Background
Twin studies indicate that personality and behavioural traits including antisocial behaviour have a strong genetic component [
1‐
4]. However, similarly to many other complex traits, the sources of their heritability have not been uncovered to a substantial degree by genome-wide association studies (GWAS) [
3,
5‐
9]. Accumulating evidence suggests that heritability not so far accessed by GWAS resides partly in under-studied forms of genetic variation including length polymorphism in arrays of short tandem repeat (STR) sequences, otherwise known as microsatellites [
10‐
13]. This is not adequately represented in GWAS because the frequency and diversity of microsatellite polymorphism are much higher than those of single nucleotide polymorphism (SNP) [
11,
14].
Many species including bacteria and yeasts are known to have harnessed the high mutability of microsatellites for regulatory purposes, and there is evidence that microsatellite functionality is far more widespread in the human genome than has traditionally been appreciated [
11‐
13,
15‐
18]. For example a recent study of expression quantitative trait loci estimated that at least 10–15% of the heritability in human gene expression levels attributable to common variants in
cis is due to microsatellite polymorphism [
12]. Candidate-gene association studies for a growing number of human regulatory microsatellites have been well replicated [
19‐
24], and mechanisms by which microsatellites function, including modification of spacing between adjacent transcription factors and other promoter elements [
25‐
27], regulation of splicing [
22,
28,
29], adoption of structural variants such as Z-DNA [
30‐
32], and modification of epigenetic signals [
12] are increasingly well understood. Despite this body of evidence, examples of phenotypic associations with microsatellites remain isolated. This is mainly due to difficulties genotyping large numbers of loci, and while methods have recently been developed that will facilitate microsatellite-based association studies comparable in scale to contemporary SNP-based GWAS, these still require extensive sequencing and/or capture probe synthesis [
11,
33].
Previous work by our group has attempted to identify likely functional microsatellites by investigating their conservation among mammalian species [
34,
35], and links between microsatellites and several neurological disorders and behavioural phenotypes suggest that they are particularly likely to be functionally important when associated with brain-related genes [
13,
36]. These considerations motivated us to investigate associations between human behavioural traits and a previously unstudied AC
12-13 microsatellite in the upstream promoter region of the T-Box Brain 1 (
TBR1) gene, which is in the top 2.5% of human microsatellites by level of conservation [
34].
TBR1 is a member of the set of regulatory genes involved in the genesis and fate determination of glutamatergic neurons [
37‐
39]. As a transcription factor it regulates as many as 124 other genes [
40,
41] including
GRIN2B, which has been implicated in attention deficits in children [
42], cognitive performance [
43], neuroticism [
44] and smoking [
45] among other psychiatric and behavioural phenotypes [
46].
TBR1 haploinsufficiency in mice results in defective axonal projections of amygdalar neurons and the impairment of social interaction, ultrasonic vocalization, associative memory and cognitive flexibility [
40]. In humans, deletion of chromosomal regions including
TBR1 is associated with intellectual disability [
47], and mutations of the gene have been linked to autism spectrum disorder [
48,
49].
TBR1 was also recently distinguished as the highest scoring locus associated with educational attainment in a study of more than 300,000 individuals [
50].
We also sought to replicate and/or expand on previous studies showing associations between psychological and/or behavioural phenotypes and three other tandem repeats. The GGN microsatellite in exon 1 of the androgen receptor (
AR) gene encodes a poly-glycine stretch associated with receptor responsiveness [
51,
52]. It has been less studied than the CAG microsatellite in the same exon, which is associated with diseases including spinobulbar muscular atrophy [
53], but its common length variants have been linked to externalizing behaviours including conduct disorder [
54] and the personality traits self-transcendence [
24,
55], aggression and impulsivity [
56]. Two intronic repeats in the
AP-2β transcription factor gene have also been studied relatively little. A minisatellite in its first intron has been associated with the impulsivity-linked phenotypes alcohol dependence [
57] and late auditory-evoked potentials [
58]. A CAAA microsatellite in the second intron of the same gene has been linked to reduced MAOB activity, which is also related to impulsivity [
59], late auditory-evoked potentials [
58], anxiety-related measures [
60] and the personality traits self-transcendence and spiritual acceptance [
61]. The
AP-2β transcription factor functions in neural development and influences brain monoaminergic systems by regulating target genes [
62,
63].
The initial aim of this study was to assess the effects on human traits relating to personality, behaviour, cognitive ability and mental health of our selected repeat polymorphisms. The most significant associations we found were between the
TBR1 microsatellite and antisocial behaviour. Following this, recent interest by two of our authors (LJH and DMF) in maternal smoking during pregnancy (MSDP) and its effects on offspring behaviour [
64,
65], led us, in conjunction with evidence showing expression of
TBR1 in the human embryonic neocortex [
37], and linking the gene to responses to prenatal cannabis exposure [
66], and maternal illness [
67], to the additional hypothesis that MSDP may modify the effects of
TBR1 microsatellite genotype.
Our test subjects were from a well-studied birth cohort, the Christchurch Health and Development Study (CHDS). The CHDS is a longitudinal study of 1265 children born in Christchurch, New Zealand who have been studied on 22 occasions from birth to age 30 [
68,
69], 570 of whom were included in the work presented here. As part of this study data were gathered on: a) MSDP; b) measures of antisocial behaviour assessed from early childhood to mature adulthood by a wide range of methods; c) potentially confounding social and contextual factors.
Results
Association analyses
We examined associations between four short tandem repeats in the
TBR1,
AR and
AP-2β genes (Table
1) and a series of 18 measures of personal characteristics, temperament and behaviours in the CHDS cohort (Table
2). The GGN microsatellite in exon 1 of the
AR gene showed modest nominally significant associations between a dichotomous (homozygous major allele GGN
23 carried by 70% of individuals vs rest) measure and two outcomes: child IQ (r = −0.104,
p = 0.046) and anxiety symptoms (r = 0.096,
p = 0.040). Those homozygous for the major allele scored higher on IQ and lower on anxiety symptoms. Neither association withstood correction for multiple comparisons. There was no apparent association with neuroticism or extroversion. The CAAA
4-5 repeat polymorphism in intron 2 of the
AP-2β gene showed nominally significant correlations with two measures: child scholastic ability (
r = 0.136,
p = 0.004) and a measure of psychotic symptoms (
r = -0.110,
p = 0.008). Carriers of the minor allele scored higher on scholastic ability at age 13 and lower on psychotic symptoms at age 18. We found no clear associations for the minisatellite in intron 1 of the
AP-2β gene. The AC
12-13 microsatellite in the promoter region of the
TBR1 gene was associated with three outcomes. The 22% of individuals who carried the minor allele (AC
13), scored higher on the measures of childhood conduct problems (
r = 0.141,
p = 0.0007), childhood anxiety/withdrawal (
r = 0.102,
p = 0.015), and hostility symptoms in adolescence (
r = 0.087,
p = 0.039). Only the association with childhood conduct problems withstood correction for multiple comparisons.
Reanalysis of the data including those of non-European ancestry identified a small number of additional nominally significant (p < 0.05) associations between the GGN microsatellite and the mental health measures of somatisation, phobic anxiety and psychoticism (r = 0.088–0.094), indicative of lower scores on these measures for those who were homozygous for the major allele. The lowest P value for these tests was 0.006, for an association between the homozygous major allele genotype and lower levels of anxiety, and there was nothing significant following Bonferroni correction. Otherwise essentially the same pattern of findings applied to the full sample.
Tests of sex by genotype interaction identified four nominally significant interactions: for AR GGN on childhood conduct problems (p = 0.013); and for AP2-β CAAA on childhood conduct problems (p = 0.043), childhood attention problems (p = 0.035) and interpersonal sensitivity in adolescence (p = 0.035). The associations with the measures of childhood behaviour problems were stronger for males, and the association with interpersonal sensitivity stronger for females. However, none of these interactions withstood Bonferroni correction. No sex interactions were observed for the minisatellite in intron 1 of the AP-2β gene or for the TBR1 microsatellite.
The most significant association we found overall was for the TBR1 microsatellite minor allele (AC13), carried by 22% of the sample, with childhood conduct problems at age 7–9. This was statistically significant after Bonferroni correction for multiple comparisons (r = 0.141, p = 0.0007), and we performed no further analyses on the other three repeats. For the reasons outlined above we examined one additional hypothesis: that there may be a joint effect of TBR1 and MSDP on childhood conduct problems, and we discovered an interaction. We then replicated the analysis by widening the outcomes to other measures of antisocial behaviour. These included measures of adolescent conduct problems (15–16 years), self-reported property/violent offences (14–25 years) and self-reported arrests or convictions (16–25 years).
Associations between smoking during pregnancy, TBR1 and subsequent antisocial behaviour
Table
3 shows the sample cross-classified according to history of MSDP (no/yes) and
TBR1 microsatellite genotype (homozygous major allele/other). For each classification the table reports the mean scores on a series of dependent variables representing measures of antisocial behaviour over the life course. The table also reports the results of regression models fitted to the data for each outcome including tests of significance of: a) the main effect of
TBR1; b) the main effect of MSDP; c) the
TBR1 by MSDP interaction. The table shows that:
1.
In three out of four cases there was a significant (p < 0.05) main effect of MSDP, reflecting a pervasive association between MSDP and higher rates of subsequent anti-social behaviour assessed up to the age of 25.
2.
There were no significant main effects for TBR1 reflecting the fact that in the absence of MSDP overall rates of antisocial behaviour did not vary with TBR1.
3.
In all analyses there was a significant (p < 0.05) MSDP by TBR1 interaction, reflecting the fact that the effects of MSDP on antisocial behaviour were more marked for the ‘other’ strata who carried at least one TBR1 minor allele than for the TBR1 homozygous major allele group.
As noted in Methods, the analysis in Table
3 is based on a classification of
TBR1 genotype in which all carriers of the
TBR1 minor allele have been classified into a single group. This was done because of the very small number (
n = 5) who were homozygous for the minor allele. At the same time it is of interest to note that elaboration of the data in Table
3 according to the number of
TBR1 minor alleles showed a pattern of data for three of the four outcomes consistent with an increasing effect of MSDP on antisocial behaviour with an increasing number of
TBR1 minor alleles (see Table S1 in Additional file
1).
A limitation of the analysis in Table
3 is that it reports multiple tests of significance, thus increasing risks of type 1 statistical errors. To address this issue the data in the table were re-analysed using a multivariate regression modelling approach to test the joint significance of effects across all outcomes (see
Methods). This analysis confirmed the presence of a just significant main effect of MSDP (F(4,520) = 2.38,
p = 0.05), the absence of a significant main effect for
TBR1 (F(4,520) = 0.57,
p = 0.59) and a significant MSDP x
TBR1 interaction (F(4,520) = 3.95,
p < 0.005).
Adjustment for confounding factors
To examine the possible effects of confounding social, economic and related factors, the analyses in Table
3 were extended to include a range of covariate factors correlated with MSDP. These factors included measures of: maternal age, maternal education, family socio-economic status, family income, family living standards, family type (single parent/two parent) and birthweight. Table
4 reports tests of the main effects and the
TBR1 by MSDP interaction after covariate adjustment. This table shows that in all cases the significant
TBR1 x MSDP interaction persisted. A multivariate regression model also confirmed the presence of a significant
TBR1 x MSDP interaction (F(4,501) = 3.3,
p = 0.01). However, the overall main effect of MSDP was no longer significant after covariate adjustment (F(4,501) = 1.64,
p = 0.16).
Table 4
Estimated regression coefficients for the effects of TBR1 genotype, MSDPand TBR1 by MSDP interaction after adjustment for confounding
Conduct problems (7–9 years) | 0.84 (1.07) | 0.43 | −0.26 (0.88) | 0.77 | 6.47 (1.86) | <0.001 |
Conduct problems (15–16 years) | −0.61 (1.19) | 0.61 | 0.83 (0.97) | 0.40 | 6.23 (2.04) | < 0.005 |
Property/violent offences (14–25 years) | −0.91 (0.47) | 0.05 | 0.65 (0.21) | < 0.005 | 1.24 (0.54) | 0.02 |
Arrests/convictions (16–25 years) | −0.32 (0.42) | 0.45 | 0.76 (0.24) | < 0.005 | 2.21 (0.49) | < 0.001 |
The effect of frequency of smoking
To take account of variations in MSDP the data were re-analysed using an ordered categorical measure of MSDP: non-smoker; 1–9 cigarettes per day; 10+ cigarettes per day. This analysis produced the same overall pattern of findings as those reported in Tables
3 and
4. After covariate adjustment the multivariate regression model again showed a significant main effect for MSDP (F(4,501) = 2.51,
p = 0.04) and a significant
TBR1 by MSDP interaction (F(4,501) = 2.85,
p = 0.02).
The effect of gender
To examine the extent to which the findings may be moderated by gender, the analyses were extended to include gender as a further factor and tests of gender x TBR1 x MSDP interaction were conducted. No significant three-way interactions were found, suggesting that the TBR1 x MSDP interaction was evident for both males and females.
The effect of ethnic stratification
To examine the implications of ethnic stratification, the data were re-analysed including an additional 91 participants with data on TBR1, MSDP and behaviour outcomes who were of non-European (Maori or Pacific Island, by self-definition) ethnic origin. The fitted regression models were extended to incorporate ethnicity as a factor and to test for ethnicity by TBR1 by MSDP interactions. While the analysis lacked statistical power to draw strong conclusions, for all outcomes the TBR1 by MSDP interaction appeared to be somewhat weaker for non-European sample members. This was reflected in the multivariate regression model which showed that, after covariate adjustment, there was a significant ethnicity by TBR1 by MSDP interaction (F(4,577) = 2.62, p = 0.034), in addition to the existing TBR1 by MSDP interaction (F(4,577) = 3.03, p = 0.017).
Collectively the results in Tables
3 and
4 and the additional analyses reported above confirm the presence of a robust and persistent
TBR1 by MSDP interaction in which MSDP had greater effects on the risks of subsequent antisocial behaviours for those with the “other” genotype who carried at least one copy of the
TBR1 minor allele when compared with those who were homozygous for the major allele. These findings held after: a) control for covariate factors; b) variation in the measurement of MSDP; c) tests of gender interaction. This interaction appeared to be stronger amongst those of European ancestry.
Linkage disequilibrium with nearby single nucleotide polymorphisms
Single nucleotide polymorphisms (SNPs) in our cohort had been genotyped as part of a genome wide association study [
78]. We examined all SNPs within 150 k bases (kb) of
TBR1 to see if any of them could explain the effects of the microsatellite. Three SNPs located 37-64 kb upstream of the gene, rs3769963, rs1116173 and rs6727917, are in at least partial linkage disequilibrium with the microsatellite polymorphism. We analysed these SNPs individually and jointly with microsatellite genotype. Individually all three SNPs showed evidence of a gene x MSDP interaction on childhood conduct problems (data not shown). Of the three, rs3769963 showed the strongest evidence of interaction and produced results that were closest to those for the microsatellite. However, in all cases the R squared values for these models (0.032-0.042) were substantially less than the R-squared value for the microsatellite model (0.058). Also, rs3769963 is located 65 kb upstream of the
TBR1 gene’s transcriptional start site, and while this is not inconsistent with a regulatory role, its level of LD with the promoter microsatellite polymorphism was 2.6 × higher than that of any other SNP within 150 kb of
TBR1. In contrast, the levels of LD between the microsatellite and its neighbouring SNPs were very low (R-squared < 0.0005 and D’ < 0.2).
Discussion
Our main finding was that the minor allele of the AC12-13 microsatellite in the upstream promoter region of the TBR1 gene interacts with MSDP to increase risk of antisocial behaviour, a putative gene by environment (G x E) interaction which remained significant after adjustment for potentially confounding social and contextual factors, gender and frequency of smoking. We found some evidence that the interaction may be stronger in individuals of European descent, but the number of non-Europeans in our cohort was too small to rule out a stochastic effect.
We also report results for two intronic repeats in the
AP-2β transcription factor gene, and a GGN microsatellite in the
AR gene’s first exon. All of these have previously been associated with personality or behaviour-related phenotypes [
24,
54‐
61]. Our finding that the most common allele of the
AR microsatellite, GGN
23, was associated with fewer anxiety-related symptoms was consistent with previous work [
24], though in our study this wasn’t significant following correction for multiple hypotheses. GGN
23, elsewhere referred to as GGC
16 [
54], is the shorter of the two most common alleles and has been associated with reduced [
51] and increased [
52] activation of androgen receptor protein in different systems. Our only nominally significant results for the two previously studied
AP-2β repeats were that carriers of the minor allele of the CAAA
4-5 microsatellite polymorphism scored lower on psychotic symptoms at age 18 and higher on scholastic ability at age 13. The former was unexpected in view of a previous study reporting higher rates of anxiety and related traits in carriers of this allele among 137 Caucasians living in Sweden [
60]. However, none of our findings in relation to this repeat were significant following Bonferroni correction.
We note that evidence for common genetic risk factors for many psychiatric, substance abuse and conduct disorder phenotypes [
79] suggests that the lack of consistent associations in our data may indicate false positives. This argument does not apply to the
TBR1 interaction, which did replicate across different ages and measures, but because this study is the first to link the
TBR1 gene with antisocial behaviour, or with the effects of MSDP, the results should be treated with caution until replication can be attempted in another cohort. This point has been highlighted by a meta-analysis showing that first reports of gene-environment interactions are often not confirmed by studies attempting replication, presumably due to under-reporting of negative results [
80]. Several considerations do, however, support the validity of our results at this stage. Firstly, our interaction hypothesis emerged from a coincidence of rare interests in conserved microsatellites, their influence on the genetics of behaviour, and the effects of MSDP. Previous attempts at finding the interaction are therefore very unlikely, and we have presented all of our negative results here. Furthermore, our sample size is substantial by the standards of G x E interaction studies [
80], and the P-value of the interaction we observed also compares favourably with previous reports of G x E interaction involving MSDP and behavioural or cognitive phenotypes [
81‐
88]. In this field, high quality, multi-method phenotype measurement similar to that used for this study is usually performed. Along with adjustment for confounding factors this provides some compensation for sample sizes being smaller than most contemporary complex trait GWAS, which are thought to be limited by insufficiently detailed phenotyping [
89].
The association between
TBR1, MSDP and antisocial behaviour may be due to modification of the effects of tobacco exposure on embryonic development by
TBR1 genotype. G x E interaction involving MSDP has been demonstrated directly in knockout mice [
90], and both MSDP and microsatellite polymorphism have been linked to epigenetic modifications in offspring [
12,
91,
92], However, the Genotype-Tissue Expression Project has so far not identified expression quantitative trait loci for
TBR1 [
93]. An alternative to the G x E interaction explanation is suggested by the likely existence of genetic factors which predispose mothers to both pregnancy smoking and antisocial behaviour [
94]. An important role for gene-gene interaction (epistasis) would be consistent with large scale studies showing that siblings who are differentially exposed to prenatal tobacco are similarly prone to antisocial behaviour [
95,
96]. Indeed, it has been suggested that promoter-associated and exonic microsatellites may be particularly likely to participate in epistasis due to their interactions with
trans-acting factors [
11].
SNPs associated with
TBR1 have not been identified by GWAS of antisocial behaviour or related phenotypes [
5‐
8]. Nevertheless, the possibility that effects of the microsatellite polymorphism were not detectable by these GWAS is supported by its very low level of LD with neighbouring SNPs, and by the absence of its effect on antisocial behaviour in individuals whose mothers didn’t smoke during pregnancy. The association we report, and the extremely high level of conservation of the
TBR1 microsatellite [
34], suggest that its putative functional importance deserves further testing. One possible approach would be to investigate the effect on transcriptional frequency of each allele using reporter plasmids in cultured neuronal cells [
97]. Such assays have been used in previous studies to show that changes in repeat number of promoter-associated poly-AC microsatellites can affect frequency of transcription [
98,
99]. However it is notable that they don’t necessarily provide an accurate model of gene expression in developing brain, and functional studies have shown cell type dependent microsatellite length effects, in one case with opposite effects in two different cell types [
100].
Acknowledgements
We thank John Pearson (University of Otago, Christchurch, NZ) for help with SNP analysis and also Sterling Sawaya (University of Colorado, USA), who led a study of conservation of microsatellites among mammalian species [
40], and suggested the
TBR1 microsatellite as a candidate genetic factor for psychological traits.