Applicability of the Mutation-Selection Balance Model to Population Genetics of Heterozygous Protein-Truncating Variants in Humans

The landscape of regional missense mutational intolerance quantified from 125,748 exomes

Missense variants can have a range of functional impacts depending on factors such as the specific amino acid substitution and location within the gene. To interpret their deleteriousness, studies have sought to identify regions within genes that are specifically intolerant of missense variation 1-12 . Here, we leverage the patterns of rare missense variation in 125,748 individuals in the Genome Aggregation Database (gnomAD) 13 against a null mutational model to identify transcripts that display regional differences in missense constraint. Missense-depleted regions are enriched for ClinVar 14 pathogenic variants, de novo missense variants from individuals with neurodevelopmental disorders (NDDs) 15,16 , and complex trait heritability. Following ClinGen calibration recommendations for the ACMG/AMP guidelines, we establish that regions with less than 20% of their expected missense variation achieve moderate support for pathogenicity. We create a missense deleteriousness metric (MPC) that incorporates regional constraint and outperforms other deleteriousness scores at stratifying case and control de novo missense variation, with a strong enrichment in NDDs. These results provide additional tools to aid in missense variant interpretation.

Latent mutations in the ancestries of alleles under selection

We consider a single genetic locus with two alleles A1 and A2 in a large haploid population. The locus is subject to selection and two-way, or recurrent, mutation. Assuming the allele frequencies follow a Wright-Fisher diffusion and have reached stationarity, we describe the asymptotic behaviors of the conditional gene genealogy and the latent mutations of a sample with known allele counts, when the count n1 of allele A1 is fixed, and when either or both the sample size n and the selection strength |α| tend to infinity. Our study extends previous work under neutrality to the case of non-neutral rare alleles, asserting that when selection is not too strong relative to the sample size, even if it is strongly positive or strongly negative in the usual sense (α→-∞ or α→+∞), the number of latent mutations of the n1 copies of allele A1 follows the same distribution as the number of alleles in the Ewens sampling formula. On the other hand, very strong positive selection relative to the sample size leads to neutral gene genealogies with a single ancient latent mutation. We also demonstrate robustness of our asymptotic results against changing population sizes, when one of |α| or n is large.

Bayesian estimation of gene constraint from an evolutionary model with gene features

Measures of selective constraint on genes have been used for many applications including clinical interpretation of rare coding variants, disease gene discovery, and studies of genome evolution. However, widely-used metrics are severely underpowered at detecting constraint for the shortest ∼25% of genes, potentially causing important pathogenic mutations to be overlooked. We developed a framework combining a population genetics model with machine learning on gene features to enable accurate inference of an interpretable constraint metric, shet. Our estimates outperform existing metrics for prioritizing genes important for cell essentiality, human disease, and other phenotypes, especially for short genes. Our new estimates of selective constraint should have wide utility for characterizing genes relevant to human disease. Finally, our inference framework, GeneBayes, provides a flexible platform that can improve estimation of many gene-level properties, such as rare variant burden or gene expression differences.

A probabilistic graphical model for estimating selection coefficient of missense variants from human population sequence data

Accurately predicting the effect of missense variants is a central problem in interpretation of genomic variation. Commonly used computational methods does not capture the quantitative impact on fitness in populations. We developed MisFit to estimate missense fitness effect using biobank-scale human population genome data. MisFit jointly models the effect at molecular level ( d ) and population level (selection coefficient, s ), assuming that in the same gene, missense variants with similar d have similar s . MisFit is a probabilistic graphical model that integrates deep neural network components and population genetics models efficiently with inductive bias based on biological causality of variant effect. We trained it by maximizing probability of observed allele counts in 236,017 European individuals. We show that s is informative in predicting frequency across ancestries and consistent with the fraction of de novo mutations given s . Finally, MisFit outperforms previous methods in prioritizing missense variants in individuals with neurodevelopmental disorders.

A mutation rate model at the basepair resolution identifies the mutagenic effect of polymerase III transcription

De novo mutations occur at substantially different rates depending on genomic location, sequence context and DNA strand. The success of methods to estimate selection intensity, infer demographic history and map rare disease genes, depends strongly on assumptions about the local mutation rate. Here we present Roulette, a genome-wide mutation rate model at basepair resolution that incorporates known determinants of local mutation rate. Roulette is shown to be more accurate than existing models. We use Roulette to refine the estimates of population growth within Europe by incorporating the full range of human mutation rates. The analysis of significant deviations from the model predictions revealed a tenfold increase in mutation rate in nearly all genes transcribed by polymerase III (Pol III), suggesting a new mutagenic mechanism. We also detected an elevated mutation rate within transcription factor binding sites restricted to sites actively used in testis and residing in promoters.

A deep catalog of protein-coding variation in 985,830 individuals

Coding variants that have significant impact on function can provide insights into the biology of a gene but are typically rare in the population. Identifying and ascertaining the frequency of such rare variants requires very large sample sizes. Here, we present the largest catalog of human protein-coding variation to date, derived from exome sequencing of 985,830 individuals of diverse ancestry to serve as a rich resource for studying rare coding variants. Individuals of African, Admixed American, East Asian, Middle Eastern, and South Asian ancestry account for 20% of this Exome dataset. Our catalog of variants includes approximately 10.5 million missense (54% novel) and 1.1 million predicted loss-of-function (pLOF) variants (65% novel, 53% observed only once). We identified individuals with rare homozygous pLOF variants in 4,874 genes, and for 1,838 of these this work is the first to document at least one pLOF homozygote. Additional insights from the RGC-ME dataset include 1) improved estimates of selection against heterozygous loss-of-function and identification of 3,459 genes intolerant to loss-of-function, 83 of which were previously assessed as tolerant to loss-of-function and 1,241 that lack disease annotations; 2) identification of regions depleted of missense variation in 457 genes that are tolerant to loss-of-function; 3) functional interpretation for 10,708 variants of unknown or conflicting significance reported in ClinVar as cryptic splice sites using splicing score thresholds based on empirical variant deleteriousness scores derived from RGC-ME; and 4) an observation that approximately 3% of sequenced individuals carry a clinically actionable genetic variant in the ACMG SF 3.1 list of genes. We make this important resource of coding variation available to the public through a variant allele frequency browser. We anticipate that this report and the RGC-ME dataset will serve as a valuable reference for understanding rare coding variation and help advance precision medicine efforts.

Scaling the discrete-time Wright-Fisher model to biobank-scale datasets

The discrete-time Wright-Fisher (DTWF) model and its diffusion limit are central to population genetics. These models can describe the forward-in-time evolution of allele frequencies in a population resulting from genetic drift, mutation, and selection. Computing likelihoods under the diffusion process is feasible, but the diffusion approximation breaks down for large samples or in the presence of strong selection. Existing methods for computing likelihoods under the DTWF model do not scale to current exome sequencing sample sizes in the hundreds of thousands. Here, we present a scalable algorithm that approximates the DTWF model with provably bounded error. Our approach relies on two key observations about the DTWF model. The first is that transition probabilities under the model are approximately sparse. The second is that transition distributions for similar starting allele frequencies are extremely close as distributions. Together, these observations enable approximate matrix-vector multiplication in linear (as opposed to the usual quadratic) time. We prove similar properties for Hypergeometric distributions, enabling fast computation of likelihoods for subsamples of the population. We show theoretically and in practice that this approximation is highly accurate and can scale to population sizes in the tens of millions, paving the way for rigorous biobank-scale inference. Finally, we use our results to estimate the impact of larger samples on estimating selection coefficients for loss-of-function variants. We find that increasing sample sizes beyond existing large exome sequencing cohorts will provide essentially no additional information except for genes with the most extreme fitness effects.

DECIPHER: Improving Genetic Diagnosis Through Dynamic Integration of Genomic and Clinical Data

DECIPHER (Database of Genomic Variation and Phenotype in Humans Using Ensembl Resources) shares candidate diagnostic variants and phenotypic data from patients with genetic disorders to facilitate research and improve the diagnosis, management, and therapy of rare diseases. The platform sits at the boundary between genomic research and the clinical community. DECIPHER aims to ensure that the most up-to-date data are made rapidly available within its interpretation interfaces to improve clinical care. Newly integrated cardiac case-control data that provide evidence of gene-disease associations and inform variant interpretation exemplify this mission. New research resources are presented in a format optimized for use by a broad range of professionals supporting the delivery of genomic medicine. The interfaces within DECIPHER integrate and contextualize variant and phenotypic data, helping to determine a robust clinico-molecular diagnosis for rare-disease patients, which combines both variant classification and clinical fit. DECIPHER supports discovery research, connecting individuals within the rare-disease community to pursue hypothesis-driven research.

Scaling the Discrete-time Wright Fisher model to biobank-scale datasets

The Discrete-Time Wright Fisher (DTWF) model and its large population diffusion limit are central to population genetics. These models describe the forward-in-time evolution of the frequency of an allele in a population and can include the fundamental forces of genetic drift, mutation, and selection. Computing like-lihoods under the diffusion process is feasible, but the diffusion approximation breaks down for large sample sizes or in the presence of strong selection. Unfortunately, existing methods for computing likelihoods under the DTWF model do not scale to current exome sequencing sample sizes in the hundreds of thousands. Here we present an algorithm that approximates the DTWF model with provably bounded error and runs in time linear in the size of the population. Our approach relies on two key observations about Binomial distributions. The first is that Binomial distributions are approximately sparse. The second is that Binomial distributions with similar success probabilities are extremely close as distributions, allowing us to approximate the DTWF Markov transition matrix as a very low rank matrix. Together, these observations enable matrix-vector multiplication in linear (as opposed to the usual quadratic) time. We prove similar properties for Hypergeometric distributions, enabling fast computation of likelihoods for subsamples of the population. We show theoretically and in practice that this approximation is highly accurate and can scale to population sizes in the billions, paving the way for rigorous biobank-scale population genetic inference. Finally, we use our results to estimate how increasing sample sizes will improve the estimation of selection coefficients acting on loss-of-function variants. We find that increasing sample sizes beyond existing large exome sequencing cohorts will provide essentially no additional information except for genes with the most extreme fitness effects.

Inferring the mode and strength of ongoing selection

Genome sequence data are no longer scarce. The UK Biobank alone comprises 200,000 individual genomes, with more on the way, leading the field of human genetics toward sequencing entire populations. Within the next decades, other model organisms will follow suit, especially domesticated species such as crops and livestock. Having sequences from most individuals in a population will present new challenges for using these data to improve health and agriculture in the pursuit of a sustainable future. Existing population genetic methods are designed to model hundreds of randomly sampled sequences but are not optimized for extracting the information contained in the larger and richer data sets that are beginning to emerge, with thousands of closely related individuals. Here we develop a new method called trio-based inference of dominance and selection (TIDES) that uses data from tens of thousands of family trios to make inferences about natural selection acting in a single generation. TIDES further improves on the state of the art by making no assumptions regarding demography, linkage, or dominance. We discuss how our method paves the way for studying natural selection from new angles.

Relating pathogenic loss-of-function mutations in humans to their evolutionary fitness costs

Causal loss-of-function (LOF) variants for Mendelian and severe complex diseases are enriched in 'mutation intolerant' genes. We show how such observations can be interpreted in light of a model of mutation-selection balance and use the model to relate the pathogenic consequences of LOF mutations at present to their evolutionary fitness effects. To this end, we first infer posterior distributions for the fitness costs of LOF mutations in 17,318 autosomal and 679 X-linked genes from exome sequences in 56,855 individuals. Estimated fitness costs for the loss of a gene copy are typically above 1%; they tend to be largest for X-linked genes, whether or not they have a Y homolog, followed by autosomal genes and genes in the pseudoautosomal region. We compare inferred fitness effects for all possible de novo LOF mutations to those of de novo mutations identified in individuals diagnosed with one of six severe, complex diseases or developmental disorders. Probands carry an excess of mutations with estimated fitness effects above 10%; as we show by simulation, when sampled in the population, such highly deleterious mutations are typically only a couple of generations old. Moreover, the proportion of highly deleterious mutations carried by probands reflects the typical age of onset of the disease. The study design also has a discernible influence: a greater proportion of highly deleterious mutations is detected in pedigree than case-control studies, and for autism, in simplex than multiplex families and in female versus male probands. Thus, anchoring observations in human genetics to a population genetic model allows us to learn about the fitness effects of mutations identified by different mapping strategies and for different traits.

Evolution and dysfunction of human cognitive and social traits: A transcriptional regulation perspective

Evolutionary changes in brain and craniofacial development have endowed humans with unique cognitive and social skills, but also predisposed us to debilitating disorders in which these traits are disrupted. What are the developmental genetic underpinnings that connect the adaptive evolution of our cognition and sociality with the persistence of mental disorders with severe negative fitness effects? We argue that loss of function of genes involved in transcriptional regulation represents a crucial link between the evolution and dysfunction of human cognitive and social traits. The argument is based on the haploinsufficiency of many transcriptional regulator genes, which makes them particularly sensitive to loss-of-function mutations. We discuss how human brain and craniofacial traits evolved through partial loss of function (i.e. reduced expression) of these genes, a perspective compatible with the idea of human self-domestication. Moreover, we explain why selection against loss-of-function variants supports the view that mutation-selection-drift, rather than balancing selection, underlies the persistence of psychiatric disorders. Finally, we discuss testable predictions.

Extreme purifying selection against point mutations in the human genome

Large-scale genome sequencing has enabled the measurement of strong purifying selection in protein-coding genes. Here we describe a new method, called ExtRaINSIGHT, for measuring such selection in noncoding as well as coding regions of the human genome. ExtRaINSIGHT estimates the prevalence of “ultraselection” by the fractional depletion of rare single-nucleotide variants, after controlling for variation in mutation rates. Applying ExtRaINSIGHT to 71,702 whole genome sequences from gnomAD v3, we find abundant ultraselection in evolutionarily ancient miRNAs and neuronal protein-coding genes, as well as at splice sites. By contrast, we find much less ultraselection in other noncoding RNAs and transcription factor binding sites, and only modest levels in ultraconserved elements. We estimate that ~0.4–0.7% of the human genome is ultraselected, implying ~ 0.26–0.51 strongly deleterious mutations per generation. Overall, our study sheds new light on the genome-wide distribution of fitness effects by combining deep sequencing data and classical theory from population genetics.

Previous work has investigated selection in the coding genome, but it is not as well characterized in the non-coding genome. By analyzing rare variants in 70k genome sequences from gnomAD, the authors detect very strong purifying selection ("ultraselection”) across the human genome, finding it in some microRNAs and coding sequences but generally rare in regulatory sequences.

Reduced reproductive success is associated with selective constraint on human genes

Genome-wide sequencing of human populations has revealed substantial variation among genes in the intensity of purifying selection acting on damaging genetic variants¹. Although genes under the strongest selective constraint are highly enriched for associations with Mendelian disorders, most of these genes are not associated with disease and therefore the nature of the selection acting on them is not known². Here we show that genetic variants that damage these genes are associated with markedly reduced reproductive success, primarily owing to increased childlessness, with a stronger effect in males than in females. We present evidence that increased childlessness is probably mediated by genetically associated cognitive and behavioural traits, which may mean that male carriers are less likely to find reproductive partners. This reduction in reproductive success may account for 20% of purifying selection against heterozygous variants that ablate protein-coding genes. Although this genetic association may only account for a very minor fraction of the overall likelihood of being childless (less than 1%), especially when compared to more influential sociodemographic factors, it may influence how genes evolve over time.

Overcoming constraints on the detection of recessive selection in human genes from population frequency data

The identification of genes that evolve under recessive natural selection is a long-standing goal of population genetics research that has important applications to the discovery of genes associated with disease. We found that commonly used methods to evaluate selective constraint at the gene level are highly sensitive to genes under heterozygous selection but ubiquitously fail to detect recessively evolving genes. Additionally, more sophisticated likelihood-based methods designed to detect recessivity similarly lack power for a human gene of realistic length from current population sample sizes. However, extensive simulations suggested that recessive genes may be detectable in aggregate. Here, we offer a method informed by population genetics simulations designed to detect recessive purifying selection in gene sets. Applying this to empirical gene sets produced significant enrichments for strong recessive selection in genes previously inferred to be under recessive selection in a consanguineous cohort and in genes involved in autosomal recessive monogenic disorders.

Mutation saturation for fitness effects at human CpG sites

Whole exome sequences have now been collected for millions of humans, with the related goals of identifying pathogenic mutations in patients and establishing reference repositories of data from unaffected individuals. As a result, we are approaching an important limit, in which datasets are large enough that, in the absence of natural selection, every highly mutable site will have experienced at least one mutation in the genealogical history of the sample. Here, we focus on CpG sites that are methylated in the germline and experience mutations to T at an elevated rate of ~10-7 per site per generation; considering synonymous mutations in a sample of 390,000 individuals, ~ 99 % of such CpG sites harbor a C/T polymorphism. Methylated CpG sites provide a natural mutation saturation experiment for fitness effects: as we show, at nt sample sizes, not seeing a non-synonymous polymorphism is indicative of strong selection against that mutation. We rely on this idea in order to directly identify a subset of CpG transitions that are likely to be highly deleterious, including ~27 % of possible loss-of-function mutations, and up to 20 % of possible missense mutations, depending on the type of functional site in which they occur. Unlike methylated CpGs, most mutation types, with rates on the order of 10-8 or 10-9, remain very far from saturation. We discuss what these findings imply for interpreting the potential clinical relevance of mutations from their presence or absence in reference databases and for inferences about the fitness effects of new mutations.

Measuring intolerance to mutation in human genetics

In numerous applications, from working with animal models to mapping the genetic basis of human disease susceptibility, knowing whether a single disrupting mutation in a gene is likely to be deleterious is useful. With this goal in mind, a number of measures have been developed to identify genes in which protein-truncating variants (PTVs), or other types of mutations, are absent or kept at very low frequency in large population samples-genes that appear 'intolerant' to mutation. One measure in particular, the probability of being loss-of-function intolerant (pLI), has been widely adopted. This measure was designed to classify genes into three categories, null, recessive and haploinsufficient, on the basis of the contrast between observed and expected numbers of PTVs. Such population-genetic approaches can be useful in many applications. As we clarify, however, they reflect the strength of selection acting on heterozygotes and not dominance or haploinsufficiency.