Surficial sediments in Vaxön are taxonomically and functionally homogenous

Our study began with the collection of four surficial sediment cores, termed SP0, SP1, SP2, and SP4 (~7–8 cm deep) from the shoreline of Vaxön, an island in the Stockholm archipelago (12 August 2023; S1 Fig). Environmental DNA was extracted from sediments at 0, 2, 4, and 6 cm depths for metagenomic analysis. Across these depths, metagenomics revealed no significant changes in bacterial community composition or functional pathways (S2A- S2E Fig; p > 0.05). Actinobacteriota dominated the communities (18.1-38.5% of reads), followed by Pseudomonadota (17.0-25.0%), Bacteroidota (1.3-2.4%), and Planctomycetota (1.0-2.2%) (S2A Fig). Neither alpha- nor beta-diversity, assessed using Shannon indices and Euclidean distances, respectively, differed significantly with depth (p > 0.05; S2A, S2B Fig). Functional annotation of metagenomic reads based on KEGG pathways showed no depth-dependent clustering (p > 0.05, Kruskal-Wallis test; S2C, S2D Fig). Predominant pathways (log₁₀ GPM > 7) included ABC transporters, quorum sensing, DNA replication and repair, glycolysis/gluconeogenesis, fatty acid metabolism, and purine metabolism (S2E Fig). Shewanella spp. populations accounted for 0.04-0.21% of total bacterial reads (S1 Table), representing more than 2.3% of Gammaproteobacteria and ranking among the 100–300 most abundant genera detected (out of >13,000). However, Shewanella abundance did not differ significantly across sediment depths (S2F Fig; p > 0.05). Together, these data indicate that (i) bacterial community composition remains largely unaltered across the upper 6 cm of sediments, and (ii) Shewanella is not a particularly dominant taxon in the surficial sediment community.

A sympatric lifestyle within surficial sediments fosters genotypic divergence in Shewanella populations

From four sediment cores collected at the shoreline in Vaxön, we isolated 60 randomly selected coexisting Shewanella strains, i.e., sympatric (designated with the prefix ‘VAX’; S2 Table). We complemented this collection with 18 Shewanella strains obtained during previous isolation efforts in the same area in 2021 (prefix ‘SP’), as well as randomly selected allopatric strains from sediment (n = 19) and water (n = 15) samples collected from distinct locations in and around Stockholm between 2021 and 2023. These sites included Nynäshamn (n = 11; prefix ‘N’), Tyresö (n = 12; prefix ‘T’), Lidingö (n = 8; prefix ‘L’), and Hagaparken (n = 3; prefix ‘H’) (S1 Fig). In total, we analyzed 112 Shewanella whole-genome sequences, including previously sequenced strains [32,33]. Of these, 32 were complete genomes, all belonging to representative members of the S. baltica complex (S2 Table).

To assess the preliminary taxonomic diversity of the Shewanella isolates retrieved from Vaxön, comprising 78 isolates from the 2021 and 2023 sampling occasions combined, whole-genome sequences were submitted to the Type Strain Genome Server (TYGS) [35], which uses 70% digital DNA-DNA hybridization (dDDH) as the threshold for species designation. Based on this indicator, 47% of the isolates were affiliated with S. baltica (n = 37), 10% to S. vaxholmensis (n = 8), 5% to S. scandinavica (n = 4), and 1% to S. septentrionalis (n = 1), whereas a remarkable 32% could not be confidently assigned to any known species (n = 25), but exhibited the closest similarity to S. baltica (Fig 1A). Thus, ~ 95% of the Vaxön isolates in this dataset were taxonomically very closely related, with the remaining strains affiliated with S. hafniensis, S. oncorhynchi, and S. holmiensis (Fig 1A). Pairwise ANIb comparisons between all these strains exhibited values that ranged from 89.6% to 100%, with two clear ANIb gaps observed between the interactions S. oncorhynchi vs. S. hafniensis (91.0-93.2%) and S. hafniensis vs. S. baltica complex members (94-94.5%) (Fig 1A). Of note, we found a distinct pattern within the S. baltica complex characterized by a continuous bimodal distribution with 60% (3,743 pairwise comparisons) falling within the 94.5%-95.9% range, while 28% (1,748 comparisons) clustered between 95.9%-97.2% (Fig 1A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Genomic diversity of S. baltica complex genomes based on ANIb relatedness.

(A-C) Frequency distributions of pairwise ANIb comparisons (0.1% windows) for (A) strains recovered from Vaxön sediment cores in 2023 (n = 60) and from previous sampling occasions in the same region (n = 18); (B) allopatric strains retrieved from other regions in or around Stockholm city: sediments (n = 19) and water (n = 15); and (C) the entire dataset, including both our isolates (108 S. baltica complex members and 1 S. hafniensis, excluding S. oncorhychi) along the type strain S. baltica CECT 323^T (in red), and those available from NCBI GenBank (n = 66, shown individually in blue for comparison). In (A), dashed areas denote comparisons with the species most closely related to the S. baltica complex strains recovered in our study, i.e., S. oncorhynchi (n = 3) and S. hafniensis (n = 1). In (A-B), pie charts showing the affiliation of the isolates as determined by TYGS were added as insets. In (B), the brown frame highlights the ANIb comparisons between strains recovered from the sediment. In (C), the all-vs-all genome-to-genome distance calculator (GGDC) analysis is embedded within the ANIb comparison, showing that ANIb values corresponding to the 70% digital DNA–DNA hybridization (dDDH) species delineation threshold fall within 95.8–96.1%, thereby defining species boundaries within the species complex.

https://doi.org/10.1371/journal.pgen.1012306.g001

We extended our analysis to a regional set of 34 S. baltica complex strains isolated from sediment and water samples collected at distinct locations across Stockholm. Given their geographic separation from the strains presented above, we treated them as allopatric strains (S1 Fig). In this set, 88% of the strains (n = 29) could be unambiguously affiliated to S. baltica by TYGS, whereas 9% (n = 3) could not be affiliated to a known species, and 3% (n = 1) circumscribed within S. scandinavica (Fig 1B). Here, the ANIb distribution revealed a trimodal pattern with 15.4% of interactions (168 pairwise comparisons) falling within the 94.8%-95.5% range, 12.4% (135 pairwise comparisons) within 95.5%-96.0%, and 68.2% (743 pairwise comparisons) within 96.0%-97.0% (Fig 1B). The first two ANIb peaks reflected interactions between sediment-sediment or sediment-water genome pairs, whereas the highest density peak (96.0%-97.0%) was observed exclusively among water-water genome comparisons, suggesting a potential role of sediments in driving genomic diversification.

ANIb pairwise comparisons across the integrated Vaxön (sympatric) and Stockholm (allopatric) genome sequence dataset corroborate the observed bimodal pattern, with high average shared genome fractions (84.8 ± 3.63%; S3A Fig). The lower ANIb peak (94.5-95.6%) accounted for ~50% of comparisons (n = 6,383), mostly involving sediment-derived isolates, with the sole exception of S. septentrionalis, isolated from surficial shore water near the sediment layer [33]. Notably, no Shewanella genome sequences from NCBI GenBank fell within this range (reviewed on 31 May 2025, S3 Table), highlighting the novelty of this population. The higher ANIb peak (centered ~96.3%) encompassed 32% of comparisons (n = 4,148), predominantly S. baltica strains, closely overlapping publicly available S. baltica genomes (~96.5% ANIb; Fig 1C). This range, bordering the canonical ~96% species delineation threshold, was benchmarked against FastANI-based intra-specific comparisons within other well-represented Shewanella clades (S. algae, 219 genomes; S. xiamenensis, 98 genomes; S. oncorhynchi, 32 genomes; S3 Table), which displayed unimodal FastANI distributions with narrower dispersion and consistently higher intra-specific similarity (>97–98% FastANI). In contrast, S. baltica (including both our isolates and publicly available genome sequences) spanned a broader FastANI range, extending toward the species boundary zone, and reflecting the exceptional intra-specific diversity of S. baltica (S3B Fig). To define species boundaries within the S. baltica complex, we determined the ANIb values corresponding to the 70% dDDH species threshold, which fell within 95.8-96.2% (Fig 1C, inset). Of note, more than half of all pairwise ANIb values fell within 94.5-96.2%, a transition zone consistent with the typical ANI gap for species delineation [4]. Taken together, these patterns represent a possible case of ongoing sympatric speciation in sediments.

Multiple differentiated lineages coexisting in surficial sediments provide a snapshot on Shewanella speciation

To assess the extent of phylogenomic differentiation within our dataset, we performed pairwise ANIb and dDDH comparisons and reconstructed phylogenies based on the core genome and core proteome content. Genomic species boundaries were assessed based on current standards [34], using the following criteria: (i) ANIb ≥ 95.8-96.2% (all vs. all); (ii) dDDH ≥ 70% (all vs. all); and (iii) phylogenomic coherence between core genome and core proteome trees. All-against-all ANIb comparisons defined three main groups (G), namely, G1, comprising mostly S. baltica strains plus several unassigned strains; G2, closely related to S. scandinavica; and G3, related to S. vaxholmensis and S. septentrionalis (Fig 2A). ANIb values between groups showed clear boundaries supporting this clustering (<96%), with intergroup similarities ranging from 94.89-95.97% (G1 vs. G2), 94.50-95.91% (G1 vs. G3), and 95.10-95.71% (G2 vs. G3) (Fig 2A). Genome-wide differentiation within the S. baltica complex was assessed using SNP-based sliding window analyses, and pairwise p-distances ranged from 0 to 0.5 (Fig 2B). Values above 0.25 indicate high genomic differentiation between the defined groups, whereas values near 0 within G2 and G3 indicate close genomic relatedness and support the emergence of new genomic species clusters within these groups (Fig 2B), aligning with the ANIb group classification.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. ANI clustering, genome-wide, and pangenome analysis of the S. baltica complex from the Stockholm region.

(A) Heatmap of all-vs-all ANIb values for the S. baltica complex. The dashed red frame within G1 denotes the strains affiliated with S. baltica. (B) Heatmap showing pairwise genome-wide differentiation between strains, based on SNPs extracted from whole-genome sequence alignments. SNPs were grouped into overlapping 10 kb sliding windows with a 2.5 kb step size. For each window, the proportion of SNPs with discordant genotypes (p-distance) was calculated for all genome pairs, producing normalized values from 0 (identical) to 0.5 according to the genomic divergence. In (A) and (B), as the closest relative, S. hafniensis (strain SP1S1-9 and three additional sequences from NCBI GenBank, S3 Table) was added as an outgroup, represented in purple. For clarity, the plots indicate the previously characterized species within the complex (S. baltica, S. scandinavica, and S. vaxholmensis). (C) Combined core genome and accessory genome tree constructed using a 90% identity and 90% coverage threshold based 3,310 core gene and protein families from the 109 genomes (including S. baltica CECT 323^T as reference). For building the accessory tree, pairwise Euclidean distances between isolates were calculated, and an average-linkage (UPGMA) method was applied. Blue labels indicate genomes isolated from the water column. Within G2 and G3, branch colors indicate distinct species previously described, as well as the genospecies (“Gs.”). Species and genospecies were delineated as explained in the main text. Accessory gene groups (AGCs) are indicated by distinct numbers and colors. Lines link corresponding genomes between the core genome and accessory genome trees, illustrating their topological relationships. Numbers on the branches denote bootstrap support values after 1,000 replications. In all the plots, groups are differentiated by color, i.e., red for G1, yellow for G2, and green for G3. The grey shadow within G1 denotes incipiently divergent genomes that could not be affiliated to S. baltica or differentiate as putative genospecies.

https://doi.org/10.1371/journal.pgen.1012306.g002

Our phylogenomic reconstructions, based on both core genome and core proteome content, comprising 3,308 core family genes (3,399,949 bp) or proteins (1,131,112 aa) from a pangenome of 18,626 genes (representing ~75% of the core genes in each genome), were highly congruent with one another and their topology was consistent with inferences drawn from overall genome relatedness indexes (OGRIs) previously introduced (Figs 2C, S4A). In total, we identified 11 putative novel genospecies (referred to as ‘genospecies’ hereafter) within G2 and G3, specifically, 9 within G2 and 2 within G3 (Fig 2C). Each lineage was genomically differentiated both from one another and from the previously proposed species S. septentrionalis, S. vaxholmensis, and S. scandinavica [32,33]. Several strains from different time points (2021 - ‘SP’ prefix, and 2023 - ‘VAX’ prefix) were affiliated with the same species or genospecies, e.g., various S. vaxholmensis strains (G3) and diverse isolates belonging to a genospecies within G2, e.g., genospecies 11 (Fig 2C). This finding evidenced stability over time of divergent lineages in this sediment ecosystem. In contrast, strains showed no clear grouping by geography or sediment depth (Fig 2C), consistent with their widespread distribution, and the taxonomic and functional homogeneity inferred from our metagenomic analyses. Thus, our phylogenomic analyses show that sympatric diversification within the S. baltica complex might be driven by the close coexistence of strains in sediments, which could be promoted by extensive lateral gene exchange in this habitat.

High rates of lineage-specific gene turnover and HGT drive accessory genome diversification in the S. baltica complex

To contextualize gene exchange in the S. baltica complex, we assessed the distribution of accessory genes across the phylogeny. We integrated an accessory genome tree derived from gene presence/absence profiles with the core genome phylogeny, which revealed distinct and group-specific accessory gene clusters (AGCs) which, overall, were congruent with the topology of the phylogenomic reconstructions (Fig 2C). Strains within G3 shared a highly similar AGC, whereas nearly every species and genospecies cluster within G2 exhibited a distinct AGC from one another. In contrast, G1 strains displayed a much higher diversity of AGCs, up to 16, which did not consistently align with the core genome tree topology (Fig 2C). The high number of AGCs observed in the S. baltica complex is consistent with its evidently open pangenome, encompassing up to 87% of the total pangenome and including over 15,000 accessory genes (Fig 3A), constituting ~25% of each genome. Despite G1 forming a cohesive cluster primarily composed of a single species, S. baltica, it harbored substantially greater accessory genome diversity than G2 and G3 combined, with Heap’s Law exponent (γ) fitted to the cumulative accessory genome curve of 0.55 for G1, compared to 0.47 for G2 and G3, confirming a more open pan genome in G1 (Fig 3A). Furthermore, the accessory genome of the S. baltica complex exceeded that of other Shewanella species such as S. algae and S. xiamenensis, and was comparable to that of S. oncorhynchi (S4B Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Gene gain, loss, and HGT patterns in the S. baltica complex pangenome.

(A) Accumulation curves of accessory genome content of the members of the S. baltica complex from this study (n = 109). Heap’s Law exponent (γ) was fitted to the cumulative accessory genome for each dataset. (B) Core genome tree, with branches colored according to the number of gene gain and loss events inferred using maximum parsimony. (C) Cumulative number of gene gain/loss events plotted against cumulative branch length, and (D) gene exchange parameters estimated by Panstripe. ‘Core’ reflects the association between gene gain/loss events and core phylogeny branch length, and ‘tip’ indicates gene exchange events occurring specifically at terminal branches. The p value within the ‘core’ and ‘tip’ boxes indicates the statistical significance of turnover events associated with the core and the tip in all the groups compared. In (A), (C), and (D), to minimize potential biases in comparisons, we merged G2 and G3 (n = 38), which, although evolutionary distinct, displayed significantly lower accessory content and rates of gene turnover compared to G1. (E) Number of HGT genes per genome (left bar) and the percentage of HGT genes relative to the total accessory genome of each genome (right bar) across the indicated groups. (F) Taxonomic classification of genes detected within RGPs, assigned using EggNOG, across the different groups. For (D) and (E), statistically significant differences are indicated by asterisks: p < 0.05 (*) and p < 0.001 (**).

https://doi.org/10.1371/journal.pgen.1012306.g003

We next studied gene turnover in S. baltica complex using the Panstripe pipeline [36]. Overall, the S. baltica complex exhibits 200-800 gene gain/loss events per genome (Fig 3B), with G1 showing significantly more events (352 ± 124) than G2 and G3 (250 ± 168; p < 0.05; Fig 3B), consistent with its broader accessory gene content (Figs 2C, 3A). Older (internal) and newer (external) lineages displayed similar rates of gene turnover along their corresponding phylogenetic branches (p > 0.05; Fig 3C), yet a pronounced ‘tip effect’ was detected across the complex, with turnover events concentrated on terminal branches, where younger lineages exchange genes more actively than older ones (p < 0.05; Fig 3D). This elevated turnover is also linked to core genome divergence (p < 0.001; Fig 3D), where G1 shows significantly higher core genome divergence from G2/G3 (p < 0.05; Fig 3D), reinforcing that genetically distinct lineages drive the greatest gene flux. To better understand this, we explored the regions of genome plasticity (RGPs), and observed that approximately 40% of accessory genes were located within putative RGPs, with each genome containing 140–380 putative HGT-derived genes (Fig 3E). Across all groups, the majority of RGPs were attributed to the family Shewanellaceae (58.7–75.6%), followed by diverse aquatic bacterial taxa including Vibrionales, Pseudoalteromonadaceae, Alteromonadaceae, and Oceanospirillales (S5B Fig). G1 contained significantly more HGT-derived genes (328 ± 65 per genome) than G2 and G3 (250 ± 65 per genome; p < 0.05; Fig 3F), consistent with its more open pangenome.

Plasmids can drive ecological adaptations and even speciation in bacteria [37]. Here, we recovered 21 complete circular plasmids (2,677–120,678 bp) from 32 complete S. baltica complex genomes, and 15 genomes carried 1–4 plasmids (S4 Table). Closely related plasmids isolated in 2021 and 2023 suggest plasmid stability in the sediment ecosystem (S5A Fig). A few plasmids clustered together based on FastANI relatedness, but these clusters spanned all strain groups (S5A Fig), indicating limited group-specificity. Plasmids were enriched in replication, mobilome, transcription, vesicular transport, and defense functions (S5B Fig), though most plasmid genes remain uncharacterized (S4 Table). Given their limited group specificity and functional profile, plasmid-borne genes are unlikely to be major drivers of accessory genome differentiation in the S. baltica complex. This indicates instead that horizontally acquired chromosomal genes represent the main source of genomic functional diversification, likely contributing to the ecological specialization of distinct lineages within the described groups through the acquisition of niche-specific traits.

Metabolic and ecological specialization differentiates groups within the S. baltica complex

We investigated genomic signatures of ecological specialization by characterizing the functional genomic repertoire of each phylogenetic group through a cluster of orthologous groups (COG) analysis. PCA of functional profiles revealed three distinct, well-separated clusters, indicating functional differentiation between groups (PERMANOVA, p = 0.001, Fig 4A). All groups shared a large core of 2,131 COGs, yet several exclusive COGs were found in each group (Fig 4B, S5 Table). G1 had the largest exclusive set with 228 COGs, consistent with the high dispersion observed in the PCA plot for this group, its large accessory genome, and its elevated gene turnover, followed by G2 with 36 COGs and G3 with 17 COGs (Fig 4B, S5 Table). Pairwise comparisons further showed 112 COGs shared exclusively between G1 and G2, 54 between G1 and G3, and only 7 between G2 and G3 (Fig 4B). Within G2, no significant functional differences were observed among constituent species or genospecies (PERMANOVA, p > 0.05, Fig 4A), whereas G3 exhibited significant different functional profile between its constituent species and genospecies (PERMANOVA, p = 0.001; Fig 4A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Metabolic specialization across the S. baltica complex.

(A) Principal Coordinate Analysis (PCoA) of COG functional profiles for all strains, colored by phylogenetic group (G1, red; G2, yellow; G3, green), with 95% confidence ellipses shown. Species and genospecies-level groupings are indicated by point shapes. Groups are significantly differentiated (PERMANOVA, p = 0.001). (B) Venn diagram depicting the distribution of COGs across the three groups. The exclusive COGs by group are listed in S5 Table. (C) COG category enrichment analysis showing significantly enriched functional categories (FDR-adjusted p < 0.05) for each group, with the total number of enriched COGs indicated above each panel. (D) RAST-SEED pathway-level enrichment analysis showing fold enrichment of significantly enriched metabolic pathways across groups (FDR-adjusted p < 0.05). The specific enriched COGs and SEED annotations by group are listed in S6 and S7 Tables. (E) Distribution of genes encoding reductases for alternative electron acceptors (AEAs) across all sequenced genomes, organized by phylogenetic groups. Presence/absence patterns are shown for DMSO (DmsA), nitrate (NapA orthologs α and β), nitrite (NrfA), thiosulfate (PhsA), sulfite (SirA), TMAO (TorA), and tetrathionate (TtrA) reductases. Proteins were grouped and color-coded according to which group of reductases they were affiliated by these functional categories as detailed in the legend. Genomes along the x-axis are ordered according to their ANIb relatedness.

https://doi.org/10.1371/journal.pgen.1012306.g004

While KEGG pathway analysis revealed no overall metabolic differences among groups, except for galactose biosynthesis and degradation pathways, which showed greater completeness in G1 (S6 Fig), group-specific functional enrichment at both the COG category and pathway levels (using the RAST-SEED annotation) revealed distinct metabolic signatures across the three groups (FDR-adjusted p < 0.05; Fig 4C, 4D; S6, S7 Tables). G1 was enriched in 59 COGs predominantly linked to amino acid transport and metabolism (n = 11) and carbohydrate transport and metabolism (n = 10), which at the pathway level were associated with significant enrichment in pyruvate metabolism, methionine degradation, and succinate dehydrogenase activity (Fig 4C, 4D; S6, S7 Tables). In contrast, G2 and G3 shared enrichment in nitrogen metabolism, pyruvate-alanine interconversions, and biosynthetic pathways for cell surface structures such as lipo-oligosaccharide core oligosaccharides (Fig 4D; S6, S7 Tables). Beyond this shared profile, each group showed distinct specializations, and G2 was enriched in 65 COGs associated with cell wall/membrane biogenesis (n = 13), transcription (n = 7), and signal transduction (n = 7), complemented by pathway-level enrichment in heavy metal resistance, CMP-N-acetylneuraminate biosynthesis, or arabinogalactan synthesis (Fig 4C,4D; S6, S7 Tables). G3 displayed the broadest enrichment with 106 COGs, dominated by cell motility (n = 32), signal transduction (n = 9), energy production (n = 8), and defense mechanisms (n = 6) with corresponding pathway-level enrichment in flagellar biosynthesis, type IV secretion systems, and restriction-modification systems (Fig 4C, 4D; S6, S7 Tables). Together, despite the isolation of strains from all three groups across all sediment depths and the lack of significant variation in community composition or bulk biological functions across the upper 6 cm of sediment, each group within the S. baltica complex exhibits distinct genomic signatures consistent with metabolic specialization, suggesting distinct ecological strategies within a shared habitat.

To explore these strategies, we next focused specifically on energy metabolism, examining genes encoding reductases for alternative electron acceptors (AEAs) to evaluate potential differences in respiratory capacities. BLASTp searches were performed using reference S. baltica complex reductase catalytic subunits for nitrate (NapA isoforms α and β), nitrite (NrfA), thiosulfate (PhsA), sulfite (SirA), tetrathionate (TtrA), dimethyl sulfoxide (DMSO; DmsA), and trimethylamine N-oxide (TMAO; TorA) against all sequenced genomes (Fig 4E, see Materials and Methods). Genes for nitrate, nitrite, tetrathionate, and thiosulfate reduction were conserved across all genomes (Fig 4E). We observed loss of TMAO reductase genes in distinct G1 strains, as well as acquisition of DMSO reduction genes in multiple G1 and G2 strains (Fig 4E; S8 Table). The sirA gene, encoding a sulfite reductase, was conserved in G3 (100%) but was only present in 56% of G2 members and 21% of G1 strains. Extension of this analysis to public G1 genome sequences revealed a similar proportion (26.5% of genomes carrying sirA), which indicates this is a globally extended characteristic within the G1 lineage. Synteny analysis showed that sirA, which is part of a large operon (sirABIGCDJKLM), is located in a RGP together with the transcriptional regulator sirR and the inversely transcribed sirEFG operon encoding a specific heme lyase required for SirA maturation [38] (S7 Fig), which altogether suggests a functional sulfite reduction gene cluster hereafter referred to as the sir cassette. We experimentally validated this genomic prediction, assessing the sulfite reduction capacity by growing representative strains in H₂S detection agar medium. The phenotypic results confirmed genome sequence-derived inferences across groups (Fig 5A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Ecological specialization across the groups of the S. baltica complex.

(A) Phenotypic assessment of sulfite reduction capacity across strains from G1 (n = 40), G2 (n = 25), and G3 (n = 12), organized by species and genospecies. Images from H₂S detection agar medium show growth on sulfite-containing media, where a dark precipitate indicates a positive sulfite reduction phenotype. Twenty-four representative G1 strains, and all G2 and G3 strains are shown. The stack-bar chart summarizes the proportion of sulfite-reducing (positive, red) and non-reducing (negative, white) strains per group. (B) Phylogenetic tree of SirA sequences retrieved from S. baltica complex genomes (S9 Table). (C) Heatmap of carbon source utilization profiles across representative strains from G1, G2, and G3, assessed using Biolog EcoPlates. (D) Mean optical density (OD₅₉₀) per carbon source across groups. Asterisks indicate significant differences between groups, as determined by pairwise Wilcoxon tests after a Kruskal-Wallis test (p < 0.05).

https://doi.org/10.1371/journal.pgen.1012306.g005

Notably, phylogenetic analysis of SirA protein sequences (including G1 strains from the NCBI GenBank dataset) revealed two distinct clades (Fig 5B). One comprised the G2 and G3 SirA proteins, whose closest ortholog was the S. xiamenensis SirA protein (98.31-98.45% sequence identity), whereas G1 SirA proteins formed a separate clade with that of S. oncorhynchi as their closest ortholog (92.87-93.00% sequence identity, Fig 5B). Only two G1 strains from the NCBI GenBank dataset (genome accessions GCA_001620325.1 and GCA_000017325.1) had the S. oncorhynchi ortholog as their closest SirA relative, both originating from the Baltic Sea. The accessions for building the phylogenetic tree are disclosed in S9 Table. At the DNA sequence level, the sir cassette in G2 and G3 strains shared 83% sequence identity with the corresponding cassette in S. xiamenensis, whereas the G1 cassette shared 79% identity with that of S. oncorhynchi. The S. oncorhynchi cassette lacks two genes present in the G1, G2, and G3 cassettes, arguing against derivation of the G1 cassette from S. oncorhynchi (S7 Fig). The genomic regions flanking the sir cassette showed conserved synteny across G1, G2, and G3 strains, as well as S. xiamenensis and S. oncorhynchi, which is compatible with recombination in shaping the distribution of these cassettes within the S. baltica complex. Known transposases were rarely detected in the vicinity of sir cassettes, providing limited support for transposase-mediated mobilization (S7 Fig).

We further expanded our analysis linking genomic differentiation with metabolic specialization by assessing carbon source utilization across representative strains using Biolog EcoPlates (G1, n = 18; G2, n = 23; and G3, n = 12). While most substrates were utilized similarly across groups, G2 strains consistently showed higher utilization of N-acetyl-D-glucosamine (consistent with their enrichment in CMP-N-acetylneuraminate biosynthesis COGs, Fig 4D) as well as pyruvic acid methyl ester, L-serine, glycyl-L-glutamic acid, Tween 40 and 80, D-cellobiose, and L-asparagine (Fig 5C, 5D). These results suggest that metabolic diversification extends beyond sulfite reduction to carbon metabolism.

Recombination drives divergence within the S. baltica complex

We reconstructed the evolutionary history of S. baltica complex strains. Our Bayesian inference based on the core proteome of the complete reference genomes (n = 32), revealed that an ancestral lineage split into two main branches, one leading to G1 and the other leading to G2 and G3 (Fig 6A). To better understand the mechanisms of this genomic diversification, we investigated whether homologous recombination plays a major role in this process, focusing on S. baltica strains recovered exclusively from sediments (n = 94). To this end, we applied two complementary approaches: (i) Gubbins, which identifies regions of elevated SNP density across the core genome alignment to detect recombination hotspots [39], and (ii) a recently described method that compares the number of identical genes (>99.8% nucleotide sequence identity) between genome pairs (observed F100) against the number expected by chance from FastANI-based relatedness, enabling inference of recent recombination frequency and gene flow within and between groups while accounting for shared ancestry [40].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Recombination events within sediment strains of the S. baltica complex from the Stockholm archipelago.

(A) Unrooted Bayesian phylogeny tree based on core-proteome sequences of representative closed genomes of the S. baltica complex (n = 32). S. baltica subspecies were collapsed.Node labels indicate highest posterior density values, and bold labels denote the different genospecies. (B) Gubbins-based recombination statistics shown as violin plots for each phylogenetic group, displaying the ratio of recombination to mutation rate (ϕ/θ), (C) the ratio of recombination events to mutation events (r/m), and (D) recombinant fragment length (bp). (E) F100 clustered heatmap showing the frequency of genes that are 100% identical between genome pairs, as a proxy of potential recombining genes. (F) F100 expected for their ANI range, i.e., F100 vs FastANI, building a GAM model using the sediment genomes (r² = 0.72). The dashed red circle denotes the members of the S. baltica complex, and turquoise dots highlight genomes that do not follow the GAM model. (G) Percentage of recombination between pairwise genomes among the different groups within the S. baltica complex with ANIb < 98%. Different letters indicate statistically significant differences between groups (p < 0.05), while groups sharing the same letter do not differ significantly (p > 0.05). (H) Pairwise estimation of r/m. In (E) and (F), F100 values were ln-transformed (ln[F100]) to improve visualization across the full range of values. In (E) and (H) different colors and frames represent the groups as previously described, indicating the previously characterized species within the complex (S. baltica, S. scandinavica, and S. vaxholmensis). S. hafniensis (SP1S1-9) and S. oncorhynchi (L-17 and SP1S1-3), also isolated from sediments in the same study areas, were used as outgroups for comparison. Genomes in the heatmap are ordered according to ANIb similarity.

https://doi.org/10.1371/journal.pgen.1012306.g006

First, to characterize the recombination dynamics within each group, we used Gubbins to compute different metrics, i.e., the ratio of recombination events to point mutations (ρ/θ; Fig 6B), the relative impact of recombination versus mutation on accumulated variation (r/m; Fig 6C) and mean of the recombinant fragment length (Fig 6D). G1 showed the lowest values across all metrics (ρ/θ = 0.008, r/m = 0.155, recombinant fragment length = 174 ± 139 bp), indicating that mutation dominates sequence diversification in this lineage. In contrast, G2 and G3 displayed substantially higher ratios and longer recombination fragments (ρ/θ = 0.162-0.267, r/m = 6.6-11.7, and recombinant fragment length = 658–734 bp), demonstrating a far greater contribution of recombination to genomic diversity (Fig 6B-6D).

Second, we performed pairwise genome comparisons using the F100-based approach, which yielded results consistent with those obtained using Gubbins (Fig 6E, 6F). As expected, F100 values dropped sharply between S. baltica complex members and more distant species isolated from Vaxön sediments, e.g., S. hafniensis (strain SP1S2-9) and S. oncorhynchi (strains L-17 and SP1S1-3), sharing less than 1% of genes [ln(F100) > 4.5] (Fig 6B). However, we observed high levels of gene sharing within the S. baltica complex strains (median 3.94; Fig 6E, 6F). These values are comparable to those observed in highly cohesive species with FastANI > 97%, both inside and outside the genus Shewanella, such as S. algae or Salinibacter ruber [41] (S7A, S7B Fig). Gene sharing was lowest in G1 (2.73 ± 0.51%) and highest in G3 (13.6 ± 9.84%), with G2 showing intermediate values (5.95 ± 2.28%). Between-group sharing was markedly higher between G2 and G3 (5.3 ± 1.32%) than between G1 and either group (~1.95 ± 1.49%), pointing to recent gene flow predominantly among the more recombinogenic lineages. Importantly, our F100 approach shows much higher gene sharing within species or genospecies than between them, supporting their delineation as discrete genomic units. To quantify this further, we counted genes classified as recently recombined across all pairwise comparisons, identifying 12,326 recombination events (62% core, 38% accessory genes), distributed across the entire genome (S8A Fig). After excluding genome pairs with ANIb > 98% (1.2% of all pairwise comparisons) to avoid overestimating recombination from very closely related strains, which exhibited high recombination levels (>45%, S8B Fig), average pairwise recombination was ~ 1.6 ± 0.5% within G1, increasing to ~4.9 ± 0.6% within G2, ~ 4.6 ± 2.3% within G3, and ~3.8 ± 1.6% between G2 and G3, with all values significantly higher than those involving G1 (p < 0.05; Fig 6G). These results indicate that G1, despite being mostly a single species (S. baltica), exhibits less gene sharing than the multi-species groups G2 and G3, reflecting extensive gene flow within the S. baltica complex, particularly between G2 and G3.

We next used F100 to estimate the r/m ratio between genome pairs and investigate recombination dynamics within the groups (Fig 6H). Within G2 and G3, strains forming part of the same species or genospecies exhibited most of the r/m ratios ranging from 1 to 100. Overall, the F100 estimates showed similar trends as Gubbins inferences, with G1 showing much lower r/m ratios than G2 and G3, in the range of 0.05 and 0.2, suggesting that point mutations predominantly drive diversification in this group. In G2 and G3, most r/m values ranged from 0.2 to 0.8 across all pairwise comparisons, showing that the divergence of these groups could be driven by ‘sexual’ speciation.

The S. baltica complex is under purifying selection

Finally, we quantified selective pressures in single-copy core genes across G1, G2, and G3 (S9A, S9B Fig). The mean dN/dS (ω) across all groups was 0.08 ± 0.11, and approximately 99.5% of core genes exhibited ω ≪ 1 (S9A Fig), indicating strong evolutionary constraint. Significant variation in mean ω was observed among groups (p < 0.05, S9A Fig). The mean ω of G3 was approximately 1.5-fold higher than that of G1 and G2, with values of 0.11, 0.08, and 0.06, respectively, and G3 contained the largest number of genes with ω > 0.5, including up to 25 core genes, and a subset of six genes inferred to be under positive selection (ω > 1) (S9B Fig), all annotated as hypothetical protein-coding genes. The overall distribution of low ω values across groups is consistent with genome-wide patterns observed in other bacterial populations [42], showing that the evolutionary dynamics within the S. baltica complex is dominated by strong purifying selection.

Sample collection, environmental DNA extraction, and bacterial isolation

Between 2021 and 2023, surficial sediment cores were aseptically collected at Vaxön (Stockholm archipelago, 59°24′01.0″ N 18°20′20.2″ E) using sterile 30 ml syringes (VWR Cat. No. 613–2035). Samples were transported to the laboratory within 2 h, and aliquots for DNA extraction and bacterial isolation were taken at 0, 2, 4, and 6 cm depth. DNA (ca. 0.5-0.8 mg) was extracted using the DNeasy PowerSoil Pro kit (Qiagen, Hilden, Germany), after mechanical disruption with FastPrep-24 5G (MP Biomedicals; 4 × 40 s at 5.5 m/s) and stored at -80 °C until sequencing. For bacterial isolation, sediment aliquots were suspended in phosphate-buffered saline (0.01 M, pH 7.4), vortexed, serially diluted, and plated (100 µl) on Lyngby’s Iron Agar (LIA) with 0.04% (w/v) L-cysteine. Plates were incubated at 28 °C for 24 h. Shewanella-like colonies, characterized by black centers with pale orangish margins, were re-streaked three times on LIA and identified to the genus level by MALDI-TOF (MALDI Biotyper Sirius, Bruker). Additional sediment and water samples were collected from Hagaparken (59.3546 N, 18.0429 E), Lidingö (59.3667 N, 18.1757 E), Nynäshamn (58.8995 N, 17.9508 E), and Tyresö (59.2324 N, 18.3110 E) between 2021 and 2022 (S1 Fig), and processed as previously described [32,33]. In total, 112 Shewanella strains were isolated and analyzed, including previously reported strains [32,33].

Whole-genome and metagenome DNA sequencing

Genomic DNA was extracted from overnight LB (Miller) cultures at 28 ºC using the DNeasy Blood & Tissue kit (Qiagen) and quantified with the Quant-iT High Sensitivity dsDNA assay (ThermoFisher, Waltham, MA, USA). Equimolarly pooled libraries were circularized using the MGI Easy Circularization kit, and 2 × 100 bp paired-end sequencing was performed on a DNBSEQ-G400 instrument (MGI, Shenzhen, China). Genome assemblies were generated with SPAdes v3.15.5 [65] via Bactopia v1.7.1 [66] or BACTpipe (https://github.com/ctmrbio/BACTpipe). For 32 representative strains, including type strains and at least one novel genospecies, Oxford Nanopore sequencing was performed to obtain complete genome sequences. High-molecular-weight DNA was extracted using the Quick-DNA HMW MagBead Kit (Zymo Research, Irvine, CA, USA), and libraries were prepared with the Rapid Barcoding Kit (SQK-RBK114.24). Sequencing was carried out on a FLO-MIN114 (R10.4.1) flow cell using a MinION Mk1B device (Oxford Nanopore Technologies), using MinKNOW v25.05.14 with high-accuracy basecalling modelling. Long reads were assembled de novo using Flye v2.9.5-b1801 [67] and Raven v1.8.3 [68]. Hybrid assemblies combining Oxford Nanopore long reads and DNBSEQ-G400 reads were generated using Unicycler v0.5.1 to optimize plasmid recovery [69,70]. Circularized replicons produced by Flye were retained whenever possible.The resulting genome sequences were polished using Medaka v1.11.3 (https://github.com/nanoporetech/medaka), Polypolish v0.6.0 [71], and POLCA within MaSuRCA v4.1.0 [72,73], and reordered using UGENE v48.1 [74] based on the Unicycler start genes database [70,74]. Assembly quality metrics were evaluated using QUAST [75] v5.0.2. Genome completeness and contamination were evaluated using CheckM [76] and ConFindr [77]. The genomes were of high quality, exhibiting an average of 6 ± 4 single nucleotide variants (SNVs) in core genes, with mean completeness of 98.76 ± 0.86% and contamination of 0.74 ± 0.63%, indicating a low likelihood of chimeric assemblies (S2 Table).

For metagenomic analysis, 2 × 150 bp paired-end sequencing was performed on a DNBSEQ-T7 (MGI). Quality-trimmed reads were classified with Kraken2 v2.1.0 [78] using the Genome Taxonomy Database (GTDB, downloaded 8 April 2022) [79]. Contigs were assembled with MEGAHIT v1.2.9 [80] and analyzed with the SqueezeMeta v1.7.2 pipeline [81]. Open reading frames (ORFs) were predicted with Prodigal v2.6.3 [82], and protein similarity searches were performed against the GenBank non-redundant (nr) protein and Kyoto Encyclopedia of Genes and Genome (KEGG) databases using DIAMOND (v2.1.10) [83]. Finally, reads were mapped back to contigs with Bowtie2 v2.5.4 [84] to assess functional coverage and abundance.

Genomic relatedness analyses

Average nucleotide identity (ANI) analyses were performed using JSpeciesWS for BLAST-based ANI (ANIb) [85] and FastANI v1.34 (‘many to many’ mode, default settings; https://github.com/ParBLiSS/FastANI). Pairwise ANIb values were hierarchically clustered to define genome groups. Publicly available genome sequences of Shewanella algae (n = 219), S. xiamenensis (n = 98), and S. oncorhynchi (n = 32, including three strains from this study) were downloaded from NCBI GenBank on 31 May 2025 and analyzed with FastANI. Digital DNA–DNA hybridization (dDDH) values were calculated using the Genome-to-Genome Distance Calculator (GGDC) [86], comparing each genome against the maximum allowed number (n = 75), and repeated for all genomes. Only results from the recommended Formula 2 were reported [87].

Pangenome analyses, accessory genome clustering and gene gain-loss dynamics

Pangenome analyses within the S. baltica complex (and the additional datasets previously mentioned) were performed on genomes annotated with Prokka v1.14.6 [88], using PPanGGOLiN v2.2 [89]. Default parameters were applied with ≥90% identity and coverage, as higher thresholds (>95%) over-split conserved genes and lower thresholds (50–80%) merged paralogs. Genes present in ≥95% of genomes were classified as core, considering that the vast majority of the genomes sequences are draft genomes. Core gene and protein family alignments were generated using fast Fourier transform (MAFFT) [90], and phylogenomic trees were reconstructed with FastTree v2.1.10 [91] using GTR + Γ4 for DNA alignments and JTT + CAT for protein alignments. Trees were bootstrap-supported with 1,000 replicates, midpoint-rooted, and visualized in iTOL v6 6 [92]. Accessory gene accumulation curves were built from the Rtab presence-absence matrix by randomly permuting genome order 100 times and averaging the cumulative number of new accessory genes at each step. Heaps’ law (n = k·N^γ) was fitted to the mean curve by non-linear least-squares regression, with γ indicating pangenome openness (closer to 0 = closed, closer to 1 = open). Using the same matrix, we constructed a hierarchical clustering dendrogram based on gene presence–absence. This matrix was also analyzed with Mandrake v1.2.2 to generate a two-dimensional embedding [93], enabling identification of accessory genome clusters (AGCs). Gene gain and loss events were modeled using Panstripe v0.3.1, using the included Gaussian distribution, which provides more robust model fitting while yielding results comparable to the compound poisson (Tweedie) model [36]. Using PanRGP as implemented in PPanGGoLIN, we identified regions of genome plasticity (RGPs), defined by PanRGP as clusters of consecutive shell or cloud genes [94], which was used as a proxy for horizontal gene transfer (HGT).

Estimation of genome-wide differentiation

Genome assemblies were aligned with Split K-mer Analysis V2 using S. baltica CECT 323^T as the reference, and single nucleotide polymorphisms (SNPs) were extracted with SNP-sites v2.5.1 [95] and stored in VCF format. SNPs were grouped into overlapping 10 kb sliding windows (2.5 kb steps), and for each window, pairwise differentiation was calculated as the proportion of discordant SNPs (p-distance). Genome-wide differentiation between each pair of genomes was obtained as the average p-distance across all windows, providing a genome-wide complement to classical population-genetic metrics ranging from 0 (identical genomes) to 1 (completely different) [96]. Pairwise distances were concatenated and clustered as for ANIb.

Bayesian phylogenetic analysis

Bayesian phylogenetic analysis was performed with BEAST X v10.5.0 [97] using the core proteome alignment of 32 complete genomes (3,396 proteins; 1,157,272 amino acids) representing distinct species and putative genospecies within the  S. baltica complex. A Blosum62 substitution model was applied, with a strict molecular clock to assume uniform substitution rates, and no time calibration was used, as the aim was to assess relative relationships rather than absolute divergence times [98,99]. The starting tree was generated under a constant-size coalescent, and a Yule speciation model constrained tree topology and branch lengths. Posterior distributions were estimated via Markov Chain Monte Carlo (MCMC) for 10,000,000 generations, sampling every 1,000 steps. Convergence was assessed with Tracer v1.7.2 [97], trees were summarized with TreeAnnotator v10.5.0 [97], and visualized using iTOL v6 [92].

Metabolic and horizontally acquired gene profiling for ecological differentiation

Genome assemblies were re-annotated using COGclassifier v1.0.5, based on COG categories (https://github.com/moshi4/COGclassifier/tree/main), and RAST SEED roles [100], by calculating the proportion of protein-coding genes assigned to each functional category (performed in KBase [101]). Taxonomic assignments for the COGs identified within RGPs were obtained using EggNOG v5.0 [102]. Functional enrichment was assessed using Kruskal-Wallis tests for COG categories and SEED roles, and fold changes were calculated as the median value per genome relative to all others. COG and RAST-annotated genomes, and COG-annotated HGT genes, were compiled into feature matrices and standardized to Z-scores prior to multivariate analyses. Group differences were assessed using permutation-based multivariate analysis of variance (PERMANOVA), and enrichment analyses used Benjamini–Hochberg false discovery rate (FDR) correction, with significance defined as FDR-adjusted p < 0.05. Finally, KEGG pathway module completeness was evaluated from KEGG orthologues (KOs) presence-absence using the kegg-pathways-completeness tool (https://github.com/EBI-Metagenomics/kegg-pathways-completeness-tool/tree/master).

Respiratory metabolism of N- and S-oxides was investigated using annotated sequences of respiratory reductases from reference species within the complex, including S. baltica OS195 (GCA_000018765.1), S. baltica CECT 323^T (GCA_900456975.1), and S. scandinavica SP2S1-2^T (GCA_031932065.1). These genomes contained reductases for nitrate (NapA; ABX48036, ABN61416, MDT3282522, MDT3279942, MEM6250911, MEM6249282), nitrite (NrfA; ABX50994, MDT3279623, MEM6250231.1), trimethylamine N-oxide (TMAO, TorA; ABX50514, MDT3281430.1, MEM6250556.1), dimethyl sulfoxide (DMSO, DmsA; ABX49397.1, MDT3282918.1), tetrathionate (TtrA; WP_006080095, MDT3279598.1, WP_311905661), thiosulfate (PhsA; MDT3282209.1, WP_311905956), and sulfite (SirA; ABX51162, MDT3279461, MEM6249422). After identifying homologs in our dataset, we constructed a custom database of these respiratory reductases designed to capture the diversity of evolutionary lineages within the complex. For the final analysis, the curated reference sequences from this database were used for BLASTp (BLAST+ v2.1.16) [103] searches with an e-value cutoff of 10^-30, and hits with ≥90% sequence identity were retained for downstream analyses. Synteny analysis of the genomic regions containing the sirA gene was performed using Clinker [104]. SirA protein sequence alignments were performed using Clustal Omega, and phylogenetic trees were reconstructed with MEGA12 using the maximum-likelihood method and the best-fit amino acid substitution model identified by the software (WAG+I), with 1,000 bootstrap replications.

Experimental metabolic characterization

The ability to reduce sulfite of G1, G2 and G3 members was assessed with an H₂S detection agar medium, consisting of 1% (w/v) basal agar medium containing 7.5 g/L NaCl, 5 g/L MgCl₂·6H₂O, 0.5 g/L KCl, 0.15 g/L CaCl₂·2H₂O, 1 g/L NH₄Cl, and 0.2 g/L KH₂PO₄, supplemented with 0.02% (w/v) casamino acids, 0.015% (w/v) FeSO₄, 40 mM sodium lactate, and 10 mM sodium sulfite. The medium was buffered with 30 mM HEPES and adjusted to pH 7.6. After 48 h of incubation at 28 ºC, the formation of black precipitates was considered a positive reaction.

In addition, carbon source utilization across representative strains of each group was evaluated using Biolog EcoPlates (G1, n = 18; G2, n = 23; and G3, n = 12). Prior to the analysis, strains were grown overnight in Miller’s LB broth, diluted 1:100, and cultured until reaching the exponential phase (ca. 0.6 OD₆₀₀). Bacteria were then pelleted, resuspended in 0.85% (w/v) NaCl, and cell densities normalized to McFarland 0.5 before transferring 150 µL to each well. Plates were incubated for 72 h, with OD₅₉₀ monitored every 12 h.

Detection of recombination events

Homologous recombination was measured using Gubbins [39], and the F100 prokaryotic recombination pipeline [40] (https://github.com/rotheconrad/F100_Prok_Recombination). Using the core genome alignment as input, we ran Gubbins separately for each of the three evolutionary groups identified in this study, applying default parameters (minimum number of SNPs to identify a recombination block = 3; minimum window size = 100 bp; maximum window size = 10,000 bp). For the F100 pipeline, reciprocal best matches (RBMs) between genome pairs were identified by BLAST [105], and the F100 metric was calculated as the proportion of RBMs that were 100% identical, serving as an indicator of recent gene exchange. Recombinant genes were classified as core (≥90% of genomes), accessory (<90% and present in multiple genomes), or genome-specific, while highly conserved genes (as the 10% of core genes with the lowest divergence) were excluded to avoid conflating conservation with recombination. Gene clustering was performed with MMSeqs2 using the same parameters as for core gene definition (--min-seq-id 0.90, -c 0.90, --cov-mode 1, --cluster-mode 2, --cluster-reassign), and pairwise genome recombination values were extracted from the pipeline output (script 03f_Recombinant_pair_analysis.py, default settings). Additional recombination parameters were calculated as the recombination-to-mutation ratio (ρ/θ) or the relative impact of recombination versus mutation on accumulated variation (r/m) within the groups was calculated using Gubbins pipeline. Pairwise r/m estimates were obtained using the empirical approach implemented in the F100 prokaryotic recombination pipeline [40], which assumes that recombined genes initially carried the average sequence divergence of the compared genomes.

Evolutionary genomic analyses

Selection acting on core genes of the S. baltica complex was assessed using codon alignments of single-copy core genes for each respective phylogenetic group [106]. Protein alignments were back-translated to codon alignments using PAL2NAL v14 to preserve reading frame information. Gene-specific phylogenies were reconstructed with FastTree v2.1.11, and branch-specific ratios of nonsynonymous to synonymous substitution rates (dN/dS, ω) were estimated using the FitMG94 model, which implements the Muse–Gaut (MG94) codon substitution model of sequence evolution, in HyPhy v2.5.64 [107]. The model was implemented to permit independent variation of dN and dS across branches, thereby accommodating lineage-specific heterogeneity in selective pressures. Confidence intervals for ω were estimated using profile likelihood, and deviations from neutrality (ω = 1) were evaluated using likelihood ratio tests comparing the unconstrained model to a nested model in which ω was fixed at 1. To ensure robust downstream inference, HyPhy outputs were filtered to retain genes with codon alignments of at least 60 codons, and estimates were excluded when dS were below 0.005 to minimize instability and avoid artificially inflated ω values associated with extremely small dS values. For core genes showing signatures of positive selection, we additionally screened for homologous recombination, and removed if it was the case, using the GARD method [108] implemented in HyPhy, since recombination can produce false signals of adaptive evolution by affecting dN/dS estimates.

General computational methods, statistics, and data availability

Data processing, plotting, and statistical analyses were performed using custom Python (v3.9.19) and Bash (GNU Bash v3.2.57) scripts, together with RStudio v2025.05.1. All analysis scripts used in this study available via the Martín-Rodríguez Lab GitHub repository (https://github.com/Martin-Rodriguez-lab/Life_in_sediments_fosters_sexual_speciation_in_the_Shewanella_baltica_complex) and multiple sequence alignments and datasets used and/or analyzed to reproduce the analysis can be found in our Zenodo for this project (https://zenodo.org/records/21233620). Genome assemblies are available in NCBI under BioProject PRJNA526057 and metagenomic raw reads under SRR36813638-SRR36813649, with individual accession numbers listed in S2 Table.