Genomic and transcriptomic analyses reveal the adaptation to semi-aquatic and aquatic life in spiders

Abstract

Spiders are largely terrestrial, but some lineages have independently adapted to marine and freshwater habitats. Although physiological and behavioral traits supporting these evolutionary transitions have been characterized, the genomic basis of adaptation to aquatic and semi-aquatic habitats in spiders remains unclear. Here, we report a chromosome-level genome assembly for the aquatic spider Argyroneta aquatica. Comparative analysis of 22 spider species supports 2 independent transitions in semi-aquatic (∼56 Mya) and aquatic (∼44 Mya) lineages. The aquatic spider shows distinct respiratory morphology compared with terrestrial spiders, including anterior spiracles and denser tracheoles. Molecular evolution analyses identified lineage-specific shifts in selective constraint or evidence of positive selection linked to trachea development (eg Wnt-1, Catenin beta, TMEM234, Arp3), hypoxia response (eg Uqcrfs1, COX5B, SLC2A3), lipid metabolism (eg Pnliprp2, ND-ACP), and osmoregulation. In the semi-aquatic spiders (Desis spp.), we detected strengthened purifying selection or positive selection on genes involved in respiratory/energy metabolism (eg SdhD, SLC2A3) and ion transporters (eg. Slc24a5). Comparative results further indicate that the A. aquatica genome may harbor bacterial-origin genes homologous to ABC transporters (39 genes) and acyl-CoA dehydrogenases (ACADs; 21 genes), which may support metabolic or transport functions in the aquatic lineage. Transcriptomic and metabolomic profiling of A. aquatica under hypoxic challenge revealed metabolic reprogramming, including shifts in glycolytic and TCA intermediates and upregulation of genes that promote fatty-acid β-oxidation (eg Slc13a5, AACS), consistent with extended anoxia tolerance (>48 h). Overall, this work provides genomic resources and highlights genomic signatures associated with semi-aquatic and aquatic adaptation in spiders.

Introduction

Most spiders are adapted to terrestrial life; however, a notable subset of species that originated on land has secondarily transitioned back into marine and freshwater habitats (Leggett et al. 2024). These reversals have occurred in both semi-aquatic and fully aquatic lineages. In this study, we define an aquatic spider as one for which the entire life cycle takes place underwater (eg the diving bell spider Argyroneta aquatica, which inhabits freshwater ponds). By contrast, we define semi-aquatic spiders as species that remain submerged for substantial periods yet must emerge onto land for essential activities such as foraging, mating, or dispersal (eg intertidal spiders such as Desis martensi, which forage terrestrially during low tide). These definitions emphasize behavioral differences rather than differences in physiological tolerance alone. For example, the marronoid clade has well-resolved phylogenetic relationships and comprises roughly 9 families (Agelenidae, Amaurobiidae, Cybaeidae, Cycloctenidae, Desidae, Dictynidae, Hahniidae, Stiphidiidae, and Toxopidae). Members of this clade exhibit diverse life-history strategies and are adapted to a range of extreme environments (Gorneau et al. 2023), including the semi-aquatic spider genus Desis (Desidae) and the aquatic spider A. aquatica (Dictynidae).

Multiple physiological adaptations to semi-aquatic and aquatic environments have been documented in spiders. Respiration is a major constraint for semi-aquatic and aquatic spiders because their respiratory organs, book lungs and tracheae, are broadly similar to those of terrestrial taxa. The semi-aquatic spider Desis resides in air-containing spaces among rocks, crevices, and/or shells within and around the nest (McQueen and McLay, 1983). In contrast, the water spider can construct a “diving bell” to breathe underwater, filling it with air carried down from the surface (Stefano et al. 2016; Schaber et al. 2023). For aquatic spiders, hypoxia tolerance is especially critical because access to air is limited (Seymour and Hetz 2011). In marine semi-aquatic spiders such as Desis, osmoregulation is a key physiological requirement, and osmotic pressure can drive a suite of physiological, behavioral, and morphological adjustments. Semi-aquatic and aquatic spiders also consume fish, whereas terrestrial spiders primarily prey on insects (Nyffeler and Pusey 2014). Together, these shifts underlie substantial morphological, physiological, and behavioral differences among terrestrial, semi-aquatic, and aquatic spiders.

However, the evolutionary processes underlying these adaptations remain unclear. Comparative genomic and molecular work in other systems suggests several possible evolutionary mechanisms:

Relaxed selection describes a decrease in the strength or effectiveness of both purifying and positive selection acting on a trait or gene (Lahti et al. 2009). During shifts to new environments, species may experience heterogeneous selective regimes across key phenotypic traits. After a trait approaches a new fitness optimum, selective pressures can weaken, producing relaxed constraints (Yoder et al. 2010; Hunt et al. 2011; Hughes 2012). In cavefish, eg, the loss of light relaxes functional constraints on visual opsin genes, often leading to higher dN/dS ratios and, in some cases, pseudogenization (Li and He, 2009; Calderoni et al. 2016).

Intensified selection refers to stronger or more effective natural selection acting on specific genes or traits. This pattern commonly emerges when populations face novel or harsher conditions that require improved physiology, more efficient behavior, or morphological specialization. In such contexts, beneficial alleles can spread quickly, promoting faster adaptive refinement and increasing fitness in the new habitat. Signals of intensified selection, therefore, suggest traits that remain functionally important and are maintained under strong constraints during continued evolutionary change. Consistent with this idea, studies of cave spiders (Leptonetela spp.) indicate that retained phototransduction pathway genes remain functional and are subject to strong selective constraints even in extreme environments (Wang et al. 2023).

Positive selection favors advantageous variants and can drive rapid adaptive change. Cetaceans (whales and dolphins), for example, underwent pronounced morphological changes from their terrestrial ancestors, including the evolution of flippers and tail flukes, alongside signatures of accelerated evolution in the genome (Yim et al. 2014; Tian et al. 2016; Noh et al. 2022). In the aquatic insect dobsonfly Neoneuromus ignobilis, genomic analyses suggest that both relaxed selection and positive selection have acted on long wavelength-sensitive and blue-sensitive opsins, thermal stress response TRP channels, and sulfotransferases (Ma et al. 2022). In spiders, emerging genomic resources have revealed the importance of molecular adaptation; for example, mitochondrial genome analysis of Desis jiaxiangi revealed positive selection on nd4 and nd6, highlighting a potential role for energy metabolism in semi-aquatic transitions (Li et al. 2021).

Symbiotic partnerships can extend an animal's physiological and metabolic capabilities beyond what is encoded in its own genome. In addition to providing nutritional advantages (Dubilier 2025), symbionts may support detoxification (Paniagua Voirol et al. 2018; Rupawate et al. 2023), osmoregulation (Cheng et al. 2023; Chen et al. 2025), or immune modulation (Wang et al. 2023), thereby broadening the host's functional repertoire without requiring genetic integration. In spiders, bacterial symbioses have been documented in which microbes contribute discrete biological functions. For example, endosymbiotic bacteria have been shown to mediate the production of certain venom proteins (Gendreau et al. 2017). In addition, host resistance to selected insecticides can be shaped by shifts in the abundance and diversity of endosymbiotic bacteria (Řezáč et al. 2025). For aquatic organisms, genes involved in detoxification, osmoregulation, and energy metabolism are particularly likely candidates for such acquisitions. Specifically, ATP-binding cassette (ABC) transporters and acyl-CoA dehydrogenases (ACADs) play key roles in cellular defense, lipid metabolism, and stress responses (Wu et al. 2011; Dermauw and Van Leeuwen 2014). We therefore investigated these 2 families in the newly assembled chromosome-level genome of the aquatic spider Argyroneta aquatica to assess their potential contribution to freshwater adaptation.

We hypothesized that multiple evolutionary processes may be involved in the adaptation of spiders to semi-aquatic and aquatic habitats, including relaxed selection, intensified selection, positive selection, and contributions from putative endosymbiotic bacteria. We provided a high-quality chromosome-scale genome assembly for the aquatic spider A. aquatica. We also incorporated publicly available genomes and transcriptomes from closely related spider species. We then conducted comparative analyses across 22 spider species, comprising the aquatic spider, 2 semi-aquatic spiders, and 19 terrestrial species within the marronoid clade. We also conducted comparative transcriptomics and metabolomics analyses with a focus on hypoxia responses in water spiders. Together, these analyses clarify how evolutionary forces have shaped semi-aquatic and aquatic adaptation in spiders.

Results

Morphological and phenotypic adaptive traits

Most spiders possess 2 distinct respiratory systems: a pair of book lungs and 1 or 2 pairs of tubular tracheae that branch extensively through the body (Foelix 2011). Book lungs lie ventrally in the anterior opisthosoma and consist of stacked air-filled chambers interspersed with lamellae containing hemolymph spaces. The tubular tracheae are positioned posterior to the book lungs, within the third abdominal segment. Gas exchange with the external environment occurs through spiracles, which are typically inconspicuous and appear as small openings.

To examine adaptive changes in respiratory morphology associated with semi-aquatic and aquatic habitats, we analyzed the respiratory systems of the semi-aquatic spider D. martensi and the only fully aquatic spider species, the water spider A. aquatica. To provide phylogenetically informed terrestrial comparisons, we selected Dictyna felis as the terrestrial counterpart of A. aquatica because both belong to Dictynidae, and we included the terrestrial congener Badumna tangae as a comparison for the semi-aquatic D. martensi, with both species belonging to Desidae (Fig. 1a).

Figure 1.

Environment, respiratory morphology, and phylogenetic relationships of the water spider Argyroneta aquatica and related spiders. a) Schematic illustrating evolutionary transitions from terrestrial to semi-aquatic habitats and from terrestrial to aquatic habitats. b) Morphological changes in respiratory systems associated with transitions from terrestrial to semi-aquatic habitats and terrestrial to aquatic habitats. Tt, tracheal tube; Ts, tracheoles; BL, book lungs. Tracheoles are extremely fine tubes, often <1 μm in diameter. c) Hypothesized evolutionary routes among terrestrial, semi-aquatic, and aquatic spiders. d) Phylogeny of A. aquatica and “marronoid” spiders, with Uloboridae (U. diversus) used as the outgroup. Divergence times are indicated at branch nodes and along the time scale. Black branches denote terrestrial taxa, blue branches denote semi-aquatic taxa, and orange branches denote aquatic taxa. Images depict A. aquatica, D. martensi, and D. japonica. Photographs: A. aquatica by Qian-Le Lu; D. martensi by Lu-Yu Wang; D. japonica reproduced from Ono (2009), with permission.

Morphological comparisons indicated that book lungs in the semi-aquatic spider D. martensi and the aquatic spider A. aquatica are thicker than those of their terrestrial relatives, although the difference is not particularly pronounced. Tracheoles are more abundant in semi-aquatic and aquatic spiders than in their terrestrial counterparts, with the greatest increase observed in the aquatic species. Moreover, the spiracles are positioned more anteriorly in the aquatic spider (Fig. 1b). This shift may be related to its tendency to carry an air-filled bubble around the abdomen to support underwater respiration.

Genomic features of the water spider

In this study, we sequenced the genome of the water spider A. aquatica. Specifically, we generated a high-quality, chromosome-level assembly for A. aquatica with a scaffold N50 of 2.44 Mbp and a total assembly size of 753.77 Mbp distributed across 11 chromosomes (Figures S2, S3, Table S2). BUSCO assessment recovered 98.2% of arthropod_odb10 conserved genes (1,013 genes) and 97.5% of arachnida_odb10 conserved genes (2,934 genes) (Table S2). By comparing with these representative spider genomes, including Amaurobius ferox, Dolomedes plantarius, and Uloborus diversus, the A. aquatica assembly shows a relatively smaller genome size but comparable scaffold contiguity and high completeness based on BUSCO metrics (Table S3). In addition, we also found that repetitive sequences comprise ∼41.53% of the genome (Figure S3, Table S4). Gene model prediction integrating homology-based searches, transcriptome evidence, and ab initio methods identified 17,632 protein-coding genes (Table S5).

Divergent evolutionary routes of semi-aquatic and aquatic spiders revealed by phylogenomic analyses

To investigate molecular adaptations associated with transitions to semi-aquatic and aquatic lifestyles, we analyzed 22 spider species within a phylogenetically well-resolved framework, primarily focusing on members of the marronoid clade (Fig. 1a, b; Table S1). This clade encompasses multiple families with diverse ecological strategies, including both the semi-aquatic genus Desis and the aquatic spider A. aquatica. Notably, semi-aquatic and aquatic lifestyles have evolved independently in their lineages. Our sampling, therefore, enables comparative analysis of 2 independent evolutionary origins from terrestrial ancestors to water-associated habitats. Due to the limited availability of genomic resources for semi-aquatic spiders, Desis represents one of the few tractable lineages; in addition to publicly available data, we generated a new transcriptome for D. martensi to improve representation within this group.

A phylogenomic tree reconstructed from 941 shared single-copy orthologs resolved relationships among these taxa, particularly within the marronoid clade (Fig. 1d). Molecular dating estimated that the marronoid spider clade originated ∼100.07 million years ago. Our results support independent origins of semi-aquatic and aquatic lifestyles from their terrestrial ancestors (Fig. 1c).

A. aquatica occurs in freshwater lakes and ponds, whereas the semi-aquatic spiders D. martensi and D. japonica inhabit intertidal coastal environments (Figure S1). Phylogenetic reconstruction placed A. aquatica as sister to a clade of terrestrial dictynid spiders (A. kudoae and D. felis), with an estimated divergence time of 43.79 million years ago (Fig. 1d). By contrast, the semi-aquatic Desis species (D. japonica, D. martensi) belong to Desidae, which diverged ∼55.82 million years ago, within Desidae, Desis species are semi-aquatic, whereas other sampled members (Metaltella simoni, Badumna longinqua) are terrestrial (Fig. 1d). These patterns indicate that both the semi-aquatic (Desis) and aquatic (A. aquatica) lineages evolved independently from terrestrial ancestors within their respective families, implying separate evolutionary origins for aquatic-associated lifestyles.

Variation in selection intensity associated with spider underwater transitions

We estimated gene-wide nonsynonymous to synonymous substitution rate ratios (dN/dS, ω) across the spider phylogeny for 4,744 shared orthologs in the semi-aquatic transition and 4,871 shared orthologs in the aquatic transition. To test for selection associated with 2 underwater transitions, we applied the RELAX framework to evaluate shifts in selective strength across genes.

For the semi-aquatic transition, we identified genes exhibiting relaxed or intensified selection in semi-aquatic species relative to the remaining taxa in “evolutionary route 1” (Fig. 1c). In total, 58 genes showed intensified selection (K > 1, q-value < 0.05), and 67 genes showed relaxed selection (K < 1, q-value < 0.05) (Table S6). Neither set displayed significant functional enrichment (Figures S4–S7). Although enrichment was not statistically supported, several genes inferred to be under intensified selection (K > 1) participate in core metabolic and cellular functions, including respiration and energy metabolism (SdhD, Pfk-2, SLC2A3), ion transport (Slc24a5), and DNA repair (Polr1b, Rad23a) (Fig. 2a, Table S6). The RELAX signal is most consistent with strengthened purifying selection rather than adaptive diversification, implying that these loci may have become increasingly conserved and functionally essential in semi-aquatic lineages. Genes showing relaxed selection included loci linked to lipid metabolism (apmap), heat-stress response (HSBP1), and DNA repair (brcc3, Mes4) (Fig. 2a, Table S6). Reduced constraint on these genes may reflect shifts in functional requirements or lineage-specific changes in selective regimes associated with the transition to semi-aquatic habitats.

Figure 2.

Genome-wide dN/dS patterns associated with transitions from terrestrial to semi-aquatic habitats and terrestrial to aquatic habitats in spiders. a) Volcano plot showing genes under relaxed or intensified selection in semi-aquatic spiders relative to terrestrial spiders. b) Volcano plot showing genes under relaxed or intensified selection in aquatic spiders relative to terrestrial spiders. c) Volcano plot depicting genes under positive selection in semi-aquatic spiders compared with terrestrial spiders. d) Volcano plot depicting genes under positive selection in aquatic spiders compared with terrestrial spiders.

For the aquatic transition, we similarly identified genes with relaxed or intensified selection in the water spider relative to taxa in “evolutionary route 2” (Fig. 1c). We detected 92 genes under intensified selection (K > 1, q-value < 0.05) and 59 genes under relaxed selection (K < 1, q-value < 0.05) (Table S7). Genes with intensified selection (K > 1) were significantly enriched for several Gene Ontology terms, including nucleolus, chromosomal part, and small molecule catabolic process (Figures S8, S9). Representative loci included genes involved in mitochondrial respiration and energy metabolism (Uqcrfs1, COX5B, UQCRC2), ion and electron transport (FUNDC1, CA2), and DNA repair (PPP4CB) (Fig. 2b, Table S7). As above, the RELAX pattern suggested strengthened purifying selection, suggesting that these cellular and metabolic genes have become increasingly conserved and functionally indispensable in the aquatic lineage. Genes exhibiting relaxed selection (K < 1) did not show significant enrichment for specific processes (Figures S10–S11), but several candidates potentially related to metabolism (GlyP, ETFA, SLC37A3), stress response (TMEM59), and DNA repair (ALKBH2, PGBD4) were identified (Fig. 2b, Table S7). Relaxation at these loci may indicate reduced functional constraint or lineage-specific shifts in selective pressures associated with adaptation to an aquatic environment.

Positive selection facilitates adaptation to underwater life

We identified 127 positively selected genes (PSGs) in semi-aquatic spiders (Table S8). These PSGs included candidates implicated in transmembrane ion transport (eg Slc24a5, slc37a3), DNA repair (eg KIN [DNA/RNA-binding protein KIN17] and Rad23a [UV excision repair protein RAD23 homolog A]), lipid metabolism and digestion (eg Pnliprp2, boca, ACADM [Medium-chain specific acyl-CoA dehydrogenase, mitochondrial], ABHD17B [Alpha/beta hydrolase domain-containing protein 17B], Acot13 [Acyl-coenzyme A thioesterase 13]), heat-stress response (eg DNAJB4 [DnaJ homolog subfamily B member 4]), and respiratory-chain/energy metabolism (eg SdhD [Succinate dehydrogenase], SLC2A3). We also detected a karyopherin gene, Kpna3 (Importin subunit alpha-4), among the PSGs (Fig. 2c). We detected 26 PSGs in the aquatic spider (Table S9). These genes included candidates related to lipid metabolism and digestion (eg ND-ACP [Acyl carrier protein, mitochondrial]) as well as the karyopherin gene Kpnb1 (Importin subunit beta-1) (Fig. 2d).

Putative bacterial-origin genes within the ABC and ACAD families

During genome annotation, we noted that the A. aquatica genome contained a high number of genes annotated as ABC transporters and ACADs compared with other spider genomes. The A. aquatica genome contains 75 annotated ABC genes, exceeding the numbers reported in terrestrial relatives such as A. ferox (26 genes) and D. plantarius (41 genes). Likewise, A. aquatica encodes 35 ACAD genes, compared with 9 in A. ferox and 12 in D. plantarius (Fig. 3a, b, Tables S10, S11).

Figure 3.

Characteristics of putative bacterial-origin ABC and ACAD genes in Argyroneta aquatica. a) Phylogenetic tree of the ABC gene family in A. aquatica and 2 terrestrial spiders, Amaurobius ferox and Dolomedes plantarius. Red branches indicate putative bacterial-origin ABC genes. b) Phylogenetic tree of the ACAD gene family in A. aquatica, A. ferox, and D. plantarius. c) Exon-number and GC-content features of ABC and ACAD genes across A. aquatica, A. ferox, and D. plantarius. d) Genomic locations of putative bacterial-origin genes on scaffold 1 of the A. aquatica assembly. e) Collinearity analysis of conserved genes flanking bacterial-origin loci in A. aquatica, A. ferox, and D. plantarius. f) Heatmap showing read support for putative bacterial-origin ABC and ACAD genes across 8 A. aquatica resequencing datasets (P1-P4, female; P5-P8, male).

Given the importance of these protein families in stress response and metabolism, we sought to characterize their genomic architecture and evolutionary origins. To further characterize ABC and ACAD family architecture in A. aquatica, we quantified exon number and the GC content for each family member (Fig. 3a–c, Tables S10, S11). A subset of loci contained only 1-2 exons, fewer than most other family members. These loci were located on unanchored scaffolds rather than assembled chromosomes, and they were distributed across scaffolds instead of forming clustered genomic arrays (Fig. 3d, e). When we mapped these loci to A. aquatica whole-genome resequencing datasets, most were supported by aligned reads (Fig. 3f).

We next combined multiple lines of evidence, including BLAST-based similarity searches, phylogenetic inference, and PCR cloning, to evaluate the origins of these ABC and ACAD loci. This integrative analysis suggested that a subset of loci may be of bacterial origin. Specifically, 39 genes annotated as ABC homologs and 21 genes annotated as ACAD homologs in A. aquatica exhibited high sequence similarity to bacterial proteins. In gene-family phylogenies, these bacterial-like sequences consistently clustered within bacterial clades and showed no close affinity to arthropod homologs (Figures S20, S21). For experimental follow-up, we selected 9 bacterial-homologous loci (6 ABC and 3 ACAD genes) for PCR amplification (Figure S22). BLAST searches against the NCBI database further indicated that these ABC and ACAD genes of bacterial origin were most similar to sequences from diverse bacterial families, including 14 (29%) from Alcaligenaceae, 10 (20.5%) from Comamonadaceae, 10 (20.5%) from Moraxellaceae, 5 (10%) from Flavobacteriaceae, 4 (8%) from Rhodocyclaceae, 2 (4%) from Chitinophagaceae, and 4 (8%) from other families or orders (Figure S19a). Notably, many of these bacterial groups are widely distributed and have been reported in freshwater environments (Figure S19b).

Notably, these bacterial-origin ABC and ACAD genes were not detectably expressed in different tissues transcriptomic data. To further verify whether they are expressed at low levels, we performed RT-PCR using a cDNA pool derived from 3 life stages (larvae, sub-adult, and adult). Despite their low abundance, several candidate loci were successfully amplified from the mixed-stage cDNA (Figure S26). These results provided evidence for putative bacterial-origin genes within the ABC and ACAD families.

Comparative transcriptomics and metabolomics of the water spider response to hypoxia stress

The water spider shows a remarkable capacity to survive for 48 h when its spiracles are sealed with paraffin. To probe the molecular basis of this hypoxia response, we performed comparative transcriptomic and metabolomic analyses of hypoxia-treated spiders (LO) versus controls maintained under normal oxygen conditions (CK) (Figure S23).

Under hypoxia stress, transcriptomic comparison of LO and CK identified 7 up-regulated genes and 17 down-regulated genes (Fig. 4a). Functional enrichment indicated that these differentially expressed genes were primarily associated with energy metabolism, including Gld (Glucose dehydrogenase), HAO1 (Hydroxyacid oxidase 1), Enox1 (Ecto-NOX disulfide-thiol exchanger 1), Slc13a5 (Solute carrier family 13 member 5), AACS (Acetoacetyl-CoA synthetase), and SLC1A2 (Excitatory amino acid transporter 2).

Figure 4.

Comparative transcriptomic and metabolomic analyses of the hypoxia-treated group (LO) and the control group (CK). a) Volcano plot showing differentially expressed genes (up-regulated and down-regulated) between the CK and LO groups based on transcriptomic data. b) Volcano plot showing differentially accumulated metabolites (up-regulated and down-regulated) between the CK and LO groups based on metabolomic data.

Using nontargeted metabolomics, we detected 579 metabolites in the water spider and identified 95 differential metabolites under hypoxia stress (57 up-regulated and 58 down-regulated) (Fig. 4b). Hypoxia treatment led to pronounced shifts in intermediates from core energy-producing pathways, including glycolysis and the TCA cycle. For example, Pyridoxal 5′-phosphate, O-Acetylcarnitine, Acetylphosphate, L-Homoserine, GTP, and 4-Hydroxyphenylpyruvic acid were up-regulated, whereas D-Erythrose 4-phosphate, Glutathione, and Succinic acid were down-regulated in the hypoxia group. These changes implicate coordinated remodeling of carbohydrate, lipid, protein, and nucleic acid metabolism in response to hypoxic stress.

Discussion

Clarifying the genomic changes that allow spiders, despite their terrestrial ancestry, to colonize underwater habitats is essential for understanding this extreme ecological transition. Here, we performed comparative analyses across spiders occupying terrestrial, semi-aquatic, and aquatic environments to study this evolutionary transition. The marronoid clade comprises species that inhabit a wide range of specialized microhabitats, including tundras, deserts, caves, intertidal zones, salt flats, and freshwater systems, spanning terrestrial, semi-aquatic, and aquatic lifestyles (Gorneau et al. 2023; Leggett et al. 2024). Our analyses centered on the aquatic A. aquatica, semi-aquatic spiders in the genus Desis, and phylogenetically proximate terrestrial taxa within the marronoid clade. A. aquatica is the only spider known to complete its entire life cycle underwater. Prior work has focused mainly on behavioral and physiological adaptations that support the aquatic lifestyle of water spiders (Seymour and Hetz 2011; Kang et al. 2014; Schaber et al. 2023). Here, we provided a high-quality chromosome-level genome assembly for the water spider A. aquatica. To address the long-standing lack of reference genomic resources for the semi-aquatic Desis lineage and closely related terrestrial marronoid spiders, we additionally assembled and annotated transcriptomic datasets for these taxa, capturing the critical independent evolutionary route to semi-aquatic life despite the limited number of available species. Nevertheless, we acknowledge that the availability of genomic and transcriptomic data for semi-aquatic spiders remains limited, constraining our sampling to the Desis lineage as a representative of semi-aquatic adaptations within the marronoid clade. While our dataset captures 2 independent evolutionary transitions to water-associated lifestyles, expanded sampling of additional semi-aquatic taxa will be necessary in the future to determine whether the molecular changes identified here reflect general patterns across independent origins or are specific to the lineages examined.

Morphological and molecular co-evolution underlying semi-aquatic and aquatic adaptation

Multiple independent transitions from terrestrial ancestors to aquatic and semi-aquatic niches highlight the capacity of spiders to adapt to the strong selective pressures imposed by abrupt environmental change (Leggett et al. 2024). In underwater or intertidal settings, spiders face acute risks of respiratory obstruction and drowning, and survival therefore depends on antiasphyxiation strategies. These strategies can involve remodeling of respiratory structures that originally evolved for terrestrial life, as well as behaviors that permit temporary submergence. For example, the semi-aquatic spider D. martensi retreats into air pockets within shells or rock crevices during high tide (Fig. 1a), where trapped air can sustain respiration. In contrast, fully aquatic spiders such as the water spider A. aquatica exhibit more specialized behavioral and respiratory adaptations: A. aquatica builds a silken “diving bell” anchored to submerged vegetation to retain an air supply underwater (Fig. 1a). Consistent with this lifestyle, anatomical observations show that spiracles (respiratory openings) are positioned more anteriorly in A. aquatica, potentially improving access to oxygen within the diving bell (Fig. 1b). Moreover, both D. martensi and A. aquatica display a higher density of tracheal tubes than their terrestrial relatives (Fig. 1b).

Such morphological innovations ultimately depend on molecular changes that optimize function in the new habitat. Our analyses identified gene categories that underlie key traits mediating adaptation to aquatic and semi-aquatic habitats. These include loci related to changes in the respiratory system (eg tracheal development genes), osmoregulation (eg ion transporter genes), dietary and energetic shifts (eg fatty acid metabolism genes), hypoxia responses, and thermal adaptation. Together, these candidate pathways identify core molecular processes that likely facilitated repeated transitions to aquatic-associated lifestyles. Beyond advancing our understanding of the origins and diversification of spiders, these results provide a comparative framework for investigating the molecular basis of adaptive evolution in other taxa subjected to extreme environmental change.

Divergent evolutionary trajectories in semi-aquatic or aquatic adaptation

The emergence of semi-aquatic and aquatic spiders from terrestrial ancestors has received extensive research attention. Because current taxonomy places these semi-aquatic and aquatic taxa across multiple families (Crews et al. 2020), the prevailing view is that both types of lifestyles have evolved independently from terrestrial spiders (Leggett et al. 2024). Consistent with this framework, our phylogenomic analyses further support separate evolutionary routes leading to the adaptation of semi-aquatic and aquatic lineages to underwater environments (Fig. 1d). Fossil-calibrated divergence time estimates also indicate that these transitions were asynchronous. Semi-aquatic adaptations arose earlier, with a node age spanning the Paleocene–Eocene transition (55.82 Mya, 95% HPD: 49.5-61.9 Mya), broadly coincident with wetland expansion associated with the Paleocene–Eocene Thermal Maximum. This timing is consistent with a scenario in which habitat reorganization under extreme climatic conditions promoted ecological opportunity, favoring behavioral shifts (eg extending nest-building into tidal zones) and physiological changes (eg increased salinity tolerance) that enabled exploitation of newly available niches. In contrast, aquatic specialization originated later in the Eocene (43.79 Mya, 95% HPD: 40.4–46.9 Mya), potentially in association with the diversification of lacustrine habitats.

Molecular adaptation of semi-aquatic spiders

In semi-aquatic spiders, we detected 58 genes under intensified selection, 67 genes under relaxed selection, and 127 PSGs relative to terrestrial relatives (Tables S5, S7). Although none of these sets showed significant functional enrichment (Figures S4–S7, S14–S15), consistent biological motifs emerged across categories, including respiratory metabolism, ion transport, DNA repair, and lipid processing. Together, these results suggest that semi-aquatic adaptation may rely on the targeted regulation of key loci rather than broad reorganization of entire pathways.

Several genes involved in aerobic respiration, including SdhD and SLC2A3, exhibited intensified selection, consistent with stronger functional constraint on energy metabolism under aquatic conditions. In insects, SdhD variants can influence HIF-1-regulated hypoxia responses (Marden et al. 2013), and SLC2A3 expression is induced under hypoxia in the Chinese mitten crab (Eriocheir sinensis) (Zhou et al. 2025). The repeated identification of SLC2A3 among PSGs further supports the idea that maintaining glucose homeostasis may be particularly important for the physiology of spiders in semi-aquatic habitats.

We also observed intensified selection in genes involved in osmoregulation. Cht10 (Probable chitinase 10) contributes to chitin turnover during molting and may affect cuticular permeability (Kim et al. 2024), whereas ACE (angiotensin-converting enzyme) has been implicated in regulating ion balance across freshwater and marine environments (Rankin et al. 2001). Scamp1 and AAK1, which function in vesicle trafficking and endocytosis, could influence membrane turnover and transporter recycling, processes that may be relevant for maintaining osmotic homeostasis in fluctuating semi-aquatic habitats.

DNA repair genes showed mixed selective signatures. Rad23a, which contributes to UV-induced DNA damage repair, displayed both intensified and positive selection, consistent with the possibility of elevated oxidative stress in coastal environments (Grønbæk-Thygesen et al. 2023). By contrast, relaxed selection on brcc3 may reflect reduced mutagenic pressure in specific microhabitats (Wang et al. 2024). The ion exchanger Slc24a5, identified under both intensified and positive selection, is known to influence pigmentation in vertebrates (Liu et al. 2024), raising the possibility that it may contribute to photoprotective responses under strong coastal UV exposure.

Genes involved in lipid metabolism also showed divergent selective trajectories. Both boca and Pnliprp2 exhibited signals of intensified selection and positive selection, consistent with roles in lipid processing and digestion. In Drosophila melanogaster, boca supports intracellular trafficking of LDL receptor family proteins (Culi and Mann, 2003), and Pnliprp2 contributes to lipid hydrolysis (Zhu et al. 2021). Collectively, these patterns are consistent with dietary and energetic remodeling, particularly altered lipid utilization, during the transition to semi-aquatic environments.

Molecular adaptation of the aquatic spider

For the aquatic transition, we detected 92 genes under intensified selection, 59 genes under relaxed selection, and 26 PSGs (Tables S7, S9). The aquatic water spider exhibits molecular signatures consistent with specialization to submerged life, particularly in respiratory and metabolic systems. Although RELAX predominantly indicates strengthened purifying selection rather than adaptive diversification, the implicated loci highlight processes that may have become increasingly important for maintaining physiological homeostasis underwater.

Genes involved in tracheal morphogenesis and gas exchange displayed lineage-specific shifts in selective regimes. For example, Wnt-1 and Catenin beta (relaxed selection) are core components of the Wnt/β-catenin pathway that regulates tracheal patterning (Llimargas and Lawrence 2001; Gerhardt et al. 2018; Brechbuhl et al., 2011). Likewise, TMEM234 (intensified) and TMEM59 (relaxed) have been linked to epithelial morphogenesis (Rock et al. 2008; Scholl et al. 2019), suggesting differential constraint across genes contributing to tracheal remodeling. Structural regulators such as Arp3 and ECE1 (intensified) influence actin organization and smooth muscle activity (De Campo et al. 2002; Zhang et al. 2005), whereas Syntaxin 5 and VEGF (intensified) contribute to tubular integrity and cell migration (Kipryushina et al. 2015; Bourne et al. 2022). Collectively, these patterns are consistent with strong functional constraints on respiratory architecture in the aquatic lineage, reflecting the need for efficient gas exchange when submerged.

Functional enrichment analyses (Figures S8–S13) indicated that genes under intensified selection and/or positive selection were significantly enriched in energy metabolism, mitochondrial respiration, and signaling pathways (eg MAPK, mTOR), whereas relaxed genes did not show clear patterns of enrichment. Representative intensified loci, including UQCRC2, GFPT2, and GPI, play central roles in glycolytic flux and mitochondrial respiration, consistent with heightened energetic demand under oxygen limitation. In contrast, relaxed selection on GlyP, SLC37A3, and ETFA may reflect partial reductions in dependence on strictly aerobic metabolic routes. Genes associated with DNA repair and redox homeostasis also exhibited lineage-specific signatures. For example, relaxed selection on ALKBH2 and PGBD4 may indicate shifts in mutagenic or repair-related pressures, whereas intensified selection on cytochrome P450 genes (CYP2J6, CYP3A9) is consistent with enhanced detoxification capacity under variable oxygen levels.

Transcriptomic and metabolomic profiles further support hypoxia-associated metabolic reprogramming (Fig. 4, Figures S23, S24, Tables S15, S16). For example, the accumulation of O-acetylcarnitine (FC = 2.55, P < 0.05) coupled with the depletion of glutathione (FC = 0.58, P < 0.05) suggests an energetic trade-off between β-oxidation and redox buffering. Up-regulation of SLC13A5 (padj < 0.05) may enhance the supply of substrates to mitochondrial metabolism, whereas down-regulation of SLC1A2 and HAO1 could impair amino-acid transport and redox balance. Although these associations remain exploratory, they suggest that A. aquatica coordinates respiratory, metabolic, and antioxidant programs to sustain aerobic function during prolonged submersion.

Putative genes of bacterial origin in the water spider

Our analyses detected 60 bacterial-homologous genes in the A. aquatica genome (39 ABC-related and 21 ACAD-related) distributed across multiple scaffolds. The putative donor families identified in our BLAST-based assignments are common in freshwater environments, which is consistent with a historical association between A. aquatica and freshwater bacteria. Similar microbe-arthropod associations have been reported in other systems, including deep-sea spiders (Dubilier 2025). Despite the low transcript abundance observed, the bacterial-origin ABC and ACAD genes are transcriptionally active, implying they represent a specialized genetic resource whose expression may be fine-tuned to specific life stages or environmental cues not captured in the present study.

From a functional perspective, the ABC transporter family is essential for maintaining osmotic balance and neutralizing environmental toxins-two of the most significant physiological hurdles for a terrestrial lineage colonizing freshwater (Albers et al., 2004). Similarly, the expansion of ACAD genes, which mediate fatty acid beta-oxidation, likely supports the high-energy demands of underwater movement and specialized foraging. The alignment of these genes expansion with our metabolomic findings (eg shifts in lipid-related intermediates like O-acetylcarnitine) further reinforces the central role of these pathways. The ABC and ACAD gene families, through the integration of bacterial-origin sequences and the molecular evolution of ancestral genes, provides the functional basis for the metabolic adaptations required for the aquatic lifestyle of A. aquatica.

In summary, our results shed light on the genomic changes that enable spiders to persist in underwater environments. By integrating signals of relaxed selection, intensified selection, positive selection, and putative bacterial-origin loci, we delineate molecular signatures associated with repeated transitions from terrestrial to semi-aquatic or aquatic habitats. Together, these findings refine our understanding of the evolutionary trajectories leading to semi-aquatic and aquatic lifestyles in spiders and contribute more broadly to comparative studies of adaptive strategies in aquatic organisms.

Methods

Morphological analysis of respiratory systems

Four spider species, Desis martensi, Badumna tangae, A. aquatica, and Dictyna felis, were examined to characterize the morphology of the respiratory system. All specimens were stored in 75% ethanol at the School of Life Sciences, Southwest University, Chongqing, China (SWUC). Following rinsing with ddH₂O, specimens were cleared by immersion in pancreatin (Álvarez-Padilla and Hormiga 2007). Respiratory structures, including book lungs and trachea, were observed and measured using a Leica M205A stereomicroscope fitted with a drawing tube, a Leica DFC450 camera, and LAS software v4.6. Additional photographs were captured with a Kuy Nice CCD camera attached to an Olympus BX53 compound microscope. Extended depth-of-field images were produced with Helicon Focus 6.7.1 software.

Sample collection

Water spider A. aquatica individuals were collected from Xilingol League, Inner Mongolia, China (Figure S1). One female Desis martensi was collected from Sanya, Hainan Province, China (Figure S1).

Genome sequencing

Ten adult female water spiders were prepared for whole-genome sequencing and 10 for Hi-C sequencing, with both groups prepared uniformly. Briefly, spiders were cleansed, dissected to remove the abdomen and stomach, processed for whole-genome sequencing (Nanopore and Illumina platforms), and stored at −80 °C.

A combination of 2 sequencing approaches including short reads on MGISEQ 2000 platform and long reads on Oxford Nanapore Technology (ONT) PromethION platform (Nextomics Bioscience, Wuhan, China) had been performed for the whole genome sequencing. Specifically, the total genomic DNA for genome sequencing was extracted using Qiagen Blood & Cell Culture DNA Mini Kit. For short reads, the DNA libraries were prepared with the MGIEasy FS DNA Library Prep Set (BGI, Shenzhen, China) following the MGISEQ protocol, a total of 212.12 Gb from 2.13 billion clean reads were obtained. For long reads, the ONT libraries were prepared following the protocol of Ligation Sequencing Kit 1D (SQK-LSK108, ONT), the N50 of the raw nanopore reads length was 25 kb.

An additional Hi-C sequencing was performed (Berry Genomics, Beijing, China). After cross-linked long-distance physical interactions and digested with MboI, the 300 bp paired-end sequencing libraries were prepared for sequencing on an Illumina HiSeq PE150 platform. The details are shown in Table S12.

An additional 8 adults water spiders (4 females and 4 males) were processed for whole genome resequencing with short reads on MGISEQ 2000 platform.

RNA sequencing

One female D. martensi was subjected to a 3-d starvation treatment and then used for transcriptome sequencing. For the water spider, transcriptome sequencing was conducted on multiple tissue types (including venom, muscle, trachea, testis, and ovary) as well as embryos collected at several developmental stages (3, 5, 6, and 19 days post fertilization); each condition included three replicates. Total RNA was isolated using TRIzol (Invitrogen) following the manufacturer's protocol. Libraries were sequenced on an Illumina NovaSeq 6000 platform (Berry Genomics, Beijing, China) to generate 150-bp paired-end reads.

Genome assembly

The water spider genome was assembled in Flye v2.5 (Kolmogorov et al. 2019) using a single round of polishing iterations and a minimum read overlap of 3,000. The draft assembly was then refined with 2 polishing rounds using NGS data in NextPolish v1.0.5 (Hu et al. 2020). Heterozygous regions were subsequently purged with Purge Haplotigs v1.1.0 (Roach et al. 2018), applying a 50% cutoff for haplotig identification. A chromosome-level assembly was produced by scaffolding with Hi-C reads using Juicer v1.6.2 (Durand et al. 2016). Potential contaminant sequences were screened by querying the assembled genome against the NCBI nucleotide (nt) and UniVec databases using HS BLASTN and BLAST + (blastn) v2.7.1 (Chen et al. 2015). Assembly quality and completeness were finally evaluated with BUSCO v5.4.5 (Simao et al. 2015) using the arthropoda_odb10 dataset (n = 1,013) and the arachnida_odb10 dataset (n = 2,934).

Genome annotation

To annotate repetitive elements in the genome, we applied a combined strategy integrating ab initio discovery and homology-based searches. First, we built a species-specific repeat library (ab initio database) with RepeatModeler v2.0.1 (Flynn et al. 2020), and then merged this library with the curated repeat collection in Repbase to create a composite reference repeat database. Genome-wide repeats were subsequently identified using RepeatMasker (Tarailo-Graovac and Chen 2009). Noncoding RNAs were identified with Infernal v1.1.2 (Nawrocki and Eddy 2013) and tRNAscan-SE v2.0.6 (Chan et al. 2021). We additionally applied the “EukHighConfidenceFilter” module in tRNAscanSE to retain high-confidence transfer RNA (tRNA) predictions.

For gene structure annotation, we combined ab initio predictions, transcriptome-supported evidence, and protein homology using Maker v2.31.10 (Holt and Yandell 2011). Ab initio gene models were generated with Augustus v3.3.3 (Hoff and Stanke 2019) and GeneMark-ES/ET/EP v4.48 (Bruna et al. 2020). To optimize model parameters, both predictors (Augustus and GeneMark-ES/ET/EP) were first trained using the BRAKER v2.1.5 pipeline (Bruna et al. 2021). Whole-body water spider transcriptome reads were aligned to the genome using HISAT2 v2.2.0 (Kim et al. 2015), and the BRAKER pipeline was run with default parameters. RNA-Seq alignments were then assembled into candidate transcript models with Stringtie v2.1.3 (Pertea et al. 2015), and these transcript predictions were supplied to Maker through the “est” option. For protein homology support, we downloaded protein sequences from the NCBI database for Stegodyphus mimosarum (GCA 000611955.2), Parasteatoda tepidariorum (GCA 000365465.3), Trichonephila clavipes (GCA 002102615.1), Drosophila melanogaster (GCA 000001215.4), Ixodes scapularis (GCA 002892825.2), Daphnia pulex (GCA 900092285.2), and Strigamia maritima (GCA 000239455.1); these sequences were concatenated and provided to the Maker pipeline.

Functional annotation of predicted genes was performed with Diamond v0.9.24 (Buchfink et al. 2015), InterProScan v5.41–78.0 (Mulder and Apweiler 2007), and eggNOG-mapper v2.0 (Cantalapiedra et al. 2021). Using Diamond in “more sensitive” mode, we searched UniProtKB/SwissProt with a maximum of one target hit per query and an e-value threshold of 1e-5. InterProScan was run against 5 domain resources: Pfam (Mistry et al. 2021), Panther (Mi et al. 2016), Gene3D (Lewis et al. 2018), Superfamily (Pandurangan et al. 2019), and the Conserved Domain Database (CDD) (Marchler-Bauer et al. 2017). Finally, eggNOG-mapper was used to query the eggNOG v5.0 database (Huerta-Cepas et al. 2019).

In addition to the newly generated omics datasets, we downloaded publicly available genomes for terrestrial spiders, including Amaurobius ferox (GCA_951213105.1) (Henriques 2024), Dolomedes plantarius (GCA_907164885.2) (Darwin Tree of Life Project Consortium 2022), and Uloborus diversus (GCF_026930045.1) (Miller et al. 2022). Because annotation files for A. ferox and D. plantarius were not publicly available, we performed de novo annotation for these 2 species, yielding 14,258 proteins for A. ferox (BUSCO completeness 74% and 76%) and 76,921 proteins for D. plantarius (80.6% arthropod conserved genes; Table S5).

RNA-seq assembly and annotation

We retrieved transcriptomic datasets for multiple spider species from the NCBI Sequence Read Archive (SRA), including the semi-aquatic spider Desis japonica (DRR296513) and the terrestrial spiders Metaltella simoni (DRR296491), Badumna longinqua (DRR297554), Adenodictyna kudoae (DRR297439), Dictyna felis (DRR297156), Brommella punctosparsa (DRR296607), Lathys maculosa (DRR297435), Cybaeus nipponicus (DRR297040), Cicurina japonica (DRR297872), Hahnia corticicola (DRR296643), Neoantistea quelpartensis (DRR297193), Agelena silvatica (DRR297956), Pireneitega luctuosa (DRR297239), Tegenaria domestica (DRR297377), Ambicodamus sp. (DRR297731), Pandava laminata (DRR297807), and Nurscia albofasciata (DRR296579) (https://www.ncbi.nlm.nih.gov/sra; Table S1). We integrated these types of RNA-seq data, including newly sequenced data for D. martensi and publicly available data retrieved from the NCBI SRA database. We further performed de novo transcriptome assembly for these spider species with RNA-seq data. Finally, we used these assembled datasets for ortholog identification across spider species and phylogenomics analysis, but not for differential expression unless otherwise specified. Raw reads were subjected to quality control using the BBTools suite v38.67 with “bbduk.sh,” trimming read ends to Q20 and removing reads shorter than 15 bp or containing >5 Ns. Cleaned reads were then assembled de novo with Trinity v.25.1 (Grabherr et al. 2011). Redundant contigs were filtered using cd-hit-est v4.8.1, and assembled transcripts were functionally annotated with Trinotate v3.0.1 (Bryant et al. 2017).

For RNA-seq datasets derived from water spider tissues and embryos at different developmental stages, we mapped sequencing reads to the water spider reference genome using HISAT2. Gene-level read counts were generated with featureCounts v1.6.4 (Liao et al. 2014), and expression values were normalized across samples as FPKM using the DESeq2 package v1.24.0 (Love et al. 2014).

Phylogenetic tree construction and divergence time estimation

Proteomes from the water spider A. aquatica and taxa representing the marronoid clade were compiled for gene cluster analyses. The dataset included 8 marronoid families: Amaurobiidae (Amaurobius ferox), Agelenidae (Agelena silvatica, Pireneitega luctuosa, Tegenaria domestica), Desidae (Desis japonica, D. martensi, Metaltella simoni, Badumna longinqua), Hahniidae (Hahnia corticicola, Neoantistea quelpartensis), Cybaeidae (Cybaeus nipponicus), Cicurinidae (Cicurina japonica), and Dictynidae (Adenadictyna kudoae, Dictyna felis, Lathys maculosa). We also incorporated 5 closely related families, including Cicurinidae (Brommella punctosparsa), Nicodamidae (Ambicodamus sp.), Titanoecidae (Pandava laminata, Nurscia albofasciata), Pisauridae (Dolomedes plantarius), and Uloboridae (Uloborus diversus) as the outgroup. Orthologous gene clusters were inferred using OrthoFinder v2.5.4 (Emms et al. 2019), resulting in single-copy orthologs and additional shared orthologous groups that were used in downstream analyses. We also extracted benchmarking single-copy orthologs from genomic and transcriptomic datasets based on BUSCO outputs, and protein sequences from 941 shared single-copy orthologs were used to construct the phylogenetic tree.

Protein sequences for each ortholog were aligned individually using MAFFT v7.471 (Katoh and Standley 2013) under the L-INS-I strategy. Alignments were subsequently trimmed with trimAl v1.4 (Capella-Gutierrez et al. 2009) using the “automated1” option, and the filtered alignments were concatenated with FASconCATG v1.04 (Kuck and Longo 2014). A maximum-likelihood (ML) phylogeny was then inferred with IQ-TREE v2.0.3 (Minh et al. 2020), using the rcluster algorithm for optimization; the numbers of replicates for ultrafast bootstrap and Shimodaira-Hasegawa (SH) approximate likelihood ratio tests were set to 1,000 and 1,000, respectively, and the optimization number was set to 10.

Divergence time calibration relied on Magalhaes (Moradmand et al. 2014; Magalhaes et al. 2020), the paleobiodb database (https://paleobiodb.org/), and the TimeTree database (http://www.timetree.org/), using constraints for Hahniinae (43–47.8 Mya), Dictynidae (100–189.7 Mya), and Entelegynae (132.9–140 Mya). Divergence times were estimated with MCMCTree in PAML v4.10.0 (Yang 2007) using 2 independent MCMC runs of 50,000 generations each, sampling every 5 generations after discarding the first 20% as burn-in.

Pattern of molecular evolution in shared orthologs

Two datasets were analyzed in this study. The first dataset represents the semi-aquatic spider evolutionary pathway (evolutionary route 1) and includes semi-aquatic and terrestrial spiders (Fig. 1c). The second dataset represents the aquatic spider evolutionary pathway (evolutionary route 2) and comprises aquatic and terrestrial spiders. To quantify patterns of molecular evolution among terrestrial, semi-aquatic, and aquatic taxa across the semi-aquatic and aquatic branches of the phylogeny, we estimated the nonsynonymous-to-synonymous substitution rate ratio (dN/dS) for each shared ortholog. Protein sequences were aligned using MAFFT v7.471 (Katoh and Standley 2013). Codon alignments were then generated from the amino-acid alignments together with the corresponding coding sequences using PAL2NAL v.14 (Suyama et al. 2006). We filtered ortholog alignments to retain those with lengths of at least 50 codons. Discrete codon substitution rates (dN/dS, ω) for water spider, semi-aquatic spider, and terrestrial spider lineages were estimated in HyPhy (Pond et al. 2005) following a previously described pipeline (Tong et al. 2022).

We next evaluated gene-wide shifts in selective strength associated with the semi-aquatic and aquatic transitions using RELAX (Wertheim et al. 2015). Genes were classified as experiencing significant relaxed selection when K < 1 (q < 0.05) and as showing significant intensified selection when K > 1 (q < 0.05). For each gene, q-values were computed as false discovery rate (FDR) estimates using the Benjamini–Hochberg procedure.

Finally, we tested for positive selection on genes associated with semi-aquatic and aquatic transitions using BUSTED-PH (https://github.com/veg/hyphy-analyses/blob/master/BUSTED-PH/README.md) (Murrell et al. 2015), with multiple-testing correction applied using the Benjamini–Hochberg procedure. We defined genes under positive selection by requiring q < 0.05 on the foreground branches (ie the aquatic spider A. aquatica, or the semi-aquatic spiders D. martensi and D. japonica), q > 0.05 on the background branches (ie 19 terrestrial spiders), and q < 0.05 for the foreground–background difference.

ABC and ACAD gene family identification

To manually curate ABC and ACAD genes, we first downloaded amino acid sequences of ABCs and ACADs predicted from D. melanogaster and Bombyx mori from NCBI. Using these references, we searched for ABC and ACAD family members in the genomes of the water spider, A. ferox, and D. plantarius.

ABC and ACAD candidates across the 4 genomes were identified with MMseqs2 v13.45111 using a blastp-like search (4 iterative rounds; e-value = 0.001). Candidate proteins were then screened with InterProScan to confirm the presence of family-specific conserved domains based on Pfam. The HMMER-Pfam classifications and functional assignments were further validated via online NCBI BLASTP searches against the nonredundant (nr) GenBank protein database. For each protein, intron-exon boundaries were inspected. Multiple hits mapping to the same genomic region were consolidated, and spurious matches were removed by re-querying with MMseqs2 using a tblastn-like search to exclude invalid protein alignments.

We next reconstructed phylogenetic trees for ABC and ACAD genes from the water spider, A. ferox, and D. plantarius. Protein sequences were aligned in MAFFT using default parameters, and poorly aligned or ambiguously homologous sites were filtered with trimAl using the heuristic option “automated1.” Gene trees were inferred in IQ-TREE under the LG substitution model with 1,000 ultrafast bootstrap replicates. Resulting phylogenies were visualized, customized, and annotated using the Evolview online tool (Subramanian et al. 2019). During online NCBI BLASTP validation, we additionally observed that some water spider ABC and ACAD proteins showed high similarity to bacterial sequences. We therefore downloaded the corresponding bacterial protein sequences for subsequent analyses.

Bacterial-origin gene identification and validation

To evaluate whether specific genes in the water spider originated from symbiotic bacteria, we applied a conservative phylogeny-based framework following Li et al. (2022).

First, we reconstructed phylogenies for the ABC and ACAD gene families. For each gene set, we combined protein sequences from A. aquatica, reference sequences from model arthropods (Drosophila melanogaster, Daphnia pulex, and Bombyx mori), and bacterial homolog proteins showing the highest similarity to candidate bacterial-origin genes in A. aquatica. Protein sequences were aligned in MAFFT using default parameters, and the resulting alignments were trimmed with trimAl to remove ambiguously aligned sites using the heuristic option “automated1.” Gene trees were then inferred with IQ-TREE under the LG substitution model with 1,000 ultrafast bootstrap replicates. In the resulting phylogenies, putative bacterial-origin genes and their putative donor bacterial sequences clustered within the same clade (Figures S20, S21).

Next, we examined ABC and ACAD expression patterns across multiple tissues (including venom, muscle, trachea, testis, and ovary) and developmental stages (3, 5, 6, and 19 d postfertilization) in the water spider to evaluate their potential biological relevance. For experimental validation, we selected 9 putative bacterial-origin genes (6 ABC genes and 3 ACAD genes). We designed PCR assays to amplify fragments spanning the flanking regions of each foreign gene, targeting amplicons of 700–1,500 bp. PCR products were evaluated by agarose gel electrophoresis to confirm expected sizes, and successful amplicons were Sanger-sequenced. A gene was considered validated when the flanking regions were successfully amplified and the Sanger reads showed >98% identity to the reference genome. Primer sequences are listed in Table S13. We also mapped the distribution of bacterial-origin genes on scaffold 1 of the water spider genome using the MG2C v2.1 online tool (Chao et al. 2021). Finally, to further support the presence of bacterial-origin genes in the genome, we quantified read support for these loci across 6 water spider whole-genome resequencing datasets using kallisto v0.48.0 (Bray et al. 2016).

RT-PCR validation of bacterial-origin genes

To verify the expression of putative bacterial-origin ABC and ACAD genes, we performed RT-PCR using a mixed cDNA pool derived from larvae, sub-adults, and adults of A. aquatica. Total RNA was extracted using FastPure Cell/Tissue Total RNA Isolation Kit V2 (Vazyme) and reverse-transcribed into cDNA. Specific primers (Table S14) were designed for 9 candidate loci. PCR was performed with 2× Taq Master Mix (Dye Plus) under the following conditions: 95 °C for 3 min, followed by 34 cycles of 95 °C for 30 s, 57 °C for 30 s, and 72 °C for 1 min, with a final extension at 72 °C for 10 min. Products with weak bands were subjected to a second round of PCR amplification to enhance the signal. The amplified products were purified and for bidirectional Sanger sequencing. By integrating sequencing read mapping results with RT-PCR and Sanger sequencing data, we validated the physical presence of the relevant genes on the specific scaffold and their corresponding transcript signals.

Hypoxia stress treatment

For the hypoxia stress experiment, the lung slits and spiracles of adult male water spiders were sealed with paraffin, and the animals were maintained in boxes at room temperature. After 48 h, spiders were flash-frozen in liquid nitrogen, the paraffin was removed, and samples were stored at −80 °C. Control individuals were handled identically except that no paraffin sealing was applied, and they were likewise kept in boxes at room temperature. Additional procedural details are provided in Figure S20. Samples were subsequently used for liquid chromatography–mass spectrometry (LC–MS) analysis and transcriptome sequencing (Suzhou PANOMIX Biomedical Tech, China). For LC–MS, each treatment group included 8 replicates, whereas transcriptome sequencing was performed with 3 replicates per group.

Differential gene expression analysis

To characterize transcriptional responses to hypoxia, we conducted a comparative transcriptomic analysis between spiders exposed to hypoxia stress and control spiders maintained under normal oxygen conditions. RNA-seq data were processed using a reference genome-guided workflow. Transcript abundance was quantified as RPKM, and differential expression was assessed using the DESeq2 package v1.24.0. Genes were defined as differentially expressed when the adjusted P-value was <0.05 and |log2ratio| ≥ 1. Volcano plots were generated in R using the “ggplot2” package.

Differential metabolite analysis

To identify metabolites altered by hypoxia treatment, we performed nontargeted LC-MS metabolomics (Suzhou PANOMIX Biomedical Tech, China) comparing spiders exposed to hypoxic versus normal oxygen conditions. Sample preparation was as follows: 1 g of tissue was precisely weighed and combined with 1,000 µL of tissue extraction solvent (75% methanol [9:1], 25% H₂O) and 3 steel balls in a 2 mL centrifuge tube. Samples were homogenized in a tissue grinder at 50 Hz for 60 s, and the homogenization step was repeated twice. Homogenates were then subjected to ultrasonic agitation for 30 min, followed by incubation in an ice bath for 30 min. After centrifugation at 12,000 rpm (13,400 × g) for 10 min at 4 °C, the supernatant was collected and evaporated to dryness. Dried extracts were reconstituted in 200 µL of 50% acetonitrile containing 2-amino-3-(2-chlorophenyl)-propionic acid (4 ppm). The reconstituted solution was passed through a 0.22 μm membrane filter and transferred to LC-MS vials for analysis. Metabolite profiles between groups were compared using multivariate and univariate approaches, including principal component analysis, clustering, and univariate testing. Volcano plots highlighting significantly changed metabolites were generated in R using the “ggplot2” package.

Author contributions

C.T., M.B. and Z.Z. designed the original research. Z.F. and C.T. performed data analysis and drafted the manuscript. B.L., L.S., T.R. and P.L. performed sample preparation and data analysis. L.W., J.G., L.C., and B.T. collected the samples and performed sample preparation. Y.C. drew the illustration. Q.H., M.D., Q.Z., X.Z., J.L., N.L. and M.I. performed formal analyses and visualization of the data. All authors revised and approved the final manuscript.

Supplementary material

Supplementary material is available at Molecular Biology and Evolution online.

Funding

This work was supported by National Key R&D Program of China (No. 2025YFE0211201), the National Science & Technology Fundamental Resources Investigation Program of China (Nos. 2025FY101101, 2025FY100802), Science & Technology Fundamental Resources Investigation Program (No. 2023FY100301), Innovative Talents International Cooperative Training Project of China Scholarship Council (No. CXXM20230124) by Ming Bai. This research is also supported by the National Natural Sciences Foundation of China (No. 32570539), the Science Foundation of School of Life Sciences SWU (Nos. 20232008071901 and 20212020110501) and the Natural Science Foundation of Chongqing (No. cstc2019jcyj-zdxmX0006) by Zhi-Sheng Zhang.

Data availability

For water spider A. aquatica, the raw Nanopore data are available on NCBI with the project of PRJNA559680 with the SRA accession numbers of SRR9994193, and the whole-body RNA accession number is SRR9994194. The other data of A. aquatica were submitted in scienceDB dataset: the genome assembly (https://cstr.cn/31253.11.sciencedb.14858), the raw Illumina data (https://cstr.cn/31253.11.sciencedb.11012), the Hi-C data (https://doi.org/10.57760/sciencedb.15431), the metabolites and transcriptomic raw data (https://cstr.cn/31253.11.sciencedb.15069), the RNA data for different stage of embryo (https://cstr.cn/31253.11.sciencedb.10923), and the RNA data of different tissues (https://cstr.cn/31253.11.sciencedb.15029). For the RNA reads of a female Desis martensi is also available on scienceDB (https://cstr.cn/31253.11.sciencedb.09032).