Chromatin loops are an ancestral hallmark of the animal regulatory genome

Abstract

In bilaterian animals, gene regulation is shaped by a combination of linear and spatial regulatory information. Regulatory elements along the genome are integrated into gene regulatory landscapes through chromatin compartmentalization^1,2, insulation of neighbouring genomic regions^3,4 and chromatin looping that brings together distal cis-regulatory sequences⁵. However, the evolution of these regulatory features is unknown because the three-dimensional genome architecture of most animal lineages remains unexplored^6,7. To trace the evolutionary origins of animal genome regulation, here we characterized the physical organization of the genome in non-bilaterian animals (sponges, ctenophores, placozoans and cnidarians)^8,9 and their closest unicellular relatives (ichthyosporeans, filastereans and choanoflagellates)¹⁰ by combining high-resolution chromosome conformation capture^11,12 with epigenomic marks and gene expression data. Our comparative analysis showed that chromatin looping is a conserved feature of genome architecture in ctenophores, placozoans and cnidarians. These sequence-determined distal contacts involve both promoter–enhancer and promoter–promoter interactions. By contrast, chromatin loops are absent in the unicellular relatives of animals. Our findings indicate that spatial genome regulation emerged early in animal evolution. This evolutionary innovation introduced regulatory complexity, ultimately facilitating the diversification of animal developmental programmes and cell type repertoires.

Main

A fundamental characteristic of animal multicellularity is the existence of specialized cell types. These cell types result from differential access to genomic information in each cell. Thus, evolutionary changes in genome regulation are proposed to be a major innovation linked to the emergence of complex multicellularity with stable cell differentiation^6,10. This idea is supported by comparative genomic analyses showing that gene innovation at the origin of animals was less extensive than previously thought^10,13, thus suggesting that an important animal innovation was the ability to coregulate existing genes in different combinations.

In bilaterian animals, genome spatial compartmentalization mediates the organization of gene neighbourhoods that can be independently regulated^3,4,6 and that are specific to different cell types¹⁴. Another mechanism contributing to elaborate gene regulation in bilaterians is the combinatorial interaction of distal cis-regulatory elements and gene promoters by means of chromatin loops that bring distant regions into spatial proximity through genome folding, contrasting with the predominant regulation by proximal promoter elements in unicellular eukaryotes¹⁰. Comparative analyses of histone posttranslational modifications have shown that candidate distal enhancer elements, as defined by chromatin features, predate the origin of bilaterian animals^8,15,16, whereas such enhancers are absent in the closest unicellular relatives of animals¹⁷. However, it is still unclear whether distal regulation in early-branching metazoans is mediated by physical interactions with gene promoters or linked to the existence of insulated gene regulatory landscapes.

To investigate the origins of animal gene regulation, here we comparatively studied chromatin architecture at subkilobase resolution in non-bilaterian animal lineages and their closest unicellular relatives of animals (Fig. 1). This includes the two phyla proposed as the sister group to all other animals^9,18: ctenophores, which are mostly pelagic, marine predators that swim using ciliated comb cells and have complex nerve nets^19,20; and sponges, which are sessile, benthic organisms that filter-feed using collared choanocyte cells^8,21. We also examined placozoans, which are millimetre-sized, flat animals that feed on microbial mats by gliding using ciliary movement and mucus secretion, controlled by peptidergic secretory cells²², and cnidarians, the sister group to bilaterians that includes jellyfishes, corals and anemones²³. Finally, we studied three unicellular relatives of animals, known as unicellular holozoans: ichthyosporeans, which are osmotrophic unicellular eukaryotes that reproduce through multinucleated coenocytes²⁴; filastereans, which are heterotrophic protists with complex life cycles, including aggregative multicellular stages^17,25 and choanoflagellates, which are heterotrophic flagellates that show both single-cell and colonial forms and are the closest living relatives to animals²⁶. The comparative analysis of chromatin maps across these lineages allows us to reconstruct the evolutionary history of genome regulation in animals.

Fig. 1: Chromatin architecture in early animal evolution.

a, Comparison of genomic features across metazoans and unicellular holozoans. For H. sapiens, we used previously published mCG methylation percentage data from H1 ESCs cells. Of note, although distal cis-regulatory elements (dCRE) were identified in Amphimedon queenslandica¹⁵, their presence in E. muelleri had not been reported previously. mCG, CG methylation; TEs, transposable elements. b, Top left, phylogenetic tree showing the taxon sampling in this study, along with the number of profiled species per clade. Top right and below, Micro-C interaction maps of specific genomic regions (S. arctica, chr. 2: 3400000–3700000, bin 1 kb; C. owczarzaki, chr. 01: 3660000–3800000, bin 400 bp; S. rosetta, chr. 21: 800000–1100000, bin 800 bp; M. leidyi, chr. 8: 15500000–15700000, bin 400 bp; E. muelleri, Emue22: 2200000–2400000, bin 800 bp; T. adhaerens, TadhH1_4: 3880000–4180000, bin 800 bp; N. vectensis, NC_064040.1: 11650000–12000000, bin 1 kb; D. melanogaster, chr. 3L: 20480000–20820000, bin 800 bp; and H. sapiens, chr. 12: 69000000–71000000, bin 5 kb), showing examples of insulation boundaries or chromatin loops. All interaction maps were balanced using ICE normalization.

Large-scale genome organization

We used Micro-C^11,12 to map genome-wide chromatin contacts at single-nucleosome resolution in representatives of non-bilaterian animal lineages (Fig. 1, Extended Data Fig. 1, Supplementary Table 1 and Supplementary Text 1): the ctenophore Mnemiopsis leidyi^19,20, the sponge Ephydatia muelleri²¹, the placozoan Trichoplax adhaerens²² and the cnidarian Nematostella vectensis²³. As outgroup species, we studied chromatin architecture in three unicellular holozoans: the ichthyosporean Sphaeroforma arctica²⁴, unicellular filasterean amoeba Capsaspora owczarzaki^17,25 and the choanoflagellate Salpingoeca rosetta²⁶. We also compared our chromatin maps with existing datasets from two bilaterians: Drosophila melanogaster²⁷ and Homo sapiens¹². To analyse our chromatin contact experiments, we first resequenced de novo and assembled to chromosome-scale the genomes of M. leidyi, E. muelleri and C. owczarzaki using a combination of Nanopore (Oxford Nanopore Technology) long-read sequencing and Micro-C data (Extended Data Fig. 2). For S. arctica, S. rosetta and T. adhaerens, we rescaffolded existing genomes^22,24,26 to chromosome level using Micro-C data. In addition, to interpret the observed contact features, we generated genome-wide maps of chromatin accessibility (assay for transposase-accessible chromatin with high-throughput sequencing or ATAC-seq), chromatin modifications (chromatin immunoprecipitation with sequencing (ChIP–seq) for H3K4me3, H3K4me2, H3K4me1) and gene expression (RNA sequencing or RNA-seq), or used published datasets when available (Supplementary Table 2). We integrated three-dimensional (3D) chromatin data with linear chromatin marks to systematically compare genome architectural features at different resolutions^3,4,7 (compartmentalization, insulation and chromatin looping) and across phylogenetically distant species with diverse genome sizes, gene densities and transposable element content (Fig. 1a).

We first analysed global chromosomal compartmentalization, which results from the spatial segregation of distinct chromatin states genome-wide (active, A; inactive, B) and is influenced by histone marks, DNA methylation and gene transcription, among other phenomena^28,29. As such, compartmentalization is often considered an intrinsic biophysical property of the chromatin driven by phase separation^30,31. To compare the degree of self-affinity and segregation between major chromatin compartments, we defined A/B compartment limits in each species. We then calculated the intensity of compartmentalization in genomic bins with compartment A and B interaction frequency in the top 20th percentile (Fig. 2a). Compartmentalization strength in each species was quantified as the ratio of homotypic (AA, BB) to heterotypic (AB) interactions (Fig. 2b). The relative resolutions were obtained by partitioning genomes into equal number of bins across species (Extended Data Fig. 3a,b), but the differences between species remained consistent regardless of the number of bins used (Fig. 2b). Furthermore, we assigned an intermediate compartment (I) to regions with weak spatial separation (Extended Data Fig. 3c,d).

Fig. 2: Chromatin compartments and insulation boundaries across species.

a, Saddle plots showing contact interactions between A and B compartments in each species, organized by eigenvector ranking. To obtain the distance-normalized matrix, the ratio of observed-over expected interactions is calculated, followed by eigenvector decomposition. The eigenvectors are oriented and sorted from the lowest (B compartment) to the highest (A compartment) values. The bins of the interaction matrix then reordered according to the rank of the eigenvector. The observed (O) and expected (E) values are averaged to create a saddle plot. The top 20% of the interaction values were used to calculate the compartment strength values shown on the saddle plots. Cowc, C. owczarzaki; Dmel, D. melanogaster; Emue, E. muelleri; Hsap, H. sapiens; Mlei, M. leidyi; Nvec, N. vectensis; Sarc, S. arctica; Sros, S. rosetta; Tadh, T. adhaerens. b, Compartment strength quantification at different relative resolutions. The barplot below shows the contribution of homotypical chromatin interactions within active (AA) and inactive (BB) chromatin states. c, Aggregate plots showing contact enrichment within a rescaled region between two insulation boundaries. The boundaries are identified using the sliding diamond window to detect the changes in contact frequencies in each genomic bin. To plot pile-ups, regions between insulation boundaries are rescaled and their normalized observed and expected contact frequencies are averaged. d, Insulation score distributions illustrating the degree of isolation between linear genomic neighbourhoods. Number of annotated strong boundaries is indicated in blue, with a vertical line representing the median value of each distribution. e, Classification of insulation boundaries using hierarchical assignment of structural and genomic features. f, Size distribution of annotated chromatin loops in each species. The boxplots show the median (centre line), 25th and 75th percentiles (box limits) and the whiskers show the range of variability, excluding outliers, which are shown as individual points. g, Annotation of chromatin loop anchors with promoter (P) and enhancer (E) signatures based on normalized H3K4me3 and H3K4me2 or H3K4me1 ChIP–seq coverage. Chromatin loop anchors with undefined (U) epigenetic signature are shown in grey.

Our analysis revealed that, with exception of M. leidyi, animal genomes were globally segregated into transcriptionally active, gene-dense compartments and transcriptionally inactive, transposable element-rich compartments, similar to what is observed in bilaterian animals (Extended Data Fig. 3d). In these species, we detected a strong separation of A and B compartments in saddle plots (Fig. 2a) and the compartment strength values above 1.8 (Fig. 2b). Moreover, these compartments encompass relatively large contiguous regions across the genome (Extended Data Fig. 3b). By contrast, unicellular holozoans and M. leidyi did not show strong separation of large A and B compartments (Fig. 2a,b), similar to what is observed in yeast³² and other protists³³. The absence of large-scale chromatin compartments in M. leidyi is unusual among animals, although it has been previously reported in certain species³⁴. This lack of compartmentalization may be due to the absence of constitutively silenced regions across different cell types. Overall, our results indicate that A/B chromosomal compartmentalization is a phylogenetically conserved feature across animal genomes.

Insulation and micro-scale contacts

We next characterized small-scale chromosomal features across species by defining spatial insulation boundaries between neighbouring loci. The boundary elements that partition genome into domains can arise from active transcription, silenced repetitive regions or binding of sequence-specific architectural proteins at insulator or tethering elements^5,27,35,36. Thus, our first goal was to identify the occurrence of insulation boundaries in each species (Fig. 2c,d and Extended Data Fig. 4), and then classify these points into different regulatory or structural features (domain boundaries, gene bodies, regulatory loops and so on) (Fig. 2e). To this end, we calculated insulation scores for each species, representing the difference in contact frequencies between each genomic bin and its neighbouring bins. We used different resolutions and sliding window sizes (Extended Data Fig. 4a,b) and, for each species, we selected the resolution and two window sizes that yielded the maximal insulation signal, indicating the strongest partitioning of the genome into isolated structural and functional domains. The median distance between successive identified boundary elements varied between 6.4 kilobases (kb) in S. rosetta and 190 kb in H. sapiens, yet the median number of genes per interval was consistently similar across species, with two to four genes (Extended Data Fig. 4c).

The presence of self-interacting domains, contiguous regions of the genome with enriched interactions, was assessed by examining the average pile-up plots between insulation boundaries (Fig. 2c). We observed weak contact enrichment between pairs of insulated boundaries in unicellular holozoans and E. muelleri. In M. leidyi, boundary elements were tethered through strong focal contacts and without intradomain interactions, contrary to what would be expected within topologically associating domains (TADs)³. By contrast, D. melanogaster showed intradomain enrichment without focal contacts, in agreement with previously reported domains³⁷. T. adhaerens and N. vectensis showed a certain degree of self-affinity within insulated neighbourhoods, as well as focal point enrichment (Fig. 2c). The degree of insulation of genomic regions could be quantified from the distribution of genome-wide insulation scores (Fig. 2d and Extended Data Fig. 4c). M. leidyi, T. adhaerens, N. vectensis, H. sapiens and S. arctica genomes contained strong boundary elements in comparison with E. muelleri and, especially, the weakly insulated genomes of C. owczarzaki and S. rosetta (Fig. 2d and Extended Data Fig. 4c).

After identifying insulation points, we investigated the genomic features associated with these boundaries (Fig. 2e and Extended Data Fig. 4d). We first assigned insulation boundaries to annotated chromatin loops, followed by the transcription start sites (TSSs) of genes not involved in chromatin looping and then accessible chromatin regions that may represent other regulatory elements. Remaining boundaries were assigned to A/B compartment limits. This analysis revealed that most insulation boundaries in unicellular holozoans and E. muelleri were associated with active TSSs (Fig. 2e), suggesting that active transcription is the main factor defining insulation in these species³⁷. By contrast, many insulation boundaries could be assigned to chromatin loop anchors in M. leidyi (77%; compared to 78% in H. sapiens human embryonic stem cells) and in T. adhaerens (38%), whereas in N. vectensis, we identified 166 chromatin loops that represented only 1.6% of insulation boundaries. The number of chromatin loops in M. leidyi (4,261) and T. adhaerens (3,065) was much higher than those found in N. vectensis (166) and D. melanogaster (313)²⁷, despite their similar genome sizes and gene densities (Fig. 1a). Loop sizes were comparable in these four species (median 21–28 kb), but much smaller than in H. sapiens (median 140 kb) with a genome 15–30 times larger (Fig. 2f). To further characterize these distal contacts, we examined genome-wide H3K4me3, H3K4me2 and H3K4me1 to classify many of the identified loop anchor sites as promoter-like elements (Fig. 2g). In M. leidyi and N. vectensis, chromatin loops predominantly occurred between promoters and enhancers (77 and 69%, respectively), similar to H. sapiens (63%). By contrast, 79% of loops in T. adhaerens connected promoters to other promoters, similarly to what is observed in D. melanogaster (49%)³⁸. Our results show that enhancer–promoter and promoter–promoter long-range chromatin loops are shared between bilaterians and early-branching animal lineages, and possibly date back to the origin of animal multicellularity.

Protists, sponges and cnidarians

In unicellular holozoans, we did not observe any spatial contact patterns indicative of chromatin loops. However, manual inspection revealed a few regions enriched in distal contacts. For example, in S. arctica, we could identify 296 self-interacting insulated domains that also contact each other (Extended Data Fig. 5a,b). These regions were depleted of active histone marks and were enriched in transposable elements, probably representing repressed chromatin domains that cosegregate (Extended Data Fig. 5c). In S. rosetta, there were 183 distally interacting regions that contained lowly expressed genes (Extended Data Fig. 5d,e) and were enriched in H3K4me1 and H3K27me3 or lacked profiled marks (Extended Data Fig. 5f). These may also represent repressed regions³⁹, albeit they do not form well-defined domains like in S. arctica. In C. owczarzaki, we observed a plaid pattern indicative of chromatin microcompartments (Extended Data Fig. 5g), reflecting the spatial cosegregation of active promoters of highly transcribed genes with a strong H3K4me3 signal (Extended Data Fig. 5h,i). These microcompartment contacts form a regional small-scale checkerboard pattern with alternating loci of high and low interactions. Furthermore, we also detected high-frequency contact domains over gene bodies of highly expressed genes (Extended Data Fig. 5j).

In the sponge E. muelleri, we identified local interactions perpendicular to the main diagonal, and visually reminiscent to fountains observed in mouse, zebrafish and C. elegans⁴⁰ (Extended Data Fig. 6a,b). Manual inspection further revealed 84 focal contacts between distal genomic loci (Extended Data Fig. 6c), including gene promoters interacting with other regions showing promoter or enhancer-like chromatin signatures (Extended Data Fig. 6d,e). These weak distal interactions occurred between extended genomic regions, in contrast to the point-to-point contacts typical of chromatin loops (Extended Data Fig. 6c). Although chromatin loops were absent in E. muelleri, we identified 243 distal cis-regulatory elements, consistent with findings in other sponge species¹⁵. These elements were characterized by chromatin accessibility, with surrounding regions showing high H3K4me1 and low H3K4me3 signals, and were mostly intergenic but close to annotated TSS (median 3.8 kb) (Extended Data Fig. 6f). This distance-to-TSS distribution was similar to that of annotated enhancer elements in M. leidyi, T. adhaerens, N. vectensis and D. melanogaster that do not form loops (median 5.6 kb, compared to 31 kb in loop-forming enhancers) (Extended Data Fig. 6f), suggesting that sponges’ enhancer elements might function by proximity without the need for stable looping⁴¹.

Genome folding in the cnidarian N. vectensis was characterized by the presence of chromatin loops, as well as weakly insulated self-interacting domains (Extended Data Fig. 7). We identified 166 chromatin loops forming both promoter–promoter and promoter–enhancer contacts, and with some loops spanning nearly 1 megabase (Mb) (Extended Data Fig. 7a–c). Chromatin loops have also been reported in the hydrozoan Hydra vulgaris⁴², suggesting they are a conserved feature in cnidarians. Notably, some of the identified chromatin loops in N. vectensis showed a one-sided stripe pattern similar to those observed in other species, which are generated by cohesin extrusion⁴³. Moreover, we identified an enriched GTGT motif (FC = 327, P = 1 × 10⁻⁴⁰) present in 32% of loop anchors (Extended Data Fig. 7d). This motif resembles sequences with G-quadruplex-forming potential⁴⁴, which have been shown to stabilize enhancer–promoter interactions in other species⁴⁵. Beyond chromatin loops, we also observed self-interacting domains in N. vectensis (Extended Data Fig. 7e). The insulation boundaries of these domains were enriched for the YY1 motif (FC = 9,016, P = 1 × 10⁻⁸⁷) (Extended Data Fig. 7f), which is known to mediate chromatin interactions^35,45. These regions represent high-frequency contacts within the same gene regulatory landscape, but are not stabilized by chromatin loops as in vertebrate TADs³, nor are they as strongly insulated as the domains defined by insulator elements in D. melanogaster²⁷.

3D promoter hubs in placozoans

Our high-resolution chromatin contract maps revealed a complex 3D genome organization in the placozoan T. adhaerens, characterized by many loop contacts forming 3D interaction hubs (Fig. 3a). To confirm this observation, we profiled chromatin contacts in a distantly related placozoan species, Cladtertia collaboinventa, which showed a very similar pattern (Fig. 3a). Most of these interactions are promoter–promoter hubs (n = 2,413 for T. adhaerens and n = 3,239 for C. collaboinventa) (Extended Data Fig. 8a). Notably, 7–10% of chromatin contacts (n = 241 for T. adhaerens, n = 394 for C. collaboinventa) connected promoters with intronic or intergenic enhancer regions (Extended Data Fig. 8a,b), revealing the presence of distal cis-regulatory elements in placozoans.

Fig. 3: Promoter hubs in placozoans.

a, Example of syntenic genomic regions in placozoans T. adhaerens (TadhH1_4: 3860000–4060000, bin 800 bp) and C. collaboinventa (chr. 4: 8983000–9183000, bin 800 bp). b, Gaudí plots projecting ATAC-seq, H3K4me3 and exon annotation signals onto a two-dimensional Kamada–Kawai graph layout (top left) represented by the top 20% of contact pairs with solid colours highlighting statistically significant regions (P < 0.05) identified using a one-sided permutation test. The high–high (HH) signal marks genomic bins enriched in signal and that are in spatial proximity with other bins enriched in signal; low–low (LL) bins are depleted in signal as well as neighbourhoods are depleted in signal; high–low (HL) and low–high (LH) are bins that are enriched in signal, but not their neighbourhood, and in reverse. c, Classification of T. adhaerens genes into three categories (GP1, GP2 and GP3) on the basis of structural and epigenetic features. Top, example regions containing genes classified into GP1, GP2 and GP3 groups. The resolution of Micro-C maps is 800 bp, maximum intensity value of ICE normalized Micro-C maps is as in a. Bottom, average loop strength between promoter regions of the genes from each groups is measured with APA. The colour bar of pile-up plots shows enrichment of observed over expected values. d, Sequence motif found in loop regions, which are also overlapping GP1 promoter regions (left panel), is present in promoter regions of orthologous GP1 genes in other placozoan species (right panel). The total number of shared orthologues is indicated. TrH2, Trichoplax sp. H2; Hhon, Hoilungia hongkongensis; HoiH23, C. collaboinventa. e, Heatmaps showing CPM normalized ATAC-seq and ChIP–seq coverage, motif scores and Mutator transposable element density ±5 kb around the TSS of GP1, GP2 and GP3 genes. Each heatmap scale starts at zero.

We identified 321 promoter hub regions in the T. adhaerens genome and 331 in C. collaboinventa, involving 1,695 and 2,191 genes, respectively, with a median of four promoters in each hub in T. adhaerens and five in C. collaboinventa. To further reconstruct the 3D organization of these hubs, we used METALoci to calculate spatial correlation between genome folding and epigenetic (ATAC, H3K4me3 ChIP) or genomic (exon annotation) features (Fig. 3b and Extended Data Fig. 8c). This analysis revealed a nested structure where accessible promoter regions were central to the 3D interactions tightly clustered in space (ATAC-seq in Fig. 3b), whereas gene bodies and the first nucleosome (H3K4me3) occupied more peripheral locations (Fig. 3b). Furthermore, genes within spatial promoter hubs were linearly grouped along the genome, resembling the arrangement of housekeeping genes observed in mouse embryonic stem cells⁴⁶. Alternatively, these structures could be associated with active transcription and the formation of micro-compartmentalized RNA polymerase II-driven transcription hotspots⁴⁷.

Notably, not all collinear genes formed promoter hubs. Following this observation, we categorized genes into three groups based on their spatial and epigenetic organization (Fig. 3c and Extended Data Fig. 8d,e). This includes group 1 genes (GP1, n = 2,978 in T. adhaerens, n = 3,973 in C. collaboinventa) that had both ATAC and H3K4me3 peaks and formed chromatin loops, with an average interaction strength in aggregate peak analysis (APA) of 1.32, indicating the enrichment of Micro-C signal at loop anchors. Group 2 genes (GP2: n = 3,681 in T. adhaerens, n = 3,119 in C. collaboinventa) also showed ATAC and H3K4me3 peaks, but lacked strong distal contacts (APA = 1.12). Last, group 3 genes (GP3: n = 3,851 in T. adhaerens, n = 4,238 in C. collaboinventa) had neither chromatin loops (APA = 0.968) nor active chromatin marks (Fig. 3c). On average, GP1 genes showed a stronger H3K4me3 ChIP–seq signal and higher expression levels compared to genes in GP2 and GP3 (Extended Data Fig. 8d) and were associated with housekeeping functions, including intracellular trafficking, translation and messenger RNA processing (Extended Data Fig. 8f). By contrast, GP3 genes were enriched in cell type-specific functions related to peptidergic cells (Extended Data Fig. 8g,h), potentially explaining the lack of chromatin features in our bulk epigenomic experiments.

To understand what distinguishes placozoan GP1 genes, we analysed loop anchor sequences in both species using genomic sequences as background. We identified an enriched motif at chromatin loop anchor regions in both placozoan species (Fig. 3d and Extended Data Fig. 8i,j) and found that GP1 promoters frequently contained insertions of Mutator DNA transposable elements (Fig. 3e and Extended Data Fig. 8e), with the terminal inverted repeat (TIR) sequence of this transposon containing the identified sequence motif. To further explore this association, we constructed a phylogenetic tree including all intact Mutator TIR sequences in four placozoan species (Extended Data Fig. 8k and Supplementary Data 1). This analysis revealed a Mutator family shared across species and with consensus TIR sequences resembling the motif found in chromatin loops anchors (Extended Data Fig. 8k). The connection between chromatin loops and the Mutator transposable element suggests a potential evolutionary and functional relationship. One possibility is that an architectural protein in placozoans evolved to recognize the sequence motif within the Mutator TIRs, leading to ‘domestication’ of these sites as regulatory elements. Alternatively, the presence of the motif and Mutator TIR sequences may indicate targeted integration of Mutator transposons into promoter regions of highly expressed genes. Overall, our analyses showed that roughly one-third of T. adhaerens and C. collaboinventa genes are part of promoter hubs mediated by chromatin loops and that these contacts are associated with the presence of conserved Mutator DNA transposons harbouring a specific sequence motif.

Enhancer–promoter loops in ctenophores

The physical architecture of M. leidyi genome is dominated by thousands of chromatin loops (n = 4,261) (Fig. 4a), primarily connecting promoter and enhancer elements (61%), as well as enhancer to enhancer regions (16%) (Fig. 2g and Extended Data Fig. 9a,b). In total, we identified 916 gene promoters participating in chromatin loops, with each promoter contacting between one (50%) and up to 15 enhancers (Fig. 4b). These enhancers are mainly located in intronic (69%) and intergenic (24%) regions at one to eight genes from the contacted promoters. We also observed the accumulation of cohesin at loop anchor sites using ChIP–seq against SMC1 cohesin subunit (Extended Data Fig. 9c). To assess whether these features are conserved across ctenophores, we profiled chromatin contacts, albeit at lower resolution, in the cydippid ctenophore Hormiphora californensis (Extended Data Fig. 9d), which diverged from the lobate ctenophore M. leidyi roughly 180 million years ago⁹. At the sampled resolution, we detected 239 strong chromatin loops in H. californensis. In both ctenophores, genes involved in chromatin loop formation showed higher expression (Extended Data Fig. 9e).

Fig. 4: Chromatin loops in the ctenophore M. leidyi.

a, Example genomic region showing chromatin loops between promoters and enhancers at 400 bp resolution. b, Left, histogram of enhancer contacts per promoter. Right, genomic location of enhancers. c, Sequence motif enriched in loop anchors. d, DNA methylation profiles centred around motifs located at promoter and enhancer loop regions, or outside loops. e, Chromatin-bound proteome of M. leidyi, showing identified proteins sorted by abundance with architectural proteins CTEP1 and CTEP2 as the most abundant zf-C2H2s. f, DAP-seq signals around GC-motif sites with high (left) versus low (right) methylation levels, and sites located within (top) or outside (bottom) of loop anchors. CTEP1 showed higher affinity for unmethylated GC-rich motifs in DAP-seq assays with native or PCR amplified gDNA (lacking methylation). g, Boxplots showing PhastCons conservation scores across three ctenophore species (B. microptera, P. bachei and H. californensis). The boxplot limits indicate the interquartile range (IQR), with the median as the middle line and whiskers extending to 1.5× IQR. Two-sided Wilcoxon rank sum test showed significant conservation differences between intergenic enhancers (n = 969) and promoters in loops (n = 778) (***P = 1.3 × 10⁻¹⁵) and between promoters in loops and promoters outside loops (n = 14,996) (***P < 2.22 × 10⁻¹⁶), whereas intergenic enhancers and promoters outside loops showed no significant difference (not significant (NS), P = 0.88). h, Syntenic conservation within M. leidyi chromatin loops compared to H. californensis. Left plot, barplot showing the fraction of conserved orthologues (OGs) in all alignable genomic regions across ctenophore species (***P = 5.5 × 10⁻⁴, chi-squared test for given probabilities). Right plot, boxplot of shared orthologues between individual genomic regions within chromatin loops (n = 115) versus in random genomic regions (n = 259) of similar size (***P = 2.4 × 10⁻⁵, Wilcoxon rank sum test with continuity correction). Boxplot limits as in g. Silhouette of H. californensis in h reproduced from PhyloPic (https://www.phylopic.org/), created by S. Haddock and K. Wothe under a CC0 1.0 Universal Public Domain licence.

To investigate whether the chromatin loops in ctenophores are formed in a sequence-specific manner, we searched for the enriched motif in loop anchors of both species, using GC-normalized genomic random sequences as a background. We identified GC-rich motif (FC = 8,522; P = 1 × 10⁻⁴⁹⁷) that was present in over 75% of loop anchors (Fig. 4c and Extended Data Fig. 9f) and at both promoter (79%) and enhancer sites (74%) involved in chromatin loops (Extended Data Fig. 9g). In addition, this motif was found in an extra 3,348 gene promoters (21% of all genes) with no chromatin loops detected (Extended Data Fig. 9g,h).

As the identified GC-rich motif contains two CpG dinucleotides, we examined DNA methylation using long-read Nanopore sequencing data. The overall methylation level in M. leidyi was low (6.8%), in agreement with previous reports using whole-genome bisulfite sequencing⁴⁸. However, at loop anchor sites motifs showed low cytosine methylation, whereas motif occurrences outside loop anchor points showed high methylation (Fig. 4d and Extended Data Fig. 9i). Thus, we propose that DNA methylation of this GC-rich motif serves as a regulatory mechanism of loop formation in M. leidyi, potentially controlling the binding of an unknown, methylation-sensitive architectural DNA-binding factor, similar to mechanisms described for CCCTC-binding factor (CTCF) and other transcription factors⁴⁹.

The presence of DNA-binding proteins was further supported by the ATAC-seq footprint profile at motif regions in loop anchors (Extended Data Fig. 9j). To identify these potential architectural proteins, we profiled the chromatin-bound proteome of M. leidyi (Fig. 4e). We then selected the most abundant zf-C2H2 domain-containing proteins and analysed their DNA-binding specificity using DAP-seq, as zf-C2H2 factors are often associated with chromatin looping in other species^2,5,35,50. This analysis identified two proteins, named here CTEP1 (Ctenophore-specific Tethering Protein 1) and CTEP2, which overlapped with 80% of detected loop anchor regions and showed strong affinity for the same GC-rich motif we had previously identified (Extended Data Fig. 9k–m). Moreover, DAP-seq confirmed that the binding of both proteins was inhibited at sites with high DNA methylation (Fig. 4f and Extended Data Fig. 9k,m). Thus, we conclude that CTEP1 and CTEP2 bind unmethylated GC-rich motif sites at chromatin loops. Notably, these proteins are conserved across ctenophore species (Extended Data Fig. 9n and Supplementary Table 3), but are absent from genomes of other metazoans.

Finally, we analysed evolutionary conservation of the sequences at the loop anchor points. To this end, we calculated genome-wide conservation scores from alignments of M. leidyi genome with three other ctenophore species (Bolinopsis microptera⁹, Pleurobrachia bachei²⁰ and H. californensis⁵¹). Chromatin loop anchors, both at intronic and intergenic regions, showed higher sequence conservation compared to other introns or random genomic regions, respectively (Fig. 4g). The promoters of genes involved in distal contacts showed lower conservation score compared to other promoters (Fig. 4g), with conservation levels similar to those of random intergenic regions. Moreover, these promoters had a high frequency of transposable element integrations and elevated DNA methylation (Fig. 4d and Extended Data Fig. 9o). Furthermore, we found that genes located within enhancer–promoter loop regions in M. leidyi have higher syntenic conservation across ctenophore species compared to other genomic regions of similar size (Fig. 4h and Extended Data Fig. 9p). Overall, the conservation of loop anchor regions across ctenophore species and the increased syntenic linkage of genes suggest that gene positioning is constrained by genome architecture. These findings indicate that the distal chromatin contacts identified in M. leidyi represent an evolutionary conserved mechanism of genome regulation present in both lobate and cydippid ctenophores.

Discussion

Genome architecture is the result of both physicochemical and regulatory processes^3,4,31. In unicellular organisms, chromatin contact patterns are shaped by the polymer nature of the chromatin fibre³² and by gene transcriptional states⁵². For example, gene body contact domains are observed in highly transcribed genes in S. cerevisiae and S. pombe⁵², and in Arabidopsis thaliana⁵³. Also, insulation boundaries resulting from highly transcribed genes in divergent orientations are described in dinoflagellate genomes³³. In unicellular holozoans, we observed similar insulation patterns around TSSs, but without evidence of further regulatory features or sequence-specific determinants associated with insulation boundaries. We also found cosegregating inactive chromatin regions in the large genome of S. arctica, and to a lesser extent in S. rosetta³⁹. By contrast, these structures are absent in unicellular organisms such as C. owczarzaki or S. cerevisiae, which both have gene-dense genomes without heterochromatic regions.

In bilaterian species, extra chromatin structures involved in gene regulation have been observed, often mediated by architectural proteins binding to specific sequences^2,5,35,50. These include discrete chromatin loops between cis-regulatory elements and promoters, mediated by tethering elements²⁷, as well as insulated gene regulatory landscapes, such as loop TADs bounded by convergent CTCF sites in vertebrates³. Notably, TAD-like domain structures can also result from the passive cosegregation of active versus inactive chromatin states^37,54, rather than being determined by sequence-specific insulation elements. Examples of these are Polycomb bodies⁵⁵ and other heterochromatic compartment domains^28,29. In early-branching animals we did not identify loop-bound TADs or any evidence of sequence-defined insulated TADs. However, we did detect chromatin loops spanning tens of kilobases and linking distal cis-regulatory elements and promoters in cnidarians, ctenophores and placozoans. In the case of ctenophores, thousands of chromatin loops link enhancers and promoters, showing that distal loops can be extremely frequent even in small genomes (roughly 200 Mb). Another example is the thousands of chromatin loops in placozoans, with even smaller genomes (roughly 100 Mb). Both placozoans and ctenophores complex looping architectures are associated with transposable elements. Although the causal relationship between transposable elements and chromatin loops is unclear, this observation suggests that complex 3D genome architectures might be influenced by lineage-specific transposable element invasion histories⁵⁶.

The mechanisms and factors responsible for loop formation in non-bilaterians and most invertebrates remain unknown⁷. The zf-C2H2 protein CTCF is the main architectural protein in vertebrates and is conserved across bilaterians. In annelids it has been associated to open chromatin regions⁵⁷ and in cephalopods it defines TAD boundaries⁵⁸. Given that CTCF is absent in non-bilaterians³⁶, other factors, possibly from the zf-C2H2 family (Extended Data Fig. 9q), might be involved in the formation of these loops. In fact, a variety of architectural proteins other than CTCF have been described in Drosophila, many of which are zf-C2H2 proteins with restricted phylogenetic distributions such as the insect-specific CP190 factor^2,50,59. Similarly, we identified two ctenophore-specific zf-C2H2 proteins (CTEP1 and CTEP2) associated with loop anchor regions in M. leidyi. It is possible that other, yet unidentified, lineage-specific zf-C2H2 proteins contribute to chromatin architecture in different animal lineages.

Globally, our findings suggest an evolutionary scenario (Fig. 5) in which chromatin compartment domains defined by transcriptional states²⁸ (but lacking sequence-specific insulation or tethering elements) were present in the unicellular ancestor of animals, as seen in extant unicellular holozoans. At the origin of animals, distal cis-regulatory elements evolved, requiring sequence-determined, stable chromatin looping mechanisms to link these enhancers with gene promoters (at least at certain distances⁴¹). This added an extra layer of regulatory complexity to cell type-specific gene regulation. The origin of this distal gene regulation would also explain the existence of regulatory-linked genomic regions showing conserved synteny⁶⁰, as observed in ctenophore regions between loop anchor points. Moreover, domains insulated by sequence elements probably originated at the root of bilaterian animals, as they are observed in vertebrates, insects and probably spiralians^57,58. In the specific case of vertebrates these domains are formed by a mechanism of CTCF-dependent loop extrusion so far not observed in any other lineage⁷, which further exemplifies the potential diversity of mechanisms involved in chromatin architecture across metazoans. Future extended taxon sampling will further refine this evolutionary scenario and help solve open questions such as whether there are conserved or lineage-specific factors involved in the establishment of chromatin loops across animals, how dynamic these structures are in development and across cell types or when did sequence-determined, insulated TADs first emerged in animal evolution.

Fig. 5: The evolution of animal regulatory genome architecture.

a, Phylogenetic tree illustrating the taxonomic distribution of 3D-chromatin features. b, Schematic depicting major innovations in animal genome regulation at different ancestral nodes. LCA, last common ancestor.

Methods

Cell and animal cultures, sample preparation and crosslinking

S. arctica coenocytic culture was grown in marine broth (Difco, 3704 g l⁻¹) at 12 °C in 25 cm² flasks. Cells were passaged every 7 days using a 1:100 dilution. To synchronize cells in the G1/early S phase, an 8-day old culture was treated with 200 mM hydroxyurea (Sigma-Aldrich, catalogue no. H8627) for 18 h in the presence of 0.3% dimethylsulfoxide (DMSO). Synchronized cells were pelleted at 2,000g for 5 min at 12 °C, washed twice with Ca²⁺/Mg²⁺-free artificial sea water (CMFSW) (10 mM HEPES (pH 7.4), 450 mM NaCl, 9 mM KCl, 33 mM Na₂SO₄, 2.5 mM NaHCO₃) and flash-frozen in liquid nitrogen. Frozen cells were then reconstituted in CMFSW and crosslinked with 1% formaldehyde (Thermo Scientific, catalogue no. 28906) for 10 min under vacuum. The crosslinking reaction was quenched with 128 mM glycine for 5 min in the vacuum desiccator, followed by a 15 min incubation on ice. Cells were pelleted at 4 °C for 10 min at 2,000g, washed once with CMFSW, reconstituted in CMFSW to the concentration of 2 M ml⁻¹ and crosslinked with 3 mM DSG (Thermo Scientific, catalogue no. A35392) for 40 min at room temperature on a rotating wheel. The reaction was quenched with 400 mM glycine for 5 min. Double-crosslinked cells were pelleted at 4 °C for 15 min at 2,000g and flash-frozen in liquid nitrogen.

C. owczarzaki strain ATCC30864 was maintained in axenic culture at 23 °C in the ATCC (American Type Culture Collection) medium 1034 (modified PYNFH medium) in 25 cm² flasks. For subculture, filopodial cells were passaged every 2–3 days using a dilution of 1:100. Before collection, filopodial cells were synchronized in G1 or early S phase by treating a filopodial culture of 70–80% confluency with 100 mM hydroxyurea (Sigma-Aldrich, catalogue no. H8627) for 18 h (ref. ⁶¹). Synchronized cells were scraped off the surface and pelleted at 2,200g for 5 min at room temperature. Collected cells were crosslinked as described in ref. ⁶². Briefly, cells were crosslinked with 1% formaldehyde (Thermo Scientific, catalogue no. 28906) in PBS for 10 min on a rotating wheel at room temperature. The crosslinking reaction was quenched with 128 mM glycine for 5 min at room temperature followed by extra incubation on ice for 15 min. The crosslinked cells were pelleted at 4 °C for 10 min at 2,000g and washed once with ice-cold PBS. Cells were diluted in PBS to the concentration of 2 M ml⁻¹ and also crosslinked with 3 mM DSG (Thermo Scientific, catalogue no. A35392) for 40 min at room temperature on a rotating wheel. The crosslinking was quenched with 400 mM glycine for 5 min. Cells were pelleted at 4 °C for 15 min at 2,000g and flash-frozen in aliquots of 2 million cells.

S. rosetta was cocultured with Echinicola pacifica bacteria in artificial sea water supplemented with 20% cereal grass media (CGM3) at 23 °C. To synchronize the cell culture in the G1 or early S phase, cells from a 3-day-old culture were pelleted at 2,000g for 10 min and diluted in 4% CGM3 in artificial sea water to the concentration of 300,000 cells per ml. Cells were treated with 0.05 mM aphidicolin (Sigma-Aldrich, catalogue no. 178273) in the presence of 0.3% DMSO. After 18 h of incubation, cells, including chain colonies, fast and slow swimmers, were pelleted at 2,000g for 15 min. To remove bacteria from the choanoflagellate culture, collected cells reconstituted in 1 ml of culture media were passed through a Ficoll layer (1.6% Ficoll (Sigma-Aldrich, catalogue no. F5415), 0.5 M sorbitol, 50 mM Tris-HCl (pH 8.8), 15 mM MgCl₂, 1% artificial sea water) by centrifugation at 1,000g for 10 min at 4 °C. Pelleted choanoflagellate cells were then double-crosslinked with 1% formaldehyde in CMFSW and 3 mM DSG in CMFSW as described above for C. owczarzaki. The crosslinked cells were pelleted at 4 °C for 15 min at 2,000g and flash-frozen in liquid nitrogen.

E. muelleri sponges gemmules were hatched and grown for 1 week in Strekal’s media⁶³ in 150 × 25 mm culture dishes (Corning, catalogue no. 353025). To isolate phagocytic choanocyte cell population, specimens were fed for 10 min with 0.5 µm fluorescent carboxylate-modified FluoSpheres (Invitrogen, catalogue no. F8813) added to Strekal’s media to final 0.02% concentration (1:100 dilution of stock 2% FluoSpheres slurry)⁶⁴. Sponges were washed once with Strekal’s media, and 1% formaldehyde solution in Strekal’s media was added to crosslink specimens for 10 min at room temperature with occasional mixing. To quench formaldehyde, 128 mM glycine was added and incubated for 5 min at room temperature and 15 min on ice. Crosslinked sponge specimens were washed twice with ice-cold Strekal’s media. Roughly 80 specimens were transferred in 5 ml of the Strekal’s media and dissociated by trituration until all tissue was removed from the gemmule husks (roughly ten trituration passages). The dissociated cell suspension was filtered through a 40-µm cell strainer, and cells were diluted to 2 M ml⁻¹ concentration. The second crosslinking was performed with 3 mM DSG (Thermo Scientific, catalogue no. 20593) in Strekal’s media for 40 min at room temperature on a rotating wheel. The reaction was quenched with 400 mM glycine for 5 min at room temperature. Crosslinked cells were pelleted at 4 °C for 15 min at 2,000g, and then resuspended in 2 ml of ice-cold Strekal’s media with 2 µg ml⁻¹ Hoechst 33342 (Thermo Scientific, catalogue no. 62249). Choanocytes were isolated using a BD FACS Aria II sorter with BD FACSDiva v.6.1.3 (BD Biosciences) as cells showing both FluoSphere fluorescence and Hoechst nuclei staining. Fluorescence-activated cell sorting (FACS) profiles were analysed with FlowJo v.10.7 (Extended Data Fig. 1b).

M. leidyi specimens were kept in 300-ml glass beakers with 5–10 individuals at 21 °C in artificial sea water (Red Sea, catalogue no. R11055) with a salinity of 27 ppt. Ctenophores were fed daily with a mixture of living rotifers (Brachionus sp.) and brine shrimps (Artemia salina). The water was exchanged once a week. For all experiments, adult lobate animals were starved for 2 days before collection. To dissociate animal tissue, roughly five adult animals (10 mm long) were transferred into CMFSW and washed twice to exchange the buffer. Animal tissue was dissociated into single cells in 5 ml of fresh CMFSW by triturating every 2 min for a total of 10 min. The efficiency of tissue dissociation was monitored under the microscope. Dissociated cells were filtered through a 40-µm cell strainer and diluted to 2 M ml⁻¹ for the subsequent formaldehyde crosslinking. Cells were crosslinked in 1% formaldehyde in CMFSW for 10 min at room temperature. The reaction was stopped with 128 mM glycine for 5 min at room temperature and 15 min on ice. Crosslinked cells were pelleted at 4 °C for 10 min at 2,000g, washed once with CMFSW and resuspended to 2 M ml⁻¹ for a second crosslinking with 3 mM DSG in CMFSW. The crosslinking reaction was stopped after 40 min of incubation at room temperature on a rotating wheel with 400 mM glycine for 5 min. The crosslinked cells were pelleted at 4 °C for 15 min at 2,000g.

H. californensis specimens from the first generation (F1) of a laboratory-reared culture at the Monterey Bay Aquarium (USA) were flash-frozen and pulverized in liquid nitrogen. Extracted cells and nuclei were filtered through a 40-µm cell strainer and pelleted by centrifugation at 4 °C for 10 min at 2,000g. Cells were double-crosslinked with 1% formaldehyde in CMFSW and 3 mM DSG in CMFSW as described for M. leidyi.

T. adhaerens and C. collaboinventa colonies were grown in 200 × 30 mm glass Petri dishes at 21 °C in artificial sea water (Red Sea, catalogue no. R11055) with a salinity of 33 ppt. Placozoans were fed once a week with unicellular algae (Pyrenomonas sp.), the water was exchanged every second week. To prepare single-cell suspension, roughly 500 animals were collected, washed twice with CMFSW and resuspended in 1 ml of CMFSW supplemented with 2 mM EDTA. Animal tissue was triturated every 2 min for a total of 10 min at room temperature. The efficiency of dissociation was monitored under the microscope. Dissociated cells were filtered through a 40-µm cell strainer, diluted to 2 M ml⁻¹ and crosslinked with 1% formaldehyde in CMFSW for 10 min at room temperature on a rotating wheel. The reaction was quenched with 128 mM glycine for 5 min at room temperature and 15 min on ice. Cells were pelleted at 4 °C for 10 min at 2,000g, washed once with CMFSW and resuspended in 3 mM DSG in CMFSW for a second crosslinking. After 40 min of incubation at room temperature on a rotating wheel, 400 mM glycine was added to stop the reaction and cells were pelleted at 4 °C for 15 min at 2,000g.

N. vectensis NvElav1::mOrange transgenic line⁶⁵ was maintained in one-third artificial sea water (Red Sea, catalogue no. R11055) with salinity of 14 ppt. To isolate NvElav1::mOrange positive cells, 1.5–2-month-old animals starved for 1 day before the experiment were crosslinked with 1% formaldehyde in Ca²⁺/Mg²⁺-free one-third sea water (one-third CMF: 17 mM HEPES (pH 7.4), 167 mM NaCl, 9 mM NaHCO₃, 3.3 mM KCl) for 10 min under vacuum. The crosslinking reaction was stopped by adding 128 mM glycine and incubating the tissue under vacuum for 5 min, followed by a 15 min incubation on ice. The crosslinked tissue was dissociated into single cells by incubating the tissue with 10 mg ml⁻¹ of Protease XIV (Sigma-Aldrich, catalogue no. P5147) in one-third CMF and 1 mM CaCl₂ for 5 min at 24 °C triturating the tissue every 1 min. The digested tissue was pelleted at 800g for 5 min, reconstituted in one-third CMF supplemented with 2 mM EDTA and 2 µg ml⁻¹ Hoechst 33342 (Thermo Scientific, catalogue no. 62249), and the trituration continued for another 5–10 min. Dissociated cells were filtered through a 40-µm cell strainer, and neurons were isolated using a BD FACS Aria II as cells showing both the mOrange signal and Hoechst nuclei staining (Extended Data Fig. 1b). Isolated NvElav1::mOrange positive cells were also crosslinked with 3 mM DSG for 40 min at room temperature.

Micro-C library preparation

Micro-C libraries were prepared as previously described^11,12 with the following modification. Double-crosslinked cells (2 million cells per sample) with 1% formaldehyde and 3 mM DSG were permeabilized with 500 µl of MB1 buffer (10 mM Tris-HCl (pH 7.4), 50 mM NaCl, 5 mM MgCl₂, 1 mM CaCl₂, 0.2% NP-40, protease inhibitor cocktail) for 20 min on ice with occasional trituration. Cells were pelleted at 4,500g for 5 min at 4 °C and washed once with MB1 buffer. To digest chromatin to a 80% monomers to 20% dimer and oligomers nucleosome ratio, an appropriate amount of MNase (Takara Bio, catalogue no. 2910a) was added (Extended Data Fig. 1a), and samples were incubated for 10 min at 37 °C with mixing at 850 rpm. The digestion reaction was stopped with 4 mM EGTA (pH 8.0) followed by incubation at 65 °C for 10 min without agitation. Cells were washed twice with ice-cold MB2 buffer (10 mM Tris-HCl (pH 7.4), 50 mM NaCl, 10 mM MgCl₂, 0.1% BSA) and pelleted at 4,500g for 5 min at 4 °C. Next, to repair the fragment ends after MNase digestion, pelleted cells were resuspended in the repair reaction mix (5 µl of 10× NEBuffer 2.1, 34 µl of nuclease-free water, 1 µl of 100 mM ATP, 2.5 µl of 100 mM DTT) supplemented with 2.5 µl of 10 U µl⁻¹ T4 PNK (NEB, catalogue no. M0201). After 15 min of incubation at 37 °C with 850 rpm agitation, 5 µl of 5 U µl⁻¹ Klenow Fragment (NEB, catalogue no. M0210) was added to generate 3′–5′ overhangs in the absence of dNTPs for a subsequent incorporation of biotin-labelled dNTPs. The reaction mixture was incubated for another 15 min at 37 °C at 850 rpm. To biotinylate DNA fragment ends, the mixture of dNTPs was added to the reaction mix (2.5 µl of 10× T4 DNA Ligase buffer, 11.875 µl of nuclease-free water, 5 µl of 1 mM Biotin-dATP (Jena Bioscience, catalogue no. NU-835-BIO14), 5 µl of 1 mM Biotin-dCTP (Jena Bioscience, catalogue no. NU-809-BIOX), 0.5 µl of a mixture of 10 mM dTTP and dGTP, 0.125 µl of 20 mg ml⁻¹ BSA). After 45 min of incubation at room temperature with interval mixing at 850 rpm, the reaction was stopped with 30 mM EDTA (pH 8.0) followed by incubation at 65 °C for 20 min without agitation. The chromatin from lysed cells and nuclei was pelleted at 10,000g for 10 min at 4 °C and washed twice with MB3 buffer (50 mM Tris-HCl (pH 7.5), 10 mM MgCl₂). Finally, the chromatin was resuspended in 1,200 µl of proximity ligation mix (920 µl of nuclease-free water, 120 µl of 10× T4 DNA Ligase buffer, 100 µl of 10% Triton X-100, 12 µl of 20 mg ml⁻¹ BSA, 36 µl of 50% PEG 4000, 12 µl of 5 U µl⁻¹ T4 DNA ligase (Thermo Scientific, catalogue no. EL0012)) and incubated at room temperature for at least 2.5 h. To remove biotin from unligated ends, pelleted chromatin was treated with 2 µl of 100 U µl⁻¹ Exonuclease III (NEB, catalogue no. M0206) for 5 min at 37 °C and agitation 850 rpm. Then, chromatin was decrosslinked and deproteinased overnight at 65 °C at 850 rpm in the presence of 350 mM NaCl, 1% SDS and 1 mg ml⁻¹ proteinase K (Roche, catalogue no. 3115879001). The DNA was purified using DNA Clean & Concentrator-5 kit (Zymo Research, catalogue no. D4014) and eluted in 50 µl of 10 mM Tris-HCl (pH 8.0) (Extended Data Fig. 1c). Next, biotinylated proximity ligated DNA fragments were captured with Dynabeads MyOne Streptavidin (Life Technologies, catalogue no. 65602). DNA ends were prepared for adapter ligation and dA-tailed using NEBNext End repair/dA-tailing mix (NEB, catalogue no. E7546). The Y-shaped Illumina adapters were ligated with NEBNext Ultra II Ligation Module (NEB, catalogue no. E7595S), and the final library was amplified using NEBNext High-Fidelity 2× PCR Master Mix (NEB, catalogue no. M0541). The final libraries were double-size selected with Ampure XP (Beckman Coulter, catalogue no. A63881) resulting in libraries ranging from 350 to 750 bp in length. The detailed Micro-C stepwise protocol is reported in Supplementary Text 1.

High molecular weight gDNA extraction for genome sequencing

Genomic DNA (gDNA) from C. owczarzaki (Cowc) strain ATCC30864 was extracted with Blood & Cell Culture DNA Mini Kit (Qiagen, catalogue no. 13323). The library was constructed by the use of Ligation Sequencing Kit (Oxford Nanopore, catalogue no. SQK-LSK109) and NEBNext Companion Module (NEB, catalogue no. E7180), and sequenced with the R9.4.1 Flow Cell set on a MinION device (Oxford Nanopore). We obtained 4.3 M reads with an estimated Oxford Nanopore N50 of 5.4 kb.

E. muelleri gDNA was isolated using the Nanobind Tissue (Circulomics, catalogue no. NB-900-701-01) from 177 mg of frozen tissue of clonal juvenile sponges hatched from overwintering cysts (gemmules). Gemmules were obtained from the head tank of the Kapoor Tunnel (Sooke Reservoir), part of the drinking water system of the city of Victoria, British Columbia, Canada²¹. Short DNA fragments of less than 10 kb were removed with Short Read Eliminator Kit (Circulomics, catalogue no. SS-100-101-01). gDNA was quantified with a Qubit fluorometer and sequenced on an Oxford Nanopore using a PromethION flow cell (R9.4), producing 5.31 million reads with an estimated Oxford Nanopore N50 of 18.97 kb.

To reduce the level of heterozygosity during the assembly of M. leidyi genome (below), an animal culture was established from a single individual through self-fertilization. High molecular weight DNA was isolated from 5–8 animals (3–5 cm) starved for 24 h before flash-freezing. Frozen tissues were powdered with mortar and pestle, dissolved in 10 ml of urea extraction buffer (50 mM Tris-HCl (pH 8.0), 7 M Urea, 312.5 mM NaCl, 20 mM EDTA (pH 8.0), 1% w/v N-lauroylsarcosine sodium salt) as described in ref. ⁶⁶ and incubated for 10 min at room temperature on a rocking platform 20 rpm. gDNA was then purified twice with a phenol-chloroform-isoamyl alcohol mixture pH 7.7–8.3 (Sigma-Aldrich, catalogue no. 77617), precipitated with 0.7 volume of 100% isopropanol and subsequently washed twice with 70% ethanol. Finally, the isolated DNA was subjected to another round of purification with Nanobind Tissue kit (Circulomics, catalogue no. NB-900-701-01), followed by short-read elimination with the Short Read Eliminator Kit (Circulomics, catalogue no. SS-100-101-01). Sequencing was performed on Oxford Nanopore using PromethION flow cell (R9.4). We obtained 4.54 million reads with an estimated Oxford Nanopore N50 of 36.84 kb.

ATAC-seq library preparation

ATAC-seq libraries from M. leidyi and from sorted choanocytes of E. muelleri were prepared using Omni-ATAC protocol as described previously⁶⁷. Briefly, two M. leidyi adult specimens were dissociated using CMFSW with 0.25% α-Chymotrypsin (Sigma-Aldrich, catalogue no. C8946). To isolate nuclei, dissociated cells were transferred into cold hypotonic ATAC lysis buffer adjusted for marine animals (10 mM Tris-HCl (pH 7.5), 35 mM NaCl, 3 mM MgCl_2, 0.1% Tween-20, 0.01% NP-40, 0.01% digitonin, 70 µM Pitstop (Abcam, AB1206875MG)). Cell lysis was stopped after 2 min by adding marine ATAC wash buffer (10 mM Tris-HCl (pH 7.5), 35 mM NaCl, 3 mM MgCl₂, 0.1% Tween-20, 1% BSA). Nuclei were then pelleted and resuspended in cold PBS buffer with 0.8 M Sorbitol. We used 50,000 nuclei per each tagmentation reaction.

To sort choanocytes of E. muelleri, 7 days posthatching sponges were fed with 0.5 µm fluorescent carboxylate-modified FluoSpheres (Invitrogen, catalogue no. F8813). After 10 min of incubation, sponges were washed twice with Strekal’s media, collected and dissociated for 15 min at 28 °C using Protease XIV (Sigma-Aldrich, catalogue no. P5147) in Strekal’s media and 1 mM CaCl₂. Cells were pelleted at 800g for 5 min at room temperature and resuspended in Strekal’s media with 2 mM EDTA (pH 8.0). Further dissociation and trituration of sponge tissue continued for another 15 min at room temperature. Cells were filtered through a 40-µm cell strainer, stained with 2 µg ml⁻¹ Hoechst 33342 and sorted using FACS. Sorted cells were lysed for 3 min in ATAC lysis buffer (10 mM Tris-HCl (pH 7.5), 10 mM NaCl, 3 mM MgCl₂, 0.1% NP-40, 0.1% Tween-40, 0.01% digitonin). For each tagmentation reaction we used 100,000 nuclei. ATAC-seq libraries were prepared as described previously⁶⁷ and sequenced on Illumina NextSeq 500 using High-Output 75 cycles.

iChIP–seq library preparation

For S. arctica, C. owczarzaki, S. rosetta, M. leidyi, E. muellleri, T. adhaerens, C. collaboinventa and N. vectensis, double-crosslinked cells, as above, were washed with PBS, resuspended in 500 µl of cell lysis buffer (20 mM HEPES (pH 7.5), 10 mM NaCl, 0.2% IGEPAL CA-630, 5 mM EDTA, protease inhibitors cocktail) and incubated on ice for 10 min. Samples were centrifuged at 16,000g for 10 min at 4 °C. The resulting pellets were resuspended in bead beating buffer (20 mM HEPES (pH 7.5), 10 mM NaCl, 5 mM EDTA, protease inhibitors cocktail), and then transferred to 0.2 ml tubes containing acid-washed glass beads (Sigma-Aldrich, G8772). Cells were lysed by vortexing five times for 30 s. The supernatant was transferred to a 1.5-ml sonication tube, SDS was added to 0.6% and samples were sonicated 3–5 cycles of 30 s on, 30 s off in a Bioruptor Pico (Diagenode) to generate 200–300 bp fragments. Chromatin was diluted with 5 volumes of dilution buffer (20 mM HEPES (pH 7.5), 140 mM NaCl), centrifuged at 16,000g for 10 min at 4 °C and stored at −80 °C before use.

Chromatin immunoprecipitation was performed as previously described⁶⁸ with the following modifications. Briefly, for each species 100 ng of chromatin was used for immunoprecipitation. The pool of chromatin was incubated for 14–16 h at 4 °C with 5 µl (1:50 dilution) of anti-H3K4me1 (Cell Signaling, catalogue no. 5326), 6 µl (3.4 µg) of anti-H3K4me2 (Abcam, catalogue no. ab32356), 2.5 µl (1:100 dilution) of anti-H3K4me3 (Millipore, catalogue no. 07-473), 5 µl (5 µg) of anti-SMC1 (Thermo Fisher, A300-055A) or 2 µl (2 µg) of anti-H3 (Abcam, catalogue no. ab1791) and recovered using a 1:1 mix of Protein A (Sigma-Aldrich, catalogue no. 16-661) and Protein G (Sigma-Aldrich, catalogue no. 16-662) magnetic beads. Immunoprecipitated complexes were washed, reverse crosslinked for 3 h at 68 °C, deproteinased and then purified using Ampure XP beads (Beckman Coulter, catalogue no. A63881). Final libraries were prepared using the NEBNext Ultra II DNA Library Prep Kit (New England BioLabs) according to the manufacturer’s protocol. ChIP–seq libraries were sequenced on Illumina NextSeq 500 sequencer using High-Output 75 cycles.

MARS-seq library preparation

Single-cell libraries were prepared from freshly dissociated and sorted choanocytes of E. muelleri as previously described⁸. To collect cells for MARS-seq libraries, 7 days posthatching, sponges were fed with 0.5 µm fluorescent carboxylate-modified FluoSpheres (Invitrogen, catalogue no. F8813). Animal tissues were dissociated and prepared for sorting as described above for ATAC-seq. Dissociated cells were sorted through FACS into four 384-well MARS-seq plates. In total, 1,536 single-cell libraries were prepared and sequenced on an Illumina NextSeq 500 using High-Output 75 cycles.

Chromatin proteomics

Chromatin proteomics samples were prepared as previously descibed⁶⁹ with minor modifications. Briefly, double-crosslinked cells of M. leidyi (1 million per replicate) were solubilized in 1 ml of lysis buffer (4 M guanidine thiocyanate, 100 mM Tris-HCl (pH 8.0), 10 mM EDTA, 2% N-lauroylsarcosine sodium salt) and incubated for 10 min. Next, before adding DNA-binding beads (Invitrogen, catalogue no. 37002D), cell lysate was mixed with 1 ml of 2-propanol. The beads were separated on a magnet and the supernatant was saved as the unbound control. The beads were washed using 1 ml of wash buffer (1:1 lysis buffer to 2-propanol ratio), transferred to a 1.5-ml sonication tube and washed again with 1 ml of 80% ethanol. The chromatin was then eluted in 200 µl of 10 mM Tris-HCl (pH 8.0) containing proteinase inhibitors (Roche, catalogue no. 04693132001) and sonicated using a Bioruptor Pico at 4 °C for 3 cycles (30 s ON, 30 s OFF). To remove RNA-binding proteins, RNase A (Roche, catalogue no. 10109142001) was added to the sonicated samples, which were then incubated at 37 °C with agitation in a thermomixer. Afterwards, chromatin was re-bound to the beads by adding 250 µl of lysis buffer, vortexing and then sequentially adding 300 µl of 2-propanol. The beads were washed twice with 1 ml of 80% ethanol, and proteins were digested on the beads using trypsin (Promega, catalogue no. V5111) and LysC (NEB, catalogue no. P8109S).

Chromatographic and mass spectrometric analysis

Samples were analysed using an Orbitrap Eclipse mass spectrometer (Thermo Fisher Scientific) coupled to an EASY-nLC 1200 (Thermo Fisher Scientific (Proxeon)). Peptides were loaded directly onto the analytical column and were separated by reversed-phase chromatography using a 50-cm column with an inner diameter of 75 μm, packed with 2-μm C18 particles (Thermo Fisher Scientific, catalogue no. ES903).

Chromatographic gradients started at 95% buffer A (0.1% formic acid in water) and 5% buffer B (0.1% formic acid in 80% acetonitrile) with a flow rate of 300 nl min⁻¹ and gradually increased to 25% buffer B and 75% A in 52 min and then to 40% buffer B and 60% A in 8 min. After each analysis, the column was washed for 10 min with 100% buffer B.

The mass spectrometer was operated in positive ionization mode with nanospray voltage set at 2.4 kV and source temperature at 305 °C. The acquisition was performed in data-dependent acquisition mode and full mass spectrometry scans with one micro-scan at resolution of 120,000 were used over a mass range of m/z 350–1,400 with detection in the Orbitrap mass analyser. Automatic gain control was set to ‘standard’ and injection time to ‘auto’. In each cycle of data-dependent acquisition analysis, following each survey scan, the most intense ions above a threshold ion count of 10,000 were selected for fragmentation. The number of selected precursor ions for fragmentation was determined by the ‘Top Speed’ acquisition algorithm and a dynamic exclusion of 60 s. Fragment ion spectra were produced by means of high-energy collision dissociation at normalized collision energy of 28% and they were acquired in the ion trap mass analyser. Automatic gain control and injection time were set to ‘Standard’ and ‘Dynamic’, respectively, and an isolation window of 1.4 m/z was used. Digested bovine serum albumin (NEB, catalogue no. P8108S) was analysed between each sample to avoid sample carryover and to assure stability of the instrument, and Qcloud⁷⁰ was used to control instrument longitudinal performance during the project.

Acquired spectra were analysed using the Proteome Discoverer software suite (v.2.5, Thermo Fisher Scientific) and the Mascot search engine (v.2.6, Matrix Science⁷¹). The data were searches against M. leidyi database and a list of common contaminants (16,042 entries)⁷² as well as all the corresponding decoy entries. For the peptide identification a precursor ion mass tolerance of 7 ppm was used for the MS1 level, trypsin was chosen as enzyme and up to three missed cleavages were allowed. The fragment ion mass tolerance was set to 0.5 Da for MS2 spectra. Oxidation of methionine and N-terminal protein acetylation were used as variable modifications whereas carbamidomethylation on cysteines was set as a fixed modification. False discovery rate in peptide identification was set to a maximum of 1%.

Peptide quantification data were retrieved from the ‘Precursor ions quantifier’ node from Proteome Discoverer (v.2.5) using 2-ppm mass tolerance for the peptide extracted ion current. The obtained values were used to calculate protein fold-changes and their corresponding P value and adjusted P values.

DAP-seq (DNA affinity purification sequencing) library preparation

The DNA-binding domains of candidate zf-C2H2 proteins were cloned from complementary DNA (cDNA) library of M. leidyi into the pIX-HALO vector using NEBuilder HiFi DNA Assembly Master Mix (NEB, catalogue no. E2621). The obtained HALO-fusion constructs were translated using the TnT SP6High-Yield Wheat Germ Protein Expression System (Promega, catalogue no. L3260). Next, an adapter-ligated DNA library was prepared from native gDNA of M. leidyi using NEBNext Ultra II FS DNA library prep kit (NEB, catalogue no. E7805) or PCR amplified gDNA. The binding to HALO-zf-C2H2 fusion proteins and recovery of adapter-ligated gDNA libraries was performed as described in ref. ⁷³. The generated DAP-seq libraries were sequenced in paired-end mode on an Illumina NextSeq 500 using High-Output 75 cycles.

De novo genome assembly and scaffolding

We made preliminary genome assemblies of C. owczarzaki from Oxford Nanopore reads basecalled by Guppy v.6.0.1 using NextDenovo v.2.5.0 (ref. ⁷⁴), Flye v.2.9.0 (ref. ⁷⁵) and NECAT v.0.0.1 (ref. ⁷⁶), which produced 20, 141 and 56 contigs including the mitochondrial genome, respectively. For the Flye assembly, we only used 5,000 bp or longer reads. We then integrated the three assemblies by manually comparing them to each other, with a help of reciprocal large-scale alignments generated with minimap2 (ref. ⁷⁷). The integrated assembly was polished with the Nanopore reads using Flye ten times, and with Illumina reads⁷⁸ using HyPo⁷⁹ twice. A chromosome-scale duplication, which was in the end included in chromosome 15 after the 3D assembly (Extended Data Fig. 2d), was temporarily removed before annotating the genome. Finally, we manually inspected the whole assembly sequence together with the mapped Illumina data, Nanopore data and the previous Sanger sequence data²⁵, and navigated them on the Integrative Genomic Viewer (IGV)⁸⁰ to find and fix errors occurred during the consensus calling. We also manually phased chimeric haplotypes for some genes using the long reads. In total, 7,937 nucleotides were manually inserted or deleted at 430 sites and 1,193 nucleotides were substituted at 1,081 sites.

We produced two new genome assemblies for E. muelleri and M. leidyi. In both cases, we used Oxford Nanopore reads after base call correction using Guppy v.5.0.17 (using the dna_r9.4.1_450bps_sup_prom.cfg configuration the super-accurate base calling model, and a filtering reads with min_qscore=10). Then, we used two different long-read assemblers (Flye v.2.9-b1768, ref. ⁷⁵ and Shasta v.0.8.0, ref. ⁸¹) and various assembly strategies (filtering by read length at 0, 10 and 50 kb), and selected the best resulting draft assemblies for each species. To that end, we evaluated the contiguity (measured using the contig N50), completeness and occurrence of uncollapsed haplotypes for each draft (Extended Data Fig. 2a,b). Contiguity was evaluated using total assembly length and contig N50. Completeness was measured with the fraction of conserved orthologues recovered by BUSCO v.5.1.2 (ref. ⁸²) (using the genome mode and the metazoa_odb10) and the fraction of mappable genes from the original assemblies (mapped using Liftoff v.1.6.1, ref. ⁸³). The presence of uncollapsed haplotypes was assessed with the distribution of per-base sequencing depths, calculated using the pbcstat utility in purge_dups v.1.2.5 (ref. ⁸⁴) (for which we remapped the input reads to the assembly with minimap2 2.18-r1015 (ref. ⁷⁷), using the -x map-ont preset for long-read mapping) (Extended Data Fig. 2e,f).

The best drafts for each species were produced using the following parameter combinations: (1) for E. muelleri, we used the Shasta assembler with the Nanopore configuration (--config Nanopore-Oct2021 flag), without filtering by read length (estimated sequencing depth roughly 100×) and (2) for M. leidyi, we used Flye with reads filtered at 50 kb (estimated sequencing depth roughly 150×), the raw Nanopore read configuration (--nano-raw flag) and an estimated total assembly size of 200 Mb.

Then, we used purge_dups to collapse putative uncollapsed haplotypes in each assembly, in the following manner: (1) we split the assembly into contigs with the split_fa utility; (2) we aligned the genome to itself with minimap2 and the -x asm5 preset; (3) we used the read alignments to the unsplit assembly (produced with minimap2 -x map-ont) to obtain the sequencing depth histogram and calculate coverage cutoffs with pbcstat and calcuts, respectively; (4) we used these cutoffs and the mapped reads to remove haplotigs and overlaps for the draft, with purge_dups proper and using two rounds of alignment chaining (-2 flag) and finally (5) we reevaluated the assembly quality using per-base sequencing depth distributions (above) and reductions in the fraction of duplicated BUSCO orthologues.

Chromosome-level assembly

To obtain chromosome-level genome assemblies, generated Micro-C libraries were mapped to de novo draft genome assemblies (C. owczarzaki, E. muelleri and M. leidyi) or current genome assemblies (T. adhaerens ASM15027v1, ref. ²², S. arctica²⁴, S. rosetta GCA_000188695.1, ref. ²⁶, C. collaboinventa⁸⁵) using Juicer v.1.6 (ref. ⁸⁶) with an option -p assembly. Proximity ligation alignments were used by 3D de novo assembly pipeline⁸⁷ to order and orient available contigs into chromosomes with the following parameters: S. arctica -r 3 --editor-repeat-coverage 10, C. owczarzaki -r 0 --editor-repeat-coverage 4, S. rosetta -r 3 --editor-repeat-coverage 2, E. muelleri -r 2 --editor-repeat-coverage 10, M. leidyi -r 2 -i 1000 --editor-repeat-coverage 2, T. adhaerens -r 3 --editor-repeat-coverage 2 and C. collaboinventa -r 3 --editor-repeat-coverage 2. The resulting assemblies were manually reviewed and corrected with Juicebox Assembly Tools⁸⁸ (Extended Data Fig. 2c–f). Finally, chromosome-level genome assemblies were polished with Medaka (v.1.5.0) to correct possible sequence errors such as indels and mismatches, as follows: (1) first, we mapped the Nanopore reads to the chromosome-level assembly using the minimap2-based mini_align utility; (2) we then used Medaka consensus to obtain consensus sequences, specifying a batch size of 200 (--batch 200 flag) and the r941_prom_sup_g507 configuration (--model flag) and (3) we merged the consensus and variant calls for all chromosomes into a polished assembly using Medaka stitch.

Genome annotation

To annotate the C. owczarzaki genome, we did not mask the repeats because the intergenic regions are very small²⁵ and, thus, masking only increased annotation failure on duplicated genes. We used BRAKER2 (ref. ⁸⁹) with OrthoDB⁹⁰ protein sequence collections as hint data, as well as with RNA-seq data from a previous study⁶¹. The three preliminary annotations, evidenced by metazoan proteins, protozoan proteins and RNA-seq data, were combined with TSEBRA⁹¹, giving rise to 9,069 annotated transcripts. Finally, we manually searched and fixed wrong annotations by navigating the assembly on IGV⁸⁰, comparing the combined annotation with the three preliminary annotations together with the mapped RNA-seq data. By this careful inspection, we modified or newly annotated 1,871 transcripts including alternatively spliced ones. Compared to the previously published proteome²⁵ (v.2), only 4,076 out of 8,792 proteins (including alternatively spliced ones) had completely matched sequences to the those predicted in this study, allowing simple amino acid mismatches probably accounting for polymorphisms.

To annotate M. leidyi genome we first downloaded developmental Illumina RNA-seq samples (GSE93977), trimmed them with fastp and built a de novo Trinity assembly, which was mapped to the genome using gmap⁹². The RNA-seq was also directly mapped to genome using HISAT2 (ref. ⁹³) with the –dta parameter, and genome-based transcriptomes were built for each sample using StringTie⁹⁴. Merged mapped RNA-seq samples were then used to find high-quality intron junctions using Portcullis. The combination of Trinity, StringTie and Portcullis intron junctions were then fed to Mikado for transcript selection. The best resulting gene models based on mapping to UniProt were then used to train an Augustus model for M. leidyi. Augustus was used for an ab initio gene prediction, using exonic hints from Mikado, intron hints from Portcullis and coding sequence hints from a MetaEuk⁹⁵ run with query fasta files combining proteins from H. californensis and UniProt. Mikado transcripts and Augustus gene models were then merged using EVidenceModeler (scores of 10 for Mikado transcripts and 2 for Augustus gene models). The resulting gene models were updated with PASA⁹⁶ to incorporate the untranslated regions from the Mikado transcripts.

To annotate S. arctica, S. rosetta, E. muelleri, T. adhaerens and C. collaboinventa genome assemblies, gene models from previous assemblies were mapped onto new coordinates using Liftoff (v.1.6.1)⁸³ with -overlap 1 -flank 1 options.

Repeat annotation

Repetitive sequences and transposable elements were annotated using EDTA (v.2.1.0)⁹⁷ with the following parameters: --sensitive 1 --anno 1 (Extended Data Fig. 2g). For H. sapiens, we used RepeatMasker (v.open-4-0-3) annotation of GRCh38 genome released by UCSC.

DNA methylation calling from Nanopore long-read sequencing data

The fast5 files obtained from the PromethION were used as input for Megalodon (v.2.5), with the Remora model dna_r9.4.1_e8 sup for 5hmc_5mc modification only on CG dinucleotides. We then built bigwig files using the bedGraphToBigWig tool from UCSC. The Megalodon CG methylation calls were compared to previously published Whole-Genome Bisulfite Sequencing remapped to the new reference genomes using Bismark (SRR8346013 and SRR10356110)^21,48. Both data sources were congruent, yet Nanopore had deeper and broader coverage, we used Megalodon methylation data for subsequent analysis.

Micro-C data processing

Micro-C data were processed using the 4D Nucleome processing pipeline⁹⁸. Briefly, raw reads were mapped to the reference genome using bwa mem (v.0.7.17-r1188) with the -SP5M option. The mapped reads were sorted and filtered with pairtools (v.0.3.0)⁹⁹. Pairs that mapped within a 2-bp distance from each other were considered duplicates. We also discarded reads mapping within the distance of 200 bp, which eliminates self-ligated pairs and reads mapping to adjacent nucleosomes. Only uniquely mapping pairs and 5′ most unique alignments of multiple ligations pairs were aggregated into 200-bp bin contact matrices and multiresolution .cool or .hic files (Extended Data Fig. 1d). Contact matrices were normalized with cooler (v.0.8.11)¹⁰⁰ using the iterative correction and eigenvector (ICE) balancing method¹⁰¹ for .cool files or with Juicer tools⁸⁶ using Knight–Ruiz balancing¹⁰² for .hic files. All contact heatmaps were visualized with either Cooltools (v.0.5.1)¹⁰³ or Coolbox (v.0.3.8)¹⁰⁴ and genome assembly heatmaps were visualized using HiGlass¹⁰⁵.

Reproducibility between replicates was estimated using the stratum-adjusted correlation coefficient (SCC) implemented in HiCRep¹⁰⁶ at resolutions of 1, 2, 5, 10, 25 and 50 kb (Extended Data Fig. 1f). The SCC scores were averaged across chromosomes. Biological replicates with SCC score estimated above 0.7 at resolutions equivalent to roughly 20,000 bins per species genome (resolution of 10 kb for S. arctica, 1 kb for C. owczarzaki, 2 kb for S. rosetta, 10 kb for E. muelleri, 10 kb for M. leidyi, 5 kb for H. californensis, 5 kb for T. adhaerens, 5 kb for C. collaboinventa and 10 kb for N. vectensis) were pooled to obtain final chromatin interaction matrices. Technical replicates were first merged, deduplicated and only then combined into the final contact maps.

The decay of the average contact frequency over genomic distance from 1 kb to 100 Mb was calculated using Cooltools (v.0.5.1)¹⁰³. The decay curves were calculated for each chromosome separately, and then averaged across chromosomes (Extended Data Fig. 1g).

Compartment analysis

Compartment analysis was performed on observed-over-expected contact maps at resolutions equivalent to 5,000, 10,000, 20,000 and 50,000 bins per species genome (Extended Data Fig. 3a) using Cooltools eigs-cis¹⁰³. We visually examined calculated eigenvectors, and, for each organism, the E1 vector corresponded to the compartmentalization pattern of contact maps. Active (A) and inactive (B) compartment types were assigned by GC content (for all species except C. owczarzaki) or H3K4me3 chromatin signal for C. owczarzaki, such that higher GC regions or positively correlated with H3K4me3 signal regions correspond to A compartment. Saddle plots were generated using the Cooltools saddle module. Specifically, the eigenvectors were sorted from lowest to highest value and combined into 40 groups according to their eigenvector value. The first (bottom 2.5% E1 values) and last (top 2.5% E1 values) groups were ignored to exclude potential outliers. The observed-over-expected value of the remaining 38 groups was averaged across all bins and chromosomes and visualized as saddle plots.

Compartment strength was calculated as the ratio of homotypic (AA + BB) over heterotypic (AB + BA) compartment contacts. We choose the top 20% of observed-over-expected values for both homotypic and heterotypic interactions. To assess the error in estimating compartment strength, we compared the compartments strength across different resolutions as well as performed visual inspection of the contact maps (Extended Data Fig. 3b). The latter showed varying degrees of accuracy in identifying compartment types between species, with the algorithm performing particularly poorly on M. leidyi due to the lack of well-defined chromatin compartmentalization in this species in our sample. Therefore, we assigned an extra intermediate compartment I to the intermediate eigenvalues close to zero. To that end, we modelled the genome-wide eigenvalues distribution as a Gaussian mixture with three components using the normalmixEM function from the mixtools R package (v.2.0.0) as described¹⁰⁷. The B–I and I–A thresholds were defined as intersection points between components (Extended Data Fig. 3c).

To characterize the distribution of genomic features in the A, I and B compartments, we calculated cumulative H3K4me3 chromatin signals and RNA-seq expression values for each compartment region. Furthermore, we estimated the percentage of bases annotated as transposable elements or coding gene regions within these compartments. All the values are presented as −log₂(1 − the value’s quantile). Thus, a normalized value of six means that the coverage is in the upper 1–2⁻⁶ quantile, that is, in the upper 1/64th of the distribution (Extended Data Fig. 3d).

Insulation profiles and boundaries classification

To compute the insulation profiles, we first determined the optimal resolution and window sizes for a target genome. To that end, we calculated insulation scores using Cooltools insulation module¹⁰³ at resolutions roughly equivalent to 50,000, 100,000, 200,000 and 400,000 genomic bins per species genome with a sliding window for each resolution that is ×5, ×10 and ×25 the applied resolution (Extended Data Fig. 4a). The resolution and two window sizes with maximum average insulation scores were considered optimal because they reflected the strongest partitioning of genomes into isolated domains. Insulation boundaries located within two bins of unmappable genomic region were removed.

Identified insulation boundaries were categorized into strong and weak using the peak prominence of their boundary strength distributions (Li threshold) as implemented in the Cooltools insulation score module. Strong boundaries were further annotated with overlapping genomic features that fall within one bin of the annotated feature from the insulation boundary. For example, if compartment boundaries were called at the resolution 5 kb, then the maximum distance to the closest insulation boundary is ±10 kb.

To estimate internal interactions within contact domains, rescaled pile-ups were generated using coolpup.py¹⁰⁸. Contact domains were defined as valleys between two strongly insulated regions, which were not further from each other than 100 kb.

Loop calling and annotation

Chromatin loops were identified using SIP v.1.6.1 (ref. ¹⁰⁹) on KR-normalized contact matrices. In M. leidyi, the SIP peak caller was applied with the following parameters: -norm KR -g 3.0 -min 2.0 -max 2.0 -mat 5000 -d 10 -res 400 -sat 0.01 -t 2000 -nbZero 6 -factor 4 -fdr 0.05 -isDroso false. For T. adhaerens, chromatin loops were called with the following parameters: -norm KR -g 5.0 -min 4.0 -max 4.0 -mat 5000 -d 20 -res 100 -sat 0.01 -t 2000 -nbZero 6 -factor 4 -fdr 0.05 -isDroso false; for C. collaboinventa: -norm VC_SQRT -g 1.5 -min 3.0 -max 3.0 -mat 5000 -d 20 -res 500 -sat 0.01 -t 2000 -nbZero 6 -factor 2 -fdr 0.05 -isDroso false; for H. californensis: -norm KR -g 2.5 -min 3.0 -max 3.0 -mat 5000 -d 10 -res 1000 -sat 0.01 -t 2000 -nbZero 6 -factor 4 -fdr 0.05 -isDroso false. Identified loops were then filtered based on APSscore, removing high-intensity signals outside the normal distribution of APSscore values. This threshold ensured accurate removal of false positive regions that corresponded to structural genomic rearrangements, such as inversions or assembly artefacts. For H. californensis, we kept only annotated loops with values greater than ten. Chromatin loops in N. vectensis and focal chromatin contacts in E. muelleri were annotated manually.

Each loop anchor was assigned a promoter or enhancer identity based on their epigenetic signature. We calculated quantile normalized counts per million (CPM) coverage of H3K4me3, H3K4me2 and H3K4me1 ChIP signals in 1-kb (T. adhaerens, C. collaboinventa, N. vectensis), 2-kb (M. leidyi, E. muelleri, D. melanogaster) or 10-kb (H. sapiens) windows from a centre of a loop anchor.

METALoci autocorrelation analysis

METALoci¹¹⁰ (v.0.3.0) analysis was applied to explore the spatial distribution and autocorrelation of epigenetic signal in T. adhaerens contact maps. For each region of interest at 800 bp resolution, the top 20% pairs of contacts were used to create a two-dimensional graph layout by means of the Kamada–Kawai algorithm¹¹¹ (Fig. 3b, top left panel) using the ‘metaloci layout’ with default parameters. Next, the signal of interest (H3K4me3 ChIP, ATAC, genic exon annotation) measured in the 800-bp genomic bin was mapped onto the built graph layout using ‘metaloci lm’ with default parameters. Spatial and epigenetic signals were embedded into Voronoi diagrams for enhanced visualization as a Gaudí plot (Extended Data Fig. 8c), and the local Moran’s index (LMI) analysis^112,113 was applied for each bin of the Gaudí plot.

According to LMI analysis, each bin is assigned to one of the four distinct groups, called LMI quadrants, based on the signal value in a bin and average signal value in its neighbourhood. If a bin and its neighbourhood have similar amounts of signal (low or high), then this bin is assigned to a low–low (blue) or high–high (red) quadrant. If a bin and its neighbourhood have different amounts of signal, then the bin is assigned to a low–high (cyan) or high–low (orange) quadrants, respectively. Significantly colocalized bins according to LMI, in which a P value is obtained using a permutation test, are highlighted by colour in the LMI scatterplots (Extended Data Fig. 8c,j, left panels). An analogous colouring scheme is applied to the Voronoi diagrams of the Gaudí plots. Hence, the highlighted blue and red bins on a Gaudí plot represent bins in which the signal is significantly colocalized in the space. Thus, ATAC-seq, H3K4me3 and motif score signals are significantly enriched inside the nested focal contacts (Fig. 3b and Extended Data Fig. 8c), whereas exons are significantly enriched outside loop contacts (Fig. 3b). METALoci code is available at the GitHub repository (https://github.com/3DGenomes/METALoci).

Motif analysis

Loop anchor regions of M. leidyi (n = 8,523) and H. californensis (n = 478) were scanned for enriched motifs with HOMER¹¹⁴ in de novo motif discovery mode. As background sequences, we used random genomic regions of equivalent size and GC content (n = 38,810 in M. leidyi and n = 49,097 in H. californensis). For motif enrichment analysis in T. adhaerens (n = 3,557) and C. collaboinventa (n = 4,037), we scanned loop anchor regions using random genomic sequences of equivalent size (n = 32,004 in T. adhaerens and n = 36,178 in C. collaboinventa) as background. Loop anchor regions of N. vectensis (n = 327) were scanned for enriched motif using random genomic sequences (n = 45,268) of equivalent size and GC content as background. In addition, we used ATAC-seq accessible neuronal promoter regions (n = 22,961) as background to scan for enriched motifs in genomic regions that overlap ATAC-seq peaks located at the non-loop insulation boundaries (n = 9,016). To annotate genomic regions with identified motifs, we used the monaLisa package¹¹⁵, selecting percentile threshold of motif scores by comparing the motif score distributions in the target regions with genome-wide motif score distributions (Extended Data Figs. 7d,f, 8i and 9f).

Whole-genome alignment and sequence conservation analyses

We evaluated the degree of sequence conservation of the M. leidyi genome by comparing it to other ctenophores (B. microptera, P. bachei and Hormiphora californiensis). To that end, we first aligned all genomes to each other using Cactus v.2.6.4 (ref. ¹¹⁶), following a progressive approach guided by the species trees of ctenophores, namely: ((M. leidyi, B. microptera), (H. californiensis, P. bachei)). Second, we used the hal2maf utility from the HAL toolkit v.2.2 (ref. ¹¹⁷) to create MAF (multiple alignment format) alignments of each chromosome, using M. leidyi as reference. To identify conserved regions in these genomes, we used the rphast v.1.6.1 implementation of the Phast toolkit¹¹⁸, as follows: (1) we used phyloFit¹¹⁹ to create an initial null model of neutral change based on the fourfold degenerate codon positions of each genome’s coding regions, using a general reversible nucleotide transition matrix and the predefined species tree; (2) we used phastCons to optimize this model using the expectation–maximization procedure, re-estimating transition probabilities and tree parameters at each step (the optimization step was performed using only the longest chromosome in each genome).

Loop synteny analysis in M. leidyi

We evaluated the degree of syntenic conservation of the loop regions in M. leidyi compared to the other ctenophore genomes, and compared it to that of length-matched regions not involved in loops. To that end, we first identified orthologous genes across the four ctenophore species (M. leidyi, B. microptera, P. bachei and H. californensis) using Broccoli v.1.2 (ref. ¹²⁰) to obtain orthologous gene pairs (step 4), using predicted peptide sequences as input (longest isoform per gene only). Within Broccoli, we used the maximum-likelihood gene tree inference algorithm (based on IQ-TREE¹²¹) and set a k-mer length of 10,000 to avoid the removal of paralogous sequences from the analysis. Second, we mapped pairs of loop anchor regions from M. leidyi (2,353 pairs of promoter–enhancer and 99 promoter–promoter loops, n = 2,452 in total) to their closest overlapping genes, and used these genes and their orthologs as anchors to map these regions to the other ctenophore genomes. In parallel, we randomly selected length-matched, non-loop overlapping regions from the M. leidyi genome to compare their synteny conservation with that of loop regions (using the randomizeRegions function in the regioneR R package¹²² to select 3× background regions, n = 7,356). Then, for each pair of species, we evaluated the synteny conservation of the foreground (loop) and background regions (random non-loops) from the point of view of the flanking synteny-anchoring genes, using two different metrics: (1) the fraction of shared orthologous genes between the flanking genes across all regions in the foreground and background sets (testing the significance of the difference using a χ² test for given probabilities) and (2) the distributions of per-region shared orthologs (tested using the one-sided Wilcoxon rank sum test with continuity correction).

ATAC-seq analysis

We used previously published datasets of ATAC-seq for C. owczarzaki¹⁷, S. rosetta³⁹, T. adhaerens¹²³, C. collaboinventa¹²³, D. melanogaster¹²⁴ and H. sapiens¹²⁵ as well as newly generated datasets for N. vectensis, M. leidyi and E. muelleri. Sequenced reads were demultiplexed and converted to fastq files using bcl2fastq v.2.20 Illumina. Raw reads were filtered and trimmed with Trimmomatic (v.0.39)¹²⁶ before mapping to the reference genome with bwa mem (v.0.7.17-r1188) and duplicates were marked with bamsormadup from biobambam2 (https://github.com/gt1/biobambam2). Using deeptools alignmentSieve aligned reads were filtered and shifted with -ATACshift, which corresponds to mate reads being shifted +4 and −5 bp for positive and negative strands, respectively. To generate nucleosome-position data tracks, nucleosome-free and nucleosome-bound regions were defined using the following length thresholds 0–120 and 150–240 bp, respectively. ATAC peaks were called with MACS2 (ref. ¹²⁷) on shifted nucleosome-free regions. Footrpint ATAC score was calculated using TOBIAS v.0.13.3 (ref. ¹²⁸).

ChIP–seq analysis

We analysed publicly available dataset for D. melanogaster¹²⁹ and H. sapiens¹³⁰ and 34 newly generated ChIP–seq datasets as described below. Raw reads after removal of 3′-adapters and quality filtering with Trimmomatic (v.0.39)¹²⁶ were aligned to the reference genome with bwa mem (v.0.7.17-r1188). Duplicated reads were marked with bamsormadup (https://github.com/gt1/biobambam2), and peaks were called using MACS2 (v.2.2.6)¹²⁷. Aggregated density plots were visualized with deeptools (v.3.1.3)¹³¹.

DAP-seq analysis

Raw reads from amplified and native gDNA fragments bound by HALO-zf-C2H2 protein fusions were analysed as described for ChIP–seq. Motif enrichment analysis was performed using HOMER¹¹⁴ in de novo motif discovery mode for MACS2 identified narrow peaks resized to 300 bp (for CTEP1 n = 14,638; for CTEP2 n = 10,615). GC- and size-normalized random genomic regions were used as background (for CTEP1 n = 25,964; for CTEP2 n = 30,744).

RNA-seq analysis

We used previously published datasets of bulk poly-A enriched RNA-seq for S. arctica¹³², C. owczarzaki⁶¹, S. rosetta³⁹, D. melanogaster¹²⁴ and H. sapiens¹³³ (Supplementary Table 2). To process data, raw reads were aligned to the reference genome using STAR (v.020201)¹³⁴ in --quantMode to estimate the number of read counts per gene. In downstream analysis, gene counts were reported as −log₂(1 − gene counts quantile).

MARS-seq analysis and single-cell expression atlases

Single-cell MARS-seq libraries generated previously^8,135 were aligned to new reference genomes of E. muelleri (GCA_049114765.1), M. leidyi (GCA_048537945.1) and N. vectensis (GCA_932526225.1) using Liftoff or de novo annotated gene models. To improve single-cell RNA-seq quantification, gene annotations for E. muelleri and M. leidyi have been extended using GeneExt¹³⁶. Briefly, MARS-seq alignment files have been subsampled to 100 M reads and MACS2 (ref. ¹²⁷) was used to call peaks using default parameters. Intergenic peaks were filtered based on the 20th percentile of the genic peak coverage and each gene was extended to the most distant peak within 5,000 nucleotides. Metacell and clustering analyses were performed as previously described⁸. The single-cell expression atlas for T. adhaerens was obtained from a previously published dataset¹²³.

Public datasets used in this study

All public datasets used in this study are listed in Supplementary Table 2. ATAC-seq, ChIP–seq and RNA-seq datasets were analysed as described above.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.