Discussion
Our study combined high-quality genome sequencing and RNA-seq data for a 17 member, three-generation family, enabling us to investigate the role and interpretability of rare noncoding variants. In contrast to low-pass approaches, high-quality full-coverage genome sequencing and patterns of Mendelian segregation provided the ability to more confidently identify and genotype rare variants within the family. More importantly, the large number of children provided us with the ability to detect eQTLs caused by rare variants specific to the family. In contrast, the power of a design that includes many small families or trios would be reduced by the overall heterogeneity of causal rare variants in each family. A further advantage is that with matched cell type and population, we were able to compare family eQTLs to population eQTLs reported by the Geuvadis Consortium.18 We identified genes that exhibit larger eQTL effect sizes in the family than in the population and demonstrated that these family-specific eQTLs are enriched with rare regulatory variants, influence more evolutionarily constrained and central genes, and are potential contributors to risk of complex disease.
One limitation of the study is that we did not observe many more large-effect eQTLs in the family than expected by chance; high FDRs were observed for all categories of large-effect eQTLs. This could suggest that there is not an overabundance of large-effect eQTLs specific to the family. It might also simply reflect limited power or imperfect comparison of effect sizes between cohorts, given that we explored by varying quantification pipelines, discovery-panel sizes, and methodologies for selecting testable variants. However, the enriched properties we identified for large-effect family eQTLs appear to be robust to such limitations, and we highlight that although there might not be a strong excess of large-effect eQTLs, the relative degree of effect between the family and population conveys meaningful properties of family eQTLs. For instance, as the degree of effect size increased in the family relative to the population, we observed an increasing enrichment of rare and potentially regulatory variants. Furthermore, such large-effect eQTLs in the family exhibited increasing enrichment in ASE, implicating a heterozygous causal variant. Additionally, the enrichment of family eQTLs among constrained and central genes was most extreme for the subset of genes in which a rare and potentially regulatory variant could be identified. These observations fit with population-genetic expectation given that rare variants can influence more essential genes because of a reduced impact of purifying selection. Furthermore, this is in contrast to the general properties of population eQTL genes; for increasing effect sizes, they have previously been shown to be less constrained and less central.19 Taken together, these results implicate an important role of rare regulatory variants in large-effect eQTLs in the family.
We compared, in addition to gene expression, ASE and alternative splicing between the family and the population. As with gene expression, we observed enrichment in rare variants for large-effect ASE and sQTLs in the family. Furthermore, we observed that evolutionary constraint and distance to splice sites for rare splicing variants was significantly informative of large splicing effects in the family. With both large-effect eQTLs and large-effect sQTLs predicted by rare variants, this study highlights existing potential for routine integration of these variants in rare-variant association tests.
Ultimately, a principal goal in genome interpretation is to develop the ability to predict the impact of all variants, including those that are rare or novel. In our study, we were able to test the importance of diverse noncoding annotations for predicting the impact of noncoding variants on gene expression. For rare variants, we identified that evolutionary constraint coupled with distance to the TSS and epigenomic information was highly informative in predicting eQTLs. For common variants, such annotations did not provide comparable predictive power. The likely reason for this difference is that common variants, regardless of genomic annotation, are very likely to be neutral, whereas rare variants have a higher prior likelihood of functional impact that can be further informed by genomic annotation. Given that no previous analyses have had access to high-quality genomes and transcriptomes in a single large human family, this study provides data to support a much-needed framework for frequency-independent evaluation of genome interpretation for noncoding variants and suggests that the impact of many rare and causal noncoding variants might be easier to predict than expected.