Symbioses have shaped life on Earth, from the origin of the eukaryotic cell to the formation of biodiversity hotspots such as coral reefs1,2. Animal chemosynthetic symbioses, where bacteria convert inorganic compounds to organic matter, are ubiquitous in marine habitats3 and fuel some of the most productive communities, such as those around hydrothermal vents4. Siboglinid worms (Annelida) often dominate deep-sea chemosynthetic environments through symbioses with environmentally acquired bacteria5,6 that adults harbour within a specialised organ called a trophosome7. Despite their ecological importance, the host’s genetic traits sustaining these symbioses have only been studied in Vestimentifera8,9,10, one of the four main lineages in Siboglinidae (Fig. 1a). The genomes of Lamellibrachia luymesi8, Paraescarpia echinospica9, Riftia pachyptila10 and Ridgeia piscesae11 have revealed a complex molecular interplay between Vestimentifera and their endosymbionts to fulfil their nutritional demands12. For example, the hosts have lost genes involved in amino acid biosynthesis8,10,11 and carbohydrate catabolism9 but expanded gene families involved in nutrient transport8, gas exchange8,9,10,13,14, innate immunity9,11 and lysosomal digestion8,9,10,15. On the other hand, there is genomic information for the endosymbionts of most major clades of Siboglinidae, including Vestimentifera, Osedax and Frenulata16,17,18. The endosymbionts of Vestimentifera and Frenulata are mixotrophs19 and show a diverse metabolic repertoire for energy production (e.g., the reductive tricarboxylic acid cycle in the endosymbionts of Vestimentifera) and nutrient biosynthesis that complements the metabolic deficiencies of, at least, the vestimentiferan host16,17. In addition, an increase in the genetic repertoire to infect and evade the host’s immunity16,17,18, transport nutrients18 and metabolise nitrogen compounds16,17 is common in endosymbionts of Siboglinidae. Notably, many of these genetic changes also occur in other distantly related chemosymbiotic animals, including bivalves20, gastropods21, and the clitellate annelid Olavius algarvensis22. Therefore, disparate animal groups have convergently evolved similar genetic mechanisms to sustain different chemosynthetic symbioses in marine ecosystems.

Fig. 1: Siboglinidae is a symbiotic annelid group.
figure 1

a Siboglinidae is a diverse clade of annelid worms that evolved chemosynthetic symbioses (left side). There are four main lineages within Siboglinidae, namely Frenulata, Osedax, Sclerolinum and Vestimentifera. Chemolithoautotrophy occurs in Frenulata, Sclerolinum and Vestimentifera, which associate with gammaproteobacteria that employ sulphur or methane to produce organic compounds in an array of marine ecosystems, from reducing sediments to methane seeps and hydrothermal vents (right side of the panel). Differently, Osedax worms (e.g., O. frankpressi; b, c) have secondarily evolved a heterotrophic association with Oceanospirillales to exploit decaying vertebrate bones. The genomic basis for the evolution of these nutritional symbioses in Siboglinidae is unclear (question marks on the left) because genomic information only exists for Vestimentifera hosts (green circles on the right). The species herein studied are highlighted in boldface. b, c Photographs of O. frankpressi in a whale bone (b; arrowheads point to O. frankpressi) and a mature female adult (c). O. frankpressi settles and colonises decaying vertebrate bones (b). There, the posterior part of the body becomes stably infected with environmentally acquired Oceanospirillales bacteria. This body part (the so-called roots) harbours the bacteria and grows to penetrate the bone, dissolving the organic components. These nutrients are absorbed and transported towards the bacteriocytes containing the endosymbionts, which will proliferate and act as food for the worm. Anterior to the root tissue there are the reproductive ovisacs and the head bears two pairs of palps.

Within Siboglinidae, the marine Osedax annelids have evolved a unique endosymbiosis23,24,25,26,27 with heterotrophic bacteria in the order Oceanospirillales18,24,28,29,30 (Fig. 1a) that allows them to obtain nutrients from bones of dead animals lying on the ocean’s floor (Fig. 1b). While Osedax shares some morphological features with other siboglinids31, including the lack of a gut, mouth and anus, Osedax contains bacteriocytes concentrated in the subepidermal connective tissue of the lower trunk that grows directly into the bone24,28 (Fig. 1c). This amorphous tissue, referred to as “roots”, expresses high levels of V-type H+-ATPase and carbonic anhydrase32, indicating acid is used to dissolve the bone matrix to access collagen and lipids, which are then absorbed across the root epithelium. Enzymatic28,29 and transcriptomic data33 support this theory by showing that the roots of Osedax express many proteases and solute carrier transporters that are thought to be involved in bone degradation and nutrient absorption, perhaps with the aid of the endosymbionts18. However, it is currently unclear whether the specialised heterotrophic symbiosis of Osedax is based on homologous genetic traits to those discovered in Vestimentifera and other chemoautotrophic invertebrates or if it relies on unique genomic adaptations. Untangling the molecular mechanisms behind this remarkable symbiosis is, therefore, central to understanding the evolution of Osedax and Siboglinidae, as well as the ecological principles and succession of bone-eating communities34.

In this study, we sequenced the genome of Osedax frankpressi Rouse, Goffredi & Vrijenhoek, 200424, as well as that of two vent-dwelling Vestimentifera, Oasisia alvinae Jones, 1985 and Riftia pachyptila Jones, 1981, and compared them with nearly 40 eukaryote and prokaryote genomes to better understand the genomic changes leading to these distinct symbiotic lifestyles. In contrast to Vestimentifera, we found that O. frankpressi has a small AT-rich genome with a reduced gene repertoire. Gene families typically expanded in chemosymbiotic hosts, such as innate immunity components, are reduced in O. frankpressi. Instead, the Osedax-Oceanospirillales symbiosis has unique genomic adaptations for bone digestion, including the loss of biosynthetic pathways of amino acids that are abundant in vertebrate bones in the host, the presence of the glyoxylate cycle in the endosymbiont that could allow the production of carbohydrates from the lipids present in vertebrate bones, and the expansion of matrix metalloproteases in the host that could aid in bone digestion. Together, our findings demonstrate that different genomic principles sustain the nutritional symbioses of Osedax and Vestimentifera, providing critical insight into the genetic and metabolic changes that have enabled symbiotic siboglinids to colonise diverse nutrient-imbalanced feeding niches.


The genome of O. frankpressi

To identify genomic signatures that could inform the genetic and physiological basis of the heterotrophic symbiosis in Osedax, we used long PacBio reads and short Illumina reads to assemble the genome of O. frankpressi24 (Supplementary Table 1). We also sequenced the genomes of two Vestimentifera from hydrothermal vents, Oasisia alvinae and R. pachyptila (Supplementary Fig. 1), complementing previous genome sequencing efforts8,9,10. We generated almost entirely haploid draft assemblies (Supplementary Fig. 2a–d), which included the circularised endosymbiont genomes of O. frankpressi and Oasisia alvinae and several epibionts associated with O. frankpressi (Supplementary Fig. 2h–j; Supplementary Table 2). Consistent with k-mer-based analyses (Supplementary Fig. 2e–g), previously reported genome size estimation for Oasisia alvinae35, and a recent genome assembly of R. pachyptila10, the assembled genomes for O. frankpressi, Oasisia alvinae and R. pachyptila span 285 Mb (1,185 scaffolds with an N50 of 426 Kb), 808 Mb (642 scaffolds with an N50 of 2.975 Mb) and 554 Mb (918 scaffolds with an N50 of 1.424 Mb) after removal of bacterial contigs, respectively (Fig. 2a; Supplementary Fig. 2k). The genome assemblies for Oasisia alvinae and R. pachyptila shows high completeness (96.9% and 95.6% BUSCO presence, respectively; Supplementary Fig. 2l; Supplementary Table 3). The assembly for O. frankpressi appeared to have lower completeness (80.1% BUSCO presence; Supplementary Fig. 2l). However, 95.62% and 97.77% of the de novo assembled transcripts from the body and root tissue mapped to the genome assembly of O. frankpressi, respectively. Accordingly, BUSCO completeness increased to a final score of 96.23% after gene annotation (Supplementary Fig. 2l) and manual curation (26 out of the 62 missing BUSCO could be manually annotated; Supplementary Data 1). Together, this suggests that the fast rates of molecular evolution in coding sequences observed in Osedax worms36 are likely responsible for the relatively low initial, assembly-based BUSCO completeness in the genome of O. frankpressi.

Fig. 2: Osedax and Vestimentifera exhibit distinct genome evolutionary trends.
figure 2

ac Plots comparing genome size (a), repeat content (b) and number of genes (c) between O. frankpressi and the four Vestimentifera with sequenced genomes. Osedax frankpressi has a smaller genome, with less genes but relatively similar repeat content. d Principal component analyses of the gene content of 28 metazoan genomes show that differently from symbiotic bivalves and gastropods, the gene content of Vestimentifera and O. frankpressi differs from slow-evolving asymbiotic species (as represented by Owenia fusiformis and C. teleta). While Vestimentifera has a unique gene content, O. frankpressi is like other fast-evolving annelid lineages. e, f Bar plots of the percentage of genes in gene families (i.e., orthogroups; e) and retained ancestral metazoan gene families (f) for ten annelid lineages. Osedax frankpressi is amongst the annelids with less genes in gene families and less retained ancestral metazoan genes. g Patterns of gene family gains (in green) and loss (in red) during the evolution of Annelida under a consensus tree topology31 and a consensus of published molecular dates8, 9. A major event of gene loss is common to all Siboglinidae. While O. frankpressi continued experiencing high rates of gene loss, a major event of gene innovation is common to all Vestimentifera. h Top five enriched gene ontology terms (Biological Process) for gene families lost (top) and expanded (bottom) in O. frankpressi. While O. frankpressi has further lost genes involved in metabolism (e.g., carbohydrate metabolism), genes involved in collagen and extracellular matrix degradation are expanded. P-values were derived from upper-tail Fisher’s exact tests.

Although the genome of O. frankpressi is ~50–75% smaller than the sequenced genomes and estimated genome sizes of Vestimentifera8,9,10,35 (Fig. 2a), the fraction of simple repeats and transposable elements in O. frankpressi (29.16%) is comparable to that of the vestimentiferan R. pachyptila (27.87%) and asymbiotic annelids with similar genome sizes (Fig. 2b; Supplementary Fig. 3a). As in Vestimentifera, the repeat landscape in O. frankpressi shows signs of expansions (Supplementary Fig. 3b), unlike in asymbiotic annelids with slow rates of molecular evolution37,38. Combining transcriptomic evidence (Supplementary Table 1) with ab initio gene prediction (Supplementary Fig. 2a), we functionally annotated 37,777 and 38,179 protein-coding transcripts in Oasisia alvinae and R. pachyptila, respectively (Supplementary Fig. 2k), which have a similar number of genes to other Vestimentifera and asymbiotic annelids8,37,38. The number of genes annotated in our assembly for R. pachyptila is higher than in a previous report10 (Supplementary Fig. 4a). Still, both annotations and assemblies are broadly equivalent (Supplementary Fig. 4b–d). Unlike Vestimentifera, O. frankpressi has a smaller repertoire of 18,657 transcripts (Fig. 2c), comparable to that of the miniaturised Dimorphilus gyrociliatus36, another annelid species with a compact genome and a streamlined gene set (14,203 genes). Therefore, O. frankpressi has the smallest genome of all sequenced siboglinids. Given the number of genes in genomes of asymbiotic annelids, gene loss rather than removal of repeat content seems to account for the genome size difference between these two lineages of Siboglinidae.

Gene gains and losses shape the evolution of Siboglinidae

To investigate gene content evolution between major lineages of Siboglinidae, we first reconstructed the gene families of 28 highly complete metazoan genomes, including seven symbiotic annelid and molluscan lineages (Supplementary Data 2). This taxonomic sampling provides sufficient resolution to infer the time of origin of each gene family while minimising potential biases in orthology inference in fast-evolving species39. A principal component analysis of the number of orthologs per gene family in the 28 species clustered the symbiotic molluscs Bathymodiolus platifrons20 and Gigantopelta aegis21 with their asymbiotic bivalve and gastropod relatives, respectively (Supplementary Fig. 5a). However, the four Vestimentifera species are markedly differentiated from the other annelid and animal genomes, and O. frankpressi is closer to heterotrophic annelids with fast rates of molecular evolution and divergent gene repertoires, such as the leech Helobdella robusta and the earthworm Eisenia andrei—which also harbour bacterial symbionts40,41,42—and the marine worm D. gyrociliatus (Fig. 2d; Supplementary Fig. 5a). Indeed, after R. pachyptila, O. frankpressi is the annelid with the second lowest percentage of genes assigned to gene families (Fig. 2e) and has only retained a fraction of ancestral metazoan gene families comparable to more rapidly evolving annelids such as H. robusta and D. gyrociliatus (Fig. 2f). Therefore, unlike symbiotic molluscs, the evolution of nutritional symbioses in Siboglinidae correlates with divergent host gene repertoires compared to their asymbiotic annelid counterparts.

To identify and characterise the evolutionary events underpinning the divergent gene repertoires of Siboglinidae, we reconstructed the patterns of gene family evolution in those 28 metazoan genomes under a consensus tree topology (Supplementary Fig. 5b). Vestimentifera and O. frankpressi share a major gene loss event involving 2270 gene families of mainly ancient origins (61.23% of the lost families originated before Metazoa and the Bilateria/Nephrozoa ancestor) (Fig. 2g) and enriched in Gene Ontology (GO) terms associated with metabolism (Supplementary Fig. 5c). This loss thus coincides with the evolution of nutritional symbioses in the last common ancestor of Siboglinidae. A high rate of gene loss continued in the O. frankpressi lineage (Fig. 2g), which ultimately accounts for its reduced gene repertoire and primarily affected genes associated with carbohydrate and nitrogen metabolism (Fig. 2h; Supplementary Fig. 5d). Notably, Vestimentifera experienced an event of gene family expansion in its last common ancestor (2,437 gene families), mainly affecting genes related to immunity, cell communication, and response to stimuli9 (Fig. 2g; Supplementary Fig. 5e). However, high lineage-specific rates of gene loss also occur in some Vestimentifera10,11, as in O. frankpressi (Fig. 2g). Compared to Vestimentifera, O. frankpressi has had few gene family gains (Supplementary Fig. 5b) but has experienced a large expansion of gene families associated with extracellular matrix remodelling and degradation (e.g., collagen degrading proteases; Fig. 2h; Supplementary Fig. 5f) in agreement with previous transcriptomic observations33. Altogether, our findings indicate that the evolution of symbiosis in Osedax and Vestimentifera relies on different host gene repertoires, one sculptured predominantly through gene loss (in O. frankpressi) and another through ancestral gene gains followed by varying, species-specific rates of gene loss (in Vestimentifera)8,9,10 (Fig. 2g).

The different genomic traits of Siboglinidae endosymbionts

To investigate the genetic and functional contribution of the endosymbionts to the nutritional symbioses of Siboglinid worms, we used our PacBio long-read data to assemble the genomes of the primary endosymbionts of O. frankpressi (Rs1 ribotype; Genome Taxonomy Database accession number Rs1 sp000416275) (Fig. 3a; Supplementary Data 3), and Oasisia alvinae (Supplementary Fig. 6; Supplementary Data 4), as well as several epibionts associated with Osedax43 (Supplementary Table 2). The circularised assembly of the endosymbionts of O. frankpressi improved the previously published genome18, revealing 95 new functional genes that provide additional insights into its symbiosis (Supplementary Data 5). Compared to deep-sea free-living relatives, the O. frankpressi endosymbiont has a genome enriched in metabolic genes for protein secretion systems, carbohydrate metabolism, and coenzyme and amino acid biosynthesis (Supplementary Data 6b). This includes additional virulence factors, such as multiple complete copies of Type 5a, 5b, and 6i secretion system pathways (Supplementary Data 6c) that are important for modulating interactions with other bacteria and eukaryotic hosts. Neptunomonas japonica, a close relative of the Oceanospirillales endosymbionts recovered from marine sediments near a whale fall, has many of the same metabolic capabilities of the endosymbionts; however, it lacks the additional secretion systems44. The Type 5a and Type 5b secretion systems are also largely absent in the endosymbionts of Vestimentifera (Supplementary Data 7g). This increase in virulence factors may reflect that Oceanospirillales repeatedly infect the roots of Osedax as it grows through bone material, unlike the trophosome of Vestimentifera, which is colonised early during host development7,45. In addition, all Siboglinidae endosymbionts contain numerous genes encoding eukaryote-like protein domains, which, interestingly, tend to be host-lineage-specific (Supplementary Data 8). Eukaryote-like proteins modulate important processes in many symbioses, including extracellular secretions, cell binding and colonisation12,46,47. Therefore, the specificity of the endosymbionts’ eukaryote-like proteins in the different lineages of Siboglinidae suggests they may be important for host and clade-specific annelid-symbiont communication, as shown in Riftia12.

Fig. 3: Osedax and its endosymbiont reconstruct the glyoxylate cycle.
figure 3

a Circular schematic representation of the genome of Osedax endosymbiont Rs1, assembled into a single contig. The plot shows the genomic location of genes involved in amino acid, lipid and vitamin/cofactor metabolism (in orange, blue and red, respectively) and the GC content (inner circle; brown colour). b Oceanospirillales endosymbionts possess the glyoxylate cycle, a metabolic pathway that can produce oxaloacetate, which can serve as the precursor to synthesise carbohydrates from the oxidation of fatty acids. This metabolic pathway could thus contribute to synthesising glucose in a diet (the bone) that is naturally poor in carbohydrates. Notably, this molecular and metabolic interaction does not occur between Vestimentifera and its symbionts because the host and microbes lack the enzyme isocitrate lyase.

Osedax frankpressi’s endosymbionts and those of Vestimentifera and Frenulata shared a broadly similar repertoire of genes involved in core cellular processes (Supplementary Data 7a). However, as we may expect for a heterotrophic microbe, the Oceanospirillales endosymbionts have significantly more genes involved in the metabolism and uptake of amino acids, coenzymes, lipids, and carbohydrates (Supplementary Fig. 7a–c; Supplementary Data 7a, e, f). This includes several complete pathways to convert oxaloacetate into ribose 5-phosphate that can be used in the biosynthesis of nucleotides and histidine, the Entner-Doudoroff and De Ley-Doudoroff pathways to catabolise carbohydrates, and multiple sugar, amino acid and oligopeptide ATP-binding transporters (Supplementary Data 7). In addition, the endosymbionts of O. frankpressi can produce all essential amino acids (including methionine and threonine) and vitamin B6, unlike Vestimentifera and Frenulata endosymbionts, as well as vitamin B2, which Vestimentifera symbionts cannot make (Fig. 4a; Supplementary Data 7d, e). Notably, the B2 pathway was considered missing in the previous draft genome of the Oceanospirillales endosymbiont18, but it is present in ours. As in some of the bacteria comprising the microbiome of degrading bones48, O. frankpressi’s endosymbionts can catabolise hydroxyproline, one of the most abundant amino acids in collagen49,50, but lacks a secreted M9 peptidase to cleave extracellular collagen (Supplementary Data 3). Finally, all endosymbionts of Vestimentifera are enriched in genes involved in chemosynthesis, most of which are absent in the heterotroph endosymbionts of Osedax (Supplementary Fig. 7c). Taken together, our results confirm and expand previous genomics efforts on the Oceanospirillales endosymbionts18, further demonstrating that Siboglinidae has partnered with metabolically versatile microbes that are suited to sustain symbioses with eukaryotes in diverse environments.

Fig. 4: Osedax’s metabolic adaptations to bone digestion.
figure 4

a Summary table of the presence (filled circles) and absence (empty crosses) of amino acid biosynthetic pathways in seven annelid genomes and O. frankpressi endosymbiont (symbiont Rs1). While Vestimentifera and asymbiotic annelids can synthesise all amino acids that are non-essential and conditional for humans, O. frankpressi shows incomplete pathways to synthetise proline, arginine, and serine (in red). Some of these amino acids are abundant in the bone (e.g., proline) and all can be produced by the symbiont (tyrosine biosynthetic pathway is truncated in the symbiont; dotted and lighter circle). bd Schematic representation (as in MetaCyc database) of the biosynthetic pathways for proline (b), serine (c) and arginine (d) indicating with red and violet circles the enzymes present in O. frankpressi and its endsymbiont, respectively. Osedax frankpressi cannot produce serine from glycolytic metabolites but can either produce serine from collagen-derived glycine or take it from the diet. In addition, O. frankpressi can only convert arginine into ornithine, producing urea as a result. e, f Heatmaps of normalised mRNA expression levels for amino acid biosynthetic enzymes (e) and glycine catabolising enzymes (f) in the body and roots of O. frankpressi. Biosynthetic enzymes (e), including the two copies of serine hydroxymethyltransferase (SHMT-a and SHMT-b) that convert glycine into serine, are more expressed in the roots than in the body of O. frankpressi. Source data for (e, f) are provided as a Source Data file.

Metabolic adaptations for bone digestion

Vertebrate bones are nutrient-imbalanced food sources enriched in lipids and proteins and deficient in carbohydrates50. Given the reduced gene repertoire of the host (Fig. 2c) and the metabolic versatility of its endosymbionts18 (Supplementary Data 7), we next explored potential molecular and metabolic interactions that could facilitate the nutritional specialisation of O. frankpressi. Combining highly sensitive profile hidden Markov Models sequence similarity searches with KEGG and COG functional annotations, we reconstructed all metabolic routes in O. frankpressi and its endosymbionts and the published genomes of Vestimentifera and their respective endosymbionts. Osedax frankpressi and Vestimentifera have similar metabolic capabilities to produce and process lipids (Supplementary Data 9). However, O. frankpressi has six incomplete pathways for carbohydrate metabolism that are intact in Vestimentifera and other asymbiotic annelids (Supplementary Data 9), consistent with the loss of gene families involved in carbohydrate metabolism (Fig. 2h). In one case (UDP-N-acetyl-D-glucosamine biosynthesis), the endosymbiont possesses the enzymes that would complement the losses in O. frankpressi (Supplementary Data 7c, Supplementary Data 9). Notably, the endosymbionts, unlike the host, lack the enzymes glycogen synthase and glycogen phosphorylase and, therefore, cannot produce glycogen (Supplementary Data 7) and possess the enzymes to complete the glyoxylate cycle (Fig. 3b, Supplementary Data 7), which allows the production of glucose from the catabolism of fatty acids and acetate51,52,53,54. This metabolic pathway does not occur in Vestimentifera because both the host and endosymbiont lack an isocitrate lyase (Fig. 3b). Therefore, the glyoxylate cycle may play a role in the metabolic interaction of Osedax and its endosymbionts by collectively converting bone lipids into carbohydrates, which are often nearly absent in bones50. Although Osedax appears to use wax esters to store energy29, the fat content of bones varies widely, and Osedax can grow in dentin (Greg W. Rouse and Shana K. Goffredi, personal observation), where lipids are a minor component. Functional studies are thus warranted to assess the nutritional and physiological relevance of this metabolic pathway in Osedax and under different nutritional sources.

Proteins, predominantly collagen50, are the core organic component of bone. Collagen is rich in proline/hydroxyproline and glycine49, and thus its amino acid composition is also imbalanced. Consistent with previous genomic analyses8,9,10, Vestimentifera and asymbiotic annelids (Owenia fusiformis and C. teleta) can produce all non-essential and conditionally essential amino acids. However, O. frankpressi cannot synthesise the amino acids proline, serine, and arginine (which are non-essential or conditional for mammals), but its endosymbionts can (Fig. 4a). Indeed, only one enzyme (pyrroline-5-carboxylate reductase) of the proline biosynthetic pathway remains, which is expressed at similar levels in the roots and the rest of the body, unlike most amino acid biosynthetic enzymes that are enriched in roots (Fig. 4b, e). Similarly, the entire pathway to synthesise serine from intermediates of glycolysis is missing in O. frankpressi (Fig. 4c). However, O. frankpressi (like other annelids) has an intact glycine cleavage system (Fig. 4f), which would favour the conversion of collagen-derived glycine into serine through serine hydroxymethyltransferase55. The two copies of this enzyme are highly expressed throughout O. frankpressi (Fig. 4f) and could provide an additional source of serine on top of those offered by the diet and endosymbionts. Therefore, O. frankpressi shows genomic-inferred metabolic adaptations to its unique bone-eating diet in its gene complement, which differs from the more intact metabolic repertoire of Vestimentifera and other asymbiotic annelids12.

The catabolism of amino acids produces ammonia, a compound that can be toxic but can also serve as a substrate for amino acid biosynthesis by both animals and bacteria. Most aquatic organisms excrete excess ammonia into the water, but a few aquatic animals and most air-breathing vertebrates shuttle ammonia into the urea cycle leading to urea production56. Osedax frankpressi lacks four urea cycle enzymes and only possesses arginase (Fig. 4d). Interestingly, the urea cycle is also incomplete in the leech Poecilobdella granulosa57, another symbiotic heterotrophic annelid with a protein-rich diet that excretes ammonia as a waste product. In O. frankpressi, the lack of CPS1 is especially significant because this enzyme is the rate-limiting step that mediates the entry of ammonia into the urea cycle; in fact, CPS1 genetic deficiency in humans leads to episodic toxic ammonia levels in the blood (“hyperammonemia”)58. However, O. frankpressi additionally lacks urease; therefore, this enzyme is not available to convert ammonia (and carbon dioxide) into urea, thus ensuring elevated internal ammonia levels. The only enzyme present in the urea cycle of O. frankpressi is arginase, which catalyses the interconversion of arginine—which the worm likely obtains from bone-derived collagen and the endosymbionts (Fig. 4a)—into ornithine and urea. Although the urea produced by this pathway can be expected to be negligible for ammonia homeostasis, the ornithine may generate putrescine and other polyamines essential for multiple cellular functions59. Therefore, the amino acid-rich diet and lack of a urea cycle almost certainly imply chronic hyperammonemia in Osedax. This would favour amino acid biosynthesis by both Osedax and their endosymbionts; however, further functional experiments are needed to test this scenario.

Lineage-specific expansions of matrix metalloproteinases

As a core component of vertebrate bones, collagen is poised to be an essential nutrient for Osedax28,29,32 and the bone-associated microbiome48. Accordingly, transcriptomic analyses uncovered numerous metalloproteases expressed in the root tissue of O. japonicus33. Our gene family evolutionary analyses also showed that genes involved in collagen catabolism and extracellular matrix organisation are expanded in the genome of O. frankpressi (Fig. 2h; Supplementary Fig. 5f). Amongst these expanded families, genes annotated as matrix metalloproteases (MMPs) are the greatest fraction (24.3%). To investigate how MMPs diversified in O. frankpressi, we extracted the reconstructed gene families and functional annotations of symbiotic and asymbiotic annelids to identify sequences containing a metallopeptidase domain (InterPro accession IPR006026). We then reconstructed a phylogeny of the metallopeptidase genes using maximum likelihood and Bayesian approaches (Fig. 5a; Supplementary Fig. 8, 9). Our analyses recovered all previously described classes of vertebrate MMPs with high statistical support (bootstrap node support >80%) (Fig. 5a, highlighted in green) and discovered eight new highly supported invertebrate-specific classes of MMPs, labelled A to H (Fig. 5a, highlighted in blue). In addition, we identified two Osedax-specific large clades of MMPs, which we referred to as MMP-Os1 and MMP-Os2 (Fig. 5a, highlighted in red). The Osedax-specific expansions are more closely related to invertebrate than to vertebrate collagenases, supporting previous enzymatic observations that suggested generic proteolysis rather than an actual collagenase activity in Osedax worms28. The majority of MMPs belonging to MMP-Os1 (37.5%) had a metallopeptidase domain combined with a C-terminal hemopexin-like repeats domain (IPR018487) thought to facilitate binding to other components of the extracellular matrix60 (Fig. 5b; Supplementary Fig. 10). As observed with the 12 MMPs reported in O. japonicus33, all but two of the 63 MMPs found in O. frankpressi are more highly expressed in root tissue than in the rest of the body (Fig. 5c). At least 43 out of 63 (68.25%) have a signal peptide. This suggests the MMPs are excreted across the root-bone interface—similar to bone-degrading osteoclast cells of vertebrate animals61—allowing Osedax to digest bone-derived collagen extracellularly and absorb the resulting nutrients through the root epithelium for direct consumption, transport to the endosymbiont for further catabolism18,32,33, or both. Therefore, the large expansion of MMPs in an otherwise reduced genome is a unique trait of Osedax that may be related to their ability to exploit bones from diverse vertebrates, hence collagens with different amino acid sequences and protease-cleavage sites.

Fig. 5: Matrix metalloproteases experienced lineage-specific expansions in O. frankpressi.
figure 5

a Phylogenetic reconstruction of animal matrix metalloproteases (MMPs) based on the metallopeptidase domain. Tree topology is based on maximum likelihood reconstruction and node bootstrap support for each major class is colour coded (white circles show an 80–89 bootstrap support; grey circles indicate a 90–99 bootstrap values and black dots highlight fully supported nodes). Vertebrate-specific MMP classes are highlighted in green and named according to existing literature161. New monophyletic clades of invertebrate MMPs are in blue and named from A to H. Osedax frankpressi experienced two independent expansions of MMPs, shown in red and named as MMP-Os1 and MMP-Os2. b Schematic drawings of the protein domain composition of the different MMP classes recovered in (a). For each class, only the most abundant domain architecture is shown. A complete characterisation of the domain composition of MMPs is in Supplementary Fig. 10. Drawings are not to scale. c Heatmap of normalised expression levels of MMPs in the body and roots of O. frankpressi. Most MMPs show higher expression levels in the roots than in the body. Source data for (c) are provided as a Source Data file.

Divergence in innate immunity repertoire

Establishing stable and specific host-bacterial associations involves innate immunity genes, which are expanded in some Vestimentifera8,9 (Supplementary Fig. 5e) and other symbiotic oligochaetes22. To identify the immune gene repertoire in O. frankpressi, we investigated the reconstructed gene families for innate immune pattern recognition receptors corresponding to six major classes, namely lectins, peptidoglycan recognition proteins, Toll-like receptors, scavenger receptors, bactericidal permeability-increasing proteins, and NOD-like receptors62. Compared to asymbiotic annelids (i.e., Owenia fusiformis and C. teleta) and Vestimentifera, O. frankpressi has fewer immunity genes in all considered classes (Fig. 6; Supplementary Table 7; Supplementary Data 1117). This includes a smaller repertoire of Toll-like receptors, which are expanded in some species of Vestimentifera8,9, and the loss of galectin and a NOD-like receptor, which is a family of cytosolic immune receptors that recognises and triggers inflammatory responses to bacterial pathogens63 that are also largely expanded in Vestimentifera9 (Supplementary Table 7; Supplementary Data 1215). Notably, there is no clear association between the expression levels of the different classes of pattern recognition receptors and the body regions and tissues of Siboglinidae. Yet, a C-type lectin is highly expressed in the root tissue of O. frankpressi (Fig. 6). Our findings indicate that O. frankpressi and Vestimentifera have different innate immune complements that are simplified in the former and generally expanded in the latter. Further research in Frenulata and Sclerolinum will inform whether this divergence in the repertoire of innate immune genes may underpin the evolution of a novel symbiotic association with Oceanospirillales bacteria in Osedax worms.

Fig. 6: O. frankpressi has a reduced innate immune gene repertoire compared to Vestimentifera.
figure 6

Heatmaps of tissue-specific normalised gene expression of innate immune genes in four species of Siboglinidae, including O. frankpressi (top) and the Vestimentifera Oasisia alvinae, R. pachyptila and P. echinospica. While Vestimentifera have relatively similar repertoires of innate immune genes, O. frankpressi has a much-reduced complement (Supplementary Table 7). Notably, innate immune genes do not show a clear tissue-specific expression within or among species of Siboglinidae. The immune repertoire and gene expression values for P. echinospica and L. luymesi is based on previously published genome resources8, 9. Source data for O. frankpressi, Oasisia alvinae and R. pachyptila are provided as a Source Data file.

A conserved developmental toolkit in Siboglinidae

In addition to lacking a gut, Vestimentifera and Osedax also lack eyes and any other sensory structure in their most anterior region, the prostomium64. Yet unlike other annelids with unusual body plans, such as the leech H. robusta38, the genomes of Vestimentifera contain a complete developmental toolkit9,10. To investigate genes involved in body patterning and organogenesis in the reduced gene set of O. frankpressi, we first focused on the repertoire of G protein-coupled receptors (GPCRs; Supplementary Data 18), a large family of evolutionarily related membrane receptors involved in an array of developmental, sensory, and hormonal processes65,66. All siboglinids show a conserved repertoire of GPCRs of class B (secretins), C (metabotropic glutamate receptors) and F (frizzled and smoothened receptors) (Supplementary Fig. 11b–d). However, Siboglinidae has a more divergent complement of rhodopsin-like receptors (class A), with five expanded clusters, one specific to O. frankpressi (Supplementary Fig. 11a, highlighted in pink). Notably, O. frankpressi and Vestimentifera have lost four GPCR families, including opsins (Supplementary Fig. 11a, highlighted in grey), suggesting an ancestral loss of light perception to these groups in parallel to the colonisation of light-deprived deep marine environments.

The bulk of the body of Siboglinidae has only two segments and the posterior end (i.e., the opisthosoma), which is often multisegmented, is lacking in Osedax31,64. Nevertheless, the complement of Hox genes—a conserved family of transcription factors that define a molecular code throughout the many trunk segments in Annelida37,67—is largely conserved in Vestimentifera, only missing the gene Antennapedia (Antp)9,10. Osedax frankpressi has a similar Hox gene repertoire, and thus the loss of Antp might have occurred in the last common ancestor of Siboglinidae (Supplementary Figs. 12a, 13a). Indeed, the number and complement of transcription factors involved in animal development are comparable in O. frankpressi, Vestimentifera and asymbiotic annelids, except for Basic Leucine Zipper Domain containing proteins (bZIP; PF00170) and zinc finger transcription factors (C2H2-Zn; PF00096), which are reduced (Supplementary Fig. 13b; Supplementary Data 19), as well as certain specific classes, such as the ParaHox genes (Supplementary Fig. 12a). Similarly, O. frankpressi retains all major developmental signalling pathways, yet it has a lower number of Notch containing proteins (Supplementary Fig. 13c, d) and a simplified repertoire of signalling ligands (Supplementary Figs. 12b, 14, 15), as also observed in the miniaturised annelid D. gyrociliatus36. Therefore, O. frankpressi and Vestimentifera show a similar and generally conserved developmental toolkit, suggesting that changes in gene regulation rather than deviations in the gene complement underpin the development of the divergent adult morphology of Siboglinidae after symbiont acquisition.

Species-specific repertoires of DNA damage repair mechanisms

Changes in the machinery that repair DNA damage can cause biases in the GC composition of the genome68,69, and such changes have been associated with genome compaction and gene loss in animals70. Osedax frankpressi has an AT-rich genome (29.08% GC content versus ~41% observed in Vestimentifera; Supplementary Fig. 2k, m) and unlike other annelids, it has three major DNA repair pathways that are largely incomplete, namely the base excision repair, the non-homologous end joining, and the Fanconi anaemia DNA repair pathway (Supplementary Figs. 12c, 16). The base excision repair pathway corrects DNA damage from base lesions caused by deamination, oxidation and methylation, and is thought to increase GC to AT base transitions when impaired71. The lack of the non-homologous end joining pathway—the most common mechanism to repair double-strand DNA breaks72—triggers the error-prone microhomology-mediated end joining pathway, which is intact in O. frankpressi and all other annelids but causes microdeletions73 (Supplementary Fig. 16f; Supplementary Data 20). Therefore, the loss of genes involved in the repair of double-strand DNA breaks and chemical base modifications might underpin the reduction in genome size and GC content observed in O. frankpressi in comparison with Vestimentifera, thus differing from other annelids with reduced genomes, such as D. gyrociliatus, whose genome eroded without changes in DNA repair pathways36.


Our data reveal additional evidence on the genetic interactions and co-dependencies of animal hosts and bacterial symbionts that have enabled distinct symbiotic lifestyles, including the exploitation of sunken vertebrate bones as a food source (Fig. 7a). Our analyses of the genomes of Oasisia alvinae and R. pachyptila confirm what was previously reported for other species of Vestimentifera8,9 and R. pachyptila itself10 and support that broadly similar genomic adaptations underpin the different symbioses of Vestimentifera, even between species occupying distinct environments, such as hydrothermal vents and methane seeps. However, compared to Vestimentifera, O. frankpressi shows a fast evolving36, divergent gene repertoire, with gene losses and expansions in key functional groups that support metabolic adaptations to its symbiotic lifestyle (Figs. 2, 3b, 4a; Supplementary Fig. 5d, f). As observed in the marine microbial assemblages on bone surfaces48, the expansion of secreted matrix metalloproteases33 (Fig. 5a) combined with the active secretion of acid in the root tissue32 are the most probable mechanisms of bone digestion by the host (Fig. 7a). The Osedax-microbe association, however, entails further molecular and metabolic interactions to overcome a nutritionally unbalanced diet that is deficient in carbohydrates but enriched in (hydroxy)proline- and glycine-rich proteins and, in some cases, lipids49,50. Most notably, our findings suggest that the Oceanospirillales endosymbionts might be able to provide Osedax with glucose through the glyoxylate cycle (Fig. 3b) and that Osedax and the endosymbionts cooperate to maintain a physiological status of hyperammonemia (Fig. 4d). The former allows the catabolism of fatty acids to produce carbohydrates, which the host could take up by digesting the endosymbionts and store as glycogen (e.g., as seen in Osedax’s oocytes74), whereas the latter could stimulate the biosynthesis of amino acids, ultimately counterbalancing the lack of carbohydrates and skewed amino acid composition in bone. Notably, the use and occurrence of the glyoxylate cycle in animals is controversial75,76 and only reported in a handful of taxa77,78, likely as a consequence of horizontal gene transfer79 and often concerning stress and a metabolic diapause, such as in the Dauer larva of nematodes51, hibernating mammals80 and bleached coral81. Indeed, Osedax, like Vestimentiferan hosts and their endosymbionts, lacks isocitrate lyase, but this enzyme is present in Osedax’s endosymbiont18 (Fig. 3b; Supplementary Data 3, 7). Therefore, the metabolic diversity of the Oceanospirillales endosymbiont may be critical to maximising the use of the imbalanced resources derived from the bones and ultimately acts as a selective pressure to acquire and maintain this microbe as the primary symbiont.

Fig. 7: The genomic basis and evolution of different nutritional symbioses in Siboglinidae.
figure 7

a Schematic drawing of the metabolic interaction for bone digestion between Osedax and its endosymbiont (red kidney-shaped ovals), which are harboured in the trophosome inside bacteriocytes. The root epidermis secretes acid to dissolve the inorganic component of the bone (via carbonic anhydrase, CA, and V-type H+-ATPase, VHA) and matrix metalloproteases (MMPs) that break collagen, one of the most abundant organic components of the bone, into amino acids and oligopeptides, which are rich in proline and glycine. These amino acids and the lipidic content of the bone are absorbed by the epidermis and used either directly by Osedax or transported to bacteriocytes. The host and endosymbiont cooperate to generate carbohydrates (in low amounts in bone) from the oxidation of fatty acids (abundant in bone and roots) through the glyoxylate cycle, most likely inside the bacteriocytes. Ultimately, these interactions transform the originally unbalanced diet into complex and diverse macronutrients, which are then taken directly or after the digestion of the bacteria by the host. b Osedax and Vestimentifera broadly show different genomic traits. Osedax has a small, AT-rich genome, with many gene losses and a reduced immune repertoire. Vestimentifera tends to show larger genomes, with a more extensive gene complement and richer innate immunity (although there is inter-specific variation for these traits, highlighted with a lighter green arrow). We hypothesise that the different nutritional relationships between the hosts and symbionts in these two groups might explain, at least partially, these genomic differences. Osedax and its endosymbiont co-depend on and compete to exploit the finite, nutritionally unbalanced diet obtained from bones, which might have favoured the evolution of an energetically “cheaper” genome in Osedax. In Vestimentifera, however, the endosymbiont acts as a primary producer, which might be able to sustain larger host genomes. Drawings are not to scale.

Symbiotic interactions can impose selective pressures that direct genome evolution—most notably in symbionts82 but also occasionally in hosts83—triggering changes in genome size (e.g., genome erosion)84, gene content85 and even DNA base composition in favour of AT-rich genomes86. Most of these changes, however, are known for strictly vertically transmitted obligate endsymbionts of insects. Our study shows that Vestimentifera and Osedax, two annelid lineages within Siboglinidae that establish environmentally acquired symbioses, show differences in genome structure and composition (Fig. 2a–c; Supplementary Fig. 2m). While Vestimentifera tends to have larger genome sizes, similar GC content to asymbiotic annelids37,38, and larger gene repertoires, O. frankpressi has a small, AT-rich genome, with a reduced gene content (Fig. 7b). In addition, these Siboglinidae crucially differ in their nutritional symbioses—chemoautotrophic in Vestimentifera and heterotrophic in Osedax—which enable them as adults to thrive in different ecological niches with different nutritional pressures. In hydrothermal vents and methane seeps, Vestimentifera relies on virtually unlimited inorganic nutrients that are exploited by the endosymbionts, which in their role as primary producers sustain long-lasting collaborative co-dependencies with their hosts3,5. Decaying bones are, however, nutritionally finite, and thus Osedax and their endosymbionts may establish a competitive co-dependency to exploit those nutritionally unbalanced resources (Fig. 7b). Moreover, the potential use of the glyoxylate cycle for energy production would be less energetically efficient than the sole catabolism of fatty acids87. Therefore, we hypothesise that the interaction between Osedax and its endosymbiont might, in turn, favour the genomic streamlining of the annelid host (Fig. 7b) so that it becomes metabolically and energetically “cheaper” and can sustain larger endosymbiotic populations for longer periods. Our findings thus suggest that incipient genome erosion can occur in hosts with horizontally acquired symbionts and that adaptive genome evolution may differ based on the type of nutritional interactions between the host and symbiont. In the future, dissecting the metabolic co-dependencies between Siboglinidae and their endosymbionts, including the Frenulata and Sclerolinum—the other two major lineages within Sibogliniade—will help to disentangle the role of neutral and adaptive selective pressures in the evolution of these fascinating, but still poorly understood, animal symbioses.


Specimen collections, gDNA extraction and sequencing

Live adult specimens of O. frankpressi, Oasisia alvinae and R. pachyptila were obtained with deep-sea specialised robots off the coasts of California and Mexico (Supplementary Fig. 1c, d). Mexican samples were collected under CONAPESCA permit PPFE/DGOPA-200/18. Ultra-high molecular weight genomic DNA (gDNA) was extracted following the Bionano Genomics IrysPrep agar-based, animal tissue protocol (Catalogue # 80002) from an entire O. frankpressi adult female, a piece of the trunk (including trophosome) of Oasisia alvinae, and a piece of the vestimentum of R. pachyptila. Long-read PacBio sequencing and short-read Illumina sequencing was performed at the Genome Centre of the University of California Berkeley in a PacBio Sequel II and Illumina Novaseq platforms (Supplementary Table 1).

Transcriptome sequencing

Total RNA from dissected tissues and body parts of Oasisia alvinae (crown, opisthosome and trophosome), and R. pachyptila (crown and trunk wall) was extracted with an NEB totalRNA Monarch kit and used for standard strand-specific RNA Illumina library prep. Libraries were sequenced to a depth of 40–50 million paired reads of 150 bases length in a NovaSeq platform (Supplementary Table 1). Publicly available datasets for O. frankpressi (NCBI short read archive accession numbers SRR2017399 and SRR2017400) were used in this study (Supplementary Table 1).

Host genome assembly and quality check

PacBio reads were used to generate an initial genome assembly with Canu v.1.888 with options ‘batOptions = “-dg 3 -db 3 -dr 1 -ca 500 -cp 50’. Two rounds of polishing using PacBio reads were performed using Pbmm2 v.1.1.0 ( and Arrow (pbgcpp v.1.9.0)89. Short genomic Illumina reads were quality filtered with FastQC v.0.11.8 and Cutadapt v.2.590, mapped to the polished assembly with BWA v.0.7.1791 and used for final polishing with Pilon v.1.2392. The polished versions of the genomes of O. frankpressi, Oasisia alvinae and R. pachyptila were used as input to BlobTools v.2.193 to identify and remove contigs with high similarity to bacteria. After decontamination, the haplotypes were purged with Purge_Dups v.1.0.194. Quality check was performed with BUSCO v.3.0.295, to estimate gene completeness of the assembly (Supplementary Table 3), QUAST v.5.0.296, and KAT v.2.4.297 to assess haplotype removal (Supplementary Fig. 2b–d) and potential bacterial remnants.

Genome size estimations

Short Illumina reads were mapped to the reference host genome assembly with BWA v.0.7.17 and KAT v.2.4.297 to count and generate a histogram of canonical 21-mers. GenomeScope298 was used to estimate the genome size and heterozygosity (Supplementary Fig. 2e–g).

Symbiont genome assembly and annotation

For O. frankpressi and Oasisia alvinae, we used Kraken2 v.2.1.099 and Krakentools v.0.199 to isolate long PacBio reads of bacterial origin. After error correction with Canu v.1.888, these PacBio reads were assembled using Metaflye v.2.9100 followed by ten polishing iterations with options “–pacbio-corr –meta –keep-haplotypes –iterations 10” and final polishing with NextPolish v.1.4.0101. The resulting assemblies were manually inspected using Bandage v.0.9.0102, binned with MaxBin2 v.2.2.7103 and quality checked with CheckM v.1.0.8104 and MetaQuast v.5.2.0105. Gene annotation was performed with Prokka v.1.14.5106 with the “—compliant” option and proteins involved in secretion systems were identified by scanning for unordered replicons using the curated HMM profiles of TXSscan in MacSyFinder v.2107. The bacterial genomes were checked for secreted proteins with eukaryotic-like domains using EffectiveELD through EffectiveDB, on default settings108. All coding sequences of the main endosymbiont ribotype for O. frankpressi, Vestimentifera and Frenulata were assigned KO numbers using BlastKOALA v.2.2109, which were used as input for KEGG Mapper v.5110 to analyse the metabolic capabilities of each symbiont. The NCBI COG database111 was used to tag functional categories to the annotated genes. Enrichment analyses of functional categories and Gene Ontology terms were performed with GSEA v.4.2.3112 and OrthoVenn2 v.2113. To compute the p-values for enriched Gene Ontology terms in a protein cluster (Supplementary Data 6, 7), a hypergeometric distribution was used to identify significantly enriched terms within each cluster of orthologous/paralogous genes. GTDB-Tk v.1.6.0114 was used for whole genome phylogenetic placement and identification of neighbouring available genomes isolated from free-living deep-sea bacteria. Circos v.0.69-9115 was used for genome assembly visualisation.

Annotation of repeats in host genomes

RepeatModeler v.2.0.1116 and Repbase117 were used to build a de novo library of repeats for the host genome of O. frankpressi, Oasisia alvinae and R. pachyptila. The predicted genes of Owenia fusiformis37 and DIAMOND v.0.8.22118 were used to filter out bona fide genes in the predicted repeats with an e-value threshold of 1e-10. Subsequently, RepeatMasker v.4.1.0119 (Supplementary Tables 46) and LTR-finder v.1.07120 were used to identify and annotate repeats, and RepeatCraft121 to generate a consensus annotation that was used to soft-mask the genome assemblies of the three annelid species. To explore the transposable element landscape, we used the online tool TEclass122 to annotate the TEs identified by RepeatModeler and the scripts “” and a custom-modified version of “”, both from RepeatMasker v.4.1.0, to estimate Kimura substitution levels, which were plotted using ggplot2 v.3.3.0123. Previously published TE landscapes were included for comparisons37.

Functional annotation of host genomes

Individual RNA-seq Illumina libraries (Supplementary Table 1) were de novo assembled with Trinity v.2.9.1124 after quality trimming with Trimmomatic v.0.35125. GMAP v.2017.09.30126 and STAR v.2.7.5a127 were used to map transcripts and quality-filtered Illumina reads to the soft-masked genome assemblies of the corresponding species. For R. pachyptila, publicly available datasets (SRA accession numbers SRR8949056 to SRR8949077) were also mapped to the soft-masked genome assembly. In addition, gene transfer format (GTF) files from the mapped reads and curated intron junctions were inferred with StringTie v.2.1.2128 and Portcullis v.1.2.2129. All RNA-seq-based gene evidence was merged with Mikado v.2.Orc2130, which produced a curated transcriptome-based genome annotation. Full-length Mikado transcripts were used to train Augustus v.3.3.3131, which was then used to generate ab initio gene predictions that incorporate the intron hints of Portcullis and the exon hints of Mikado. In addition, Exonerate v.2.4.0132 was used to produce spliced alignments of the curated proteomes of Owenia fusiformis, C. teleta and L. luymesi that were used as further exon hints for Augustus. Finally, the Mikado RNA-seq-based gene evidence and the ab initio predicted Augustus gene models were merged with PASA v.2.4.1133. A final, curated gene set was obtained after removing spurious gene models and genes with high similarity to transposable elements. Gene completeness and annotation quality were assessed with BUSCO v.3.0.295. Trinotate v.3.2.1134, PANTHER v.1.0.1047 and the online tool KAAS135 were used to functionally annotate the curated gene sets.

Comparison of R. pachyptila assemblies

Overall genomic stats were obtained with BUSCO v.3.0.295, QUAST v.5.0.296 and AGAT v.0.5.0136. We used minimap2 v.2.17 to align our R. pachyptila assembly with the assembly previously reported10 and the R package pafr to generate a dot-plot representation of the sequence similarity between the two versions. In addition, we reassembled all transcriptomic evidence published elsewhere12 using Trinity v.2.9.1124 and cd-hit v.4.8.1137. To identify one-to-one orthologs between genomic and transcriptomic resources, we used a reciprocal best BLAST hit approach with BLAST v.2.12.0+ 138. Finally, we used PFAMscan v.1.6139 to identify and quantify distinct Pfam domains in the different assemblies.

Gene family evolutionary analyses

The non-redundant proteomes of O. frankpressi, Oasisia alvinae and R. pachyptila together with 25 high-quality genomes spanning major groups of the animal tree (Supplementary Data 2) were used to construct orthogroups with OrthoFinder v.2.5.2140 using DIAMOND v.2.0.9118 with “–ultra-sensitive” option. The OrthoFinder output and a published Python script36 were used to infer gene family evolutionary dynamics at each node and tip of the tree. Gene Ontology term enrichment analyses for expanded and lost gene families were performed with the R package “TopGO” v.2.42.0. The number of orthologs per gene family and species as generated by OrthoFinder was used to perform a Principal Component Analysis with R built-in functions.

Reconstruction of host metabolic pathways and developmental gene sets

PANTHER and Pfam annotations obtained through PANTHER v.1.0.1047 and Trinotate v.3.2.1134, respectively, were used to assess for the presence of each enzyme involved in the synthesis of amino acids, vitamin Bs, nitrogen metabolism, glycine degradation, matrix metalloproteases, transcription factors and DNA repair pathways in an array of annelid species. A combination of BlastKOALA109 and KofamKOALA141 was used to annotate the host and endosymbiont genomes for the analysis of the lipid and carbohydrate metabolism. Information about each step in a pathway was collected from MetaCyc142, KEGG143 and PANTHER47 databases. To analyse the tissue-specific expression of candidate genes in O. frankpressi, Oasisia alvinae and R. pachyptila, quality-filtered short Illumina reads were pseudo-mapped to the filtered gene models of each species with Kallisto v.0.46.2144 to quantify transcript abundances as Transcripts per Kilobase Million (TPM) values. The R libraries ggplot2 v.3.3.0123 and pheatmap v.1.0.12 ( were used to plot expression and abundance heatmaps.

Reconstruction of innate immune repertoires

The OrthoFinder output was used to identify gene families of innate immune pattern recognition receptors of O. frankpressi, Vestimetifera and two asymbiotic annelids, Owenia fusiformis and C. teleta, with the published pattern recognition receptors of Vestimetifera9 as baits (Supplementary Data 1117). PANTHER and Pfam annotations (see above) of the target proteins were further used to remove sequences that were too short or lacked target domains. TPM expression values (see above) and TBtools v.1.042145 were used to plot gene expression heatmaps.

Reconstruction of the G protein-coupled receptor (GPCR) repertoire

Transcriptomes of the focal species were downloaded and processed as described elsewhere146. Multiple sequence alignments of rhodopsin type GPCRs (PF00001), secretin type GPCRs (PF00002), glutamate type GPCRs (PF00003) and frizzled type GPCRs (PF01534) were downloaded from the Pfam webpage ( and used to create HMM profiles using hmmer-3.1b2147. HMMer search was performed with an e-value cut-off of 1e-10. The online version of CLANS ( was used for the initial BLAST comparison for the cluster analysis and edges below 1e-10 for secretin, glutamate and frizzled type GPCRs and 1e-20 for rhodopsin type GPCRs were removed. The java offline version of CLANS148 was then used for the cluster analysis. The p-value for clustering was set to 1e-25. Singletons and group-specific sequence clusters with less than five sequences and no annotation (using Linkage clustering for identification) were removed. The highly vertebrate-specific expanded olfactory GPCR type-A receptors were also deleted as these showed no connections and strongly repulsed all other sequences. Gene clusters were annotated according to the presence of characterized sequences of Drosophila melanogaster, Homo sapiens, Danio rerio and Platynereis dumerilii.

Orthology assignments

MAFFT149 with default options was used to align candidate sequences to a curated set of proteins that we obtained either from previous studies36,150 or manually from UniProt151. Conserved protein domains were retained by trimming by hand the alignment in Jalview152 and the resulting sequences were re-aligned in MAFFT with the “L-INS-I” algorithm149. After a final trim to further remove spurious regions with trimAI v.1.4.rev15153, FastTree v.2.1.10154 with default options and IQ-Tree v.2.2.0-beta155 (for matrix metalloproteases) using the options “-m MFP -B 1000”, were used to infer orthology relationships. In addition, for the matrix metalloproteases, the posterior probabilities were obtained from Bayesian reconstructions in MrBayes v.3.2.7a156, which were performed using as a prior the LG matrix157 with a gamma model158 with four categories to describe sites’ evolution rate. Four runs with eight chains were run for 20,000,000 generations. FigTree v.1.4.4 ( and Adobe Illustrator were used to edit the final trees. CD-Search159 with default options and the Conserved Domain Database (CDD)160 were used to annotate protein domains in the predicted matrix metalloproteases.

Reporting summary

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