Genome description

Genome structure and annotation.

We present a highly contiguous, complete chromosome-scale diploid assembly of a single Nanomia septata colony collected near Monterey, California (Fig 1B). Each of the two haplotypes has 8 chromosomes (1794 and 1796 Mbp, respectively) and a BUSCO score of ~92% (Table 1 and Fig 1). A karyotype confirmed there are 2n = 16 chromosomes (Fig 1C). We designate each chromosome with a number 1–8. Haplotype 1 has 96.7% of the bases in 8 chromosomes, and haplotype 2 has 96.9% of bases in 8 chromosomes. The haplotypes are very similar to each other, with haplotype 2 having a slightly higher completeness and assembly quality (Table 1). Haplotype 2 also had fewer gaps than the alternate, and contained more contigs within the chromosome-scale scaffolds, resulting in fewer orphan scaffolds. Thus, we chose to use haplotype 2 for subsequent analyses, including annotation (Table 2). We created multiple genome annotations, including one using the Nanomia septata Iso-seq data. We found that the annotation made with the Iso-seq called 198k genes, most of which we found to not be real genes. Thus, we chose to use the more conservative genome annotation that we present here, with the caveat that we may be missing some genes. In this annotation the total length of introns in the final assembly is 71.3Mb and total length of exons is 37.4Mb (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Nanomia septata genome assembly statistics.

https://doi.org/10.1371/journal.pone.0351247.t001

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Nanomia septata annotation of haplotype 2 (primary).

https://doi.org/10.1371/journal.pone.0351247.t002

We manually curated the Homeobox and Wnt gene annotation, given their importance in cnidarian development (body and axis patterning) [23–26] and potential relevance to the unique colony-level growth of siphonophores. The original annotation had 38 of these genes and we found an additional 22 more using the Iso-sequencing of Nanomia septata. In total, Nanomia septata has 60 homeobox genes, split across 8 classes of genes (S1 Table). In comparison, other hydrozoans have on the order of 60−80 homeobox genes (S1 Table). More specifically, Nanomia septata has 17 Antp genes, including 4 Hox-like (HOXL) and 13 NK-like (NKL) members. Among HoxL genes, we identified one Hox gene (Cnox-1, located on chromosome 4), an unclassified Hox-like gene, and the two paraHox genes Gsx (Cnox-2) and Cdx (Cnox-4). We found 13 Nk-like genes, including Nk-1, Nk-2 (three copies), and Nk-6 genes (S1 Fig). In addition, the genome has nine Wnt genes identified as Wnt 1, 2, 3, 5 (two copies), 6, 7, 8 and A (S2 Fig). These have all been added so that in total, the annotation of haplotype 2 has 26,323 protein-coding genes (the first version had 26,301 genes).

Differential gene expression across zooids

A previous bulk RNA-seq study assessed the differential expression of genes across different types of developing and mature zooids in 7 species using short read transcriptomes assembled with TRINITY as references [8]. We re-mapped the Nanomia RNA-seq reads used in that study to our new genome and annotation (six tissue/zooid types, some of which were split between developing and mature). These previous data are from specimens of the species now referred to as Nanomia septata, the same species that we sequence the genome of here. That manuscript was published before Nanomia septata was split from Nanomia bijuga, so they were referred to as Nanomia bijuga in this previous publication. Our remapping rates range from 32–51% for each RNA-seq sample, with a mean of 41%. The results from the original paper [8] are robust to the improved gene references. For example, female gonodendron and nectophores group together (S3 Fig). We included the male gonodendron dataset that was excluded in the original paper and found that it also closely groups with the female gonodendron and nectophores (S3 Fig), consistent with all these zooids being highly modified medusae that are retained within the colony. The pneumatophore, a specialized structure for buoyancy that is not a zooid, clusters closely with both gonodendra and nectophores (S3 Fig). Lastly, gastrozooids and palpons, which are both types of specialized polyps, have more similar expression to each other than to other zooids.

These re-mapped data also provide an updated list of genes with zooid-specific expression. We find that there is a total of 3,816 genes that have significant differential expression among zooid tissues and stages (S2 Table). Of these, 3,782 genes have expression specific to a single zooid type, mostly with increased expression in either mature female gonodendra or developing nectophores (S2 Table). Similar to the original study [8], genes with elevated expression in gastrozooids include many with functions related to digestion. With our new results, we also identify GOs (gene ontologies) for toxins that are abundant in female gonodendra (GO:0019835, GO:0030286, GO:0032991), nectophores (GO:0030286), palpons (GO:0004866) and the pneumatophore (GO:0004866) (S2 Table).

The remaining 34 out of 3,816 total genes have differential expression that is shared across multiple zooid types (S3 Table). Unsurprisingly, developing nectophores have shared differential expression with the mature pneumatophore and the mature female gonodendra, all of which cluster together in the PCA (S3 Table and S3 Fig). These shared differentially expressed genes have GOs for developmental processes such as morphogenesis of a branching structure and morphogenesis of epithelium. Specifically, TAGLN3, an actin gene, is shared and upregulated between the female gonodendra and pneumatophore, and UNC45A, a muscle development gene, is shared and upregulated between the female gonodendra and nectophore (S3 Table). Fox/forkhead genes have elevated expression in female gonodendra and developing nectophores, consistent with their role in medusa development observed in other cnidarians [27]. This updated list of differentially expressed genes in zooids will be useful for pursuing future in-depth zooid studies, including spatial expression experiments and single cell sequencing (S2 Table).

Additionally, we assessed the relationship between physical proximity of genes and the covariance of their expression across zooid types. The goal was to find genes that are close together in the genome and have similar expression across zooids. To do this we derived a new metric we call the expression proximity product (see methods). For each pair of genes, the expression proximity product is near zero if the genes are far apart or have low correlation of expression across zooids. They are near one if they are physically close and have high positive correlation of expression. They are near negative one if they are physically close and have strong negative correlation.

We found multiple regions in the genome with a high expression proximity product, i.e., regions that are enriched for genes with correlated expression across zooid types (S4 Fig 4). In all cases, these genes had elevated expression specific to male gonophores. These genes included spermatogenic genes (S4 Fig 4).

Repeat analysis

63.7% of the genome is composed of repetitive elements (Fig 2; S4 Table and S5 Table). This is consistent with previous estimates based on k-mers [21]. Notably, the proportion of DNA elements was high, with TcMar-Tc1, Kolobok-T2, Maverick, PiggyBac, and hAT-Charlie being abundant (Fig 2B). In addition, many other TEs are also abundant in the genome (Fig 2B,C; S5C Fig and S6 Fig; S4 Table). Furthermore, when examining the substitution levels of subfamilies of each transposon, we found a large number of subfamilies with low divergences (Fig 2A). This suggests that the Nanomia septata genome contains many young/active transposons. Additionally, other species belonging to Nanomia have been reported to exhibit a wide range of genome sizes [21].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Repeat landscape of the Nanomia septata genome.

A. Genome coverages and sequence divergences of TE families. The x-axis depicts the degree of Kimura substitution within repeat elements compared to consensus sequences, serving as an indicator of their relative age. The y-axis shows the genome coverage of each repeat family throughout the genome. Repeats with ancient activity are located towards the graph’s right, while those with more recent activity are found on the left side. B. Pie chart representing proportional representation of each TE family. Note that 36.3% of the genome consists of non-repetitive regions, while the remaining 63.7% are repetitive elements. C. Absolute genome coverage of top 50 TEs. Arrows indicate the eukaryotic Core-TEs. The eukaryotic Core-TEs collectively occupy 28.1% (505 Mb) of the Nanomia septata genome.

https://doi.org/10.1371/journal.pone.0351247.g002

Previously, Kon-Nanjo et al. [13] identified 14 highly conserved active TE families (A-TEs), mainly composed of DNA elements (hAT-Ac, hAT-hATm, hAT-Tip100, hAT-Charlie, TcMar-Tc1, TcMar-Fot1, CMC-Transib, CMC-EnSpm, Sola-2, Maverick, and MULE-MuDR), that are found across a wide range of eukaryotic genomes. In the Nanomia septata genome, these A-TEs collectively occupy 22.3% (401 Mb) of the genome (Fig 2C; S5 Fig 5). We also compared nine species of metazoans including two bilaterians (Pecten maximus and Branchiostoma floridae), seven cnidarians (Nematostella vectensis, Acropora millepora, Rhopilema esculentum, Nanomia septata, Hydractinia symbiolongicarpus, Hydra viridissima, and Hydra vulgaris). Among these species, Nanomia septata exhibited greater genome coverage by a wider variety of TE families, indicating a multi-TE family expansion (Cluster #1 in S5C Fig and S6 Fig). Additionally, 13.7% of repeats across the entire genome could not be classified, suggesting that they may be specific to Nanomia septata or siphonophores more broadly.

Comparative cnidarian genomics

Many features of synteny are conserved across more than 500M years of cnidarian evolution, but are modified in the reduced karyotypes of Siphonophora (Fig 3A). A chromosome scale assembly of the siphonophore Physalia physalis was recently produced for a population analysis of Physalia [18] and is considered here. As there is detectable homology between n = 10 chromosomes in Physalia physalis (3 GB) and n = 8 chromosomes in Nanomia septata (1.7 GB), siphonophore genomes have shared ancestral chromosomal fusions. Differences between Physalia physalis and Nanomia septata reveal evolutionary changes within Siphonophora. While overall the chromosomal homologies are conserved there were still a few detectable fusions or translocations. These suggest that the reduced karyotype in Nanomia septata (n = 8) and Physalia physalis (n = 10) could have arisen by some fusions of the homologs of Physalia physalis chromosomes 2 and 3 to form LG 1, chromosomes 8 and 10 to form LG 3, and fusions of chromosome 6 and 4 to form LG 6 in Nanomia septata (Fig 3D). However, more translocations and mixing events could have occurred that are difficult to pin down with only two siphonophore genomes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Syntenic transitions towards Nanomia septata karyotype.

A. Ribbon plot of Nanomia septata and other cnidarians with chromosomes below. Chromosome labels with their accession numbers can be found in S7 Table. B-D. Oxford dot plots (ODP) of pairwise hydrozoan comparisons. B. ODP of Hydractinia symbiolongicarpus chromosomes vs Physalia physalis chromosomes. C. ODP of Hydractinia symbiolongicarpus chromosomes vs. Nanomia sepatata chromosomes. D. ODP of Physalia physalis chromosomes vs Nanomia septata chromosomes. E. Summary panel of siphonophores, Nanomia septata and Physalia physalis, and closely related hydrozoans that show where rearrangements, fusion and consolidation happened along the branches. Each protein ortholog is colored by the ancestral linkage group it was described in (BCnS [20]). Other protein orthologs are in black. Significance of linkage is denoted by a p-value (if significant opaque, otherwise transparent). P-value cutoffs are below panel E.

https://doi.org/10.1371/journal.pone.0351247.g003

To investigate the origins of the siphonophore karyotype, we compared these genomes to their outgroups. Because Hydra vulgaris and Hydractinia symbiolongicarpus form a paraphyletic group with respect to siphonophores [19], their many shared features provide extensive information about the state at the start of the siphonophore stem branch (Fig 3A). These shared ancestral features include 15 chromosomes with highly conserved synteny (Fig 3A), and a typically small cnidarian genome size. Given the similarities between these two species, we make comparisons to Hydractinia symbiolongicarpus as a proxy for the ancestral state.

We compared the Hydractinia symbiolongicarpus and two siphonophore genomes, Physalia physalis and Nanomia septata, using oxford dot plots (Fig 3B-D). Comparison among these three species indicate that there is not a 1:1 correspondence between Hydractina symbiolongicarpus chromosomes (and other cnidarians) and any chromosomes in either siphonophore species. However, the Physalia physalis genome is more similar than Nanomia septata to Hydractinia symbiolongicarpus, showing many fusions of complete chromosomes and clear syntenic boundaries between them on Physalia physalis chromosomes. As such, there are discrete areas of homology between pieces of Hydractinia symbiolongicarpus chromosomes and pieces of Physalia physalis chromosomes, such as chromosome 15 in Hydractinia symbiolongicarpus which corresponds to chromosome 10 in Physalia physalis (Fig 3B). Only some inter-chromosomal translocations (e.g., Physalia physalis chromosome 6 and chromosome 9) were observed. This indicates a more ancestral state and a less mixed state of Physalia physalis. Furthermore, this shows that the genome of Nanomia septata has undergone more extensive reorganization losing syntenic identities through extensive mixing along the newly fused chromosomes (Fig 3C and 3D). The chromosomal mode of evolution in siphonophores thus corresponds to the “consolidation” process describing synteny loss primarily through extensive chromosomal fusion-with-mixing and karyotype reduction [28]. Supporting this, there is stronger statistical support for chromosomes between Hydractina symbiolongicarpus and Physalia physalis (S6 Table: Fig 3_Chr_Values_PanelB), while support values for chromosomes shared between Hydractina symbiolongicarpus and Nanomia septata are marginal (S6 Table: Fig 3_Chr_Values_PanelC). To understand the degree of scrambling of the previously described ancestral linkage groups (ALGs; bilaterian-cnidarian-sponge and unicellular-metazoan linkage groups), we visualize these ALGs in Hydractinia symbiolongicarpus, Physalia physalis, Nanomia septata (S7 Fig) which provides additional support for more extensive genome reorganization in Nanomia septata. All the supporting values for the pairwise analyses above are contained in S6 Table and chromosome labels in S7 Table.

Together, these findings suggest the role of chromosomal fusions, followed by mixing, as one of the major drivers in synteny loss in siphonophores (“consolidation” [28]) (Fig 3). Whether and how TE activity has contributed to within-chromosomal differences and mixing in Physalia compared to Nanomia, however, requires more chromosomal-scale species sampling to discern.

Genomic diversity in Nanomia

To assess the broader context of diversity and population structure in Nanomia, we performed whole-genome short-read sequencing on 24 specimens collected over 19 years from the Atlantic Ocean (Virginia and Rhode Island), Pacific Ocean (Washington, California, Gulf of California, and Hawai’i), and Mediterranean Sea (Villefranche-sur-mer). Phylogenetic analyses of CO1 (S8 Fig) indicate that these specimens belong to Nanomia septata (7 samples from California, 3 from Washington), Nanomia cara (1 sample from Rhode Island), Nanomia bijuga (3 from Rhode Island, 2 from Hawai’i, 1 from Villefranche-sur-mer, 3 from Gulf of California), and one undescribed species from Hawai’i consistent with Nanomia sp. 1 from Hosia et al. [7].

We assembled mitochondrial genomes from all sampled Nanomia and found them all to have the same conserved gene order recently reported for Nanomia septata and Nanomia bijuga [21]. The phylogenetic relationships of the full-length mitochondrial genomes of Nanomia (Fig 4A) are consistent with the relationships based on 16S and CO1 reported by Hosia et al. [7], with improved support (bootstrap values at all internal nodes 100), and with our CO1 analyses (S8 Fig). Nanomia cara is sister to all other Nanomia, and Nanomia septata is sister to a clade of Nanomia bijuga and Nanomia sp. 1 (Fig 4).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Population genomics of Nanomia.

A. Phylogeny based on complete mitochondrial genomes. B. Map of collecting locations, which are coded by color. Colors in other panels correspond to locations here. C. PCA of all Nanomia specimens newly sequenced in this study. D. PCA of Nanomia septata specimens only. E. PCA of Nanomia bijuga samples only. All PCAs use the Nanomia septata reference genome. Analyses with species specific transcriptome references are shown in S10 Fig. Map made with Natural Earth (public domain) in R.

https://doi.org/10.1371/journal.pone.0351247.g004

We also used k-mer distributions (S9 Fig) to confirm and extend recent observations of genome sizes of Nanomia septata and Nanomia bijuga [21]. New samples of Nanomia septata (n = 10, including the genome animal) have genome sizes between 1.4–1.8 GB and have similar repeat and heterozygosity statistics consistent with the single individual analyzed previously [21]. The single sample of Nanomia cara had a 1.7 GB genome, providing critical new information on a species for which no previous genome statistics were available. K-mer analyses of all Nanomia bijuga samples, as well as the specimen of Nanomia sp. 1 show that these species have much smaller genomes of about 0.7 GB and a lower fraction of repetitive elements. Taken together, these results indicate that there are two Nanomia species, Nanomia septata and Nanomia cara, with large genomes of 1.7 GB, and two, Nanomia bijuga and Nanomia sp. 1 with much smaller genomes that are about half the size at 0.7 GB. In combination with the phylogenetic results that show Nanomia septata sister to Nanomia bijuga and Nanomia sp. 1 and Nanomia cara sister to all of these species (Fig 4), these analyses suggest that the most recent common ancestor of Nanomia had a genome size of about 1.7 GB, and that there was then a large reduction in genome size along the branch that gave rise to Nanomia bijuga + Nanomia sp. 1 (Fig 4).

Combined population genomic analyses (see Methods), based on mapping each genome to the chromosome scale reference genome for Nanomia septata, clearly differentiate the species identified above (Fig 4C). There is also strong geographic structure to population diversity within species. Within Nanomia septata (Fig 4D), samples collected near Friday Harbor, Washington are distinct from those collected near Monterey, California. Within Nanomia bijuga (Fig 4E), we observe a separation between Pacific specimens and other specimens on PC1, and between the Mediterranean specimen and other specimens on PC2 in the graph (Fig 4E).

Reads from Nanomia septata map well to the Nanomia septata reference genome (85%–93% properly paired), but reads from Nanomia cara (44%), Nanomia bijuga (22–67%), and Nanomia sp. 1 (44%) map poorly (Fig 4). Previous work in Physalia found that results using transcriptome and genome references are highly similar [18]. To test whether results were robust to choice of reference sequence in Nanomia, we generated transcriptomes (Iso-seq) for Nanomia bijuga and Nanomia septata and used these as a reference for mapping reads from those species (S10 Fig). The PCAs of variation using species-specific transcriptome references are largely consistent with those derived when mapping to the reference genome (S10 Fig).

Nanomia species have distinct but overlapping ranges (Fig 4 and S10C Fig). For example, Nanomia septata is present across the North Pacific, from Japan, Alaska, and California. Likewise, Nanomia bijuga is found in parts of the Pacific, including Hawai’i, Japan, and Mexico, but also in the Atlantic and Mediterranean. Japan is a region of overlap among Nanomia septata and Nanomia bijuga and Nanomia sp. 1 ([7]). Additionally, Nanomia sp. 1 and Nanomia bijuga were both collected in Hawai’i, and Nanomia cara and Nanomia bijuga were both collected in Rhode Island.

Genome description

Karyotype

The karyotyping protocol was adapted from [44] with modifications as described below. We dissected the nectosomal and siphosomal growth zones (NGZ and SGZ) and basigasters (the base of gastrozooids closest to the stem and the origin point of cnidocytes) from four Nanomia sp. (Nanomia V4433 and Nanomia specimens 1, 2, and 3 from diffusion tube culture, labeled as is because no sampler number available) and placed them into small petri dishes (60 mm). Several basigasters were put into 1 hour colchicine treatment in the dark and the remaining basigasters, NGZ, and SGZ were put into a 5 hour colchicine treatment in the dark. Half of the basigasters from the 1 hour colchicine treatment were washed with DI water, by adding water in via drops and immediately pipetting it up. The rest of the basigasters in the 5 hour colchicine incubation were split up and washed with either a mix of DI water: filtered sea water (1:1) ratio or only DI water. Both were immediately pipetted up. We made Carnoy’s Fix fresh and added (3:1 methanol: glacial acetic acid) enough to cover the entire petri dish, which was then put on wet ice for 30 minutes.

Samples in petri dishes were transferred to slides and the liquid was removed. 20µl of 60% glacial acetic acid were added over the tissue, which was covered with siliconized coverslips using Sigmacote. We squashed preps manually using maximum pressure. Squashed preps were then placed overnight in a humidified chamber within a 4℃ refrigerator, and sealed with parafilm. The next day, siliconized coverslips were removed, samples were placed on dry ice, and each slide was washed with ~2mL 1X PBS to cover samples and then stained with DAPI 1:5000. Slides were imaged via confocal.

Transcriptome sequencing

New transcriptomes were sequenced for two specimens, fresh tissue of Nanomia septata (Nanomia-culture-1) and frozen tissue of Nanomia bijuga (YPM-IZ-110217). We extracted RNA with the RNAqueous Total RNA Isolation Kit (Invitrogen) following the standard protocol as described in Church et al. [18]. We then removed DNA using the Turbo DNAse kit (Invitrogen).

Our samples were sent to Keck Microarray Shared Resource at Yale University for PacBio library prep and Iso-Sequencing, which is as follows. 400 ng of total RNA with a RIN of >7 was used as input. First-strand cDNA synthesis and PCR amplification of cDNA products were performed using the NEBNext Single Cell/Low Input cDNA Synthesis & Amplification Module (NEB kit), in conjunction with the Iso-Seq Express Template Switching Oligo and cDNA PCR Primer from the PacBio Iso-Seq Express Oligo Kit (101-737-500), following the PacBio Iso-Seq protocol. A minimum of 200 ng of amplified cDNA was used to prepare SMRTbell libraries with the SMRTbell Prep Kit 3.0 (102-396-000). The SMRTbell libraries were prepared according to the SMRT Link Sample Setup instructions, using the Polymerase Binding Sequel® II Binding Kit 3.1, and loaded onto the Sequel II for sequencing.

Delivered reads were put through our lab’s treeinform workflow as described in Church et al. [18] and at https://github.com/dunnlab/isoseq.

Genome sequencing and initial assembly

Genome HiFi sequencing, Dovetail Omni-C proximity ligation, initial assembly, and annotation were completed by Cantata Bio on specimen YPM-IZ-111756 (sampling number D1399-SS10-NA-1).

Two combined DNA preps for the Nanomia septata sample were made, both of which used a CTAB protocol followed by the Qiagen mini kit. For the first prep, 180 mgs of sample incubated in 10 mls of CTAB, which yielded 8860ng of DNA. For the second prep, 500 mgs of sample incubated in 20 mls of CTAB, which yielded 8440 ng of DNA. DNA samples were quantified using Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA). The PacBio SMRTbell library (~20kb) for PacBio Sequel was constructed using SMRTbell Express Template Prep Kit 2.0 (PacBio, Menlo Park, CA, USA) using the manufacturer’s recommended protocol. The library was bound to polymerase using the Sequel II Binding Kit 2.0 (PacBio) and loaded onto PacBio Sequel II. Sequencing was performed on PacBio Sequel II 8M SMRT cells (2 were used). This produced 4,568,345 HiFi reads totalling 54.2GB in length.

PacBio CCS reads were assembled with Hifiasm1 v0.15.4-r347 using default parameters. Blast results of the Hifiasm output assembly against the nt database were used as input for blobtools2 v1.1.1 [45] and scaffolds identified as possible contamination were removed from the assembly. Finally, purge_dups3 v1.2.5 was used to remove haplotigs and contig overlaps [46].

Structural annotation

Chromatin was fixed in place with formaldehyde in the nucleus. Fixed chromatin was digested with DNase I and then extracted, chromatin ends were repaired and ligated to a biotinylated bridge adapter followed by proximity ligation of adapter containing ends. After proximity ligation, crosslinks were reversed and the DNA purified. Purified DNA was treated to remove biotin that was not internal to ligated fragments. Sequencing libraries were generated using NEBNext Ultra enzymes and Illumina-compatible adapters. Biotin-containing fragments were isolated using streptavidin beads before PCR enrichment of each library. The Omni-C library was sequenced on an Illumina HiSeqX platform to produce ~ 30x sequence coverage.

The de novo HiFi assembly and Dovetail OmniC library reads were used as input data for HiRise, a software pipeline designed specifically for using proximity ligation data to scaffold genome assemblies [47,48]. Dovetail OmniC library sequences were aligned to the draft input assembly using bwa (https://github.com/lh3/bwa). The separations of Dovetail OmniC read pairs mapped within draft scaffolds were analyzed by HiRise to produce a likelihood model for genomic distance between read pairs, and the model was used to identify and break putative misjoins, to score prospective joins, and make joins above a threshold [49].

Assembly refinement and Quality assessment

Juicebox (JBAT) [50] was used to manually curate and fix mis-assemblies and re-orient some contigs. We followed a similar pipeline used to manually curate the Octopus vulgaris genome [51], including using Chromap [52] to map the Omni-C data with a quality cutoff of Q0 and creating .hic files using 3d-dna [53] and artisanal (https://bitbucket.org/bredeson/artisanal) for use in Juicebox. In total, we went through 8 rounds of manual refinement of the genome. Scripts are deposited in our GitHub repository.

To review the quality of our assembly, we ran BUSCO metazoa_odb10 [54, 55] on our assembly and NCBI FCS.

Gene annotation

Our gene annotation was completed by Cantata Bio as follows. Coding sequences from Hydra vulgaris (PRJNA703404), Hydractinia symbiolongicarpus (https://research.nhgri.nih.gov/hydractinia/), Nematostella vectensis (PRJNA667495) and Rhopilema esculentum (PRJNA523480 and PRJNA512552) were used to train the initial ab initio model for Nanomia septata using the AUGUSTUS software v2.5.5 [56]. Six rounds of prediction optimisation were done with the software package provided by AUGUSTUS [56]. The same coding sequences were also used to train a separate ab initio model for Nanomia septata using SNAP v2006-07–28 [57]. RNA-seq reads were mapped onto the genome using the STAR aligner software v2.7 [58] and intron hints generated with the bam2hints tools within the AUGUSTUS software. MAKER [59], SNAP [57] and AUGUSTUS [56] (with intron-exon boundary hints provided from RNA-seq) were then used to predict for genes in the repeat-masked reference genome. To help guide the prediction process, Swiss-Prot peptide sequences from the UniProt database [60] were downloaded and used in conjunction with the protein sequences from Hydra vulgaris, Hydractinia symbiolongicarpus, Nematostella vectensis and Rhopilema esculentum to generate peptide evidence in the Maker pipeline. Only genes that were predicted by both SNAP and AUGUSTUS softwares were retained in the final gene sets. To help assess the quality of the gene prediction, AED scores were generated for each of the predicted genes as part of the MAKER pipeline. Genes were further characterized for their putative function by performing a BLAST [61–63] search of the peptide sequences against the UniProt database. tRNAs were predicted using the software tRNAscan-SE v2.05 [64].

Gene trees

We developed a Snakemake workflow to build gene trees using proteomes from Nanomia septata, two choanoflagellates, and 19 other animals (porifera, ctenophore, placozoa, ecdysozoa, spiralia, and chordata; listed in the github repo under: “analyses_annotation/gene_trees/config/download_targets.tsv”). The workflow downloads protein sequences, functionally annotates them using eggNOG-mapper v2.1.10 [65, 66] and infers individual gene trees using Orthofinder v2.5.4 [67, 68] with default parameters. It merges all gene trees in a master text file. We identified key genes that are conserved across Metazoa, with a focus on Cnidaria, including Homeobox and Wnt genes. We identified orthogroups containing homeobox sequences using human homeobox genes functionally annotated by eggNOG-mapper as landmarks. We corroborated Nanomia septata orthofinder results by identifying homeodomain-containing proteins running HMMER v3.4 (hmmer.org) with the hidden Markov model hd60.hmm from Zwarycz et al. [69], following Steinworth et al. [70]. In order to accurately identify Wnt and Hox-related genes, we used their gene trees from Orthofinder as a starting point for a more detailed phylogenetic analysis.

We retrieved 1,114 protein sequences from the orthogroup containing Antp homeobox genes and combined them with 111 previously identified Hox/Hox-related sequences from Khalturin et al. [71]. We aligned these 1,225 sequences with the L-INS-i algorithm using MAFFT v7 [72], which accounts for one conserved domain and long gaps. After manually trimming the variable regions and keeping only the 60 aa homeodomain, we added 270 Hox/Hox-related homeodomain sequences from Steinworth et al. [70] using MAFFT with the option –add fragments. We trimmed this alignment using trimAl (options -gt 0.9 -seqoverlap 90 -resoverlap 0.1) [73] and inferred a maximum likelihood tree using IQTree2 [74] with the evolutionary model LG + R8, selected by ModelFinder [75]. Ultrafast bootstrap branch support values were calculated based on 1,000 replicates. This tree was rooted on the stem branch of Nk-like genes.

For inferring the Wnt gene tree, we first combined all protein sequences assigned to Wnt orthogroups (325 sequences), 13 Wnt sequences from the hydrozoan jellyfish Clytia hemisphaerica previously identified by Condamine et al. [76], and 12 Wnt sequences from the cubozoan jellyfish Tripedalia identified by Khalturin et al. [76]. We removed two sequences from Orbicella faveolata that were exceptionally long (XP_020618804.1 and XP_020632229.1) and one sequence from Trichoplax adhaerens that was extremely short (TriadP8072). We aligned the remaining 347 sequences and 7 DASH sequences (for structural homology, see [77] using the FFT-INS-ii algorithm on MAFFT v7 [72]. Using IQTree2 [74], we selected the best evolutionary model (LG + R9) and inferred a maximum likelihood tree, with ultrafast bootstrap branch support values based on 1,000 replicates and midpoint rooted.

Additional Homeobox genes that we found manually (not originally included in the annotation) have been added to annotation files (transcript, protein and gff). These sequences were found in the Nanomia septata Iso-sequencing and were checked before their addition.

Differential expression and zooid covariance/co-expression analyses

We re-examined previously published zooid-specific bulk RNA-seq data [8] using salmon v1.10.1 [78]. Counts were then imported with tximport v1.26.1 into DESeq2 v1.38.3. Our scripts also produced figures to show the variance/comparison between different zooid types and developmental stages and how they relate to one another (S3 Fig).

We next sought to identify any genes that are close to each other physically and have strong expression covariance across zooids. This analysis was implemented in the script in the repository https://github.com/dunnlab/nanomia_genome/blob/main/analyses_de/expression-proximity.ipynb. First, for each gene we average the normalized counts across replicates for each combination of tissue and stage that was sequenced. This generated a dataframe with one row per gene and one column per tissue and stage. We ordered genes by chromosome and start site. The gene expression covariance matrix was calculated based on the averaged normalized expression values. We then created a physical distance matrix between genes, where the distance is nan if the genes are on different chromosomes and the number of genes away if they are on the same chromosome, e.g., 0 for self, 1 for adjacent, etc. We then derived a proximity matrix from this distance matrix by passing it through a Gaussian kernel and taking the inverse. In the proximity matrix, the value is 0 if the genes are on different chromosomes, 1 on the diagonal (i.e., the proximity of a gene to itself is 1), and has a smooth transition from 1 to 0 as the genes become more distant. The standard deviation of the kernel was set such that proximity is 0.5 for genes that have a distance of 50. Finally, we took the element-wise product of the expression covariance matrix and the proximity matrix, producing a matrix we refer to as the expression proximity product. Values in this expression proximity product are near zero if the genes are far apart or have low correlation. They are near one if they are physically close and have high positive correlation, and near negative one if they are physically close and have strong negative correlation.

Custom repetitive element library generation and TE detection

Since the Nanomia septata genome presented here is the first genome assembly for the genus Nanomia, the number of repetitive elements registered in databases, such as Dfam, is limited relative to typical model animals [79]. Thus, we generated a custom repeat library using RepeatModeler v2.0.5 [79, 80]. Inactivated TEs accumulate mutations, including base substitutions, insertions, and deletions [81, 82]. Because of this, when RepeatClassifier, a submodule of RepeatModeler, performs a search against Dfam using rmblast, the number of hits can be extremely low, resulting in a library containing many uncharacterized sequences (“Unknown” repeats). This is particularly evident for non-model species whose repetitive elements are not well-represented in Dfam. To address this challenge, we used an approach similar to that used in our previous studies [13, 80]. First, we generated a custom repetitive element library using RepeatModeler v2.0.5 [80] with the Nanomia septata genome assembly as the query. Then, we obtained all Dfam HMM profiles of TEs from the Dfam database, version 3.8 (https://dfam.org/home). Because the Dfam HMM profile dataset for TEs is a huge data size, we divided these HMM profiles into smaller chunks, each containing 25,000 HMM profiles. Each chunk of HMM profiles was indexed using hmmpress [83,84]. Then, we performed an HMM-based search against the Dfam HMM profiles using nhmmscan with default parameters, using the custom repeat library as the query [83,84]. For each query sequence, unclassified query sequences were assigned names using the annotation of the hit with the lowest E-value (≤ 0.05). Using the updated custom repetitive element library, we conducted RepeatMasker [85] on the Nanomia septata genome assembly with the options of -parallel 70 -gff -a -dir -xsmall. For other species, we obtained genome assemblies from the NCBI genome database, and performed RepeatModeler [80] and RepeatMasker [85] in the same manner.

Comparative cnidarian genomics

Population genetics and mitochondrial genomes

Nanomia samples (sample identifiers NA28–38 and WS2–10) were sequenced via Illumina NovaSeq S4 2X150 bp PE. Coverage ranged from 250 million reads to 900 million reads (S11 Fig). Genome size was estimated for samples NA28–38 and WS1–10, for which there was sufficient coverage to achieve an accurate estimate. For this analysis, we used a Snakemake workflow [93] to count k-mers using Jellyfish v2.3.0 [94] and Genomescope v2.0 [95, 96] as detailed in Ahuja et al. [21].

For all samples, libraries were trimmed for Illumina adapters using Trimmomatic v0.39 [97] and library quality was assessed using FastQC v.0.11.9 [98]. Mitochondrial genomes were assembled from a subset of 10,000,000 read pairs using GetOrganelle v.1.7.7.0 [99], using as a seed the SRR23143273 and SRR23143286 publicly available mitochondrial (mt) genomes from Nanomia bijuga sample CWD116 and Nanomia sp. California NA19 [21], and using the animal_mt database and default parameters. All mitochondrial genomes were assembled as linear, with the exception of YPM-IZ-35039; upon visual inspection of mitogenome alignment, this was considered to be a misassembly, and we manually corrected this genome. The mitogenome for YPM-IZ-107008 was assembled in three fragments. For downstream analyses the largest fragment was used. Mitochondrial genomes were annotated with MITOS2 [100] and tRNAscan [64].

A mitochondrial genome was also assembled from PacBio HiFi reads for the genome reference specimen by blasting CO1 from the 2 seed Nanomia mt genomes and identifying the appropriate contigs. Assembled mitochondrial sequences were combined with publicly available mt genomes for Nanomia (OQ957219.1, OQ957196.1), Halistemma (OQ957197.1, OQ957195.1), and Agalma (OQ957203.1, OQ957191.1, OQ957190.1, OQ957189.1), and then aligned using MAFFT, --adjustdirectionaccurately option v.7.505 [72] and a phylogeny was inferred using IQTree2 software [74], model autoselected [75] and 1,000 ultrafast bootstraps [101]. Assembled mitochondrial sequences are available on our GitHub repository at https://github.com/dunnlab/nanomia_genome, which is also archived at https://doi.org/10.5281/zenodo.19629162.

In addition, individual marker sequences for the mitochondrial cytochrome oxidase I (CO1) gene were assembled from a subset of 10 million reads using in silico PCR as implemented in sharkmer (available at https://github.com/caseywdunn/sharkmer). Sequences were combined with publicly available Nanomia sequences, as published in Hosia et al. [7], along with sequences for Halistemma and Agalma from NCBI. Sequences were aligned with MAFFT, and gene trees inferred with IQTree2, as described above. Assembled CO1 sequences are available on our GitHub repository.

Population structure was analyzed following the workflow described in Church et al. [18]. In brief: libraries were assessed for contamination with Kraken2 [102]; sample kinship was evaluated using Plink2 [103]; reads were then mapped to the reference assembly using BWA [104] and alleles were called using BCFTools v1.16 [105]; genotype likelihoods were calculated using ANGSD v.0.935 [106]; principal components of genomic variation were analyzed on genotype likelihoods using PCANGSD v1.21 [107]. We repeated this analysis for Nanomia bijuga and Nanomia septata separately, using assembled transcriptomes from PacBio Iso-Seq data as the reference, and compared our results to those using the Nanomia septata reference genome.