Comparative single-cell analysis of black rockfish (Sebastes schlegelii) ovarian stroma reveals placenta-like adaptations toward viviparity

Abstract

The black rockfish (Sebastes schlegelii) is a viviparous teleost with high fecundity and prolonged female sperm storage for about six months despite lacking a specialized storage organ. During gestation, its ovarian stroma undergoes a functional shift from supporting oocyte growth to maintaining embryo development, resembling the maternal decidua in mammals. Here, we present a single-cell atlas of the ovarian stroma across unfertilized, normal-gestation, and abnormal-gestation states. Cross-species integration with human and mouse datasets reveals convergence on decidualization-like programs during early pregnancy, while highlighting lineage-specific implementation by distinct cell types. We further implicate annexins in ovarian sperm storage and show that abnormal-gestation is associated with lymphatic endothelial skewing and lipid-handling signatures, consistent with stromal lipid recapture. Together, these findings provide a framework for the maternal contribution to teleost placentation and reveal convergent, placenta-like adaptations underlying viviparity.

Single-cell analysis of black rockfish ovarian stroma reveals placenta-like cellular and molecular adaptations, highlighting convergent mechanisms underlying the evolution of viviparity in teleosts.

Introduction

The transition from external to internal embryogenesis in vertebrates—viviparity—is a repeatedly evolved shift from yolk-based nutrition (lecithotrophy) to maternal provisioning (matrotrophy)¹. Far from a simple retention of eggs within the body, this transition demands a profoundly coordinated reprogramming of maternal endocrine, immune, vascular, and stromal systems to support embryonic growth while maintaining physiological homeostasis². As one of the ultimate forms of parental investment, this orchestrated state shields offspring from environmental extremes and predation, albeit at the expense of classic life-history trade-offs such as reduced fecundity, limited dispersal, and elevated energetic costs^3,4. Across fishes, reptiles, and mammals, viviparity has arisen independently many times, which reflects strong convergent selection for intra-maternal embryonic protection^5,6. These evolutionary transitions clearly required major changes in the morphology and physiology of the reproductive tract. To accommodate these demands, the placenta evolved as a transient organ that enables sustained maternal–fetal exchange while preserving maternal homeostasis. Following Mossman, the placenta is defined as the functional apposition or fusion of fetal membranes with the maternal reproductive tract that enables sustained physiological exchange between mother and offspring⁷.

Beyond this functional definition, placentas exhibit extraordinary morphological diversity. While mammals typically possess highly specialized chorioallantoic placentas characterized by intimate maternal–fetal blood apposition or fusion⁸, structurally diverse analogs have evolved independently across vertebrates. These include specialized omphaloplacentas in reptiles^9,10, oviductal “uterine-milk” adaptations in amphibians^11,12, and yolk-sac placentas in sharks¹³. Viviparous teleosts, however, present a fundamentally unique evolutionary paradigm: they lack a dedicated uterus entirely, and embryogenesis occurs in situ within the ovary. Consequently, to sustain this prolonged internal development, the ovary itself, often in coordination with transient embryonic absorptive organs, must undergo a profound functional transition to act as a placenta. Historically, research on teleost placentation has predominantly focused on striking embryonic adaptations, including trophotaenia, vascularized fins, and gills, that interface with a hypertrophied maternal epithelium^14–18. In contrast, the regulatory mechanisms governing the maternal compartment itself remain remarkably underexplored. This knowledge gap is particularly glaring in species lacking conspicuous embryonic specializations. In such cases, the maternal ovarian stroma must serve as the sole source of post-fertilization provisioning and immunological protection. This raises a fundamental evolutionary and physiological question: How does the maternal ovarian stroma functionally reprogram to sustain prolonged gestation, and do these maternal adaptations functionally converge with mammalian decidualization paradigms?

The black rockfish serves as an unparalleled model to address these questions. As a highly fecund viviparous teleost, it nourishes several hundred thousand embryos over a 1-month gestation period without obvious embryonic placental specializations. Importantly, its matrotrophy index (MI) is approximately 1.8, indicating that embryos acquire nearly twice their initial yolk energy through maternal provisioning¹⁹. These observations imply a massive post conception state transition within the ovarian stroma—from an oocyte-supporting niche to a placenta-like environment. Adding to this complexity, the ovarian stroma of the black rockfish is also specialized for sperm storage for ~6 months (November to April)^20,21. Notably, sperm are activated upon entry into the ovarian lumen and are actively maintained by the ovarian stroma²². This implies remarkable, stage-specific functional plasticity: pre-fertilization, the stroma supports long-term, low-damage sperm maintenance; post-fertilization, the same compartment transitions towards a decidual/placental-like phenotype that enables nutrient transfer, immune-stromal remodeling, and vascular reprogramming. Together, these features make the black rockfish a compelling model for resolving the maternal programs that underwrite teleost placentation and understanding the broader evolutionary trajectories of vertebrate viviparity.

Beyond its biological significance, the black rockfish also holds considerable economic value as a commercially important marine species in East Asia. However, in aquaculture practice, high rates of embryonic mortality, particularly intraovarian fetal death, remain a major obstacle that severely limits the sustainability of the industry. Combined with our anatomical observations, these reproductive failures are thought to be closely associated with placental insufficiency and maternal–fetal interaction defects (e.g., reduced ovarian vascularization in abnormal-gestation), underscoring the need to better understand the maternal contributions to pregnancy maintenance in this species. Accordingly, elucidating the mechanisms that govern maternal placental functions in the black rockfish is essential not only to fill a critical gap in our understanding of viviparous teleost placentation and its evolution, but also to provide a biological foundation for mitigating reproductive failures and improving aquaculture productivity.

To date, investigations into the placenta-like structures of the black rockfish have been largely restricted to histological observations and bulk transcriptomics^23,24. While informative, bulk transcriptomics approaches inherently obscure the massive cellular heterogeneity, rare populations, and intricate intercellular crosstalk driving gestation. Single-cell RNA sequencing (scRNA-seq) offers a powerful resolution to these limitations. Although recent scRNA-seq studies have profiled the ovaries of several teleost species^25–27, these foundational atlases have predominantly focused on oocyte development and maturation. Consequently, the non-germline compartment, specifically the complex ovarian stroma composed of dynamic immune, endothelial, and fibroblast lineages, has remained largely overlooked. This knowledge gap is particularly critical for viviparous teleosts such as the black rockfish, where the stroma must functionally transform to orchestrate prolonged sperm storage, continuous embryo nourishment, and maternal–fetal immune tolerance.

Here, we apply scRNA-seq to define the single-cell landscape of the ovarian stroma in the viviparous teleost black rockfish across unfertilized, normal-gestation, and abnormal-gestation states. To our knowledge, this constitutes the first single-cell atlas of the ovarian stroma in a viviparous teleost across distinct reproductive states. Comparing conditions, we delineate pregnancy-associated shifts in stromal cell composition and function and reconstruct the cell-cell communication networks that maintain a viable ovarian gestational niche. Cross-species integration with human and mouse decidual compartments highlights conserved versus lineage-specific maternal programs, and our analyses predict sperm-stroma interactions that may underlie ovarian sperm storage. Together, these analyses provide a comprehensive framework for the maternal contributions to teleost placentation.

Results

Single-cell atlas of black rockfish ovarian stroma

To understand how the maternal ovarian stroma is cellularly reprogrammed to support intraovarian embryonic nutrition, and to identify the microenvironmental defects that lead to gestation failure, we profiled ~65,000 stromal cells spanning unfertilized, normal-gestation, and abnormal-gestation states (Figs. 1C and S1A–C). By using established markers, 14 ovarian stromal cell types were identified: macrophages (e.g., c1qa and mrc1b); endothelial cells (e.g., eltd1, cdh5 and vwf); perivascular cells (e.g., rgs5, cald1 and tpm1); granulosa cells (e.g., fshr, cyp19a1 and inha); theca cells (e.g., anpep and igfbp5); fibroblasts (e.g., fst, ecm1 and angptl2); epithelial cells (e.g., cldn3 and cldn7); T cells (e.g., trac, cd28, and cd3e); B cells (e.g., ebf1 and pax5); erythroid cells (e.g., hbz and hbbe2); dendritic cells (e.g., xcr1 and cd8b); mast cells (e.g., tpsab1 and il1rl1); neutrophils (e.g., alox5ap and ncf4) and oocyte-like cells (e.g., h1foo and zp4) (Fig. 1F; Supplementary Data 3). Using the normal-gestation dataset as a reference, we then performed anchor-based integration separately for the unfertilized and abnormal-gestation datasets. This approach prevented algorithmic over-correction, accurately preserving the massive structural shifts and biological heterogeneity inherent to these distinct reproductive phases (Fig. 1D, E).

Fig. 1. Integrated single-cell atlas of black rockfish ovarian stroma across unfertilized, normal-gestation, and abnormal-gestation states.

A Histological comparison of unfertilized, normal-pregnancy, and abnormal-pregnancy ovaries under gross macroscopic inspection (top row) and light microscopy (bottom row). Scale bars: 200 μm. B Transmission electron microscopy (TEM) comparing granulosa cell ultrastructure. Top (Blastula stage): Granulosa cells exhibiting apoptotic hallmarks, including severe perinuclear chromatin condensation and extensively dilated endoplasmic reticulum (white arrowheads). Bottom (Unfertilized): Healthy granulosa cells displaying an intact nucleus with evenly distributed euchromatin. Scale bars: 2 μm (left panels), 500 nm (top right), and 1 μm (bottom right). C Study workflow. A total of 65,055 high-quality cells were retained for single-cell RNA-seq integration and downstream analysis. D UMAP embedding of the normal-gestation and unfertilized datasets shown as a mixed integration (left) and split by condition (top right). Each dot is a single cell; colors denote annotated cell types. The bottom-right bar plot shows the average cell-type composition for each condition. E UMAP embedding of the normal-gestation and abnormal-gestation datasets shown as a mixed integration (left) and split by condition (top right), with colors indicating cell types. The bottom-right bar plot summarizes the average cell-type composition per condition. F Dot plot of marker gene expression across annotated cell populations. Dot size represents the fraction of cells expressing the gene within a cluster; color intensity reflects the scaled average expression. UMAP uniform manifold approximation and projection, zp zona pellucida, grc granulosa cell, n nucleus, ld lipid droplet, Epi epithelial cell, Endo endothelial cell.

To determine how the local follicular support system adapts to the onset of pregnancy, we first examined the cellular and structural remodeling within the follicle. A defining feature of the transition into gestation was the massive restructuring of the follicle support system. By comparing unfertilized and normal-gestation samples, we observed that granulosa cells were rarely detected in the gestational dataset, with theca cells emerging during gestation (Figs. 1D and S1D). Consistent with this transcriptomic cellular shift, transmission electron microscopy (TEM) of ovarian follicles at the embryonic blastula stage provided direct ultrastructural validation of granulosa cell apoptosis. Specifically, these cells exhibited classical apoptotic hallmarks, including perinuclear chromatin margination and condensation, disorganization of the endomembrane system (characterized by dilated endoplasmic reticulum [ER] and fragmented Golgi apparatus), cytoplasmic condensation, irregular cell contours, and the formation of apoptotic bodies (Fig. 1B). This indicates that the onset of gestation triggers a programmed clearance of the pre-ovulatory granulosa layer.

We next asked whether broader tissue remodeling occurs beyond these local follicular changes. By comparing the relative cell-type proportions between normal-gestation and both the unfertilized and abnormal-gestation states (Fig. 1D, E), we found that gestation induces a profound global reallocation of immune and vascular populations. Specifically, permutation-based statistical testing (Fig. S1D, E) revealed distinct and significant compositional shifts. After the onset of gestation, the relative proportions of macrophages, endothelial cells, epithelial cells, perivascular cells, and dendritic cells generally increased, whereas lineages such as T cells, B cells, and neutrophils decreased (false discovery rate (FDR) < 0.05). During normal-gestation, macrophages constituted the most abundant stromal lineages, followed by fibroblasts and endothelial cells; by contrast, in the unfertilized and abnormal-gestation states, the most represented lineages were T cells and epithelial cells, respectively, and endothelial cells were reduced in both (Fig. 1D, E). Together, this baseline atlas reveals a global reallocation of immune, vascular, and stromal lineages after gestation onset, providing a strong rationale for the targeted dissection of the immune–vascular modules that distinguish successful from failed gestation.

Macrophages contribute to stromal angiogenesis during gestation

Having mapped the global cellular shifts, we next asked which specific stromal immune–vascular programs are essential for pregnancy maintenance. Gross inspection and light microscopy revealed that normal-gestation ovaries contained a richly vascularized stroma supporting well-developed embryos. In contrast, abnormal-gestation ovaries exhibited a sparsely vascularized stroma with predominantly degenerated embryos; furthermore, their follicular contents appeared as small clumps and lost normal morphology (Figs. 1A and S1F). Consistent with these tissue-level differences, single-cell profiling further indicated that macrophages were the most abundant stromal lineage during normal-gestation, and endothelial cells displayed activation consistent with the observed vascular phenotype (Figs. 1D, E and S1D, E). Driven by this phenotypic and transcriptomic correlation, we focused first on the macrophage and endothelial compartments to define their states, interactions, and contributions to stromal vascularization and gestational outcome.

To determine how macrophages functionally adapt to the gestational environment, we first compared their subpopulation dynamics before and after fertilization. Subclustering revealed that post-fertilization macrophage (Mφ) composition shifted toward pro-angiogenic, tissue-resident states after fertilization. Comparing unfertilized and normal-gestation ovaries by subclustering revealed that post-fertilization macrophages were dominated by ANGPTL4⁺ Mφ (M2/TRM-like), followed by CD40LG⁺ Mφ (M2/TRM-like), PTGES⁺ Mφ, CXCR5⁺ Mφ, and CENPF⁺ Mφ (a proliferating subset). In contrast, unfertilized ovaries were enriched for MERTK⁺ Mφ and TNF⁺ Mφ (Supplementary Data 1 and 3). Monocle2 pseudotime ordering together with CytoTRACE-based developmental-potential scoring placed unfertilized-state macrophages upstream, with higher CytoTRACE scores in unfertilized clusters that declined along trajectories toward gestational macrophage states (Figs. 2A and S2A–C). Subcluster-resolved inspection at single-gene resolution showed that gestational macrophages preferentially expressed angiogenesis-associated programs (e.g., vegfa, kdr, tnf, flt1, pdgfb, sod2) (Fig. 2B). To quantitatively validate this observation, we computed an Angiogenesis Module Score (Supplementary Data 9), confirming that gestation-enriched macrophages (e.g., ANGPTL4+, PTGES+, and CD40LG+ subclusters) displayed a significantly stronger pro-angiogenic transcriptomic program than unfertilized baseline states (P < 0.0001; Fig. S2G).

Fig. 2. Macrophages contribute to stromal angiogenesis during gestation.

A UMAP embedding of the macrophage compartment subset from the unfertilized and normal-gestation datasets, colored by macrophage subclusters. The black arrow marks the inferred transition trajectory from unfertilized-enriched to gestation-enriched macrophage states. The z-axis represents UMAP_3 in the three-dimensional embedding. B UMAP feature plots illustrating the normalized expression and spatial distribution of representative, individual angiogenesis-associated genes (vegfa, flt1, kdr, pdgfb, tnf, sod2) across the macrophage compartment. Color intensity indicates the normalized per-cell expression level. C Pseudotime-ordered heatmap of differentially expressed genes (DEGs) across the macrophage trajectory. Rows represent individual genes, and the x-axis (columns) explicitly represents individual cells ordered sequentially by Monocle-inferred pseudotime. Genes are hierarchically grouped into four modules. Representative enriched pathways for each module are annotated. D Aggregated incoming/outgoing signaling strength across unfertilized, normal-gestation, and abnormal-gestation samples. Each node denotes a cell type; node size scales with its total incoming or outgoing signal. E Circle plots of cell-cell communication network in normal-gestation samples. Edge width reflects interaction weights/strength between cell types. F Violin plots showing the expression of ligand/receptor genes mediating macrophage–endothelial communication (from G) in normal-gestation samples. Each subpanel includes its own x-axis cell-type labels to allow direct identification of the corresponding cell types. The y-axis indicates normalized log-transformed expression levels. Violin width reflects the density distribution of cells expressing each gene within a given cell type. Colors are used solely to visually distinguish cell types. G Bubble plot of ligand–receptor pairs from macrophages (source) to recipient stromal cell types in normal-gestation ovary. The color bubbles indicate the computed communication probability (mean). Dot size indicates statistical significance (P value); because a stringent threshold was applied (P < 0.01), all displayed interactions are highly significant, resulting in uniform dot sizes.

We next asked how this pro-angiogenic macrophage program is derailed during pregnancy failure. Pseudotime analysis across normal versus abnormal-gestation indicated a progression from normal-gestation macrophage states toward abnormal-gestation states (Fig. S2D–F). Clustering the top 1000 pseudotime-variant genes into four modules followed by Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment, we uncovered a functional dichotomy. Normal-gestation macrophages were highly enriched for vascular support pathways (e.g., HIF-1, TNF, VEGF signaling), whereas abnormal-gestation macrophages were enriched for metabolic and biosynthetic pathways (e.g., Carbon metabolism, Folate biosynthesis, Glycine/serine/threonine metabolism) (Fig. 2C). This suggests a pathological shift: the loss of pro-angiogenic capacity and a pivot toward aberrant metabolic remodeling in failed gestation.

To definitively map how macrophages direct this vascularization, we evaluated cell-cell communication networks. CellChat analyses performed separately for the three conditions showed that normal-gestation featured macrophages as the strongest senders and endothelial cells as the strongest receivers of intercellular signals; in unfertilized ovaries, fibroblasts and T cells dominated outgoing and incoming signaling, respectively; in abnormal-gestation, macrophages remained the strongest senders, but fibroblasts became the main receivers (Fig. 2D, E). To systematically and objectively assess this critical macrophage–endothelial crosstalk, we extracted the exhaustive set of significant ligand–receptor (L–R) pairs (P < 0.01) mediating macrophage-to-endothelial communication (Supplementary Data 10). In normal-gestation, this unbiased macrophage-centric L–R network highlighted: (i) FN1b-SDC4 interactions broadly with multiple lineages; (ii) a TNFSF-HDR axis unique with erythroid cells; and (iii) macrophage–endothelial circuits enriched for VEGF-VEGFR pairs (vegfab-kdr, vegfab-flt1, and vegfab-[flt1+kdr]), which are canonical drivers of endothelial proliferation, and TNFSF14-TNFRSFA, a highly specific immune–vascular axis critical for vascular homeostasis and remodeling (Fig. 2F, G). Notably, tnfsf14 expression was essentially restricted to macrophages, whereas its receptor tnfrsfa, was specific to endothelial cells (Fig. 2F).

Collectively, these results support a model in which successful teleost gestation is associated with a macrophage-centered pro-angiogenic program and macrophage–endothelial paracrine wiring, whereas abnormal-gestation is accompanied by attenuation or rewiring of these vascular-support modules.

A gestation-specialized steroidogenic perivascular population provides a maternal endocrine support module

Beyond immune-mediated vascularization, progestogen signaling is a conserved prerequisite for pregnancy maintenance across vertebrates^28,29. However, viviparous teleosts lack the specialized trophoblast-derived endocrine tissues seen in mammalian placentas. We therefore asked which maternal stromal lineages supply this essential endocrine support during black rockfish gestation. A global inspection of steroid-related transcripts across cell types revealed high expression of steroidogenic genes (e.g., hsd3b1, cyp17a1) in perivascular cells (PVC) (Fig. S3). Upon reclustering the PVC compartment, we resolved a gestation-specialized subcluster, annotated as CYP17A1+ PVCs (highlighted by black dashed circles in the corresponding UMAPs), characterized by selective upregulation of canonical steroidogenesis genes, including star, star2, cyp11a1, cyp17a1/a2, cyp21a2, and hsd3b1 (Fig. 3A, B, E).

Fig. 3. Characterization of a gestation-specialized steroidogenic perivascular cell population.

A UAMP of perivascular cells (PVCs) subsetted from unfertilized and normal-gestation datasets after subclustering. Each dot is a single cell; colors denote PVC subclusters. The legend (bottom right) lists subcluster names with cell counts. The black circle highlights the steroidogenic subcluster with specific upregulation of steroid biosynthesis genes. B UMAPs split by condition (unfertilized vs normal-gestation), with cells colored by PVC subclusters. The black circle marks the subcluster with specific enrichment of steroidogenesis genes. C qPCR of progesterone-synthesis genes (star, cyp11a1, hsd3b1) across ovarian-stroma states. X-axis: condition; Y-axis: expression relative to 18S rRNA (ΔΔCt). Normal: normal-gestation; Abnormal: abnormal-gestation. Data are presented as mean ± SEM (n = 3 biologically independent samples per group, each averaged from technical triplicates). Significance: P < 0.05 (*), P < 0.01(**), P < 0.001(***), P < 0.0001(****), ns not significant. D Spatial localization of HSD3B1 protein by IHC in the ovarian stroma across unfertilized (left) and normal-gestation (right) states. Top panels show lower-magnification overviews, and bottom panels display the corresponding higher-magnification views of the dashed boxed regions. Positive signal appears yellow-brown; black arrowheads indicate representative HSD3B1-positive cells (e.g., follicle cells and perivascular cells). Scale bars: 50 μm (top), 20 μm (bottom). E UMAPs projecting normalized expression of steroidogenesis-related genes onto the PVC landscape, color intensity reflects per-cell expression. cap capillary.

To orthogonally validate this transcriptomic prediction, and given that progesterone (P4) is a principal progestogen supporting pregnancy in mammals, we prioritized three key enzymes in the P4 pathway, STAR, CYP11A1, and HSD3B1, for orthogonal validation. qPCR on stromal fractions from unfertilized, normal-gestation, and abnormal-gestation ovaries showed high star and cyp11a1 expression in normal-gestation (vs. unfertilized and abnormal; P < 0.01) and high hsd3b1 expression in normal vs. abnormal-gestation (P < 0.01), with a non-significant tendency vs. unfertilized (Fig. 3C). To confirm these findings at the spatial and protein levels, we performed immunohistochemistry (IHC) for HSD3B1 across unfertilized and normal-gestation states. While HSD3B1 protein was primarily localized to the follicle cells surrounding oocytes prior to fertilization to support oocyte maturation, its spatial expression shifted dramatically during pregnancy. In normal-gestation ovaries, this analysis localized robust HSD3B1 protein expression specifically to the perivascular niche, definitively corroborating the emergence of a steroidogenic PVC population during gestation (Fig. 3D).

Collectively, these findings reveal that the black rockfish ovarian stroma has evolved a gestation-specialized perivascular steroidogenic module. This functionally answers our initial question by demonstrating that the maternal perivascular niche acts as a critical endocrine compartment to sustain pregnancy, compensating for the absence of trophoblast-based placental endocrine tissues in this teleost.

Cross-species analysis reveals functional convergence between the placenta-like ovarian stroma of black rockfish and mammalian early decidua

Prior microscopy in black rockfish found no specialized maternal–fetal exchange organ and no intimate tissue apposition comparable to the mammalian hemochorial placenta²⁴. We therefore asked whether, despite the lack of direct physical contact, the black rockfish placenta-like ovarian stroma functionally converges with the mammalian early decidua (the pre-hemochorial stage, e.g., mouse E5.5-E9.5). This mammalian stage operates prior to intimate contact and is characterized by progesterone-driven tissue remodeling, antioxidative defense, angiogenesis, and immune tolerance.

To test this evolutionary convergence hypothesis, we integrated single-cell datasets from black rockfish, human, and mouse with SAMap, and visualized the joint embedding (UMAP) by species and by cell-type annotation (Figs. 4A and S4A). A cross-species Sankey summary of cell-type similarities (Fig. 4B) revealed many-to-many cell-type correspondences across clades. Notably, all three species shared macrophages, neutrophils, dendritic cells, B cells, T cells, endothelial cells, and fibroblasts—a conserved set of immune, vascular, and stromal lineages that collectively organize the ovarian/decidual microenvironment.

Fig. 4. Cross-species analysis reveals functional convergence between the placenta-like ovarian stroma of black rockfish and mammalian early decidua.

A UMAP of the integrated manifold combining human, mouse, and black rockfish single-cell profiles. Each point is a cell; colors denote species. B Sankey plot summarizing cell-type transcriptomic mappings across species. To facilitate accurate cross-species tracking, all connecting ribbons are uniformly colored according to the central black rockfish cell lineages, illustrating their homology mappings to both the mouse (left) and human (right) datasets. Edges are shown if weight ≥0.05; for each source node, the top three outgoing mappings are retained. Macro macrophage, Mono monocyte, DSC decidual stromal cells, EC endothelial cell, FIB fibroblast, MFC myofibroblasts. C Comparative histological analysis (AB-PAS staining) of the embryonic epidermis between the yolk-dependent guppy (P. reticulata, left) and the matrotrophic black rockfish (S. schlegelii, right). Black arrowheads highlight robust PAS-positive (magenta) glycogen accumulation specifically along the S. schlegelii epidermis, providing direct evidence for maternal nutrient absorption, which is absent in the guppy. Alcian Blue marks cartilaginous structures (e.g., gills). Dashed boxes indicate regions magnified in the bottom row. Scale bars: 200 μm (top), 100 μm (middle), and 50 μm (bottom). D Immunofluorescence (IF) visualization of GYS1 spatial localization in the ovarian stroma. Representative images compare GYS1 distribution (magenta, Alexa Fluor 555) between unfertilized (top) and normal-gestation (bottom) states. In the unfertilized control, GYS1 strictly localizes to the follicle cells enclosing the oocyte. Upon gestation, GYS1 expression undergoes profound spatial and structural remodeling within the stromal compartment to support pregnancy. Nuclei are counterstained with DAPI (cyan). Scale bars: 60 μm. E Schematic model comparing the establishment of placenta/placenta-like structures in viviparous teleosts (e.g., P. reticulata, S. schlegelii) versus mammals (e.g., H. sapiens, M. musculus), highlighting functional convergence prior to the hemochorial stage.

We next investigated how specific stromal populations adapt to these analogous microenvironments, beginning with theca cells. Black rockfish theca cells—which proliferate after mating²⁴—align with angiogenesis-biased decidual stromal cells (DSC) in mouse early decidua and with the human endometrial fibroblast-like (FIB1) subcluster, which is enriched for protein synthesis/secretion and endometrial receptivity markers^30,31. To quantitatively substantiate this functional convergence across distinct mammalian features, we employed a dual validation strategy. First, unsupervised Gene Ontology (GO) and KEGG enrichment analyses of black rockfish theca cell markers revealed highly significant enrichment in pathways driving angiogenesis and hypoxia response (e.g., angiogenesis, HIF-1 signaling pathway), decidua-like hormone responsiveness and remodeling (e.g., female pregnancy, Relaxin signaling pathway, extracellular matrix disassembly), and massive protein processing/secretion (e.g., Protein processing in endoplasmic reticulum), natively mirroring the secretory and receptive state of mammalian early decidua (Fig. S5B, C). Second, to validate the alignment with murine pro-angiogenic DSCs, we performed module scoring using the AddModuleScore function in Seurat based on a conserved angiogenesis gene signature. This quantitative module scoring confirmed that black rockfish theca cells possess a significantly higher and specific pro-angiogenic transcriptomic program compared to other ovarian stromal lineages (P < 0.0001; Fig. S5A; Supplementary Data 7). Consistently, CellChat in the normal-gestation ovary showed that theca cells signal to endothelium via VEGF (vegfab-flt1/kdr) and JAM (jam3b-jam2a/3b) pathways, and engage other cells through CDH and AGRN axes (Supplementary Data 2).

Having identified the decidua-like features of theca cells, we next evaluated the functional trajectory of granulosa cells. Granulosa cells, which produce E2 pre-fertilization and undergo attrition during gestation, mainly mapped to mouse pre-decidual stromal cells (pre-DSCs), a state annotated with “epithelial cell proliferation” and mesenchymal-to-epithelial transition (MET) during decidualization; they also matched human fibroblast 3 (FIB3), enriched for ECM organization and cell-cycle/proliferation signatures^30,31. Given that our single-cell capture of granulosa cells predominantly reflects their healthy, unfertilized state prior to gestational attrition, we performed GO and KEGG enrichment analyses to validate their baseline functional wiring. Exactly aligning with their pre-fertilization role, these cells were robustly enriched for ovarian steroidogenesis and estrogen signaling pathway, functionally underpinning their E2 production. More importantly, consistent with their cross-species mapping to matrix-remodeling and MET-undergoing mammalian pre-DSCs/FIB3, black rockfish granulosa cells exhibited highly active structural reorganization networks, marked by significant enrichment in ECM-receptor interaction, focal adhesion, bicellular tight junction, and Wnt/TGF-beta signaling (Fig. S5D, E).

Because mouse decidualization relies on the dynamic state conversion of endometrial stromal fibroblasts (eSFs) to pre-DSCs and subsequently to DSCs, we asked whether black rockfish fibroblasts undergo a parallel state transition. Black rockfish fibroblasts aligned with mouse eSFs and pre-DSCs. Motivated by this, we subclustered the ovarian fibroblasts into five states (SF1-SF5) (Fig. S4B). SF4 predominated in the unfertilized condition, whereas pregnancy fibroblasts mainly occupied the non-SF4 states (Fig. S4B). CytoTRACE and pseudotime analyses placed SF4 at the root of the trajectory, with pregnancy-biased states deriving from SF4 (Fig. S4C). Consistent with a MET-like progression reminiscent of mammalian decidualization, mesenchymal/stromal marker genes (e.g., dcn, pdgfrb) were enriched in SF4, while epithelial marker genes (e.g., cdh1, cldn3, cldn4, epcam) were upregulated in the non-SF4 clusters (Fig. S4D, E). We also identified a pregnancy-specialized, immune-interacting fibroblast subtype (SF2) that co-expressed immune genes alongside canonical fibroblast markers and showed KEGG enrichment for Chemokine signaling; Fc gamma R–mediated phagocytosis; B cell receptor signaling; and NF-kappa B signaling (Fig. S4H). Potential doublets were excluded using DoubletFinder (Fig. S4F).

Finally, a hallmark of early decidualization is the maternal secretion of nutrients, such as glycogen, to support the conceptus. Given that the black rockfish is highly matrotrophic and the exact non-yolk maternal nutrient source remains unclear, we asked which cell types might mediate this provision. Our cross-species analysis revealed that the black rockfish epithelial cells uniquely aligned with the mouse nourishing DSC population (the primary glycogen source during early decidualization). To investigate this further, we inspected the normal-gestation scRNA-seq dataset and observed glycogen-biosynthetic genes to be expressed across several cell types, with the highest levels specifically concentrated in the epithelium (Fig. S4G). Consistently, immunofluorescence (IF) for the key glycogen-synthesis enzyme GYS1 was positive in the pregnancy ovarian stroma (Fig. 4D). Furthermore, AB-PAS staining revealed limited stromal glycogen storage but prominent glycogen signal in embryonic epidermis, suggesting a division of labor in which maternal stroma synthesizes glycogen while embryos uptake it at the surface (Fig. 4C). This is in line with prior research that ovarian fluid supports embryonic nutrition in black rockfish¹⁹. As a yolk-reliant (lecithotrophic) comparison, the poecilid guppy (Poecilia reticulata) processed in parallel lacked epidermal glycogen uptake under identical histochemical conditions (Fig. 4C).

In summary, while human and mouse fertilization occurs in the oviduct and early development relies on endometrial decidualization prior to placentation, viviparous teleosts such as black rockfish and guppy develop in situ within the ovary. Our cross-species analysis indicates that the pregnancy-induced ovarian stroma of black rockfish undergoes cell-state reallocations and paracrine rewiring that functionally converge on mammalian early decidual programs—including angiogenesis, immune-stromal crosstalk, MET-like stromal transitions, and maternal glycogen provision/embryonic uptake—even in the absence of intimate maternal–fetal contact (Fig. 4E). Together, these data validate our evolutionary hypothesis, supporting an evolutionary convergence toward decidua-like support modules in an ovary-based pregnancy system.

Ovarian sperm storage: ligand–receptor signaling and ANXAs upregulation

The black rockfish stores sperm within the ovary for up to ~6 months prior to fertilization, despite lacking a specialized sperm storage organ (such as a seminal receptacle). We therefore asked how the ovarian stroma interacts with spermatids to maintain their prolonged viability. To map the paracrine networks supporting this unique niche, we utilized CellChat to infer intercellular crosstalk between spermatids and the surrounding ovarian stromal lineages. The analysis revealed bidirectional signals between spermatids and epithelial, fibroblast, and granulosa cells, and a unidirectional signal from spermatids to mast cells. Notably, we identified two sperm to stroma pathways: SEMA4C (sema4c-plxnb2b) and AGRN (agrn-dag1), and three stroma to sperm pathways: ncWNT (wnt5b-fzd6), EPHB (efnb1-ephb2b), and EPHA (efna5a-ephb2b/epha3) (Fig. 5A, B). Beyond paracrine signaling, physical docking is crucial for prolonged sperm survival. In mammals, sperm transiently bind oviductal epithelium to form a reservoir; work in bovine and other species implicates apical annexins (ANXAs) as a sperm-binding receptor platform on the oviductal epithelium, supporting viability and phased release³². Given this conserved mechanism, we hypothesized that a homologous annexin-based binding platform is deployed within the black rockfish ovarian microenvironment to enable its organ-independent sperm storage.

Fig. 5. Ovarian sperm storage: ligand–receptor signaling and ANXAs upregulation.

A Circle plot of cell-cell communication between spermatids and ovarian stromal cell types. Edge width reflects interaction weights/strength between cell types. B Bubble plot of ligand–receptor pairs linking spermatids with stromal cell types. The color gradient indicates the computed communication probability (mean). Dot size indicates statistical significance (P value); because a stringent threshold was applied (P < 0.01), all displayed interactions are highly significant, resulting in uniform dot sizes. C Heatmap of SsANXAs gene expression across the ovarian stroma. X-axis: reproductive stages (PRM, pre-mating; POM, post-mating). Y-axis: individual SsANXAs. Heatmap colors represent row-scaled normalized expression levels (Z-scores). A value of “0” indicates the mean expression level of a gene across all samples, with red and blue indicating standard deviations above and below the mean, respectively. D Predicted tertiary structures of the five SsANXAs upregulated at POM. E Phylogenetic analysis of ANXAs across vertebrates; SsANXAs are highlighted with red shading. F Violin plots showing SsANXA expression distributions across cell types in Unfertilized scRNA-seq samples. The y-axis represents normalized log-expression levels. The width of each violin reflects the probability density of cells expressing the gene, and the categorical color coding serves solely to visually distinguish distinct cell types. G Schematic model of the ovarian sperm-storage microenvironment in black rockfish. Small colored dots schematically represent ligand–receptor-mediated intercellular signaling interactions corresponding to the indicated communication pairs.

To test this hypothesis, we first conducted a genome-wide survey and identified 15 anxa genes in the black rockfish (anxa1a/b, anxa2a/b, anxa3a/b, anxa4, anxa5a/b, anxa6, anxa11a/b, anxa13.1/2, anxa14) (Fig. 5E). To determine if these genes respond to the presence of sperm, we analyzed bulk transcriptomes of the ovarian stroma pre- versus post-mating. Strikingly, we observed a distinct post-mating upregulation of specific members (anxa1a, anxa2a, anxa2b, anxa5a, anxa11a) (Fig. 5C), nominating them as primary candidates for the storage phase. To structurally validate these candidates, we also predicted 3D structures for representative ANXAs, confirming their canonical functional domains (Fig. 5D). Finally, we asked whether the spatial expression of these annexins aligns with potential sperm-tissue encounter sites. In our scRNA-seq dataset, 14/15 anxa genes were detectable across the ovarian microenvironment. Their expression was predominantly concentrated in granulosa cells, epithelial cells, fibroblasts, dendritic cells, and macrophage compartments, with anxa1a, anxa2a, anxa4, anxa5b, and anxa13.1 being the highest-expressed members (Fig. 5F). This expression pattern places annexins precisely at stromal-immune-epithelial interfaces where prolonged sperm docking and maintenance are anatomically required. To orthogonally validate this bioinformatic prediction, dual-color fluorescence in situ hybridization (FISH) confirmed the distinct co-localization of anxa1a transcripts within cldn3+ epithelial cells lining the ovarian compartments (Fig. S7). This provides direct histological evidence for a specialized, annexin-expressing epithelial microenvironment.

Together, these structural and transcriptomic, histological data suggest a putative ovarian sperm-storage niche in the black rockfish. Specifically, the SEMA4C, AGRN, ncWNT, EPHA, and EPHB pathways, alongside post-mating ANXA upregulation and broad stromal expression, and precise epithelial localization, outline a coherent ovarian sperm-storage niche in black rockfish (Fig. 5G).

Abnormal-gestation is associated with lymphatic endothelial skewing and lipid recapture

Having established the critical role of robust angiogenesis in maintaining pregnancy, we next sought to characterize the vascular regression observed during abnormal-gestation and understand how the surviving endothelial networks functionally adapt to gestational failure. Consistent with the sparse vasculature observed macroscopically via stereomicroscopy, our scRNA-seq data revealed that the overall proportion of endothelial cells (ECs) in abnormal-gestation ovaries was drastically reduced to ~50% of that observed in normal-gestation, suggesting functional reprogramming of the endothelial compartment. To decode the functional reprogramming of this diminished endothelial compartment, we subclustered ECs from normal and abnormal samples into 11 subtypes (Fig. 6A): lymphatic EC (LEC), arterial EC (AEC), capillary EC (CAP), venous EC (VEC), F3⁺ EC, ALAS2⁺ EC, and an uncharacterized group.

Fig. 6. Abnormal-gestation is associated with lymphatic endothelial skewing and lipid recapture.

A UAMP of endothelial cells (ECs) subsetted from normal and abnormal-gestation datasets after subclustering. Each dot is a cell; colors denote EC subclusters. The legend (bottom right) lists subcluster names with cell counts. The black circle highlights the subcluster with lipid-handling genes. Condition-split UMAPs (normal vs abnormal-gestation), colored by EC subcluster(right). The black circle marks the lipid-handling subcluster in the abnormal-gestation condition. AEC: artery endothelial cell; CAP: capillary endothelial cell; VEC: vein endothelial cell; LEC: Lymphatic endothelial cell. B UMAPs projecting normalized expression of lymphatic endothelial markers (e.g., lyve1, flt1) and vascular endothelial markers (e.g., cd44, rgcc, sema3g, cdh5); color intensity reflects per-cell expression. C Pesudotime-ordered heatmap of DEGs for LEC and AEC/CAP, arranged by shared kinetics inferred with Monocle. Gene modules were identified by hierarchical clustering; representative DEGs and enriched KEGG pathways per module are annotated at right. D UMAP of PVCs subsetted from normal and abnormal-gestation datasets after subclustering; colors denote PVC subclusters, with the legend listing names and cell counts. Condition-split UMAPs (normal vs abnormal-gestation) for PVCs, colored by subcluster identity (bottom). E UMAPs showing normalized expression of FABP4⁺ and RGS5⁺ perivascular cell programs; color intensity reflects per-cell expression.

This subclustering revealed a massive lineage shift. Notably, LEC (marked by e.g., flt4, lyve1) dominated the abnormal sample, whereas normal-gestation was enriched for AEC/CAP and the unknown EC cluster (marked by e.g., cd44, rgcc, sema3g, cdh5) (Fig. 6A, B). To trace the functional bifurcation between these healthy and failing vascular states, we performed Branch Expression Analysis Modeling (BEAM) pseudotime analysis contrasting the LEC branch against the blood-vascular branch (using the top 1000 branch-dependent genes) (Fig. 6C). Crucially, to validate the directionality of this pathological transition, we deployed RNA velocity analysis. The inferred velocity streamlines were broadly consistent with the pseudotime trajectory and suggested a transcriptomic transition from the canonical blood-vascular pool toward the lymphatic-skewed state during abnormal-gestation (Fig. S6). The resulting heatmap demonstrated that LEC branches preferentially expressed lipid-handling genes (e.g., fabp4, fabp1) with KEGG enrichment in Autophagy and Fat digestion and absorption pathways. Conversely, the blood-vascular branch upregulated classical angiogenesis/vascular genes (e.g., eltd1, kdr, flt1), enriched for HIF-1 signaling and immune-related pathways (leukocyte transendothelial migration; antigen processing and presentation) (Fig. 6C). To determine whether the surrounding perivascular niche co-adapts to this lipid-scavenging endothelial state, we examined the perivascular compartment. Consistently, subclustering identified a distinct FABP4+ perivascular cell (PVC) population that was highly biased toward abnormal-gestation (Fig. 6D, E).

Together, these data indicate a pathological adaptation. By shifting the endothelial landscape toward a lymphatic program that facilitates lipid capture, the maternal stroma functionally compensates for a diminished blood-vascular supply. This lymphatic-skewed program likely provides a necessary route for the maternal resorption of unfertilized oocytes and nonviable embryos, finalizing our model of how the ovarian stroma responds to both successful and failed pregnancies.

Discussion

Our single-cell atlas of the black rockfish ovarian stroma across unfertilized, normal-gestation, and abnormal-gestation states delineated a molecular, state-dependent remodeling of a unique placenta-like interface. Rather than a static structure, this ovary-based maternal microenvironment operates as a highly dynamic, state-dependent platform. Prior to fertilization, the stroma is specialized for prolonged sperm preservation. Following gestation onset, a massive functional reprogramming is triggered: the follicular architecture is dismantled, macrophages adopt pro-angiogenic phenotypes, and a gestation-specialized perivascular lineage emerges to provide crucial steroidogenic support, collectively orchestrating a microenvironment tailored for embryo survival. Conversely, under abnormal-gestation, this supportive niche actively transitions into a scavenging state characterized by lymphatic skewing and lipid recapture. Crucially, cross-species integration reveals that despite evolving independently and lacking a dedicated uterus, the black rockfish ovarian stroma repeatedly co-opts core mammalian placentation modules. Together, these findings offer a comparative window into how vertebrates independently assemble placenta-like phenotypes on the evolutionary path to viviparity.

A central implication of this study is that teleost viviparity in black rockfish depends on functional repurposing of canonical ovarian structures. In vertebrate ovaries, granulosa cells (GCs) and theca cells (TCs) classically support folliculogenesis and steroidogenesis prior to ovulation: GCs promote oocyte growth and produce estradiol and inhibin, whereas TCs provide androgen precursors and paracrine support^33–35. In mammalian models, theca cells are strictly restricted to oogenesis, as pregnancy subsequently proceeds within a distinct uterus. In black rockfish, however, embryogenesis occurs in situ within the ovary, and the follicular compartment must therefore be remodeled rather than discarded after fertilization. Our single-cell and ultrastructural data support this transition. Granulosa cells undergo widespread apoptosis after gestation onset, whereas theca cells expand. Although these surviving cells retain canonical theca-lineage markers (e.g., anpep and igfbp5), their physiological role appears to shift substantially during pregnancy. Consistent with previous observations, this expanded theca population upregulates genes associated with mammalian placentation-like programs, including plgf, plac8, and tpbg, during gestation²⁴. These findings suggest that, in intraovarian-gestating teleosts lacking a uterus, theca cells are not simply regressing remnants of ovulation, but are repurposed into a gestation-supportive stromal population.

Beyond structural remodeling, successful gestation strictly depends on sustained endocrine support. We therefore investigated the source of progestogen synthesis in this atypical pregnancy system and identified a perivascular cell population expressing key steroidogenic genes (e.g., star, star2, cyp11a1, cyp17a1/a2, cyp21a2, and hsd3b1). qPCR and IHC both support the interpretation that these cells form a steroidogenic niche within the gestational ovarian stroma. In mammals, luteinization, triggered by the ovulatory LH surge, drives granulosa and theca interna differentiation into the corpus luteum, whose progesterone is essential for establishing and maintaining early decidualization³⁶. Unlike mammals, which later shift progesterone production to the placental syncytiotrophoblast (fetal in origin), teleosts lack both a trophoblast lineage and trophoblast-based placental architecture³⁷. Our data suggest that black rockfish compensates through a maternal-side solution: a perivascular steroidogenic program within the ovarian stroma that locally supplies progestogens during gestation. This represents a clear example of functional convergence at the physiological level. Although black rockfish lacks the specialized gestational structures found in mammals, it appears to achieve a comparable outcome—local progestogen-supported pregnancy maintenance—by repurposing an alternative maternal stromal lineage. In this framework, the steroidogenic PVC program may provide the endocrine support needed to shift the ovarian stroma from an oocyte-focused environment to a gestation-supportive, placenta-like interface.

Downstream of this endocrine transition, the maternal stroma appears to assemble a decidua-like support system for angiogenesis and embryonic nourishment. In mammalian early pregnancy, progesterone-dependent decidualization is accompanied by extensive vascular remodeling and maternal immune specialization, with uterine natural killer (uNK) cells playing a major role in orchestrating pre-implantation vascular adaptation^38,39. Indeed, uNK cells comprise a large proportion of leukocytes in the secretory endometrium and early decidua, where they regulate vascular change and tissue remodeling^40–42. In contrast, our black rockfish dataset identifies macrophages as the dominant gestational immune compartment, enriched for angiogenesis-associated programs such as vegfa, kdr, tnf, flt1, pdgfb, and sod2. CellChat analysis further identifies a macrophage-to-endothelium TNFSF14 axis that is specific to this interaction pair; notably, in mammals, TNFSF14 (LIGHT) also participates in decidual crosstalk through TNFRSF14/HVEM signaling^43,44. Although we did not detect a clear uNK-like population, whether because of genuine absence or unresolved marker divergence, the macrophage-centered circuitry suggests that distinct maternal immune effectors may execute functionally analogous remodeling programs in different vertebrate lineages. In parallel with vascular remodeling, the maternal stroma must also support embryonic nutrition. Because teleosts lack a trophoblast lineage, nutrient transfer beyond the initial yolk must be mediated entirely by maternal tissues. In a congener, Sebastes taczanowskii, embryos absorb yolk-like material from ovarian fluid via the hindgut⁴⁵. Mammals similarly exhibit an early phase of endometrial provisioning before establishment of a fully invasive placenta, including nutrient-supportive decidual cell states in mice³¹. Taken together, these observations support a model in which the black rockfish ovarian stroma achieves key placenta-like functions—vascular remodeling and early nutrient support—through a maternal interface that differs architecturally from mammalian decidua yet converges with it at the functional level.

Before this gestational remodeling begins, however, the black rockfish ovary must first perform a distinct but equally important role: prolonged sperm storage. Prolonged female sperm storage is phylogenetically widespread across vertebrates and shapes cryptic female choice and sperm competition. In some taxa with long storage windows (e.g., birds, up to ~4 months), specialized sperm storage tubules (SSTs) at the utero-vaginal junction (UVJ) create antioxidant-rich, immunosuppressive niches⁴⁶. Most mammals lack discrete SSTs; instead, sperm are retained at species-specific epithelial sites, including the vagina, uterus, or utero-tubal junction, where quiescence is maintained and timed release is regulated^47,48. Within these mammalian reservoirs, sperm bind oviductal epithelium, undergo a reduction in intracellular Ca²⁺ and motility, delay capacitation, and are ultimately released peri-ovulation in response to progesterone-linked cues^49–51. Analogously, in the black rockfish, the proliferating ovarian epithelium forms peri-follicular crypts after mating that retain sperm until fertilization^21,22. Mechanistically, annexins emerge as compelling candidates for mediating this critical epithelial-sperm interface. For instance, ANXA1, ANXA2, ANXA4, and ANXA5 localize to apical oviductal epithelium and reduce sperm binding to epithelial explants in cattle⁵². Similarly, our data reveal that anxa1a, anxa2a/b, anxa5a, and anxa11a are upregulated during the storage phase in the black rockfish. This parallel suggests a deeply conserved, annexin-centric molecular module that stabilizes sperm and prevents premature capacitation across divergent vertebrate lineages. Consequently, these molecules constitute tractable targets for functional testing and could inform the development of annexin-guided, serum-free formulations as a practical route to in vitro sperm storage in aquaculture.

While the normal gestational stroma successfully mimics a decidual interface, our analysis of abnormal-gestation reveals the physiological safety nets deployed when this system fails. We observed a drastic reduction in blood-vascular endothelial cells and an active transcriptomic trajectory toward a lymphatic-skewed endothelial state enriched for lipid-handling pathways, accompanied by the emergence of FABP4+ PVCs. For a highly fecund, viviparous marine teleost like the black rockfish, gestation represents an extraordinary maternal energetic investment. If embryonic development fails broadly, passive loss of nonviable embryos and associated nutrient reserves would likely impose a major fitness cost. Macroevolutionary studies in viviparous teleosts, including Poeciliidae, have emphasized that the evolution of complex placenta-like matrotrophy is constrained by the energetic burden of reproduction, making mechanisms that mitigate such costs especially important⁵³. Within this context, the lymphatic and lipid-scavenging shift observed here may represent an active maternal recovery program. By promoting lymphatic drainage and lipid recapture, the stroma may facilitate resorption of nonviable embryos, unreleased oocytes, or unutilized yolk-derived resources. Thus, in this high-investment reproductive strategy, lipid scavenging is better interpreted not as a passive byproduct of tissue degeneration, but as a potentially adaptive mechanism for reclaiming maternal physiological resources and preserving future reproductive capacity.

In conclusion, the black rockfish ovarian stroma appears to function in a state analogous to early gestational support in mammals prior to the establishment of intimate hemochorial placentation. By maintaining this support state for an extended period, it provides a natural model of prolonged decidua-like function within a non-uterine reproductive organ. Remarkably, this single maternal compartment integrates pre-fertilization sperm storage, local progestogen synthesis by steroidogenic perivascular cells, coordinated immune–vascular remodeling, and, under pathological conditions, nutrient recapture through lymphatic-skewed stromal reprogramming. Because black rockfish lacks a fetal trophoblast lineage, its ovarian stroma illustrates how core gestational functions can be assembled through maternal-side evolutionary innovation alone. More broadly, this system offers a comparative framework for understanding the flexible routes by which vertebrates evolve placenta-like support mechanisms, while also providing a foundation for future aquaculture applications, including improved sperm preservation and breeding control.

Methods

Ethics statement and experimental animals

All animal procedures complied with national and institutional guidelines and were approved by the Animal Research and Ethics Committees of Ocean University of China prior to study initiation.

Adult black rockfish (1200 ± 300 g) were procured from marine net-cages in the northern Yellow Sea (Shandong Province, China). The experimental individuals described below were anesthetized with ethyl 3-aminobenzoate methanesulfonic acid (MS-222, 200 mg/L) before sacrifice. Ovarian tissues were collected in May and December 2024 to obtain stromal samples representing unfertilized, normal-gestation (the embryo was at the pigment stage), and abnormal-gestation states. Specifically, during our sampling of pregnant females (n = 36), abnormal-gestation was identified based on a naturally occurring incidence (approximately 44%; Fig. S1G) of spontaneous embryonic mortality. Gestational states were unambiguously scored via macroscopic inspection using two consistent criteria: (1) embryonic viability (live, developing embryos versus opaque, resorbing embryos), and (2) maternal vascularization (densely vascularized, bright red stroma versus avascular, pale stroma; Fig. S1F).

Tissue collection and cell preparation

Single-cell suspensions were prepared following established protocols with minor modifications²⁶. Briefly, after MS-222 anesthesia, blood was removed via the caudal (tail) vein using a 10 mL syringe. The coelom was opened, the ovaries were dissected free from the viscera, and immediately transferred to ice-cold phosphate-buffered saline (PBS). Ovaries were gently perfused with PBS to minimize residual blood cells. The ovarian wall was opened to expose the ovarian stroma; depending on reproductive state, oocytes or embryos were carefully removed by microdissection so that only stromal tissue was retained.

Approximately 3 mL (wet volume) of ovarian stromal tissue was minced with sterile scissors and digested in 10 mL Trypsin-EDTA Solution (0.25% trypsin, 0.02% EDTA; Biosharp, Beijing, China) with gentle trituration using a Pasteur pipette for 4 min at room temperature. Digestion was quenched with Leibovitz’s L-15 (Biosharp) supplemented with 10% FBS (G-Clone, Beijing, China). The cell suspension was passed through a 40 μm strainer and centrifuged at 400 × g, 4 °C for 6 min. Pellets were washed once in ice-cold PBS under the same conditions and finally resuspended in PBS.

Cell counts and viability were determined with a hemocytometer using 0.4% trypan blue, mixed 1:1 with the cell suspension. Viability was calculated as the percentage of unstained (viable) cells among the total counted cells.

Single-cell RNA-seq library preparation and sequencing

Single-cell gene-expression libraries were prepared with the Chromium Next GEM Single Cell 3′ v3.1 workflow (10x Genomics) according to the manufacturer’s instructions. Gel beads carrying unique molecular identifiers (UMIs) and cell barcodes were loaded near saturation so that most GEMs contained one cell and one bead. After exposure to lysis buffer, polyadenylated RNAs were hybridized to bead-bound oligo-dT primers. Emulsions were broken, and beads were recovered into a single tube for reverse transcription. During cDNA synthesis, each molecule was tagged at the cDNA 5′ end (corresponding to the mRNA 3′ end) with its UMI and cell barcode to record the cell of origin. Bead-bound cDNA then underwent second-strand synthesis, adapter ligation, and universal amplification. Libraries were constructed from randomly fragmented whole-transcriptome amplification products, thereby enriching the 3′ ends linked to the UMI/cell barcode. All remaining steps followed the standard Chromium Single Cell 3′ v3.1 (10X Genomics, no. PN-1000121) protocol. Libraries were quantified using a Bioanalyzer 2100 with a High Sensitivity DNA chip (Agilent) and a Qubit High Sensitivity DNA assay (Thermo Fisher Scientific), and sequenced on an Illumina NovaSeq with 2 × 150 bp paired-end reads.

Single-cell RNA-seq data processing

Raw base-call files were processed with Cell Ranger (10x Genomics, v9.0.1; default parameters) and aligned to the black rockfish reference genome (GCA_014673565.1). The resulting UMI count matrix was analyzed with the R packages Seurat (v5.1.0)⁵⁴. We filtered features and cells using min.cell = 3 and min.features = 300, removing genes detected in <3 cells and cells with <300 detected genes. Per-condition quality control thresholds were then applied as follows: Normal-gestation libraries, retain cells with 200–4000 detected genes and 200–14,000 total UMI counts; Unfertilized libraries, retain cells with 200–3700 detected genes and 200–17,000 total UMI counts; Abnormal-gestation libraries, retain cells with 200–4500 detected genes and 200–20,000 total UMI counts. Thresholds were empirically determined from the density distributions of gene counts and total UMIs for each specific library to strictly trim extreme long-tailed outliers (representing potential multiplets) and dense low-count peaks (representing empty droplets), and barcodes outside these ranges were discarded (a representative example of this distribution-based QC filtering is shown in Fig. S1H, I).

Data were normalized with NormalizeData, highly variable genes (HVGs; 2000 features) were selected using FindVariableFeatures and data were scaled with ScaleData. Principal component analysis (PCA) was performed with RunPCA. Elbow plots indicated diminishing variance contributions beyond PC ≈ 20; therefore, PCs 1-20 (dims = 1:20) were used for neighbor-graph construction (FindNeighbors), clustering (FindClusters), and UMAP embedding (RunUMAP). For cross-sample comparisons, datasets were integrated with Harmony via the IntegrateLayers function (method = HarmonyIntegration). To ensure quantitative rigor, integration was strictly anchored on the pre-computed principal component space (orig.reduction = “pca”) derived from the top 2000 HVGs, effectively mitigating batch effects prior to graph construction and clustering. Clustering resolution was quantitatively optimized by evaluating a sequence of resolutions (ranging from 0.1 to 1.0) using the clustree R package. By inspecting the stability of cluster partitioning across this gradient, we objectively identified the optimal dataset-specific resolution for each analysis step. Cell identities were primarily assigned using canonical marker panels visualized by Dotplot across candidate clusters; the FindAllMarkers results were used as supportive evidence to refine boundaries and resolve ambiguous groups. Crucially, to identify robust cluster-specific markers, we applied strict quantitative cutoffs using the Wilcoxon rank-sum test. Genes were considered significant markers only if they met the following pre-defined criteria: (1) a minimum log2(fold change) > 0.25; (2) expression in at least 25% of cells within the target cluster (min.pct = 0.25); (3) strictly positive enrichment (only.pos = TRUE); and (4) a Bonferroni-adjusted P value < 0.05. The marker panels used for cell-type annotation are displayed in Fig. 1F. The corresponding FindAllMarkers results for each cluster are reported in Supplementary Data 3.

To evaluate statistical differences in cell-type proportions between different gestational states, we utilized the scProportionTest R package⁵⁵. This permutation-based framework calculates the Log2 Fold Difference (log2FD) of cell proportions between two compared conditions. Statistical significance was assessed using 10,000 random permutations. A compositional shift was considered statistically significant if the FDR was <0.05 and the absolute log2FD was >0.58 (equivalent to a fold change >1.5).

Complementary scRNA-seq datasets from prior studies were integrated to increase sample diversity and facilitate cross-species inference. Specifically, we used maternal-derived decidual tissue (maternal part of placenta) from early human pregnancy (GSE183837), maternal-derived decidual tissue from early mouse pregnancy (GSE226417), and male black rockfish testis (GSE252438). Processed count matrices were reanalyzed with the same pipeline described above, and cell types were assigned by referencing the source publication’s cluster annotations and curated marker genes.

Finally, to facilitate interactive data exploration and public accessibility of our processed single-cell transcriptomic atlases, we utilized the ShinyCell R package⁵⁶ to construct user-friendly, web-based visualization portals.

Reconstruction of single-cell trajectories and RNA velocity

The Monocle2 package (v 2.26.0)⁵⁷ and CytoTRACE (v 0.3.3)⁵⁸ were used for trajectory construction and pseudotime inference. For Monocle2, the intersection of HVGs identified by Seurat and genes with high dispersion calculated by Monocle2 was used. Genes were retained if detected at ≥0.1 expression in ≥10 cells. Dimensionality reduction used DDRTree, followed by cell ordering with Monocle2’s default parameters. The ordered cells were visualized in a trajectory using the plot_cell_trajectory function, which plots the minimum spanning tree. CytoTRACE was used to estimate developmental potential and re-root the trajectory on the state with the highest CytoTRACE score. CytoTRACE was run with default settings after filtering to genes detected in ≥5 cells. Pseudotime-associated genes were identified with Monocle2 using differentialGeneTest. For macrophage subclusters, the top 1000 genes (ranked by q-value) were partitioned into four expression modules for pseudotime heatmap visualization, and each module underwent GO and KEGG pathway enrichment analyses with KOBAS (v 3.0)⁵⁹. For endothelial cells, branch-dependent regulators distinguishing lymphatic endothelial cells from blood endothelial cells were identified using BEAM in Monocle2. The top 200 branch-dependent genes were grouped into four clusters for heatmaps, followed by KOBAS enrichment per cluster; representative pathway genes are highlighted in Fig. 6C.

To investigate cellular dynamics and state transitions, RNA velocity analysis was performed using the scVelo package in Python. Matrices of spliced and unspliced transcript counts were loaded from loom files. Preprocessing was conducted by filtering and normalizing the data to retain the top 2000 HVGs. First- and second-order moments were computed across 30 nearest neighbors using 30 principal components. RNA velocity was then estimated using the stochastic model. Finally, a velocity graph was constructed, and the inferred velocity vectors were projected as streamlines directly onto the UMAP embeddings previously generated by Seurat to visualize the directional flow of endothelial differentiation.

Cell-cell communication analysis

Intercellular signaling was inferred with the CellChat package (v 2.1.2)⁶⁰. Given the absence of a black rockfish-specific L–R compendium, we used CellChatDB. zebrafish as the reference. To enable this cross-species analysis, black rockfish sequences were mapped to zebrafish orthologs using DIAMOND with an e-value threshold of 1e-5. Approximately 82.3% of the expressed genes in our dataset were successfully mapped to the zebrafish genome, ensuring a robust foundation for the subsequent L–R predictions (Supplementary Data 8).

For each sample, a CellChat object was created, the zebrafish database was assigned, and default preprocessing was applied (gene filtering and identification of over-expressed genes/interactions), followed by computation and aggregation of communication probabilities. Two-sample comparisons were performed by merging the per-sample CellChat objects with mergeCellChat.

For the targeted analysis of sperm-ovarian stroma crosstalk, we subset late spermatids from the differentiating-stage testis and merged them with the unfertilized ovarian stroma dataset from female black rockfish to predict L–R pairs relevant to sperm storage. Unless stated otherwise, default parameters were used.

Cross-species comparative analysis

We integrated scRNA-seq datasets from human, mouse, and black rockfish using SAMap v1.0.15 (Python 3)⁶¹. Human and mouse samples comprised maternal decidua in early pregnancy prior to the formation of a functional hemochorial placenta (maternal compartment). The human dataset was an endometrial biopsy 5 days after ovulation (ultrasound-dated; equivalent to LH+7, i.e., the window of implantation), totaling 6789 cells. The mouse dataset covered E5.5-E9.5 with 33,288 cells. The black rockfish dataset combined unfertilized and normal pregnancy ovarian stroma, totaling 35,377 cells. Fetal-derived trophoblast cells were excluded prior to cross-species alignment.

Seurat objects were converted to .h5ad using SeuratDisk in R (SaveH5Seurat). Orthology maps were generated with map_genes.sh (SAMap utilities) to compute pairwise protein correspondences among the three species (human: GCF_000001405.40; mouse: GCF_000001635.27; black rockfish: study reference proteome). SAMap was initialized from species-specific SAM objects and run with pairwise=True. The resulting integrated manifold (sm.samap) was used for co-embedding and cross-species cell-type correspondence; annotations from each species were projected onto the shared UMAP for visualization. To summarize cross-species correspondences, we drew a Sankey diagram using SankeyNetwork in networkD3 (R). Before plotting, we kept edges with weight ≥ 0.05 and retained the top 3 outgoing edges per source node.

To quantify the expression levels of conserved functional programs at the single-cell resolution, we utilized the AddModuleScore function implemented in Seurat. Specifically, a conserved angiogenesis gene signature (detailed in Supplementary Data 7) was used to calculate an Angiogenesis Module Score for each stromal cell. Statistical differences in module scores across distinct cell lineages (e.g., theca cells versus fibroblasts, epithelial cells, and granulosa cells) were evaluated using the nonparametric Wilcoxon rank-sum test, with P < 0.0001 considered statistically significant.

Bulk RNA-seq data processing

Ovarian stroma from black rockfish at pre-mating (PRM; n = 3) and post-mating (POM; n = 3) stages was downloaded from the CNGB Nucleotide Sequence Archive (CNSA; accession ID CNP0000222). Raw reads were adapter-/quality-trimmed with fastp⁶² and aligned to the black rockfish reference genome using HISAT2⁶³. Alignments were sorted and indexed with SAMtools⁶⁴, and gene-level counts were obtained using featureCounts⁶⁵. Differential expression between POM vs PRM (PRM as the reference) was performed on raw counts with DESeq2⁶⁶. Genes with Benjamini-Hochberg FDR < 0.05 and |log2 fold-change| ≥ 1 were considered differentially expressed genes (DEGs). Transcripts per million (TPM) were then computed for visualization only, and heatmaps of selected genes were plotted in R using the base heatmap function.

Gene identification and bioinformatics analysis

Because the ANXA family is well curated in Danio rerio, we first collected zebrafish ANXA protein sequences (Supplementary Data 4) and used them as queries to search the black rockfish predicted proteome with BLASTP in TBtools⁶⁷ with default parameters. In parallel, we performed a domain-based screen using the Pfam annexin domain (PF00191): candidate proteins were required to contain PF00191 as determined by InterProScan. The BLAST- and domain-based results were merged and deduplicated. In total, 15 black rockfish ANXA genes (SsANXAs) were retained (Supplementary Data 5). To place SsANXAs into subfamilies, ANXA protein sequences were retrieved from NCBI for representative model species: Homo sapiens, Mus musculus, Gallus gallus, Xenopus laevis, Danio rerio, Oryzias latipes, Gasterosteus aculeatus, and two viviparous teleosts: Sebastes umbrosus and Poecilia reticulata. All the identified SsANXAs proteins were aligned at the amino-acid level using MUSCLE implemented in MEGA12⁶⁸ with default parameters. The phylogenetic tree was constructed using the neighbor-joining algorithm with the Jones–Taylor–Thornton substitution model. Rate variation across sites was set to Uniform Rates, and lineage patterns were assumed homogeneous. Gaps/missing data were handled by Complete Deletion. Branch support was assessed by the bootstrap method with 1000 replicates. Trees were visualized with iTOL (https://itol.embl.de/). Basic protein properties and chromosomal locations of SsANXAs were obtained using TBtools. Results are summarized in Supplementary Data 5 and Fig. S8. Predicted 3D structures of SsANXAs were generated with AlphaFold3 (https://alphafold.com/) and visualized in PyMOL (https://pymol.org/).

TEM observation

Black rockfish ovarian tissues (from unfertilized and blastula-stage samples) were fixed in 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.2–7.4) at 4 °C, rinsed in buffer, and post-fixed in 1% osmium tetroxide with 1.25% potassium ferrocyanide. Subsequently, ultrathin sections of selected areas were mounted on formvar-coated grids, stained with uranyl acetate, and the ultrastructure of the ovarian stroma was examined by TEM using a JEOL-1400Flash electron microscope.

AB-PAS staining

Ovaries (including pigment-stage embryos and surrounding ovarian stroma) from black rockfish and guppy were dissected and fixed in 4% paraformaldehyde (PFA) in PBS at 4 °C for 16 h. Tissues were processed through a graded ethanol series, cleared in xylene, and embedded in paraffin. Sections (5 μm) were cut on a rotary microtome, mounted on slides, baked at 65 °C for 1 h, deparaffinized in xylene, and rehydrated through graded ethanol to water. For Alcian blue staining, each section received ~100 μL of Alcian blue (pH 2.5) in a humid chamber for 1 h, followed by a 3-min rinse in running tap water. Sections were then oxidized with periodic acid (~100 μL/ section) for 10 min in the dark, rinsed in distilled water on a shaker (5 min), and incubated with Schiff reagent (~100 μL/ section) for 1 h at 37 °C (protected from light). After rinsing in distilled water (5 min), slides were counterstained with hematoxylin for 30 s, differentiated in acid alcohol (1% HCl in 70% ethanol) for 30 s, rinsed twice in tap water, and blued in running tap water for 5 min. Finally, sections were dehydrated through graded ethanol, cleared in xylene, and mounted with neutral resin. Images were acquired at 10× and 20× magnification.

FISH

Sense and antisense riboprobes for cldn3 and anxa1a were synthesized using DIG-dUTP and Bio-dUTP, respectively (Roche Diagnostics, Mannheim, Germany). Primers for probe sequences are listed in Supplementary Data 6. All procedures were performed under RNase-free conditions.

Ovarian stroma from early gestation was fixed in 4% paraformaldehyde (PFA) in PBS at 4 °C for 16 h, processed through a graded ethanol series, cleared in xylene, and embedded in paraffin. Sections (7 μm) were cut on a rotary microtome and mounted on charged slides. Slides were deparaffinized in xylene, rehydrated through a graded ethanol series, and rinsed in PBS. To properly prepare the tissues, sections were quenched for endogenous peroxidase activity, permeabilized with a Proteinase K (PK) working solution, and blocked for endogenous biotin. The slides were then equilibrated in hybridization buffer (50% formamide, 10% dextran sulfate, 5×SSC, 5× Denhardt’s, 0.5 mg/mL tRNA). After pre-hybridization at 55 °C for 1 h in a humid chamber, a mixture of the respective antisense (or sense) probes was applied in hybridization buffer (100 μL/section) and incubated overnight at 55 °C. Sections were washed in decreasing-stringency SSC buffers and rinsed in PBS, then blocked with 10% goat serum in PBS (Invitrogen) for 30 min.

For sequential signal detection, slides were first incubated with an HRP-conjugated anti-DIG antibody (Roche; 1:500 in blocking buffer) and rinsed twice with sterile PBS for 5 min each time. The first fluorescent TSA detection was performed using a tyramide kit with Alexa Fluor 555 (Invitrogen, Carlsbad, CA, USA) for approximately 30 min. To prevent signal crosstalk, residual peroxidase activity was subsequently quenched again. The slides were then incubated with HRP-conjugated streptavidin (Roche; 1:500 in blocking buffer) to detect the biotinylated probe. Following additional PBS washes, the second signal was developed using a tyramide kit with Alexa Fluor 488 (Invitrogen) for approximately 30 min. Nuclei were counterstained with DAPI in an antifade mounting medium (G-CLONE, Beijing, China). Images were captured using an Olympus BX53F fluorescence microscope (Olympus, Japan).

IHC and IF

Paraffin sections (ovarian stroma from unfertilized and normal-gestation blastula-stage samples) were deparaffinized in xylene, rehydrated through a graded ethanol series to water, and rinsed in PBS. Heat-induced epitope retrieval was performed in 10 mM sodium citrate buffer (pH 6.0) using a microwave oven for 30 min. For IHC, endogenous peroxidase activity was quenched with 3% H₂O₂ in methanol for 10 min. Sections were then blocked with 5% BSA in PBS for 30 min at 37 °C.

To detect HSD3B1 (a key enzyme for progesterone synthesis), sections were incubated overnight at 4 °C with a rabbit anti-HSD3B1 monoclonal antibody (Abmart, T58791S; 1:1000 in 5% BSA/PBS). After four washes with PBST (PBS containing 0.1% Tween-20), sections were incubated for 1 h at room temperature with an HRP-conjugated goat anti-rabbit IgG (H + L) secondary antibody (Yamay, Shanghai, China; 1:500). Color development was carried out using a DAB chromogen kit (Njjcbio, Nanjing, China) in the dark (~3 min, monitored under a microscope), followed by a PBST wash to stop the reaction. Slides were counterstained with hematoxylin, dehydrated, cleared in xylene, and mounted with neutral balsam. Bright-field images were acquired using a standard light microscope.

IF was used to determine the spatial localization of GYS1, a key enzyme involved in glycogen synthesis, at the protein level. The section preparation and antigen retrieval procedures for IF were identical to those used for IHC, except that the endogenous peroxidase quenching step was omitted. After blocking, the sections were incubated with a rabbit anti-GYS1 polyclonal antibody (Proteintech, 10566-1-AP, Wuhan, China; 1:2000, diluted in blocking buffer) overnight at 4 °C. Sections were then washed and incubated with a GF555-conjugated goat anti-rabbit IgG (H + L) (Yamay) for 1 h at room temperature in the dark. Subsequently, the samples were counterstained with DAPI in an antifade mounting medium (G-CLONE, Beijing, China). Fluorescence images were captured using an Olympus BX53F fluorescence microscope (Olympus, Japan).

RNA extraction, cDNA synthesis, and quantitative real-time PCR (RT-qPCR)

Ovarian stroma tissues of unfertilized, normal-gestation, and abnormal-gestation groups were placed in 1 mL of TRIzol reagent (Vazyme, Nanjing, China) with solid glass beads and homogenized using a high-throughput tissue lyser (DHSbio, Beijing, China). Total RNA was then extracted according to the manufacturer’s protocol. RNA integrity was assessed by agarose gel electrophoresis, and concentration and purity were determined with a biophotometer (OSTC, Beijing, China). Total RNA was reverse transcribed into complementary DNA (cDNA) using the HiScript III 1st Strand cDNA Synthesis Kit (Vazyme, Nanjing, China) following the manufacturer’s instructions.

RT-qPCR was performed on a StepOnePlus Real-Time PCR System (Thermo Fisher Scientific, USA) using the ChamQ SYBR Color qPCR Master Mix (Vazyme, Nanjing, China) according to the manufacturer’s instructions. All primers used in the present study are listed in Supplementary Data 6. Cycling conditions were as follows: 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s. Relative gene expression was calculated using the 2^−ΔΔct method, normalized to 18S rRNA. All data are presented as the mean ± SEM. Statistical analyses were performed using an unpaired two-tailed Student’s t-test in GraphPad Prism, and differences were considered significant at P < 0.05.

Statistics and reproducibility

Sample sizes, statistical tests, and significance thresholds are provided in the “Methods,” figure legends, or both. RT-qPCR data are presented as mean ± SEM from biologically independent samples, with each biological replicate averaged from technical triplicates. Histological and imaging-based experiments were repeated in at least three biologically independent samples with consistent results. Replicates are defined as biologically independent animals or independently collected samples unless otherwise specified.

Reporting summary

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

Author contributions

Xin Qi: conceptualization, funding acquisition, software, resources, writing—review and editing; Xinlin Yang: conceptualization, investigation, writing—original draft; Jianshuang Li: data curation; Bingyan Zheng: data curation; Chong Zhang: methodology; Haishen Wen: conceptualization; Yun Li: conceptualization.

Peer review

Peer review information

Communications Biology thanks Hatice Osmanbeyoglu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Tobias Goris. A peer review file is available.

Data availability

The raw single-cell RNA sequencing data (FASTQ files) and the processed gene expression matrices generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) database under the accession number GSE324772. To facilitate interactive data exploration and visualization of the single-cell transcriptomic atlases of the black rockfish ovarian stroma, we have developed user-friendly web applications using ShinyCell. These atlases can be freely accessed at the following URLs:1. Pregnancy Initiation Atlas (Unfertilized vs. Normal Gestation): https://blackrockfish.shinyapps.io/Pregnancy_Initiation_Atlas/ 2. Pregnancy Maintenance Atlas (Normal vs. Abnormal Gestation): https://blackrockfish.shinyapps.io/Pregnancy_Maintenance_Atlas/. Supplementary numerical data associated with this study are provided in Supplementary Data. All other data supporting the findings of this study are available within the paper and its Supplementary Information or from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-026-10195-y.