CDCA7 is characterized by the unique zf-4CXXC_R1 domain (Pfam PF10497 or Conserved Domain Database [CDD] cl20401) (Lu et al., 2020; Mistry et al., 2021). Conducting a BLAST search using human CDCA7 as a query sequence against the Genbank protein database, we realized that no zf-4CXXC_R1-containing proteins are identified in the classic model organisms Drosophila melanogaster, Caenorhabditis elegans, Schizosaccharomyces pombe and Saccharomyces cerevisiae, which are all also known to lack any DNMTs and genomic 5mC (Zemach et al., 2010). However, we identified a protein with a zf-4CXXC_R1 motif in the bumblebee Bombus terrestris and the thale cress A. thaliana, which have both maintenance and de novo DNMTs (Bewick et al., 2017b; Chan et al., 2005; Li et al., 2018; Zemach et al., 2010; Figure 1 and Figure 1—source data 1). Based on the reciprocal best hits (RBH) criterion using human HELLS as a query sequence (see Methods) (Ward et al., 2014), HELLS-like proteins were identified in A. thaliana (DDM1) and S. cerevisiae (Irc5) in line with previous reports (Litwin et al., 2017), as well as in B. terrestris. No clear HELLS-like proteins were identified in D. melanogaster, C. elegans, or S. pombe. This pilot-scale analysis based on the RBH criterion led us to hypothesize that the evolutionary maintenance of CDCA7 is linked to that of HELLS and DNMTs. In order to statistically validate this hypothesis, we set out to systematically identify and classify homologs of CDCA7, HELLS and DNMTs in a broad range of evolutionary lineages.

We first characterized evolutionary conservation of CDCA7 family proteins in vertebrates, where 5mC, DNMT1, and DNMT3 are highly conserved. A BLAST search against the Genbank protein database identified two zf-4CXXC_R1 domain-containing proteins, CDCA7/JPO1 and CDCA7L/R1/JPO2, throughout Gnathostomata (jawed vertebrates; Figure 2A, Figure 2B, Figure 2—figure supplement 1, and Figure 1—source data 1). In frogs (such as Xenopus, but not all amphibians), and some fishes (such as Astyanax mexicanus and Takifugu rubripes), a third paralog CDCA7e exists. CDCA7e is the only CDCA7-like protein that can be detected in Xenopus eggs (Jenness et al., 2018), and thus likely represents a form specific to oocytes and early embryos in these species. Among twelve conserved cysteine residues originally reported in the zf-4CXXC_R1 domain (Ou et al., 2006), the 12th cysteine residue is not conserved in Rhincodon typus (whale shark) CDCA7 and in Xenopus laevis CDCA7e. In general, the 12th cysteine residue of the zf-4CXXC_R1 domain is least conserved among CDCA7 homologs within and outside of vertebrates, such that we do not consider it a key component of the zf-4CXXC_R1 domain (see below). Considering that the jawless fish Petromyzon marinus (sea lamprey) and other invertebrates commonly possess only one CDCA7 family gene (Figure 2, see Figure 6 and Figure 1—source data 1), the CDCA7 paralogs may have emerged in jawed vertebrates. Overall, the zf-4CXXC_R1 sequence is highly conserved among vertebrate CDCA7 homologs, including three residues that are mutated in ICF3 patients (R274, G294 and R304 in human CDCA7) (Figure 2B; Thijssen et al., 2015). Note that these amino acid positions of the ICF3 mutations in human CDCA7 are based on the previously reported sequence of an NP_665809 (Thijssen et al., 2015), which is annotated as isoform 2 (371 amino acids), whereas we list the isoform 1 (NP_114148) with 450 amino acids in this study (Figure 2A).

While the presence of four CXXC motifs in the zf-4CXXC_R1 domain is reminiscent of a classic zinc finger-CXXC domain (zf-CXXC, Pfam PF02008), which binds to nonmethylated CpG (Long et al., 2013), their cysteine arrangement is distinctly different, perhaps reflecting the capacity of zf-4CXXC_R1 domain to recognize nucleosomes (Jenness et al., 2018) and potentially specific epigenetic marks or DNA sequences. In vertebrate CDCA7 paralogs, 11 conserved cysteines are arranged as CXXCX₁₀CX₄CX₇CXXCX₁₉CXXCX₃CXCXXC. In contrast, in the classic zf-CXXC domain eight cysteines are arranged as CXXCXXCX_4-5CXXCXXCX_8-14CX₄C (Long et al., 2013). Apart from the zf-4CXXC_R1 domain, vertebrate CDCA7 family proteins often, but not always, contain one or two Lens epithelium-derived growth factor (LEDGF)-binding motif(s) (Figure 2A and Figure 2C), defined as ([E/D]XEXFXGF; Tesina et al., 2015). It has been reported that human CDCA7L and CDCA7 both interact with c-Myc but apparently via different regions, a leucine zipper and a less-defined adjacent segment that overlaps with a bipartite nuclear localization signal, respectively (Gill et al., 2013; Huang et al., 2005). This leucine zipper sequence is highly conserved among vertebrate CDCA7 family proteins (Figure 2A and Figure 2D). In contrast to zf-CXXC domain-containing proteins such as KDM2A/B, DNMT1, MLL1/2, and TET1/3, the vertebrate CDCA7 family proteins do not contain any predicted enzymatic domains (Huang et al., 2005; Maertens et al., 2006; Tesina et al., 2015).

A BLAST sequence homology search identified three classes of zf-4CXXC_R1 domain-containing proteins in Arabidopsis thaliana (Figure 3). Class I proteins have one zf-4CXXC_R1 domain with no other detectable domains listed in CDD or Pfam/InterPro (Paysan-Lafosse et al., 2023). Strikingly, all three ICF-associated residues (R274, G294, and R304 in human CDCA7) as well as all eleven characteristic cysteines of the zf-4CXXC_R1 domain are conserved, with the exception that the position of the fourth cysteine is shifted two residues toward the N terminus. We define the domain with the conservation of the signature cysteine residues and three ICF-associated residues as ‘class I zf-4CXXC_R1’ and hypothesize that proteins with class I zf-4CXXC_R1 are CDCA7 orthologs (including their highly homologous paralogs in this report).

Class II proteins contain a zf-4CXXC_R1 domain, a DDT domain and a WHIM1 domain. These proteins were previously identified as DDR1-3 (Dong et al., 2013). DDT and WHIM1 domains are commonly found in proteins that interact with SNF2h/ISWI (Aravind and Iyer, 2012; Li et al., 2017; Yamada et al., 2011). Indeed, it was reported that Arabidopsis DDR1 and DDR3 interact with the ISWI orthologs CHR11 and CHR17 (Tan et al., 2020). Among the eleven cysteine residues in the zf-4CXXC_R1 domain of these proteins, the position of the fourth residue is shifted towards the C-terminus. The ICF-associated glycine residue (G294 in human CDCA7, mutated to valine in ICF3 patients) is replaced by isoleucine. We define the zf-4CXXC_R1 containing a substitution at this glycine residue as class II zf-4CXXC_R1.

Class III proteins are longer (~1000 amino acid) and contain an N-terminal zf-4CXXC_R1 domain and a C-terminal JmjC domain (Pfam, PF02373), which is predicted to possess demethylase activity against histone H3K9me2/3 (Saze et al., 2008). While all 11 cysteine residues can be identified, there are deletions between the 4th and 5th cysteine and 6th and 7th cysteine residues. None of the ICF-associated residues are conserved in the class III. One of these class III proteins is IBM1 (increase in bonsai mutation 1), whose mutation causes the dwarf “bonsai” phenotype (Saze et al., 2008), which is accompanied with increased H3K9me2 and DNA methylation levels at the BONSAI (APC13) locus. Double mutants of ddm1 and ibm1 exacerbate the bonsai phenotype, indicating that DDM1 and IBM1 act independently to regulate DNA methylation (Saze et al., 2008). Another class III protein is JMJ24, which harbors a RING finger domain in addition to the 4CXXC and JmjC domains. This RING finger domain promotes ubiquitin-mediated degradation of the DNA methyltransferase CMT3, and thus opposes DNA methylation (Deng et al., 2016).

Homologs of these three classes of CDCA7 proteins found in Arabidopsis are widely identified in green plants (Viridiplantae), including Streptophyta (e.g. rice, maize, moss, fern) and Chlorophyta (green algae; Figure 3—figure supplement 1 and Figure 1—source data 1). Other variants of zf-4CXXC_R1 are also found in Viridiplantae. In contrast to green plants, in which the combined presence of HELLS/DDM1-, CDCA7- and DNMT- orthologs is broadly conserved, no zf-4CXXC_R1-containing proteins can be identified in red algae (Rhodophyta; Figure 1—source data 1, see Figure 5 and Figure 5—figure supplement 1).

Although S. pombe and S. cerevisiae genomes do not encode any CDCA7 family proteins, a BLAST search identified various fungal protein(s) with a zf-4CXXC_R1 domain. Among the zf-4CXXC_R1-containing proteins in fungi, 10 species (Kwoniella mangroviensis, Coprinopsis cinere, Agaricus bisporus, Taphrina deformans, Gonapodya prolifera, Basidiobolus meristosporus, Coemansia reversa, Linderina pennispora, Rhizophagus irregularis, Podila verticillate) harbor a class II zf-4CXXC_R1 domain with two notable deviations (Figure 4A). First, the space between the third and fourth cysteine residues is variable. Second, the fifth cysteine is replaced by aspartate in Zoopagomycota (Basidiobolus meristosporus ORX82853, Coemansia reversa PIA14937, Linderina pennispora ORX71196) (Figure 4). As these proteins do not contain any additional CDD-annotated domain like vertebrate and plant CDCA7 orthologs (except for Taphrina deformans CCG81269, which also has a CHROMO domain), we define this fungal protein family as CDCA7F, which forms a distinct clade in our phylogenetic analysis of zf-4CXXC_R1 sequence alignment (Figure 2—figure supplement 1 and Figure 3—figure supplement 1).

Besides CDCA7F, several fungal species encode a protein with a diverged zf-4CXXC_R1 domain, including those with a JmjC domain at the N-terminus (Figure 4B, Figure 1—source data 1). This composition mimics the plant class III proteins, for which the JmjC domain is located at the C-terminus. Among these proteins, it was suggested that Neurospora crassa DMM-1 does not directly regulate DNA methylation or demethylation but rather controls the deposition of histone H2A.Z and/or H3K56 acetylation, which inhibit spreading of heterochromatin segments with methylated DNA and H3K9me3 (Honda et al., 2010; Zhang et al., 2022). Thus, the known roles of these divergent variants of zf-4CXXC_R1 are not directly associated with DNA methylation.

To systematically identify CDCA7 and HELLS homologs in the major eukaryotic supergroups, we conducted a BLAST search against the NCBI protein database using human CDCA7 and HELLS protein sequences. To omit species with a high risk of false negative identification, we selected species containing at least 6 distinct proteins with compelling homology to the SNF2 ATPase domain of HELLS, based on the assumption that each eukaryotic species is expected to have 6–20 SNF2 family ATPases (Flaus et al., 2006). Indeed, even the microsporidial pathogen Encephalitozoon cuniculi, whose genome size is a mere 2.9 Mb, contains six SNF2 family ATPases (Flaus et al., 2006). As such, we generated a panel of 180 species encompassing all major eukaryote supergroups (5 Excavata, 18 SAR [2 Rhizaria, 6 Alveolata, 10 Stramenopiles]), 1 Haptista, 1 Cryptista, 15 Archaeplastida [3 Rhodophyta and 12 Viridiplantae], 4 Amoebozoa, 136 Opisthokonts [34 Fungi, 3 Holozoa, and 99 Metazoa] (Figure 5—figure supplement 1, and Figure 1—source data 1).

To annotate HELLS orthologs, a phylogenetic tree was constructed from a multiple sequence alignment of the HELLS homologs identified based on the RBH criterion alongside other SNF2-family proteins of H. sapiens, D. melanogaster, S. cerevisiae, and A. thaliana. If HELLS orthologs are correctly identified (i.e. without erroneously including orthologs of another SNF2-subfamily) they should cluster together in a single clade. However, the sequence alignment using the full-length protein sequence failed to cluster HELLS and DDM1 in the same clade (Figure 5—figure supplement 2). Since HELLS and other SNF2-family proteins have variable insertions within the SNF2 ATPase domain, multiple sequence alignment of the SNF2 domains was then conducted after removing the insertion regions, as previously reported (Flaus et al., 2006). By this SNF2 domain-only alignment method, all HELLS orthologs formed a clade, separated from CHD1, ISWI, SMARCA2/4, SRCAP, and INO80 (Figure 5—figure supplement 3). An independent phylogenetic tree construction with a maximum-likelihood based method and 1000 bootstrap replicates confirmed these assignments (see Methods, Figure 5—source data 1 and Figure 5—source data 2).

A BLAST search with the human CDCA7 sequence across the panel of 180 species identified a variety of proteins containing the zf-4CXXC_R1 domain, which is prevalent in all major supergroups (Figure 5, Figure 5—figure supplement 1, and Figure 1—source data 1). Each of these identified proteins contains only one zf-4CXXC_R1 domain. The resulting list of CDCA7 BLAST hits were further classified as prototypical CDCA7 orthologs if they preserve the criteria of the class I zf-4CXXC_R1 (signature 11 cysteine residues and the three ICF-associated residues) (Figure 3—figure supplement 1). A phylogenetic tree analysis of zf-4CXXC_R1 domains from diverse species confirmed that these CDCA7 orthologs with the class I zf-4CXXC_R1 domain are clustered under the same clade (Figure 2—figure supplement 1, also see Methods, Figure 2—source data 1 and Figure 2—source data 2). CDCA7 orthologs are broadly found in the three supergroup lineages (Archaeplastida, Amoebozoa, Opisthokonta; Figure 5, Figure 5—figure supplement 1, and Figure 1—source data 1). In Excavata, the amoeboflagellate Naegleria gruberi encodes a protein that is a likely ortholog of CDCA7 with an apparent C-terminal truncation (XP_002678720), possibly due to a sequencing error (Figure 3—figure supplement 1). In SAR, CDCA7 orthologs are absent from all available genomes, except stramenopile Tribonema minus, which encodes a distantly related likely ortholog of CDCA7 (KAG5177154). These conservations suggest that the class I zf-4CXXC_R1 domain in CDCA7 was inherited from the LECA.

In addition to the class I zf-4CXXC_R1 domain, we also identified divergent zf-4CXXC_R1 domains across eukaryotes, although metazoan species only contain CDCA7 orthologs (and their paralogs such as CDCA7L and CDCA7e) with the exception of the sponge Amphimedon queenslandica, which also encodes a protein (XP_019849176) with the class II zf-4CXXC_R1 domain, DDT, and WHIM1 domain, the composition of which is similar to plant DDR1-3 (Figure 3, Figure 5, Figure 3—figure supplement 1, and Figure 1—source data 1). The Amoebozoa Acanthamoeba castellanii encodes a protein (XP_004340890) containing the class II zf-4CXXC_R1 domain as well as the CHROMO domain (cd00024) (Figure 3—figure supplement 1 and Figure 1—source data 1). This domain combination is also found in the Taphrina deformans CDCA7F, CCG81269 (Figure 4). Proteins with a combinatory presence of a diverged zf-4CXXC_R1 domain and the JmjC domain are found in plants, fungi, Amoebozoa, and Naegleria gruberi (Figure 1—source data 1).

Despite the prevalence of the zf-4CXXC_R1 domain and its variants in eukaryotes, no zf-4CXXC_R1 domain was found in prokaryotes and Archaea. This is in contrast to SNF2 family proteins and DNA methyltransferases, which can be identified in prokaryotes and Archaea (Colot and Rossignol, 1999; Flaus et al., 2006; Huff and Zilberman, 2014; Ponger and Li, 2005), pointing toward a possibility that the zf-4CXXC_R1 domain emerged to deal with unique requirement of eukaryotic chromatin.

A simple RBH approach is not practical to classify eukaryotic DNMT proteins due to the presence of diverse lineage-specific DNMTs (Huff and Zilberman, 2014). Therefore, we collected proteins with a DNMT domain within the panel of 180 eukaryote species, and then extracted the DNMT domains from each sequence (based on an NCBI conserved domains, Dcm [COG0270], Dcm super family [cl43082], AdoMet_MTases superfamily [cl17173]). Generating a phylogenetic tree based on the multisequence alignment of the DNMT domains, we were able to classify the majority of all identified DNMTs as previously characterized DNMT subtypes according to their sequence similarity (Figure 5—figure supplement 4 and ). These DNMT subtypes include DNMT1; DNMT3; the plant specific de novo DNA methyltransferases DRM1-3; the ‘true’ plant DNMT3 orthologs Yaari et al., 2019; the plant-specific CMT Bewick et al., 2017a; the fungi-specific maintenance methyltransferase Dim-2 and de novo methyltransferase DNMT4 Bewick et al., 2019a; Nai et al., 2020; the SNF2 domain-containing maintenance methyltransferase DNMT5 Dumesic et al., 2020; Huff and Zilberman, 2014; DNMT6 (a poorly characterized putative DNMT identified in Stramenopiles, Haptista and Chlorophyta) (Huff and Zilberman, 2014), and the tRNA methyltransferase TRDMT1 (also known as DNMT2; Figure 5—figure supplement 4). In this report, we call a protein DNMT3 if it clusters into the clade including metazoan DNMT3, plant DNMT3, and DRM. We also identified other DNMTs, which did not cluster into these classes. For example, although it has been reported that DNMT6 is identified in Micromonas but not in other Chlorophyta species such as Bathycoccus and Ostreococcus (Huff and Zilberman, 2014), we identified Chlorophyta-specific DNMTs that form a distinct clade, which seems to be diverged from DNMT6 and DNMT3. We temporarily called this class Chlorophyta DNMT6-like (Figure 5—figure supplement 4). Other orphan DNMTs include the de novo DNA methyltransferase DNMTX in fungus Kwoniella mangroviensis (Catania et al., 2020), and an uncharacterized DNMT in N. gruberi (XP_002682263).

The classification of homologs of CDCA7, HELLS and DNMTs across the panel of 180 eukaryotic species reveals that they are conserved across the major eukaryote supergroups, but they are also dynamically lost (Figure 5, Figure 5—figure supplement 1 and Figure 1—source data 1). We found 40 species encompassing Excavata, SAR, Amoebozoa, and Opisthokonta that lack CDCA7 (or CDCA7F), HELLS and DNMT1. Species that encode the set of DNMT1, UHRF1, CDCA7, and HELLS are particularly enriched in Viridiplantae and Metazoa. A clear exception in Amoebozoa is Acanthamoeba castellanii, whose genome also encodes these four proteins and is reported to have methylated cytosines (Moon et al., 2017). N. gruberi is an exceptional example among Excavata species, encoding a putative CDCA7 (XP_002678720) as well as HELLS; an orphan DNMT protein (XP_002682263), and a UHRF1-like protein. DNMT1, DNMT3, HELLS, and CDCA7 seem to be absent in other Excavata lineages, although Euglenozoa variants of DNMT6 and cytosine methylation are identified in Trypanosoma brucei and Leishmania major (Huff and Zilberman, 2014; Militello et al., 2008). The yellow-green algae Tribonema minus is the only SAR species that encodes a putative CDCA7 (KAG5177154), along with DNMT5 and HELLS (Figure 5, Figure 2—figure supplement 1, and Figure 3—figure supplement 1).

Among the panel of 180 eukaryote species, we found 82 species that encode CDCA7 (including fungal CDCA7F) (Figure 5—figure supplement 1 and Figure 1—source data 1 tab6). Strikingly, all 82 species containing CDCA7 (or CDCA7F) also harbor HELLS. Almost all CDCA7 encoding species also possess DNMT1. Exceptions are: the yellow-green algae T. minus and the Mamiellophyceae lineage of the green algae/Chlorophyta (Bathycoccus prasinos, Ostreococcus lucimarinus, Micromonas pusilla), which lost DNMT1 but possess DNMT5; the amoeboflagellate N. gruberi, which encodes an orphan DNMT and UHRF1, and the fungus T. deformans, which encodes UHRF1 and DNMT4. In contrast, 20 species (e.g. S. cerevisiae) possess only HELLS, while 10 species (e.g. the silk moth Bombyx mori) retains only DNMT1 among the set of DNMT1, CDCA7 and HELLS. These observations indicate that the function of CDCA7 is strongly linked to HELLS and DNMT1, such that the presence of CDCA7 depends on HELLS and DNMT1, while CDCA7 is easier to lose than DNMT1 and HELLS. Compared to the apparent HELLS/DNMT1-dependent existence of CDCA7, DNMT3 seems to be more dispensable; besides Fungi, which lack DNMT3, 16 species (e.g., the bed bug Cimex lectularius and the red paper wasp Polistes canadensis) possess the set of CDCA7, HELLS, and DNMT1 but not DNMT3/DNMT3-like proteins (Figure 5—figure supplement 1 and Figure 1—source data 1 tab6).

To quantitatively assess coevolution of DNMTs, CDCA7 and HELLS, we performed CoPAP analysis on the panel of 180 eukaryote species (Figure 6—figure supplement 1; Cohen et al., 2013). The analysis was complicated due to the lineage-specific diverse DNMT classes (e.g. Dim2, DNMT5, DNMT6 and other plant specific DNMT variants) and divergent variants of zf-4CXXC_R1. Considering this caveat, we conducted CoPAP analysis of five DNMTs (DNMT1, Dim-2, DNMT3 [including DRM], DNMT5, DNMT6), UHRF1, HELLS, CDCA7, proteins with class II zf-4CXXC_R1, and proteins with zf-4CXXC_R1 and JmjC. As fungi-specific CDCA7F contains class II zf-4CXXC_R1, and all the other class II zf-4CXXC_R1 containing proteins were identified in species that also possess CDCA7, we conducted CoPAP against two separate lists; in the first list (Figure 6—figure supplement 1A) CDCA7F was included in the CDCA7 category (i.e. considered as a prototypical CDCA7 ortholog in fungi), whereas in the second list (Figure 6—figure supplement 1B) CDCA7F was included in the class II zf-4CXXC_R1 category. As positive and negative controls for the CoPAP analysis, we also included subunits of the PRC2 complex (EZH1/2, EED and Suz12), and other SNF2 family proteins SMARCA2/SMARCA4, INO80 and RAD54L, which have no direct role related to DNA methylation, respectively. As expected for proteins that act in concert within the same biological pathway, both CoPAP results showed significant coevolution between DNMT1 and UHRF1, as well as between the PRC2 subunits EZH1 and EED. Suz12 did not show a significant linkage to other PRC2 subunits by this analysis, most likely due to a failure in identifying diverged Suz12 orthologs, such as those in Neurospora and Paramecium (Jamieson et al., 2013; Miró-Pina et al., 2022). The evolutionary linkage between DNMT5 and DNMT6 were also observed (Figure 6—figure supplement 1). In the result of CoPAP analysis against the first list, where CDCA7F was included in the CDCA7 category, CDCA7 exhibited a coevolutionary linkage to DNMT1 (Figure 6—figure supplement 1A). In the second CoPAP, where CDCA7F was included in the class II zf-4CXXC_R1 category, CDCA7 exhibited linkage to UHRF1 and HELLS (Figure 6—figure supplement 1B). Notably, none of these proteins show an evolutionary association with the PRC2 proteins or other SNF2 family proteins, while the coevolutionary linkage between CDCA7 and the DNMT1-UHRF1 cluster was reproducible.

We next conducted the CoPAP analysis against a panel of 50 Ecdysozoa species, where DNA methylation system is dynamically lost in multiple lineages (Figure 6A; Bewick et al., 2017b; Engelhardt et al., 2022), yet the annotation of DNMTs, UHRF1, CDCA7 and HELLS is unambiguous. As a negative control, we included INO80, which is dynamically lost in several Ecdysozoa lineages, such as C. elegans (Figure 5, Figure 6A and Figure 5—figure supplement 1). As expected, CoPAP analysis showed a highly significant coevolutionary interaction between DNMT1 and UHRF1 (Figure 6B). In addition, HELLS interacts with DNMT1, UHRF1 and CDCA7. In contrast, no linkage from INO80 or DNMT3 was determined. Altogether, these CoPAP analyses reproducibly found coevolutionary linkages between DNMT1-UHRF1 and CDCA7-HELLS, although the exact interaction topology is sensitive to the selection of species and protein classification methods.

CoPAP analysis detected the coevolutionary linkage between CDCA7 and DNMT1-UHRF1, rather than DNMT3. We were therefore intrigued by the apparent absence of CDCA7 and DNMT3 in two insect species, the red flour beetle Tribolium castaneum and the braconid wasp Microplitis demolitor, whose genomic DNA does not have any detectable 5mC despite the presence of DNMT1 (Bewick et al., 2017b; Schulz et al., 2018; Zemach et al., 2010; Figure 5). To further validate the co-loss of CDCA7 and DNMT1/DNMT3 in insects, we focused on the Hymenoptera clade (including M. demolitor), for which genome synteny has been reported (Li et al., 2021). Indeed, a striking synteny is observed in the genome region surrounding CDCA7 among the parasitic wood wasp (Orussus abietinus) and Aculeata species (bees Bombus terrestris and Habropoda laborlosa, and the eusocial wasp Polistes canadensis), which diverged ~250 MYA (Li et al., 2021; Peters et al., 2017; Figure 7 and Figure 7—source data 1). In these species, CDCA7 is located between Methyltransferase-like protein 25 homolog (MET25, E) and Ornithine aminotransferase homolog (OAT, F). In fact, the gene cluster containing LTO1 homolog (D), MET25 (E), OAT (F), and Zinc finger protein 808 (ZN808, G) is highly conserved in all the analyzed Hymenoptera species, but not outside of Hymenoptera (e.g. Drosophila).

However, CDCA7 is absent from this gene cluster in parasitoid wasps, including Ichneumonoidea wasps (braconid wasps [M. demolitor, Cotesia glomerata, Aphidius gifuensis, Fopius arisanus] and ichneumon wasp [Venturia canescent]) and chalcid wasps (Copidosoma floridanum, Nasonia vitripennis) (Figure 7 and Figure 7—source data 2). Among Ichneumonoidea wasps, CDCA7 is undetectable in the genome of braconid wasps, while CDCA7 is located on a different chromosome in the ichneumon wasp Venturia canescens. In chalcid wasps, CDCA7 is present at a genome segment between Artemis (I) and Chromatin accessibility complex protein 1 (CHRC1, J). Supported by the chromosome-level genome assembly for Cotesia glomerata and Aphidius gifuensis (Feng et al., 2020; Pinto et al., 2021), the synteny analysis strongly indicates that CDCA7 is lost from braconid wasps. Intriguingly, braconid wasps co-lost CDCA7 either with DNMT3 (in M. demolitor, Cotesia glomerata, Cotesia typhae and Chelonus insularis) or with DNMT1 and UHRF1 (in Fopius arisanus, Diachasma alloeum and Aphidius gifuensis) (Figure 7 and Figure 7—source data 2). Notably, in addition to M. demolitor (which lacks CDCA7 and DNMT3), it has been reported that little or no 5mC can be detected in Aphidius ervi (which lacks CDCA7, DNMT1 and UHRF1, Figure 7—source data 2; Bewick et al., 2017b). In contrast, all Hymenoptera species that are known to have detectable 5mC possess CDCA7 along with DNMT1, UHRF1, DNMT3, and HELLS, except for Polistes canadensis, which has all but DNMT3 (Figure 7 and Figure 7—source data 2). This co-loss of CDCA7 and DNA methylation (together with either DNMT1-UHRF1or DNMT3) in braconid wasps suggests that evolutionary preservation of CDCA7 is more sensitive to DNA methylation status per se than to the presence or absence of a particular DNMT subtype. Consistent with this idea, among the panel of 180 eukaryote species, none of the 17 species where absence of genomic 5mC has been experimentally shown encodes CDCA7 (Figure 1—source data 1 tab8) (Aliaga et al., 2019; Antequera et al., 1984; Bewick et al., 2019a; Bewick et al., 2017b; Cock et al., 2010; Engelhardt et al., 2022; Geyer et al., 2011; Mondo et al., 2017; Noordhoek et al., 2018; Schulz et al., 2018; Singh et al., 2018; Zemach et al., 2010).

A list of 180 eukaryote species was manually generated to encompass broad eukaryote evolutionary clades (Figure 1—source data 1). Species were included in this list based on two criteria: (i) the identification UBA1 and PCNA homologs, two highly conserved and essential proteins for cell proliferation; and (ii) the identification of more than 6 distinct SNF2 family sequences. Homologs of CDCA7, HELLS, UBA1, and PCNA were identified by BLAST search against the Genbank eukaryote protein database available at National Center for Biotechnology Information using the human protein sequence as a query (NCBI). Homologs of human UHRF1, ZBTB24, SMARCA2/SMARCA4, INO80, RAD54L, EZH2, EED, or Suz12 were also identified based on the RBH criterion. To get a sense of genome assembly level of each genome sequence, we divided ‘Total Sequence Length’ by ‘Contig N50’ (‘length such that sequence contigs of this length or longer include half the bases of the assembly’; https://www.ncbi.nlm.nih.gov/assembly/help/). In the species whose genome assembly level is labeled as ‘complete’, this value is close to the total number of chromosomes or linkage groups. As such, as a rule of thumb, we arbitrarily defined the genome assembly ‘preliminary’, if this value is larger than 100. In Figure 5, these species with preliminary-level genome assembly were noted as boxes with dotted outlines.

CDCA7 homolog identification and annotation BLAST search was conducted using human CDCA7 (NP_114148) as the search query against NCBI protein database. The obtained list of CDCA7 homologs was classified based on the conservation of eleven cysteine and three ICF-associated residues in the zf-4CXXC_R1 domain, as described in Results. This classification was further validated based on their clustering in a phylogenetic tree built from the CLUSTALW alignment of the zf-4CXXC_R1 domain identified by NCBI conserved domains search (Higgins and Sharp, 1988; Lu et al., 2020; Thompson et al., 1994; Figure 2—figure supplement 1), using MacVector (MacVector, Inc). Jalview was used to color-code amino acids based on conservation and amino acid types (Waterhouse et al., 2009). The cluster of class I zf-4CXXC_R1 domain-containing proteins (where all three ICF-associated residues are conserved) was segregated from other variants of zf-4CXXC_R1-containing proteins except for the moss Physcomitrium XP_024393821 and green algae Bathycoccus XP_007512509 and Chlamydomonas GAX81623, which has class II zf-4CXXC-R1 (where the ICF-associated glycine residue is substituted). To further assess this validation, another phylogenetic tree was built by the optimal maximum likelihood-based model selected by IQ-TREE 2.2.2.6 (Minh et al., 2020), using the zf-4CXXC_R1 domain alignment generated and selected by Muscle v5 (Figure 2—source data 1; Edgar, 2022). A consensus tree was then constructed from 1000 bootstrap trees using UFBoot2 (Figure 2—source data 2; Hoang et al., 2018). The phylogenetic tree built by this second method deviated all class II and other zf-4CXXC_R1 variants from CDCA7 orthologs with class I zf-4CXXC_R1 (Figure 2—source data 2).

HELLS homologs were first identified according to the RBH criterion. Briefly, a BLAST search was conducted using human HELLS as the query sequence, after which protein sequences of obtained top hits (or secondary hits, if necessary) in each search were used as a query sequence to conduct reciprocal BLAST search against the Homo sapiens protein database. If the top hit in the reciprocal search returned the original input sequence (i.e. human HELLS), it was temporarily annotated as an orthologous protein. If HELLS showed up as a next best hit, it is temporarily listed as a “potential HELLS ortholog”. To further validate the identified HELLS orthologs, full length amino acid sequences of these proteins were aligned using CLUSTALW in MacVector with homologs of the Snf2 family proteins SMARCA2/SMARCA4, CHD1/CHD3/CHD7, ISWI, RAD54L, ATRX, HLTF, TTF2, SHPRH, INO80, SMARCAD1, SWR1, MOT1, ERCC6, and SMARCAL1, which were also identified and temporarily annotated with a similar reciprocal BLAST search methods. The phylogenetic tree generated by this full-length alignment separated the clade containing HELLS, DDM1, INO80, SMARCA2/SMARCA4, CHD1/CDH3/CHD7, and ISWI from other Snf2 family proteins (Figure 5—figure supplement 2). This alignment was used to define the conserved SNF2 domain and variable linker regions in the putative HELLS orthologs and other homologous SNF2-family proteins (INO80, SMARCA2/SMARCA4, CHD1/CDH3/CHD7, and ISWI). The variable linker regions were then removed from selected proteins for each kingdom/phylum to conduct the secondary CLUSTALW alignment in MacVector, from which a phylogenetic tree was generated (Figure 5—figure supplement 3). A distinct cluster of HELLS and other SNF2 family proteins can be identified from the phylogenetic tree, hence confirming that the annotation of HELLS orthologs based on the reciprocal BLAST search method is reasonable. Exceptions are Leucosporidium creatinivorum ORY88017 and ORY88018, which did not cluster within the HELLS clade of the phylogenetic tree. However, we decided to annotate L. creatinivorum ORY88017 and ORY88018 as HELLS orthologs since among other L. creatinivorum SNF2-family proteins in this species these two proteins are most similar to human HELLS while clear orthologs of other L. creatinivorum SNF2 family proteins, CHD1 (ORY55731), ISWI (ORY89162), SMARCA2/4 (ORY76015), SRCAP/SWR1 (ORY90750), and INO80 (ORY91599), can be identified. To further validate the phylogenetic tree generated in Figure 5—figure supplement 3, another phylogenetic tree was built by the optimal maximum likelihood-based model selected by IQ-TREE 2.2.2.6 (Minh et al., 2020), using the alignment generated and selected by Muscle v5 (Figure 5—source data 1; Edgar, 2022). A consensus tree was then constructed from 1000 bootstrap trees using UFBoot2 (Figure 5—source data 2; Hoang et al., 2018). The topology of the phylogenetic tree build by this second method was consistent with the original tree shown in Figure 5—figure supplement 3.

Proteins with a DNA methyltransferase domain were identified with BLAST searches using human DNMT1 and DNMT3A. Additional BLAST searches were conducted using human DNMT2, C. neoformans DNMT5, N. crassa Dim-2 and DNMT4, Thalassiosira pseudonana DNMT6 (XP_002287631; Huff and Zilberman, 2014), and Kwoniella mangroviensis DNMTX (XP_019001759; Catania et al., 2020). DNMT domains were extracted from selected proteins that represent each kingdom/phylum and DNMT type (including ambiguous classes) based on a NCBI conserved domains search and were aligned with CLUSTALW to build a phylogenetic tree in MacVector. To further assess this validation, another phylogenetic tree was built by the optimal maximum likelihood-based model selected by IQ-TREE 2.0.3 (Minh et al., 2020), using the alignment generated and selected by Muscle v5 (; Edgar, 2022). A consensus tree was then constructed from 1000 bootstrap trees using UFBoot2 (; Hoang et al., 2018). The topology of the phylogenetic tree built by this second method was largely consistent with the original tree except for placement of orphan DNMTs. In both trees, Vitrella brassicaformis CEM15752, Fragilariopsis cylindrus OEU08290, OEU15938, Thalassiosira pseudonana XP_002295472, and Tribonema minus KAG5185060 form a clade, which, only in the IQ-TREE generated tree, splits from the canonical DNMT4 clade. As proteins in this clade have been previously annotated as DNMT4 (Huff and Zilberman, 2014), we followed this classification and present this IQ-TREE-based phylogenetic tree of DNMTs in Figure 5—figure supplement 4, which was illustrated using the ETE3 toolkit.

The published method was used (Cohen et al., 2013). The curated list of orthologous proteins listed in Figure 1—source data 1 was first used to generate a presence-absence FASTA file. Next, a phylogenetic species tree was generated from all orthologous protein sequences listed in Figure 1—source data 1 using the ETE3 toolkit. For this, protein sequences were retrieved using the rentrez Bioconductor package and exported to a FASTA file alongside a COG file containing gene to orthologous group mappings. ETE3 was used with the parameters -w clustalo_default-trimal01-none-none and -m cog_all-alg_concat_default-fasttree_default and the resulting tree exported in Newark format. CoPAP was run using default parameters and results visualized using Cytoscape. Code and files required for CoPAP input generation as well CoPAP parameters and output results can be found in our Github repository (https://github.com/RockefellerUniversity/Copap_Analysis copy archived at Carroll, 2023). A method to calculate p-value for CoPAP was described previously (Cohen et al., 2012). Briefly, for each pair of tested genes, Pearson’s correlation coefficient was computed. Parametric bootstrapping was used to compute a p-value by comparing it with a simulated correlation coefficient calculated based on a null distribution of independently evolving pairs with a comparable exchangeability (a value reporting the likelihood of gene gain and loss events across the tree).

As negative and positive controls for the CoPAP analysis, we identified several well-conserved protein orthologs across the panel of 180 eukaryotic species, including Snf2-like proteins SMARCA2/SMARCA4, INO80, and RAD54L (Flaus et al., 2006), as well as subunits of the polycomb repressive complex 2 (PRC2), which plays an evolutionary conserved role in gene repression via deposition of the H3K27me3 mark. PRC2 is conserved in species where DNMTs are absent (including in D. melanogaster and C. elegans) but is frequently lost particularly in several lineages of SAR and Fungi (Sharaf et al., 2022). Among the four core subunits of PRC2, we focused on the catalytic subunit EZH1/2, EED, and SUZ12, since the fourth subunit RbAp46/48 has a PRC2-independent role (Margueron and Reinberg, 2011). We are aware that the reciprocal BLAST search missed previously reported highly divergent functional orthologs of SUZ12 in Neurospora (Jamieson et al., 2013), and EED and Suz12 in Paramecium (Miró-Pina et al., 2022). However, we did not attempt to use these divergent homologs of EED and SUZ12 as baits to expand our search in order to consistently apply our homology-based definition of orthologs.

The mapping of gene loci is based on the information available on the Genome Data Viewer (https://www.ncbi.nlm.nih.gov/genome/gdv). Genome positions of listed genes are summarized in Figure 7—source data 1.

Artworks of species images were obtained from https://www.phylopic.org/, of which images of Daphnia, Platyhelminthes, Tribolium and Volvox were generated by Mathilde Cordellier, Christopher Laumer/T. Michael Keesey, Gregor Bucher/Max Farnworth and Matt-Crook, respectively.