Effective population size does not explain long-term variation in genome size and transposable element content in animals

Reviewer #1 (Public Review):

Summary:

One enduring mystery involving the evolution of genomes is the remarkable variation they exhibit with respect to size. Much of that variation is due to differences in the number of transposable elements, which often (but not always) correlates with the overall quantity of DNA. Amplification of TEs is nearly always either selectively neutral or negative with respect to host fitness. Given that larger effective population sizes are more efficient at removing these mutations, it has been hypothesized that TE content, and thus overall genome size, may be a function of effective population size. The authors of this manuscript test this hypothesis by using a uniform approach to analysis of several hundred animal genomes, using the ratio of synonymous to nonsynonymous mutations in coding sequence as a measure of the overall strength of purifying selection, which serves as a proxy for effective population size over time. The data convincingly demonstrates that it is unlikely that effective population size has a strong effect on TE content and, by extension, overall genome size (except for birds).

Strengths:

Although this ground has been covered before in many other papers, the strength of this analysis is that it is comprehensive and treats all the genomes with the same pipeline, making comparisons more convincing. Although this is a negative result, it is important because it is relatively comprehensive and indicates that there will be no simple, global hypothesis that can explain the observed variation.

Weaknesses:

In several places, I think the authors slip between assertions of correlation and assertions of cause-effect relationships not established in the results. In other places, the arguments end up feeling circular, based, I think, on those inferred causal relationships. It was also puzzling why plants (which show vast differences in DNA content) were ignored altogether.

Reviewer #2 (Public Review):

Summary:

The Mutational Hazard Hypothesis (MHH) is a very influential hypothesis in explaining the origins of genomic and other complexity that seem to entail the fixation of costly elements. Despite its influence, very few tests of the hypothesis have been offered, and most of these come with important caveats. This lack of empirical tests largely reflects the challenges of estimating crucial parameters.

The authors test the central contention of the MHH, namely that genome size follows effective population size (Ne). They martial a lot of genomic and comparative data, test the viability of their surrogates for Ne and genome size, and use correct methods (phylogenetically corrected correlation) to test the hypothesis. Strikingly, they not only find that Ne is not THE major determinant of genome size, as is argued by MHH, but that there is not even a marginally significant effect. This is remarkable, making this an important paper.

Strengths:

The hypothesis tested is of great importance.

The negative finding is of great importance for reevaluating the predictive power of the tested hypothesis.

The test is straightforward and clear.

The analysis is a technical tour-de-force, convincingly circumventing a number of challenges of mounting a true test of the hypothesis.

Weaknesses:

I note no particular strengths, but I believe the paper could be further strengthened in three major ways.

(1) The authors should note that the hypothesis that they are testing is larger than the MHH. The MHH hypothesis says that
(i) low-Ne species have more junk in their genomes and
(ii) this is because junk tends to be costly because of increased mutation rate to nulls, relative to competing non/less-junky alleles.

The current results reject not just the compound (i+ii) MHH hypothesis, but in fact any hypothesis that relies on i. This is notably a (much) more important rejection. Indeed, whereas MHH relies on particular constructions of increased mutation rates of varying plausibility, the more general hypothesis i includes any imaginable or proposed cost to the extra sequence (replication costs, background transcription, costs of transposition, ectopic expression of neighboring genes, recombination between homologous elements, misaligning during meiosis, reduced organismal function from nuclear expansion, the list goes on and on). For those who find the MHH dubious on its merits, focusing this paper on the MHH reduces its impact - the larger hypothesis that the small costs of extra sequence dictate the fates of different organisms' genomes is, in my opinion, a much more important and plausible hypothesis, and thus the current rejection is more important than the authors let on.

(2) In addition to the authors' careful logical and mathematical description of their work, they should take more time to show the intuition that arises from their data. In particular, just by looking at Figure 1b one can see what is wrong with the non-phylogenetically-corrected correlations that MHH's supporters use. That figure shows that mammals, many of which have small Ne, have large genomes regardless of their Ne, which suggests that the coincidence of large genomes and frequently small Ne in this lineage is just that, a coincidence, not a causal relationship. Similarly, insects by and large have large Ne, regardless of their genome size. Insects, many of which have large genomes, have large Ne regardless of their genome size, again suggesting that the coincidence of this lineage of generally large Ne and smaller genomes is not causal. Given that these two lineages are abundant on earth in addition to being overrepresented among available genomes (and were even more overrepresented when the foundational MHH papers collected available genomes), it begins to emerge how one can easily end up with a spurious non-phylogenetically corrected correlation: grab a few insects, grab a few mammals, and you get a correlation. Notably, the same holds for lineages not included here but that are highly represented in our databases (and all the more so 20 years ago): yeasts related to S. cerevisiae (generally small genomes and large median Ne despite variation) and angiosperms (generally large genomes (compared to most eukaryotes) and small median Ne despite variation). Pointing these clear points out will help non-specialists to understand why the current analysis is not merely a they-said-them-said case, but offers an explanation for why the current authors' conclusions differ from the MHH's supporters and moreover explain what is wrong with the MHH's supporters' arguments.

(3) A third way in which the paper is more important than the authors let on is in the striking degree of the failure of MHH here. MHH does not merely claim that Ne is one contributor to genome size among many; it claims that Ne is THE major contributor, which is a much, much stronger claim. That no evidence exists in the current data for even the small claim is a remarkable failure of the actual MHH hypothesis: the possibility is quite remote that Ne is THE major contributor but that one cannot even find a marginally significant correlation in a huge correlation analysis deriving from a lot of challenging bioinformatic work. Thus this is an extremely strong rejection of the MHH. The MHH is extremely influential and yet very challenging to test clearly. Frankly, the authors would be doing the field a disservice if they did not more strongly state the degree of importance of this finding.

Reviewer #3 (Public Review):

The Mutational Hazard Hypothesis (MHH) suggests that lineages with smaller effective population sizes should accumulate slightly deleterious transposable elements leading to larger genome sizes. Marino and colleagues tested the MHH using a set of 807 vertebrate, mollusc, and insect species. The authors mined repeats de novo and estimated dN/dS for each genome. Then, they used dN/dS and life history traits as reliable proxies for effective population size and tested for correlations between these proxies and repeat content while accounting for phylogenetic nonindependence. The results suggest that overall, lineages with lower effective population sizes do not exhibit increases in repeat content or genome size. This contrasts with expectations from the MHH. The authors speculate that changes in genome size may be driven by lineage-specific host-TE conflicts rather than effective population size.

The general conclusions of this paper are supported by a powerful dataset of phylogenetically diverse species. The use of C-values rather than assembly size for many species (when available) helps mitigate the challenges associated with the underrepresentation of repetitive regions in short-read-based genome assemblies. As expected, genome size and repeat content are highly correlated across species. Nonetheless, the authors report divergent relationships between genome size and dN/dS and TE content and dN/dS in multiple clades: Insecta, Actinopteri, Aves, and Mammalia. These discrepancies are interesting but could reflect biases associated with the authors' methodology for repeat detection and quantification rather than the true biology.

The authors used dnaPipeTE for repeat quantification. Although dnaPipeTE is a useful tool for estimating TE content when genome assemblies are not available, it exhibits several biases. One of these is that dnaPipeTE seems to consistently underestimate satellite content (compared to repeat masker on assembled genomes; see Goubert et al. 2015). Satellites comprise a significant portion of many animal genomes and are likely significant contributors to differences in genome size. This should have a stronger effect on results in species where satellites comprise a larger proportion of the genome relative to other repeats (e.g. Drosophila virilis, >40% of the genome (Flynn et al. 2020); Triatoma infestans, 25% of the genome (Pita et al. 2017) and many others). For example, the authors report that only 0.46% of the Triatoma infestans genome is "other repeats" (which include simple repeats and satellites). This contrasts with previous reports of {greater than or equal to}25% satellite content in Triatoma infestans (Pita et al. 2017). Similarly, this study's results for "other" repeat content appear to be consistently lower for Drosophila species relative to previous reports (e.g. de Lima & Ruiz-Ruano 2022). The most extreme case of this is for Drosophila albomicans where the authors report 0.06% "other" repeat content when previous reports have suggested that 18%->38% of the genome is composed of satellites (de Lima & Ruiz-Ruano 2022). It is conceivable that occasional drastic underestimates or overestimates for repeat content in some species could have a large effect on coevol results, but a minimal effect on more general trends (e.g. the overall relationship between repeat content and genome size).

Another bias of dnaPipeTE is that it does not detect ancient TEs as well as more recently active TEs (Goubert et al. 2015). Thus, the repeat content used for PIC and coevolve analyses here is inherently biased toward more recently inserted TEs. This bias could significantly impact the inference of long-term evolutionary trends.

Author response:

Reviewer #1:

Summary:

One enduring mystery involving the evolution of genomes is the remarkable variation they exhibit with respect to size. Much of that variation is due to differences in the number of transposable elements, which often (but not always) correlates with the overall quantity of DNA. Amplification of TEs is nearly always either selectively neutral or negative with respect to host fitness. Given that larger effective population sizes are more efficient at removing these mutations, it has been hypothesized that TE content, and thus overall genome size, may be a function of effective population size. The authors of this manuscript test this hypothesis by using a uniform approach to analysis of several hundred animal genomes, using the ratio of synonymous to nonsynonymous mutations in coding sequence as a measure of the overall strength of purifying selection, which serves as a proxy for effective population size over time. The data convincingly demonstrates that it is unlikely that effective population size has a strong effect on TE content and, by extension, overall genome size (except for birds).

Strengths:

Although this ground has been covered before in many other papers, the strength of this analysis is that it is comprehensive and treats all the genomes with the same pipeline, making comparisons more convincing. Although this is a negative result, it is important because it is relatively comprehensive and indicates that there will be no simple, global hypothesis that can explain the observed variation.

Weaknesses:

In several places, I think the authors slip between assertions of correlation and assertions of cause-effect relationships not established in the results.

Several times in the text we use the expression “effect of dN/dS on…” which might indeed suggest a causal relationship. The phrasing refers to dN/dS being used in the regression as an independent variable that can be able to predict the variation of the dependent variables genome size and TE content. We are going to rephrase these expressions so that correlation is not mistaken with causation.

In other places, the arguments end up feeling circular, based, I think, on those inferred causal relationships. It was also puzzling why plants (which show vast differences in DNA content) were ignored altogether.

The analysis focuses on metazoans for two reasons: one practical and one fundamental. The practical reason is computational. Our analysis included TE annotation, phylogenetic estimation and dN/dS estimation, which would have been very difficult with the hundreds, if not thousands, of plant genomes available. If we had included plants, it would have been natural to include fungi as well, to have a complete set of multicellular eukaryotic genomes, adding to the computational burden. The second fundamental reason is that plants show important genome size differences due to more frequent whole genome duplications (polyploidization) than in animals. It is therefore possible that the effect of selection on genome size is different in these two groups, which would have led us to treat them separately, decreasing the interest of this comparison. For these reasons we chose to focus on animals that still provide very wide ranges of genome size and population size well suited to test the impact of drift.

Reviewer #2:

Summary:

The Mutational Hazard Hypothesis (MHH) is a very influential hypothesis in explaining the origins of genomic and other complexity that seem to entail the fixation of costly elements. Despite its influence, very few tests of the hypothesis have been offered, and most of these come with important caveats. This lack of empirical tests largely reflects the challenges of estimating crucial parameters.

The authors test the central contention of the MHH, namely that genome size follows effective population size (Ne). They martial a lot of genomic and comparative data, test the viability of their surrogates for Ne and genome size, and use correct methods (phylogenetically corrected correlation) to test the hypothesis. Strikingly, they not only find that Ne is not THE major determinant of genome size, as is argued by MHH, but that there is not even a marginally significant effect. This is remarkable, making this an important paper.

Strengths:

The hypothesis tested is of great importance.

The negative finding is of great importance for reevaluating the predictive power of the tested hypothesis.

The test is straightforward and clear.

The analysis is a technical tour-de-force, convincingly circumventing a number of challenges of mounting a true test of the hypothesis.

Weaknesses:

I note no particular strengths, but I believe the paper could be further strengthened in three major ways.

(1) The authors should note that the hypothesis that they are testing is larger than the MHH. The MHH hypothesis says that

(i) low-Ne species have more junk in their genomes and

(ii) this is because junk tends to be costly because of increased mutation rate to nulls, relative to competing non/less-junky alleles.

The current results reject not just the compound (i+ii) MHH hypothesis, but in fact any hypothesis that relies on i. This is notably a (much) more important rejection. Indeed, whereas MHH relies on particular constructions of increased mutation rates of varying plausibility, the more general hypothesis i includes any imaginable or proposed cost to the extra sequence (replication costs, background transcription, costs of transposition, ectopic expression of neighboring genes, recombination between homologous elements, misaligning during meiosis, reduced organismal function from nuclear expansion, the list goes on and on). For those who find the MHH dubious on its merits, focusing this paper on the MHH reduces its impact - the larger hypothesis that the small costs of extra sequence dictate the fates of different organisms' genomes is, in my opinion, a much more important and plausible hypothesis, and thus the current rejection is more important than the authors let on.

The MHH is arguably the most structured and influential theoretical framework proposed to date based on the null assumption (i), therefore setting the paper up with the MHH is somehow inevitable. Because of this, in the manuscript, we mostly discuss the peculiarities of TE biology that can drive the genome away from the MHH expectations, focusing on the mutational aspect. We however agree that the hazard posed by extra DNA is not limited to the gain of function via the mutation process, but can be linked to many other molecular processes as mentioned above. In a revised manuscript, we will make the concept of hazard more comprehensive and further stress that this applies not only to TEs but any nearly-neutral mutation affecting non-coding DNA.

(2) In addition to the authors' careful logical and mathematical description of their work, they should take more time to show the intuition that arises from their data. In particular, just by looking at Figure 1b one can see what is wrong with the non-phylogenetically-corrected correlations that MHH's supporters use. That figure shows that mammals, many of which have small Ne, have large genomes regardless of their Ne, which suggests that the coincidence of large genomes and frequently small Ne in this lineage is just that, a coincidence, not a causal relationship. Similarly, insects by and large have large Ne, regardless of their genome size. Insects, many of which have large genomes, have large Ne regardless of their genome size, again suggesting that the coincidence of this lineage of generally large Ne and smaller genomes is not causal. Given that these two lineages are abundant on earth in addition to being overrepresented among available genomes (and were even more overrepresented when the foundational MHH papers collected available genomes), it begins to emerge how one can easily end up with a spurious non-phylogenetically corrected correlation: grab a few insects, grab a few mammals, and you get a correlation. Notably, the same holds for lineages not included here but that are highly represented in our databases (and all the more so 20 years ago): yeasts related to S. cerevisiae (generally small genomes and large median Ne despite variation) and angiosperms (generally large genomes (compared to most eukaryotes) and small median Ne despite variation). Pointing these clear points out will help non-specialists to understand why the current analysis is not merely a they-said-them-said case, but offers an explanation for why the current authors' conclusions differ from the MHH's supporters and moreover explain what is wrong with the MHH's supporters' arguments.

We agree that comparing dispersion of the points from the non-phylogenetically corrected correlation with the results of the phylogenetic contrasts intuitively emphasizes the importance of accounting for species relatedness. Just looking at the clade colors in Figure 2 makes immediately stand out that a simple regression hides phylogenetic structure. We will stress this in the discussion to make the point clear.

(3) A third way in which the paper is more important than the authors let on is in the striking degree of the failure of MHH here. MHH does not merely claim that Ne is one contributor to genome size among many; it claims that Ne is THE major contributor, which is a much, much stronger claim. That no evidence exists in the current data for even the small claim is a remarkable failure of the actual MHH hypothesis: the possibility is quite remote that Ne is THE major contributor but that one cannot even find a marginally significant correlation in a huge correlation analysis deriving from a lot of challenging bioinformatic work. Thus this is an extremely strong rejection of the MHH. The MHH is extremely influential and yet very challenging to test clearly. Frankly, the authors would be doing the field a disservice if they did not more strongly state the degree of importance of this finding.

We respectfully disagree with the reviewer that there is currently no evidence for an effect of Ne on genome size evolution. While it is accurate that our large dataset allows us to reject the universality of Ne as the major contributor to genome size variation, this does not exclude the possibility of such an effect in certain contexts. Notably, there are several pieces of evidence that find support for Ne to determine genome size variation and to entail nearly-neutral TE dynamics under certain circumstances, e.g. of particularly strongly contrasted Ne and moderate divergence times (Lefébure et al. 2017; Mérel et al. 2024; Tollis and Boissinot 2013; Ruggiero et al. 2017). The strength of such works is to analyze the short-term dynamics of TEs in response to Ne within groups of species/populations, where the cost posed by extra DNA is likely to be similar. Indeed, the MHH predicts genome size to vary according to the combination of drift and mutation under the nearly-neutral theory of molecular evolution. Our work demonstrates that it is not true universally but does not exclude that it could exist locally. Moreover, defense mechanisms against TEs proliferation are often complex molecular machineries that might or might not evolve according to different constraints among clades. We have detailed these points in the discussion.

Reviewer #3:

Summary

The Mutational Hazard Hypothesis (MHH) suggests that lineages with smaller effective population sizes should accumulate slightly deleterious transposable elements leading to larger genome sizes. Marino and colleagues tested the MHH using a set of 807 vertebrate, mollusc, and insect species. The authors mined repeats de novo and estimated dN/dS for each genome. Then, they used dN/dS and life history traits as reliable proxies for effective population size and tested for correlations between these proxies and repeat content while accounting for phylogenetic nonindependence. The results suggest that overall, lineages with lower effective population sizes do not exhibit increases in repeat content or genome size. This contrasts with expectations from the MHH. The authors speculate that changes in genome size may be driven by lineage-specific host-TE conflicts rather than effective population size.

Strengths

The general conclusions of this paper are supported by a powerful dataset of phylogenetically diverse species. The use of C-values rather than assembly size for many species (when available) helps mitigate the challenges associated with the underrepresentation of repetitive regions in short-read-based genome assemblies. As expected, genome size and repeat content are highly correlated across species. Nonetheless, the authors report divergent relationships between genome size and dN/dS and TE content and dN/dS in multiple clades: Insecta, Actinopteri, Aves, and Mammalia. These discrepancies are interesting but could reflect biases associated with the authors' methodology for repeat detection and quantification rather than the true biology.

Weaknesses

The authors used dnaPipeTE for repeat quantification. Although dnaPipeTE is a useful tool for estimating TE content when genome assemblies are not available, it exhibits several biases. One of these is that dnaPipeTE seems to consistently underestimate satellite content (compared to repeat masker on assembled genomes; see Goubert et al. 2015). Satellites comprise a significant portion of many animal genomes and are likely significant contributors to differences in genome size. This should have a stronger effect on results in species where satellites comprise a larger proportion of the genome relative to other repeats (e.g. Drosophila virilis, >40% of the genome (Flynn et al. 2020); Triatoma infestans, 25% of the genome (Pita et al. 2017) and many others). For example, the authors report that only 0.46% of the Triatoma infestans genome is "other repeats" (which include simple repeats and satellites). This contrasts with previous reports of {greater than or equal to}25% satellite content in Triatoma infestans (Pita et al. 2017). Similarly, this study's results for "other" repeat content appear to be consistently lower for Drosophila species relative to previous reports (e.g. de Lima & Ruiz-Ruano 2022). The most extreme case of this is for Drosophila albomicans where the authors report 0.06% "other" repeat content when previous reports have suggested that 18%->38% of the genome is composed of satellites (de Lima & Ruiz-Ruano 2022). It is conceivable that occasional drastic underestimates or overestimates for repeat content in some species could have a large effect on coevol results, but a minimal effect on more general trends (e.g. the overall relationship between repeat content and genome size).

There are indeed some discrepancies between our estimates of low complexity repeats and those from the literature due to the approach used. Hence, occasional underestimates or overestimates of repeat content are possible. As noted, the contribution of “Other” repeats to the overall repeat content is generally very low, meaning an underestimation bias. We thank the reviewer for providing this interesting review. We will emphasize it in the discussion of our revised manuscript.

Not being able to correctly estimate the quantity of satellites might pose a problem for quantifying the total content of junk DNA. However, the overall repeat content mostly composed of TEs correlates very well with genome size, both in the overall dataset and within clades (with the notable exception of birds) so we are confident that this limitation is not the explanation of our negative results. Moreover, while satellite information might be missing, this is not problematic to test our a priori hypothesis since we focus our attention on TEs, whose proliferation mechanism is very different from that of tandem repeats.

Finally, divergence from the consensus can be estimated only for TEs. Therefore, recently active elements do not include simple and tandem repeats: yet the results based on recent TE content are very similar to those based on the overall repeat content.

Another bias of dnaPipeTE is that it does not detect ancient TEs as well as more recently active TEs (Goubert et al. 2015). Thus, the repeat content used for PIC and coevolve analyses here is inherently biased toward more recently inserted TEs. This bias could significantly impact the inference of long-term evolutionary trends.

Indeed, dnaPipeTE is not good at detecting old TE copies due to the read-based approach, biasing the outcome towards new elements. We agree on TE content being underestimated, especially in those genomes that tend to accumulate TEs rather than getting rid of them. However, the sum of old TEs and recent TEs is extremely well correlated to genome size (Pearson’s correlation: r = 0.87, p-value < 2.2e-16; PIC: slope = 0.22, adj-R2 = 0.42, p-value < 2.2e-16). Our main result therefore does not rely on an accurate estimation of old TEs. In contrast, we hypothesized that recent TEs could be interesting if selection acted on TEs insertion and dynamics rather than on non-coding DNA. Our results demonstrate that this is not the case: it should be noted that in spite of its limits for old TEs, dnaPipeTE is especially fitting for this specific analysis as it is not biased by very repetitive new TE families that are problematic to assemble. We will clearly emphasize the limitation of dnaPipeTE and discuss the consequences on our results in the discussion of the revised manuscript.

Finally, in a preliminary analysis on the dipteran species, we show that the TE content estimated with dnaPipeTE is generally similar to that estimated from the assembly with earlGrey (Baril et al. 2024) across a good range of genome sizes going from drosophilid-like to mosquito-like (Pearson’s correlation: r = 0.88, p-value = 3.22e-10; see also the corrected Supplementary Figure S2 below). While for these species TEs are probably dominated by recent to moderately recent TEs, Aedes albopictus is an outlier for its genome size and the estimations with the two methods are largely consistent. However, the computation time required to estimate TE content using EarlGrey was significantly longer, with a ~300% increase in computation time, making it a very costly option (a similar issue is applicable to other assembly-based annotation pipelines). Given the rationale presented above, we decided to use dnaPipeTE instead of EarlGrey.