Single-cell and spatial transcriptomics characterisation of RSPO2⁺ nucleus pulposus cells reveals a WNT/FN1–CD44 degenerative axis and therapeutic targets in IVDD

Abstract

Background

Intervertebral disc degeneration (IVDD) is a common cause of chronic low back pain, imposing a significant economic and physiological burden on individuals and society worldwide. Although dysregulation of the WNT/β-catenin pathway is considered an important factor contributing to the dysfunction of nucleus pulposus (NP) cells and degradation of the extracellular matrix, the mechanisms by which specific subgroups of NP cells are activated and the maintenance of excessive activation of specific pathways remain unclear.

Methods

We combined single-cell RNA sequencing and spatial transcriptomics with human disc specimens, a rat tail-compression model using static 1.5 MPa loading, an independent rat needle-puncture time-course model, primary NP-cell experiments using continuous static hydrostatic pressure (0.2 MPa for 24 h), WNT-pathway perturbation, molecular docking and dynamics, and targeted siRNA. Computational analyses were interpreted at the cell/spot level, whereas animal-level inference used the stated biological replicates.

Results

Single-cell transcriptomics identified R-spondin 2 (RSPO2) as a selective marker of a homeostatic NP-resident Cluster 1 compartment whose proportion contracted after mechanical injury, while RSPO2-related WNT/FN1–CD44 signalling scores increased across expanded degenerative effector populations. Human panels were used as representative cross-sectional comparisons without inferential between-grade testing. In NP cells, 0.2 MPa pressure and exogenous RSPO2 enhanced WNT/β-catenin, FN1–CD44, matrix-catabolic and apoptotic responses; IWR-1 and DKK1 attenuated these changes. Molecular docking, 100-ns dynamics and a forward co-immunoprecipitation were consistent with an RSPO2–LGR4 association. RSPO2 worsened degeneration in the 1.5 MPa rat compression model, whereas combined IWR-1 treatment was protective. Spatial transcriptomics provided descriptive maps across the needle-puncture time course, and machine-learning models showed internal spot-level discrimination of the prespecified RSPO2-associated labels.

Conclusion

RSPO2 is a candidate regulator, rather than only a marker, of an IVDD-associated NP state. The perturbation data support an RSPO2-associated WNT amplification program coupled to FN1–CD44/MMP3 activation, but the proposed ordering remains provisional pending independent-siRNA and rescue validation.

The translational potential of this article

This study identifies an RSPO2-associated NP state and a candidate WNT/FN1–CD44 programme linked to IVDD. The findings provide a testable basis for future WNT-directed intervention studies while recognising the limits of acute rat models and cross-species data.

Graphical abstract

1. Introduction

Low back pain (LBP) is a common musculoskeletal disorder. According to the Global Burden of Disease data from 2020, more than 600 million people worldwide are currently affected by LBP, and predictive analysis shows that this number will increase to 843 million by 2050 [1]. LBP is a leading cause of years lived with disability (YLD) worldwide [1] and a major condition contributing to global rehabilitation needs [2]. Degenerative lumbar spine disease contributes substantially to the global clinical burden [3]. Intervertebral disc degeneration (IVDD) is a major contributor to LBP and reflects progressive loss of disc homeostasis [4]. Although some progress has been made in studying the mechanism of IVDD, the current treatment strategies aim merely to relieve symptoms and cannot prevent or reverse the degeneration process [5]. This treatment gap highlights the urgent need to clarify the specific molecular mechanisms of IVDD and identify new effective intervention targets.

The nucleus pulposus (NP) is the gel-like structure at the centre of the intervertebral disc, which maintains the internal homeostasis of the tissue by balancing the synthesis and degradation of the extracellular matrix. Advancements in single-cell RNA sequencing (scRNA-seq) research have changed our understanding of the biological characteristics of NP cells and revealed significant heterogeneity within the cellular microenvironment [6,7]. A landmark discovery within the intervertebral disc was the identification of distinct subpopulations of chondrocytes, including stable cells, regulatory cells, and effector cells, as well as the specific PROCR + stem cells with strong trilineage differentiation ability [8]. Spatial transcriptomics analysis further enriched this cellular landscape, identifying cathepsin K (CTSK) as a stem cell marker; stem cells are concentrated mainly in the extracellular region of the cartilage disc [9]. Importantly, CD44⁺ chondrocytes are a functional cell subset, whose expression decreases with ageing and disease progression [10]. As degeneration intensifies, homeostatic and progenitor-like NP states may be lost, while stress- and matrix-remodelling programmes become more prominent [6,8,9]. Related senescence-associated remodelling has also been described in osteoarthritic cartilage and meniscus [11], although its relevance to IVDD remains unresolved. Different specific pathological chondrocyte subpopulations may balance disease development via distinct signal cascades, but the identities of these cell subpopulations have not been clearly elucidated. The WNT/β-catenin signalling cascade connects the mechanical stress within the intervertebral disc, cellular dysfunction, and key mediators of cartilage matrix degradation metabolism. During the process of IVDD, multiple upstream regulators modulate β-catenin activation, including periostin (POSTN), which is aberrantly expressed in degenerated discs and promotes NP-cell apoptosis by activating WNT/β-catenin signalling [12]. Increased mechanical loading (compressive stress) also leads to upregulation of WNT/β-catenin, thereby inhibiting cell proliferation and promoting cell apoptosis [13]. Overactivation of the WNT pathway also leads to the expression of downstream matrix metalloproteinases and to reduced synthesis of type II collagen or aggrecan. This subsequently leads to cell ageing and increased apoptosis, thereby collectively disrupting the stability of the cartilage matrix [14,15]. Targeting the p300/FOXO3/Sirt1 axis to inhibit WNT/β-catenin signalling has had significant protective effects in both in vivo and in vitro IVDD models [16]. Additionally, non-coding RNAs, such as the circITCH/miR-17-5p/SOX4 regulatory circuit, can regulate WNT pathway activity, and increased circITCH levels lead to SOX4 upregulation, thereby triggering extracellular matrix degradation driven by β-catenin [17]. Although these advances have been made, the upstream regulators that initiate the overactivation of the WNT pathway in specific NP subpopulations during the early degeneration process remain unclear.

R-spondin 2 (RSPO2) is a secreted glycoprotein that enhances canonical WNT/β-catenin signalling through LGR4/5/6 receptors and the ZNRF3/RNF43 E3 ubiquitin ligase module, thereby stabilising Frizzled receptors and increasing cellular sensitivity to WNT ligands [18]. In bone and joints, RSPO2 can be strongly activated after injury and coordinates pathological interactions among synovial fibroblasts, macrophages, and chondrocytes through WNT/β-catenin overactivation [19]. The RSPO2–LGR4 signalling pathway regulates the WNT inhibitor DKK1 through Gαq and β-catenin pathways, indicating a complex regulatory role in musculoskeletal tissues [20]. Moreover, bioinformatics and experimental studies have identified fibronectin (FN1) as an extracellular-matrix-related regulatory gene in IVDD [[21], [22], [23]]. During disc degeneration, FN1 deposition and fibronectin-fragment generation can promote matrix-catabolic responses and the expression of degrading enzymes [24,25]. CD44 is a hyaluronan receptor that can also interact with osteopontin and participate in WNT-related signalling [26,27]. Single-cell analyses identify SPP1-associated communication changes during IVDD [22], whereas fibronectin expression and extracellular-matrix remodelling have been reported in degenerated disc tissue [23]. In addition, age-associated changes in CD44-positive NP cells may contribute to the loss of matrix homeostasis [10,28].

Based on converging evidence, we hypothesised that an Rspo2-expressing NP cell state contributes to IVDD by amplifying WNT/β-catenin signalling. This signalling upregulates FN1 expression and activates the FN1–CD44 axis, driving extracellular matrix (ECM) remodelling. Because clinically obtainable early-stage human disc tissue is scarce, temporally defined rat injury models were used to resolve early RSPO2-associated responses, whereas human specimens provided cross-sectional comparison across the clinically available degeneration grades. This design permits mechanistic investigation of early events but does not assume temporal equivalence between acute rat injury and idiopathic human IVDD. To test the proposed pathway relationship, we combined molecular, cellular, and organismal approaches. ScRNA-seq of whole-disc tissue from a pressure-induced rat IVDD model, followed by NP-focused analyses, characterised cellular heterogeneity and identified degeneration-associated cellular programmes. CellChat analysis revealed RSPO2-associated communication networks in the degenerating disc, which we evaluated in human disc samples and a pressure-overload rat model. In primary NP cells, we applied mechanical pressure (0.2 MPa) together with RSPO2 exposure, WNT inhibition (IWR-1 and DKK1), and targeted knockdown to test the proposed pathway-level responses. Molecular docking and 100 ns molecular dynamics simulations examined protein–protein interactions involving RSPO2–LGR4, FN1–CD44, RSPO2–ZNRF3, and IWR-1–tankyrase. To evaluate these findings in vivo, we administered RSPO2 with or without a WNT inhibitor via intradiscal injection in a rat compression model and assessed imaging, histopathological, and molecular outcomes. Machine-learning models were used to evaluate the internal discriminative value of RSPO2-associated features, and descriptive plots of public human IVDD GWAS summary statistics provided genetic context. Our multi-scale study supports the interpretation of RSPO2-associated NP cells as a degeneration-associated state rather than a lineage-defined progenitor population and provides a testable framework for WNT-directed intervention.

2. Materials and methods

2.1. Ethics statement, animal use, and human disc specimens

All animal experiments in this study were approved by the Experimental Animal Ethics Association of Anhui Medical University (approval number: LLSC-20242216), and conformed to the guidelines of the ARRIVE protocol. Eight-to ten-week-old male Sprague–Dawley (SD) rats weighing 220–250 g were raised in a pathogen-free environment. The housing environment consisted of a 12-h light/dark cycle, a temperature range of 22–24 °C, a humidity level of 50–60%, and a supply of sufficient food and water.

Human lumbar disc tissues were obtained from 24 patients who underwent spinal surgery at the First Affiliated Hospital of Anhui Medical University for vertebral fracture or lumbar disc herniation. Preoperative MRI was used for Pfirrmann grading. Written informed consent was obtained from all participants or their legal guardians, and the protocol was approved by the Ethics Committee of the First Affiliated Hospital of Anhui Medical University. Specimens were used for immunofluorescence as described below; human panels in Fig. 2 are representative and were not included in inferential between-grade statistical testing.

Fig. 2.

RSPO2 expression correlates with intervertebral disc degeneration in human specimens and the rat pressure model. A Representative T2-weighted MRI images of the human lumbar spine showing Pfirrmann grades I–II (mild), III–IV (moderate), and V (severe) disc degeneration. B Immunofluorescence co-staining for FN1 (green) and RSPO2 (red) in human disc specimens across Pfirrmann grades. Scale bars: 0.5 mm (5×, upper panels) and 0.2 mm (20×, lower panels). C Representative X-ray and MRI images of the rat caudal spine from Sham and Operation groups at 4 weeks, with HE and Safranin O staining. Scale bars: 5 mm (X-ray and MRI) and 500 μm (HE and Safranin O). D Triple immunofluorescence for FN1 (purple), CD44 (green), and MMP3 (red). E Triple immunofluorescence for FN1 (purple), CD44 (green), and RSPO2 (red). Scale bars: 0.5 mm (5×) and 0.05 mm (40×). F Quantification of disc height index, Pfirrmann grade, and histological score. Panels A–B show representative images from the human cohort (total n = 24) and were not included in an inferential between-grade comparison. For panel F, data are mean ± SD; n = 6 animals per group. Comparisons used unpaired two-tailed Student's t-tests; ***P < 0.001.

2.2. Reagents and antibodies

Primary antibodies used included the following: anti-RSPO2 (Abcam, ab203510; 1:1000 western blot (WB), 1:200 immunohistochemistry (IHC)), anti-β-catenin (Cell Signaling Technology, 8480; 1:1000 WB, 1:200 immunofluorescence (IF)), anti-MMP3 (Affinity Biosciences, AF0217; 1:1000 WB, 1:100 IF), anti-aggrecan (Affinity Biosciences, DF7561; 1:1000 WB, 1:200 IF), anti-collagen II (Affinity Biosciences, AF0135; 1:1000 WB, 1:200 IF), anti-FN1 (Proteintech, 15613-1-AP; 1:1000 WB, 1:200 IF), anti-CD44 (Affinity Biosciences, DF6392; 1:1000 WB, 1:150 IF), anti-Bcl2 (Cell Signaling Technology, 15071; 1:1000 WB), anti-Bax (Cell Signaling Technology, 2772; 1:1000 WB), anti-β-actin (Proteintech, 66009-1-Ig; 1:5000 WB), and anti-GAPDH (Proteintech, 10494-1-AP; 1:5000 WB). Secondary antibodies included HRP-conjugated goat anti-rabbit and anti-mouse IgG (Beyotime), and Alexa Fluor 488/594/647-conjugated goat anti-rabbit IgG (Invitrogen) at 1:500. Pharmacological reagents included the following: recombinant rat RSPO2 protein (R&D Systems), tankyrase inhibitor IWR-1 (Selleck, S7086), and recombinant DKK1 protein (R&D Systems, 5439-DK).

2.3. Animal models of intervertebral disc degeneration

Rat Tail Compression Model. Animals were randomly allocated to four groups (n = 6 per group; 24 rats total): sham, operation + vehicle, operation + RSPO2, and operation + RSPO2 + IWR-1. Rats were anaesthetised with 1% pentobarbital sodium (4 mL/kg, intraperitoneally; 40 mg/kg). Under fluoroscopic guidance, Kirschner wires were inserted through the Co5 and Co6 vertebral bodies, and a calibrated spring was placed to generate static 1.5 MPa compression. Spring length and loading were checked and recalibrated before use. The sham-operation group underwent the same anaesthesia, fluoroscopic positioning, and Kirschner-wire placement procedure without spring compression. In figures and text, 'operation,' 'degeneration,' and 'mechanical pressure' refer to the pressure-loaded degeneration group; 'operation' is used as the primary label for consistency. Immediately after modelling, discs were injected with recombinant RSPO2 (100 ng in 10 μL PBS), IWR-1 (1 μM in 10 μL), RSPO2 + IWR-1, or vehicle control. Discs were harvested 4 weeks later. Imaging, histological, and immunofluorescence analyses used n = 6 animals per group; western-blot densitometry in Fig. 8B–F was based on n = 3 quantified tissue lysates per group.

Fig. 8.

In vivo perturbation supports RSPO2-mediated disc degeneration through the WNT/β-catenin/FN1–CD44 axis. A Western blot of CD44, FN1, β-catenin, MMP3, and ACAN in disc tissues. B–F Protein quantification. G Triple immunofluorescence for FN1 (purple), CD44 (green), and MMP3 (red). H Triple immunofluorescence for FN1 (purple), CD44 (green), and RSPO2 (red). Scale bars: 0.5 mm (5×) and 0.05 mm (40×). I–J Fluorescence-intensity quantification. For panels B–F, data are mean ± SD from n = 3 quantified tissue lysates per group. For panels I–J, data are mean ± SD from n = 6 animals per group. Comparisons used one-way ANOVA with Tukey's post hoc test; ns, P ≥ 0.05; *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Rat Needle Puncture Model. An independent cohort of 24 rats was randomly allocated to four post-injury time-point groups (n = 6 per group). A 21-gauge needle was inserted 5 mm into the Co7-Co8 NP, rotated 360°, and held for 10 s before withdrawal. Discs were collected at 0, 1, 3, and 28 days post-injury. Histological scoring used the six animals assigned to each time point, whereas spatial transcriptomics used a prespecified subset of three animals per time point (one section per animal).

2.4. Imaging assessment

Four weeks after the operation, sagittal T2-weighted magnetic resonance images of the rat caudal spine were obtained using a 3.0-T MRI system, and degeneration was graded according to the Pfirrmann classification. X-ray images were acquired to calculate the disc height index. For histological evaluation, disc tissues were fixed, decalcified, embedded in paraffin, sectioned, and stained with haematoxylin-eosin (HE) and Safranin O/Fast Green (SO/FG). Histological degeneration was scored using a modified Thompson grading system by blinded observers.

2.5. Single-cell RNA sequencing

Four weeks after pressure modelling, whole caudal discs were harvested from sham and operation rats (n = 3 animals per group; six sequenced samples in total) and enzymatically digested to obtain single-cell suspensions. Single-cell libraries were prepared using the 10x Genomics Chromium platform and sequenced on an Illumina platform. Data were processed with Cell Ranger (v6.0) and analysed using Seurat (v4.0). Before doublet/multiplet and apoptotic or other low-quality cell removal, quantified cell yields ranged from 13,028 to 17,699 per sample; 9037–15,785 cells per sample were retained after quality control. Across samples, the mean UMI count per cell ranged from 5330 to 9,166, the mean detected gene count per cell ranged from 1817 to 2,241, and the mean mitochondrial UMI fraction ranged from 5.50% to 7.40%. Cells with fewer than 200 detected genes or a mitochondrial read ratio exceeding 10% were excluded. Unsupervised dimensional reduction and clustering of the full dataset yielded 19 clusters with broad reference annotations including macrophages, fibroblasts, B cells, stem cells, endothelial cells, innate lymphoid cells, neutrophils, monocytes, and dendritic cells. The NP-focused subset displayed in Supplementary Fig. 1–3 comprised 3216 cells across 11 clusters. CellChat and pseudotime analyses were used to explore cell–cell communication and cell-state dynamics.

2.6. Spatial transcriptomics

For the rat puncture model, spinal discs were obtained at 0 days, 1 day, 3 days and 28 days after injury (the full cohort had n = 6 animals per time point; spatial transcriptomics used one section from each of three animals per time point). The spinal discs were embedded in OCT, and frozen sections were made on the Visium spatial gene expression slide (10x Genomics). Spot-level counts were generated using Space Ranger and analysed using Seurat. Spatial regions were identified using the BayesSpace R package (v1.4.0) with q = 3 clusters (domains D0, D1, and D2), 15 principal components, gamma = 3 (spatial smoothing parameter), and 10,000 MCMC iterations. Domain identity was biologically annotated by marker enrichment: D0 represented the degenerative core enriched for Mmp3/Mmp13 and RSPO2-associated activity, D1 represented a transitional region, and D2 represented relatively preserved tissue enriched for Acan and Col2a1.

2.7. Primary nucleus pulposus cell culture and mechanical stress

Primary nucleus pulposus (NP) cells were isolated from rat caudal discs by digestion with 0.25% trypsin-EDTA and 0.2% collagenase II, and then cultured at 37 °C in 2% O2 and 5% CO2. Adherent NP cells were exposed to continuous static hydrostatic pressure at 0.2 MPa for 24 h; unpressurised sham-loading controls were handled under identical conditions without pressure. Where indicated, recombinant RSPO2 (100 ng/mL), IWR-1 (10 μM), or DKK1 (100 ng/mL) was added to the culture medium 2 h before pressure loading and maintained throughout the pressure protocol; these concentrations were selected based on published NP-cell studies and preliminary titration experiments. For loss-of-function experiments, adherent NP cells at 30-50% confluence in six-well plates were transfected with one HPLC-purified siRNA duplex targeting Rspo2 (siRspo2-1500), one targeting Fn1 (siFn1-378), or a non-targeting control (siNC). All duplexes and Rfect V2 siRNA Transfection Reagent were obtained from Changzhou Baidai Biotechnology Co., Ltd. (Changzhou, China). Per well, 60 pmol siRNA and 7.5 μL Rfect V2 were separately diluted in 125 μL Trans Enhancer. After 5 min, the solutions were combined, incubated for 15 min, and added as a 250 μL complex to 2.5 mL antibiotic-free complete medium without subsequent medium replacement (approximately 22 nM siRNA based on the final well volume). The duplex sequences (sense/antisense, 5′-3′) were siRspo2-1500, CAACCGCACGUGUGGAUUUAATT/UUAAAUCCACACGUGCGGUUGTT; siFn1-378, GGGAAGCAUUAUCAGAUAAAUTT/AUUUAUCUGAUAAUGCUUCCCTT; and siNC, UUCUCCGAACGUGUCACGUTT/ACGUGACACGUUCGGAGAATT. At 24 h after transfection, cells were exposed to continuous static hydrostatic pressure at 0.2 MPa for a further 24 h and harvested immediately for western blotting (48 h after transfection); reduction of the corresponding target protein was verified in the same experiments. A second sequence-independent siRNA and an siRNA-resistant rescue construct were not evaluated. Quantification was based on three independent cell experiments.

2.8. Western blotting and Co-IP

The processed cells or rat tissue sections were homogenised using RIPA buffer to extract proteins. Then, 20–40 μg of protein from each sample was separated by SDS-PAGE and transferred onto a PVDF membrane. Bands were detected using enhanced chemiluminescence and quantitatively analysed using ImageJ. For Co-IP, mechanically stressed primary NP cells were lysed in IP lysis buffer containing protease inhibitors, and the clarified supernatant was incubated overnight at 4 °C with an anti-RSPO2 antibody. Protein A/G agarose beads were then added, followed by washing and elution of bound proteins. Eluted material was analysed by western blotting with an anti-LGR4 antibody; input and IgG controls were included.

2.9. Immunofluorescence, immunohistochemistry, and TEM

Cells and paraffin-embedded disc sections were fixed in 4% paraformaldehyde, permeabilised with 0.3% Triton X-100 and blocked with 5% serum. Imaging was performed using a Leica TCS SP8 confocal microscope. For immunohistochemistry, antigen retrieval was performed using citrate buffer, and staining was visualised with a DAB kit. For transmission electron microscopy, treated NP cells were seeded in six-well plates, digested and collected, and the supernatant was discarded. One millilitre of electron-microscopy fixation solution was added gently; cells were fixed at room temperature for 1 h and then transferred to 4 °C overnight. After routine specimen preparation, the samples were examined by TEM.

2.10. Flow cytometry

Apoptosis was analysed by Annexin V-FITC/PI staining in three independent cell experiments. NP cells were stained with Annexin V-FITC and propidium iodide (BD Biosciences) and analysed on a FACSCanto II flow cytometer. Data were processed with FlowJo.

2.11. Molecular docking and molecular dynamics simulation

Three-dimensional structures were obtained from the Protein Data Bank or AlphaFold. Molecular docking was performed with AutoDock Vina for the RSPO2–LGR4, RSPO2–ZNRF3, FN1–CD44, and IWR-1–tankyrase complexes. The top-ranked RSPO2–LGR4 pose was then subjected to a 100 ns molecular-dynamics simulation using GROMACS with the AMBER ff14SB force field in a TIP3P water box. Binding free energy for the simulated RSPO2–LGR4 complex was computed using the molecular mechanics/generalised Born surface area (MM/GBSA) method.

2.12. Machine learning and descriptive GWAS visualisation

Classification models were built to distinguish a custom RSPO2-positive label from RSPO2-negative spatial units using signature-gene expression, pathway/module scores, time-point indicators, and quality-control covariates as features. Within each time point, the custom positive label was assigned by ranking k-nearest-neighbour-smoothed Rspo2 expression to prespecified positivity rates (Control, 1.65%; Day 1, 3.42%; Day 3, 5.63%; Day 28, 9.63%). Multiple algorithms (random forest, XGBoost, LightGBM, and gradient boosting), a soft-voting ensemble, and a holdout-stacking model were evaluated using a single stratified 80:20 spot-level train–test split (random seed 42). Permutation-based feature importance and local model-agnostic explanations were used for interpretation [29]. Because spots from the same tissue sections could occur in both partitions, these metrics represent internal spot-level discrimination rather than section-level or external validation. Spot-level linear models were interpreted as association analyses. For the exploratory mediation graphic, ordinary least-squares models with HC3-robust standard errors were fitted to the Rspo2_smooth → Wnt_target_smooth → FN1CD44_smooth path, with adjustment for time point, log1p library size, and mitochondrial UMI fraction. The indirect effect and its 95% percentile interval were estimated from 500 spot-level bootstrap resamples (random seed 1). Because spots were nested within tissue sections, this analysis was treated as exploratory and does not provide a biological-sample-level mediation estimate or causal inference. Publicly available human IVDD GWAS summary statistics were plotted descriptively for genetic context; because the cellular reference is rat-derived, these plots do not provide cell-state-specific or direct genetic evidence in the animal model.

2.13. Statistical analysis

Quantitative data are expressed as the mean ± SD. Unless otherwise specified in a figure legend, n denotes independent biological replicates (independent cell experiments or animals). Comparisons between two groups used unpaired two-tailed Student's t-tests; multiple-group comparisons used one-way ANOVA with Tukey's post hoc test. Cell- and spot-level computational analyses are identified as such in the corresponding legends and were not equated with independent biological replication. P < 0.05 was considered statistically significant. Significance symbols are defined in each applicable figure legend; panels without inferential testing are described as descriptive.

3. Results

3.1. Single-cell RNA sequencing identifies an RSPO2-Defined NP compartment and degenerative signalling program

To comprehensively characterise the cellular heterogeneity within the intervertebral disc tissue under pathological conditions, we performed scRNA-seq on whole-disc tissues (encompassing the NP, annulus fibrosus, and adjacent compartments) harvested from rats subjected to mechanically induced degeneration and sham operation controls. Unsupervised clustering of the full dataset yielded 19 clusters with broad cell-type annotations. For the NP-focused analyses presented in Supplementary Fig. 1–3, 3216 cells were resolved into 11 transcriptionally distinct clusters, with proportions varying markedly between the sham and operation groups (Supplementary Fig. 1A–B). Hierarchical clustering of cluster-defining marker genes resolved the population identities, including Cluster 1 — the homeostatic NP-resident cell compartment characterised by Crabp2, Fndc1, Kera, Igf1, Col14a1, Clec2dl1, Col8a2, Gpr15lg, and Rrad expression (Supplementary Fig. 1C). Within this Cluster 1 compartment, Rspo2 emerged as a highly selective marker in the cell-level analysis (log2FC = 0.85, pct.1 = 30.9%, pct.2 = 6.4%), with feature and violin plots confirming that Rspo2 expression was largely restricted to a sub-fraction of Cluster 1 cells while remaining near-absent in all other clusters (Supplementary Fig. 1D–E). CellChat information flow profiling revealed pronounced upregulation of COLLAGEN, FN1, SPP1, LAMININ, ANGPTL, TENASCIN, THBS, PERIOSTIN, PTN, and MIF pathways in the operation group. At the same time, CADM signalling was predominantly enriched under sham conditions, indicating a global rewiring of intercellular communication networks during degeneration (Supplementary Fig. 1F).

To further dissect how RSPO2⁺ cells engage neighbouring NP subpopulations, we performed comprehensive CellChat-based intercellular communication analysis comparing sham and operation conditions. Module scoring of ECM-related and WNT signalling pathway gene sets across the UMAP landscape demonstrated spatially coordinated activation, with regions of high ECM and WNT activity overlapping with RSPO2⁺ cell territories (Supplementary Fig. 2A–B). STRING-based protein–protein interaction analysis placed RSPO2, WNT2, FZD9, FN1, and CD44 within the same inferred interaction network, providing association-level support for pathway connectivity (Supplementary Fig. 2C). Comparative information flow analysis indicated substantial enhancement of FN1, COLLAGEN, THBS, SPP1, NCAM, and TENASCIN signalling in the operation condition (Supplementary Fig. 2D). Cell-level exploratory correlation panels showed positive associations between Rspo2 and WNT pathway activity and between Fn1 and Cd44, together with inverse associations between CD44/WNT scores and structural ECM scores (Supplementary Fig. 2E). The displayed coefficients and P values are panel-level exploratory outputs; panel-specific filtering and effective sample sizes were not treated as animal-level replicates. Chord diagrams of the FN1 signalling network demonstrated dense and topologically simplified connectivity in operation tissue, with FN1–CD44, FN1–SDC4, and FN1–integrin (ITGAV/ITGA4/ITGA5/ITGB1) ligand–receptor pairs becoming dominant communication routes (Supplementary Fig. 2F). Quantitative analysis of communication pathway changes around the RSPO2-defined Cluster 1 revealed substantial upregulation of COLLAGEN, FN1, SPP1, MIF, and THBS pathways in operation tissue, accompanied by selective downregulation of CD99 and ANNEXIN signalling (Supplementary Fig. 2G). Outgoing signalling pattern analysis identified COLLAGEN, FN1, SPP1, THBS, LAMININ, and TENASCIN as the dominant secretion programs in pressure-stimulated NP cells (Supplementary Fig. 2H). Network-level interaction analysis indicated that the RSPO2-defined compartment occupies a central hub position with extensive bidirectional communication links to all other NP cell subpopulations under operation conditions (Supplementary Fig. 2I). Collectively, these analyses position the RSPO2-defined cellular program as a candidate communication hub associated with multi-pathway crosstalk in the degenerative niche.

Quantitative analysis of each cell cluster revealed that Cluster 1, the homeostatic Rspo2-expressing NP-resident cell compartment, contracted from approximately 40% in the sham group to less than 10% in the operation group, whereas the degenerative effector clusters, including Clusters 2 and 5, increased in this dataset (Supplementary Fig. 3A). Importantly, this pattern reflects the loss of a stable NP-resident homeostatic identity after mechanical injury rather than simple depletion of the RSPO2 signalling program. Further module scoring showed that the RSPO2 pathway, FN1–CD44 axis, WNT signalling, and ECM degradation scores expanded across multiple effector clusters in the operation group, particularly Clusters 2, 6, 8, 9, and 11 (Supplementary Fig. 3B–E). The apoptosis score also showed a moderate but consistent increase in most operation clusters (Supplementary Fig. 3F). Monocle pseudotime analysis revealed a branched differentiation continuum spanning NP subpopulations, with WNT signalling activity progressively rising from homeostatic toward degenerative cell states (Supplementary Fig. 3G–H). Branch-dependent gene-module analysis further resolved three major transcriptional programs (Supplementary Fig. 3I). Around Cluster 1, COLLAGEN, FN1, SPP1, MIF, ADGRE, APP, and THBS represented the major pathways with enhanced interaction intensity; CD99 was relatively enriched in the sham group, whereas CXCL emerged as an operation-specific output signal (Supplementary Fig. 3J). Overall, these results indicate that mechanical injury contracts the stable Rspo2-expressing NP-resident cell compartment while propagating a degeneration-associated signalling program across a broader NP network.

3.2. Spatial transcriptomics maps RSPO2-Associated features across disc degeneration stages

To complement the pressure-model scRNA-seq analysis with anatomical information from an independent degeneration model, we performed spatial transcriptomics on rat NP tissues collected from the needle-puncture model at control, day 1, day 3, and day 28 after injury. Because this dataset captured spatially resolved gene-expression territories within the NP region rather than dissociated whole-disc cell clusters, RSPO2 signal was interpreted at the level of RSPO2-expressing spatial units and pathway activity. The categorical RSPO2-positive labels shown in Supplementary Fig. 4A–B were assigned to prespecified within-time-point proportions for descriptive visualisation and were not used to estimate biological prevalence. Gene-expression heatmaps and pathway-score panels descriptively summarised WNT components (Ctnnb1, Lef1, Tcf7), catabolic markers (Mmp3, Mmp13, Adamts4), and anabolic markers (Acan, Col2a1, Sox9) across the displayed spatial units (Supplementary Fig. 4C–D). Interaction-network, progression-model, and spatial-mapping panels were treated as exploratory summaries rather than animal-level inference (Supplementary Fig. 4E–H).

UMAP visualisation of the complete spatial-transcriptomic dataset displayed the prespecified RSPO2-positive proportions across the four time points (1.6%, 2.4%, 3.3%, and 9.6%); these percentages define the descriptive labels and do not constitute an independently estimated temporal increase (Fig. 1A). Gene-set and pathway panels were examined as exploratory spot-level summaries of IVDD-related and WNT features (Fig. 1B and C). The Ripley's L-function panel was retained as a descriptive spatial summary and was not used for animal-level inference (Fig. 1D). Pseudotime analysis revealed coordinated expression of key pathway genes, and ligand-receptor analysis highlighted FN1–CD44 as a prominent interaction associated with RSPO2-positive regions (Fig. 1E and F). The corresponding inferred cell–cell communication network indicated that RSPO2-positive territories occupy central positions (Fig. 1G). The trajectories of Ctnnb1, Fn1, Cd44, Mmp13, Col2a1, Sox9, Acan, and Wnt2 illustrated gene-expression patterns along the displayed pseudotime (Fig. 1H). The integrated pathway-score summary and schematic model illustrate coordinated multi-pathway activation associated with RSPO2-positive territories (Fig. 1I). Spatial domain analysis of day 28 tissue identified three anatomical regions: D0, representing the degenerative core; D1, representing the transitional zone; and D2, representing relatively preserved tissue. RSPO2-positive territories and the FN1–CD44 signalling axis were enriched in D0 (Supplementary Fig. 5). Taken together, these exploratory spatial panels complement the pressure-model scRNA-seq analysis at the level of tissue-pattern visualisation, but they do not establish expansion of an RSPO2-positive lineage or a biologically estimated increase in prevalence.

Fig. 1.

Descriptive spatial mapping of RSPO2-associated features across disc-degeneration stages. A UMAP visualisation coloured by disease stage and RSPO2 status, with quantification of RSPO2-positive spatial units. B–C Gene-set scores and pseudotime pathway dynamics. D Ripley's L-function analysis of RSPO2-positive spatial clustering across disease stages. E Selected gene-expression trends along pseudotime. F–G Top ligand–receptor pairs and the cell–cell communication network. H Individual gene-expression dynamics along pseudotime. I Normalised pathway-score summary and schematic model. For the descriptive spatial visualisations in this figure and Supplementary Fig. 4, RSPO2-positive labels were assigned to the within-time-point proportions displayed in the figures (Control, 1.6%; Day 1, 2.4%; Day 3, 3.3%; Day 28, 9.6%); this label is distinct from the k-nearest-neighbour-smoothed custom label used for Fig. 10 machine-learning analysis and does not independently estimate biological prevalence. Spatial-transcriptomic data were obtained from n = 3 animals per time point (one section per animal). The cell-/spatial-unit-level summaries are descriptive; no animal-level inferential statistical test is reported for this figure.

3.3. Elevated RSPO2 expression correlates with IVDD severity in human specimens and in the pressure-induced rat model

To assess whether RSPO2-associated staining was evident in clinically available human tissue, we examined representative intervertebral disc specimens spanning Pfirrmann grades (Fig. 2A). Representative immunofluorescence images showed qualitatively stronger RSPO2 and FN1 staining in more degenerated specimens; no inferential between-grade comparison was performed (Fig. 2B). In the animal experiments of the pressure model, X-ray and MRI evaluations at 4 weeks showed that the intervertebral discs in the operation group had marked narrowing of the intervertebral space and decreased T2-weighted signal intensity (Fig. 2C). Furthermore, histological analysis revealed structural damage of the intervertebral discs and loss of proteoglycans. The triple immunofluorescence results showed that, in the degenerated NP region, FN1, CD44, and MMP3 were upregulated (Fig. 2D), and RSPO2 was widely co-localised with FN1 and CD44 in the degenerated intervertebral discs (Fig. 2E). Quantitative results showed that the disc height index in the operation group was significantly decreased (P < 0.001), with a higher Pfirrmann grade (P < 0.001) and an increased histological degeneration score (P < 0.001) (Fig. 2F).

3.4. Mechanical pressure activates the RSPO2/WNT/β-catenin/FN1–CD44 axis in NP cells

To determine whether mechanical stress could directly trigger the degenerative signalling cascade observed in vivo, primary NP cells were exposed to continuous static hydrostatic pressure at 0.2 MPa for 24 h. Western blotting showed that pressure loading markedly reduced COL2A1 expression, while increasing MMP3, FN1, and CD44 levels, indicating a shift from matrix maintenance toward matrix catabolism (Fig. 3A–D). Pressure also promoted an apoptotic phenotype, as reflected by decreased Bcl2 and increased Bax expression (Fig. 3E–G). We next examined whether RSPO2 was involved in this pressure-induced response. Molecular docking illustrated structural compatibility between FN1 and CD44, supporting the possibility that FN1–CD44 signalling participates in the downstream catabolic process (Fig. 3H). In parallel, western blotting showed that pressure increased RSPO2 expression and activated β-catenin signalling, and this effect was further enhanced by exogenous RSPO2 treatment (Fig. 3I–K). Immunofluorescence staining confirmed stronger RSPO2 and β-catenin signals after pressure stimulation, with more evident cytoplasmic and perinuclear accumulation in the pressure + RSPO2 group (Fig. 3L). Triple immunofluorescence further showed increased co-localisation of RSPO2 and CD44 under pressure, particularly after RSPO2 supplementation (Fig. 3M). Consistently, protein-protein interaction analysis placed RSPO2 within a regulatory network connected to WNT ligands, FN1, and CD44 (Fig. 3N).

Fig. 3.

Mechanical pressure activates the RSPO2/WNT/β-catenin/FN1–CD44 signalling axis in nucleus pulposus cells. A Western blot of COL2A1, MMP3, FN1, CD44, Bcl2, and Bax under Control, 0.2 MPa, and 0.2 MPa + RSPO2 conditions. B-G Densitometric quantification. H Molecular-docking model of the FN1–CD44 interaction. I-K RSPO2 and β-catenin western blotting and quantification. L-M Immunofluorescence. Scale bar: 0.1 mm. N PPI network. For panels B-G and J-K, data are mean ± SD; n = 3 independent cell experiments per group. Comparisons used one-way ANOVA with Tukey's post hoc test; ns, P ≥ 0.05; *P < 0.05, **P < 0.01, and ***P < 0.001.

3.5. Exogenous RSPO2 treatment amplifies the degenerative phenotype through WNT/β-catenin activation

We next investigated whether exogenous RSPO2 could amplify the pressure-induced degenerative phenotype. Western blotting showed that RSPO2 treatment increased FN1, β-catenin, CD44, MMP3, and Bax expression, while reducing Bcl2 expression (Fig. 4A–G). The combination of hydrostatic pressure and RSPO2 produced a stronger response than either stimulus alone. Representative flow-cytometry plots showed a pattern consistent with increased apoptosis after RSPO2 exposure and hydrostatic pressure, particularly with combined treatment (Fig. 4H); no independent rate comparison was performed in this panel. Immunofluorescence further confirmed enhanced co-localisation of RSPO2 and FN1 after RSPO2 or pressure treatment, especially in the combined group (Fig. 4I–K). These findings indicate that RSPO2 does not merely accompany pressure-induced degeneration but amplifies WNT/β-catenin activation, FN1–CD44 axis engagement, and apoptosis in NP cells.

Fig. 4.

Exogenous RSPO2 treatment amplifies the degenerative phenotype and apoptosis in NP cells. A Western blot of FN1, β-catenin, CD44, MMP3, Bcl2, and Bax. B–G Protein quantification. H Flow-cytometry plots (Annexin V-FITC/PI). I–K Immunofluorescence co-staining for RSPO2 (green) and FN1 (red). Scale bar: 10 μm. Panel H shows representative flow-cytometry plots from three independent cell experiments; no quantitative apoptosis-rate summary or inferential test is shown for this panel. For panels B–G and J–K, data are mean ± SD; n = 3 independent cell experiments per group. Comparisons used one-way ANOVA with Tukey's post hoc test; ns, P ≥ 0.05; *P < 0.05, **P < 0.01, and ***P < 0.001.

3.6. Molecular dynamics simulation and Co-IP support RSPO2–LGR4 association

Previous studies have shown that RSPO2 can enhance WNT signalling by binding to LGR4, LGR5, and LGR6 receptors [18]. Among these receptors, we chose LGR4 as the primary candidate for two reasons: (i) in our scRNA-seq data, Lgr4 showed the highest expression among the three receptors in NP precursor-like cells, whereas Lgr5 and Lgr6 were expressed at very low levels; and (ii) LGR4 has been linked to RSPO2-dependent regulation of DKK1 and musculoskeletal tissue remodelling [20]. To study the RSPO2–LGR4 interaction at the atomic level, we conducted molecular docking and 100 ns molecular dynamics simulations. Docking predicted contacts between RSPO2 and several LGR4 residues, including ARG-156, ALA-138, ASP-137, and GLU-147 (Fig. 5A and B). Molecular-dynamics simulation indicated that the docked complex reached an apparent equilibrium within 20 ns and remained conformationally stable in silico over the 100-ns trajectory (Fig. 5C). Root-mean-square fluctuation and radius-of-gyration analyses further supported the stability of the binding interface, with the radius of gyration maintained at approximately 4.8–5.0 nm (Fig. 5D and E). Solvent-accessible surface area and inter-protein distance analyses were consistent with a stable binding interface throughout the simulation (Fig. 5F and G). Hydrogen-bond analysis showed 8–14 intermolecular bonds throughout the simulation, and binding-energy decomposition identified VAL-205, ARG-156, and LEU-186 as major contributors (Fig. 5H and I). Free-energy profile analysis showed a clear energy minimum, and simulation snapshots at 0, 50, and 100 ns showed consistent binding conformations (Fig. 5J–M). To further support these computational predictions, we performed co-immunoprecipitation (Co-IP) assays in primary NP cells under mechanical pressure. Immunoprecipitation with an anti-RSPO2 antibody co-precipitated LGR4, whereas the IgG control did not (Fig. 5N), supporting a physical association between RSPO2 and LGR4 in NP cells.

Fig. 5.

Molecular-dynamics simulation and Co-IP support an RSPO2–LGR4 association. A Three-dimensional docking model of RSPO2–LGR4. B Two-dimensional interaction diagram. C RMSD trajectory over 100 ns. D RMSF per residue. E Radius-of-gyration trajectory. F SASA changes. G–H Inter-protein distance and hydrogen bonds. I Per-residue binding energy. J–M Free-energy landscape, RMSD versus Rg, electrostatic surface, and representative snapshots at 0, 50, and 100 ns. N Representative Co-IP blot in NP cells: immunoprecipitation with an anti-RSPO2 antibody co-precipitated LGR4, with input and IgG lanes as controls. Panels C–M were derived from the 100-ns molecular-dynamics trajectory. No densitometric quantification or inferential statistical test was performed for panel N.

3.7. WNT pathway inhibition reverses RSPO2-Mediated degenerative changes

Because no specific small-molecule inhibitor of RSPO2 is currently available, we inhibited the downstream WNT/β-catenin pathway through which RSPO2 exerts its effects. We used IWR-1, a tankyrase inhibitor, which stabilises axin and promotes β-catenin degradation, to test whether the RSPO2-induced phenotype was WNT dependent. Immunofluorescence showed that hydrostatic pressure combined with RSPO2 increased FN1 expression, whereas IWR-1 reversed this change (Fig. 6A). Additional fluorescence staining showed that IWR-1 reduced FN1 and CD44 levels and decreased caspase-3 activation (Fig. 6B). Transmission electron microscopy showed that pressure and RSPO2 induced mitochondrial swelling and autophagosome accumulation, with partial ultrastructural preservation after IWR-1 treatment (Fig. 6C).

Fig. 6.

WNT-pathway inhibition by IWR-1 reverses RSPO2-mediated degenerative changes. A Confocal immunofluorescence showing RSPO2 (green) and FN1 (red) in NP cells. Scale bar: 10 μm. B Quadruple immunofluorescence for DAPI (blue), FN1 (red), CD44 (green), and CASPASE3 (purple). Scale bar: 100 μm. C TEM ultrastructural analysis. Scale bar: 500 nm. Representative microscopy images are shown; no quantitative comparison or inferential statistical test was performed in this figure.

Because RSPO2 enhances canonical WNT signalling through the LGR4/5/6–ZNRF3/RNF43 regulatory module, we next tested whether downstream WNT inhibition could reverse RSPO2-induced FN1–CD44 axis activation. We first compared IWR-1 with DKK1, an LRP5/6 antagonist, and found that both inhibitors reduced FN1, CD44, MMP3, and β-catenin levels, with the greatest suppression observed with combined treatment (Fig. 7A–E). In this context, RSPO2–ZNRF3 docking illustrated structural compatibility with the canonical R-spondin mechanism involving ZNRF3 sequestration, whereas IWR-1–tankyrase docking illustrated structural compatibility with the pharmacological target through which IWR-1 promotes β-catenin degradation (Fig. 7F and G). Immunofluorescence showed that WNT pathway inhibition reduced FN1, RSPO2, and β-catenin co-localisation (Fig. 7H). Within the single-reagent knockdown experiments, siRSPO2 reduced RSPO2, β-catenin, FN1, CD44, and MMP3 levels and restored COL2A1 expression relative to pressure + siNC, whereas siFN1 mainly reduced FN1, CD44, and MMP3, with smaller changes in RSPO2 and β-catenin (Fig. 7I and J). Together with the inhibitor experiments, these patterns are consistent with a model in which RSPO2/WNT signalling precedes FN1–CD44/MMP3 activation; however, they do not by themselves establish a definitive linear hierarchy.

Fig. 7.

WNT-pathway inhibition and siRNA-mediated knockdown attenuate RSPO2-mediated FN1–CD44 axis activation. A Western blot of FN1, CD44, MMP3, and β-catenin showing the effects of WNT inhibitors (Ctrl, DKK1, IWR-1 + DKK1, and IWR-1 groups). B-E Protein quantification of panel A. F RSPO2–ZNRF3 docking model. G IWR-1–tankyrase docking model. H Representative immunofluorescence for FN1, RSPO2, and β-catenin after WNT-pathway inhibition. Scale bar: 50 μm. I Western blot analysis of RSPO2, β-catenin, FN1, CD44, MMP3, and COL2A1 in Ctrl, 0.2 MPa + siNC, 0.2 MPa + siRSPO2, and 0.2 MPa + siFN1 groups. J Descriptive bubble-plot summary of the relative protein expression shown in panel I. For panels B-E, data are mean ± SD; n = 3 independent experiments per group. Comparisons used one-way ANOVA with Tukey's post hoc test; ns, P ≥ 0.05; *P < 0.05, **P < 0.01, and ****P < 0.0001. Panels I-J were based on n = 3 independent cell experiments using one siRNA reagent per target; panel J summarises the relative protein-expression pattern in panel I, and no additional inferential test was applied.

3.8. In vivo perturbation supports RSPO2-Mediated disc degeneration through the WNT/β-catenin/FN1–CD44 axis

To further evaluate the proposed pathway in vivo, we performed intradiscal injections of RSPO2 alone or in combination with IWR-1 in the pressure model. Western blotting showed that RSPO2 treatment exacerbated the pressure-induced upregulation of CD44, FN1, β-catenin, and MMP3 and reduced ACAN expression, whereas IWR-1 partially reversed these changes (Fig. 8A–F). Immunofluorescence showed more widespread co-localisation and upregulation of FN1, CD44, and MMP3 in the operation + RSPO2 group, while IWR-1 weakened this response (Fig. 8G–J).

MRI assessment showed preserved disc signal in controls, progressive signal loss after operation, further deterioration after RSPO2 injection, and partial restoration after combined RSPO2 + IWR-1 treatment (Fig. 9A). Pfirrmann grading and disc height index quantification confirmed more severe degeneration after operation and partial structural improvement with IWR-1 co-treatment (Fig. 9B and C). HE and Safranin O staining further confirmed disrupted disc architecture after operation, worsened pathology after RSPO2 treatment, and better tissue preservation after RSPO2 + IWR-1 treatment (Fig. 9D and E). We also performed HE and Safranin O/Fast Green staining at control, day 1, day 3, and day 28 to validate the degeneration trajectory of the independent needle-puncture model used for spatial transcriptomics. These sections showed a clear time-dependent progression from preserved NP architecture at baseline to focal puncture-related matrix disruption on day 1, progressive NP shrinkage and fibrosis on day 3, and severe proteoglycan loss with fibrotic replacement by day 28. Animal-level histological scores showed a stepwise increase in degeneration severity across the needle-puncture time course (n = 6 animals per time point; Fig. 9F and G). The displayed comparisons were evaluated by one-way ANOVA followed by Tukey's multiple-comparisons test.

Fig. 9.

Imaging and histological assessment support the protective effects of WNT-pathway inhibition in vivo and illustrate the needle-puncture degeneration time course. A Representative MRI and X-ray images from Ctrl, Operation, Operation + RSPO2, and Operation + RSPO2 + IWR-1 groups in the rat pressure model. Scale bar: 5 mm. B Pfirrmann-grade quantification. C Disc-height index. D HE and Safranin O staining of pressure-model discs. Scale bar: 1 mm. E Histological degeneration scores for the pressure-model groups. F HE and Safranin O/Fast Green staining of rat discs from the needle-puncture model at Ctrl, Day 1, Day 3, and Day 28, corresponding to the spatial-transcriptomic time points. Scale bar: 1 mm. G Histological degeneration scores across the needle-puncture time course. For panels B–C, E, and G, data are mean ± SD; each dot represents one animal (n = 6 animals per group or time point). Comparisons used one-way ANOVA with Tukey's post hoc test; *P < 0.05, **P < 0.01, and ***P < 0.001.

3.9. Machine learning shows internal spot-level discrimination; GWAS plots are descriptive

To further evaluate the internal discriminative value of RSPO2-associated features, we applied multiple machine-learning methods to the spatial-transcriptomic data. The custom label reproduced the prespecified positivity rates across IVDD stages (Fig. 10A and B), and spot-level comparisons identified differentially expressed genes between RSPO2-positive and RSPO2-negative spatial units (Fig. 10C–F). The top RSPO2-associated signature genes distinguished these internally defined states (Fig. 10G and H), and correlation analyses identified positive spot-level associations between RSPO2, WNT pathway activity, and FN1–CD44 expression (Fig. 10I and J). Panel K shows an exploratory spot-level mediation model for the Rspo2_smooth → Wnt_target_smooth → FN1CD44_smooth path; the displayed bootstrap interval is not a biological-sample-level confidence interval and does not support causal inference. In a single stratified 80:20 spot-level train–test split, the evaluated algorithms achieved AUC values greater than 0.92 (Fig. 10L and M). RSPO2-associated features ranked among the leading predictive variables (Fig. 10N and O), and adjusted spot-level regression identified associations between RSPO2-positive status and WNT/FN1–CD44, catabolic, NP-identity, and ECM-fibrosis modules (Fig. 10P). These analyses demonstrate internal spot-level discrimination and require validation across independent animals and external cohorts.

Fig. 10.

Machine learning identifies RSPO2-associated features with internal spot-level discriminative value. A UMAP visualisation of custom RSPO2-positive and RSPO2-negative spatial units. B RSPO2 positivity rate across Control, Day 1, Day 3, and Day 28. C–F Volcano plots comparing RSPO2-positive and RSPO2-negative spatial units at each time point. G Heatmap of core RSPO2-positive signature genes upregulated at least three time points. H Heatmap of the top 40 RSPO2-positive signature genes across time points. I Correlation between smoothed RSPO2 expression and WNT-pathway activity. J Correlation between WNT-pathway activity and FN1–CD44-axis expression. K Exploratory spot-level mediation graphic. L–M ROC and precision–recall curves for models predicting the custom RSPO2-positive status. N–O Permutation-importance ranking and local model-agnostic explanation. P Forest plot showing associations between RSPO2-positive status and fibrosis/catabolic modules after adjustment for time point and quality-control covariates. Within each time point, custom RSPO2-positive labels were assigned by ranking k-nearest-neighbour-smoothed Rspo2 expression to prespecified positivity rates (Control, 1.65%; Day 1, 3.42%; Day 3, 5.63%; Day 28, 9.63%); panel B therefore displays the label definition rather than an independently estimated prevalence. Model features comprised signature-gene expression, pathway/module scores, time-point indicators, and quality-control covariates. In panels C–F, Welch's two-sample t-tests were applied to spot-level log1p-CPM values and P values were adjusted using the Benjamini–Hochberg method; FDR < 0.05 was considered significant. Solid lines in panels I–J denote spot-level linear fits. Panel K shows an exploratory spot-level mediation model fitted by ordinary least squares with HC3-robust standard errors and adjusted for time point, log1p library size, and mitochondrial UMI fraction. The indirect effect for the Rspo2_smooth → Wnt_target_smooth → FN1CD44_smooth path and its displayed 95% percentile interval were obtained from 500 spot-level bootstrap resamples (random seed 1). Because spatial spots are nested within tissue sections, this interval is not a biological-sample-level confidence interval and the model does not support causal inference. Panels L–M report soft-voting and holdout-stacking performance in a single stratified 80:20 spot-level train–test split (random seed 42), not an independent cohort. Panel N shows LightGBM permutation importance from three repeats, and panel P reports OLS coefficients with HC3-robust 95% confidence intervals adjusted for time point and quality-control covariates. Spatial-transcriptomic data were obtained from n = 3 animals per time point (one section per animal); spots are the analytic units and do not constitute independent biological replicates.

Supplementary Fig. 6 provides descriptive visualisations of publicly available human IVDD GWAS summary statistics, including Manhattan, QQ, volcano, effect-size, uncertainty, and chromosome-level displays. These source-level plots are presented for genetic context only; no cell-state-specific integration or cross-species causal inference is claimed.

4. Discussion

Rather than identifying RSPO2 simply as a marker that rises in degenerated tissue, our data support an injury-responsive signalling model in which RSPO2 increases the sensitivity of NP cells to canonical WNT cues and favours persistence of an FN1–CD44-associated catabolic state. The key biological change may therefore be the propagation of an RSPO2-associated programme across the injured NP microenvironment, rather than expansion of a stable, lineage-defined RSPO2-positive population.

Mechanistically, this interpretation is compatible with established R-spondin biology. R-spondins bind LGR4/5/6 and neutralise the ZNRF3/RNF43 brake on Frizzled abundance, thereby amplifying rather than independently initiating WNT signalling [[18], [19], [20],30]. RSPO2 could thus convert an injury-induced WNT signal into a more sustained response. Previous IVDD studies have implicated mechanical loading and several endogenous regulators in WNT/β-catenin activation [13,16,[31], [32], [33], [34], [35]], whereas fibronectin deposition or fragmentation promotes matrix-catabolic responses and CD44 participates in WNT-related matrix communication [21,[23], [24], [25], [26], [27], [28]]. Our perturbation data connect these observations by supporting a model in which RSPO2-associated WNT amplification precedes FN1–CD44/MMP3 activation, although the proposed sequence is not yet equivalent to a fully resolved linear epistatic cascade.

This framework also helps reconcile an apparent difference between the transcriptomic datasets. The Rspo2-enriched homeostatic Cluster 1 contracted in the pressure-model scRNA-seq data, whereas exploratory spatial panels illustrated RSPO2-associated features across needle-puncture stages. Because the analyses resolve different units—dissociated cell states versus multicellular spatial units—and use different injury models, these findings should not be interpreted as simple expansion of an RSPO2-positive lineage. Instead, they are consistent with loss of a homeostatic NP identity accompanied by broader activation of an RSPO2-associated tissue programme. This interpretation extends prior descriptions of heterogeneous NP and progenitor-like populations [6,8,9,36,37], while the lineage relationship between RSPO2-positive cells and PROCR- or CD24-positive populations remains unresolved.

Within the mechanical-injury context, RSPO2 may operate alongside, rather than replace, the PIEZO1–NF-κB–periostin loop, senescence-associated signalling, autophagy dysregulation, and vesicle-mediated regulation described previously [[38], [39], [40], [41], [42]]. Gain-of-function, WNT-inhibitor, and siRNA responses converged on the same pathway model in our rat NP-cell system, and Co-IP was consistent with RSPO2 participating in an LGR4-containing complex. However, the single-reagent knockdown design supports, but does not definitively establish, the proposed RSPO2→WNT→FN1→CD44 ordering. WNT-pathway inhibition and future targeted interference with RSPO2–LGR4 or FN1–CD44 interactions should accordingly be viewed as candidate strategies for further preclinical testing [32,34].

Interpretation of the early time course requires a clear distinction between experimental injury and human disease. Needle puncture imposes an abrupt, synchronised annular and NP injury with a known time zero and generates an acute microenvironment of matrix disruption, sterile inflammation, and wound-repair signalling, which makes early changes at days 1–3 experimentally observable. Idiopathic human IVDD develops more gradually and asynchronously under ageing, repetitive mechanical loading, impaired nutrient exchange, low-grade inflammation, and cumulative matrix damage. RSPO2-associated states in patients may therefore be triggered heterogeneously and over a substantially longer interval than in the rat model. Because clinically obtained early-stage (Pfirrmann I–II) human tissue was limited and disease onset cannot be dated, the rat spatial series defines a mechanistic post-injury sequence but cannot assign an equivalent RSPO2 activation time to human idiopathic IVDD.

Additional limitations concern the resolution and generalisability of the evidence. A second sequence-independent siRNA and an siRNA-resistant rescue construct were not tested, so residual off-target effects cannot be excluded. Spatial transcriptomics used one section from each animal per time point, and the interval from day 3 to day 28 was not profiled; regional heterogeneity and intermediate remodelling states may therefore be under-resolved. Cell-communication analysis remains computational inference, the machine-learning models require validation across independent animals and external cohorts, and the descriptive GWAS source plots do not provide cell-state-specific or direct human cellular evidence. Site-directed mutagenesis of the predicted RSPO2–LGR4 interface would also be needed to test the specificity of that interaction. Together with anatomical and biomechanical differences between rat caudal and human lumbar discs, these limitations mean that our in vivo results provide mechanistic support, not direct evidence of activation timing or clinical efficacy in human IVDD.

5. Conclusions

In conclusion, this study identifies RSPO2 as a candidate regulator, rather than only a marker, of NP degeneration. Integrated cellular, spatial, and perturbation evidence supports a model in which RSPO2-associated WNT amplification precedes FN1–CD44/MMP3 activation. Although this ordering requires confirmation with sequence-independent knockdown and rescue experiments, it provides a testable framework for investigating disease progression and WNT-directed intervention.

Ethics declaration

Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed.

This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place. This study was approved by the Ethics Committee of the First Affiliated Hospital of Anhui Medical University. (Approval No. LLSC-20242216)

This study was conducted in accordance with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines. This study was approved by the Experimental Animal Ethics Association of Anhui Medical University. (Approval No. LLSC-20242216)

Data availability

Single-cell and spatial (Visium) transcriptomic data generated in this study are available from the corresponding author on reasonable request. The human disc-degeneration GWAS summary statistics shown descriptively in Supplementary Fig. 6 are publicly available from the NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/) under accessions GCST90077921 and GCST90077922 (UK Biobank exome-sequencing study [43]). Any additional data supporting the findings of this study are available from the corresponding author on reasonable request.

Author contributions

Conception and design, Q.L., G.L., and K.B.; financial support, R.Z., F.L., and C.S.; provision of study materials, Q.L., L.K., P.J., and C.Z.; collection and assembly of data and samples, all authors; data analysis and interpretation, Q.L., G.L., K.B., and C.Z.; manuscript writing, Q.L., G.L., K.B., and C.S.; project supervision, R.Z., F.L., and C.S. All authors read and approved the final manuscript.

Ethics approval and consent to participate

Animal experiments were approved by the Experimental Animal Ethics Association, Anhui Medical University (LLSC-20242216). Human lumbar disc tissues from 24 patients were used with written informed consent under approval of the Ethics Committee of the First Affiliated Hospital of Anhui Medical University (LLSC-20242216). All methods followed relevant guidelines and regulations.

Declaration of AI and AI-assisted technologies in the writing process

Statement of the Use of AI and AI-assisted Technologies in the Writing Process. During the preparation of this work, the authors used Claude (Anthropic) and OpenAI Codex for language polishing, grammar checking, document consistency review, and improving readability. The authors reviewed and edited the content as needed and take full responsibility for the content of the published article. AI was not used for data analysis, interpretation of results, or generation of scientific conclusions.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No: 82272551 and 81772408), and the Graduate Research and Innovation Program of Anhui Medical University (No. YJS20240033).

Declaration of competing interests

The authors declare no competing interests.

Abbreviations

IVDD

Intervertebral disc degeneration

LBP

Low back pain

NP

Nucleus pulposus

ECM

Extracellular matrix

scRNA-seq

Single-cell RNA sequencing

RSPO2

R-spondin 2

LGR4

Leucine-rich repeat-containing G-protein-coupled receptor 4

FN1

Fibronectin 1

CD44

Cluster of differentiation 44

MMP

Matrix metalloproteinase

ACAN

Aggrecan

DHI

Disc height index

WNT/β-catenin

WNT/beta-catenin signalling pathway

IWR-1

Inhibitor of WNT response 1

DKK1

Dickkopf 1

PPI

Protein–protein interaction

TEM

Transmission electron microscopy

MRI

Magnetic resonance imaging

UMAP

Uniform manifold approximation and projection

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

AUC

Area under the curve

RMSD

Root-mean-square deviation

RMSF

Root-mean-square fluctuation

SASA

Solvent-accessible surface area

LIME

Local interpretable model-agnostic explanations.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Supplementary Fig. 1: scRNA-seq identifies NP-cell populations in IVDD — A UMAP of 11 NP-cell clusters; B cluster proportions (Sham versus Operation); C marker-gene heatmap; D feature plots of Cluster 1 markers (Rspo2, Kera, Crabp2, Fndc1, Clec2dl1, Col8a2, Col14a1, Gpr15lg, Igf1, and Rrad); E violin plots of marker expression; F CellChat information flow (Sham versus Operation). The NP-focused analysis included 3216 cells derived from six sequenced rat samples (n = 3 animals per group); cluster-specific cell counts are shown in panel A. All panels are descriptive, and no inferential statistical test is displayed.

Supplementary Fig. 2: Communication analysis around RSPO2-positive NP cells — A–B ECM and WNT module scores on UMAP; C STRING PPI network (RSPO2–WNT2–FZD9–FN1–CD44); D differential information flow; E correlation analyses (Rspo2–WNT, Fn1–Cd44, and related comparisons); F FN1-signalling chord diagrams (Sham versus Operation); G pathway changes around the RSPO2-defined Cluster 1; H outgoing-signalling pattern heatmap; I cell–cell interaction network in Operation. The analysis used the same 3216-cell NP-focused subset derived from six sequenced rat samples (n = 3 animals per group). In panel E, each point represents a cell; rank-correlation summaries and the displayed P values are exploratory panel-level outputs, and panel-specific filtering and effective sample sizes are not treated as animal-level replicates. The remaining panels are descriptive.

Supplementary Fig. 3: NP-cell dynamics, gene scoring, and pseudotime — A cluster-proportion changes; B–F split violin plots of RSPO2 pathway, FN1–CD44, WNT, ECM-catabolic, and apoptosis scores; G–H Monocle pseudotime with WNT-activity overlay; I gene-module heatmap; J differential signalling-role scatter for Cluster 1. The analysis used the same 3216-cell NP-focused subset derived from six sequenced rat samples (n = 3 animals per group). All panels are descriptive cell-level summaries; no inferential statistical test is displayed.

Supplementary Fig. 4: Descriptive spatial characterisation of RSPO2-associated NP features — A–B RSPO2-positive spatial-unit quantification; C–D expression heatmap and pathway scores; E RSPO2–WNT–FN1–CD44 interaction network; F disease-progression model; G integrated disease-progression analysis; H spatial-distribution maps of RSPO2 expression. For panels A–D, RSPO2-positive labels were assigned to the within-time-point proportions displayed (Control, 1.6%; Day 1, 2.4%; Day 3, 3.3%; Day 28, 9.6%); this descriptive label is distinct from the k-nearest-neighbour-smoothed custom label used in Fig. 10. The phrase ‘progressive increase’ in the figure title refers only to these prespecified display proportions and not to an estimated biological increase. Spatial-transcriptomic data were obtained from n = 3 animals per time point (one section per animal). Panels A–D are interpreted as descriptive cell-/spot-level summaries, and panels E–H are descriptive network, trajectory, and spatial-mapping summaries. Spatial units do not constitute independent biological replicates; displayed dots, error bars, and star-like graphical annotations are descriptive plot elements rather than significance symbols and are not used for animal-level inference.

Supplementary Fig. 5: Spatial-domain and FN1–CD44 interaction analysis — A Domain mapping into D0 (degenerative core, enriched for Mmp3/Mmp13 and RSPO2 activity), D1 (transitional zone), and D2 (relatively preserved tissue, enriched for Acan and Col2a1). D0* denotes the degenerative domain of interest; the asterisk is part of the domain label, not a significance symbol. The RSPO2-positive overlay in panel A shows the displayed top 9.9% of Day 28 spots and is used for spatial visualisation. B–F Domain-differential expression and RSPO2-positive enrichment. G FN1–CD44 spatial scoring. H–K Neighbourhood-enrichment and distance-decay analyses. Spatial-transcriptomic data were obtained from n = 3 animals per time point (one section per animal). Panels A–I show the Day 28 domain analysis, whereas panels J–K summarize Control, Day 1, Day 3, and Day 28. For panels B and F, domain-differential results were based on spot-level gene-detection rates comparing D0 spots (n = 410) with non-D0 spots (n = 1503); Benjamini–Hochberg-adjusted values are displayed as FDR. For the neighbourhood-enrichment panels, the odds ratios shown in the original plots are Fisher-exact neighbour-pair summaries with 95% confidence intervals; displayed annotations are exploratory and are not interpreted as animal-level significance. In panels I and K, points show the mean score per distance bin and error bars show mean ± SEM across spots within each bin. Spot-level tests and neighbourhood summaries are exploratory and do not constitute independent biological replicates at the animal level.

Supplementary Fig. 6: Descriptive visualisation of publicly available human IVDD GWAS summary statistics - A Manhattan plot; B QQ plot; C volcano plot; D effect size versus standard error; E distribution of effect sizes; F distribution of -log10(P); G GWAS hits per chromosome; H top 10 absolute effect sizes; I effect size by sign. Each point represents a variant from the publicly available GWAS summary statistics (GCST90077921 and GCST90077922). Displayed P values were inherited from the source summary statistics; no cell-state-specific integration, additional inferential test, or significance symbols are claimed in this figure.