisma@cantab.net (B.C.), sb205@cam.ac.uk (S.B.-C.)
Grant sponsors: Target Autism Genome (TAG); Nancy Lurie Marks (NLM) Family Foundation; Elizabeth Rausing Family Trust; Medical Research
Council, UK, and ENSACP (N-EURO).
three functional categories: (A) sex hormone-related
genes; (B) genes involved in neural development and
connectivity; and (C) genes involved in social and
emotional responsivity (Table I). We searched for com-
mon genetic variants (single nucleotide polymorphisms
(SNPs)) on the assumption that autistic traits are
continuously distributed in the general population
[Constantino & Todd, 2005; Sung et al., 2005]. Each of
the three functional categories derives from a clear
neurocognitive theory of ASC. The fetal androgen theory
[Baron-Cohen, Lutchmaya, & Knickmeyer, 2004] suggests
that genes involved in sex steroid synthesis and transport
might be related to ASC. The neural connectivity theory
[Belmonte et al., 2004], based on evidence from rat and
human brains suggests that the key abnormality in autism
might be related to neural growth and connectivity.
Therefore, genes involved in neural growth, synaptogenesis,
and synapse stabilization were included in our set of
candidates. Finally, the social–emotional responsivity theory
[Chakrabarti, Kent, Suckling, Bullmore, & Baron-Cohen,
2006; Dawson et al., 2002] suggests that the aberrant social
behavior patterns noticed in ASC might be related, in part,
to genes that are known to modulate social behavior in
animals. The rationale of choice for all genes, together with
relevant gene function, is described in detail in the online
Supplementarymaterial (S1). Some of these genes have been
associated with autism in previous genetic studies, and these
are indicated in bold in Table I. These 68 candidate genes
were tested in two experiments.
In Experiment 1, we measured autistic traits in a
population-based sample of volunteers without any
psychiatric diagnoses, to test whether any of these genes
Table I. List of All Genes Included in the Association Study, Along with Brief Functional Roles Where Known
Neural development and connectivity
NGF, BDNF, NTF3, NTF5, NGFR, NTRK1, NTRK2, NTRK3,
TAC1, IGF1, IGF2
Neuronal survival, differentiation and growth.
RAPGEF4 Growth and differentiation of neurons. Mutations associated with classic autism.
VGF Upregulated directly by NGF and expressed in neuroendocrine cells.
VEGF Promotes cell growth and migration, especially during angiogenesis and vasculogenesis, often
observed during hypoxia. Modulated directly by PTEN.
ARNT2 Neural response to hypoxia.
NLGN1, NLGN4X, AGRIN Synapse formation and maintenance in CNS neurons. NLGN4X mutaions have been linked to autism.
NRCAM Neuronal adhesion and directional signalling during axonal cone growth.
EN-2(AUTS1) Neuronal migration and cerebellar development. EN-2 has been previously linked to ASCs in several studies.
HOXA1 Hindbrain patterning. Mixed evidence suggests a link with ASCs.
Social and emotional responsivity
OXT, OXTR, AVPR1A, AVPR1B Linked to social attachment behavior in humans and other mammals. AVPR1A and OXTR have
previously been associated with ASCs.
CNR1, OPRM1, TRPV1 Mediate endogenous reward circuits, in tandem with dopaminergic pathways. Implicated in
underlying rewarding features of social interactions.
MAOB Synaptic breakdown of dopamine and serotonin. Suggested links with social cognition.
WFS1 Mutations linked to affective disorders. Overexpressed in amygdala during fear response, though
exact functional role is not known.
GABRB3, GABRG3, GABRA6, ABAT Mediate inhibitory (GABA-ergic) neurotransmission as well as play a role in early cortical
development. GABRA6 is expressed strongly in the cerebellum; GABRB3, GABRG3, ABAT have all
been associated with ASCs.
VIPR1 Suggested involvement in neural pathway underlying pheromone processing. Mutations associated
with social behavioral abnormalities in mice. Its endogenous ligand (VIP) shows an
overexpression in neonatal children with autism.
Sex hormone biosynthesis, metabolism and transport
DHCR7 Metabolism of cholesterol: precursor for sex hormones (mutations associated with near-universal
presence of ASC).
CYP1A1, CYP1B1, CYP3A, CYP7A1, CYP11A, CYP11B1,
CYP17A1, CYP19A1, CYP21A2, POR
Synthesis of sex hormones such as progesterone, estrogen, cortisol, aldosterone and testosterone.
CYP21A2 and POR mutations associated with CAH.
HSD11B1, HSD17B2, HSD17B3, HSD17B4 Local regulation of sex steroids.
STS, SULT2A1, SRD5A1, SRD5A2 Steroid hormone metabolism.
SHBG, SCP2, TSPO, SLC25A12, SLC25A13 Intracellular transport of sex steroids as well as their important precursors and/or metabolites.
Mixed evidence suggests an association of SLC25A12 with classic autism.
AR Intracellular receptor for testosterone.
ESR1, ESR2 Receptors for estrogen.
CGA, CGRPR, LHB, LHRHR, LHCGR, FSHB Regulation of reproductive functions.
Genes marked in bold indicate those previously linked to ASC through genetic linkage/association studies. For list of SNPs chosen from each gene, see
Table II. Yellow [dark grey]5Neural growth genes; Light grey5 Social responsivity genes; Pink [medium grey]5 Sex steroid genes. [Color table can be
viewed online at www.interscience.wiley.com]
2 Chakrabarti et al./Genes for autistic traits and/or ASC INSAR
were associated with autistic traits. Our primary measure
of autistic traits is the Autism Spectrum Quotient (AQ)
[Baron-Cohen, Wheelwright, Skinner, Martin, & Clubley,
2001b], a 50-point self-report scale with a quasi-normal
distribution in the general population. The AQ has good
reliability and validity, with 80% of people with AS
scoring above 32/50 compared to 1.5% of controls, and
males scoring higher than females [Baron-Cohen et al.,
2001b]. AQ results have been replicated cross-culturally,
and it is independent of IQ, age, education, major
personality traits [Wakabayashi, Baron-Cohen, & Wheel-
wright, 2006], and scores above 32 is an excellent
predictor of AS diagnosis [Woodbury-Smith, Robinson,
Wheelwright, & Baron-Cohen, 2005]. AQ shows herit-
ability of !57% in twins [Hoekstra, Bartels, Verweij, &
Boomsma, 2007], and in parents of children with ASC
[Bishop et al., 2004].
Our second measure of autistic traits focused on
individual differences in empathy. Empathy is a core
deficit in ASC [Baron-Cohen, 1995]. The Empathy
Quotient (EQ) [Baron-Cohen & Wheelwright, 2004] is a
valid and reliable measure of empathy [Lawrence, Shaw,
Baker, Baron-Cohen, & David, 2004], females scoring
higher than males, and 81% of people with AS scoring
less than 30/80 compared to 12% of controls [Baron-
Cohen & Wheelwright, 2004]. Twin studies have
established a genetic basis for empathy in humans
[Zahn-Waxler, Robinson, & Emde, 1992]. Specific genes
have also been implicated in empathy-related behavior in
humans and other animals [Champagne et al., 2006]. The
AQ and EQ allow a wider net to be cast in capturing genes
underlying autistic traits.
In addition to the questionnaire measures, we used two
performance measures related to autistic traits: the ‘‘Read-
ing the Mind in the Eyes’’ (Eyes) Test of empathy [Baron-
Cohen, Joliffe, Mortimore, & Robertson, 1997; Baron-
Cohen, Wheelwright, Hill, Raste, & Plumb, 2001a] on
which people with ASC score below average; and the
Embedded Figures Test (EFT) of attention to detail [ Jolliffe
& Baron-Cohen, 1997] on which people with ASC score
above average. Both are normally distributed in the
population and are candidate endophenotypes, since
parents and siblings of children with ASC show mild
deficits on the Eyes test and above average performance on
the EFT [Baron-Cohen & Hammer, 1997; Dorris, Espie,
Knott, & Salt, 2004; Losh & Piven, 2007]. The EFT and Eyes
tests tap highly specific components within autistic traits
(emotion–recognition, and attention to detail), strength-
ening measures of these within the AQ and EQ.
In Experiment 2, we tested the same 68 genes for
association in a sample of people with clinically diag-
nosed AS, the ‘‘high-functioning’’ subgroup on the
autistic spectrum, in a case–control design. Almost all
previous genetic studies have been conducted on samples
that included both people with classic (Kanner’s) autism
as well as those with a diagnosis of Autism Spectrum
Disorder, an approach that dates back to when ASC were
regarded as rare. Today ASC is thought of as relatively
common (!1%) [Baird et al., 2006; Baron-Cohen et al.,
2009]. If autism is the extreme of normally distributed
autistic traits, then AS is the more logical subgroup to
investigate, since restricting the case sample to AS
removes a suite of comorbid features. This increases
the power to detect genes underlying autistic traits,
independent of genes underlying learning difficulties
or language delay. Family pedigrees of AS suggest
heritability [Gillberg, 1991] and one full genome scan
of AS has been conducted [Ylisaukko-oja et al., 2004],
revealing strong linkage peaks at 1q21–22, 3p14–24, and
13q31–33.
We predicted that genes from each of the three
functional categories would show significant association
with caseness (certain alleles being more common in
the AS group relative to population controls) and/or
with individual differences on the phenotypic measures
(AQ, EQ, the Eyes, and the EFT), in the population
sample.
Materials and Methods
Samples
Individuals (n5 349; 143 males and 206 females, mean
age522.5 years, SD52.6 years) free of any neurological/
psychiatric diagnoses were recruited by advertisement
from a student population. A student population in-
evitably means IQ distribution is not representative, but
this is unlikely to introduce confounds since scores on
the AQ [Billington, Baron-Cohen, & Wheelwright, 2007],
EQ [Lawrence et al., 2004], Eyes Test [Baron-Cohen et al.,
2001a] and EFT (unpublished data) are all independent of
IQ. Participants were included only if they reported
Caucasian ancestry for three generations. They filled in
the AQ and EQ online, and the results were comparable
to those reported previously by our and other groups. The
mean EQ score was 44.1 (range: 9–75, mean score for
males: 36.1, mean score for females: 49.6), and the mean
AQ score was 16.43 (range: 3–36, mean score for males:
18.01, mean score for females: 15.33). AQ and EQ scores
showed a modest negative correlation (Spearman’s
rho5"0.55, Pr0.01). A subset of this sample (n596)
completed an online version of the Eyes Task and the EFT.
In addition, n5 174 cases (140 males and 34 females,
mean age523.2 years, SD514.6 years) with a formal
diagnosis of AS (based on DSM-IV or ICD-10) from
independent clinicians were recruited through our online
database. Cases were excluded if they had comorbid
major psychiatric conditions (psychosis, schizophrenia,
or bipolar disorder), or if they reported incompatible
diagnostic features (such as a history of language delay or
learning difficulties), or if they were self-diagnosed, or if
INSAR Chakrabarti et al./Genes for autistic traits and/or ASC 3
the clinician making the diagnosis was not affiliated to a
recognized specialist psychiatric clinic. DSM-IV and ICD-
10 criteria were used rather than ADI-R/ADOS, as the
latter were designed to diagnose classic autism and their
accuracy in diagnosing AS has not been confirmed. To
our knowledge, this is the largest reported sample for a
genetic association study of AS. As a check on diagnosis,
approximately half of the cases of AS (91 of 174) also
filled in the AQ. One would expect 80% of cases with AS
to score equal to or more than 32 [Baron-Cohen et al.,
2001b]. Out of the 91 cases, 73 (80.2%) scored at
this level, confirming this sample was comparable
to other published samples. As a check on IQ of the
AS sample, approximately 10% of the AS group (n519)
were randomly selected and administered the WASI
full scale IQ test. This revealed a mean full-scale IQ
score of 119.5 (SD521.1), which is comparable to that
of a larger pool of typical student participants that
these volunteers were drawn from. Ethnicity information
was available for 106/174 (60.9%) of the cases, all of who
were Caucasian for at least three generations. To control
for possible confounds from missing ethnicity data
among cases, w2 analyses were performed for each SNP
from Table II, comparing cases with and without
ethnicity information. This revealed no differences in
allele frequencies between these two groups for all but
two SNPs, consistent with genetic homogeneity of the
cases.
SNP Selection
SNPs (216) with a minor allele frequency (MAF) Z0.2 in
the Caucasian population were chosen, to ensure ade-
quate power given our sample size, which was fixed by
external constraints prior to the study. This approach of
selecting multiple common SNPs per gene has the
advantage of checking for informative associations both
directly and indirectly [Collins, Guyer, & Chakravarti,
1997]. SNPs (from dbSNP build 123) were chosen
randomly from across the whole gene, including UTRs
and introns. The number of SNPs per gene varied with
the gene size and number of commercially available ABI
assays, and is detailed in Table II. The median SNP density
across all genes was one SNP per 14.1 kb; 125 of these
SNPs have been genotyped in one or more populations in
the HapMap database (Release 23a). We used TAGGER to
estimate the coverage, which revealed that 40 SNPs were
in strong LD with SNPs genotyped in the HapMap
database. These SNPs covered 7.3% of HapMap variation
at r240.8 and 13.26% at r240.5, for SNPs with
MAF40.001. The remaining 176 of our SNPs did not
exhibit a strong LD with the genotyped SNPs in the
HapMap database, and hence the true coverage is
considerably higher than our estimate. It should be
noted that in absence of complete polymorphism data
on the same sample, it is not possible to estimate the
actual gene coverage.
All volunteers contributed mouth swabs for DNA
extraction. These were anonymized and DNA was
genotyped for the SNPs (see Table II) using standard
PCR-based assays (TaqMans SNP genotyping assays,
Applied Biosystems Inc., CA). The genotyping call rate
was 93.35% across all samples. Concordance for duplicate
samples was 99.8%. No SNP showed a significant
deviation from Hardy–Weinberg Equilibrium at Po0.001.
The following experiments were performed:
1. An association study for AQ and EQ was
conducted on the population sample (n5349)
using nonparametric (Kruskal–Wallis) analysis
of variance for each SNP, since neither the AQ
nor the EQ were normally distributed in our
sample (Anderson–Darling statistic53.26). w2
statistics and asymptotic P-values (two-tailed)
were generated from this test. A sex-specific
analysis was conducted for all X-linked genes. A
similar analysis for the EFT and Eyes tasks was
undertaken in a subset of this sample (n596),
using univariate ANOVA for each SNP, since
neither EFT and Eyes task scores deviated
sign