Replication of linkage at chromosome 20p13 and identification of suggestive sex-differential risk loci for autism spectrum disorder

Autism spectrum disorders (ASDs) are a group of neurodevelopmental conditions characterized by severe social impairment that affect 1 in 88 individuals [1]. Genetic factors have long been known to contribute significantly to ASD risk based on twin studies [2], the sibling recurrence risk [3, 4], and elevated rates of comorbid ASD in populations with a wide variety of monogenic syndromes such as Fragile X or Timothy Syndrome [5, 6]. ASDs are also known to present heterogeneously across the population of affected individuals, and results from recent genetic studies strongly suggest that genetic risk factors for ASD are similarly diverse. Copy number variant (CNV) and exome sequencing studies of sporadic ASD cases from single-incidence (“simplex”) families have found numerous novel, de novo risk variants [7‐13], and no significant signal for rare inherited variation. Estimates based on these findings project that approximately 1,000 genes are likely to contribute to ASD etiology.

While a highly productive approach for gene discovery, the study of simplex families is designed to identify mostly the non-inherited genetic component of ASD risk: rare variants resulting from de novo mutations, in which variants arise in the germ cell and are not carried by the mother or father. However, evidence of high heritability for ASD [14], high sibling recurrence risk [3, 4], and aggregation of subthreshold ASD-like phenotypes in families [15‐18] suggest that inherited genetic variation also plays a significant role in ASD etiology.

Additionally, while germline mutations, potentially shared between affected siblings, may also plausibly affect ASD risk in multiplex families, current evidence suggests that rare de novo CNV events are more prevalent among sporadic cases than cases from multiple-incidence (“multiplex”) families [13]. Largely, however, the specifics of the genetic architecture of ASDs that differ between simplex and multiplex families are unknown. Therefore, studies of familial transmission to identify regions of genetic linkage in multiplex families remain an important approach to identifying predisposing genes.

Another important clue to ASD etiology lies in its consistently male-biased prevalence [19]. There is an approximately 4:1 male bias, a phenomenon that is likely driven, or at least influenced, by the actions of sex-specific biological factors, such as sex chromosomes or steroid hormones that potentiate and attenuate ASD risk in males and females, respectively [20]. Indeed, several ASD and intellectual disability risk genes have been identified on the X chromosome [6, 21], including FMR1[22], NLGN4X, and NLGN3[23], demonstrating that in some cases ASD may be X-linked. However, the proportion of ASD cases currently attributable to X-linked variants remains insufficient to account for the degree of male bias observed in ASD prevalence. The notion of female protective factors on a broader scale is supported by the observation of an increased proportion of autistic females relative to males carrying variants of large effect size, such as large CNVs or deleterious single nucleotide variants (SNVs) [8, 11, 13]. However, with few exceptions [24], it is unknown which specific autosomal risk variants are differentially penetrant by sex, thus contributing to the sex bias in ASD prevalence. For example, a greater number of variants may be associated with ASD risk in males as compared to the number of variants that also, or specifically, confer risk to females. The discovery of such sex-differential risk loci would provide genetic clues for investigation of the biological mechanisms driving the ASD male bias. However, because these signals are likely masked by heterogeneity within sex-mixed cohorts, stratification of multiplex ASD family cohorts by proband sex may facilitate the identification of novel, sex-differential loci harboring inherited risk variants.

Most previous linkage studies of ASD have used relatively small samples (<350 families) and markers with coarse resolution and incomplete information [25‐31], and there has been little agreement between studies in the reported findings. Furthermore, the larger studies [32, 33] combine subjects drawn from several diverse populations with different ascertainment schemes. Multiplex family samples ranging from 109 to 753 families from the Autism Genetics Resource Exchange (AGRE) cohort have been previously tested for linkage [25‐28, 33, 34], and the linked risk loci reported by these analyses, with a few exceptions [25, 28], show little agreement. This may be due to genetic heterogeneity, small sample sizes, or sparse marker coverage.

In this study, we used a pruned set of single nucleotide polymorphism (SNP) markers providing nearly complete linkage information on all autosomes and the X chromosome (information content greater than 0.976 for 99.5% of regions covered by SNP genotypes) for our analyses in the largest available AGRE sample. Specifically, we performed non-parametric linkage testing in three stages to: 1) identify novel risk loci in a discovery sample of 847 multiplex families; 2) confirm loci identified in this discovery sample, or reported by earlier analyses of AGRE samples, in an extended sample of 1,008 multiplex families; and 3) test for formal replication of genome-wide significant linkage signals in the independent portion of families not tested in the corresponding discovery studies (Figure 1).

For all stages, we also applied a stratification approach used previously [25, 28, 32] to identify sex-differential loci, now using a sample three times as large as the original and a panel of markers that are substantially more informative [28]. We then used the family-based association transmission disequilibrium test (TDT) on genotyped and imputed SNPs within linked regions to identify common variants conferring increased risk for ASD. With the addition of 343 AGRE families beyond those tested in the most recent linkage analysis using subjects from AGRE [33], this comprises the largest linkage study for ASD, and sex-differential risk, of families from a single, uniformly ascertained cohort. This also allows us to attempt replication of previous linkage findings.

By testing the largest available, uniformly ascertained ASD family sample (n = 1,008 multiplex families) for linkage, we aimed to identify novel ASD risk loci not identified by the previous, smaller analyses of AGRE samples, and/or to confirm the loci reported by these analyses. As genotype data were collected in two stages, we use the stage 1 sample to identify linked and sex-differential risk loci genome-wide, and the combined sample (union of stages 1 and 2) to evaluate identified loci from stage 1 and earlier reports (discovery studies) for confirmation of linkage in the largest available AGRE genotype data set.

To identify and support genomic loci likely to contain variants contributing to ASD risk in multiplex families from the AGRE collection, we performed linkage analyses in all and sex-stratified subsets of multiplex families followed by targeted association testing in 1,521 families (1,008 multiplex) from the AGRE cohort. The strongest linkage signal that we identified was on chromosome 20p13, which exceeded a LOD score of 3.0 in our combined sample of 1,008 multiplex families. At 20p13, we also replicated in an independent sample from the same AGRE cohort a previous report of significant linkage at this locus [33]. Analyses of sex-defined family subgroups and randomization testing for signals from these subgroups identified a locus at chromosome 1p31.3 that showed significant linkage in the MO subgroup in both the stage 1 and combined samples, and loci at chromosomes 8p21.2 and 8p12 that showed significant linkage in the FC subgroup only in the stage 1 sample. No genotyped or imputed common SNPs within any linked region proved to be significantly associated with ASD, an observation consistent with a model where the influence of multiple loci of very small effect size, or of rare variants, contributes to ASD.

The linkage signal at 20p13 is especially noteworthy as it was the most significant signal that we observed in our combined sample with a LOD_All.Com of 3.02, and it also meets the criterion for an independent replication of genome-wide significant linkage [43]. In addition, as both this study and the study by Weiss and colleagues [33] were conducted using the AGRE collection, a sample that has been uniformly ascertained and evaluated over time, this is a clear replication. A 2-LOD interval from the combined sample peak spans such potential risk candidate genes as NRSN2 [GenBank:NM_024958] (0.3-LOD drop from peak), a gene expressed throughout the cerebral cortex, thalamus, hypothalamus, and in Purkinje cells [58], and CSNK2A1 [GenBank:NG_011970] (0.5-LOD drop from peak), which encodes a protein involved in the regulation of circadian rhythms [59].

For nearly all linked regions identified in the discovery samples, including 20p13, LOD scores decreased when analyzed in the combined sample despite the increased sample size, consistent with previous observations of genetic heterogeneity [60‐62]. However, it is interesting to note that those signals that were confirmed or replicated in the combined or independent sample were initially observed by tests of relatively large discovery samples. This is because, like association testing, a very large number of small pedigrees are critical to identify robust linkage signals [63]. In smaller samples, signal fluctuations between analyses are caused by extreme heterogeneity of genetic risk loci for ASD [60, 61, 64], such that each analysis identifies different, but potentially true, risk loci. Since linkage peaks represent deviations from expected proportions of affected family members sharing two, one, or zero alleles identical by descent at a particular locus, linkage analyses for complex traits are particularly sensitive to the composition of families included in any one test. If, as projected, there are close to 1,000 risk genes for ASD [12], then the chances that a sufficient proportion of families in current sample sizes share the same risk locus are low.

We recognize that the power we have to detect linkage in our analyses is for complete linkage with loci of modest genetic effects on ASD risk [63], and so it is not surprising that we do not observe genome-wide significant linkage in either the stage 1 or extended, combined stage samples, and that most suggestive signals fail to increase in the extended sample. However, we emphasize that only now are we approaching the sample sizes necessary to detect significant linkage in the face of locus heterogeneity in cohorts comprised of small pedigrees, and that any possible reduction of locus heterogeneity (for example, by testing families ascertained by the same cohort as is done here), will be key to identify replicable linkage signals.

Rational stratification of cohorts into subgroups based on a shared trait may also facilitate the discovery of risk loci by increasing the relative homogeneity of specific genetic risk factor(s) in the subgroup. Here, we stratified our sample by the sex of the probands within each family to identify loci with sex-differential relationships with ASD risk, and found significant risk loci at chromosome 1p31.3 (MO), 8p21.2 (FC), and 8p12 (FC). At 1p31.3 and 8p12, linkage signals from the MO and FC subsets, respectively, were significantly stronger than the signal from the full, non-stratified cohort, suggesting that this sex-based stratification approach can indeed reduce the genetic heterogeneity within each subgroup that would otherwise obscure signals at these loci.

We note that these loci should not be interpreted as simply sex-specific. Although we cannot say that the MO group is perfectly restricted to families who carry solely male-specific variants and therefore only have, and would ever have, affected male children, we do assume that the MO subgroup is substantially enriched for families who carry risk variants that are more penetrant in males. Thus, signals specific to the MO subset are more likely to be male-specific. In contrast, since the FC group includes affected brothers of autistic girls, FC signals are best interpreted not as female-specific, but as female-affecting risk loci, in accord with a hypothesis that only a subset of ASD risk loci are penetrant in females [7, 9‐13, 65].

The significant MO and FC signals identified here implicate regions containing promising candidate genes that warrant further exploration by targeted re-sequencing (for example, [66‐69]). The MO peak at 1p31.3 is located directly over NFIA [GenBank:NG_011787], whose gene product has transcription factor activity and has been implicated in central nervous system development [55, 56]. Rare deletions encompassing this gene have been identified in subjects with ASD [70], as well as de novo mutations [7]. The FC peak at 8p21.2 spans several candidates, including STC1 [GenBank:NG_029711] (1.1-LOD drop from peak), which encodes a glycoprotein regulated by calcium that may act to protect neurons from ischemia and hypoxia [71], and neurofilament genes NEFM [GenBank:NG_008388] and NEFL [GenBank:NG_008492] (both 1.1-LOD drop from peak) whose products likely function in transport to neuronal projections [72]. Potentially relevant to sex-differential risk, GNRH1 [GenBank:NG_016457] (0.7-LOD drop from peak) is also located within this linkage region, and mutations in this gene are likely to affect gonadal function [73], perhaps differentially modulating downstream manifestation of ASD risk factors in males and females. The neighboring FC peak at chromosome 8p12 is located directly over NRG1 [GenBank:NG_012005] (0-LOD drop from peak), a known schizophrenia risk gene [74, 75].

A similar stratification approach for the identification of male-specific and female-affecting ASD risk loci has been successfully applied previously by Stone and colleagues [28] to an early iteration of the AGRE cohort, as well as by Szatmari and colleagues [32] to a sample that included a subset of AGRE families, and Lamb and colleagues [30] and Schellenberg and colleagues [29] to other ASD family samples. No convergence in linkage signals was observed across these studies. We predicted that our ability to identify novel, sex-differential ASD risk loci in the present study would be aided by a greater than three-fold increase in subjects exclusively from AGRE as compared to those used by Stone and colleagues [28] and by the increase in coverage afforded by dense SNP data in lieu of several hundred microsatellite markers. However, the results from our sex-stratified analyses do not reach genome-wide significance and also do not align with findings from earlier sex stratification analyses, including Stone and colleagues [28] who reported a linkage peak at 17q11-q21 in MO families that was subsequently replicated [25]. This variability between studies is again consistent with extreme risk locus heterogeneity in ASD [76, 77], with each analysis of a different combination of ASD families identifying different linked regions.

Under the assumption of genetic heterogeneity, for complex traits such as ASD, it is possible that the various loci identified by analyses of different family sets flag true sites for ASD risk in a proportion of the families tested. To pursue this, linked loci will need to be investigated more closely to identify the precise variants that effectively increase ASD risk. This has so far proven challenging, as association testing of densely mapped common SNPs within linkage peaks has failed to definitively identify risk variants, both in the present study and in previous work [33].

Although our analyses are likely underpowered to identify regionally significant associations with common variants of small effect size, it is alternatively possible that rare variants, not explicitly tested here, contribute to the heritable component of ASD risk. For example, rare variants, shared between siblings but private to each nuclear family, may cluster in the same gene or set of genes. Since rare variants are less likely to be tagged by common SNPs, they should be more readily localized by allele-agnostic linkage analyses than association testing. As in gene discovery studies of sporadic ASD cases, sequencing of functional genomic features in linked regions will be necessary to identify rare variants and evaluate their role in familial ASD risk. In either case, larger family-based cohorts will be needed to improve power.

We conclude that the use of linkage analyses in multiplex family cohorts has complementary utility to genome-wide association studies for the investigation of the familial, inherited contribution to ASD risk. This is especially the case in the context of rare variants in human disease (for example, [78, 79]). Additionally, the use of a sex stratification approach facilitates the identification of risk loci that are differentially associated with ASD in families with autistic sons versus daughters. However, further work is needed to determine which gene(s) or genetic features within linked regions, especially at chromosome 20p13, replicated here, harbor the variants responsible for increasing familial and sex-differential genetic risk for ASD. Exploring this in detail via targeted sequencing in large cohorts will be necessary to elucidate the common versus rare genetic contributions to ASD.