Limited role of generation time changes in driving the evolution of the mutation spectrum in humans

Curation statements for this article:
  • Curated by eLife

    eLife logo

    eLife assessment

    This study, of interest to population geneticists and evolutionary biologists alike, aims at investigating temporal variation in patterns of germline mutation during the evolution of human populations. The authors suggest that shifts in mutation spectra occur frequently, over a few thousands of generations, possibly as a consequence of changes in environmental exposure, or of genetic modifiers. There are several important aspects of methodology that need to be clarified, and several additional tests have to be done to confirm that the reported observations are not the result of methodological artifacts. The paper also overstates certain weaknesses of previously published papers on mutation spectrum evolution as well as the generation time hypothesis; correcting these oversimplifications would more accurately capture what the paper's new analyses add to the state of knowledge in these areas.

This article has been Reviewed by the following groups

Read the full article See related articles

Abstract

Recent studies have suggested that the human germline mutation rate and spectrum evolve rapidly. Variation in generation time has been linked to these changes, though its contribution remains unclear. We develop a framework to characterize temporal changes in polymorphisms within and between populations, while controlling for the effects of natural selection and biased gene conversion. Application to the 1000 Genomes Project dataset reveals multiple independent changes that arose after the split of continental groups, including a previously reported, transient elevation in TCC>TTC mutations in Europeans and novel signals of divergence in C>Gand T>A mutation rates among population samples. We also find a significant difference between groups sampled in and outside of Africa in old T>C polymorphisms that predate the out-of-Africa migration. This surprising signal is driven by TpG>CpG mutations and stems in part from mis-polarized CpG transitions, which are more likely to undergo recurrent mutations. Finally, by relating the mutation spectrum of polymorphisms to parental age effects on de novo mutations, we show that plausible changes in the generation time cannot explain the patterns observed for different mutation types jointly. Thus, other factors – genetic modifiers or environmental exposures – must have had a non-negligible impact on the human mutation landscape.

Article activity feed

  1. Author Resposnse

    Reviewer #2 (Public Review):

    This manuscript reassesses the strength of evidence for rapid human germline mutation spectrum evolution, using high coverage whole genome sequencing data and paying particular attention to the potential impact of confounders like biased gene conversion. The authors also refute some recently published arguments that historical changes in the age of reproduction might explain the existence of such mutation spectrum changes. My overall impression is that the paper presents a useful new angle for studying mutation spectrum evolution, and the analysis is nicely suited to addressing whether a particular model such as the parental age model can explain a set of observed polymorphism data. My main criticism is that the paper overstates certain weaknesses of previously published papers on mutation spectrum evolution as well as the generation time hypothesis; correcting these oversimplifications would more accurately capture what the paper's new analyses add to the state of knowledge in these areas.

    As part of the motivation for the current study, the introduction states in lines 97-99 that "it thus remains unclear if the numerous observed [mutation spectrum] differences across human populations stem from rapid evolution of the mutation process itself, other evolutionary processes, or technical factors." This seems to overstate the uncertainty that existed prior to this study, given that Speidel, et al. 2021 found elevated TCC>TTC fractions in ancient genomes from a specific ancient European population, which seems like pretty airtight evidence that this historical mutation rate increase really happened. In addition, earlier papers (Harris 2015, Mathieson & Reich 2016, Harris & Pritchard 2017) already presented analyses rejecting the hypothesis that biased gene conversion or genetic drift could explain the reported patterns-in fact, the Mathieson & Reich paper reports one mutation spectrum difference between populations that they conclude is an artifact caused by the Native American population bottleneck, but they conclude that other mutation spectrum differences appear more robust.

    We completely agree with the reviewer that there has been compelling evidence from multiple independent groups supporting transient elevation of TCC>TTC mutation rate in Europeans. Beyond the TCC signal, however, the mechanisms underlying the observed differences in mutation spectrum across populations remain unclear. In particular, several biological and technical factors impact the mutation spectrum and none of the previous studies have investigated their effects, independently or altogether. Thus, it remains unclear if the mutation rate is evolving rapidly across populations, or if one or more factors (like biased gene conversion) differ across groups or over evolutionary time. Our analysis framework attempts to control these effects together to more reliably investigate the effects of various factors and examine when and how often there has been evolution of mutation rate over the course of human evolution.

    As the authors acknowledge in the discussion of their own results, biased gene conversion and non-equilibrium demography are difficult confounders to deal with, and neither previous papers nor the current paper are able to do this in a way that is 100% foolproof. The current manuscript makes a valuable contribution by presenting new ways of dealing with these issues, particularly since previous papers' work on this topic was often confined to supplementary material, but it seems appropriate to acknowledge that earlier papers discussed the potential impacts of biased gene conversion and demographic complexity and presented their own analyses arguing that these phenomena were poor explanations for the existence of mutation spectrum differences between populations.

    For the most part, I found the paper's introduction to be a useful summary of previous work, but there are a few additional places where the limitations of previous work could be described more clearly. I'd suggest noting that the data artifacts discovered by Anderson-Trocmé, et al. were restricted to a few old samples and that the large differences the current manuscript focuses on were never implicated as potential cell line artifacts. In addition, when the authors mention that their new approach includes "minimiz[ing] confounding effects of selection by removing constrained regions and known targets of selection" (lines 106-107), they should note that earlier papers like Harris & Pritchard 2017 also excluded conserved regions and exons.

    We agree with the reviewer that some of the previous work also attempted to account for the contributions of selection or other factors in post hoc ways; we now acknowledge this in the Results section more explicitly. However, we note that our contribution is in introducing a framework to account for these effects a priori and then assess if there are differences in mutation spectrum across populations and over the course of human evolution. In particular, an innovation of our framework is to better control for the effect of gBGC, which has not been done in previous studies.

    One innovative aspect of the current paper's approach is the use of allele ages inferred by Relate, which certainly has advantages over using allele frequencies as a proxy for allele age. Though the authors of Relate previously used this approach to study mutation spectrum evolution, they did not perform such a thorough investigation of ancient alleles and collapsed mutation type ratios. I like the authors' approach of building uncertainty into the use of Relate's age estimates, but I wonder about the validity of assuming that the allele age posterior probability is distributed uniformly between the upper and lower confidence bounds. Can the authors address why this is more appropriate than some kind of peaked distribution like a beta distribution?

    The lower and upper bounds of the allele age reported by Relate reflect the start and end points of the branch that the mutation falls on in the reconstructed genealogical tree. If Relate does a perfect job in reconstructing the tree and estimating the branch lengths, the mutation age should be uniformly distributed in the inferred interval. It is unrealistic that Relate can perform perfectly in tree building, and there is likely considerable uncertainty and even bias in the time to endpoints of the branch. Unfortunately, Relate does not report the uncertainty in the lower and upper bounds of the mutation age, so we were not able to model the posterior distribution of the allele age properly. However, assuming a uniform distribution of the mutation age between the upper and lower confidence bounds should be valid to first approximation.

    I would also argue that the statement on line 104 about Relate's reliability is not yet supported by data-there is certainly value in using Relate ages to investigate mutation spectrum change over time and compare this to what has been seen using allele frequencies, but I don't think we know enough yet to say that the Relate ages are definitely more reliable. Relate's estimates might be biased by the same processes like selection and demography that make allele frequencies challenging to interpret. The paper's statements about the limitations of allele frequencies are fair, but there is always a tradeoff between the clear drawbacks of simple summary statistics and the more cryptic possible blind spots of complicated "black box" algorithms (in the case of Relate, an MCMC that needs to converge properly). DeWitt, et al. 2021 noted that the demographic history inferred by Relate doesn't accurately predict the underlying data's site frequency spectrum, indicating that the associated allele ages might have some problems that need to be better characterized. While testing Relate for biases is beyond the scope of this work, the introduction should acknowledge that the accuracy and precision of its time estimates are still somewhat uncertain.

    We agree with the reviewer and have now added a paragraph in the Discussion highlighting some issues of Relate regarding mutation age estimation and ancestral allele polarization.

    The paper's results on C>T mutations in Europeans versus Africans are a nice confirmation of previous results, including the observation from Mathieson & Reich that neither SBS7 nor SBS11 is a good match for the mutational signature at play. More novel is the ancient mutational signature enriched in Africa and the interrogation of the ability of parental age to explain the observed patterns. I just have a few minor suggestions regarding these analyses:

    1. I like the idea of using maternal age C>G hotspots to test the plausibility of the maternal age as an explanatory factor, but I think this would be more convincing with the addition of a power analysis. Given two populations that have average maternal ages of 20 and 40, and the same population sample sizes available from 1000 Genomes, can the authors calculate whether the results they'd predict are any different from what is observed (i.e. no significant differences within the maternal hotspots and significant differences outside of these regions)?

    We thank the review for this suggestion. We performed simulations to estimate the power of observing significant inter-population differences within and outside the maternal C>G mutation hotspots, under the assumption that all differences in the mutation spectrum between the two populations are related to the parental age (i.e., generation time). We found that, because of the extraordinarily strong maternal age effects in the maternal mutation hotspots, the power for detecting variation in C>G/T>A ratio due to change in generation age is much greater within maternal hotspots than outside, despite the smaller total size of the maternal hotspot regions (and hence fewer SNPs; Figure 3 – figure supplement 4). For example, even with an age difference of five years, there is nearly 100% power to detect significant differences in the maternal hotspots, compared to <12% for regions outside the maternal hotspots. In other words, if inter-population differences in the mutation spectrum are driven by differences in maternal age across populations, we should have enough power to observe a signal in the maternal hotspot regions alone, the lack of which (Figure 2C) strongly suggests that maternal age is not driving these signals.

    1. Is it possible that the T>C/T>G ratio is elevated in all variants above a certain age but shows up as an African-specific signal because the African population retains more segregating variation in this age range, whereas non-African populations have fixed or lost more of this variation? Since Durvasula & Sankararaman identified putative tracts of super-archaic introgression within Africans, is it possible to test whether the mutation spectrum signal is enriched within those tracts?

    The observation that the T>C / T>G signal is driven by TpG>CpG mutations (which might be mis-polarized CpG transitions) casts a doubt on the signal. Given the unresolved technical issue, we have now removed any discussion of the biological explanations behind the signal and instead focus on describing the challenges with ancestral allele polarization under context-dependent mutation rate variation.

    1. Although Coll Macià, et al. argued that generation time is capable of explaining all mutation spectrum differences between populations, including the excess of TCC>TTC in Europeans, Wang et al. argue something slightly different. They exclude TCC>TTC and the other major components of the European signature from their analysis and then argue that parental age can explain the rest of the differences between populations. I think the analysis in this paper convincingly refutes the Coll Macià, et al. argument, but refuting the Wang, et al. version would require excluding the same mutation types that are excluded in that paper.

    Although we did not present an analysis that explicitly excludes TCC>TTC mutations, our analysis still shows that generation time alone cannot explain the remaining variations in the mutation spectrum observed (Figure 4). Specifically, the temporal trend of T>C/T>G ratio would suggest a decreasing generation time of Europeans with time, whereas the C>G/T>A ratio suggests the opposite. In addition, the power analysis for C>G maternal hotspots (suggested by the reviewer) further supports that the inter-population differences observed cannot be entirely driven by differences in parental ages. These observations, which do not involve TCC>TTC mutations, strongly suggest that generation time is not the sole or primary driver of differences in mutation spectrum across populations. Further, our analysis shows that several technical issues and biological processes, in addition to changes in life history traits can lead to changes in the mutation spectrum of polymorphisms. Therefore, inferring generation time using changes in mutation spectrum is not straightforward as Wang et al. proposed, because generation time is not the only or dominant factor impacting mutation spectrum.

  2. eLife assessment

    This study, of interest to population geneticists and evolutionary biologists alike, aims at investigating temporal variation in patterns of germline mutation during the evolution of human populations. The authors suggest that shifts in mutation spectra occur frequently, over a few thousands of generations, possibly as a consequence of changes in environmental exposure, or of genetic modifiers. There are several important aspects of methodology that need to be clarified, and several additional tests have to be done to confirm that the reported observations are not the result of methodological artifacts. The paper also overstates certain weaknesses of previously published papers on mutation spectrum evolution as well as the generation time hypothesis; correcting these oversimplifications would more accurately capture what the paper's new analyses add to the state of knowledge in these areas.

  3. Reviewer #1 (Public Review):

    This study aims at investigating temporal variation in patterns of germline mutation during the evolution of human populations. For this purpose, the authors analyzed polymorphism data from the 1000 Genomes project. They inferred the age of each derived variant using Relate, a newly developed method that reconstructs local genealogies based on phased haplotype sequences and estimates allele ages (Speidel et al. 2019).

    Speidel et al. (2019) already had used their method to explore temporal variation in mutation patterns. Their analysis had confirmed the transient elevation in non-CpG C>T mutations in Europeans compared to African and Est Asians previously described by Harris (2015). However, Speidel et al. did not push their study very far, notably because of the difficulty of distinguishing the effects of changes in mutation patterns from those of GC-biased gene conversion (gBGC).

    Here Gao et al. carefully accounted for gBGC to further explore variation in mutation patterns. As expected, they confirmed the previously described European-specific mutational shift. In addition, they identified two novel interpopulation differences in the mutation spectrum. This suggests that shifts in mutation spectra occur frequently, over a few thousands of generations. The reasons (environmental or genetic) for these recurrent shifts are not known, but the authors convincingly show that they cannot be explained by changes in the age of reproduction over time.

    I found this manuscript very well written and very interesting. There is however an important point in their results that seems very puzzling. Indeed, the authors report that among mutations that are estimated to be old (>28800 generations), the ratio of T>C over T>G differs significantly in African samples compared to non-African samples (Fig. 2A). This difference is unexpected given that these old mutations largely predate the out-of-Africa migration (<3000 generations), and hence are a priori expected to be largely shared across populations. Curiously, this pattern is driven by variants for which the derived allele is observed in both Neanderthals and Denisovans (ND11 variants) (while ND01 and ND10 variants do not contribute to this pattern; Fig. 2D, SupFig 2.8). The authors hypothesize that the T>C/T>G ratio was higher in one or more populations in the remote past and those ancient groups contribute variable amounts of ancestry to contemporary populations. However, I do not understand how this model can account for the fact that ND10 and ND01 variants behave differently from ND11 variants (ND10 and ND01 variants are also expected to be emerged prior to the split of modern humans and archaic hominins).

    It is possible that I misunderstood something, but in any case, there are several points in the methodology that have to be clarified. Notably, it is not clear to me if the reported pattern is driven by variants that are specific to the African samples, or if it is also observed among variants that are shared across populations. Furthermore, I suspect that polarization errors (notably at CpG sites) might be responsible for this pattern.

    In summary, this manuscript reports very interesting observations, but several additional tests have to be done to check whether they are real or if they might result from methodological artefacts.

  4. Reviewer #2 (Public Review):

    This manuscript reassesses the strength of evidence for rapid human germline mutation spectrum evolution, using high coverage whole genome sequencing data and paying particular attention to the potential impact of confounders like biased gene conversion. The authors also refute some recently published arguments that historical changes in the age of reproduction might explain the existence of such mutation spectrum changes. My overall impression is that the paper presents a useful new angle for studying mutation spectrum evolution, and the analysis is nicely suited to addressing whether a particular model such as the parental age model can explain a set of observed polymorphism data. My main criticism is that the paper overstates certain weaknesses of previously published papers on mutation spectrum evolution as well as the generation time hypothesis; correcting these oversimplifications would more accurately capture what the paper's new analyses add to the state of knowledge in these areas.

    As part of the motivation for the current study, the introduction states in lines 97-99 that "it thus remains unclear if the numerous observed [mutation spectrum] differences across human populations stem from rapid evolution of the mutation process itself, other evolutionary processes, or technical factors." This seems to overstate the uncertainty that existed prior to this study, given that Speidel, et al. 2021 found elevated TCC>TTC fractions in ancient genomes from a specific ancient European population, which seems like pretty airtight evidence that this historical mutation rate increase really happened. In addition, earlier papers (Harris 2015, Mathieson & Reich 2016, Harris & Pritchard 2017) already presented analyses rejecting the hypothesis that biased gene conversion or genetic drift could explain the reported patterns-in fact, the Mathieson & Reich paper reports one mutation spectrum difference between populations that they conclude is an artifact caused by the Native American population bottleneck, but they conclude that other mutation spectrum differences appear more robust. As the authors acknowledge in the discussion of their own results, biased gene conversion and non-equilibrium demography are difficult confounders to deal with, and neither previous papers nor the current paper are able to do this in a way that is 100% foolproof. The current manuscript makes a valuable contribution by presenting new ways of dealing with these issues, particularly since previous papers' work on this topic was often confined to supplementary material, but it seems appropriate to acknowledge that earlier papers discussed the potential impacts of biased gene conversion and demographic complexity and presented their own analyses arguing that these phenomena were poor explanations for the existence of mutation spectrum differences between populations.

    For the most part, I found the paper's introduction to be a useful summary of previous work, but there are a few additional places where the limitations of previous work could be described more clearly. I'd suggest noting that the data artifacts discovered by Anderson-Trocmé, et al. were restricted to a few old samples and that the large differences the current manuscript focuses on were never implicated as potential cell line artifacts. In addition, when the authors mention that their new approach includes "minimiz[ing] confounding effects of selection by removing constrained regions and known targets of selection" (lines 106-107), they should note that earlier papers like Harris & Pritchard 2017 also excluded conserved regions and exons.

    One innovative aspect of the current paper's approach is the use of allele ages inferred by Relate, which certainly has advantages over using allele frequencies as a proxy for allele age. Though the authors of Relate previously used this approach to study mutation spectrum evolution, they did not perform such a thorough investigation of ancient alleles and collapsed mutation type ratios. I like the authors' approach of building uncertainty into the use of Relate's age estimates, but I wonder about the validity of assuming that the allele age posterior probability is distributed uniformly between the upper and lower confidence bounds. Can the authors address why this is more appropriate than some kind of peaked distribution like a beta distribution?
    I would also argue that the statement on line 104 about Relate's reliability is not yet supported by data-there is certainly value in using Relate ages to investigate mutation spectrum change over time and compare this to what has been seen using allele frequencies, but I don't think we know enough yet to say that the Relate ages are definitely more reliable. Relate's estimates might be biased by the same processes like selection and demography that make allele frequencies challenging to interpret. The paper's statements about the limitations of allele frequencies are fair, but there is always a tradeoff between the clear drawbacks of simple summary statistics and the more cryptic possible blind spots of complicated "black box" algorithms (in the case of Relate, an MCMC that needs to converge properly). DeWitt, et al. 2021 noted that the demographic history inferred by Relate doesn't accurately predict the underlying data's site frequency spectrum, indicating that the associated allele ages might have some problems that need to be better characterized. While testing Relate for biases is beyond the scope of this work, the introduction should acknowledge that the accuracy and precision of its time estimates are still somewhat uncertain.

    The paper's results on C>T mutations in Europeans versus Africans are a nice confirmation of previous results, including the observation from Mathieson & Reich that neither SBS7 nor SBS11 is a good match for the mutational signature at play. More novel is the ancient mutational signature enriched in Africa and the interrogation of the ability of parental age to explain the observed patterns. I just have a few minor suggestions regarding these analyses:

    1. I like the idea of using maternal age C>G hotspots to test the plausibility of the maternal age as an explanatory factor, but I think this would be more convincing with the addition of a power analysis. Given two populations that have average maternal ages of 20 and 40, and the same population sample sizes available from 1000 Genomes, can the authors calculate whether the results they'd predict are any different from what is observed (i.e. no significant differences within the maternal hotspots and significant differences outside of these regions)?

    2. Is it possible that the T>C/T>G ratio is elevated in all variants above a certain age but shows up as an African-specific signal because the African population retains more segregating variation in this age range, whereas non-African populations have fixed or lost more of this variation? Since Durvasula & Sankararaman identified putative tracts of of super-archaic introgression within Africans, is it possible to test whether the mutation spectrum signal is enriched within those tracts?

    3. Although Coll Macià, et al. argued that generation time is capable of explaining all mutation spectrum differences between populations, including the excess of TCC>TTC in Europeans, Wang et al. argue something slightly different. They exclude TCC>TTC and the other major components of the European signature from their analysis and then argue that parental age can explain the rest of the differences between populations. I think the analysis in this paper convincingly refutes the Coll Macià, et al. argument, but refuting the Wang, et al. version would require excluding the same mutation types that are excluded in that paper.