📄 evolutionary biology

Dynamics of mutators of arbitrary dominance in humans

This paper presents a population genetic model for mutator alleles with arbitrary dominance to analyze human germline mutation rate modifiers, finding that most known variants are consistent with purifying selection while suggesting that recessive mutators are likely underrepresented in current discovery efforts due to detection biases.

Original authors: Saeidi, M., Sella, G., Przeworski, M., Milligan, W. R.

Published 2026-09-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Saeidi, M., Sella, G., Przeworski, M., Milligan, W. R.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Every living thing carries a hidden instruction manual written in DNA, a code that dictates how an organism grows, functions, and passes its traits to the next generation. Occasionally, typos occur in this manual as cells divide to create sperm or eggs. These typos, known as mutations, are the raw material for evolution, allowing species to adapt over millennia, but they are also the root cause of many heritable diseases. While the rate at which these errors happen is generally kept low by the body's repair mechanisms, it is not a fixed constant. Recent discoveries have shown that some people carry specific genetic variants that act as "mutators," essentially breaking the brakes on the error-correction system and causing a much higher number of new mutations to appear in their children. Because these extra mutations are usually harmful, natural selection should theoretically keep such mutator genes rare, yet scientists have found several examples in humans that persist in the population. The question remains: why do these harmful variants exist, and what does their presence tell us about the hidden rules governing how mutation rates evolve?

A team of researchers at Columbia University set out to answer this by building a mathematical model to track the fate of these mutator genes in human populations. They focused on a specific trait of these genes: how they behave when a person carries just one copy versus two copies. In genetics, this is known as dominance. Some harmful genes show their effects even if a person has only one copy, while others hide silently until a person inherits two copies. The researchers wanted to see if the known human mutators fit the pattern of being kept in check by natural selection, or if something else was allowing them to survive. They took the seven mutator genes already identified in humans—genes like XPC, MPG, POLE, and MUTYH, which are involved in repairing DNA or copying it—and ran simulations to predict how common these genes should be if the only force acting on them was the harm they caused to future generations.

The results offered a clear picture for most of the known mutators. For six of the seven genes, the observed frequency in the human population was well fit by the model, consistent with purifying selection arising solely from their effects on germline mutation rates, although the data was also consistent with a wide range of parameters. This suggests that these genes are indeed being kept in check by natural selection, which removes them from the population because they generate too many harmful errors. The model showed that the strength of this selection depends heavily on whether the mutator acts in a dominant or recessive way. If a mutator is dominant, meaning it causes problems even when a person has just one copy, it is quickly weeded out and stays very rare. If it is recessive, hiding its effects in people with only one copy, it can drift to higher frequencies before natural selection catches it. The data indicated that the mutators in genes like MPG and XPC likely act as recessive variants, which explains why they are found at low but detectable levels in the population.

One gene, however, refused to fit the model. A specific variant in the MUTYH gene was found to be much more common in humans than the researchers' calculations predicted, even when they adjusted for different population histories and selection strengths. The authors could not explain this discrepancy, noting that the variant appears to be an outlier that does not follow the standard rules of selection acting solely on germline mutation rates. This suggests that there may be other factors at play for this specific gene, or that our understanding of the mutation rate at that site is incomplete.

The study also turned its attention to how scientists find these mutators in the first place. Most discoveries so far have come from studying families, or trios, where a child has an unexpectedly high number of new mutations. The researchers simulated this process to see what kind of mutators are most likely to be caught by this method. They found a surprising bias: current family-based studies are much more likely to find mutators that act in a semi-dominant way—those that cause problems even in people with just one copy—because these individuals are more common in the population than those with two copies of a recessive mutator. Yet, the two most famous mutators discovered this way, in the XPC and MPG genes, turned out to be recessive. This contradiction led the researchers to a significant conclusion: there must be a vast number of recessive mutators with large effects that we have not yet found. Because recessive mutators hide in people with a single copy, they are harder to spot in family studies, meaning the known examples are just the tip of the iceberg.

Ultimately, the work provides a new framework for understanding the genetic architecture of mutation rates in humans. It suggests that the variation in how often new mutations occur is not driven by a few common genes, but by a complex landscape of many rare variants, most of which are recessive and hidden from view. While the model successfully explained the behavior of most known mutators, the stubborn case of the MUTYH variant serves as a reminder that nature often holds exceptions to our rules. By combining these mathematical models with real-world data, scientists can better predict where to look next, perhaps focusing on older fathers or specific family structures to uncover the many hidden mutators that shape our genetic future.

Technical Summary: Dynamics of Mutators of Arbitrary Dominance in Humans

Problem Statement
Germline mutations are the substrate for evolution and the cause of heritable disease, yet the mutation rates themselves are evolving traits. While theoretical frameworks exist to explain differences in mutation rates across species (e.g., the "drift-barrier" hypothesis), there is a lack of models describing the dynamics of mutation rate modifiers (mutators) within a species. Specifically, existing models often assume mutators act additively (semi-dominantly) or ignore the effects of genetic drift. Furthermore, the role of dominance coefficients in shaping the frequency and detection probability of mutators remains poorly understood. Recent empirical studies have identified specific human mutators (e.g., in XPC, MPG, POLE, POLD1, MUTYH), but it is unclear whether their observed frequencies and dominance patterns are consistent with purifying selection acting solely through their effects on germline mutation rates, or if other factors (such as somatic effects or pleiotropy) are involved.

Methodology
The authors develop a population genetic model of mutation rate modifiers with arbitrary dominance coefficients ( $h$ ) in a diploid, panmictic population.

Model Structure: The model assumes a genome with $G$ sites, a fraction $f$ of which are under strong selection. Individuals carry deleterious alleles that reduce fitness multiplicatively. Mutation rate is determined by $M$ modifier sites. A mutator allele increases the genome-wide mutation rate by $\omega$ in homozygotes and $h\omega$ in heterozygotes relative to the anti-mutator.
Dynamics: The model incorporates mutation, selection, and genetic drift (Wright-Fisher sampling). It approximates the expected selection coefficient ( $s$ ) against a mutator based on the additional deleterious alleles it generates and their linkage to the mutator. The stationary distribution of mutator frequencies is derived analytically using the first two moments of frequency change, accounting for the dominance coefficient.
Parameterization: The model is parameterized using human demographic history (ancestral $N_e \approx 20,000$ ; South Asian, African, and European histories) and estimated mutation rates ( $\hat{u} \approx 1.25 \times 10^{-8}$ ).
Empirical Validation: The model is applied to seven known human mutators identified in recent pedigree studies (XPC, MPG, POLE, POLD1, MUTYH, REV1, LIG1). For each, the authors estimate the effect size (fold increase in de novo mutations, DNMs) and simulate predicted frequency distributions under varying selection and dominance scenarios.
Detection Probability: The authors simulate a "trio-based" discovery approach (screening ~22,000 trios for offspring with unexpectedly high DNM counts) to determine how effect size and dominance influence the probability of detecting a mutator parent.

Key Results

Frequency Dynamics: The model identifies three regimes based on the scaled selection parameter ( $2N_es$ ): effectively neutral, nearly neutral, and strongly selected. In the strongly selected regime, dominance significantly affects frequency: semi-dominant mutators are kept at lower frequencies than recessive ones because selection acts on heterozygotes. However, recessive mutators contribute less to the mean population mutation rate when rare because they are hidden from selection in heterozygotes.
Fit to Observed Data:
- For six of the seven known mutators (XPC, MPG, POLE, POLD1, and two MUTYH variants), the observed allele frequencies are well-fit by the model under the assumption that selection arises solely from germline mutation effects. However, the likelihood surface is flat across a wide range of selection coefficients, meaning the data cannot strongly reject neutrality or precisely estimate the strength of selection.
- For the MUTYH variant G368D, the observed frequency is significantly higher than predicted by the model under plausible parameters, suggesting an unexplained discrepancy (potentially due to underestimated local mutation rates or other factors).
- For XPC and MPG, the observed frequencies are consistent with purely recessive fitness effects, though partial dominance cannot be excluded due to limited data (single variants per gene).
Detection Bias: The study reveals a strong bias in trio-based discovery methods. For a given effect size, semi-dominant mutators are much more likely to be detected than recessive ones because heterozygous carriers (who are more common for semi-dominant alleles) express the phenotype. Recessive mutators require homozygosity to be detected, which is rare unless the allele frequency is high or consanguinity is present.
Inference on Genetic Architecture: The fact that the first two mutators discovered via trio screening (XPC and MPG) appear to be recessive, despite the model predicting that semi-dominant mutators are easier to detect, implies that the number of modifier sites producing large-effect recessive mutators likely exceeds the number of those producing semi-dominant mutators by at least an order of magnitude.

Significance and Claims
The paper provides a theoretical framework for interpreting the properties of mutators in humans and other species. Its primary contributions are:

Generalization of Mutator Models: It extends previous models by allowing for arbitrary dominance coefficients, moving beyond the assumption of semi-dominance.
Interpretation of Empirical Data: It demonstrates that the frequencies of most known human mutators are consistent with a model of purifying selection driven by germline mutation load, though precise parameter estimation remains difficult due to the "flatness" of the likelihood landscape for single-variant observations.
Insight into Discovery Bias: It highlights that current pedigree-based discovery methods are inherently biased toward detecting semi-dominant mutators. The observation of recessive mutators in these studies suggests a hidden abundance of large-effect recessive modifiers in the human genome.
Genetic Architecture: The findings suggest that the genetic architecture of germline mutation rate variation is highly polygenic, with a preponderance of recessive modifiers for large effect sizes, consistent with broader trends in deleterious mutations.

The authors modestly note that their analyses are limited by the small number of known mutators and the inability to rule out alternative parameter values (such as different selection strengths or pleiotropic somatic effects). They conclude that while the model fits the data, further insights require larger cohort sizes, analysis of multiple variants within genes (e.g., loss-of-function aggregates), and potentially integrating models of somatic mutation effects.

More like this