The ongoing dissection of the genetic architecture of autistic spectrum disorder

As mentioned above, the MZ twin concordance rate for ASD is 70 to 90% and the DZ concordance is similar to the sibling recurrence rate of 0 to 10%, with the extreme values probably being representative of smaller sample sizes [4, 5, 39]. Given these numbers, there are two possible explanations for the data: a polygenic model involving several genes interacting with environmental factors, or de novo mutations with limited transmission. Most de novo mutations will occur in affected individuals and will not be transmitted; however, they can be transmitted from unaffected parents to multiple affected siblings as a result of maternal transmission of mutations on the X chromosome to male offspring [40, 41] or of gonadal mosaicism [24, 42], which together would result in higher sibling recurrence rates. Furthermore, it is also theoretically possible for transmission to occur from asymptomatic parents via autosomal recessive inheritance patterns or through mutations in imprinted genes. Although features of ASD have been identified in other syndromic epigenetic disorders [43], ASD-specific examples of highly penetrant genetic variants whose transmission was enabled by unaffected family members via imprinting have yet to be discovered.

Although perfect concordance between autistic MZ twins is not seen, this does not necessarily imply that environmental factors must play a role in producing or preventing the ASD phenotype. Discordance could occur as a result of variable X-linked inactivation [44], somatic mosaicism for a de novo mutation that occurred early in development in only one of the two developing zygotes, or autosomal dominance with variable expressivity, potentially resulting from stochastic events during embryogenesis.

To date, there have been three major genome-wide association studies (GWAS) aimed at identifying common variants that predispose to ASD. The initial study investigated 780 families (3101 subjects) and a second cohort with 1204 affected individuals (all of European ancestry) and identified significant association of six single-nucleotide polymorphisms (SNPs) with ASD to chromosome 5p14, an intergenic region between cadherins 9 and 10. These genes are involved in neuronal cell adhesion, and could be viewed as promising candidates [45]. The second study involved a combined linkage and association study of 1031 families. Although no promising linkage results were discovered, an association to an SNP near the gene SEMA5A was shown, and this was combined with expression analysis detailing decreased expression of this gene in the brains of patients with ASD [46]. The third study, comprising 1558 individuals found genome-wide evidence for association at the MACROD2 locus, although the authors had difficulty maintaining this signal in a replication cohort [47].

Notably, despite the robust sample sizes of these studies, none of these groups were able to replicate each other's results and each of the studies found modest effect sizes ranging from 0.55 to 1.2. Furthermore, these studies were carried out on modern platforms containing tagged SNPs for common CNVs, therefore it is also unlikely that such CNVs will explain an appreciable fraction of idiopathic ASD [48].

There have also been a countless number of candidate gene association studies performed in the past, which have suffered from the same curse of inconsistent replication. Despite these repeated efforts, there are still no common variants that could be inarguably associated with ASD in the same manner in which common variants have recently been identified for other complex traits (it should be noted that successful association studies have generally studied traits in which little selection occurs against the phenotype). If the genetic architecture of ASD is dominated by de novo mutations, followed by strong selection against the phenotype, it is clear that association studies will never identify clear signals, as these causal mutations will never have time to establish themselves in linkage disequilibrium with SNPs on a genotyping platform at a population-wide level. However. it is theoretically possible for an association to develop between a common variant and a region prone to a non-homologous allelic recombination that could lead to the ASD phenotype. Although abstract, this is not an unprecedented example [49‐54]. Overall, the lack of replication in these association studies would indicate that if common variants are still responsible for a significant fraction of idiopathic ASD, the number of common variants involved would have to be relatively high and their effect sizes would have to be considerably small, as has been recently shown[55].

There have also been numerous non-parametric linkage studies combining multiple small families, which have produced inconsistent results, with linkage peaks over the entire genome [56, 57]. The benefit of non-parametric linkage is that this method should also identify rare variants or transmitted mutations on different haplotypes at a small number of loci. A problem would arise under an architecture dominated by de novo mutations on a genome-wide level, as most sample cohorts would be biologically and genetically heterogeneous. Chromosomal regions of haplotype or allele sharing between siblings will be very large, therefore highly penetrant shared haplotypes between siblings could potentially be uniformly distributed throughout the genome in any ASD cohort, resulting in random noise during linkage analysis.

Obtaining global disease incidence data for ASD is difficult, as diagnostic disparities coupled with little clinical investigation of ASD in certain regions is a hindrance. However, studies have failed to conclusively reveal an uneven distribution of ASD from any ethnic background [3]. One of the many perplexing aspects of ASD is its relatively high frequency and high heritability, despite strong selection against the phenotype. This might be explained by strong selection for disease-predisposing alleles, as is seen for example in sickle-cell disease. However, it would be expected that such a strong selective pressure would be related to specific environments, which differ in different parts of the world, and so would predict an uneven global distribution, which is not the case for ASD. Under a polygenic model, this could only be reconciled by the existence of many allelic variants with low effect sizes that confer a susceptibility to ASD dating back to a common founder population for all of humanity. These variants would have had to be spread globally through small human populations as they migrated to various locations around the world. Information gleaned from the International Hapmap project has shown that a considerable amount of human genetic variation is common between several populations around the globe, indicating that a substantial proportion of variants date back to a common founder population [58]. Therefore, on the surface, this concept is plausible.

In small isolated founder populations, such as those likely to be found in many of our ancestral migratory movements, the genetic variation is reduced, with a concomitant increase in homozygosity at all loci, as founder effects coupled with genetic drift act to reduce overall genetic variation. This would posit that under a polygenic model, the allelic variants that produce the ASD phenotype must have combined more frequently to produce individuals that would be modernly diagnosed as autistic. In order for these variants to have remained globally at a similar frequency, there could not have been any selection acting against them, or any fixation by genetic drift during small population migrations and settlements, otherwise an uneven disease distribution would be presently observed.

In our modern environment, autistic individuals frequently survive into adulthood, yet strong negative selection still occurs against the phenotype, as evidenced by their low reproductive rates [59]. A gene pool is a reflection of successful reproductions between ancestral genomes that were composed of allelic variants whose collective actions conferred advantages to their historical possessors. The persistence of allelic variants under negative selection throughout evolutionary history in all human populations around the globe disagrees with evolution at the genetic level, and requires an explanation. Considering the low transmission of the ASD-conferring allelic variants, it is thus very unlikely that the remaining cases of idiopathic ASD could explain an even disease distribution around the globe.

It is already known that males are more prone to developing the autistic phenotype than females, by a ratio of approximately 4:1 [60]. It has been suggested that intrinsic differences between male and female brains may be a reason why one sex is more vulnerable to the ASD variants [61]. This explanation may apply equally well to a de novo hypothesis as a polygenic model. A previous study highlighted that in high-risk families, males are more likely than their female relatives to inherit the condition, despite their shared genetic background [12]. The selective advantage of diploidy, as a result of its protective mechanism against de novo mutations, is intuitive, and it would therefore follow that hemizygous males would be prone to the effect on the X chromosome. There is also evidence that the X chromosome has a higher proportion of genes involved in brain development and cognition than of autosomal genes [62, 63]. A recent review highlighted that 6 of 26 genes that show evidence of causal, monogenic dysfunction in non-syndromic ASD are located on the X chromosome [64], which might be considered too low a number to fully explain the skewed sex ratio, given the degree of scrutiny the X-chromosome has received [65]. However, it must be noted that defining the precise number of currently identified ASD alleles can be difficult, given the degree of variable expressivity in neurodevelopmental disorders and the degree of clinical overlap between ASD and other known syndromic conditions. Using another recent review, which developed a list containing every gene identified in ASD and all conditions that currently share a clinical overlap with features of ASD, the proportion of genes on the X chromosome was listed at 45 of 103 [66]. Whether females are resistant to contributing high-penetrance autosomal variants or whether a greater proportion of X-chromosome mutations will be discovered must be left for future research. It should also be noted that de novo mutations occurring on the X chromosome in females can be briefly transmitted to produce affected males, unless unfavorable X-chromosome inactivation produces a phenotype in females. However, it would be rare for the transmission to occur through multiple generations, as the penetrant mutation is unlikely to be transmitted by male offspring.

There have been many studies showing that first-degree relatives of individuals diagnosed with ASD can often show some of the broader phenotypic traits of autism [67, 68]. These studies were carefully conducted with proper controls, and were able to achieve statistical significance consistently.

A polygenic model would nicely account for this finding, proposing that allelic ASD variants for every autistic trait circulate in the population. The child diagnosed with ASD would be considered unfortunate to have acquired by chance a higher number of these allelic variants, which caused them to surpass the threshold in the phenotypic triad. The existence and transmission of harmful allelic variants through multiple generations has been discussed in a previous section. It has also been suggested that the allelic variants predisposing to the phenotypic triad are inherited separately [69]; however, this poses a perplexing question: how do the allelic variants predisposing to the phenotypic triad tend to co-occur with such high frequency? The existence of numerous tightly linked haplotypes located throughout the genome, containing clusters of ASD-conferring allelic variants that have evaded natural selection, is unlikely. There could be an unknown mechanism that leads to this higher than expected co-segregation, or perhaps even a synergistic relationship between them. However, mutations in several genes have already been discovered whose global brain expression results in global deficits in all three domains [13, 40, 41, 70], which questions the necessity of investigating this avenue.

Autosomal dominant mutations in genes that are involved in embryonic development can result in an incredibly diverse array of phenotypes. Mutations in the Sonic Hedgehog protein (SHH), for instance, result in holoprosencephaly, a condition whose behavioral and cognitive phenotype can be similar to ASD. Mutations in this gene can result in dramatically different phenotypes, from cyclopia in one family member to slight midline abnormalities in another [71]. If the phenotype of mutant SHH-associated holoprosencephaly were defined on narrowed thresholds, there would undoubtedly be pure cases of holoprosencephaly, with siblings showing the broader spectrum of holoprosencephalic features. The mechanism that leads to a phenotype with variable expressivity at most loci is unknown; it is therefore possible that other genetic variants and environmental factors may influence the ASD phenotype when such a gene is mutated.