Replication-origin inversions accompany compositional skew and phylogenetic artifacts in fan worm mitochondrial genomes Authors: Renyu Zheng, Yanan Sun, Qinghua Bao, Xuwen Wu, Kuidong Xu Published: 2026 ABSTRACT SummaryInversion of the origin of replication (ORI) in metazoan mitochondrial genomes (mitogenomes) can induce compositional biases misleading phylogenetic inferences. Fan worms (Sabellida, Annelida) exhibit atypical positive guanine-cytosine (GC) skew, a presumed ORI indicator, yet structural evidence remains elusive. We analyzed 38 complete Sabellida mitogenomes, including 20 newly sequenced, and identified inverted replication-associated motifs in the control region of species with positive GC skew species, providing structural evidence for ORI. Throughout Sabellida evolution, three major ORI events arose independently, accompanied by pronounced shifts in GC skew, synonymous codon usage, hydrophobic amino acid frequencies, and elevated nonsynonymous substitution rates in ORI-affected genes. Additionally, extensive gene rearrangement correlated with ORI events, elevated GC skew and substitution rates, but reduced adenine-thymine (AT) skew. ORI lineages show artificial clustering, confounding phylogeny. We propose that ORI events consistently co-occur with compositional skew, structural variations, and gene rearrangements. Coalescent-based methods integrating nuclear and mitochondrial loci are recommended to resolve ORI-affected phylogenies.Keywords: Sabellida, replication origin inversion, mitochondrial DNA, gene rearrangement, GC skew Introduction The widely distributed polychaete Sabellida Levinsen, 1883, commonly known as “fan worms” or “flowers of the sea,” is a group of tube-dwelling suspension feeders characterized by a sessile tube and a versatile, flower-like radiolar crown.1 Despite creating biodiverse habitats,2 fan worm aggregations can also become biofoulers or invasive species via marine transport.3 The order Sabellida comprises three families: Fabriciidae Rioja, 1923; Sabellidae Latreille, 1825; and Serpulidae Rafinesque, 1815.1 The monophyly of Sabellida and its families has been well-established and is strongly supported by multi-locus and transcriptomic datasets.4,5,6,7,8 Despite this consensus, mitochondrial genomes (mitogenomes), a molecular tool often advocated for balancing simplicity and precision, have consistently failed to recover the monophyly of both the order Sabellida and the family Sabellidae.9,10,11 Yet the underlying cause of this discrepancy remains unknown. The inversion of the origin of replication (ORI) has been recognized as one of the factors compromising mitogenome-based phylogenetic accuracy in several major metazoan lineages, such as Arthropoda12 and Annelida.13 ORI events can induce mutational bursts in nucleotide sequences, thereby violating the assumption of nucleotide frequency stationarity underlying model-based phylogeny.14,15 However, both ORI events themselves and their associated nucleotide skewness remain largely overlooked in phylogenetic and evolutionary studies.12 The definitive identification of ORI events requires physical replication mapping,16 which is hardly scalable for large comparative studies. Currently, two complementary lines of evidence are used for indirect inference: inversions of conserved motifs in the control region and associated flipping of nucleotide skew, e.g., a shift in GC skew from negative to positive.12,17 Such nucleotide skewness reflects strand-specific mutation pressure during mitochondrial DNA replication.12,18 Standard mitochondrial DNA replication, through spontaneous deamination of adenine and cytosine on the single-stranded parental H-strand (adenine-to-guanine and cytosine-to-thymine), consistently enriches one strand with guanine and thymine and the other with adenine and cytosine, producing a predictable skew pattern.19 However, an ORI event reverses this pattern, leading to the fully flipped skewness observed in some metazoan mitogenomes.20,21,22 Significant deviation from the background skew, even without a sign change, has also been used as presumptive evidence for ORI inference.12 While motif-based evidence has been detected in crustaceans and insects,12,22 its occurrence in annelids has so far been inferred preliminarily from flipped GC skew. For instance, a positive GC skew was proposed as evidence for an ancestral ORI event in Serpulidae and Sabellidae.13 However, this hypothesis was challenged by recently published Sabellidae mitogenomes, which exhibit a negative GC skew consistent with the pattern found in most other annelids.23,24 This variety necessitates further structural confirmation of ORI events and a re-evaluation of their evolutionary scope within Sabellida. Pronounced positive GC-skew in certain Sabellida lineages correlates with extensive gene rearrangements from the conserved annelid putative ground pattern (PGP).10 This pattern has been documented in Sabella spallanzanii,25 Bispira melanostigma,9 and all sequenced species of Serpulidae.11,26,27 The co-occurrence of these genomic features offers a clue for the hypothesized role of ORI events in annelid mitogenomes rearrangement.10 We hypothesize that ORI events have occurred multiple times in Sabellida and underlie the observed mitochondrial features, including atypical nucleotide skewness and amino acid composition. These changes lead to biased phylogenetic estimates based on mitogenomic sequences. We further propose that ORI events commonly co-occur with extensive mitochondrial gene order rearrangements. Since complex rearrangements are unlikely to reverse or arise convergently,15,28 the distinct gene orders in Sabellida mitogenomes may thus represent a potentially valuable source of phylogenetic signal. Current sampling within Sabellida is severely limited, with mitogenomes available for only 12 of the 128 described genera, covering only 10% of known diversity.1 To address the gap, we sequenced mitogenomes of 20 Sabellida species, representing 13 genera across Sabellidae and Serpulidae. We aim to: (1) identify ORI events in Sabellida by analyzing nucleotide skewness and replication-associated motifs; (2) characterize the impact of ORI events on sequence composition and structural architecture; and (3) evaluate the effects of ORI events on phylogenetic reconstruction. Results General features of Sabellida mitochondrial genomes Of the 20 species newly sequenced (Table 1), a single mitogenome contig was assembled in 17 species, with 9 of them circularized. Contigs length ranges from 14,101 base pairs (bp) in Laonome cf. indica to 22,344 bp in Protula sp. 1 (Table 2). Two contigs were recovered in Parasabella cf. microphthalma (14,912 + 1,578 bp), Hypsicomus sp. (13,101 + 10,897 bp), and Hydroides bulbosa (6,620 + 14,770 bp) (Table 2). In each case, the 2 contigs carry complementary gene sets and together contain the full complement of 13 protein-coding genes (PCGs), 2 ribosomal RNAs (rRNAs), and 22 transfer RNAs (tRNAs); all other species returned a single contig with the complete gene set (Figure 1). Gene duplications were observed in Parasabella sp. 1 (duplicated trnL1), Hypsicomus sp. (duplicated trnL2), and Hydroides bulbosa (duplicated trnR) (Figure 1). All mitogenomes contained large non-coding regions (NCRs) (>100 bp), ranging in number from 1 in Branchiomma cf. pererai, Eudistylia vancouveri, and Laonome cf. indica to 12 in Protula sp. 1 (Figure 1). Table 1Specimen collection details and GenBank accession numbers for the sequences generated in this studyFamilySpeciesCollection localityVouchersGenBank accessions18S28SMitogenomesSabellidaeAcromegalomma sp.Mulan bay, ChinaMBM288569PX778526PX778546PX939022–Bispira sp. 1Lingshui, ChinaMBM288570PX778527PX778547PX939015–Bispira sp. 2Qingdao, ChinaMBM288571PX778528PX778548PX939016–Branchiomma cf. pereraiLianzhou bay, ChinaMBM288572PX778529PX778549PX939017–Branchiomma sp.Lingshui, ChinaMBM288573PX778530PX778550PX939018–Eudistylia vancouveri Kinberg, 1866Qingdao, ChinaMBM288574PX778531PX778551PX939019–Hypsicomus sp.aLingshui, ChinaMBM288575PX778535PX778555PX939026PX939027–Laonome cf. indicaLianzhou bay, ChinaMBM288576PX778536PX778556PX939020–Laonome sp.Qingdao, ChinaMBM288577PX778537PX778557PX939023–Myxicola sp.Qingdao, ChinaMBM288578PX778538PX778558PX939021–Parasabella sp. 1Wenchang, ChinaMBM288579PX778540PX778560PX939024–Parasabella cf. microphthalmaaQingdao, ChinaMBM288580PX778539PX778559PX832396PX832397–Parasabella sp. 2Qingdao, ChinaMBM288581PX778541PX778561PX939025SerpulidaeFicopomatus shenzhensis Li, Wang & Deng, 2012Qionghai, ChinaMBM288582PX778532PX778552PX970955–Ficopomatus sp.Wenchang, ChinaMBM288583PX778533PX778553PX970953–Hydroides bulbosa ten Hove, 1990aLingshui, ChinaMBM288584PX778534PX778554PX970959PX970960–Pileolaria sp.Lingshui, ChinaMBM288585PX778542PX778562PX970956–Protula sp. 1Hong Kong, ChinaMBM288586PX778543PX778563PX970957–Protula sp. 2Hong Kong, ChinaMBM288587PX778544PX778564PX970958–Spirobranchus sp.Hong Kong, ChinaMBM288588PX778545PX778565PX970954aThe mitochondrial genome was assembled into two complementary contigs that together constitute the complete mitogenome, each deposited under a separate GenBank accession number. Specimen collection details and GenBank accession numbers for the sequences generated in this study The mitochondrial genome was assembled into two complementary contigs that together constitute the complete mitogenome, each deposited under a separate GenBank accession number. Table 2Contig information, nucleotide content, and nucleotide skewness of the newly assembled mitochondrial genomesSpeciesContig length (bp)aGC content (%)AT skewGC skewCategoryWGbAllPCGs3rdPCGstRNAsrRNAsWGAllPCGs3rdPCGstRNAsrRNAsWGAllPCGs3rdPCGstRNAsrRNAsEudistylia vancouveri14,969(C)37.738.333.337.334.8−0.05−0.100.080.060.12−0.24−0.29−0.56−0.03−0.12N.A.Myxicola sp.17,488(C)36.237.931.837.935.7−0.02−0.070.110.050.12−0.23−0.22−0.43−0.05−0.16N.A.Acromegalomma sp.16,20341.040.633.242.943.0−0.10−0.140.000.060.04−0.33−0.38−0.66−0.06−0.20N.A.Parasabella sp. 120,34842.542.738.542.942.6−0.01−0.060.180.060.15−0.35−0.41−0.65−0.09−0.19N.A.Parasabella cf. microphthalma14,91241.141.234.242.943.8−0.08−0.120.030.030.06−0.31−0.35−0.61−0.06−0.21N.A.–1,57840.5N.A.N.A.50.044.40.07N.A.N.A.0.070.10−0.12N.A.N.A.−0.20−0.15–Parasabella sp. 215,39845.145.143.044.045.4−0.04−0.090.140.040.11−0.36−0.40−0.68−0.09−0.21N.A.Bispira sp. 215,503(C)38.840.437.436.534.30.17−0.22−0.13−0.05−0.06−0.220.130.250.090.18IIIBispira sp. 115,288(C)38.539.736.134.838.7−0.06−0.140.04−0.020.140.41−0.16−0.240.06−0.23IIBranchiomma cf. pererai15,311(C)36.737.829.931.939.5−0.05−0.130.030.020.130.34−0.13−0.180.05−0.20IIBranchiomma sp.15,312(C)37.838.932.933.138.8−0.06−0.120.04−0.020.130.37−0.14−0.190.07−0.20IIHypsicomus sp.13,10135.535.529.240.836.2−0.21−0.28−0.29−0.040.050.230.200.590.140.08I–10,89737.236.530.036.943.1−0.22−0.28−0.30−0.160.080.350.330.630.260.17–Laonome cf. indica14,101(C)34.434.524.136.135.0−0.20−0.25−0.17−0.060.020.320.350.600.240.20ILaonome sp.14,51236.537.030.437.635.3−0.20−0.26−0.17−0.020.010.400.430.760.300.27IFicopomatus shenzhensisc17,81530.331.221.034.629.8−0.26−0.36−0.40−0.10−0.030.350.320.560.270.28IFicopomatus sp.15,775(C)29.430.119.330.334.5−0.16−0.26−0.19−0.030.040.370.400.670.260.18IHydroides bulbosa6,62046.143.240.551.347.0−0.16−0.20−0.39−0.040.000.320.360.500.170.17I–14,77044.043.242.249.5N.A.−0.33−0.41−0.54−0.13N.A.0.450.470.740.26N.A.–Pileolaria sp.16,62154.652.957.448.347.0−0.16−0.27−0.250.050.060.300.300.360.220.24IProtula sp. 122,34441.240.736.946.546.2−0.20−0.32−0.39−0.11−0.010.440.430.770.300.17IProtula sp. 220,31642.041.840.144.947.1−0.23−0.32−0.35−0.130.000.410.410.680.280.24ISpirobranchus sp.17,455(C)39.439.332.543.043.4−0.24−0.33−0.37−0.11−0.050.350.360.580.270.25IThe contigs were categorized as having no GC skew flip (N.A.) or belonging to one of three flip subtypes (I–III).aSuperscript (C) indicates circularized contigs generated using NOVOPlasty; linear contigs were generated using MitoFinder.bWG, The whole mitochondrial genome.cAs a frameshift error was observed in the nad1 gene of Ficopomatus shenzhensis from the MitoFinder output, the SPAdes-assembled contig was used for this species. Contig information, nucleotide content, and nucleotide skewness of the newly assembled mitochondrial genomes The contigs were categorized as having no GC skew flip (N.A.) or belonging to one of three flip subtypes (I–III). Superscript (C) indicates circularized contigs generated using NOVOPlasty; linear contigs were generated using MitoFinder. WG, The whole mitochondrial genome. As a frameshift error was observed in the nad1 gene of Ficopomatus shenzhensis from the MitoFinder output, the SPAdes-assembled contig was used for this species. Mitochondrial gene arrangement and photo of species Hollow circle between genes indicates large non-coding regions (>100 bp). Different colors indicate putative conserved gene clusters in Pleistoannelida. Circularized mitochondrial genomes are aligned starting from cox1; linear contigs retain their original gene order. Most Sabellida mitogenomes were adenine-thymine (AT)-rich, except Pileolaria sp. (GC content = 54.6%). In PCGs, third codon positions showed lower GC contents than the average (Table 2). Six species from the subfamily Myxicolinae7 (genera Eudistylia, Myxicola, Parasabella, and Acromegalomma) exhibited strongly negative PCG GC skew (from −0.41 to −0.22); three Sabellinae species (Bispira sp. 1, Branchiomma cf. pererai, and Branchiomma sp.) showed moderately negative values (from −0.16 to −0.13); all Serpulidae species, together with Bispira sp. 2, Hypsicomus sp., Laonome cf. indica, and Laonome sp. exhibited positive PCG GC skew (Table 2). PCG AT skew values were negative across all mitogenomes. Third codon positions of PCGs showed GC skew deviating farther from zero than the average, while AT skew showed no consistent shift. Compared to the PCGs, the tRNA and rRNA genes exhibited both GC and AT skews approaching zero (Table 2). ORI-affected genes exhibited significantly higher median GC skew (+0.36 vs. −0.35) and lower AT skew (−0.30 vs. −0.09) than non-ORI genes (Table 3). Table 3Per-protein coding gene and per-taxon statistical comparisons between ORI and non-ORI groupsGroupNRange (mean) [median]Shapiro-Wilk pTest (one-sided)p valueGC skew (per gene)non-ORIa175−0.02 to −0.63 (−0.35) [−0.35]5.99 × 10−2 (NS)Student’s t test:ORI > non-ORI4.17 × 10−223 (∗∗∗)–ORIa319+0.06 to +0.82 (+0.36) [+0.36]9.33 × 10−2 (NS)––AT skew (per gene)non-ORI175−0.28 to +0.14 (−0.09) [−0.09]1.11 × 10−1 (NS)Wilcoxon rank-sum test:ORI < non-ORI8.09 × 10−65 (∗∗∗)–ORI319−0.54 to −0.02 (−0.30) [−0.30]1.43 × 10−2 (∗)––Common interval distanceb (per taxon)non-ORI100 to 178 (94.40) [123]6.32 × 10−3 (∗∗)Wilcoxon rank-sum test:ORI > non-ORI2.25 × 10−3 (∗∗)–ORI28142 to 180 (168.93) [173]5.91 × 10−5 (∗∗∗)––Kac (per gene)non-ORI162d0.16 to 0.93 (0.49) [0.50]1.72 × 10−3 (∗∗)Wilcoxon rank-sum test:ORI > non-ORI7.07 × 10−15 (∗∗∗)–ORI3190.20 to 0.99 (0.63) [0.65]8.54 × 10−8 (∗∗∗)––Ka/Ksc (per gene)non-ORI1620.04 to 0.79 (0.20) [0.16]1.65 × 10−10 (∗∗∗)Wilcoxon rank-sum test:ORI > non-ORI2.90 × 10−10 (∗∗∗)–ORI3190.04 to 0.97 (0.27) [0.25]6.62 × 10−8 (∗∗∗)––Significance levels: NS, not significant; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001.aNon-ORI/ORI genes, PCGs with negative/positive GC skew; Non-ORI/ORI taxa, mitogenomes with PCG GC skew below/above the significantly elevated threshold (−0.17).bCalculated against the putative ground gene order of Pleistoannelida.10cKa, nonsynonymous substitution rate; Ka/Ks, ratio of nonsynonymous to synonymous substitution rate.dThe 13 PCGs of Manayunkia occidentalis were used as the reference for calculating Ka and Ka/Ks values and were excluded from the Non-ORI group. Per-protein coding gene and per-taxon statistical comparisons between ORI and non-ORI groups Significance levels: NS, not significant; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001. Non-ORI/ORI genes, PCGs with negative/positive GC skew; Non-ORI/ORI taxa, mitogenomes with PCG GC skew below/above the significantly elevated threshold (−0.17). Calculated against the putative ground gene order of Pleistoannelida.10 Ka, nonsynonymous substitution rate; Ka/Ks, ratio of nonsynonymous to synonymous substitution rate. The 13 PCGs of Manayunkia occidentalis were used as the reference for calculating Ka and Ka/Ks values and were excluded from the Non-ORI group. The canonical start codon ATG was commonest in most PCGs, followed by GTG, ATT, TTG, ATC, and ATA (Table S8). Complete stop codons (TAA and TAG) were commonly used, while partial stop codons (TA or T) were frequent in cytochrome c oxidase subunits I and II (cox1/2), cytochrome b (cytb), and NADH dehydrogenase subunit 4L (nad4L) (Table S8). Relative synonymous codon usage (RSCU) analyses revealed significant shifts in synonymous codon usage within Serpulidae, notably involving glutamic acid, for which the codon GAA was replaced by GAG), as well as glycine (GGA→GGG), lysine (AAA→AAG), methionine (AUA→AUG), and glutamine (CAA→CAG) (Figure 2A). Comparison of synonymous codon usage and amino acid frequency among mitochondrial genomes of Spionidae (outgroup), Fabriciidae, Sabellidae, and Serpulidae (A) Comparison of relative synonymous codon usage (RSCU) across the four families. Colored segments represent the mean RSCU values of individual synonymous codons across species within each family, stacked according to their corresponding amino acids. Error bars indicate standard deviation (SD) among species. No SD is shown for Fabriciidae because only one species was included (n = 1). The horizontal axis lists the 62 analyzed codons, grouped by their corresponding amino acid; the vertical axis shows the RSCU value. (B) Hierarchical clustering of amino acid frequencies across genera. Numbers in parentheses indicate the count of species per genus (shown for genera with more than one species); roman numerals or “N.A.” before genus names denote the GC skew flip type. S1/S2 and L1/L2 were separated in analyses. A, alanine; C, cysteine; D, aspartic acid; E, glutamic acid; F, phenylalanine; G, glycine; H, histidine; I, isoleucine; K, lysine; L1 and L2, leucine; M, methionine; N, asparagine; P, proline; Q, glutamine; R, arginine; S1 and S2, serine; T, threonine; V, valine; W, tryptophan; Y, tyrosine. Leucine (L, including L1 and L2) was the most frequent amino acid (11.6%–17.7%), while cysteine (C) was the least (0.6%–3.4%) (Figure 2B). Based on frequency variation, the 20 amino acids were grouped into 4 clusters (Figure 2B). The frequency of cluster A (alanine [A], serine [S2], threonine [T], isoleucine [I], methionine [M], and proline [P]) and cluster D (L1) decreased in Serpulidae and three genera of Sabellidae (Hypsicomus, Bispira, and Laonome), while cluster B (phenylalanine [F], glycine [G], valine [V], and L2) increased. Cluster C (cysteine [C], aspartic acid [D], arginine [R], glutamine [Q], glutamic acid [E], histidine [H], lysine [K], asparagine [N], tryptophan [W], tyrosine [Y], and S1) remained low in all taxa (Figure 2B). Most ATP synthase subunit 8 (atp8) genes in Sabellidae had a conserved N-terminal motif MPHL, except for Bispira melanostigma and Bispira sp. 2, which had MPQL. By contrast, atp8 genes in Fabriciidae and most Serpulidae had a truncated N-terminal motif MP, except Ficopomatus sp. (MF) and Pileolaria sp. (MG). Species of Laonome, Pileolaria, and Ficopomatus had distinctly shorter atp8 proteins (26–40 amino acids) compared to the regular size of >50 amino acids (Figure S1). Atp8 genes in B. melanostigma and Pileolaria sp. exhibited the highest pairwise p-distance across Sabellidae and Serpulidae, respectively. Compared to other families, Serpulidae had the highest p-distances; within each family, p-distances were comparable (Figure S2). Conserved sequence motifs in control region of Sabellida mitochondrial genomes Three conserved sequence motifs were identified with an E-value threshold of <1 × 10−5 (Figures 3 and S3). Consistent with findings in crustaceans,12 all three motifs were assigned with the Gene Ontology (GO) term “regulation of DNA-template transcription” (Figures 3A–3C). Motif 1 was additionally associated with terms “DNA replication initiation” and “sequence-specific DNA binding” (Figure 3A and Table S9) and was predicted to form a stem-loop structure across multiple species (Figure S4), underscoring its specific association with the replication mechanism. A total of 672, 1,063, and 2,209 copies of motifs 1, 2, and 3, respectively, were identified with a p value threshold of 1 × 10−4 (Figure 3D). Motifs 1 and 2 copies exhibited strong strand bias with respect to cox1 GC skew: in mitogenomes with a negative GC skew in cox1, the copies were predominantly located on the opposite strand (82.5% and 90.3%, respectively), while in those with a positive GC skew in cox1, the copies were predominantly located on the same strand (90.6% and 94.1%, respectively) (Figure 3D). In contrast, motif 3 copies showed nearly random strand distribution: 54.0% same vs. 46.0% opposite in mitogenomes with a negative GC skew in cox1, and 48.6% same vs. 51.4% opposite in mitogenomes with a positive GC skew in cox1. Conserved sequence motifs putatively associated with replication mechanism in Sabellida mitochondrial genomes (A–C) Identification and GO functional annotation of conserved motifs 1–3, showing only terms highly relevant to mitochondrial replication and their matching frequency among six databases (complete annotations in Table S9). (D) Strand distribution of the three motifs in mitogenomes with negative and positive GC skew in cox1. (E) Distribution of motif identification outcomes for high-confidence “circularized only” vs. inclusive “all” datasets. For each motif, the stacked bar chart shows the percentage of mitogenomes assigned to the defined outcome categories (“precise hit,” “ambiguous hit,” or “failed”). (F) Cumulative p value distributions of the three identified putative functional motifs across Sabellida mitogenomes. The horizontal axis represents -lg(p) (with original p values labeled); the vertical axis represents the cumulative percentage of motifs (0%–100%). (G) Spatial distribution and strand orientations of the three motifs within the largest NCR of the 16 circularized mitogenomes (except Ficopomatus enigmaticus which is shown within its non-largest NCR). Species names followed by “NO” and “O” denote non-ORI (negative GC skew in cox1) and ORI (positive GC skew in cox1) taxa, respectively. Letters “p” (precise hit) and “a” (ambiguous hit) next to each motif indicate the classification result. At least one putative functional copy (precise/ambiguous hit) was identified in each mitogenome (Table S10). Among the “circularized only” dataset (verified NCR integrity), motif 1 exhibited the highest precise hit rate (68.8%), motif 3 exhibited the lowest failed rate (0.0%), while motif 2 exhibited both the lowest precise (31.3%) and the highest failed rate (43.7%) (Figure 3E and Table S10). In the “all” dataset including linear contigs, motifs 1 and 3 showed the expected reduction in precise hit (from 68.8% to 45.5% and from 43.8% to 36.4%, respectively) and increase in failed rate (from 12.4% to 22.7% and from 0.0% to 6.8%, respectively), reflecting their sensitivity to NCR integrity. In contrast, motif 2 exhibited an increased precise hit (from 31.3% to 36.4%) and decreased failure (from 43.7% to 27.2%), highlighting its limited specificity. In the cumulative p value distribution, motif 1 accounted for the highest proportion of hits in the most stringent significance bin (<1 × 10−6), a threshold previously applied by Jakovlić et al.12 (Figure 3F). Putative functional copies of the three motifs were identified in the largest NCR of each circularized mitogenome, with variable presence and order across species. Motifs 1 and 3 were commonly present (13 of 16 and 15 of 16 species, respectively), while motif 2 was frequently absent in 9 of 16 species. Despite the higher occurrence of motif 3, motif 1 yielded more precise hits (11 precise + 2 ambiguous) than motif 3 (7 precise + 8 ambiguous) (Figure 3G). Nevertheless, in a few species with motifs 1 and 2 absent (e.g., Amphiglena cf. mediterranea, Laonome cf. indica, and Ficopomatus enigmaticus), only motif 3 was detected. Cross-validation against different assembly methods yielded fully aligned motif hits (21 of 22) in the largest NCRs of circularized mitogenomes. The single exception was a motif fragmented in the reference-guided assembly, resulting in its classification shifting from a precise to an ambiguous hit, while its orientation and relative position remained unchanged (Figure S5). Contrasting phylogenetic relationships in Sabellida For each of the nuclear-only and the mitogenome matrices, the ML and BI analyses yielded identical topologies (Figure 4). The nuclear tree recovered monophyletic Serpulidae (ultrafast bootstrap [UFBoot]/posterior probability [PP] = 100/1) and Sabellidae (UFBoot/PP = 91/1). Within Serpulidae, the monophyletic Serpulinae (clade i, UFBoot/PP = 96/0.99) was recovered as sister of Filograninae + Spirorbinae group (clade ii, UFBoot/PP = 90/0.98) (Figure 4). Two subgroups within Serpulinae, representing Serpulini (Hydroides, clade ia, UFBoot/PP = 100/1) and Ficopomatini (Ficopomatus, Marifugia, and Spirobranchus, clade ib, UFBoot/PP = 100/1), were inferred with well-support. Within Sabellidae, two main clades were inferred with well-support (clade iii, UFBoot/PP = 94/1 and clade iv, UFBoot/PP = 89/0.98). Clade iii was further split into two subclades. One included a moderately supported monophyletic group of five genera that had been attributed to Sabellinae, including Hypsicomus, Bispira, Sabella, Branchiomma, and Sabellastarte (clade iiia, UFBoot/PP = 80/1). However, relationships among the five genera remained uncertain owing to low support. The other subclade (clade iiib, UFBoot/PP = 100/1) included a sister group of Eudistylia-Pseudopotamilla, which had been recently attributed to the subfamily Myxicolinae.7 The remaining Myxicolinae species involved in the analyses clustered into clade iv (UFBoot/PP = 89/0.98), inferring sister relationship of well-supported clades Parasabella-Acromegalomma-Amphiglenina (UFBoot/PP = 100/1) and Laonome-Chone-Myxicola (UFBoot/PP = 94/1). Discordance between nuclear (left) and mitochondrial (right) phylogenies of Sabellida and per-gene GC skew pattern of each taxon Maximum likelihood trees are shown with outgroups hidden. Node support values are given in the order: UFBoot/SH-aLRT/posterior probability (PP). Asterisks (∗) stand for full support; pound signs (#) stand for low values (UFBoot/SH-aLRT <50, PP < 0.5). Branches in the mitochondrial tree highlighted in red bold indicate topological conflicts with the nuclear tree. Clade scopes of Sabellinae and Myxicolinae follow Tilic et al.7 The right column displays per-gene GC skew bubble plots and GC skew flip types for each taxon. Flip types are classified as: N.A., all genes with negative GC skew; type I, all genes with positive GC skew; type II, minority of genes with positive GC skew; type III, majority of genes with negative GC skew. Orange boxes in the bubble plots highlight individual genes with a negative GC-skew but significantly elevated above the threshold. However, ML and BI analyses based on mitochondrial genes yielded topologies incongruent with nuclear trees in several clades (Figure 4). For instance, the family Sabellidae was non-monophyletic; instead, Serpulidae was nested within Sabellinae as sister to the Hypsicomus-B. -melanostigma group (UFBoot/PP = 90/1), which was sister to a clade comprising (Bispira sp. 1 + Bispira sp. 2)-Sabellastarte-Branchiomma-Sabella (UFBoot/PP = 83/1) (Figure 4). On the other hand, terminals attributed to Myxicolinae were recovered as monophyletic in the mitochondrial gene tree with low support (clade IV, UFBoot/SH-like approximate likelihood ratio test [SH-aLRT]/PP = 64/100/1). Within Myxicolinae, two subclades, “(Parasabella + Acromegalomma)-Amphiglenina-Laonome” (clade IVa, UFBoot/SH-aLRT/PP = 64/100/1) and “(Eudistylia-Pseudopotamilla)-(Chone-Myxicola)” (clade IVb, UFBoot/PP = 100/1), were recovered, respectively. The ASTRAL-derived species tree supported the monophyly of the families Serpulidae (local posterior probability [LocalPP] >0.99) and Sabellidae (LocalPP = 0.95), and subfamilies Serpulinae (clade AI, LocalPP >0.99), Filograninae (LocalPP >0.99), Myxicolinae (clade B + C, LocalPP = 0.91), and Sabellinae (clade D, LocalPP = 0.79) (Figure 5A). The phylogenetic positions of serpulids were consistent between the species tree and mitochondrial tree. Within Myxicolinae, both trees produced congruent topologies when Pseudopotamilla-Eudistylia and Chone-Myxicola were constrained as sister groups, though with low support (clade C, LocalPP = 0.39) (Figure 5A). Relationships within Sabellinae remained unresolved due to a lack of strong nodal support. Mitochondrial gene rearrangements and GC skew flip patterns across Sabellida (A) Scenarios of mitochondrial gene rearrangements and hypothetical ancestral gene order in Sabellida. The cladogram is based on the species tree topology; the black dashed line indicates the constrained clade; red and yellow dots indicate well-supported nodes (local posterior probability ≥0.95) and moderately supported nodes (0.75 ≤ local posterior probability <0.95), respectively. Newly sequenced species are highlighted with red squares. Green stars mark the three inferred ORI events. Gene orders are shown for protein-coding and rRNA genes only. Colors denote putative conserved gene clusters inherited from the Pleistoannelida ancestor; a minus sign indicates genes located on the complementary strand relative to cox1. Rearrangement events include I (inversion); T (transposition); TDRL (tandem-duplication-random-loss). Ordered and unordered rearrangement events are shown in square and round brackets. The unconstrained species tree is provided in Figure S8. (B) PCG GC skew of each mitogenome. Black circles represent the mean GC skew across all 13 PCGs in each mitogenome; the dashed line marks the threshold for a significantly elevated GC skew (baseline mean + 2 × SD). Red and blue bars represent the mean GC skew of G-skewed and C-skewed PCGs per taxon, respectively. Numbers next to the bars indicate the count of PCGs included in each mean. (C–F) Schematic illustrations of the four GC skew flip types (N.A. and types I–III), showing the orientation of motif 1 and the strand distribution of PCGs that together produce each pattern. Substitution saturation analyses showed a strong linear correlation between patristic and pairwise distances (R2 = 0.936, Figure S6). No excessively long branch (>20 × median length) was detected in any species (Table S11); therefore, no taxon was flagged as potentially affected by long-branch attraction (LBA) effects. GC skew patterns and phylogenetic distribution in Sabellida mitogenomes The per-gene distribution of GC skew revealed four distinct patterns across Sabellida mitogenomes. The pattern with all PCGs uniformly negative, consistent with the ground pattern observed in most annelids, was designated as N.A. Type I GC skew flip denoted uniformly positive GC skew across all PCGs. Type II denoted a minority of PCGs (cox1–3, atp8, and nad3) with positive GC skew. Type III denotes a majority of PCGs positive, which is the complementary set to type II (Figure 4). When mapped to the species tree, species with a flipped per-gene GC skew (uniform or partial) were clustered into three lineages: Serpulidae, Sabellinae, and the genus Laonome (Figure 5A). At the level of 13 PCGs, all N.A. taxa fell below the threshold of significantly elevated GC skew (−0.17). Type II taxa, although remaining negative, all exceeded this threshold. Types I and III species exhibited fully flipped values, surpassing the threshold by a wider margin (Figure 5B). The four skew patterns corresponded to distinct strand distributions of PCGs. In N.A. and type I, all 13 PCGs resided on a single strand and uniformly exhibited negative and positive GC skew, respectively (Figures 5B–5D). In type II, the five PCGs exhibiting positive GC skew were all located on the strand congruent with cox1 (Figures 5B and 5E). In type III, the eight PCGs exhibiting positive GC skew were all located on the strand opposite to cox1 (Figures 5B and 5F). Comparison between mitochondrial gene arrangements Despite frequent variation in gene arrangement across Sabellida mitogenomes and the absence of globally conserved gene blocks, several locally conserved blocks were detected (Figures 1 and 5A). For instance, the Sabellidae genera Acromegalomma and Parasabella shared four conserved blocks “cox1-trnM-trnY,” “cytb-nad2-trnW-trnL2,” “trnT-trnQ-rrnS,” and “nad5-atp8-cox3”; genera Sabella, Sabellastarte, Branchiomma, and Bispira (except B. melanostigma) shared block “(-rrnL)-(-rrnS)-cox1” (Figures 5A and S7). In Serpulidae, the genus-specific blocks “nad5-trnE-nad4” in Hydroides, “trnR-trnP-nad1-cox1” and “cytb-trnC-nad4L-trnG-trnH-nad2” in Spirobranchus were identified. Additionally, “trnA-nad1” was shared by Pileolaria and Protula (Figure S7). In the tRNA-excluded gene arrangements, only three species (M. occidentalis, E. vancouveri, and C. infundibuliformis) strictly retained the PGP of Pleistoannelida (Figure 5A).10 TreeREx analysis against the species tree inferred that the hypothetical ancestral gene order of Serpulidae (cox1-nad2-atp8-rrnS-rrnL-nad6-nad5-nad3-cytb-cox2-atp6-nad4-cox3-nad1-nad4L) and Sabellidae (cox1-cytb-cox2-atp8-nad1-nad3-nad4L-rrnS-nad2-cox3-rrnL-nad6-atp6-nad5-nad4) were retained in Pileolaria sp. and Hypsicomus sp., respectively (Figure 5A). The putative ancestral gene order for the Sabellidae + Serpulidae clade was inferred to be identical to that for the ancestral Sabellidae. Pairwise common interval distances for mitochondrial gene arrangements ranged from 0 to 180 (Table S12), with distances to the ground gene order (PGP of Pleistoannelida) showing a significantly higher median in ORI-affected taxa than in non-ORI taxa (173 vs. 123; Table 3). Comparison of species-level gene arrangements with the ground gene order revealed a strong phylogenetic signal in common interval distances, as indicated by both Blomberg’s K (1.116, p = 0.001) and Pagel’s λ (>0.999, p < 0.001) (Table S13). Selective pressure and mitogenomic correlates of gene rearrangement For each PCG, Ka and Ka/Ks values were significantly higher in the putatively ORI-affected genes (GC skew >0) than in the unaffected genes (GC skew <0), except for atp8 (Figures S14 and S15). Across all 13 PCGs combined, ORI-affected genes exhibited a higher median Ka (0.65 vs. 0.50) and Ka/Ks (0.25 vs. 0.16), with both differences being statistically significant (Wilcoxon test, p = 7.07 × 10−15 and 2.90 × 10−10, respectively) (Figure 6A and Table 3). Among all PCGs, atp8 had the highest Ka (0.81 in both groups) and Ka/Ks (0.54 vs. 0.46, ORI-affected vs. not-affected) values, while cox1 had the lowest Ka (0.34 vs. 0.19) and Ka/Ks (0.10 vs. 0.05) (Figure 6A). Selective pressure and mitogenomic correlates of gene rearrangement (A) Ka and Ka/Ks values for each PCG, grouped by ORI-affected (GC skew >0) and Not-affected (GC skew <0). Data in the left panel are presented as mean + SD. Comparisons were performed using Student’s t test when both groups satisfied normality and Wilcoxon rank-sum test otherwise. The right panel shows overall comparisons of Ka and Ka/Ks between the two groups, with data pooled across all 13 PCGs and displayed as violin plots. Significance levels: NS, not significant; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001. Detailed statistics are provided in Table S14 (Ka) and S15 (Ka/Ks). (B) Correlation between common interval gene order distance and nine mitogenomic features, evaluated at both per-gene and whole-genome (WG) levels. Spearman’s rank correlation was used due to the non-uniform distribution of common interval distances. The corresponding p values are shown within each cell. Warm and cool colors indicate positive and negative correlations, respectively, with color intensity corresponding to the p value categories shown in the figure. Among nine mitogenomic features, GC content and polar amino acid content showed no significant overall correlation with gene rearrangement (Spearman’s ρ, WG: p > 0.05), although scattered significant correlations were observed in a few isolated PCGs (GC content: atp8, cox3, and rrnS; polar amino acid content: cytb, nad1, and nad4) (Figure 6B). Highly rearranged mitogenomes exhibited significantly decreased AT skew (except atp8 and cox2), increased GC skew (except cox1–3 and nad3), elevated Ka (Kn) and Ka/Ks values (except atp8 and nad3), and significantly varied mitogenome size (p < 0.001), gene length variation in ten genes was accordingly significant (atp6/8, cox3, nad2–5, nad 4L, rrnS, and rrnL), with higher counts and proportions (p < 0.001) of large NCRs (Figure 6B). Discussion ORI events arose independently at least three times in Sabellida Our analyses off 38 Sabellida mitogenomes, including 20 newly sequenced here, revealed that 28 species spanning 13 genera (6 of Serpulidae and 7 of Sabellidae) exhibited significant PCG GC skew elevation above the threshold (Figure 5B), with uniformly flipped (type I) or partially flipped (types II and III) at per-PCGs level (Figure 4). This provides a compositional indicator of ORI events. We further identified three replication-associated conserved motifs from a discovery training set comprising the control regions of circularized Sabellida mitogenomes. Among them, motif 1 is uniquely annotated with GO term DNA replication initiation, exhibits strong strand bias tightly coupled with GC skew, achieves the most stable high-quality hits across phylogenetically diverse Sabellida mitogenomes, especially in circularized ones (Figure 3), and is predicted to form a stem-loop structure across multiple species (Figure S4). We therefore focus on motif 1 as the primary structural indicator of ORI events. Contrary to prior predictions to locate conserved motifs on the L-strand, i.e., minus (opposite to cox1 orientation) and plus (same as cox1 orientation) strand for mitogenomes with and without flipped GC skew, respectively,13 we found these conserved motifs consistently on the H-strand (plus/minus strand for mitogenomes with/without flipped GC skew, respectively) (Figures 3D and 3G). This further supports their role in DNA primase recognition and the initiation of L-strand synthesis as OL, a short element whose reported size closely matches motif 1 identified here (∼30 vs. 21 bp).15,16 Although short-read sequencing and reference-guided assembly can complicate the recovery of fine NCR structures, motif identification appears robust across mixed-method dataset (Table S3), and was cross-validated against the largest NCR assemblies (Figure S5). Integrating with the GC skew elevation, motif 1 orientation and phylogeny, we inferred that ORI events arose independently three times within Sabellida: ORI i at the common ancestor of Serpulidae (clade A), ORI ii in the genus Laonome (clade BII), and ORI iii at the common ancestor of Sabellinae (clade D) (Figure 5A). The observed four GC skew flipped patterns are driven by the re-location of replication-associated motif 1, and modulated by independent gene inversions. In taxa with no GC skew flip (type N.A.), motif 1 (OL) resides on the complementary strand of all genes. Replication thus initiates from OH on the opposite strand, with the coding strand serving as the leading strand. All 13 PCGs remain double-stranded during replication, protected from spontaneous adenine-to-guanine deamination and resulting in a consistent negative GC skew (Figure 5C). In type I GC skew flip, motif 1 inverts to the coding strand, shifting all PCGs to the lagging strand, where single-strand exposure drives a uniform positive GC skew (Figure 5D). In type II, motif 1 inverts to the ancestral coding strand, combined with independent gene inversions that return a subset of genes to the leading strand. These genes escape the ORI effect and retain a negative GC skew, while the non-inverted genes (e.g., cox1–3, atp8, and nad3) remain on the lagging strand and a positive GC skew (Figure 5E). Conversely, type III retains the ancestral orientation of motif 1. While sharing an identical gene strand distribution with type II, type III exhibits a complementary flipped GC skew subset in which genes flipped in type II switch back to a negative GC-skew, and vice versa (Figure 5F). However, the phylogenetic position of this type within a motif 1-inverted clade (all congeners show type I or II) supports an ancestral motif 1 inversion (ORI event) in Sabellinae (Figures 5A and 5B). We propose a two-step process: an initial ORI event (ORI iii) at the common ancestor of Sabellinae, followed by a second ORI event that restored the motif to its original strand. The second event reversed the GC skew flip induced by ORI iii, producing a GC skew pattern opposite to that of type II mitogenomes. Three of the five genes showing a negative GC skew in type III remain significantly elevated above the threshold (Figure 4), suggesting that the second ORI may be relatively recent, with the negative GC skew not yet having accumulated beneath the baseline typical of type N.A. taxa. This two-step ORI event process is analogous to that reported in Isopoda.12 ORI-affected Sabellida mitogenomes exhibit biased sequence compositions During mitochondrial replication, prolonged single-strand exposure on the lagging strand drives a consistent adenine-to-guanine mutation bias, especially at third codon positions with minimal functional constraints.19 This trend is prominently observed in Serpulidae (Figure 2A), where an ancestral ORI event shifted all PCGs to the lagging strand. Mechanistically, the ORI-mediated switch of all genes to the lagging strand exposes adenine to deamination during replication, converting it to hypoxanthine (hX). Since hX pairs with cytosine, the process ultimately leads to the adenine-to-guanine substitution on the original strand.19 The shifts in synonymous codon usage may be accompanied by increases in the relative expression abundance of tRNA transcripts carrying cytosine at the third position of the anticodon. Such coadaptation has been hypothesized to enhance translational efficiency and accuracy, thus providing a fitness advantage in gene expression.29 A distinct shift in amino acid frequency is observed in the PCGs of ORI taxa, most prominently in those with a uniform (type I) GC skew flip (Figure 2B). The shift is most pronounced in hydrophobic residues (clusters A, B, and D), while hydrophilic residues (cluster C) remain conserved. This pattern deviates from the prediction that elevated GC skew and reduced AT skew should broadly alter the frequencies of amino acids encoded by guanine- and thymine-rich codons (C, V, F, G, and W) and by cytosine- and adenine-rich codons (Q, T, P, H, N, and K). Instead, only 5 of the 11 corresponding amino acids showed such changes. This discrepancy likely reflects functional constraints: hydrophobic residues drive protein folding and structural stability through their sequestration in protein cores, while hydrophilic residues mediate critical functions of proteins via forming hydrogen bonding, such as enzyme catalysis and ligand binding.30 The strong purifying selection observed in PCGs likely acts to preserve critical hydrophilic residues to maintain functional integrity, thereby allowing greater accumulation of hydrophobic substitutions that modulate structural flexibility. Mitochondrial protein variation is pronounced in ORI lineages, particularly in atp8, whose elevated nonsynonymous substitution rate is consistent with its most relaxed purifying selection (Figure 6A). In these lineages, atp8 proteins exhibit high sequence variability, reduced protein length, or a truncated initiation motif. Despite this variability, its N-terminal transmembrane helical domain remains conserved (Figure S1), suggesting that core structural or functional integrity is likely maintained. Confounding effects of ORI events on Sabellida phylogenetic reconstruction Our expanded mitogenomic dataset recovered artifactual topologies that conflict with nuclear gene evidence, notably across ORI-affected Serpulidae and Sabellinae (Figure 4). These artifactual relationships correspond to unresolved clades in prior studies,9,23 suggesting that ORI events may represent a potential confounding factor in mitogenomic phylogenetic reconstruction of Sabellida. An ORI event has been reported to interfere with phylogenetic reconstruction by inducing LBA effects14 or promoting attraction between lineages with homoplastic nucleotide skewness.12,31 ORI-induced strand-switching prolongs single-strand exposure and leads to the elevated substitution rates observed in ORI-affected genes (Figure 6A), which can generate long branches susceptible to phylogenetic artifacts.13,18 Although LBA has been hypothesized as a potential cause,11,13 our empirical analyses, including substitution saturation and assessments of exceptionally long branches, did not support LBA as the sole or primary driver of this topological ambiguity (Figure S6 and Table S11). The artifacts are more plausibly explained by homoplastic GC skew in lineages that have undergone ORI events, as proposed by Jakovlić et al.12 In Sabellida, the combined effects of ORI events and gene strand distribution produce strong, strand-specific mutation pressures that generate four distinct GC skew flip types. This mechanistic framework can explain the primary phylogenetic artifact: Serpulidae and clade IIIa of Sabellidae were artifactually grouped due to their homoplastic type I GC skew flip; this initial grouping subsequently clustered with Sabellinae lineages with partial GC skew flip (type II and III), forming a larger artifactual clade (Figure 4). Additionally, the weak support for the placement of Laonome in clade IV and Bispira sp. 2 in clade IIIb likely arises from their heterogeneous GC skew flip types within these clades, constituting another form of phylogenetic artifact associated with ORI-related processes. To mitigate the phylogenetic artifacts, we compared several compensatory strategies, including concatenating nuclear and mitogenome matrices, refining partitioning, and excluding third codon positions; however, none of these treatments resolved the artifactual clustering in our phylogeny (Figures S9–S11), leaving ORI-induced homoplastic GC skew as the most plausible remaining driver. Although the two-loci nuclear dataset recovered monophyletic Sabellidae and Serpulidae, it provided limited resolution for interspecific relationships and yielded markedly low nodal support for shallow-level relationships (Figure 4). On the other hand, the coalescent-based species tree inferred via the hybrid-weighted model in ASTRAL successfully recovered Serpulidae and Sabellidae as monophyletic, offering the most effective mitigation of ORI-driven artifacts. This approach assigns higher weights to nuclear gene trees with shorter branches, enabling the stable nuclear signal to dominate deeper nodes while simultaneously retaining the informational density of mitochondrial trees for resolving closely related taxa with homoplastic compositions. Thus, to extract reliable phylogenetic signal from ORI-affected mitogenomes, the most effective strategy involves integrating conserved nuclear loci (e.g., 18S and 28S) within a coalescent-based framework, such as ASTRAL, rather than relying on direct concatenation, which is more susceptible to compositional bias. Notably, when the ORI event is ancestral to the entire focal clade (e.g., Serpulidae), the resulting compositional homogeneity could function as a shared derived trait. In such cases, reliable internal phylogenetic inference can be performed within that clade only. ORI events show correlation with mitochondrial gene rearrangements Mitochondrial gene rearrangements have been attributed to a complex interplay of molecular drivers27,32,33,34 and ecological factors.14,35 In annelids, molecular drivers are considered predominant, with reduced mitochondrial efficiency acting as an underlying cause of both gene rearrangements and associated sequence and structural variation.10 Our findings in Sabellida further suggest that ORI events are consistently accompanied by mitochondrial gene rearrangements, and both are intrinsically linked to reduced mitochondrial efficiency and the observed mitogenomic variability. The energy-conserving lifestyles of Sabellida (e.g., tube-dwelling and sessile filter feeding) likely reduce both the demand for high-efficiency oxidative phosphorylation and the selective constraints acting upon it. Consistent with the significantly elevated Ka/Ks values observed in ORI-affected PCGs (Figure 6A) and highly rearranged mitogenomes (Figure 6B), this relaxation of selection may permit a decline in mitochondrial efficiency, which in turn increases the propensity for replication errors and structural instability, thereby creating conditions permissive for gene rearrangements and ORI events.12 Mitogenomes with extensive gene rearrangements also show the strongest correlation with ORI-associated molecular features, particularly with significantly elevated Ka and GC skew (Figure 6B), and those with significantly elevated PCG GC skew exhibit significantly greater gene order deviation from the ground pattern (Wilcoxon test, p = 2.25 × 10−3; Table 3). Given that the mitochondrial control region serves as both the essential replication origin and a rearrangement hotspot,34,36 ORI events and gene rearrangements are likely closely linked. The structural consequences of their co-occurrence can be reflected in increased mitogenome size and increased frequency of large NCRs (Figure 6B), although the causal direction between the two remains an open question. These rearrangements carry significant phylogenetic signal (Blomberg’s K and Pagel’s λ; Table S13) and can yield phylogenetically informative patterns. For instance, extensive gene inversions coupled with a shared “(-rrnL)-(-rrnS)-cox1” block indicate a close affiliation among four genera (Sabella, Bispira [except B. melanostigma], Sabellastarte, and Branchiomma), which aligns with the species tree (Figure 5A, clade D) and published phylogenies.7,37,38 However, gene arrangements generally provide lower resolution than sequence-based methods, partly because major lineages in Sabellida are separated by deep nodes carrying dense tandem duplication random loss (TDRL)-dominated rearrangements (Figure 5A), which disrupt the continuity of gene order changes across clades and limit cross-lineage comparability. This limited resolution is compounded by the near-saturation of rearrangements relative to the ground pattern (Figure S12). Therefore, despite its potential, the utility of gene order data remains largely confined to shallow, species- or genus-level resolutions. Further exploration is needed to rigorously define its reliable applications within a robust phylogenetic framework. In summary, our study supports the occurrence of ORI events in Annelida, a major metazoan phylum, by providing the first structural evidence (conserved motif inversions) and corroborating the inferences based on GC skew. We revealed at least three independent ORI events within Sabellida, corresponding to Serpulidae, the subfamily Sabellinae, and the genus Laonome. These events are associated with pronounced, multi-level mitogenomic variation in synonymous codon usage, hydrophobic amino acid composition, and protein sequences. Such compositional shifts can disrupt phylogenetic reconstruction, often introducing systematic errors associated with exceptionally long branches. Nevertheless, mitogenomes remain a viable phylogenetic resource for Sabellida under appropriate strategies, especially coalescent methods integrating nuclear markers for ORI-affected lineages. Notably, clades with a shared ancestral ORI event (e.g., Serpulidae) exhibit internal compositional homogeneity, enabling reliable mitogenomic phylogeny within the group. We propose that ORI events, reduced mitochondrial efficiency, and elevated gene rearrangement are intercorrelated features accompanying the extensive architectural diversity in these mitogenomes, and that these distinct gene arrangements can serve as complementary markers for resolving phylogenetic relationships within specific Sabellida lineages. Limitations of the study Several limitations warrant acknowledgment. First, although the evidence presented here demonstrates that ORI events, altered compositional skew, phylogenetic artifacts, and extensive gene rearrangements strongly co-occur in Sabellida mitogenomes, their causal direction remains largely unresolved. Whether ORI events act as the initial structural perturbation in the control region to trigger progressive mitogenome-wide rearrangements, or conversely, extensive gene rearrangements disrupt the control region and precipitate ORI events, is still open. Second, as the dominant source of current mitogenome assemblies, next-generation sequencing (NGS) inherently struggles to resolve fine-scale NCR structures and recover complete control regions, thereby impeding the detection of ORI events through the identification of structural inversions in the control region. Therefore, our motif-based approach remains the most practical current option, though direct physical replication mapping would further substantiate these structural inferences. Third, limited within-genus sampling hinders analysis of compensatory mutations of the identified motifs among closely related Sabellida species, which would reveal structural constraint of these motifs independent of functional annotations and support their role as ORI indicators. Nevertheless, multiple occurrences of ORI events with distinct GC skew subtypes make Sabellida an ideal model taxon for ORI research. Our work prompts two further questions: (1) When did ORI events arise in Sabellida, and what environmental or geological drivers were involved? (2) Did these events contribute to the high diversity of this group? Addressing these will require efforts in three directions: first, expanding mitogenomic sampling across more representative Sabellida genera and systematically screening for ORI signatures to delineate the full evolutionary scope of each ORI; second, constructing a robust, time-calibrated phylogenomic reference tree to establish the chronology of ORI events; and finally, conducting integrated multi-omics comparisons between lineages with and without ORIs to evaluate their functional and evolutionary consequences. Resource availability Lead contact Further information and requests can be directed to the lead contact, Dr. Yanan Sun (sunyn@qdio.ac.cn). Materials availability This study did not generate new reagents. Data and code availability •The 18S/28S rRNA and mitochondrial genome sequences supporting this study have been deposited at GenBank under the accession numbers listed in Table 1. The 18S/28S rRNA and mitochondrial genome sequences supporting this study have been deposited at GenBank under the accession numbers listed in Table 1. •The raw sequencing reads and corresponding BioSample metadata are available at NCBI BioProject: PRJNA1394476. The raw sequencing reads and corresponding BioSample metadata are available at NCBI BioProject: PRJNA1394476. •Figures S1–S12 and Tables S1–S15 are included in Document S1 (supplemental information). Additional supplementary datasets and analysis files generated in this study are available on Mendeley Data: https://doi.org/10.17632/b3j79s2wj7.1. Figures S1–S12 and Tables S1–S15 are included in Document S1 (supplemental information). Additional supplementary datasets and analysis files generated in this study are available on Mendeley Data: https://doi.org/10.17632/b3j79s2wj7.1. •We have no unreported custom computer code or algorithms. We have no unreported custom computer code or algorithms. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments This work was financially supported by the Shandong Provincial Natural Science Foundation (grant no. 2023HWYQ-101), the Natural Science Foundation of China (grant no. 42306132), and the Taishan Scholars Program (grant no. tsqn:202306288). This work was supported by the Oceanographic Data Center, Institute of Oceanology, Chinese Academy of Sciences. We are indebted to three anonymous reviewers for their constructive criticisms and helpful comments. We are grateful to Juhao Wang, Dr. Yanjie Zhang, and Yuxue Ju for their assistance in sample collection. Author contributions Conceptualization, R.Z., Y.S., and K.X.; resources, R.Z., Y.S., Q.B., X.W., and K.X.; methodology, R.Z. and Y.S.; formal analysis, R.Z.; data curation, R.Z.; visualization, R.Z.; writing – original draft, R.Z. and Y.S.; writing – review and editing, R.Z., Y.S., and K.X.; supervision, Y.S. and K.X.; funding acquisition, Y.S. and K.X. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERBiological samples20 Sabellida specimensThis studyNCBI BioProject: PRJNA1394476; NCBI BioSample: SAMN54367504–SAMN54367523Critical commercial assaysDNeasy Blood & Tissue KitQiagenCat#69504Deposited dataNewly sequenced Sabellida mitochondial genomesThis studyGenBank: PX832396–PX832397, PX939015–PX939027, PX970953–PX970960 (Refer to Table 1)Newly sequenced Sabellida 18S rRNA genesThis studyGenBank: PX778526–PX778545 (Refer to Table 1)Newly sequenced Sabellida 28S rRNA genesThis studyGenBank: PX778546–PX778565 (Refer to Table 1)Raw reads of Sabellida samplesThis studyNCBI BioProject: PRJNA1394476; NCBI SRA: SRR36695159–SRR36695178Publicly available mitogenome sequences (135) of LophotrochozoanNCBI GenBankRefer to Tables S1 and S2Publicly available mitogenome (24), 18S (23), and 28S (23) sequences of Sabellida and SpionidaeNCBI GenBankRefer to Table S3Additional supplemental materialThis studyMendeley Data: https://doi.org/10.17632/b3j79s2wj7.1Software and algorithmsFastp v.0.20.1Chen et al.39https://github.com/OpenGene/fastpTrimmomatic v.0.33Bolger et al.40https://github.com/usadellab/TrimmomaticMitoFinder v.1.4.1Allio et al.41https://github.com/RemiAllio/MitoFinderMEGAHIT v.1.2.7Li et al.42https://github.com/voutcn/megahitMITOS2 v.2.1.8Donath et al.43https://usegalaxy.eu/?tool_id=mitos2NOVOPlasty v.4.3.5Dierckxsens et al.44https://github.com/ndierckx/NOVOPlasty/releasesOpen Reading Frames FinderNCBI GenBankhttps://www.ncbi.nlm.nih.gov/orffinder/HMMER v.3.0Finn et al.45https://github.com/EddyRivasLab/hmmerDeepTMHMM v.1.0Hallgren et al.46https://dtu.biolib.com/DeepTMHMMARAGORN v.1.2.41Laslett and Canback47https://github.com/lab7arriam/aragorn/blob/main/aragorn1.2.41.c.cBarrnap v.0.9Torsten Seemannhttps://github.com/tseemann/barrnapSPAdes v.4.0.0Prjibelski et al.48https://github.com/ablab/spades/releasesBLASTNAltschul et al.49https://blast.ncbi.nlm.nih.gov/Blast.cgi?PROGRAM=blastn&PAGE_TYPE=BlastSearch&LINK_LOC=blasthomeKaKs_Calculator v.3.0Zhang50https://github.com/Chenglin20170390/KaKs_Calculator-3.0ParaAT v.2.0Zhang et al.51https://github.com/Icrus-Dev/ParaAT-2.0-WindowsMEME v.5.5.8Bailey and Elkan52https://meme-suite.org/meme/tools/memeGOMO v.5.5.8Buske et al.53https://meme-suite.org/meme/tools/gomoFIMO v.4.11.2Grant et al.54https://meme-suite.org/meme/tools/fimoMAFFT v.7.505Katoh and Standley55https://mafft.cbrc.jp/alignment/software/MACSE v.2.06Ranwez et al.56https://github.com/ranwez/MACSE_V2_PIPELINESGblocks v.0.91bTalavera and Castresana57https://github.com/evolbioinfo/dockerfiles/blob/master/gblocks/v0.91b/DockerfiletrimAl v.1.2Capella-Gutiérrez et al.58https://github.com/inab/trimal/releasesPhyloSuite v.1.2.3Zhang et al.59; Xiang et al.60https://github.com/dongzhang0725/PhyloSuite/releasesModelFinder v.2.2.2Kalyaanamoorthy et al.61https://github.com/iqtree/iqtree2IQ-TREE v.2.2.0Minh et al.62https://github.com/iqtree/iqtree2MrBayes v.3.2.7aRonquist et al.63https://github.com/NBISweden/MrBayes/releasesASTRAL-IIIZhang et al.64https://github.com/smirarab/ASTRALFrevoluzer distmat v.0.1.8Bernt et al.65; Bergeron et al.66https://usegalaxy.eu/root?tool_id=toolshed.g2.bx.psu.edu/repos/iuc/revoluzer_distmat/revoluzer_distmat/0.1.6+galaxy0TreeRex v.1.85Bernt et al.67https://siks.informatik.uni-leipzig.de/185-0-TreeREx.htmlR v.4.3.2The R Project for Statistical Computinghttps://github.com/R-project Experimental model and study participant details Wild-collected Sabellida specimens (20 species; Table 1) were used in this study. Fan worm individuals were scraped off using a stainless-steel spatula from the intertidal zone along the Chinese coast. Specimens were carefully removed from their tubes and immediately preserved in 95% ethanol. Genotype, age/developmental stage, and sex were not determined for the wild-collected specimens, and no laboratory maintenance or husbandry was performed. These species are neither endangered on the CITES catalog nor collected from a natural reserve, so no specific collection permissions or licenses were required. All collection and sequencing work was strictly executed under local legislation and related laboratory regulations to protect wild resources. No laboratory model organisms, cell lines, or human participants were used. Ethical approval was not applicable, as this study involved only wild-collected invertebrate specimens. As this study focuses on species-level phylogenetic and morphological analyses, no experimental grouping or sex/gender-related comparisons were performed; thus, sex-associated analyses were not applicable. Further collection and preservation procedures are provided in Method Details. Method details Sample collection, DNA extraction and genome skimming Twenty specimens belonging to eight genera of Sabellidae and five genera of Serpulidae were collected along the Chinese coast from 2014 to 2025 (Table 1). Specimens were fixed and preserved in 95% ethanol, and deposited in the Marine Biological Museum of the Chinese Academy of Sciences (MBMCAS), situated within the Institute of Oceanology, Chinese Academy of Sciences (IOCAS), Qingdao, China. Genomic DNA were extracted with DNeasy® Blood and Tissue kit (Qiagen Inc., Valencia, CA) following the manufacturer’s protocol. The animal whole-genome library preparation and whole-genome sequencing (WGS) were carried by Novogene Bioinformatics Technology Co., Ltd (Beijing, China). Paired-end libraries were with an average insert size of 300 bp were constructed and sequenced on the Illumina Novaseq X plus platform to generate approximately 10 Giga bases (Gb) of raw sequencing data. Fastp v.0.20.139 and Trimmomatic v.0.3340 were applied to remove adapters and low-quality sequences in raw reads. Mitochondrial assembly and annotation Trimmed reads were assembled using MitoFinder v.1.4.141 with MEGAHIT v.1.2.742 assembler and 39 published mitogenomes of Lophotrochozoan as reference of contig identification (Table S1). Assembled mitochondrial contigs were annotated using MITOS2 v.2.1.8 on EU Galaxy server (https://usegalaxy.eu/) with code table Invertebrate (5)’ and reference ‘RefSeq89 Metazoa’.43 For circularization, additional assemblies were run in NOVOPlasty v.4.3.544 using the cox1 fragment extracted from the MitoFinder assembly of the same specimen as a seed. To minimize seed dependence for a de novo assembly, extension was initiated from short k-mers (27–39 bp) of the seed by setting “Extend seed directly = no”. Circularized contigs replaced their linear counterparts and were re-annotated using MITOS2. Since MITOS2 failed to annotate atp8 of serpulids, we performed hidden Markov models (HMM) searches following Sun et al. 2021.11 All possible open reading frames (ORFs) were predicted by Open Reading Frames Finder (https://www.ncbi.nlm.nih.gov/orffinder/), translated with the invertebrate mitochondrial genetic code, and searched against an atp8 HMM profile derived from 135 Lophotrochozoan mitogenomes (Table S2) using HMMER v.3.0.45 HMMER searches were conducted with all heuristic filters were turned off (-max) to accept low score matches. The highest score ORFs were checked for the transmembrane helices of a putative atp8 by TMHMM-2.0 and DeepTMHMM.46 Supplementary annotation of tRNAs and rRNAs was conducted with ARAGORN v.1.2.4147 and Barrnap v.0.9 (https://github.com/tseemann/barrnap), respectively. All annotations were manually examined and curated in Geneious Prime 2024.0.5. The sequencing technology, assembly method, and circularization status for each newly sequenced mitogenome are summarized in Table S3. Assembly of 18S rRNA and 28S rRNA genes De novo assembly of the trimmed raw reads was performed with SPAdes v.4.0.0,48 followed by the filtration for eukaryotic rRNA genes (18S and 28S rRNA) with Barrnap v.0.9. The genomic origin of remaining contigs with a minimum sequence coverage of 150× was confirmed by BLASTN alignments.49 All sequences generated in this study (18S, 28S and mitogenomes) have been deposited in GenBank (accession numbers listed in Table 1). Sequence composition, codon usage, and evolutionary selection analyses For each mitogenome, GC content was calculated by (G + C)/(A + T + G + C) × 100%, with AT and GC skew were calculated by (A-T)/(A + T) and (G-C)/(G + C), respectively.18 We calculated the GC skew over all PCGs as the primary indicator for ORI events, which is directly comparable regardless of NCR completeness. To provide a quantitative framework for identifying significant GC skew shifts, we adopted the 2 × SD approach of Jakovlić et al. 202112: Using the PCGs GC skew of taxa where all 13 PCGs exhibit negative GC skew as baseline, we set the threshold for “significantly elevated GC skew” at the baseline mean +2 × SD (−0.17) (Table S4). The relative synonymous codon usage (RSCU) value of each mitogenome was calculated by RSCUij=xij1ni∑j=1nixij, where xij is the actual occurrence of the jth codon for the ith amino acid, 1ni∑j=1nixij is the average occurrence of all alternative codons for the ith amino acid.68 Ka and Ka/Ks values of PCGs were calculated in KaKs_Calculator v.3.050 and ParaAT v.2.051 with automated model averaging (MA). Ka/Ks values were interpreted as follows: >1 indicating positive selection, = 1 indicating neutral evolution, and <1 indicating purified selection.69 Non-coding substitution rates (Kn) of two rRNA genes (rrnS and rrnL) were calculated in KaKs_Claculator v.3.0. Identification of conserved sequence motif in mitochondrial control region To identify replication-associated motifs across Sabellida mitogenomes, we constructed a dataset (“All”) comprising 38 available Sabellida species (20 newly sequenced and 18 retrieved from GenBank). Six species of Spionidae, the most closely related family to Sabellida,8,70 were selected as an outgroup (Table S3). Within this dataset, 16 mitogenomes (9 newly sequenced and 7 retrieved) were confirmed as circularized and separately designated as the “Circularized only” dataset (Table S5). The prevalence of multiple large (>100 bp) non-coding regions (NCRs) in Sabellida mitogenomes hindered the identification of functional control regions.11 Therefore, we constructed a subset from the “Circularized only” dataset comprising species with ≤3 large NCRs (except Hydroides norvegica with 6 and Spirobranchus sp. with 7, Table S5). Spirobranchus giganteus, although circularization was not confirmed, was additionally included because its control region showed continuous coverage.27 For each mitogenome in the subset, the largest NCR was extracted to search for conserved sequence motifs on both strands (-revcomp) using MEME v.5.5.8, with an E-value threshold of 0.05.52 The three conserved motifs with highest E-value were assigned a series of GO terms using GOMO.53 We subsequently searched for the three motifs across the two datasets (“Circularized only” and “All”) using FIMO v.4.11.2, applying a p-value threshold of 1 × 10−4.54 To define the single putative functional motif from multiple copies per mitogenome, we applied two criteria: (1) located in any NCR. (2) found on the same strand as cox1 (+) when cox1 exhibits positive GC-skew, and on the opposite strand (−) when cox1 exhibits negative GC-skew. cox1 was chosen as the reference for both motif orientation and preliminary ORI screening by GC skew, because it has the highest sequence conservation and is conventionally designated as the start of the plus strand.14 Copies were evaluated in ascending order of their p-value (from lowest to highest) and classified into three categories based on the first valid copy: “Precise hit”, the most significant copy (lowest p-value) was confirmed as the putative functional motif; “Ambiguous hit”, a putative functional motif was confirmed, but from a less significant copy (non-minimal p-value); “Failed”, no functional motif was confirmed. For species with multiple contigs, each contig was searched independently, and all motif candidates were then pooled together and ranked by p-value as a single set, following the same criteria described above. To assess the robustness of motif identification and strand assignment across assembly strategies, we independently applied the same FIMO pipeline to the largest NCR of each NOVOPlasty (de novo) and MitoFinder (reference-guided) assembly for the 9 species with newly assembled and circularized mitogenomes (Table 2). Phylogenetic analyses Phylogenetic analyses included the same 38 Sabellida mitogenomes in the “All” dataset, with Spionidae species as outgroup (Table S3). Nucleotide sequences of two nuclear genes (18S, 28S) and 15 mitochondrial genes (13 PCGs, rrnS and rrnL), were used in the phylogenetic analyses. For species with complementary contigs, genes were extracted from the respective contig and included as a single taxon. Each gene was aligned separately using MAFFT v.7.505 under “L-INS-I” strategy.55 To exclude poorly aligned regions, the PCG alignments were subsequently optimized with MACSE v.2.0656 and trimmed with Gblocks v.0.91b.57 The rRNA alignments were trimmed with trimAl v.1.2.58 The trimmed sequences were concatenated into two matrices partitioned by genes using PhyloSuite v.1.2.359,60: matrix 1 including all mitochondrial genes, and matrix 2 including 18S and 28S genes only. Both data matrices were of 100% completeness and were analyzed independently. ModelFinder v.2.2.261 identified the best-fit edge-linked partition model71 under Bayesian information criterion (BIC) (Table S6). Maximum-likelihood (ML) trees were inferred using IQ-TREE v.2.2.062 with 5000 ultrafast bootstrap (UFBoot) replicates72 and 1000 SH-like approximate likelihood ratio test (SH-aLRT) replicates.73 Bayesian inference (BI) was performed in MrBayes v.3.2.7a,63 running 1,500,000 generations with sampling every 500 generations, burn-in was set to 25%. Substitution saturation and potential long-branch attraction (LBA) of the generated ML trees were evaluated in PhyloSuite v.1.2.3.60 Substitution saturation was quantified as the R-squared between patristic and pairwise distances.60 The long-branch score of taxon i was calculated using the formual LBi=(PDi¯PDall¯−1)×100, where PDi¯ and PDall¯ denotes the mean pairwise patristic distance of taxon i and all taxa, respectively.74 Taxa with branch lengths exceeding 20 times the median were flagged as highly susceptible to LBA artifacts.60 The species tree was inferred from the single gene tree derived from the fifteen mitochondrial and two nuclear genes. Dataset of each gene were processed following the phylogenetic analyses described above. ML analysis for each gene was conducted by IQ-Tree v.2.3.6 with automatic model selection62 (Table S7). The species tree was reconstructed using hybrid-weighted ASTRAL v.1.10.2.375 implemented in ASTRAL-III,64 with branch support estimated as local posterior support (LocalPP). If a clade of interest was not recovered in the preliminary outputs, it would be enforced to monophyletic and computed the hidden support (localPP). Gene order analyses A gene order matrix was linearized to start with cox1. When complementary contigs were present for the same mitogenome, each was oriented by reference to the complete gene order of closely related taxa before merging. The matrix was then imported to CREx and “distmat” tools65 on the EU Galaxy server to calculate the common interval distances66 with four pairwise rearrangement scenarios considered: I (Inversion), T (Transposition), iT (inverse Transposition) and TDRL (Tandem Duplication Random Loss). Parsimonious gene rearrangement scenarios were mapped to the species tree using TreeREx67 to reconstruct ancestral gene orders. Three consistency methods (0-consistentency; k-consistency and parsimonious k-consistency) were applied in the TreeREx analyses. Only PCGs and rRNA genes were included in CREx and TreeREx analyses. Statistical exploration between gene rearrangements and mitogenomic features We assessed statistical correlation between gene rearrangement levels and nine mitogenomic features: GC content, polar amino acid content, AT skew, GC skew, non-synonymous substitution rate (Ka), non-synonymous to synonymous rate ratio (Ka/Ks), gene length variation, and frequency and proportion of large NCR (>100 bp). To quantify gene length variation, we used its log-transformed absolute value. In our dataset, all linear contigs contain a complete set of 37 mitochondrial genes. For these contigs, the NCR fragments at the 5′ and 3′ termini most likely represent a single NCR split by an assembly breakpoint, and were counted as one when calculating NCR frequency and proportion. This rule does not affect any other analyses. For species with multiple contigs, whole-mitogenome compositional parameters were computed as the length-weighted average of the contigs. The mitogenome of fabriciid M. occidentalis was used as both the benchmark of mitogenomic features and gene order in Sabellida. Gene rearrangement levels were quantified using pairwise common interval distances. Because the common interval distances were non-uniformly distributed, Spearman’s rank correlation was used to provide a robust, non-parametric assessment of the relationship between gene rearrangement and mitogenomic features. Statistical comparisons of GC skew, AT skew, Ka, Ka/Ks, and common interval distance to the ground pattern between ORI and non-ORI groups were performed using Student’s t-tests when normality held, and Wilcoxon tests otherwise. Quantification and statistical analysis Statistical analyses were performed using R (v.4.3.2). Data are presented as mean, median, and range (Table 3), mean +standard deviation (SD) (Figure 6A), or boxplots showing medians, means, interquartile ranges, whiskers, and outliers (Figure S2). Significantly elevated GC skew against the baseline mean was defined as 2 × SD (Table S4). For pairwise group comparisons (Figure 6A and Table 3), normality was first assessed using the Shapiro-Wilk test. When both groups satisfied normality, Student’s t test was applied; otherwise, the Wilcoxon rank-sum test was used (Tables S14 and S15). Statistical correlations (Figure 6B) were assessed using Spearman’s rank correlation to accommodate the non-uniform distribution. Substitution saturation was quantified as the R2 between patristic and pairwise distances (Figure S6). Potential long-branch attraction was assessed by flagging taxa with branch lengths exceeding 20 times the median (Table S11). For all statistical analyses, significance levels were denoted as: NS, not significant; ∗, p < 0.05; ∗∗, p < 0.01; ∗∗∗, p < 0.001. Further statistical information can be found in the Method Details. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.117497. Contributor Information Yanan Sun, Email: sunyn@qdio.ac.cn. Kuidong Xu, Email: kxu@qdio.ac.cn. Supplemental information Document S1. Figures S1–S12 and Tables S1–S15 References Associated Data Supplementary Materials Document S1. Figures S1–S12 and Tables S1–S15 Data Availability Statement •The 18S/28S rRNA and mitochondrial genome sequences supporting this study have been deposited at GenBank under the accession numbers listed in Table 1. The 18S/28S rRNA and mitochondrial genome sequences supporting this study have been deposited at GenBank under the accession numbers listed in Table 1. •The raw sequencing reads and corresponding BioSample metadata are available at NCBI BioProject: PRJNA1394476. The raw sequencing reads and corresponding BioSample metadata are available at NCBI BioProject: PRJNA1394476. •Figures S1–S12 and Tables S1–S15 are included in Document S1 (supplemental information). Additional supplementary datasets and analysis files generated in this study are available on Mendeley Data: https://doi.org/10.17632/b3j79s2wj7.1. Figures S1–S12 and Tables S1–S15 are included in Document S1 (supplemental information). Additional supplementary datasets and analysis files generated in this study are available on Mendeley Data: https://doi.org/10.17632/b3j79s2wj7.1. •We have no unreported custom computer code or algorithms. We have no unreported custom computer code or algorithms. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.