Combined evidence from evolutionary, population-genetic and disease studies supports strong association between germline hypomethylation and selective structural mutability. Genome-wide, both relative and attributable risks of structural mutations due to methylation deserts are at least comparable to the corresponding statistical risks due to LCR-mediated NAHR. Our results show that 23% of human-specific evolutionary rearrangements are associated with hypomethylation. Methylation deserts comprise a total of 1% of the genomic sequence and contain about 10% of the 522 submicroscopic human-specific structural rearrangements identified by primate genome comparisons.
The evolutionary findings are generally consistent with the results of analyses of CNVs in the human population. Our analysis reveals a two-fold genome-wide enrichment for deletions and duplications between DP-LCRs, the signature pattern of LCR-mediated NAHR. While the enrichment is statistically significant, the fraction of structural variation statistically attributable to NAHR is small, approximately 2.5%. We show that methylation deserts exhibit higher association with CNVs (~9%) and contain a disproportionately high fraction of CNVs that have high structural heterozygosity. The population-based analyses reveal less striking enrichment patterns than the evolutionary analyses. This may be explained by the fact that population based studies were generally of lower resolution (array-based, unlike sequence-based evolutionary analyses), were limited to copy-number changes, and were biased against rare variants.
By demonstrating a higher association of structural mutability with hypomethylation than with NAHR, our results underscore the potential relative contribution of the role of microhomology-mediated break-induced repair in structural genomic instability 
which is consistent with replication based mechanisms such as FoSTeS 
, MMBIR 
, and serial replication slippage (SRS) 
rather than NAHR.
Our results are consistent with the concept of a structural selective “mutability profile”, an epigenomic phenotype marked by the variation in germline methylation levels along the genome. Three questions regarding this mutability are of particular interest: heritability, mechanism, and evolution.
First, does inter-individual variation in methylation-associated selective mutability profiles exist and if it does, is it heritable? As a first step toward answering these questions, we have generated preliminary results tentatively suggesting that inter-individual variation in selective structural mutability may be associated with methylation deserts (Text S1
section 6 and Figure S17
The second open question is the mechanism behind the selective mutability profile. One conceivable mechanism is genetic variation in DNA-break inducing base-excision repair enzymes involved in germline-specific demethylation 
. Another possibility may involve unrepaired DNA breaks associated with active transcription because methylation deserts are highly transcribed in germline. Yet another possibility may be that transcription factors mediate structural rearrangements by bending chromatin, creating looping structures and DNA breaks, analogously to the role played by estrogen and androgen receptors in mediating structural instability in hormonally regulated tumors 
. One specific possibility opened by this model is that selective structural mutability may be affected by the cellular and organismal environment and may be controlled experimentally or even therapeutically.
Finally, assuming selective mutability profile variation is heritable, the question of its evolution arises (for a recent survey of the topic of “evolution of evolvability” see 
). Specifically, does selective mutability evolve mostly neutrally by random drift? If not, what may be the nature of selection pressure acting on it? Assuming that selection indeed plays a role, it is useful to consider the payoff (higher probability of developing a favorable mutation that ultimately becomes fixed in the population) and risk (of mutation causing disease). A selective mutability profile with excess mutability concentrated in the loci with low payoff/risk ratios would then be less likely to produce mutations that ultimately become fixed than a mutability profile with mutability concentrated in the loci with high payoff/risk ratios. The latter would therefore be favored by selection.
One testable corollary of this payoff/risk model is that de novo
mutations will tend to cause diseases related to the phenotypes that are under positive selection in the human population. Assuming that brain function is under selection in the human population, this corollary predicts high incidence of brain-related diseases such as schizophrenia, bipolar disorder, autism, epilepsy, developmental delay and cranial features due to rare and de novo
mutations. Our findings that the rare and de novo
CNV variants in the individuals suffering from these diseases indeed concentrate within methylation deserts is consistent with this corollary. These findings suggest a novel type of connection between evolution and human disease 
The payoff/risk model is also consistent with highly mutable loci being responsible for tissue-specific phenotypes. This is because a mutation in a locus regulating a tissue-specific phenotype may not confer much risk to other tissues. The enrichment within methylation deserts that we observed for genes with tissue-specific patterns of expression and for transcription factors involved in cellular differentiation is therefore consistent with this payoff/risk model.