Introduction

DNA damage, the precursor to mutations, occurs between 1,000 to 100,000 times daily in each cell ^1–9. This poses a continual risk of generating mutations that could negatively impact health and fitness. To address this challenge, organisms rely on proteins that detect and repair DNA damage. Whether organisms can preferentially repair and reduce mutation rates in functionally important genomic regions such as coding regions is both a mechanistic and theoretical question critical to understanding mutation rate evolution ^10–13.

Research in humans, largely from the field of cancer biology, has revealed specific mechanisms that recruit DNA repair proteins to epigenomic marks found in the coding regions of active genes ^14–34. These consist of DNA repair proteins fused to histone reader domains, which bind to specific histone posttranslational modification (PTMs). For example, the N-terminus of human DNA mismatch repair protein MSH6 (homolog of bacterial and archaeal MutS³⁵) is fused to a PWWP histone reader domain that binds to H3K36me3 (Figure 1a-c) ^36,37. H3K36me3, in humans, is highly enriched in the exons of active genes, while MSH6 plays a key role in several DNA repair pathways. Consequently, targeting MSH6 via its PWWP domain to H3K36me3 reduces mutation rates in coding regions ^{18,19,29,38,39}. This is especially true and well-studied in somatic tissues where mutations in coding sequences are the primary source of cancer driver mutations ^29,40.

Figure 1.

In plants, H3K4me1 is the most enriched histone PTM, marking coding regions of actively expressed and essential genes ^41–48. Studies on de novo mutation in the model plant Arabidopsis thaliana (Arabidopsis hereafter) and rice have found that H3K4me1 exhibits the strongest association with regions of low mutation rates, akin to H3K36me3-associated hypomutation observed in humans ^48–51. Indeed, mutation rates in Arabidopsis are lower in gene bodies ^{49,50,52–60}. Importantly, the N-terminus of MSH6 in Arabidopsis was found to be fused to a Tudor domain that binds H3K4me1, which is distinct from the H3K36me3-binding PWWP domain fused to the human MSH6 (Figure 1a-d) ^48,49,58,61. Experiments have confirmed that genic H3K4me1-associated hypomutation is significantly diminished in the absence of MSH6 in Arabidopsis⁴⁸. This observation aligns with mutation accumulation studies in MSH2 deficient plants (the protein that dimerizes with MSH6 to form the mismatch repair protein complex MutSα ^62,63), which, like Arabidopsis lacking MSH6, exhibit elevated mutation rates in H3K4me1-rich gene bodies ⁵⁶.

Thus, humans and Arabidopsis have distinct histone readers fused to MSH6, PWWP, and Tudor, respectively, which each bind histone PTMs in genic regions and contribute to lower mutation rates. However, the evolution of these systems across the tree of life remains unknown.

The Tudor and PWWP domains of Arabidopsis and Human MSH6, respectively, are each members of the ‘Royal Family’ of histone readers, which also includes Agenet, Chromo, and MBT domains⁶⁴. These histone readers are found across eukaryotes in numerous proteins (often those involved in chromatin regulatory feedbacks, for example, when fused to histone methyltransferases ⁶⁵). PWWP, Tudor, and other Royal Family histone reader domains are believed to have arisen and diversified from a common eukaryotic ancestor before the emergence of major eukaryotic clades ^64,65. The differences in binding sites, structure, and amino acid sequence between the MSH6 Tudor and PWWP domains suggest they do not share a common origin in an ancestral MSH6 protein with histone reader functionality (Figure 1c,d). Furthermore, the presence of Tudor and PWWP domains in multiple other proteins across Arabidopsis and humans indicates that these domains likely evolved separately before they were incorporated into MSH6 ^64–66. Thus, the N-terminus fusions of Tudor and PWWP domains to MSH6 may reflect independent evolutionary events, a hypothesis that has yet to be tested with broad phylogenetic analyses.

Here, we investigated histone reader fusions to MSH6 across the eukaryotic tree of life to elucidate their origins, diversity, and potential adaptive value. Specifically, we test for evidence of evolutionary convergence⁶⁷, selection of functional amino acids in the MSH6 histone readers, and co-evolution of MSH6 histone readers with other traits.

Results & discussion

MSH6 independently evolved distinct histone readers in eukaryotes

Our first aim was to determine the evolutionary history and diversity of histone reader fusions to MSH6. We identified putative MSH6 orthologs in 4075 eukaryotic organisms, in which the presence and absence of histone reader could be confidently called, emphasizing the detection and removal of false negatives (Materials and Methods, Supplemental Table 1). We observed no cases where MSH6 orthologs showed evidence of containing both PWWP and Tudor domains. The vast majority (99% and 97% of PWWP and Tudor, respectively) of species in which MSH6 histone reader domains were identified showed evidence of histone readers from both blastp and ab initio domain prediction (Supplemental Table 2, Supplemental Table 3).

These data provided the basis to examine the distribution of MSH6 histone reader domains across the eukaryotic tree of life and characterize their evolutionary origins and history (Figure 2, Figure S1, Figure S2). We found that fungi, along with most taxa diverging from the last eukaryotic common ancestor, lack PWWP or Tudor domains in their MSH, as also seen in the MutS homologs of bacteria and archaea (Figure 1a, Figure S3). This finding is consistent with a reader-less MSH6 being the ancestral state in eukaryotes, supporting independent evolutionary histories of PWWP and Tudor domain fusions to MSH6.

Figure 2.

Metazoa evolved MSH6-PWWP

MSH6-PWWP domain fusions were only found in Eumetazoa, with the distribution of presence/absence suggesting either multiple gains or loss events. While the presence of the PWWP domain is highly conserved in Deuterostomes (98%), major Protostome clades, such as Endopterygota (flies, beetles, wasps, ants) and Nematodes, lack a histone reader (Figure 2, Figure S1, Figure S2). Our filtering criteria excluded species with evidence of potential false negatives (MSH6 incorrectly identified as lacking a reader) due to possible genome annotation errors (Materials and Methods). Still, individual species that lack evidence of a histone reader in MSH6 are interpreted cautiously until they can be independently verified to be free of assembly or annotation artifacts. Nonetheless, clear phylogenetic patterns exemplified by whole clades lacking any MSH6 reader are likely reliable, along with confirmed reader-less MSH6 orthologs observed in model organisms such as Drosophila melanogaster, Caenorhabditis elegans, and Saccharomyces cerevisiae (Figure 1a-b, Figure S1). These findings are also comparable to broader evolutionary trends in Protosomes, where losses of a number of DNA repair proteins have been observed in various taxa^68,69.

Determining the ancestral state of MSH6 reader presence in Eumetazoa and whether the variability observed across taxa represents multiple gains versus multiple losses is challenging due to the uncertainty about the relative rate at which these transitions might occur. Losses may be more mutationally likely (i.e., deletion events) than gains (e.g., translocations from another PWWP-containing protein). Ancestral state reconstruction assuming equal rates (ER) suggests multiple gain events, whereas all-rates-different (ARD) models indicate multiple losses (Figure S1, Figure S2). The discovery of variation in MSH6 histone reader domains enforces the importance of exercising caution in generalizing experimental findings on mutation rate heterogeneity and DNA repair mechanisms across taxa ⁷⁰. This diversity also has implications for understanding variation in mutation rates and spectra across the Tree of Life ¹³. Once lost, an MSH6 reader may be improbable to be re-gained.

Plants and their relatives evolved MSH6-Tudor

MSH6 fusions to Tudor domains are unique to land plants and other photosynthetic organisms, including green algae, brown algae, and cryptomonad algae. In land plants (Streptophyta), MSH6-Tudor fusions are highly conserved (96%) (Figure 2, Figure S1, Figure S2). Some green algae (Chlorophyta), along with some diatoms, brown algae (Stramenopiles), and other SAR (Stramenopiles + Alveolata + Rhizaria) lineages lack the Tudor domain (Figure 2). It is interesting to note that the species lacking MSH6-Tudor includes Chlamydomonas reinhardtii, a model for algal genomics, which also lacks the H3K4me1 enrichment of gene bodies that is otherwise conserved in higher land plants ^44,71. We also do not observe MSH6-Tudor in any red algae (Rhodophyta). Like in animals, the ancestral state of the MSH6 domain presence is somewhat uncertain, especially given that several groups containing Tudor domains, including Stramenopiles, have potentially gained genes from green algae via ancient endosymbioses ^72,73. However, the phylogenetic relationship of MSH6 orthologs does not support horizontal gene transfer from green algae: a maximum likelihood tree of MutS domains I-V shows the MSH6 ortholog of Stramenopiles sister to other SAR taxa that lack a Tudor domain (Figure S4).

Plants (Viridiplantae) contain a unique MutS ortholog, MSH7, which arose via an ancient duplication of MSH6 ^74–76. We identified putative MSH7 orthologs in plants and queried whether Tudor domains were observed (Materials and Methods). We found no plant MSH7 orthologs containing Tudor domains, suggesting that the Tudor was acquired by MSH6 after the duplication that gave rise to MSH7 (Figure S6). or that MSH7 lost the Tudor domain during or after duplication in the common ancestor of plants. Thus, the MSH6-Tudor fusion is probably ancient and predates the separation of the major clades of Viridiplantae, possibly in the common ancestor of Viridiplantae, crytomonads, and SAR.

Testing for additional fusions of MutS homologs to histone readers

We used ab initio domain prediction to investigate whether additional, previously undescribed, histone reader fusions have occurred with MSH6. This analysis yielded no known candidate histone reader domains beyond Tudor and PWWP (Supplemental Table 3). We further expanded our search to other potential MutS homologs, and while we did identify a number of intriguing fusions to non-MutS domains, we found none with obvious histone readers.

For example, we identified a fusion between MSH3 and N-acetylglucosaminyl transferase component (Gpi1) (e.g., UniProt: G3Y5U7) shared across Aspergillus fungi, with unknown functional importance. We also found a fusion of MSH6 with ESCO acetyltransferase and Zinc-Finger binding domains in the Ustilaginomycetes (smut fungi) (Figure S5). These domains are not known to be histone readers per se; however, ESCO proteins have been previously described to bind chromatin by DNA-binding of a charged N-terminus intrinsically disordered region ⁷⁷. While not histone readers, these findings are a reminder of the potential for undiscovered fusions between chromatin interacting domains and other proteins involved in DNA repair, a potentially valuable direction for future research.

In summary, we find compelling evidence of convergent evolution in the origins of distinct histone-binding domains, specifically to MSH6 in eukaryotes⁶⁷. These fusions of histone readers to MSH6 provide a unique opportunity to study the evolutionary dynamics of epigenome-recruited DNA repair systems at a Tree of Life scale.

Conservation of domain architectures and sequence

We next set out to further test if the PWWP and Tudor domain fusions to MSH6 exhibit evidence of being subject to natural selection. Patterns of conservation can also provide mechanistic insight into the adaptive value and biochemical function of MSH6-associated histone readers.

Constraint on linker region

In both human and Arabidopsis, the MutS repair domains are separated from the histone reader by a (MobiDB-lite annotated) intrinsically disordered “linker” region (Figure 1a-b, Figure 3a). To better understand this region, we examined its sequence composition and length across all species containing the respective domains (Figure 3a). Regardless of histone reader (PWWP or Tudor), we found that MSH6 linker regions were significantly (p<2×10^-16) enriched for the charged amino acids lysine (K), aspartate (D), and glutamate (E), consistent with a role in promoting disordered protein regions and potentially other molecular interactions⁷⁸. As expected, these linker regions were >36 times more likely to be predicted as intrinsically disordered than histone reader or MutS-containing protein regions in silico⁷⁹ (Z=782.5, p<2×10^-16).

Figure 3.

We also observed a notable degree of conservation in linker sequence lengths (Figure 3a). It is difficult to establish a strong null model with which to compare disordered region length variation against neutrality. However, we did note that Tudor domains are significantly closer (shorter linker region) to MutS domain I than PWWP (Tudor mean = 201 a.a., PWWP mean = 217 a.a, t = 13.527, p-value < 2.2×10^-16, Figure 3a). This is interesting given that the Arabidopsis MSH6 Tudor and human PWWP domains bind to methylated lysines on the 4th (most distal to histone/DNA) and 36th (more proximate) lysines, respectively. The length of the disordered and potentially flexible linker region may be constrained to tether MSH6 at an optimal distance to chromatin, a hypothesis necessitating further experimental testing.

Constraint on histone interacting domain amino acids

We then investigated patterns of evolutionary constraint on the reader domains themselves. If the histone PTM binding affinities are subject to selection, we would expect to observe the conservation of amino acids involved in interactions with the H3 peptides, as seen in the evolution of sites involved in protein-protein interactions⁸⁰. We compared MSH6 Tudor and PWWP domains across all species. For each amino acid position within the domains, we identified the consensus amino acid (the most common residue) and calculated both the conservation (the frequency of the consensus amino acid) and Shannon entropy (a measure of amino acid diversity at each position).

For both Tudor and PWWP domains, amino acids involved in H3 PTM binding were among the most conserved residues, consistent with purifying selection on histone reader binding affinities (Figure 3b-c). We estimated the distance between amino acids and the H3 peptide by superimposing predicted structures to experimentally determined PWWP and Tudor structures that captured H3 PTM binding, H3K36me3 and H3K4me1, respectively (Materials and Methods). We found that amino acids predicted to be in close proximity to the H3 peptides are significantly more conserved, both in PWWP and Tudor reader domains (PWWP: R=0.55, p=2×10^-9; Tudor: R=0.56, p=4×10^-6). (Figure 3b-c). Thus, evidence of purifying selection was enriched at the histone-binding interface. Several distal amino acids were also highly conserved. This conservation may reflect roles in proper folding or other interactions, such as between PWWP and DNA directly ⁸¹.

In summary, evolutionary conservation observed in MSH6 domain architectures, as well as Tudor and PWWP domains themselves, exhibit signatures of selection, especially for histone binding.

Species with MSH6 histone readers show distinct genomic traits

The preceding analyses revealed convergent evolution and conservation of MSH6 histone readers (Figure 3). However, not all Eukaryotes gained (or retained) MSH6 histone readers (Figure 2, Figure S3, Figure S4). To study this variability further, we tested for correlated evolution of MSH6 reader presence/absence with genomic architectures and other traits toward understanding what might be behind the observed phylogenetic patterns.

Contrasting genomic traits

Theory and experimental work have observed that hypermutators and changes in mutation biases can be favored by selection under certain conditions ^82–84. Whether this can help explain variability in MSH6 reader presence in eukaryotes is a compelling, albeit challenging, question to answer. Previous theory based on the drift-barrier hypothesis predicts that the adaptive value of promoting DNA repair in coding regions may be contingent on genome architectures¹³. To test this, we compared a suite of genomic traits between species (Figure 4, Figure S7, Table S4). We found that species with MSH6 fused to histone readers have (more than an order of magnitude) larger genomes. Previous investigations have described substantial evolutionary variation in intron/exon architectures across eukaryotes ^85,86. After accounting for phylogeny, we found significant co-evolution of MSH6 readers with genome size, marked especially by greater intron content, both in size and number per gene (Figure 4b-c, Figure S7a). This was also the case when only Protosome animals were examined, where the loss of MSH6-PWWP may have occurred (Figure S7b). These observations are consistent with the hypothesis proposed by Lynch et al. (2023)¹³ that “the benefits of targeting DNA repair to coding regions may be greatest for species with a large fraction of non-coding genome content.”

Figure 4.

For species where data were available^87,88, we also compared several metrics relevant to life history and population genetics. Species without MSH6 histone reader domains have smaller body sizes, greater estimated population sizes, and experience significantly lower germline SNP mutation rates (Figure 4d). While interesting, these associations were not significant after accounting for phylogeny. Measurements of these characteristics across more species with a broader taxonomic range may be needed for more robust tests for co-evolution with reader domains in phylogenetic contexts.

In summary, we observed significant co-evolution with genome features like genome size and intron number per gene and found differences in population and organismal characteristics in species with and without reader fusions to MSH6. Whether these reflect a cause or consequence of epigenome-recruited repair or instead reflect correlated responses to shared evolutionary forces remains an open question. Still, these results highlight interesting connections between DNA repair systems and the evolution of genome architectures. These patterns are consistent with the selective benefits and costs of such systems being contingent on factors like non-coding genome size, as previously hypothesized¹³. More sampling of traits and mutation rates, coupled with theoretical analyses, are needed to integrate these findings through the models of the drift-barrier hypothesis to better determine what evolutionary processes underlie these associations^{10–13,49,89}.

These findings inspire a number of outstanding questions: How conserved are epigenomes across the tree of life? Are there yet-to-be-discovered mechanisms of epigenome-recruited DNA repair systems? Are there biological or evolutionary costs associated with epigenome-recruited mismatch repair? What is the relative importance of these mechanisms for somatic versus germline mutation in different organisms? Do the distributions of fitness effects between coding and non-coding sequences differ among organisms? Finally, what is the extent of mutation rate heterogeneity across different organisms’s genomes? These and other questions, in light of the findings presented here, provide a roadmap for future empirical and theoretical work.

Conclusions

The recruitment of DNA mismatch repair by H3K36me3 (MSH6-PWWP) in humans and by H3K4me1 in Arabidopsis (MSH6-Tudor), now investigated at a tree of life scale, reveals a compelling example of evolutionary convergence⁶⁷. Previous experiments support the role of these mechanisms in recruiting mismatch repair to functionally important genomic regions, at least in Humans and Arabidopsis, with consequences on mutation rates ^{17,37,38,48,56}. The phylogenetic distribution of PWWP and Tudor fusions to MSH6 support their independent evolutionary origins, a hallmark of evolution by natural selection. Patterns of evolutionary conservation reveal selection for histone binding and potentially on the regions that link reader domains with MutS repair domains, providing hypotheses about biochemical functions. We also find evidence of clades that never gained or have lost MSH6 histone readers, especially of PWWP in major clades with Protosome animals, coupled with differences in genome architectures and other characteristics. This co-evolution may provide insight into the evolutionary forces acting on these mechanisms, reflecting their costs and benefits under different conditions. In summary, the convergent evolution of epigenome targetting in MSH6 suggests conditionally adaptive mechanisms, inspiring further questions about the mutational consequences and evolution of DNA repair systems across the tree of life.

Materials and methods

Defining MSH6 orthologs

We downloaded all annotated (both NCBI RefSeq and GenBank) public eukaryotic genomes and proteomes on NBCI (4,938 species) on August 9, 2023 (Figure S8). For each proteome, we queried using the protein sequence of human MSH6 (blastp with eval threshold of 0.01). Each of the resulting hits was blasted back against the human proteome. Proteins were considered potential orthologs of human MSH6 if the top hit in the human proteome was human MSH6. We thus removed proteins that show evidence of belonging to other members of the MSH protein family, though these were retained for downstream analysis to test for the presence of histone readers. We also performed the same reciprocal blastp analysis with Arabidopsis MSH6. Proteins were considered MSH6 only if the top hit of the reverse blast was the Arabidopsis MSH6 protein. These predicted MSH6 orthologs for each species were then defined as those that were classified as orthologs of both human and Arabidopsis MSH6. Finally, to further confirm MSH6 orthology, we ran ab initio of each predicted MSH6 ortholog using InterProScan with all available databases⁹⁰. The final set of MSH6 orthologs was determined by requiring them to be called MSH6-specific orthologs by both reciprocal blastp queries and also contain ab initio annotated MutS domains. This removed potential false positives from downstream analysis in which non-MSH proteins containing PWWP or Tudor domains would be incorrectly called. Because genome annotation errors can lead to false negatives of MSH6 presence in a species, we ignored all species lacking a definitive MSH6 ortholog in all downstream analyses.

Identifying Tudor and PWWP domains in MSH6 orthologs

For each species proteome, we queried for the PWWP domain of the human MSH6 protein (Figure S8). We also queried for the Tudor domain of the Arabidopsis MSH6 protein. From these results, we annotated MSH6 orthologs as containing Tudor and PWWP domains if the same protein was detected by MSH6 blastp searches (preceding step) and significant (e score) matches for specific queries of PWWP and Tudor domains. To further improve the annotation of Tudor and PWWP domains, we analyzed the results of ab initio domain prediction from InterProScan with “CDD,” “Pfam,” “SMART,” and “ProSiteProfiles” databases. MSH6 orthologs were thus annotated as containing a Tudor domain if they contained a significant match to the Tudor domain query or if they were annotated as having a Tudor domain by InterProScan. Likewise, MSH6 orthologs were thus annotated as containing a PWWP domain if they contained a significant match to the PWWP domain query or if they were annotated as having a PWWP domain by InterProScan. Species were classified as containing a Tudor domain or a PWWP domain in MSH6 if any of their MSH6 orthologs contained such domains. This approach was conservative in that it prioritized against calling false negatives. That is, to preclude misclassifying species as lacking a histone reader when, in reality, one is present. For example, in a polyploid species, we might find multiple putative MSH6 orthologs, which vary in the presence/absence of a reader domain. If at least one MSH6 ortholog contained a reader domain, it was classified as having that domain at the species level.

Removing species with evidence of genome annotation errors

We considered several possible scenarios from which genome annotation errors could cause false negatives in identifying reader domains in MSH6 orthologs (Figure S8).

First, genome annotations could lead to incorrect splits of MSH6 into several genes. In this case, the MSH6 gene may be found adjacent to a “gene” that contains a Tudor domain or PWWP domain. To identify such cases, we compared the location of MSH6 orthologs to results from querying (blastp) Tudor and PWWP domains. Cases in which a neighboring gene (within 5kb) contained a Tudor or PWWP domain were classified as ambiguous and removed from subsequent analyses.

Second, genes could be incompletely annotated such that transcribed regions (e.g., whole exons) are excluded from the predicted gene model. We thus used tblastn to query Tudor and PWWP domains on the entire genome, which were translated into all reading frames of each genome. Species in which a sequence matching a Tudor or PWWP domain within 5 kb of an MSH6 ortholog were classified as ambiguous and removed from subsequent analyses.

Finally, we considered the possibility that MSH6 orthologs with histone reader domains may be missed from annotation entirely. To identify and remove such cases, we used tblastn to query MSH6 protein sequences from human and Arabidopsis orthologs. Cases where potential unannotated orthologs (matching MSH6 > 200aa, to exclude matches to potential reader domain itself, such as in other proteins which contain these domains) were defined as cases where a tblastn query for Tudor or PWWP domain was also found within 5 kb of MSH6 tblastn result. For such cases, the species’ domain call was classified as ambiguous and excluded from downstream analyses.

Despite these steps, we could not preclude other sources of misannotation that might lead to false negatives for domain presence in a single species. For example, fragmentation during genome assembly could lead to edge cases where the repair domain (i.e., MutS domains) of MSH6 and its reader domain (if present) were assembled on different scaffolds or contigs, making it impossible to accurately control against the subsequent effects of misannotation. Similar issues might arise from miss-assemblies that cause false separation from MSH6 repair domains from reader domains. Thus, we interpret single species lacking a reader domain, especially when nested within a clade in which closely related species that contain such a domain, with caution until they can be validated.

MSH6 homologs in non-eukaryotes

We first examined the structure and domain architectures of MutS homologs in Escherichia coli (strain K12) (Uniprot: P23909) and Thermoplasmatales archaeon (Uniprot: A0A256YTS1), which showed no reader domains (in structure nor in Interproscan annotations). We queried, with blastp, the entire NCBI Archaeal protein database for the Arabidopsis and Human MSH6 proteins and checked for any that aligned to Tudor or PWWP regions of MSH6. We also queried Archaeal proteomes for the Tudor and PWWP domains directly, which yielded no hits. We repeated this specifically for Asgard Archaea, described as the closest relatives to Eukaryotes ⁹¹.

Ancestral state reconstruction

To estimate the likely ancestral states, we used the ace function from ape in R, which estimates marginal ancestral states for a discrete trait (in this case, MSH6 reader domain state) with the re-rooting method described by Yang et al. (1995)⁹². For this, we used the time-calibrated eukaryotic species tree (2004 species) available from TimeTree v5 ⁹³, setting the edge.length + 1×10^-5 to account for 116 branches with length=0. We compared ancestral states estimated with “equal rates” (ER) and “all-rates-different” (ARD) models. We also estimated ancestral states for “PWWP” vs. “Tudor” vs. “None,” as well as “PWWP” vs. “None” and “Tudor” vs. “None” separately.

Natural variation in Tudor and PWWP domains

Tudor and PWWP domains (± 5aa from respective blastp match) of MSH6 from all species were extracted from protein fasta files using blastp hits to define boundaries of domains. The resulting unique sequences were aligned with MUSCLE with the msa function from the R package msa. We used the Arabidopsis Tudor and human PWWP as reference frames to define alignment positions along the domains and thus ignored insertions relative to these. From these alignments, we calculated the frequency of each amino acid at each position and the corresponding Shannon entropy H(X) as

Where p(x) represents the frequency of an amino acid x from alphabet X (20 amino acids and gap) at position X. We also calculated conservation at each position as the proportion of the most frequent amino acid, with these amino acids representing, together, the consensus sequence across the domain.

Modeling Structure of Tudor and PWWP domains interface with histone PTMs

Intrinsically disordered regions (IDRs) in all MSH6 proteins were predicted with ALBATROSS ⁷⁹. To examine the size, sequence composition, and IDR enrichment of the linker sequence separating MutS domains from histone readers, we used results from InterProScan to identify positions of well-annotated protein domains (one each of MutS I through V, Tudor, and PWWP, if present). Amino acid enrichment in linker regions was estimated by comparing amino acid counts within linker regions across species with the expectation of uniformity in amino acid frequencies. IDR enrichment was estimated by comparing the frequency of overlaps between linkers and predicted IDRs with overlaps between repair or reader domains with predicted IDRs. Given the disparities in species numbers with each domain architecture, we randomly selected 200 species with each domain architecture to visualize (Figure 3a). These were plotted with domain positions centered around the start of MutS domain I. We compared the estimated size of the linker sequence between all PWWP and Tudor domain-containing species by measuring the length between the InterProScan annotated reader and MutSI domains.

Next, we approximated the distance between reader domain amino acids and H3 peptides. To obtain these estimates, we superimposed the Human MSH6 PWWP domain predicted by Alphafold ⁹⁴ onto PDB 5ciu ⁹⁵, the experimental structure of H3K36me3 bound by the DNMT3B PWWP domain, using the Pairwise Structure Alignment tool from RCSB with the TM-align method (https://www.rcsb.org/alignment). We superimposed the Arabidopsis MSH6 Tudor domain to PDB 7DE9⁴⁵, the experimental structure of H3K4me1 bound by the PDS5C(RDM15) Tudor domain. From the resulting structural alignments, we estimated the distance of each amino acid in the histone reader domain to the corresponding H3 peptide chain by calculating the minimum Euclidean distance between atoms in each amino acid to atoms in the H3 peptide. We compared the relationship between the estimated distances of each amino acid with their evolutionary constraint (Shannon entropy). Similar results were found when we compared distance with conservation.

Comparisons between species with and without reader domains in MSH6

For each species’ genome and its corresponding annotations (averaging summary statistics across annotations when multiple were available), we estimated the total genome size by the assembly length in base pairs and calculated the total base pairs of protein-coding sequence, total base pairs of non-protein-coding sequences, the average number of coding exons per protein-coding gene, average length of coding exons, average length of introns per gene, number of introns, average proportion of introns in gene bodies, and average length of proteins. To validate genome size, we compared genome assembly size with previously reported C-values ⁹⁶ and found high concordance (r=0.97, p<2×10^-16). As a further control against assembly artifacts influencing results, we also tested contig N50 values as reported in NCBI in the same manner as other variables. Other traits: Lifespan was compiled by the AnAge database ⁹⁷. Germline mutation rates previously compiled by Wang & Obbard (2023)⁸⁸. Body size (“log10_body_mass_pred”) and population size (“log10_popsize”) estimates were generated by

Buffalo (2021)⁸⁷. Of these, we used both log₁₀ transformed and ranked values to account for non-normality. We first performed binomial generalized linear models contrasting species with and without Tudor or PWWP histone reader domains in R with glm(family=“ binomial”). We then performed phylogenetically constrained contrasts using the time-calibrated eukaryotic species tree (2434 species) reported from TimeTree v5 ⁹³, setting the edge.length + 1×10^-5 to account for 116 branches with length=0. We compared histone reader presence/absence classification as a function of the above metrics with the phyloglm function (btol = 1000, method=“logistic_MPLE”, log.alpha.bound=0) from the R package phylolm V2.6.2 and the binaryPGLMM (default settings) function from the ape V5.7.1 package ^98,99. We also modeled the relationship between readers and traits with comparative.data (vcv=TRUE, vcv.dim=2) and pgls functions (lambda=‘ML’) from the geiger V2.0.11 package in R and phylolm function from the phylolm package (model=“kappa”).

Data availability

Genomes and annotations analyzed here are available on NCBI and listed in Table S1. A summary of data used to classify MSH6 histone readers is found in Table S2. Interproscan results for putative MutS homologs are found in Table S3. Species-level genome and trait information is found in Table S4.

Code availability

Code for this research is maintained at github.com/greymonroe/MSH6_evolution.

Supplemental figures

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Figure S5.

Figure S6.

Figure S7.

Figure S8.

Acknowledgements

We thank Dan Sloan, Detlef Weigel, and Michael Lynch for valuable discussions of this work. The Monroe Lab (J.G.M) is supported by FFAR grant ICRC20-0000000014, USDA-NIFA grant 108681-Z5327202, a UC Davis STAIR grant, NSF Grants 2317191 and 2338236, and CPRB grant HG-2022-35. The research was conducted at the University of California Davis, which is located on land that was the home of the Patwin people for thousands of years (https://diversity.ucdavis.edu/land-acknowledgement-statement).

Additional files

Supplementary Tables 1-4. Table S1. Summary of presence/absence of histone reader domains and genomic traits across species Table S2. Evidence for the presence/absence of MSH6 histone reader domain Table S3. Interproscan results Table S4. Species-level genomic and other characteristics