The gene dose makes the disease

Abstract

A long-standing challenge in genomics has been to identify causal genes within rare copy-number variant regions that are intolerant to altered dosage. In this issue, Collins et al. perform a meta-analysis of almost a million individuals to identify dosage-sensitive segments and genes conferring risk for a range of disease phenotypes.

We are in a watershed moment in human genetics. Deep genotypic and detailed clinical data are becoming readily available across disease ascertainments and in unselected populations, allowing researchers to perform association studies of genes and phenotypes with ever-increasing statistical power. This is particularly exciting for the study of rare copy number variants (CNVs), which collectively involve deletion and duplication of thousands of potential dosage-sensitive genes and account for about 25% of individuals diagnosed with neurological and developmental disorders (Iyer & Girirajan, 2015). Historically, causal genes for CNVs associated with syndromic disorders have been mapped through rare chromosomal events, such as translocations to identify NSD1 in Sotos syndrome and atypical deletions to identify RAI1 in Smith-Magenis syndrome (Fig. 1A) (Iyer & Girirajan, 2015). Atypical deletions have also been used to identify several dosage-sensitive genes in Williams–Beuren syndrome (WBS), where each gene within the deletion interval contributes to distinct phenotypes of the disorder (Fig. 1B) (Iyer & Girirajan, 2015). However, contrary to initial claims, many rare CNVs are pleiotropic and thus an individual region can be associated with multiple nosologically discrete disorders, including developmental delay, autism, and schizophrenia. While mapping causal genes has been successful for CNVs leading to the same constellation of clinical features, gene discovery within pleiotropic CNVs has been particularly challenging. Therefore, large samples from multiple ascertainments are necessary to untangle CNVs with pleiotropic effects and to identify genes that drive specific phenotypes. In this issue of Cell, Collins and colleagues accrue data from approximately one million individuals in both disease and control cohorts to identify disease-associated dosage-sensitive regions, and fine-map clinical features to individual genes within these segments across the genome (Collins et al., 2022).

Figure 1. Mechanisms of pathogenicity of disease for dosage-sensitive genes within CNVs.

(A) Phenotypes observed in both deletions and duplications can be caused by a single gene, such as RAI1 in Smith-Magenis and Potocki-Lupski syndromes. (B) Multiple dosage-sensitive genes within a CNV region can also individually contribute to distinct phenotypes, as in Williams-Bueren syndrome. Individual genes could be either bidirectionally sensitive or uniquely haploinsufficient (gene B) or triplosensitive (gene A). (C) Phenotypes in CNV disorders can also result from interactions of multiple dosage-sensitive genes, as reported in model systems for 16p11.2 deletion. (D) Interactions of one or more dosage-sensitive genes with modifier variants in the genetic background, including both rare and common variants, can also modulate CNV phenotypes. (E) The expression of dosage-sensitive genes can be affected by parental imprinting, as in Prader-Willi and Angelman syndromes, where the parent of origin of the CNV allele affects the manifested phenotypes.

The authors first rigorously harmonized microarray CNV calls across 17 independent cohorts, standardized phenotypic annotations to derive 54 neurological and non-neurological features, and binned each cohort into seven meta-cohorts based on ascertainment, study design, sequencing platform, and processing pipelines. Although imperfect, this process creates a uniform dataset for relatively unbiased association studies. The authors assessed enrichment for deletions or duplications across 200-kilobase pair sliding windows of the genome and identified 163 genomic regions with significant disease associations, including 53 previously associated with disease. Many of these CNV regions were associated with more than one phenotype, and these regions with pleiotropic effects were enriched for genes under mutational constraint compared to CNVs associated with a single phenotype. This observation is also in line with previous studies where known dosage-sensitive genes such as EHMT1 and KANSL1 were shown to carry an excess of de novo single nucleotide variants (SNVs) in individuals with neurodevelopmental disorders (Coe et al., 2019). Further, an excess of pathogenic SNVs in affected individuals has also bolstered support for several candidate genes, including MAPK3 within the 16p11.2 region (Coe et al., 2019). Similarly, for highly penetrant CNV regions associated with neurological phenotypes, Collins et al. found that genes within deletions were enriched for de novo protein-truncating variants and duplicated genes were enriched for de novo missense mutations in children with related disorders. This observation supports the notion that heterozygous loss-of-function mutations recapitulate deletion phenotypes and, interestingly, suggests that missense mutations may mimic duplication phenotypes, such as heterozygous mutations or duplications encompassing MYH11 in the 16p13.11 region that increase the risk for thoracic aortic aneurysms (Kuang et al., 2011). Collins et al. also identified 12 genomic regions with significant disease association that did not overlap any known protein-coding exons, indicating potential pathogenicity mechanisms outside of gene dosage, such as structural alterations that disrupt regulatory regions.

The authors further examined enrichment of individual CNV genes towards disease and used a Bayesian fine-mapping algorithm to identify at least one candidate gene in 55% of the significant loci associated with disease. They not only confirmed previous associations for several canonical causal genes such as RAI1, but also found new associations for further study, such as GMEB2 with central nervous system defects. Interestingly, seven genes that were previously associated with disease through haploinsufficiency were also enriched for duplications, including ANKRD11 where haploinsufficiency leads to Cornelia de Lange syndrome and this study found its duplication to be associated with growth abnormalities. These results show that some CNV genes may be bidirectionally dosage-sensitive, which could explain the “mirror phenotypes” observed in reciprocal deletions and duplications of the same region, such as the opposing alterations in height and weight observed with CNVs in the distal 1q21.1 region (Owen et al., 2018). Some significant CNV regions also contained multiple candidate genes, indicating a contiguous gene model, as in WBS, or even potential additive interactions of genes within the region, a phenomenon that has been observed in model systems (Fig. 1B). Furthermore, for almost half of all significant regions, the authors were unable to nominate any candidate genes. This suggests mechanisms that limit detection of dosage-sensitive genes, such as synergistic effects of two or more genes that may not individually have significant effects (Fig. 1C). For example, complex interactions and oligogenic effects have been proposed for dosage-sensitive genes within the 16p11.2 deletion, which contribute to neuronal, behavioral, and craniofacial phenotypes in model systems (Iyer et al., 2018; McCammon et al., 2017). Additionally, the ultimate phenotypic outcome of dosage-sensitive genes may be modulated by variants in regulatory regions or elsewhere in the genetic background, mutations unmasking recessive effects of specific genes, or parental imprinting effects which can further obfuscate detection of dosage-sensitive genes (Fig. 1) (Albers et al., 2012; Iyer & Girirajan, 2015; Pizzo et al., 2021).

Moving beyond phenotype associations, using an empirical Bayesian approach combined with eight machine learning methods, Collins et al. modeled genome-wide dosage sensitivity and identified 2,987 haploinsufficient and 1,559 triplosensitive genes, including 911 bidirectionally dosage-sensitive genes, far more than that identified from past studies (Coe et al., 2019). The authors further explored the properties of dosage-sensitive genes by correlating genetic features with predicted dosage-sensitivity. They found that bidirectionally dosage-sensitive genes were mutationally constrained, and haploinsufficient and triplosensitive genes can be distinguished by their length, density of nearby genes, and nearby regulatory elements. While these preliminary findings are exciting, further work is needed to fully understand the causes of gene dosage-sensitivity.

Although no clinical therapies for CNV disorders currently exist, there has been encouraging work suggesting that therapies targeting specific CNV genes can rescue neurological phenotypes, and the candidate genes identified by Collins and colleagues represent a roster of new potential targets (Martin Lorenzo et al., 2021). Additionally, future work could build on the findings from Collins et al. by incorporating whole genome sequencing (WGS) for more accurate assessments of CNV breakpoints and including samples from diverse ancestries to account for varying genetic architecture predisposing to rearrangements. Further, phenotyping could move away from traditional disease definitions towards assessment of specific quantitative traits representing distinct phenotypic domains to associate with dosage-sensitive genes. Finally, integrating information from model systems including patient-derived cells, gene network analyses, and animal models with results from association studies is key to untangling the molecular effects of gene dosage alterations underpinning CNV disorders. Overall, Collins and colleagues pave the way for new research into elucidating the mechanisms behind CNV pathogenicity by prioritizing dosage-sensitive regions and candidate genes causal for human disease.