Phylogenomics of the Epigenetic Toolkit Reveals Punctate Retention of Genes across Eukaryotes

Abstract

Epigenetic processes in eukaryotes play important roles through regulation of gene expression, chromatin structure, and genome rearrangements. The roles of chromatin modification (e.g., DNA methylation and histone modification) and non-protein-coding RNAs have been well studied in animals and plants. With the exception of a few model organisms (e.g., Saccharomyces and Plasmodium), much less is known about epigenetic toolkits across the remainder of the eukaryotic tree of life. Even with limited data, previous work suggested the existence of an ancient epigenetic toolkit in the last eukaryotic common ancestor. We use PhyloToL, our taxon-rich phylogenomic pipeline, to detect homologs of epigenetic genes and evaluate their macroevolutionary patterns among eukaryotes. In addition to data from GenBank, we increase taxon sampling from understudied clades of SAR (Stramenopila, Alveolata, and Rhizaria) and Amoebozoa by adding new single-cell transcriptomes from ciliates, foraminifera, and testate amoebae. We focus on 118 gene families, 94 involved in chromatin modification and 24 involved in non-protein-coding RNA processes based on the epigenetics literature. Our results indicate 1) the presence of a large number of epigenetic gene families in the last eukaryotic common ancestor; 2) differential conservation among major eukaryotic clades, with a notable paucity of genes within Excavata; and 3) punctate distribution of epigenetic gene families between species consistent with rapid evolution leading to gene loss. Together these data demonstrate the power of taxon-rich phylogenomic studies for illuminating evolutionary patterns at scales of >1 billion years of evolution and suggest that macroevolutionary phenomena, such as genome conflict, have shaped the evolution of the eukaryotic epigenetic toolkit.

Significance

Eukaryotic organisms evolved complex epigenetic processes to orchestrate gene expression and genome dynamics. By applying a taxon-rich phylogenomic approach, including adding transcriptome data from several lineages of understudied microeukaryotes, we identify homologs of the epigenetic gene toolkit in diverse lineages across the eukaryotic tree of life. We show that gene families involved in chromatin modification and the processing of non-protein-coding RNAs originated in the last eukaryotic common ancestor. However, the distribution of epigenetic genes across eukaryotes now reflects a punctate pattern, with differential conservation of genes across lineages and functional categories. The analyses here suggest that macroevolutionary phenomena, such as genome conflict and/or adaptations to diverse living styles, shaped the epigenetic toolkit and phenotypic diversity within eukaryotes.

Introduction

Throughout the last decades, it has become increasingly clear that epigenetic modifications play major roles in regulating the expression of the genotype in a wide range of eukaryotic taxa (e.g., Wolffe and Matzke 1999; Bird 2007; Goldberg et al. 2007). The existence of epigenetic mechanisms expands upon the idea of a linear relationship between genotypes and phenotypes and can challenge Mendelian inheritance (e.g., Katz 2006). Epigenetics can modify gene expression, including completely silencing genes and mobile genetic elements, and also be responsible for altering genome structures (e.g., Bernstein and Allis 2005; Heard and Martienssen 2014). The effects of these epigenetic processes range from cell differentiation to genomic imprinting and, in cases where they malfunction, disease (e.g., Jiang et al. 2004; Gluckman et al. 2009; Handel et al. 2010). Epigenetics also plays a role in shaping genome architectures through DNA rearrangement/elimination and polyploidization in diverse lineages of eukaryotes (e.g., Liu and Wendel 2003; Maurer-Alcalá and Katz 2015). In addition to impacting individual cells or organisms, epigenetics likely also acts across generations, influencing the evolution of populations and species (e.g., Smith and Ritchie 2013; Smith et al. 2016) and may contribute to rapid adaptive responses (e.g., Rey et al. 2016). Overall, its effects can be summed up as creating a variety of phenotypes from the same genotype.

The term “epigenetics” was first introduced by Waddington (1942) to refer broadly to the expression of the phenotype during development. Ever since, its definition has been subject to intense discussion (e.g., Haig 2004; Bird 2007; Goldberg et al. 2007; Stotz and Griffiths 2016) and generally includes both well-known processes (i.e., histone modifications and DNA methylation) as well as a variety of poorly known genetic phenomena (i.e., paramutation, transgenerational effects). Today’s textbook definition is that epigenetics refers to heritable phenotypic changes that arise without change in the underlying DNA sequence (e.g., Tollefsbol 2017). However, here we use Denise Barlow’s broader definition of epigenetics as “all the weird and wonderful things that cannot [yet] be explained by genetics” (McVittie 2006).

The molecular processes of epigenetics can be roughly assigned to two classes: chromatin modifiers (e.g., DNA methylation and histone modifications; e.g., Razin and Riggs 1980; Ng and Bird 1999) and non-protein-coding RNAs (npc-RNAs, RNA interference: microRNAs, Piwi interacting RNAs, and small interfering RNAs; e.g., Sharp 2001; Shabalina and Koonin 2008; Peng and Lin 2013; Bond and Baulcombe 2014). Of the two classes, chromatin modifiers are currently understood more deeply. Through mechanisms such as the addition or removal of methyl or acetyl groups to nucleotides or histones, chromatin modifiers can silence or activate genes by producing physical changes to chromatin accessibility (e.g., Fuks 2005). A large number of enzymes are known to be involved in these processes, including DNA and histone methyltransferases, histone acetyltransferases, and deacetylases as well as the members of the Polycomb-group proteins (e.g., Fuks 2005; Zemach and Zilberman 2010; Maumus et al. 2011; Di Croce and Helin 2013; Aravind et al. 2014; Rastogi et al. 2015; Vogt 2017).

In contrast, npc-RNAs act through sequence-specific gene silencing and their targets include viral genes, transposons, and eukaryotic genes in both germline and somatic cells (Shabalina and Koonin 2008; Peng and Lin 2013). They have been argued to have originated in genome screening and defense (Obbard et al. 2009). Based on previous analyses, the genes involved in generating npc-RNAs appear widespread across eukaryotes and the most prominent members include ARGONAUTE, PIWI, the RNases III DROSHA and DICER as well as RNA-dependent RNA polymerases and RNA helicases (Sharp 2001; Peng and Lin 2013; Li and Patel 2016).

Though epigenetic processes are best understood in plants and animals, many components of the epigenetic toolkit are also found in other lineages across the eukaryotic tree of life (e.g., Maurer-Alcalá and Katz 2015) and an extensive epigenetic machinery was likely present already in the last eukaryotic common ancestor (LECA) as key elements can be traced back to prokaryotic systems of secondary metabolism and genome conflict (Iyer et al. 2008; Aravind et al. 2014). Authors such as Fedoroff (2012), Lisch (2009), and Klobutcher and Herrick (1997) have also hypothesized that epigenetic processes originally arose as a means to restrict the spreading of transposable elements within genomes and only later were their roles expanded to other dynamic genome processes. Despite the importance of epigenetics for the development and evolution of eukaryotic lineages, knowledge on these processes in nonmodel lineages remains scarce. Especially for many clades of microbial eukaryotes, including Rhizaria, Amoebozoa, and diverse ciliates, details on epigenetic gene families remain unknown, even though these groups are known for complex genome dynamics that likely involve epigenetics (e.g., Parfrey et al. 2008; Croken et al. 2012).

The combination of advances in single-cell ‘omics (e.g., Kolisko et al. 2014; Saliba et al. 2014), large-scale sequencing (e.g., Massana et al. 2015), and phylogenomics (e.g., Ceron-Romero et al. 2019) now allow for easy access and exploration of data from uncultivable microeukaryotes. Among the clades with the greatest paucity of data are Amoebozoa, Rhizaria, and Ciliophora (with the exception of models such as Tetrahymena and Paramecium; reviewed in Maurer-Alcalá et al. 2018), which are now included in this study. Though genomes are well-sampled for pathogens (e.g., Acanthamoeba and Entamoeba) and model lineages (e.g., Physarum and Dictyostelium) within Amoebozoa, clades such as the shell-building Arcellinida lack ‘omics data. The situation is similar within the Rhizaria, where the lack of human parasites within this major eukaryotic clade likely contributes to the dearth of data (Grattepanche et al. 2018).

To investigate macroevolutionary patterns of the epigenetic toolkit across the eukaryotic tree of life, we analyze epigenetic gene families using PhyloToL (Ceron-Romero et al. 2019). PhyloToL was specifically designed for the investigation of the heterogenous evolutionary patterns observed among diverse eukaryotic clades, spanning 1.8 billion years of evolution. We combine PhyloToL with a taxon-rich data set to assess homology and generate both multiple sequence alignments (MSAs) and gene trees. PhyloToL (Ceron-Romero et al. 2019) also allows for the removal of contaminants that are frequent in ‘omics data sets. For our analyses, we included a maximum of 278 transcriptomes and 182 genomes representing 460 species from all major eukaryotic clades. We also include a limited set of 89 bacterial genomes and 25 archaeal genomes. In addition to the genomes and transcriptomes obtained from publicly available databases, such as GenBank and OrthoMCL, we added single-cell transcriptomes from diverse clades of microbial eukaryotes for understudied taxa from Amoebozoa and SAR (Stramenopila, Alveolata and Rhizaria) in order to improve taxonomic coverage. We analyzed a total of 118 epigenetic gene families that are involved in either chromatin modification or npc-RNAs. Our intention is to characterize the distribution of the epigenetic toolkit across the eukaryotic tree of life, especially targeting microbial eukaryotic clades that remain understudied.

Results

Distribution of the Epigenetic Toolkit across Major Eukaryotic Clades

Based on the literature, we analyzed 179 genes in the eukaryotic epigenetic toolkit as those that play major roles in either chromatin modification or npc-RNA processes. These 179 genes fall into 118 gene families as defined by the database OrthoMCL (Li et al. 2003; table 1 and supplementary table S1, Supplementary Material online), which is the starting point for gene family delineation in PhyloToL (Ceron-Romero et al. 2019). This focal set of genes is both incomplete and biased as epigenetics has so far been best studied in plants (e.g., Finnegan et al. 1998; Rapp and Wendel 2005), animals (e.g., Fazzari and Greally 2004; Glastad et al. 2011), and only a few other eukaryotic lineages (e.g., Grewal 2000; Aramayo and Selker 2013).

Table 1.

Summary of Epigenetic Gene Families and Their Functional Categories

Category	Subcategory	Pathway/Process	# Gene Families
Chromatin modification	DNA methylation	DNA methyltransferases	12
		Methyl-DNA binding	3
	Histone modification	Lysine acetyltransferases	10
		Lysine deacetylases	10
		Lysine methyltransferases	20
		Lysine demethylases	12
		Arginine methyltransferases	9
		Polycomb-group proteins	12
		Others	6
npc-RNAs	NA	Non-protein-coding RNAs	24

Note.—Shown are the two main categories of epigenetic processes, chromatin modification and npc-RNAs, which are split into subcategories and associated pathways/processes. The number of gene families for each representative pathway is indicated. In total, we analyzed 118 gene families. Details on individual genes and their functions are shown in supplementary table S1, Supplementary Material online.

To evaluate the distribution of the epigenetic toolkit across eukaryotes, we analyzed the presence/absence of the 118 gene families in up to 574 species sampled from all major eukaryotic clades plus a limited number of bacteria and archaea (table 2 and supplementary table S2, Supplementary Material online). The data set includes 69 newly sequenced transcriptomes of six species of Arcellinida (Amoebozoa), three species of Ciliophora (Alveolata), and 14 species of Rhizaria, which substantially increases taxonomic coverage for these understudied clades (sequences available at GenBank SRA BioProject PRJNA637648). To assess the effect of taxon sampling on macroevolutionary patterns, we compared the results obtained for four different data sets: 1) ALL: all 574 taxa that passed a quality cutoff; 2) INFORMED: taxonomically informed, “even” subsample across clades with 25 taxa each; 3) RANDOM: random subsample of 25 taxa per major eukaryotic clade; and 4) GENOME: only taxa for which we had whole genome data (232 total), which allowed us to rule out missing data in transcriptomes as a major driver of the observed patterns. All 118 gene trees were generated for the taxon sets ALL, INFORMED, and GENOME. The RANDOM set, on the other hand, failed to produce a gene tree for the methyl-DNA binding protein MECP2 (OG5_140477), as this gene family had too few taxa for tree inference.

Table 2.

Eukaryotic and Prokaryotic Lineages Included in the Analysis

Major Clade	Nested Clades	ALL	INFORMED	RANDOM
Stramenopila (St)	Diatoms, Bikosea, Blastocystida, Chrysophytes, Eustigmatophytes, Labyrinthulomycetes, Oomycetes, Brown Algae, Pinguiophyceae, Raphidophytes, Synchromophytes, Synurophytes	77 (13/64)	25 (6/19)	25 (4/21)
Alveolata (Al)	Apicomplexa, Chromerida, Ciliates, Dinoflagellates, Perkinsozoa	87 (28/59)	25 (11/14)	25 (13/12)
Rhizaria (Rh)	Cercozoa, Foraminifera, Sticholonchida	31 (2/29)	25 (1/24)	25 (2/23)
Archaeplastida (Pl)	Green Algae and plants, Glaucophytes, Red Algae	59 (20/39)	25 (12/13)	25 (12/13)
Orphan lineages (EE)	Apusozoa, Breviatea, Centroheliozoa, Cryptomonads, Haptophytes, Katablepharids	42 (3/39)	25 (2/23)	25 (3/22)
Excavata (Ex)	Euglenozoa, Fornicata, Heterolobosea, Jakobida, Malawimonadidae, Oxymonadida, Parabasalia	31 (20/11)	25 (14/11)	25 (15/10)
Amoebozoa (Am)	Archamoeba, Discosea, Mycetozoa, Stereomyxa, Tubulinea	36 (8/28)	25 (5/20)	25 (5/20)
Opisthokonta (Op)	Choanoflagellates, Fungi, Ichthyosporea, Metazoa	97 (88/9)	25 (22/3)	25 (24/1)
Archaea (Za)	Archaeoglobi, Asgard group, Bathyarchaeota, Crenarchaeota, Halobacteria, Korarchaeota, Methanobacteria, Methanococci, Methanomicrobia, Methanopyri, Nanoarchaeota, Thaumarchaeota, Thermococci, Thermoplasmata	25 (25/0)	25 (25/0)	25 (25/0)
Bacteria (Ba)	Actinobacteria, Proteobacteria, Aquificae, Bacilli, Bacteroidia, Chlamydiales, Chlorobi, Chloroflexia, Clostridia, Cyanobacteria, Cytophagia, Deinococcus-Thermus, Dictyoglomi, Fusobacteriia, Nitrospira, Planctomycetes, Spirochaetia, Tenericutes, Thermotogae, Verrucomicrobia	89 (89/0)	25 (25/0)	25 (25/0)

Note.—The names and abbreviations used throughout the manuscript for the major eukaryotic clades, bacteria, and archaea. Shown are exemplary nested clades for each major clade and the number of species included in the different taxon subselections. Numbers in parentheses indicate genomes and transcriptomes, respectively. For details on chosen species, their taxonomy, and accession numbers, see supplementary table S2, Supplementary Material online.

The sizes of the gene trees are highly variable (fig. 1), indicating complex patterns of distribution of the toolkit across eukaryotic lineages. Among the three larger data sets (ALL, INFORMED, and RANDOM), we observe a consistent pattern of presence/absence of gene families across major clades. For example, all three data sets yielded similar numbers of gene families that seem to have existed already before the last eukaryotic common ancestor (pre-LECA; 20–28 gene families, defined as present in all but one major eukaryotic clade, bacteria, and/or archaea) or that were present in LECA (34–39 gene families, defined as present in all but one major eukaryotic clade, supplementary table S3, Supplementary Material online). This indicates that taxon choice did not have a substantial impact on our interpretation. The only exception is the GENOME data set that generally shows lower values (supplementary table S3, Supplementary Material online), which corresponds to the low number of whole genomes available for some major clades (e.g., only two whole genomes were publicly available for Rhizaria and eight for Amoebozoa, supplementary table S2, Supplementary Material online). Given the overall similarity among data sets, we provide the results for all four data sets in the supplementary files (supplementary table S4, Supplementary Material online) and focus the rest of our study on results from the INFORMED subsample where the even distribution of species allows better comparisons across major clades.

Fig. 1.

Exemplary epigenetic gene families showing variability in size. (A) The phylogenetically restricted Polycomb-related gene SUZ12, (B) the punctate npc-RNA-related gene family DICER, and (C) the complex Lysine deacetylase HDAC1 gene family. Single gene trees do not generate well-resolved phylogenetic relationships across the ∼1.8 billion years of eukaryotic evolution, and these trees are included to show the variability in conservation of the epigenetic genes across eukaryotes. Taxon selection is the INFORMED data set and taxa are colored by major clades: Stramenopila (St) = blue, Alveolata (Al) = yellow, Rhizaria (Rh) = gray, Archaeplastida (Pl) = olive, orphans (EE) = dark blue, Excavata (Ex) = red, Amoebozoa (Am) = light blue, Opisthokonta (Op) = dark gray, Archaea (Za) = blue gray, and Bacteria (Ba) = dark red. The trees were manually rooted on bacteria, fungi, or metazoa depending on which lineages were present.

Overall, patterns of conservation of epigenetic gene families are complex (fig. 2). As expected, given the relatively large number of studies, Opisthokonta (Op) and Archaeplastida (Pl) contain the highest number of gene families with 109 and 97 out of 118, respectively. We identified 86 gene families in Amoebozoa (Am), 85 in Rhizaria (Rh), 83 in Stramenopila (St), 76 in Alveolata (Al), and 84 among the nonmonophyletic orphan lineages (i.e., EE, “everything else”). A striking difference is that the 25 species within Excavata (Ex) contain only 53 gene families, the smallest number among all eukaryotic clades (fig. 2). Bacteria (Ba) and archaea (Za) only contain a few of the gene families analyzed, which is as expected given the eukaryotic focus of this study.

Fig. 2.

Differential conservation of the epigenetic toolkit across major eukaryotic clades with relative paucity of gene families in Excavata. Each column represents the presence/absence pattern per major clade (MC, abbreviations of the major clades as in table 2), and the rows represent the 118 epigenetic gene families sorted by degree of conservation across the tree of life (for the exact order of gene families see supplementary table S5, Supplementary Material online). The numbers on top indicate the number of gene families (GF) present in each major clade. Shown are the results for the INFORMED taxon selection (250 species) and the presence (blue) and absence (white) of the epigenetic gene families in the major eukaryotic clades, bacteria, and archaea. There is a striking difference in degree of conservation among epigenetic gene families: about half of them seem to have been present already before LECA (pre-LECA) or in LECA, whereas the other half are more restricted. Another strong signal is the absence of the majority of gene families in the Excavata (highlighted by a red star).

We identified three distinct patterns from the presence/absence analysis of gene families in major eukaryotic clades (fig. 2): 1) Pre-LECA gene families that are present in six of the seven eukaryotic clades (Op, Pl, Al, St, Rh, Am, and/or Ex) as well as in bacteria and/or archaea, 2) LECA gene families that are present in six of the seven major eukaryotic clades but absent in the sampled bacteria/archaea; and 3) the remaining gene families that are found in one to five of the eukaryotic clades. In total, 21 of the 118 gene families meet the pre-LECA criteria for the INFORMED taxon selection (fig. 2). Of these, 17 gene families are part of the 94 gene families involved in chromatin modification pathways and the remaining four are among the 24 gene families involved in npc-RNA processes. A total of 39 of the 118 gene families can be assigned to LECA, of which 31 have functions related to chromatin modification and eight to npc-RNAs (fig. 2 and supplementary tables S1, S4, and S5, Supplementary Material online). The remaining 58 gene families have variable distributions among the major eukaryotic clades (49 of 58, >1 MC label fig. 2 and supplementary tables S4 and S5, Supplementary Material online) or are specific to a certain major clade (nine of 58, one MC label fig. 2). Of these, 46 gene families are involved in chromatin modification and 12 in npc-RNA processes.

We further assessed the relationship of gene function and patterns of conservation (fig. 3). Of the gene families belonging to chromatin modification pathways, the degree of conservation appears to depend on function: lysine deacetylases and acetyltransferases show a high degree of conservation, as the majority of gene families in these categories are designated to pre-LECA/LECA (90% and 80%, respectively). Lysine demethylases, arginine methyltransferases, and a group of other histone-modification proteins all have around 50% of their respective gene families likely present in pre-LECA/LECA. In contrast, lysine methyltransferases only have 45% pre-LECA/LECA gene families and the Polycomb-related gene families show the least degree of conservation among the chromatin modifiers with only 25% present in LECA. Instead, 42% of the Polycomb-related gene families are in fewer than six but more than one major eukaryotic clade and 33% are even restricted to one clade (fig. 3). For the npc-RNA-related gene families, 50% are conserved as they fall in the pre-LECA and LECA data sets, whereas DNA methylation gene families are a less conserved functional class with 26% of the gene families in pre-LECA/LECA, 53% in between one to five major eukaryotic clades, and 20% in only one major clade.

Fig. 3.

Conservation of epigenetic gene families shows differences across functional categories. Epigenetic gene families classified by functional category, as shown in table 1, show variable numbers of conserved genes. “DNA methylation” and “non-protein-coding RNAs” represent higher level categories, comprising a variety of genes with different functions. To the right, the number of gene families (GF) is listed for each category. For each category, the percentage of pre-LECA (black) and LECA (gray) gene families is indicated, as well as the gene families that are present in fewer major clades (MC, white) or are even restricted to one major clade (crosshatched). Data are based on the results of the INFORMED taxon selection.

Distribution of the Epigenetic Toolkit at the Species Level

To assess species-specific patterns of gene family presence/absence, we repeated the analysis on the 250 species in the INFORMED data set and mapped the data onto a phylogeny generated from a concatenation of 391 housekeeping gene families (nonepigenetic genes that are widespread across eukaryotes and likely were present already in or before LECA, fig. 4). First, we evaluated the quality of our data by assessing presence/absence of 118 housekeeping gene families (i.e., the same number as in our epigenetic set) that we chose randomly from among the 391 gene families used for the phylogenomic analysis (see Materials and Methods). The housekeeping gene families are present in almost all species sampled here, demonstrating the overall good quality of data in our INFORMED data set, which includes 121 transcriptomes among the 250 species (fig. 4). Though four of the 200 eukaryotic species contained none or only one of the 118 epigenetic gene families, some of the other species with only transcriptome data are among the samples with the greatest numbers of gene families (supplementary table S2, Supplementary Material online). The INFORMED data set contains the newly generated transcriptomes of five species of Arcellinida (Amoebozoa) and 10 species of Rhizaria, which is a subset of our newly added transcriptome data as described above. Orphan lineages like the Apusozoa and Malawimonas have both few epigenetic and few housekeeping gene families, suggesting data quality plays a role here.

Fig. 4.

Comparison of the presence/absence between epigenetic and housekeeping gene families in eukaryotic species shows punctate retention of the epigenetic toolkit. The phylogenetic tree contains the eukaryotic species of the INFORMED data set (196 species, four species were removed due to low data quality) and is a concatenated tree based on the 391 housekeeping gene families (see Materials and Methods for details). The color coding of the major clades (MC) follows the colors in figure 1, genome taxa are in bold. Two Malawimonas species were originally classified as Excavata but fell among the orphan lineages in the tree. The panel on the left shows the presence (gray) or absence (white) for 118 of the 391 housekeeping gene families (columns) in each of the eukaryotic taxa (rows). The panel on the right shows the presence (blue) or absence (white) of the 118 epigenetic gene families. The orphan lineages are disregarded in counting the number of major clades.

At the species level, the same overall pattern emerges as for the level of major clades, with the greatest numbers of gene families found within species of Opisthokonta and/or Archaeplastida and the fewest among Excavata (fig. 4 and supplementary table S2, Supplementary Material online). Among Opisthokonta, animal species show a high degree of similarity in the composition of their epigenetic toolkits (fig. 4). The same is true for the species of fungi, yet compared with animals their toolkit contains fewer gene families. Among Archaeplastida, the toolkit of green algae is homogeneous across species and can be differentiated from the toolkit of the red algae and glaucophytes (fig. 4 and supplementary tables S4 and S5, Supplementary Material online). The three SAR clades as well as the Amoebozoa appear similar in the composition of their toolkits and there are no obvious lineage specific patterns given our taxon sampling. As with the clade-based analyses, the size of the Excavata toolkit is overall smaller than in other eukaryotes, with Euglenozoa and the other Excavata showing a distinctive subset of gene families (fig. 4 and supplementary tables S4 and S5, Supplementary Material online).

Punctate Distribution of Many Epigenetic Gene Families

We observe a punctate distribution pattern among eukaryotes for many epigenetic gene families. Here, punctate refers to gene families that are widespread across eukaryotic lineages (i.e., present in three or more major clades) and yet are found in only a small number of species per major clade. Among the pre-LECA/LECA gene families (i.e., those present in at least six and often all major clades), there are cases where gene families are retained in only 24 out of the 250 species (i.e., the gene family OG5_135026, RNA helicase). This punctate pattern can be seen in some individual gene trees (fig. 1B) as well as in the presence/absence data at the species level (fig. 4). The punctate pattern is apparent when the presence/absence data for the epigenetic gene families is compared with the housekeeping gene families, which show a more homogeneous distribution across the same eukaryotic species (fig. 4).

Two possibilities to explain the punctate distribution of gene families include 1) functional constraints are similar across lineages but gene loss is higher among epigenetic genes than housekeeping genes and 2) punctate genes are evolving rapidly such that homologs now fail to meet the criteria for homology assessment necessary to generate MSAs and gene trees. To distinguish between these possibilities, we calculated the average branch length for each of the gene trees for the epigenetic gene families and compared them with our housekeeping gene set. In the first scenario (i.e., change in pattern of gene loss), branch lengths from nodes to tips may not be significantly different, whereas in the second case (i.e., rapid evolution of epigenetic genes), branch lengths are expected to be longer. For this, we classified the epigenetic trees in three categories, big (>100 sequences), medium (26–100 sequences), and small (≤25 sequences). Although the big trees and the housekeeping gene trees have similar branch lengths, the medium and small trees have increasingly longer average branch lengths (fig. 5).

Fig. 5.

Differences between the average branch lengths of the housekeeping and epigenetic gene trees. We calculated average branch lengths for every tree of the housekeeping and epigenetic gene families based on the INFORMED taxon selection. The epigenetic trees are clustered into three groups (big, middle, and small) based on the number of branches they contain. Statistical analysis (Shappiro Wilk, big: P > 0.5102, middle: P > 0.8219, small: P > 0.496, and housekeeping: P < 0.002036) and analyses of QQ plots (supplementary fig. S1, Supplementary Material online) of the four data sets suggest that the data likely are normally distributed. The means of all four data sets are significantly different from each other (Welch’s t-test, supplementary table S6, Supplementary Material online), even though the difference between the housekeeping and big trees is smaller than between all other combinations. The box-whisker plots include medians for each data set.

To compare mean branch lengths across these trees, we used a parametric test, Welch’s t-test. The data points of the three epigenetic categories showed a normal distribution according to a Shappiro Wilk test (big: P > 0.5 and n = 31, middle: P > 0.8 and n = 60, and small: P > 0.4 and n = 27) and QQ plots (supplementary fig. S1, Supplementary Material online). In contrast, the housekeeping gene families do not fit expectations for normal distribution (P < 0.005 and n = 391; supplementary fig. S1, Supplementary Material online), which is likely due to the large number of data points that lead to a high sensitivity to deviations from normality. Under Welch’s t-test, the means of each category (i.e., housekeeping, big, medium, and small; fig. 5) are statistically significantly different from every other category (supplementary table S6, Supplementary Material online).

To address the possibility that we failed to include rapidly evolving members of smaller gene families, we used BLAST to identify additional sequences for three of the npc-RNA gene families (DICER, PIWI, and ARGONAUTE) but found that few added genes survived Guidance analysis, the MSA tool we use to assess homology (see methods). For example, an alignment of sequences from 11 gene families that we identified as potential DICER homologs did not survive Guidance (supplementary table S7, Supplementary Material online). When we forced the genes to align using MAFFT and checked the result by eye, we saw little evidence for homology, consistent with either rapid evolution or the independent origin of these genes. We obtained a similar result for PIWI genes: combining eight potential homologs, only three survived Guidance and the resulting tree indicated deep divergence between gene families consistent with ancient paralogy rather than lost nested homologs (supplementary fig. S2 and table S7, Supplementary Material online). The forced alignment of the potential ARGONAUTE homologs retained six out of eight gene families that fall into two clades in the tree (supplementary fig. S2 and table S7, Supplementary Material online). However, each of the taxa present is also represented by the “main” ARGONAUTE gene family and so inclusion of the divergent genes would not have changed our assessment of presence/absence of this gene family. In sum, manually combining additional gene families does not add any further information to the macroevolutionary patterns of the epigenetic genes.

Paralogs

We find a trend toward higher numbers of sequences per species per gene family (i.e., paralogs) in the housekeeping genes than in the epigenetic genes, though the absolute number of paralogs is confounded here by observation of only highly expressed genes in the transcriptome data. We repeatedly subsampled 60 gene families (100 repetitions) from the housekeeping data set and compared them with the 60 pre-LECA/LECA epigenetic gene families. The overall trend of more sequences in the housekeeping gene families was significant for 93 out of the 100 iterations of the analysis (Sign test, H_a: epigenetic < housekeeping, P < 0.05, supplementary table S8, Supplementary Material online). The major clades responsible for this trend are Stramenopila, Rhizaria, Archaeplastida, Excavata, and Amoebozoa (Mann–Whitney, H_a: epigenetic < housekeeping, P < 0.05 for more than 65/100 iterations). Although Alveolata show no evident trend with high data dispersion, Opisthokonta show the opposite trend with more sequences in the epigenetic genes than in the housekeeping genes (Mann–Whitney, H_a: epigenetic > housekeeping, P < 0.05 for more all 100 iterations; supplementary table S8, Supplementary Material online).

Discussion

Our taxon-rich analyses yield three main insights: 1) a rich epigenetic toolkit existed in LECA, containing genes for both chromatin modification and npc-RNA processes; 2) the toolkit is differentially conserved among major eukaryotic clades with a notable paucity of genes within Excavata; and 3) in contrast to the housekeeping gene families, many epigenetic gene families show a punctate distribution in that they are widespread across eukaryotes but retained in only a few species.

Presence of the Epigenetic Toolkit in LECA

Because epigenetic processes play fundamental roles in many eukaryotes, several authors have proposed the existence of a widespread, ancient epigenetic toolkit (e.g., Cerutti and Casas-Mollano 2006; Parfrey et al. 2008; Shabalina and Koonin 2008; Aravind et al. 2014; Maurer-Alcalá and Katz 2015). Previous analyses have largely focused on a narrow sampling of lineages (e.g., animals and plants; Finnegan et al. 1998; Fazzari and Greally 2004; Rapp and Wendel 2005; Glastad et al. 2011), leaving the majority of eukaryotic diversity understudied. However, data from a limited sample of microeukaryotes and phylogenomic approaches suggested that epigenetics is not restricted to multicellular organisms, but present in microbial lineages as well and may indeed have been present already in LECA (e.g., Aravind et al. 2014). Epigenetic processes play a role in the complex genome dynamics of microbial lineages, such as changes in ploidy level (up to thousand copies of the genome) in some lineages of Rhizaria and Alveolata (Parfrey et al. 2008) and/or separation of the genome into germline and soma within one cell (e.g., Ciliophora; Prescott1994; Katz 2001). Other lineages have a parasitic lifestyle that involves frequent changes to their chromatin structures and gene expression profiles (e.g., Croken et al. 2012), which have been shown to be influenced by epigenetic processes as well (e.g., Liu et al. 2007; Cortes et al. 2012; Croken et al. 2012; Chalker et al. 2013). Yet, for many microbial eukaryotic lineages, it remained unclear if these processes and the underlying epigenetic genes correspond to gene families present in animals and/or plants, or if they evolved independently.

Our taxon-rich phylogenomic approach allows us to provide a more detailed depiction of the conservation of epigenetic processes across eukaryotes and supports the hypothesis of a toolkit in LECA as all major eukaryotic clades contain gene families of all functional categories as defined in this study (fig. 6 and supplementary table S5, Supplementary Material online). Coupling PhyloToL (Ceron-Romero et al. 2019), which allows rapid homology assessment and the generation of MSAs and gene trees, with single-cell transcriptome data of uncultivable microbial eukaryotes in Rhizaria, Amoebozoa, and ciliates, allowed us to provide additional detail to the evolution of eukaryotic epigenetic gene families.

Fig. 6.

Distribution of the epigenetic gene families by functional categories in the major eukaryotic clades. Shown is the percentage of gene families per functional category that each major eukaryotic clade contains, based on the INFORMED taxon selection. Color coding of the major eukaryotic clades follows the colors in figure 1. Noteworthy is the limited number of gene families related to methylation processes and Polycomb-group proteins in the Excavata.

Our analyses indicate that the retention of epigenetic genes varies by functional categories, with gene families related to histone modifications, especially acetylation and deacetylation, being over-represented in pre-LECA/LECA, whereas the Polycomb-group proteins and DNA methylation genes are retained in fewer lineages (e.g., fig. 3 and supplementary table S5, Supplementary Material online). Gene families involved in processes like lysine acetylation/deacetylation are used in post-translational modifications in bacteria and archaea (e.g., Christensen et al. 2019) and have been co-opted to serve in chromatin modification in eukaryotes. The Polycomb-group proteins, on the other hand, appear to be a eukaryotic invention as members such as the protein SUZ, chromobox proteins (CBX), enhancer of zeste (EZH), and the Polycomb-group ring finger proteins (PCGF) are found only among eukaryotes (supplementary tables S1 and S4, Supplementary Material online). Early work on Polycomb-group proteins demonstrated their roles in cell differentiation and development and so they were originally assumed to be restricted to multicellular lineages (animals and plants; e.g., Kohler and Villar 2008). However, core components of the Polycomb Repressive Complex 2 (PRC2) also exist in unicellular eukaryotes, such as the green alga Chlamydomonas and the diatom Thalassiosira (Shaver et al. 2010). Our analysis extends on this as we find PRC2 components (e.g., Nurf55, ESC, and EZH; supplementary tables S1 and S4, Supplementary Material online) in a wide range of unicellular lineages (e.g., especially among Stramenopila and Rhizaria). The most parsimonious explanation, therefore, is that a basic set of Polycomb-group proteins was already present in LECA and has been lost or has evolved rapidly and beyond recognition where they appear absent. Intriguingly, some have argued that Polycomb-group proteins originated as defense against mobile genetic elements and only later they took on the more specific roles in multicellular lineages (Shaver et al. 2010). For DNA methylation systems, it has been suggested that they have been transferred from bacteria to eukaryotes several times independently and that some components may have been lost in individual lineages (Ponger and Li 2005; Iyer et al. 2008; Zemach and Zilberman 2010). Our study is consistent with this idea, because—despite much wider taxon sampling—we also observe the DNA methylation gene families to be less widespread across eukaryotes (figs. 3 and 6).

Smaller Toolkit Size in the Excavata

Phylogenomic analyses demonstrate a notable paucity of genes among Excavata, despite the fact that complete genomes exist for many of these species (i.e., we can rule out failure to detect signal from incomplete transcriptome data; supplementary table S2, Supplementary Materialonline). Excavata lack the majority of Polycomb-group gene families, which are also sparse in other major eukaryotic clades (fig. 6). More surprising, most Excavata also lack gene families with conserved functions related to methylation (e.g., lysine methyltransferases and demethylases, DNA methylation; fig. 6 and supplementary table S4, Supplementary Material online). The smaller toolkit size in Excavata could be due to several factors discussed in detail below: 1) Excavata exhibit unusual genome structures, suggesting that their chromatin may be regulated differently; 2) the parasitic and thus often anoxic/microaerophilic lifestyle of many sampled Excavata may be incompatible with epigenetic processes involving methylation some of which require oxygen; or 3) if Excavata are at the root of the eukaryotic tree of life (He et al. 2014), some functions of the epigenetic toolkit may have expanded after their divergence.

Unusual genome structures within Excavata may underlie the smaller number of epigenetic gene families. Among the Excavata, members of the Kinetoplastida exhibit an unusual genome organization, with protein-coding genes arranged in large polycistronic transcription units that are processed post-transcriptionally through trans-splicing (e.g., Belli 2000; El-Sayed et al. 2005; Clayton 2019). In addition, histone sequences in Excavata, and especially of the Trypanosomatids, are highly divergent from those of other eukaryotes (Sullivan et al. 2006). These structural peculiarities suggest that processes underlying chromatin modification in Excavata may also be divergent from other eukaryotes. Even though histone modifications governed by epigenetic processes exist within Excavata, the specific patterns of these marks, that is, the “histone code”, differ from conserved eukaryotic patterns (Sullivan et al. 2006; Croken et al. 2012). Elias and Faria (2009) do report roles of npc-RNA processes in gene regulation in some Trypanosomatids. Although we find support for the existence of some npc-RNA gene families in Excavata, some such as ARGONAUTE are represented by a divergent “ARGONAUTE-like” gene family (OG5_149426) instead of the more widespread ARGONAUTE gene family (OG5_127240; supplementary table S4, Supplementary Material online). Together, these data suggest that unusual genome structures may have led to divergent epigenetic strategies within Excavata.

A second possible explanation for the smaller set of epigenetic gene families within Excavata is that gene families underlying methylation processes (e.g., the DNA methylase DNMT and lysine demethylases KDM; supplementary table S4, Supplementary Material online) may have been reduced in parasites that can live in low-oxygen environments. For example, DNA methylation seems to be absent in the Excavata genus Giardia (Lagunas-Rangel and Bermudez-Cruz 2019), whereas histone acetylation and npc-RNAs are important for its encystation and expression of surface proteins for host immune evasion (Prucca et al. 2008; Carranza et al. 2016; Ortega-Pierres et al. 2018). Similar patterns are found in other anaerobic parasites, such as Trypanosoma gondii (Excavata), and even two Apicomplexans (Plasmodium and Cryptosporidium, Alveolata; Croken et al. 2012). In human tumor cells and germinating rice, low or anoxic conditions lead to aberrant DNA methylation patterns, suggesting that these epigenetic processes require oxygen as substrate (Bhandari et al. 2017; Narsai et al. 2017; Camuzi et al. 2019). Together these data suggest that the anaerobic life style of many Excavata may have an influence on the composition of the epigenetic toolkit similar to how a microaerophilic lifestyle is thought to be related to altered genome structures and gene expression in a range of human parasites (Vanacova et al. 2003).

Though the position of the root of the eukaryotic tree of life is still debated, one hypothesis is that it lies within Excavata, and specifically between Discoba (i.e., Euglenozoa, Heterolobosea, Tsukubea, and Jakobea) and the rest of eukaryotes (He et al. 2014). If this hypothesis were true, the smaller epigenetic toolkit in Excavata could be an indicator that the epigenetic functions expanded in the remainder of the eukaryotes after the divergence of the Excavata. However, the position of the root within Excavata may be the result of phylogenetic artifacts such as long-branch attraction, and alternative roots such as between Unikonta and Bikonta (Stechmann and Cavalier-Smith 2003; Derelle et al. 2015) and between Opisthokonta and the other eukaryotes (Stechmann and Cavalier-Smith 2002; Katz et al. 2012) are still valid hypotheses (reviewed by Burki et al. [2020]).

The Epigenetic Toolkit Shows a Pattern of Punctate Distribution across Eukaryotes

We observe a punctate distribution pattern of many epigenetic gene families (fig. 4). Most strikingly, gene families that we conservatively define as being present in pre-LECA/LECA (i.e., those in more than five of seven major eukaryotic clades) are not present in many of the sampled lineages, which stands in stark contrast with the high conservation of housekeeping genes in the same data set (fig. 4). We see a similar pattern among the more “recent” gene families as some are present in two or more major clades but only in a few of the species sampled (fig. 4). Similarly, we see fewer paralogs among epigenetic gene families as compared with housekeeping genes (supplementary table S8, Supplementary Material online). Two possible explanations for this punctate pattern include 1) genes may have been lost in some lineages and/or 2) epigenetic genes evolve rapidly in some lineages and are no longer detected as homologs in our phylogenomic approach.

Distinguishing between these two explanations is challenging due to both data availability and the definitions used for gene family membership. Though assessing cases of gene loss especially is hampered by the lack of whole genome data from many eukaryotic lineages, our analyses of the limited set of whole genome data show the same punctate distribution of genes (supplementary table S4, Supplementary Material online). Consistent with the hypothesis of rapid evolution of epigenetic gene family members, we did observe longer branch lengths (i.e., from tips to first node) in smaller (i.e., more punctate) gene families as opposed to larger gene families (fig. 5), but phylogenetic artifacts and biases likely contribute to this pattern. More fundamentally, gene loss can occur in a continuum, from the accumulation of numerous mutations that impact homology assessment to the complete elimination of genes from within genomes. Hence, some “lost” members of epigenetic gene families may have changed sufficiently to be excluded as members of their ancestral gene families.

Macroevolutionary Phenomena May Underlie the Distribution of Epigenetic Gene Families among Eukaryotes

We hypothesize that the punctate distribution pattern of genes in the epigenetic toolkit is the result of genome conflict, either as a defense against mobile genetic elements and/or as a regulator of germline/soma differentiation. Some epigenetic processes are believed to have originated as mechanisms for defense against viruses and other mobile genetic elements (e.g., Fedoroff 2012), and the relatively rapid rates of some epigenetic genes (e.g., those involved in processing npc-RNAs) may be the result of an arms race between host and intruder genomes (e.g., Obbard et al. 2009). Epigenetic genes also play a role in germline-soma distinctions. For example, ciliates rely on complex epigenetic processes to drive germline/soma distinction and DNA elimination throughout their lifecycle (e.g., Liu et al. 2007; Maurer-Alcalá and Katz 2015; Pilling et al. 2017).

Another macroevolutionary pattern that may explain the punctate distribution of genes in the epigenetic toolkit is their potential role in differential adaptation and reproductive isolation. A growing number of studies find differences in epigenetic marks (e.g., methylomes) of populations that are exposed to different environmental conditions (e.g., Marsh and Pasqualone 2014; Johnson and Kelly 2020; Wogan et al. 2020) and in some cases these differences seem to be correlated with reproductive isolation (e.g., Smith et al. 2016; Blevins et al. 2017). Further, by regulating gene expression, epigenetic modifications can produce phenotypic plasticity, which selection may act upon (Rey et al. 2016) and which in turn can lead to reproductive isolation and ultimately to speciation. Finally, the possibility of inter-generational or trans-generational inheritance of epigenetic marks or npc-RNAs (as reviewed by Boskovic and Rando [2018] and Perez and Lehner [2019]) may enhance the possibility of adaptation. Epigenetics, therefore, may allow for adaptation of species to changing environmental conditions (Rey et al. 2016).

Materials and Methods

All approaches taken for data acquisition and data analysis are summarized here, and we refer the reader to the Supplementary Material online for details on methods.

Data Acquisition

We identified genes involved in epigenetic processes by delving into the literature describing the molecular basis of epigenetics (Fuks 2005; Anantharaman et al. 2007; Peters and Meister 2007; Hollick 2008; Shaver et al. 2010; Maumus et al. 2011; Fedoroff 2012; Bond and Baulcombe 2014; Rastogi et al. 2015; Li and Patel 2016; Vogt 2017) and searching databases such as Pfam (https://pfam.xfam.org/, last accessed November 2, 2018) and KEGG (www.genome.jp/kegg/, last accessed November 2, 2018; supplementary table S1, Supplementary Material online). We used the resulting list of genes to identify the corresponding OG (orthologous groups) numbers in the OrthoMCL database (Li et al. 2003), which correspond to the gene families in the phylogenomic pipeline PhyloToL (Ceron-Romero et al. 2019). In total, we identified 179 genes that group into 118 distinct gene families (table 1) and we ran PhyloToL to search for homologs of these epigenetic gene families in all major eukaryotic clades, plus a limited number of bacteria and archaea.

In addition to the sequence data included in PhyloToL (retrieved from either GenBank, RefSeq, or OrthoMCL; supplementary table S2, Supplementary Material online), we added 69 transcriptomes from understudied clades within SAR and Amoebozoa that we generated to increase taxonomic sampling. Because these microbial eukaryotes are not currently cultivable, we used a single-cell whole transcriptome amplification approach and assessed the quality of the resulting data based on the presence of at least 100 of 391 housekeeping gene families (supplementary table S1, Supplementary Material online). This approach resulted in the final number of 574 taxa, 296 of which are represented by whole genomes and 278 by transcriptomes (table 2 and supplementary table S2, Supplementary Material online). We subsampled these data in three different ways to test the robustness of our analyses to taxon selection (supplementary table S2, Supplementary Material online). We then used PhyloToL to produce MSAs and gene trees for each of the epigenetic gene families for all four taxon selections. We also repeated this analysis for the 391 housekeeping genes.

Data Analysis

As described in detail in the Supplementary Material online, we used custom Python scripts (github.com/Katzlab/Epigenetics) to count the number of species per major clade that appeared in each gene family tree as well as their number of paralogs (supplementary table S4, Supplementary Material online). We repeated this analysis for all four taxon sets and used the resulting data to estimate which gene families were present in LECA or even before (supplementary tables S3 and S4, Supplementary Material online). We assessed the evolutionary history of gene families in relationship with their grouping into certain functional categories (fig. 3). We also calculated the branch length of each gene tree (fig. 5 and supplementary tables S1 and S6, Supplementary Material online) and compared the number of paralogs in the epigenetic gene families versus the housekeeping gene families (supplementary table S8, Supplementary Material online), using methods described in the Supplementary Material online.

Data Availability

All sequenced transcriptomes are available on GenBank under the SRA BioProject PRJNA637648. The scripts used in the analyses of the data are available under github.com/Katzlab/Epigenetics.

Supplementary Material

Supplementary data are available at Genome Biology and Evolution online.