Tumor-derived GDF15 induces CCN3⁺ Schwann cells to promote cancer pain in pancreatic cancer

Abstract

Tumor–neural crosstalk contributes to the remodeling of the tumor microenvironment, yet how tumors engage peripheral glial networks, particularly Schwann cells (SCs), to drive chronic pain remains unclear. Here, we identify a specialized cellular communication network factor 3-positive (CCN3⁺) SC subpopulation that promotes tumor innervation and contributes to pain in pancreatic ductal adenocarcinoma (PDAC). We demonstrate that cancer cell–derived growth differentiation factor 15 (GDF15) drives expansion of CCN3⁺ SCs and induces glycolytic reprogramming via the GDNF family receptor alpha-like (GFRAL) receptor. Mechanistically, GFRAL activation triggers the protein kinase B (AKT)–runt-related transcription factor 2 (RUNX2) cascade, upregulating the glycolytic enzyme muscle-type phosphofructokinase (PFKM) in CCN3⁺ SCs, which enhances tumor innervation and pain sensitization. Targeted inhibition of GDF15–GFRAL signaling in CCN3⁺ SCs significantly alleviates PDAC-associated pain. Together, these findings reveal a perineural–metabolic axis driven by glycolytic reprogramming in SCs and highlight a promising therapeutic strategy for PDAC-associated pain.

Chronic pain in pancreatic ductal adenocarcinoma (PDAC) remains poorly understood. Here, the authors show that cancer-derived GDF15 drives the expansion of a specialized CCN3⁺ Schwann cell subpopulation, which in turn promotes sensory nerve sensitization and contributes to pain in PDAC.

Introduction

In pancreatic ductal adenocarcinoma (PDAC), which is a leading cause of cancer-related mortality^1–4, chronic pain is a clinical symptom affecting approximately 80% of patients⁵. Chronic pain significantly complicates PDAC treatment⁶, and its severity is directly correlated with decreased survival and poor prognosis³. However, the mechanisms underlying PDAC-induced chronic pain remain unclear, and no effective treatment is currently available.

Neuroinvasion, defined as tumor cell infiltration of the myelin sheath, which is primarily composed of Schwann cells (SCs), occurs in up to 98% of pancreatic cancer cases^7–9. Cancer cell invasion induces profound neural remodeling that triggers inflammatory cytokine cascades driving neuropathic pain^10,11. Whether neuroinvasion-associated structural nerve alterations directly contribute to cancer-associated pain¹² in PDAC remains unknown.

SCs support axons and form myelin sheath in the peripheral nervous system^13,14. Beyond their structural role, SCs display pronounced plasticity and immunomodulatory functions that critically shape cancer pathogenesis and dissemination through perineural invasion¹⁵. SCs are also involved in cancer-associated pain by secreting various neurotrophic factors, such as NGF, and regulating the expression and function of receptors and ion channels, including P2X, P2Y, TLRs, and TRP cation channels^16–23. However, specialized SCs in PDAC and the molecular mechanisms underlying chronic pain remain elusive.

In this study, we aimed to identify chronic pain-regulating SC subpopulations in PDAC and to elucidate the mechanisms underlying their generation.

Results

Increased CCN3⁺ SCs in pancreatic cancer tissues contribute to chronic pain in PDAC

Following in situ injection of syngeneic K8484 PDAC cells into the pancreas, mice developed robust nociceptive phenotypes, evidenced by markedly increased cumulative withdrawal responses, response frequencies to Von-Frey filaments, and hunching scores, along with significantly decreased paw withdrawal latency to heat and total distance traveled, indicating successful establishment of a chronic pain model in PDAC mice (Fig. 1A–G). Orthotopic injection of an alternative pancreatic cancer cell line (KPC) also recapitulated prominent chronic pain in PDAC mice (Supplementary Fig. 1A–E). A similar pattern was observed in female mice, which exhibited chronic pain comparable to that in males (Supplementary Fig. 1F-I).

Fig. 1. Increased CCN3⁺ SCs contribute to PDAC-induced chronic pain.

A Schematic diagram of the chronic pain model induced by pancreatic cancer in mice (Created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889). The syngeneic K8484 PDAC cancer cells were injected into the pancreas in situ. B Specimen diagrams of the sham group pancreas and the pancreatic tumor of mice. Scale bars: 1 cm. C–G The painful behavior was evaluated by detecting spontaneous pain (hunching score at week 5) in mice, movement restriction (total distance traveled at week 5), mechanical allodynia (cumulative withdrawal response and response frequency measured weekly) and thermal hyperalgesia (the paw withdrawal latency to heat measured weekly). C Kruskal-Wallis test with Dunn’s post hoc test. Cliff’s Delta (δ) = −1.00. ***P = 0.0006 vs Sham group. n = 7 mice per group. D unpaired two-tailed t-test. Cohen’s d = 4.90, 95% CI [4.500 to 6.314]. ****P < 0.0001 vs Sham group. n = 7 mice per group. E two-way repeated measures ANOVA. η² (Sham vs PDAC) = 0.537. 95% CI [104.4 to 120.7]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001 vs. sham group. n = 7 mice per group. F two-way repeated measures ANOVA. η² (Sham vs PDAC) = 0.486. 95% CI [28.34 to 37.95]. Šidák’s post hoc test for multiple comparisons: ***P = 0.004, ****P < 0.0001 vs sham group. n = 7 mice per group. G two-way repeated measures ANOVA. η² (Sham vs PDAC) = 0.467. 95% CI [3.516 to 4.430]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001 vs sham group. n = 7 mice per group. H Representative immunofluorescence staining and quantification of PGP9.5 or S100β in sham group pancreas and pancreatic tumor of PDAC mice. Scale bars: 50 μm. Representative three-dimensional (3D) immunofluorescence staining of PGP9.5 or S100β in pancreatic tumor of PDAC mice. PGP9.5 was analyzed by a two-tailed unpaired t-test. Cohen’s d = 2.074. 95% CI [0.922 to 4.410]. **P = 0.0072 vs sham group. n = 6 (Sham), 5 (PDAC). S100β was analyzed by a two-tailed unpaired t-test. Cohen’s d = 2.012. 95% CI [0.901 to 2.886]. **P = 0.0019 vs sham group. n = 6 mice (Sham), 5 mice (PDAC). I Representative immunofluorescence staining and quantification of PGP9.5 or S100β in the adjacent normal tissue and pancreatic tumor in PDAC patients. Scale bars: 50 μm. Representative three-dimensional (3D) immunofluorescence staining of PGP9.5 or S100β in pancreatic tumors of PDAC patients. PGP9.5 was analyzed by a two-tailed unpaired t-test. Cohen’s d = 1.153. 95% CI [0.195 to 3.888]. *P = 0.0332 vs. ADJ group. n = 5 patients (ADJ), 8 patients (PDAC). S100β was analyzed by a two-tailed unpaired t-test. Cohen’s d = 2.112. 95% CI [1.629 to 5.389]. **P = 0.0017 vs. ADJ group. n = 8 patients (ADJ), 5 patients (PDAC). J, K Statistical analysis of the correlations between the number of nerve fibers and the number of SCs in PDAC mice and PDAC patients. PDAC mice were analyzed by Pearson correlation (two-tailed). r = 0.894. R² = 0.800. 95% CI [0.658 to 0.970]. ****P < 0.0001. n = 12 mice. PDAC patients were analyzed by Pearson correlation (two-tailed). r = 0.606. R² = 0.368. 95% CI [0.050 to 0.876]. *P = 0.037. n = 12 patients. L AP frequency of dorsal root ganglia in sham group mice and pancreatic tumor-bearing PDAC mice. two-way repeated measures ANOVA. η² (Sham vs PDAC) = 0.463. 95% CI [0.928 to 2.563]. ***P = 0.0002 vs. sham group. n = 5 mice (13 independent records) per group. M AP Rehobase of dorsal root ganglia in sham group mice and pancreatic tumor-bearing PDAC mice. two-tailed unpaired t test. Cohen’s d = 1.094. 95% CI [−46.12 to −6.187]. *P = 0.0124 vs. sham group. n = 5 mice (13 records) per group. N Statistical analysis of the correlations between the protein levels of S100β and mechanical allodynia in PDAC mice. Pearson correlation (two-tailed). r = 0.915. R² = 0.838. 95% CI [0.593 to 0.985]. **P = 0.0014. n = 8 mice. O Identifications of 3 subclusters of SCs from single-cell sequencing dataset of patients (GSE212966) from the GEO public database (adjacent normal tissues and tumor tissues from 3 PDAC patients). P The proportion of 3 Schwann cell subclusters including CCN3⁺ SCs, FOSB⁺ SCs and ANGPTL7⁺ SCs in adjacent normal tissues and tumor tissues from single-cell sequencing dataset of PDAC patients (GSE212966). Q Representative double immunofluorescence staining showed the colocalization of CCN3 and S100β in the pancreatic tumor in PDAC patients (n = 5 patients). Scale bars: 50 μm. R Representative double immunofluorescence staining and quantification showed the colocalization of CCN3 and S100β in sham group pancreas and pancreatic tumor of PDAC mice. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 9.954. 95% CI [1.850 to 2.257]. ****P < 0.0001 vs. sham group. n = 7 mice (Sham), 13 mice (PDAC). S–U Movement restriction, spontaneous pain, mechanical allodynia and thermal hyperalgesia in PDAC mice after intratumoral co-injection of AAV-CNP-Cre and AAV-Ccn3-DIO-EGFP/DTA. S two-tailed unpaired t test. Cohen’s d = 6.426. 95% CI [3.638 to 6.112]. ****P < 0.0001 vs. AAV-Ccn3-DIO-EGFP group. n = 5 mice per group. T Mann-Whitney U test (two-sided) with Dunn’s post hoc test. Cliff’s delta = −1.000. 95% CI [−3.000 to 3.000]. **P = 0.0079 vs. AAV-Ccn3-DIO-EGFP group. n = 5 mice per group. U two-way repeated measures ANOVA. Von-Frey cumulative response: η² (AAV-Ccn3-DIO-EGFP vs AAV-Ccn3-DIO-DTA) = 0.065. 95% CI [27.94 to 50.46]. Šidák’s post hoc test for multiple comparisons: ***P = 0.0003, ****P < 0.0001 vs. AAV-Ccn3-DIO-EGFP group. n = 5 mice per group. Von-Frey response frequency: η² = 0.172. 95% CI [13.94–27.66]. Šidák’s post hoc test for multiple comparisons: ***P = 0.0001, ****P < 0.0001 vs. AAV-Ccn3-DIO-EGFP group. n = 5 mice per group. thermal hyperalgesia: η² = 0.083. 95% CI [−2.068 to −1.129]. Šidák’s post hoc test for multiple comparisons: ***P = 0.0002, ****P < 0.0001 vs. AAV-Ccn3-DIO-EGFP group. n = 5 mice per group. Data are shown as the mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

As PDAC shows significantly increased numbers of nerve fibers in the cancer tissue and SCs provide local structural and nutritional support for promoting nerve fiber growth, we explored the distribution of nerves and SCs in patients with PDAC and PDAC model mice. Compared with the sham group, the number of nerve fibers (PGP9.5⁺ signal) and SCs (S100β⁺ cells) in the pancreatic tumors of PDAC model mice was significantly increased, and the SCs were co-located with nerve fibers (Fig. 1H). Importantly, the number of nerve fibers and co-located SCs in patients with PDAC was significantly increased in the pancreatic tumors compared with those in the adjacent tissues (Fig. 1I). Statistical analysis revealed a positive correlation between the numbers of nerve fibers and SCs in PDAC model mice and patients with PDAC (Fig. 1J, K).

Although nerve density and pain are known to be positively correlated, pain induction often requires sensory neuron sensitization, rather than a mere increase in nerve quantity. Patch-clamp electrophysiological recordings of primary dorsal root ganglion (DRG) neurons from PDAC mice revealed significantly increased action potential frequency (AP frequency), along with a significantly decreased action potential rheobase (AP Rheobase), indicating neuronal sensitization (Fig. 1L, M). Characterization of the subtypes of proliferating nerve fibers indicated a significant increase in TRPV1- or CGRP-expressing fibers, which mediate pancreatic cancer-associated pain^24,25, within the pancreatic tumors of PDAC mice (Supplementary Fig. 1 J, K). Notably, TRPV1⁺/CGRP⁺ nerve fibers were significantly increased in tumor tissues than in adjacent tissues in patients with PDAC (Supplementary Fig. 1 L, M). In pancreatic tumors, SCs were co-labeled with the specific markers S100β and SOX, or S100β and GAP43 (Supplementary Fig. 1 N, O). Western blot assays confirmed that S100β expression was elevated in the pancreatic cancer tissue of PDAC mice (Supplementary Fig. 1P). Moreover, S100β expression levels were positively correlated with chronic pain severity in PDAC mice (Fig. 1N).

To elucidate the mechanism of SC involvement in PDAC-induced chronic pain, we analyzed a single-cell sequencing dataset of patients (GSE212966) from the GEO public database. After applying harmony integration for batch effect correction, TSNE visualization identified 11 clusters, including SCs, in tumor tissues and adjacent normal tissues from three PDAC patients (Supplementary Fig. 2A). SCs were identified by specific markers S100B, SOX10, and GAP43, and further divided into three subclusters: CCN3⁺ SCs, FOSB⁺ SCs, and ANGPTL7⁺ SCs (Fig. 1O, Supplementary Fig. 2B, F). Among these, the proportion of CCN3⁺ SCs was significantly increased in the pancreatic cancer tissues of patients with PDAC (Fig. 1P). Moreover, the Schwann cell subpopulation with high CCN3 expression was validated in an independent public dataset (GSE278688) (Supplementary Fig. 2C–E). Cell trajectory analysis revealed that CCN3⁺ SCs were enriched at the terminal stage of the trajectory (Supplementary Fig. 2G), suggesting that CCN3⁺ SCs represent a specialized Schwann cell subpopulation that expands within the tumor environment under pathophysiological conditions. Among CCN3⁺ SC signature genes, myelinating Schwann cell-specific genes such as Mbp, Mag, and Egr2 showed significantly reduced expression, whereas the expression of non-myelinating Schwann cell-specific genes such as Plp1, Gfap, Ncam1, and Ngfr (encoding p75NTR) was elevated (Supplementary Fig. 2H). Immunofluorescence results showed low MBP expression and high p75NTR expression in CCN3⁺ SCs (Supplementary Fig. 3A, B), further supporting the notion that CCN3⁺ SCs predominantly exhibit a non-myelinating state. However, repair‑associated Schwann cell genes such as Gap43 and Ngfr (also non‑myelinating markers), as well as pro‑pain, pro‑inflammatory, and pro‑fibrotic genes including Ngf, Bdnf, Spp1, and Fn1, were all highly expressed (Supplementary Fig. 2H). These genes are typically expressed at low levels in non-myelinating SCs, suggesting that CCN3⁺ SCs represent a specialized cellular subpopulation in the PDAC tumor microenvironment.

To elucidate the role of CCN3⁺ SCs in PDAC-induced chronic pain, we first verified CCN3 expression in SCs from patients with PDAC and PDAC model mice. Double immunofluorescence labeling revealed a substantial presence of CCN3⁺ SCs (CCN3⁺S100β⁺ cells) in the tumor tissues of patients with PDAC (Fig. 1Q). Moreover, the number of CCN3/S100β double-positive cells was significantly higher in PDAC mice compared with those in the sham group (Fig. 1R). Similarly, the number of nerve fibers and CCN3⁺S100β⁺ cells in the tumors of KPC mice was significantly increased (Supplementary Fig. 3C, D). We observed that during the first week following PDAC induction, a stage where mice showed no signs of pain, CCN3⁺ SCs were virtually absent within the tumors (Supplementary Fig. 3E–G). By the third and fifth weeks, the number of CCN3⁺ SCs progressively increased, concomitant with a gradual increase in pain severity (Supplementary Fig. 3E–G). Importantly, intratumoral co-injection of AAV-CNP-Cre and AAV-Ccn3-DIO-DTA to specifically ablate CCN3⁺ SCs alleviated mechanical pain (cumulative withdrawal response and response frequency), spontaneous pain (hunching score), thermal hyperalgesia (paw withdrawal latency to heat), and movement restriction (total distance traveled) (Fig. 1S–U). The efficacy of this targeted intervention was further supported by the observation that gabapentin administration yielded a similar and rapid reversal of these pain behaviors (Supplementary Fig. 3H–K). Overall, increasing CCN3⁺ SCs in pancreatic cancer tissues contributes to PDAC-induced chronic pain.

Cancer cells induce CCN3⁺ SC proliferation to promote nerve fiber growth and sensitization in PDAC

To elucidate the mechanism underlying the increase in CCN3⁺ SCs in pancreatic cancer, we reanalyzed the single-cell sequencing dataset (GSE212966) and dissected the specialized gene expression profiles in SCs. Many neurotrophic factors (such as Btc, Igfbp5 and Crlf1) related to cell proliferation and cell migration were significantly upregulated in the CCN3⁺ SCs of patients with PDAC (Supplementary Fig. 2F). Proliferation and migration assays showed that EdU-positive cell numbers and RSC96 cell (Schwann cell line) migration increased significantly following incubation with K8484 (PDAC) cell supernatant (SN) (Supplementary Fig. 4A, B). Importantly, flow cytometry analysis showed an increase in the proportion of CCN3⁺S100β⁺ cells among RSC96 cells following incubation with K8484 cell supernatant (Fig. 2A). Western blot assays and quantitative real-time PCR (qPCR) further demonstrated that CCN3 protein and mRNA expression was significantly elevated in RSC96 cells incubated with the K8484 cell supernatant (Supplementary Fig. 4D, E). Overall, these results suggest significant proliferation and migration of CCN3⁺ SCs in PDAC. Furthermore, by incubating ND7/23 cells derived from the DRG with conditioned medium (CM) collected from the co-culture of the K8484 cell supernatant and RSC96 cells, we examined whether CCN3⁺ SCs promoted nerve fiber growth and pain sensitization. The neurite length of ND7/23 cells significantly increased after incubation with the CM (Supplementary Fig. 4C). Additionally, CM treatment markedly upregulated TRPV1 and CGRP expression in ND7/23 cells (Supplementary Fig. 4F, G). More importantly, CM incubation significantly increased the AP frequency and decreased the AP Rheobase in mouse DRG neurons (Fig. 2B, C). Together, these findings indicate that cancer cells drive CCN3⁺ SC expansion, which in turn promotes nerve fiber growth and neuronal sensitization in PDAC.

Fig. 2. Cancer cells induce CCN3⁺ SC proliferation to promote nerve fiber growth and sensitization in PDAC.

A Flow cytometry analysis showing the proportion of CCN3⁺S100β⁺ cells in RSC96 cells following the incubation of 10% DMEM or K8484 cell supernatant. two-tailed unpaired t-test. Cohen’s d = 4.664. 95% CI [4.100 to 7.646]. ****P < 0.0001. n = 6 independent experiments per group. The gating strategy is shown in Supplementary Fig. 15. B, C AP frequency and AP Rehobase of dorsal root ganglia following incubation with comditional medium from co-culture of K8484 and RSC96 cells. B two-way ANOVA. η² (K8484 + RSC96 CM vs RSC96 + RSC96 CM) = 0.071. 95% CI [1.791 to 2.379]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001 vs. RSC96 + RSC96 CM group. n = 13 independent records (RSC96 + RSC96 CM), 10 independent records (K8484 + RSC96 CM). C two-tailed unpaired t test. η² (RSC96 + RSC96 CM vs K8484 + RSC96 CM) = 0.270. Cohen’s d = 1.214. 95% CI [−52.80 to −7.660]. *P = 0.0111 vs. RSC96 + RSC96 CM group. n = 13 independent records (RSC96 + RSC96 CM), 10 independent records (K8484 + RSC96 CM). D Schematic diagram showed that AAV-CNP-shCcn3 or AAV-CNP-scrambled shRNA was intratumorally injected into a pancreatic tumor (Created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889). E Specimen diagrams and quantification showing the pancreatic tumor of mice following the intratumoral injection of AAV-CNP-shCcn3 or AAV-CNP-scrambled shRNA. Scale bars: 1 cm. two-tailed unpaired t test. Cohen’s d = 3.748. 95% CI [−0.975 to −0.427]. ***P = 0.0004 vs. AAV-CNP-shNC group. n = 5 mice per group. F Representative staining showing the expression of AAV-encoded EGFP and S100β fluorescence in tumor, pancreas, and DRG tissues following intratumoral injection of AAV-CNP-shCcn3 (n = 5 mice). Scale bars: 50 μm. G, H Mechanical allodynia, spontaneous pain, and movement restriction in PDAC mice after intratumoral injection of AAV-CNP-shCcn3 or AAV-CNP-scrambled shRNA. G Von-frey cumulative response: two-way repeated measures ANOVA. η² = 0.144. Šidák’s post hoc test for multiple comparisons: ***P = 0.0008, ****P < 0.0001 vs. AAV-CNP-shNC group. n = 6 mice group. Von-frey response frequency: two-way repeated measures ANOVA. η² = 0.154. Šidák’s post hoc test for multiple comparisons: **P = 0.0013, ***P = 0.0002 vs. AAV-CNP-shNC group. n = 6 mice per group. Spontaneous pain: Mann-Whitney U test (two-sided). Cliff’s delta = −0.837. 95% CI of median difference [−1.000 to 1.000]. *P = 0.0152 vs. shNC group. n = 7 mice per group. H two-tailed unpaired t-test. Cohen’s d = 2.759. 95% CI [1.310 to 4.231]. **P = 0.0024 vs. AAV-CNP-shNC group. n = 5 mice per group. I Representative immunofluorescence staining and the quantification showed that inhibition of CCN3 by using AAV-CNP-shCcn3 prevented the increased co-expression of the PGP9.5 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 5.896. 95% CI [−3.466 to −2.093]. ****P < 0.0001 vs. AAV-CNP-shNC group. n = 5 mice per group. J, K Representative staining and quantification showed that intratumoral injection of AAV-CNP-shCcn3 inhibited the increased colocalization of TRPV1 and PGP9.5 or CGRP and PGP9.5 in the pancreatic tumor of PDAC mice. Scale bars: 50 μm. TRPV1: two-tailed unpaired t test. Cohen’s d = 3.463. 95% CI [−3.796 to −1.473]. ***P = 0.0008 vs. AAV-CNP-shNC group. n = 5 mice per group. CGRP: two-tailed unpaired t test. Cohen’s d = 4.607. 95% CI [−3.202 to −1.605]. ***P = 0.0001 vs. AAV-CNP-shNC group. n = 5 mice per group. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

To investigate how CCN3⁺ SCs promote nerve fiber growth in PDAC, we silenced CCN3 in RSC96 cells using siCcn3 (Supplementary Fig. 4H). CCN3 knockdown significantly reduced SC proliferation and migration, indicated by decreased EdU⁺ cells and impaired motility in response to K8484 cell supernatant (Supplementary Figs. 4I, 5A). CM from the coculture of K8484 cells and siCcn3-treated RSC96 cells markedly suppressed neurite outgrowth in ND7/23 cells (Supplementary Fig. 5B). In parallel, CCN3 knockdown in SCs significantly reduced TRPV1 and CGRP expression in the cocultured ND7/23 cells (Supplementary Fig. 5C, D). To establish CCN3 specificity, we performed rescue experiments and found that re-expression of wild-type (WT) Ccn3 restored SC proliferation and pro-neurite activity following siCcn3 knockdown (Supplementary Fig. 5E, F). In contrast, a signaling-deficient mutant (Mut) Ccn3 failed to rescue these phenotypes (Supplementary Fig. 5E, F). Overall, these results demonstrate that CCN3 is an essential element required for mediating the tumor-induced SC proliferation and pro-neurite outgrowth.

We further knocked down CCN3 expression in SCs in vivo. Following intratumoral injection of AAV‑CNP‑shCcn3, AAV‑mediated EGFP expression co‑localized with SCs (S100β⁺ cells) within the tumor tissue, and the viral transduction efficiency reached 92.03% (Fig. 2F). In contrast, no EGFP signal was detected in the adjacent normal pancreatic tissue or within the DRG in PDAC mice (Fig. 2F). Intratumoral delivery of AAV-CNP-shCcn3 significantly reduced CCN3 mRNA and protein levels in tumor-infiltrating SCs (Supplementary Fig. 5G, H), demonstrating that intratumoral delivery effectively targeted tumor-associated SCs. Importantly, SC-specific CCN3 knockdown significantly inhibited tumor growth (Fig. 2D, E) and decreased nerve fiber density, indicated by reduced PGP9.5-positive signals, which included TRPV1⁺/CGRP⁺ nerve fibers (Fig. 2I–K). Further, chronic pain in the PDAC model mice was markedly alleviated (Fig. 2G, H). Together, these results suggest that cancer cells contribute to CCN3⁺ SC proliferation, thereby promoting nerve fiber growth and chronic pain in PDAC.

GDF15–GFRAL signaling mediates PDAC-induced chronic pain by driving CCN3⁺ SC proliferation and subsequent nerve growth and sensitization

PDAC originates from the malignant transformation of epithelial cells^26–32, identified by labeling Krt19, Ceacam6, and Muc1 (Supplementary Fig. 2B). To elucidate the molecular mechanisms of cancer cell-mediated SC proliferation and migration, we performed cell-cell communication analysis using a single-cell sequencing dataset (GSE212966). Compared with adjacent normal tissues, the number and levels of cellular interactions in tumor tissues were significantly altered (Fig. 3A). Moreover, the expression levels of genes involved in the GDF signaling pathway were markedly higher in tumor tissues than in adjacent normal tissues (Supplementary Fig. 6 A). Further analysis of intercellular communication showed that among all the enhanced signaling pathways in patients with PDAC, the levels of GDF signaling from epithelium-derived cells to SCs were significantly increased (Fig. 3B). The transforming growth factor beta (TGF-β) superfamily member GDF15 plays an important role in the pathogenesis of diseases such as diabetes and cancer^33–45. In the present study, GDF15 expression was elevated in the tumor tissues from patients with PDAC relative to that in the adjacent normal tissues (Fig. 3C). Western blot assays also confirmed increased GDF15 expression in the tumors of PDAC mice (Fig. 3D). Differentially expressed gene analysis further revealed that upregulated Gdf15 was predominantly expressed in epithelium-derived cells (Supplementary Fig. 6B, C). Moreover, plasma GDF15 levels were quantified, and numerical rating scale (NRS) and visual analogue scale (VAS) pain scores, reflecting the intensity of pain experienced by patients, were obtained from 120 patients with PDAC. Compared with the no-pain group, patients in the pain group showed significantly elevated plasma GDF15 concentrations, with no significant sex-based differences (Fig. 3E, F). The plasma levels of GDF15 in patients with PDAC were positively correlated with both NRS pain scores (R² = 0.6443) and VAS pain scores (R² = 0.6091) (Fig. 3G, Supplementary Fig. 6D). Receiver operating characteristic curve analyses demonstrated that plasma GDF15 levels could serve as a potential biomarker for pain in patients with PDAC (AUC = 0.8615) (Fig. 3H). Prior studies indicate that intratumoral innervation promotes cancer progression^46–48, and that an increased density of neural distribution within tumors is associated with pain⁴⁹. In our cohort, patients in the pain group showed a significantly higher metastasis rate (Fig. 3I). Furthermore, PDAC patients with metastatic progression showed significantly elevated plasma GDF15 concentrations, which effectively predicted pain development (Fig. 3J). GDF15 expression in PDAC mice was positively correlated with S100β expression or Von-Frey cumulative responses (Fig. 3K, L). Collectively, these results suggest that the GDF15 signaling pathway from cancer cells to SCs may contribute to chronic pain in PDAC.

Fig. 3. Increased GDF15 signaling pathway from cancer cells to SCs is linked to clinical pain in PDAC.

A Cell communication analysis showed the alterations in the number and levels of cellular interactions in adjacent normal tissues or tumor tissues from a single-cell sequencing dataset of PDAC patients (GSE212966). B Signal pathway strength analysis showed the levels of GDF signaling pathways from epithelium-derived cells to SCs in adjacent normal tissues and tumor tissues from a single-cell sequencing dataset of PDAC patients (GSE212966). C Representative staining images and the quantification of GDF15 in the adjacent normal tissue and pancreatic tumor in PDAC patients. Scale bars: 100 μm. two-tailed unpaired t test. Cohen’s d = 3.002. 95% CI [1.938 to 68.00]. *P = 0.0417 vs. normal group. n = 3 patients (ADJ), 4 patients (PDAC). D Representative immunoblot images showing the protein level of GDF15 in sham group pancreas and the pancreatic tumor of PDAC mice. two-tailed unpaired t-test. Cohen’s d = 3.154. 95% CI [0.201 to 0.546]. **P = 0.0011 vs. sham group. n = 5 mice per group. E The quantification showing the plasma GDF15 levels of PDAC patients in the no-pain group and pain group. two-tailed unpaired t-test. Cohen’s d = 1.023. 95% CI [2.151 to 3.593]. ****P < 0.0001 vs. no-pain group. n = 44 (No-pain), 76 (Pain). F The quantification showing no significant sex-based differences observed in the plasma GDF15 levels of PDAC patients from the no-pain group and pain group. two-way ANOVA with Šidák’s post hoc test. No significant interaction or sex effect was found. n = 120 total patients. G Statistical analysis of the correlations between the plasma levels of GDF15 and VAS pain score of PDAC patients. Pearson correlation (two-tailed). r = 0.780. R² = 0.609. 95% CI [0.699 to 0.842]. ****P < 0.0001. n = 120 patients. H ROC curve analysis showing that plasma GDF15 levels were a potential biomarker for distinguishing patients with chronic pain induced by PDAC. ROC curve analysis. AUC = 0.862. 95% CI [0.792 to 0.931]. two-tailed DeLong’s test, ****P < 0.0001. n = 44 patients (No-pain), 76 Patients (Pain). I The quantification showing the metastasis percentage of PDAC patients from the no-pain group and pain group. Fisher’s exact test (two-sided). ****P < 0.0001. n = 44 patients (No-pain), 76 patients (Pain). J The quantification showing the plasma GDF15 levels of PDAC patients from the no-metastasis group and the metastasis group. two-tailed unpaired t test. Cohen’s d = 0.523. 95% CI [0.408 to 2.286]. **P = 0.0053 vs. No-meta group. n = 32 patients (Meta), 88 patients (No-meta). K Statistical analysis of the correlations between the protein level of GDF15 and S100β in the sham group and PDAC mice. Pearson correlation (two-tailed). r = 0.898. R² = 0.807. 95% CI [0.670 to 0.971]. ****P < 0.0001. n = 12 mice. L Statistical analysis of the correlations between the protein level of GDF15 and mechanical allodynia in the sham group and PDAC mice. Pearson correlation (two-tailed). r = 0.814. R² = 0.663. 95% CI [0.452 to 0.946]. **P = 0.0013. n = 12 mice. Data are shown as the mean ± SEM. *P < 0.05, **P < 0.01, ****P < 0.0001. Source data are provided as a Source Data file.

As a secreted cytokine, GDF15 participates in biological processes such as cell growth by binding to its specific receptor GFRAL^41,42. To determine whether the GDF15-GFRAL pathway is involved in CCN3⁺ SC proliferation and nerve growth, we first screened a specific siGdf15 and knocked down GDF15 in K8484 cells using siGdf15 (Supplementary Fig. 7A). Following incubation of RSC96 cells with the supernatant of K8484 cells treated with siGdf15, the number of EdU-positive cells, number of migrating cells, and proportion of CCN3⁺S100β⁺ cells in RSC96 cells were significantly reduced relative to those in the scramble group (Fig. 4A, Supplementary Fig. 7B, C). Furthermore, the fiber length of ND7/23 cells was significantly reduced following incubation with conditioned medium from the co-culture of siGdf15-treated K8484 and RSC96 cells (Supplementary Fig. 7D). Moreover, GDF15 knockdown in tumor cells significantly suppressed the TRPV1 and CGRP expression in co-cultured ND7/23 cells (Supplementary Fig. 7E, F). We further established a stable PDAC cell line by shGdf15 knockdown and performed a CCK‑8 proliferation assay. The results demonstrated that GDF15 knockdown impaired the proliferative capacity of PDAC cells (Supplementary Fig. 7I). Further, the injection of K8484 cells treated with shGdf15 inhibited tumor growth (Fig. 4B, C). Importantly, compared with the scramble group, injecting K8484 cells treated with shGdf15 into the pancreas in situ prevented the increased number of CCN3⁺ SCs (CCN3⁺S100β⁺ cells) (Fig. 4D) and the amount of nerve fiber (PGP9.5-positive signaling), including TRPV1⁺/CGRP⁺ nerve fibers (Fig. 4E, Supplementary Fig. 7G, H), while it also partially alleviated PDAC-induced chronic pain (Fig. 4F–I).

Fig. 4. GDF15 inhibition reverses cancer-induced CCN3⁺ SCs proliferation, nerve fiber growth, and PDAC pain.

A Flow cytometry analysis showed that the proportion of CCN3⁺S100β⁺ cells in RSC96 cells was reduced following the incubation of K8484 cell supernatant treated with siGdf15. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484-siGdf15 SN vs. K8484-siNC SN: Mean difference = –3.050. 95% CI [–5.664 to –0.436]. *P = 0.0186. n = 6 independent experiments per group. The gating strategy is shown in Supplementary Fig. 15. The schematic diagram was created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889. B Schematic diagram showing the syngeneic K8484 PDAC cancer cells treated with shGdf15 were injected into the pancreas in situ (Created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889). C Specimen diagrams and quantification showing the pancreatic tumor of mice following the pre-application of shGdf15 on K8484 cells. Scale bars: 1 cm. two-tailed unpaired t test. Cohen’s d = 3.754. 95% CI [−0.807 to −0.355]. ***P = 0.0004 vs. shNC group. n = 5 mice per group. D Representative double staining images and the quantification showed that the pre-application of shGdf15 on K8484 cells inhibited the number of CCN3⁺S100β⁺ cells. Scale bars: 20 μm. two-tailed unpaired t test. Cohen’s d = 3.363. 95% CI [−3.735 to −1.297]. **P = 0.0014 vs. shNC group. n = 5 mice per group. E Representative immunofluorescence staining showed the pre-application of shGdf15 on K8484 cells inhibited the increased co-expression of the PGP9.5 and S100β in the pancreatic tumor of PDAC mice. Scale bars: 20 μm. two-tailed unpaired t test. Cohen’s d = 2.348. 95% CI [−2.910 to −0.524]. *P = 0.0105 vs. shNC group. n = 5 mice per group. F-I Mechanical allodynia, spontaneous pain and movement restriction were explored in PDAC mice following injection of K8484 cells treated with scrambled shRNA or shGdf15 into the pancreas in situ. (F) two-way repeated measures ANOVA. η² (shNC vs shGdf15) = 0.072. 95% CI [–45.03 to –25.37]. Šidák’s post hoc test for multiple comparisons: **P = 0.0012, ****P < 0.0001 vs. shNC group. n = 5 mice per group. (G) two-way repeated measures ANOVA. η² (shNC vs shGdf15) = 0.138. 95% CI [−25.60 to −9.597]. Šidák’s post hoc test for multiple comparisons: **P = 0.0042 (week 4); **P = 0.0011 (week 5) vs. shNC group. n = 5 mice per group. (H) Mann-Whitney U test (two-sided). Cliff’s delta = −0.880. 95% CI of median difference [−1.000 to 1.000]. *P = 0.0317 vs. shNC group. n = 5 mice per group. (I) two-tailed unpaired t test. Cohen’s d = 4.816. 95% CI [2.638 to 5.336]. ***P = 0.0001 vs. shNC group. n = 5 mice per group. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

We also explored the role of GFRAL (the GDF15 receptor) by using siGfral in RSC96 cells to knock down GFRAL expression (Supplementary Fig. 8A). Following incubation of siGfral-treated RSC96 cells with the supernatant of K8484 cells, the number of EdU-positive cells, the number of migrating cells, and the proportion of CCN3⁺S100β⁺ cells in RSC96 cells were significantly reduced relative to those in the scramble group (Fig. 5A, Supplementary Fig. 8B, C). Furthermore, the fiber length of ND7/23 cells was significantly reduced following incubation with conditioned medium from the co-culture of K8484 and RSC96 cells treated with siGfral (Supplementary Fig. 8D). GFRAL knockdown in SCs significantly suppressed TRPV1 and CGRP expression in cocultured ND7/23 cells (Supplementary Fig. 8E, F).

Fig. 5. The GFRAL of SCs mediates PDAC-induced chronic pain by promoting CCN3⁺ SCs proliferation and nerve fiber growth.

A Flow cytometry analysis showed the proportion of CCN3⁺S100β⁺ cells in RSC96 cells following the incubation of K8484 cell supernatant treated with siGfral. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484 + RSC96-siGfral SN vs. K8484 + RSC96-siNC SN: 95% CI [–8.109 to –4.644]. ****P < 0.0001. n = 6 mice per group. The gating strategy is shown in Supplementary Fig. 15. The schematic diagram was created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889. B Schematic diagram showed that AAV-CNP-shGfral or AAV-CNP-scrambled shRNA was intratumorally injected into pancreatic tumor (Created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889). C Specimen diagrams and quantification showing the pancreatic tumor of mice following the intratumoral injection of AAV-CNP-shGfral or AAV-CNP-scrambled shRNA. Scale bars: 1 cm. two-tailed unpaired t test. Cohen’s d = 2.688. 95% CI [−1.238 to −0.303]. **P = 0.0052 vs. shNC group. n = 5 mice per group. D Representative staining and quantification showed that intratumoral injection of AAV-CNP-shGfral inhibited the increased colocalization of CCN3 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + AAV-CNP-shGfral vs. PDAC + AAV-CNP-shNC: 95% CI [–1.853 to –1.171]. ****P < 0.0001. n = 5 mice per group. E Representative immunofluorescence staining and quantification showed that inhibition of GFRAL by using AAV-CNP-shGfral prevented the increased co-expression of the PGP9.5 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + AAV-CNP-shGfral vs. PDAC + AAV-CNP-shNC: 95% CI [–4.276 to –1.334]. ***P = 0.0003. n = 5 mice per group. F–I Mechanical allodynia, spontaneous pain and movement restriction in PDAC mice after intratumoral injection of AAV-CNP-shGfral or AAV-CNP-scrambled shRNA. F two-way ANOVA. η² = 0.202. Šidák’s post hoc test for multiple comparisons: **P = 0.0093, ****P < 0.0001 vs. AAV-CNP-shNC group. n = 6 mice group. G two-way ANOVA. η² = 0.220. Šidák’s post hoc test for multiple comparisons: **P = 0.0025, ***P = 0.0001 vs. AAV-CNP-shNC group. n = 6 mice group. H Mann-Whitney U test (two-sided). Cliff’s delta = −0.778. 95% CI of median difference [−1.000 to 1.000]. *P = 0.0325 vs. shNC group. n = 6 mice per group. I two-tailed unpaired t test. Cohen’s d = 4.816. 95% CI [2.638 to 5.336]. ***P = 0.0001 vs. shNC group. n = 5 mice per group. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

GFRAL was widely expressed in the SCs of patients with PDAC and PDAC model mice (Supplementary Fig. 9B, C). GFRAL was highly expressed in CCN3⁺ SCs but showed low expression in FOSB⁺ SCs and ANGPTL7⁺ SCs (Supplementary Fig. 9A). Following intratumoral injection of AAV‑CNP‑shGfral, AAV‑mediated EGFP expression co‑localized with SCs (S100β⁺ cells) within the tumor tissue and the viral transduction efficiency was 90.09% (Supplementary Fig. 9D). Across both the K8484 and KPC models, intratumoral injection of AAV-CNP-shGfral inhibited tumor growth (Fig. 5B, C, Supplementary Fig. 9E). Knockdown of GFRAL in SCs by intratumoral injection of AAV-CNP-shGfral reduced the number of CCN3⁺ SCs (CCN3⁺S100β⁺ cells) and the amount of nerve fibers (PGP9.5-positive signaling) including TRPV1⁺/CGRP⁺ nerve fibers in PDAC mice (Fig. 5D, E, Supplementary Fig. 9F, G, L, M). Importantly, intratumoral injection of AAV-CNP-shGfral alleviated chronic pain in both the K8484 and KPC models of PDAC mice, as indicated by significantly decreased cumulative withdrawal response, response frequency, and hunching score and a marked increase in total distance traveled (Fig. 5F–I, Supplementary Fig. 9H-K). Notably, intratumoral injection of AAV‑CNP‑shGfral at Week 5 when PDAC pain had peaked, still alleviated pain over the subsequent two weeks (Supplementary Fig. 9N). Overall, these results suggest that the interaction of GDF15 and GFRAL between cancer cells and SCs mediates PDAC-induced chronic pain by promoting CCN3⁺ SC proliferation and nerve fiber growth.

Increased GFRAL-mediated PFKM contributes to chronic pain by promoting CCN3⁺ SC proliferation and the subsequent nerve growth and sensitization

To elucidate the molecular mechanism of GFRAL signaling pathway-mediated SC proliferation in PDAC, we performed GO enrichment analysis of differential signaling pathways in CCN3⁺ SCs based on a single-cell sequencing dataset (GSE212966). Metabolism-related signaling pathways such as glucolipid binding, proteoglycan binding, and glycosaminoglycan binding were significantly upregulated in CCN3⁺ SCs (Fig. 6A). scMetabolism algorithms further showed significant upregulation of carbohydrate metabolism-related signaling pathways, especially glycolysis (Fig. 6B). Consistently, glycolytic proton efflux rate (PER), which indicates glycolytic activity, increased significantly in SCs after incubation with the supernatant of K8484 cells (Fig. 6C). Among the key enzymes in glycolysis, the mRNA levels of Hk1, Hk2, Pfkm, Pfkl, and Pfkp in SCs were significantly increased after incubation with the supernatant of K8484 cells (Fig. 6F). Following incubation with supernatant of K8484 cells treated with siGdf15, the increase in glycolytic PER was inhibited in RSC96 cells (Fig. 6D), and Pfkm expression was inhibited rather than that of Hk1, Hk2, Pfkl, and Pfkp (Fig. 6F). Immunofluorescence staining showed that the number of PFKM-expressing SCs increased significantly in patients with PDAC (Fig. 6I) and PDAC mice (Fig. 6J). Furthermore, both PFKM and GFRAL were co-expressed in SCs (Fig. 6L), and knockdown of the GDF15 receptor GFRAL using siGfral inhibited the K8484 cell supernatant-induced Pfkm and glycolytic PER upregulation (Fig. 6E, G). Pre-application of ML251 (a PFK inhibitor) inhibited the K8484 cell supernatant-induced increase in glycolytic PER, while reducing the proportion of CCN3⁺S100β⁺ cells among RSC96 cells (Fig. 6H, K). Additionally, PFKM knockdown in SCs using siPfkm significantly reduced the proportion of CCN3⁺ SCs among co-cultured SCs (Supplementary Fig. 10A, B).

Fig. 6. GFRAL-mediated PFKM contributes to glycolytic reprogramming and CCN3⁺ SCs expansion.

A GO enrichment analysis of differential signaling pathways in CCN3⁺ SCs from a single-cell sequencing dataset of PDAC patients (GSE212966). B scMetabolism algorithms score of 3 Schwann cell subclusters, including CCN3⁺ SCs, FOSB⁺ SCs and ANGPTL7⁺ SCs from a single-cell sequencing dataset of PDAC patients (GSE212966). C The glycolic proton efflux rate was examined in RSC96 cells following the incubation of 10% DMEM or K8484 cell supernatant. two-way ANOVA (two-sided). η² (K8484 SN vs. 10% DMEM) = 0.052. 95% CI [130.1 to 159.0]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001 vs. 10% DMEM group. n = 8 independent experiments per group. D The glycolic proton efflux rate was examined in RSC96 cells following the incubation of 10% DMEM or K8484 cell supernatant treated with siGdf15 or scrambled siRNA. two-way ANOVA (two-sided). η² (K8484-siGdf15 supernatant vs. K8484-siNC supernatant) = 0.014. 95% CI [−75.61 to −57.62]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001 vs. K8484-siNC supernatant group. n = 8 independent experiments per group. E The glycolic proton efflux rate was examined in RSC96 cells following the incubation of 10% DMEM or K8484 cell supernatant treated with siGfral or scrambled siRNA. two-way ANOVA (two-sided). η² (K8484 supernatant+RSC96-siNC vs. K8484 supernatant+RSC96-siGfral) = 0.034. 95% CI [−112.3 to −85.37]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001. n = 8 independent experiments per group. F The mRNA level of key enzymes in glycolysis (Hk1, Hk2, Pfkm, Pfkl, Pfkp, Pkm) was examined in RSC96 cells following the incubation of 10% DMEM or K8484 cell supernatant treated with siGdf15 or scrambled siRNA. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484 SN vs. 10% DMEM: **P = 0.0013 (Hk1); **P = 0.0066 (Pfkl), ***P = 0.0003 (Pfkm), ****P < 0.0001 (Hk2, Pfkp). K8484-siGdf15 vs. K8484-siNC: *P = 0.0162 (Pfkm). n = 3 independent experiments per group. G The Pfkm mRNA level in RSC96 cells following the incubation of 10% DMEM or K8484 cell supernatant treated with siGfral or scrambled siRNA. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484 supernatant+RSC96-siGfral vs. K8484 supernatant+RSC96-siNC: 95% CI [−0.6808 to −0.2873]. ***P = 0.0002. n = 3 independent experiments per group. H The glycolic proton efflux rate was examined in RSC96 cells treated with ML251 or DMSO following the incubation of K8484 cell supernatant. two-way ANOVA (two-sided). η² (K8484 supernatant+ML25 vs. K8484 supernatant+DMSO) = 0.037. 95% CI [–113.5 to –85.62]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001. n = 8 independent experiments per group. I Representative double immunofluorescence staining and quantification showing the colocalization of PFKM and S100β in the adjacent normal tissue and pancreatic tumor in PDAC patients. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 1.511. 95% CI [1.485 to 7.526]. **P = 0.0059 vs. ADJ group. n = 8 patients (ADJ), 11 patients (PDAC). J Representative double immunofluorescence staining and quantification showing the colocalization of PFKM and S100β in sham group pancreas and the pancreatic tumor of PDAC mice. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 2.877. 95% CI [1.806 to 3.775]. ****P < 0.0001 vs. Sham group. n = 8 mice (Sham), 11 mice (PDAC). K Flow cytometry analysis showed the proportion of CCN3⁺S100β⁺ cells in RSC96 cells treated with ML251 or DMSO following the incubation of K8484 cell supernatant. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484 supernatant+ML251 vs. K8484 supernatant+DMSO: 95% CI [–8.813 to –2.667]. ***P = 0.0002. n = 6 independent experiments per group. The gating strategy is shown in Supplementary Fig. 15. The schematic diagram was created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889. L Representative immunofluorescence staining and analysis showed the colocalization of PFKM, GFRAL, and S100β in the pancreatic tumor in PDAC patients (n = 5 patients). Scale bars: 50 μm. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

PFKM inhibition by intratumoral injection of ML251 suppressed the increase in the number of CCN3⁺ SCs and nerve fibers, including TRPV1⁺/CGRP⁺ nerve fibers, and the chronic pain induced by GFRAL overexpression in SCs (Fig. 7A–E, Supplementary Fig. 10C, D). Importantly, applying ML251 inhibited the increase in the number of CCN3⁺ SCs and nerve fibers containing TRPV1⁺/CGRP⁺ nerve fibers, while relieving chronic pain in PDAC mice (Fig. 7F–J, Supplementary Fig. 10E, F). Furthermore, intratumoral injection of AAV-CNP-shPfkm markedly reversed the exacerbated pain and tumor growth induced by GFRAL overexpression in SCs (Supplementary Fig. 11A–C). The intratumoral injection of AAV-CNP-shPfkm also significantly alleviated pain and tumor growth in PDAC mice (Supplementary Fig. 11D–F). Notably, intratumoral injection of AAV‑CNP‑shPfkm at Week 5, still alleviated pain over the subsequent two weeks (Supplementary Fig. 11G). Correspondingly, in a PDAC mouse model established with GDF15-knockdown tumor cells and intratumoral injections of AAV-CNP-Pfkm to overexpress PFKM in SCs, PFKM overexpression was found to restore the pain and tumor growth in PDAC mice alleviated by GDF15 knockdown in tumor cells (Fig. 8A–C). Furthermore, intratumoral injection of AAV-CNP-Pfkm alone to overexpress PFKM in SCs exacerbated both pain and tumor growth in PDAC mice (Fig. 8D–F). Overall, upregulation of the GDF15/GFRAL-mediated key enzyme PFKM in the glycolysis process of CCN3⁺ SCs contributes to chronic pain by promoting CCN3⁺ SCs proliferation and nerve fiber growth in PDAC.

Fig. 7. GFRAL-mediated PFKM upregulation in SCs is indispensable for tumor innervation and pancreatic cancer pain.

A Representative staining and quantification showed that ML251 suppressed the increase in the number of CCN3⁺ SCs induced by intratumoral injection of AAV-CNP-Gfral in the pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + AAV-CNP-Gfral + ML251 vs. PDAC + AAV-CNP-Gfral+Saline: 95% CI [–5.924 to –2.494]. ****P < 0.0001. n = 5 mice per group. B Representative staining and quantification showed that ML251 suppressed the increased co-expression of the PGP9.5 and S100β induced by intratumoral injection of AAV-CNP-Gfral in pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + AAV-CNP-Gfral + ML251 vs. PDAC + AAV-CNP-Gfral+Saline: 95% CI [–10.51 to –1.911]. **P = 0.0060. n = 5 mice per group. C–E Mechanical allodynia and movement restriction in PDAC mice were explored following intratumoral injection of AAV-CNP-Gfral and ML251. C two-way ANOVA. η² = 0.047. Šidák’s post hoc test for multiple comparisons: **P = 0.0037, ****P < 0.0001 vs. PDAC + AAV-CNP-Gfral+Saline. n = 4 mice per group. D two-way ANOVA. η² = 0.070. Šidák’s post hoc test for multiple comparisons: **P = 0.0058, ****P < 0.0001 vs. PDAC + AAV-CNP-Gfral+Saline. n = 4 mice per group. E one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + AAV-CNP-Gfral + ML251 vs. PDAC + AAV-CNP-Gfral+Saline: 95% CI [–3.171 to –0.3854]. *P = 0.0151. n = 4 mice per group. F Representative staining and quantification showed that intratumoral injection of ML251 decreased the colocalization of CCN3 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + ML251 vs. PDAC+Saline: 95% CI [–3.510 to –0.9691]. **P = 0.0014. n = 5 mice per group. G Representative staining and quantification showed that the increased co-expression of PGP9.5 and S100β was inhibited by ML251 in pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + ML251 vs. PDAC+Saline: 95% CI [–5.114 to –1.716]. ***P = 0.0005. n = 5 mice per group. H–J Mechanical allodynia and movement restriction in PDAC mice were explored following intratumoral injection of ML251. H two-way ANOVA. η² = 0.065. Šidák’s post hoc test for multiple comparisons: ***P = 0.0005, ****P < 0.0001 vs. PDAC+Saline. n = 4 mice per group. I two-way ANOVA. η² = 0.090. Šidák’s post hoc test for multiple comparisons: ***P = 0.0009 (week 4); ***P = 0.0004 (week 5) vs. PDAC+Saline. n = 4 mice per group. J one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC + ML251 vs. PDAC+Saline: 95% CI [–5.439 to –1.025]. **P = 0.0069. n = 4 per group. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Fig. 8. GDF15-mediated upregulation of PFKM in SCs is essential for driving nerve fiber sensitization and pancreatic cancer pain.

A Specimen diagrams and quantification showing the pancreatic tumor of mice following the pre-application of shGdf15 on K8484 cells and intratumoral injection of AAV-CNP-Pfkm or AAV-CNP-EGFP. Scale bars: 1 cm. two-tailed unpaired t test. Cohen’s d = 4.699. 95% CI [0.391 to 0.806]. ***P = 0.0002 vs. PDAC-shGdf15 + AAV-CNP-EGFP group. n = 5 mice per group. B, C Mechanical allodynia and movement restriction in mice following the pre-application of shGdf15 on K8484 cells and intratumoral injection of AAV-CNP-Pfkm. B Von-frey cumulative response: two-way repeated measures ANOVA. η² (PDAC-shGdf15 + AAV-CNP-EGFP vs. PDAC-shGdf15 + AAV-CNP-Pfkm) = 0.046. 95% CI [25.85 to 39.75]. ****P < 0.0001 vs. PDAC-shGdf15 + AAV-CNP-EGFP group. n = 5 mice per group. Von-frey response frequency: two-way repeated measures ANOVA. η² (PDAC-shGdf15 + AAV-CNP-EGFP vs. PDAC-shGdf15 + AAV-CNP-Pfkm) = 0.103. 95% CI [10.28 to 21.72]. *P = 0.0148 (week 3); *P = 0.0148 (week 5), ***P = 0.0004 vs. PDAC-shGdf15 + AAV-CNP-EGFP group. n = 5 mice per group. C two-tailed unpaired t test. Cohen’s d = 5.089. 95% CI [−4.274 to −2.199]. ****P < 0.0001 vs. PDAC-shGdf15 + AAV-CNP-EGFP group. n = 5 mice per group. D Specimen diagrams and quantification showing the pancreatic tumor of mice following the intratumoral injection of AAV-CNP-Pfkm or AAV-CNP-EGFP. Scale bars: 1 cm. two-tailed unpaired t test. Cohen’s d = 4.876. 95% CI [0.836 to 1.546]. ****P < 0.0001. n = 5 mice per group. E, F Mechanical allodynia and movement restriction in PDAC mice were explored following intratumoral injection of AAV-CNP-Pfkm. E Von-frey cumulative response: two-way repeated measures ANOVA. η² (PDAC + AAV-CNP-EGFP vs. PDAC + AAV-CNP-Pfkm) = 0.018. 95% CI [–32.64 to –16.96]. Šidák’s post hoc test for multiple comparisons: **P = 0.003, ****P < 0.0001 vs. PDAC + AAV-CNP-EGFP group. n = 5 mice per group. Von-frey response frequency: two-way repeated measures ANOVA. η² (PDAC + AAV-CNP-EGFP vs. PDAC + AAV-CNP-Pfkm) = 0.065. 95% CI [7.659 to 19.54]. Šidák’s post hoc test for multiple comparisons: **P = 0.0037 (week 4); **P = 0.0037 (week 5) vs. PDAC + AAV-CNP-EGFP group. n = 5 mice per group. F two-tailed unpaired t test. Cohen’s d = 6.118. 95% CI [−7.318 to −4.239]. ****P < 0.0001 vs. PDAC + AAV-CNP-EGFP group. n = 5 mice per group. G ELISA analysis reveals the secretion levels of NGF, BDNF, and IGFBP5 in conditioned medium from the co-culture of siGdf15-treated K8484 cells and RSC96 cells. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484-siNC+RSC96 CM vs. K8484-siGdf15 + RSC96 CM: *P = 0.0207, ***P = 0.0002. n = 3 independent experiments per group. H ELISA analysis reveals the secretion levels of NGF, BDNF, and IGFBP5 in conditioned medium from the co-culture of K8484 cells and siGfral-treated RSC96 cells. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484 + RSC96-siNC vs. K8484 + RSC96-siGfral CM: **P = 0.0016, ****P < 0.0001. n = 3 independent experiments per group. I ELISA analysis reveals the secretion levels of NGF, BDNF, and IGFBP5 in conditioned medium from the co-culture of K8484 cells and siPfkm-treated RSC96 cells. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. K8484 + RSC96-siNC vs. K8484 + RSC96-siPfkm CM: **P = 0.0031, ****P < 0.0001. n = 3 independent experiments per group. J, K AP frequency and AP Rehobase of dorsal root ganglia treated with GW441756 following incubation with comditioned medium from co-culture of K8484 and RSC96 cells. J two-way ANOVA. η² (K8484 + RSC96 CM + GW44175 vs. K8484 + RSC96 CM) = 0.083. 95% CI [–2.477 to –1.940]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001 vs. K8484 + RSC96 CM group. n = 10 independent records per group. K two-tailed unpaired t test. Cohen’s d = 1.261. 95% CI [4.875 to 39.13]. *P = 0.0147 vs. K8484 + RSC96 CM group. n = 10 independent records per group. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

We further investigated the influence of SC glycolysis on neural function. The mRNA levels of various neurotrophic factors (including Ngf, Bdnf, Igfbp5, Btc and Crlf1) increased significantly in RSC96 cells after incubation with K8484 PDAC cell supernatant (Supplementary Fig. 11H). ELISA showed that NGF, BDNF, and IGFBP5 secretion in the conditioned medium of SCs was significantly increased after co-culturing with tumor cells, with NGF being the most abundantly secreted (Supplementary Fig. 11I). Moreover, knockdown of either GDF15 in tumor cells or GFRAL in SCs inhibited the secretion of NGF and BDNF in the conditioned medium, whereas IGFBP5 secretion remained largely unchanged (Fig. 8G, H). More importantly, PFKM knockdown in SCs significantly suppressed NGF and BDNF secretion in the co-culture system (Fig. 8I). Concurrently, pretreatment of primary mouse DRG neurons with GW441756 (a specific NGF receptor inhibitor) reversed the increased AP frequency and the decreased AP Rheobase induced by the co-culture conditioned medium (K8484 + RSC96 CM group data replotted from Fig. 2B, C) (Fig. 8J, K). Overall, glycolysis in CCN3⁺ SCs promotes neurotrophic factor secretion and sensory neuron sensitization, ultimately contributing to PDAC-associated pain.

The GDF15-GFRAL pathway mediates PDAC-induced chronic pain by regulating AKT signaling pathway activity

The GDF15-GFRAL pathway can activate intracellular PI3K-AKT signaling^33,50. In the present study, KEGG enrichment analysis showed a significant upregulation of the PI3K-AKT signaling pathway in CCN3⁺ SCs according to the single-cell sequencing datasets (GSE212966) (Fig. 9A). Western blot assays showed that PFKM and CCN3 expression was significantly upregulated with increased p-AKT levels in RSC96 cells following incubation with K8484 cell supernatant (Supplementary Fig. 12A). However, incubation with the supernatant of K8484 cells treated with siGdf15 significantly inhibited the increased p-AKT, PFKM, and CCN3 in RSC96 cells (Supplementary Fig. 12B). Correspondingly, following incubation of siGfral-treated RSC96 cells with the supernatant of K8484 cells, the expression of the p-AKT, PFKM and CCN3 in RSC96 cells was inhibited (Supplementary Fig. 12C). Furthermore, pre-application of AKTi1/2 (a specific inhibitor of AKT) also reduced PFKM and CCN3 upregulation in RSC96 cells following incubation with the K8484 cell supernatant (Supplementary Fig. 12D). Further, AKT knockdown in SCs using siAkt significantly decreased the proportion of CCN3⁺ SCs within the co-cultured SCs (Supplementary Fig. 12E, F). Thus, GDF15‑GFRAL promotes CCN3⁺ SC proliferation by activating the AKT signaling pathway. Considering the complexity of AKT-dependent proliferative signals, we examined the cell proliferation-associated canonical AKT downstream pathways, including S6 phosphorylation in the mTOR pathway and Cyclin D1 expression, which is involved in cell cycle regulation. Conditioned medium from K8484 tumor cells promoted S6 phosphorylation and Cyclin D1 expression (Supplementary Fig. 13A–D), confirming that the tumor environment contains factors that can activate proliferative pathways. Crucially, GDF15 knockdown in tumor cells or GFRAL knockdown in SCs did not diminish these proliferative signals (Supplementary Fig. 13A–D). Importantly, injecting K8484 cells treated with shGdf15 into the pancreas in situ prevented the increased number of pAKT-expressing SCs (pAKT⁺S100β⁺ cells); further, GFRAL knockdown in SCs by intratumoral injection of AAV-CNP-shGfral also reduced the quantity of pAKT-expressing SCs (Fig. 9B, C). Intratumoral injection of AKTi1/2 reduced the increased quantity of CCN3⁺ SCs (CCN3⁺S100β⁺ cells) (Fig. 9D) and the amount of nerve fiber (PGP9.5-positive signaling) (Fig. 9E), while partially alleviating PDAC-induced chronic pain (Fig. 9F–H). Overall, these results suggest that GDF15-mediated AKT activation contributes to chronic pain by regulating CCN3⁺ SC numbers in PDAC.

Fig. 9. GDF15-GFRAL pathway drives pancreatic cancer pain via regulating the activity of AKT pathway.

A KEGG enrichment analysis of differential signaling pathways in CCN3⁺ SCs from single-cell sequencing dataset of PDAC patients (GSE212966). B Representative double staining image and quantification showed pre-application of shGdf15 on the K8484 cells inhibited the increased colocalization of p-AKT and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 2.592. 95% CI [−3.288 to −0.748]. **P = 0.0064 vs. shNC group. n = 5 mice per group. C Representative staining and quantification showed that intratumoral injection of AAV-CNP-shGfral inhibited the increased colocalization of p-AKT and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 3.811. 95% CI [−3.627 to −1.454]. ***P = 0.0007 vs. PDAC + AAV-CNP-shNC group. n = 5 mice per group. D Representative staining and quantification showed that intratumoral injection of AKTi1/2 inhibited the increased colocalization of CCN3 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC+AKTi1/2 vs. PDAC+Saline: 95% CI [–4.002 to –0.2343]. *P = 0.0249. n = 5 mice per group. E Representative staining and quantification showed that intratumoral injection of AKTi1/2 inhibited the increased colocalization of PGP9.5 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. one-way ANOVA with Tukey’s post hoc test for multiple comparisons. PDAC+AKTi1/2 vs. PDAC+Saline: 95% CI [–4.028 to –0.9998]. **P = 0.0011. n = 5 mice per group. F–H Mechanical allodynia and spontaneous pain in PDAC mice following intratumoral injection of AKTi1/2. F Von-frey cumulative response: two-way ANOVA. η² = 0.193. Šidák’s post hoc test for multiple comparisons: ***P = 0.0003 (week 3); ***P = 0.0003 (week 4), ****P < 0.0001. n = 5 mice per group. G two-way ANOVA. η² = 0.256. Šidák’s post hoc test for multiple comparisons: **P = 0.0011, ****P < 0.0001. n = 5 mice per group. H Mann-Whitney U test (two-sided). Cliff’s delta = −0.889. 95% CI of median difference [−2.000 to 2.000]. **P = 0.0065 vs. PDAC + AAV-CNP-shNC group. n = 6 mice per group. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

AKT-mediated Runx2 phosphorylation promotes PFKM transcription in CCN3⁺ SCs in PDAC

To examine the transcriptional mechanisms underlying PFKM upregulation, we employed the SCENIC algorithm for the single-cell sequencing dataset (GSE212966) and identified the distinctively activated transcription factors within CCN3⁺ SCs, such as MAF, RUNX2, and HMGA1 (Fig. 10A). The analysis of regulon specificity score identified RUNX2 was the most specific transcription regulator for CCN3⁺ SCs, characterized by a target gene set uniquely enriched within this subpopulation (Fig. 10B). Western blot assays indicated that p-RUNX2 levels increased in RSC96 cells after incubation with K8484 cell supernatant (Supplementary Fig. 14A). Furthermore, RUNX2 knockdown using siRunx2 inhibited the K8484 cell supernatant-induced PFKM upregulation and increased CCN3⁺S100β⁺ cells in RSC96 cells (Supplementary Fig. 14B–D). Across both the K8484 and KPC models, intratumoral injection of AAV-CNP-shRunx2 inhibited tumor growth (Fig. 10C, D, Supplementary Fig. 14F). Importantly, RUNX2 knockdown by intratumoral injection of AAV-CNP-shRunx2 blocked the PDAC-induced increase in CCN3⁺ SC number, nerve fiber growth, and chronic pain in both K8484 and KPC models (Fig. 10E–J, Supplementary Fig. 14G–L). Thus, RUNX2 activation contributes to chronic pain via transcriptional regulation of PFKM expression. Analysis of the promoter sequence of Pfkm using the JASPAR database revealed the putative binding site of RUNX2 was present on the Pfkm promoter (Fig. 10K). Chromatin immunoprecipitation further showed RUNX2 recruitment to the Pfkm promoter was significantly increased at the specific binding site after incubation with the K8484 cell supernatant (Fig. 10L). Furthermore, incubation of SCs with recombinant GDF15 protein, even in the absence of tumor cell supernatant, promoted RUNX2 phosphorylation (Supplementary Fig. 14E). Moreover, AKT inhibitor treatment effectively blocked the GDF15-induced increase in RUNX2 phosphorylation (Supplementary Fig. 14E), indicating AKT involvement in GDF15 signaling-induced RUNX2 phosphorylation. Next, we explored the role of AKT pathway in RUNX2-mediated PFKM expression. Co-immunoprecipitation demonstrated increased interaction between AKT and p-RUNX2 following incubation with the K8484 cell supernatant (Fig. 10M). Moreover, AKTi1/2 application prevented the increase in Runx2 phosphorylation (Supplementary Fig. 14A) and the interaction between AKT and p-RUNX2 (Fig. 10N), while inhibiting the increased recruitment of RUNX2 to the Pfkm promoter following incubation with the K8484 cell supernatant (Fig. 10L). Overall, the AKT signaling pathway promoted Runx2 phosphorylation, thereby promoting Pfkm transcription and enhancing cellular glycolysis in CCN3⁺ SCs.

Fig. 10. AKT-regulated Runx2 phosphorylation promotes the PFKM transcription in CCN3⁺ SCs in PDAC.

A The SCENIC algorithm showed the identification of the distinctively activated transcription factors within SCs from a single-cell sequencing dataset of PDAC patients (GSE212966). B The analysis of regulon specificity score showed the most specific transcription regulator for SCs from single-cell sequencing dataset of PDAC patients (GSE212966). C Schematic diagram showed that AAV-CNP-shRunx2 or AAV-CNP-scrambled shRNA was intratumorally injected into a pancreatic tumor (Created in BioRender. Guojun, C. (2026) https://BioRender.com/d25l889). D Specimen diagrams and quantification showing the pancreatic tumor of mice following the intratumoral injection of AAV-CNP-shRunx2 or AAV-CNP-scrambled shRNA. Scale bars: 1 cm. two-tailed unpaired t test. Cohen’s d = 2.876. 95% CI [−1.108 to −0.108]. *P = 0.0238 vs. AAV-CNP-shNC group. n = 5 mice (AAV-CNP-shNC), 4 mice (AAV-CNP-shRunx2). E Representative double immunofluorescence staining and quantification showing intratumoral injection of AAV-CNP-shRunx2 inhibited the increased colocalization of CCN3 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 2.450. 95% CI [−3.911 to −0.785]. **P = 0.0085 vs. PDAC + AAV-CNP-shNC group. n = 5 mice per group. F Representative double immunofluorescence staining and quantification showing intratumoral injection of AAV-CNP-shRunx2 inhibited the increased colocalization of PGP9.5 and S100β in pancreatic tumor of PDAC mice. Scale bars: 50 μm. two-tailed unpaired t test. Cohen’s d = 2.197. 95% CI [−3.855 to −0.571]. *P = 0.0145 vs. PDAC + AAV-CNP-shNC group. n = 5 mice per group. G–J Mechanical allodynia, spontaneous pain and movement restriction in PDAC mice after intratumoral injection of AAV-CNP-shRunx2 or AAV-CNP-scrambled shRNA. G, H Von-frey cumulative response: two-way repeated measures ANOVA. η² (PDAC + AAV-CNP-shNC vs PDAC + AAV-CNP-shRunx2) = 0.053. 95% CI [−45.97 to −21.23]. Šidák’s post hoc test for multiple comparisons: ***P = 0.0004, ****P < 0.0001 vs. PDAC + AAV-CNP-shNC group. n = 5 mice per group. Von-frey response frequency: two-way repeated measures ANOVA. η² (PDAC + AAV-CNP-shNC vs. PDAC + AAV-CNP-shRunx2) = 0.074. 95% CI [−21.81 to −6.991]. Šidák’s post hoc test for multiple comparisons: *P = 0.0277, ****P < 0.0001 vs. PDAC + AAV-CNP-shNC group. n = 5 mice per group. I Mann-Whitney U test (two-sided). Cliff’s delta = −0.720. 95% CI of median difference [−2.000 to 2.000]. P = 0.0635 (ns) vs. PDAC + AAV-CNP-shNC group. n = 5 per group. J two-tailed unpaired t test. Cohen’s d = 3.464. 95% CI [1.507 to 4.656]. **P = 0.0030 vs. PDAC + AAV-CNP-shNC group. n = 4 mice per group. K The pattern showing the putative binding site of RUNX2 to the Pfkm promoter by using the JASPAR database. L Chromatin immunoprecipitation showed the interaction between RUNX2 and Pfkm promoter in RSC96 cells treated with AKTi1/2 or DMSO following the incubation of K8484 cell supernatant. two-way ANOVA. η² (RUNX2 vs. lgG) = 0.881. 95% CI [3.107 to 3.323]. Šidák’s post hoc test for multiple comparisons: ****P < 0.0001 vs. lgG group. n = 3 independent experiments per group. M Representative co-immunoprecipitation images showing the interaction between AKT and p-RUNX2 in RSC96 cells following the incubation of 10% DMEM or K8484 cell supernatant (n = 3 independent experiments). The samples derive from the same experiment but different gels for IP and WCL were processed in parallel. N Representative co-immunoprecipitation images showing the interaction between AKT and p-RUNX2 in RSC96 cells treated with AKTi1/2 or DMSO following the incubation of K8484 cell supernatant (n = 3 independent experiments). The samples derive from the same experiment but different gels for IP and WCL were processed in parallel. Data are shown as the mean ± SEM. ns, not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Discussion

Long-term abdominal pain is a significant clinical symptom in patients with PDAC, along with increased nerve density mediated by nerve infiltration. SCs play crucial roles in neural support and myelin formation. However, whether and how SCs mediate PDAC-induced chronic pain remains unclear. In the present study, we identified a CCN3⁺ SC sub-cluster by analyzing a single-cell sequencing dataset from patients with PDAC. The number of CCN3⁺ SCs was significantly increased in both patients with PDAC and PDAC mice, and specific ablation of CCN3⁺ SCs attenuated PDAC-induced chronic pain.

During pancreatic cancer progression, pancreatic nerves show notable changes, including increased size and density⁴⁸. SCs are an important component of pancreatic nerves⁵¹. Tumor environment components, such as tumor-associated macrophages, are known to activate SCs in PDAC⁵². Here, we found that incubation with K8484 supernatant (from mouse pancreatic cancer cells) increased CCN3 expression and the proportion of CCN3⁺S100β⁺ cells in RSC96 cells (SCs), suggesting that cancer cells promoted CCN3⁺ SC proliferation. Studies have revealed that tumor cells demonstrate a propensity to grow toward nerves and promote nerve endings to specifically extend toward them⁵³. We observed that incubation with conditioned medium from the co‑culture of K8484 and RSC96 cells significantly increased the neurite length of ND7/23 cells (a DRG‑derived line), elevated the AP frequency, and reduced the AP Rheobase in mouse primary DRG neurons, suggesting that the interaction between SCs and cancer cells promoted nerve fiber growth and sensitization. This interaction allowed more SCs to engage with tumor cells, potentially promoting CCN3⁺ SC proliferation and increasing nerve fiber excitability, thereby mediating chronic pain.

Cell communication studies further indicated enhanced GDF signaling between tumor cells and SCs in PDAC. As a member of the transforming growth factor beta (TGF-β) superfamily, GDF15 is typically expressed at low levels in most somatic tissues under physiological conditions⁵⁴. We observed a significant increase in GDF15 expression in both PDAC patients and PDAC mice, predominantly in epithelial-derived cells, and this upregulation was positively correlated with chronic pain. This finding is consistent with reports that GDF15 expression is often increased in various cancers, including those of the digestive system⁵⁵. Clinically, as plasma GDF15 levels in patients with PDAC were positively correlated with pain scores, GDF15 could serve as an effective biomarker for pain in these patients. It should be noted that multivariate models incorporating stage, metastasis, and treatment variables are still needed to determine the predictive value of GDF15. Although GDF15 expression is associated with cancer cell proliferation and progression⁵⁶, current research suggests that GDF15 upregulation in pancreatic tissue may mediate PDAC-induced chronic pain. Notably, we observed a significant association between pain and metastasis. The relationship between pain and metastasis has been reported⁵⁷. Pain-induced neural remodeling may generate a microenvironment that favors metastatic dissemination^57,58, while metastatic burden could in turn exacerbate pain through mechanisms such as nerve invasion or inflammation^57,59. In our study, more aggressive tumors may concurrently produce elevated levels of GDF15 and possess heightened metastatic potential. GDF15-GFRAL-driven CCN3⁺ SCs may act as a central hub linking pain and tumor metastasis.

GDF15 precisely binds to its receptor, GFRAL, to form the GDF15-GFRAL complex⁶⁰. Our results indicate that SCs express a large amount of GFRAL and that the interaction between GDF15 and GFRAL is significantly increased in patients with PDAC and PDAC mice. Furthermore, the knockdown of GDF15 in K8484 cells or GFRAL in SCs blocked the increase in the number of CCN3⁺ SCs and the fiber length of ND7/23 cells in vitro. Further, in vivo treatment with shGdf15 in cancer cells or intratumoral injection of AAV-CNP-shGfral prevented the increase in the number of CCN3⁺ SCs and nerve fiber length, and attenuated PDAC-induced chronic pain. In our PDAC model, local GDF15 signals through peripheral GFRAL on SCs to drive pain, likely without reaching cachexia-inducing systemic levels. This suggests its pro-nociceptive effect precedes central metabolic actions. We identify a peripheral pain pathway (GDF15/GFRAL/metabolic reprogramming/neurotrophic secretion) distinct from its central role in appetite regulation.

The GDF signaling pathway participates in various biological functions, including the inflammatory response and energy metabolism, by activating the PI3K-AKT pathway and MAPK cascades^{33,50,61–63}. Surprisingly, GO and metabolic pathway analyses of CCN3⁺ SCs indicated significant upregulation of the glycolytic pathway in these cells, and KEGG analysis revealed upregulation of the PI3K-AKT pathway. The PI3K-AKT pathway is an important signaling pathway that participates in various physiological and pathological functions by regulating biological processes^64–69. We found that PDAC significantly upregulated the key glycolytic enzyme PFKM in SCs; the gain/loss of function results further confirmed that the GDF15-GFRAL interaction regulated PFKM upregulation by activating the AKT pathway, facilitating CCN3⁺ SC formation and thus promoting pain. AKT inhibition may alleviate pain through multiple mechanisms beyond SCs, including modulation of neuronal excitability, suppression of neuroinflammation, and reduction of tumor burden^70–72. However, broad-spectrum AKT inhibitors often result in limited efficacy and significant toxicity. Our study identifies RUNX2 and PFKM as pain pathway-specific downstream effectors in SCs. Targeting these nodes may enable selective inhibition of pro-nociceptive reprogramming and represent a more precise therapeutic strategy for PDAC-associated pain.

Glycolysis in CCN3⁺ SCs promotes secretion of neurotrophic factors, such as NGF, and sensitizes sensory neurons. This is supported by the effectiveness of anti-NGF therapy in pancreatic cancer pain models²⁴. We propose that the efficacy of anti-NGF therapy may be mediated partially by blocking the tumor-induced GDF15/GFRAL/glycolysis signaling axis in CCN3⁺ SCs.

The runt-related transcription factor (RUNX) family comprises three closely related transcription factors: RUNX1, RUNX2, and RUNX3⁷³. RUNX2, a key transcription factor essential for chondrocyte and osteoblast differentiation and bone formation, is also detected in non-skeletal tissues and cells, such as the brain and T cells⁷⁴. Additionally, RUNX2 promotes astrocyte transformation and scar formation following spinal cord injury⁷⁵. Here, we identified RUNX2 as the most specific transcription factor regulating CCN3⁺ SC function. Furthermore, our results indicate that the AKT signaling pathway promotes RUNX2-mediated Pfkm transcription, enhances CCN3⁺ SC proliferation, and mediates chronic pain. This is consistent with reports that SC migration and axonal regeneration are inhibited after specific RUNX2 knockout following sciatic nerve injury⁷⁶. As RUNX2 may be stronger specificity for the Pfkm promoter than for other glycolytic enzymes (such as Hk1 and Hk2), the GDF15‑GFRAL signaling selectively affects PFKM. This should be tested in future studies by comparing the binding affinity of RUNX2 with the promoters of various glycolytic enzymes.

Although tumor burden increases (accompanied by elevated GDF15 levels) with tumor progression⁷⁷ and may contribute to tumor-associated pain, the analgesic effect of specifically ablating CCN3⁺ SCs, their temporal expansion relative to pain onset, and the neuron‑sensitizing effect of their conditioned medium collectively indicate CCN3⁺ SCs are a key driver of pain in PDAC mice.

Other mechanisms may coexist: (1) GDF15 may directly bind to and sensitize sensory nerves; (2) inflammation-inducible GDF15 may indicate the presence of inflammatory mediators, which initiate pain signaling; and (3) the acidic, lactate-rich PDAC microenvironment itself may produce pain by activating acid-sensing ion channels (ASICs)⁷⁸. With tumor progression, increased GDF15 release further induces CCN3⁺ SCs. CCN3⁺ SCs drive intratumoral sensory nerve growth and sensitization, thereby underpinning the algogenic effect of the acidic, lactate‑rich milieu.

Furthermore, CCN3⁺ SCs highly express chemokines (e.g., SPP1, FN1) capable of recruiting/polarizing monocytes/macrophages, which can release pro‑inflammatory cytokines sensitizing nociceptors^79,80. The GDF15‑GFRAL axis may also modulate the expression of key inflammatory cytokines (e.g., TNF‑α, CCL2) and matrix metalloproteinases (e.g., MMP2, MMP9) in SCs (data available upon request), potentially amplifying neuroinflammation and tissue remodeling in the PDAC microenvironment. In contrast, other inflammatory mediators such as IL-6, PGE2, CXCL10, and CXCL12 were not regulated by the GDF15-GFRAL pathway (data available upon request), suggesting that the pro-nociceptive effects of this axis are mediated through distinct mechanisms.

Our evidence confirms that GDF15-GFRAL signaling promotes CCN3⁺ SCs proliferation, ultimately driving tumor innervation and pain. However, this study has several limitations. First, the preference of CCN3⁺ SCs for promoting specific neuronal subtypes remains to be investigated. Second, although CCN3⁺ SCs enhance nociceptive hypersensitivity and provide a foundation for tumor neuroinvasion, their direct role in promoting tumor cell growth remains unclear. Third, SC subpopulations in PDAC likely represent distinct cell subsets that expand under different conditions within the tumor environment, but this was not explored, necessitating future lineage-tracing studies to establish the developmental relationships among SC states. The selectivity of the RUNX2-PFKM axis in SCs requires future validation by comparing RUNX2 binding affinity to promoters of various glycolytic enzymes. The measurement of single-cell glycolytic flux remains inferential and will require in vivo genetic or metabolic tracing approaches. Additionally, because surgical PDAC samples almost always come from symptomatic patients, the baseline for pain is uniformly high, which precludes effective correlation analysis.

Overall, this study demonstrates significant translational potential for clinical applications. We identify GDF15 as a potential biomarker of cancer-induced pain in PDAC, which may guide clinical pain management strategies. Our early interventions (week 2 post-tumor implantation, a time point when pain had already manifested) attenuated pain maintenance, while our late interventions (week 5) reversed established pain behaviors. These findings suggest that targeting this pathway may hold translational promise not only for early intervention but also for symptomatic relief in patients with established pain. Furthermore, the availability of clinically advanced GDF15-GFRAL-targeting antibodies (e.g., ponsegromab, visugromab)^81,82 provides an opportunity to evaluate their analgesic efficacy in future studies, which may ultimately contribute to improving the quality of life of these patients.

Taken together, cancer cell-derived GDF15 drives an expansion of CCN3⁺ SCs and induces glycolytic reprogramming via the GFRAL receptor. Mechanistically, GFRAL activation triggers the AKT–RUNX2 signaling cascade, leading to upregulation of the glycolytic enzyme PFKM in CCN3⁺ SCs. This metabolic reprogramming enhances tumor innervation and pain sensitization, thereby sustaining chronic pain.

Methods

This study received approval from the Ethics Committee of the Sun Yat-sen University Cancer Center (B2024-758-01). All animal experiments adhered to the National Institutes of Health guidelines for the ethical care and treatment of animals, and all experimental protocols were approved by the Animal Care and Use Committee of the Sun Yat-sen University Cancer Center (L025503202505011).

Sex as a biological variable

Our study examined male and female patients (120 PDAC patients, aged 24-83). Sex was determined from medical records (assigned, not self-report). The results did not differ significantly between the sexes. Gender was not considered in this study. Written informed consent was obtained from all participants. Both male and female C57BL/6 mice were used in this study. Based on prior research⁸³ and our experimental results, no significant sex-based differences were observed in PDAC-associated pain behavior in mice.

PDAC patient plasma and tissues

Peripheral blood samples were collected from 120 PDAC patients from Sun Yat-sen University Cancer Center with clinically assessed pain scores between 2020 and 2024 (Supplementary Table 1). Plasma was isolated for subsequent analysis. PDAC tissues were ethically procured from patients who were undergoing surgical intervention at the Sun Yat-sen University Cancer Center. Following surgical removal, these tissues were subjected to fixation using a 4% solution of paraformaldehyde. They were then dehydrated and embedded in paraffin wax. The specimens were then conserved at ambient temperature for subsequent analysis and research applications. Written informed consent was obtained from all participants for human plasma and tissues. The study received approval from the Ethics Committee of the Sun Yat-sen University Cancer Center (B2024-758-01).

Clinical pain scores

Clinical pain was assessed using both the numerical rating scale (NRS) and the visual analog scale⁸⁴. For the NRS, patients were asked to rate their average pain intensity over the past 24 hours by selecting a whole number from 0 (no pain) to 10 (worst imaginable pain). For the VAS, patients marked their pain level on a 100-mm horizontal line anchored by “no pain” at 0 mm and “worst pain imaginable” at 100 mm; the distance from the left anchor was measured in millimeters. Based on the scores, patients were stratified into two groups: the Pain group (NRS > 0 or VAS > 0 mm) and the No‑Pain group (NRS = 0 and VAS = 0 mm).

Animals

C57BL/6 (C57BL/6 J) mice (6 weeks old), with a weight range of 20 to 25 grams, were ethically obtained from the Guangdong Medical Laboratory Center in Guangzhou, China. These animals were domiciled within a controlled environmental chamber that maintained a stable thermal and hygrometric milieu (22–25 °C and 40–60% humidity), simulating a natural 12-hour light/dark photoperiod. Mice were fed standard rodent chow (XB-SL5, XB-BIO), and the mice had continuous access to water to fulfill their physiological requirements without restriction. This research adhered to the National Institutes of Health guidelines for the ethical care and treatment of animals, and all experimental protocols were approved by the Animal Care and Use Committee of the Sun Yat-sen University Cancer Center (L025503202505011).

Chronic pain model induced by PDAC

C57BL/6 mice were administered anesthesia via intraperitoneal injection of a 1% solution of sodium pentobarbital (volume range: 180-200 μl per animal). Following induction of anesthesia, the mice were depilated, and a laparotomy was performed to gain access to the abdominal cavity, thereby exposing the pancreas. Mouse K8484 pancreatic cancer cells, cultured to optimal confluence, were harvested and suspended in an equal volume ratio (1:1) with Matrigel™ matrix (356234, Corning) to form a homogenous mixture. This cell-matrix suspension was then locally injected into the pancreatic tissue of the C57BL/6 mice at a concentration of 7.5 × 10^4 cells per 100 μl per animal. As a control, physiological saline was injected in an equivalent volume. The abdominal incision was subsequently closed, the skin was sutured, and the surgical site was disinfected. Postoperatively, the mice were returned to their standard husbandry conditions for continued observation and feeding. All animal experiments were approved by the Animal Care and Use Committee of Sun Yat-sen University Cancer Center (approval number L025503202505011). The maximal permitted tumor size was 2 cm in diameter. This limit was not exceeded in the experimental animals. Mice were euthanized when tumor diameter reached 2 cm or if they showed signs of severe distress (e.g., weight loss >20%, lethargy, inability to reach food or water). No animals reached these endpoints during the study.

Behavior tests

Von-Frey assay

The Von-Frey filament test, a widely recognized method for assessing mechanical nociception⁸⁵, was employed to quantify the tactile allodynia in C57BL/6 mice. Specifically, the mice were positioned within a 20 cm × 20 cm enclosure constructed from transparent acrylic, featuring a metallic grid flooring, elevated 20 cm from the laboratory bench surface. A series of calibrated Von Frey monofilaments, ranging from 0.008 to 0.16 grams, was applied perpendicularly to the midsection of the left ventral quadrant for a duration of 3 seconds, exerting a force sufficient to elicit filament deflection. Each subject underwent a stimulation protocol with each filament for 5 times, with an inter-stimulus interval of no less than 3 minutes to prevent sensitization. The occurrence of hunchback posturing or claw withdrawal was documented as indicative of a positive nociceptive response. The composite response score was calculated by multiplying the total number of affirmative responses by a factor of 20. Furthermore, the response frequency at the 0.16 g threshold was ascertained by determining the ratio of affirmative responses out of the total stimuli administered. The murine Von-Frey assay was systematically monitored at weekly intervals post-induction of a pancreatic cancer-mediated chronic pain model, spanning from the initial week through to the fifth week of observation.

Hunching score

The hunching score serves as a quantitative tool for assessing the intensity of spontaneous nociception in murine subjects⁸⁶. This method involves the visual evaluation of murine spinal posture under ambient lighting, with particular attention to the curvature of the spine during both stationary and ambulatory states. The scoring system categorizes the degree of hunchback into a five-tiered scale based on the severity of spinal curvature: Neutral alignment (0 points): Characterized by an absence of spinal curvature, with the murine spine maintaining a straight configuration. Subtle curvature (1 point): A slight spinal deviation detectable at specific vantage points, yet not overtly apparent. Moderate curvature (2 points): A pronounced spinal deviation discernible from various perspectives, without impeding the execution of fundamental locomotive activities. Pronounced curvature (3 points): A marked spinal deviation, potentially accompanied by severe spinal deformities, which may compromise the murine capacity for movement. Profound curvature (4 points): An extreme spinal deviation, likely accompanied by severe spinal deformities, significantly impairing the murine mobility. All hunching scores in this study were assessed in the fifth week following the establishment of the PDAC mouse model.

Open field test

Following 30 minutes of habituation to the testing room, individual mice were gently placed in the corner of a square arena (40 × 40 × 40 cm) under standardized illumination (50 lux). Spontaneous locomotor activity was recorded for 5 min using an automated tracking system. Behavioral parameters, including total distance traveled, velocity, center zone duration (20 × 20 cm central area), and rearing frequency, were quantified. The apparatus was thoroughly cleaned with 70% ethanol between trials to eliminate olfactory cues. All tests were conducted during the light phase of the diurnal cycle to minimize circadian variability. All open-field tests in this study were conducted in the fifth week following the establishment of the PDAC mouse model.

Hot plate test

The hot plate test was performed to assess thermal nociception. Mice were acclimated to the testing room for at least 30 minutes. The metal plate surface was maintained at 53 °C. Each mouse was placed individually in the center of the plate, and the latency to the first nocifensive behavior (hind‑paw lick, shake, or jump) was recorded with a 30-second cut‑off to prevent tissue damage. The plate was cleaned with dilute ethanol between trials. Each mouse was tested twice with a 15 min interval, and the average latency was calculated. All tests were conducted by an experimenter blinded to the treatment groups. The hot plate test was systematically monitored at weekly intervals post-induction of a pancreatic cancer-mediated chronic pain model, spanning from the initial week through to the fifth week of observation.

Gabapentin analgesic treatment

A single intraperitoneal injection of the clinical analgesic gabapentin (20 mg/kg in saline) or vehicle (saline) was administered to PDAC-bearing mice at week 5 post-model establishment, after stable pain hypersensitivity was confirmed. Behavioral assessments, including Von-Frey assay, hunching score, hot plate test, and open field test, were assessed before injection (baseline) and at 1, 2 and 4 hours post-injection. All behavioral tests were performed by an experimenter blinded to the treatment groups.

Recombinant adeno-associated virus or inhibitor administration

AAV-CNP (Schwann cell‑specific promoter)-mediated short hairpin RNAs (shRNAs) specifically designed to target the Schwann cell Ccn3, Gfral, Pfkm and Runx2, as well as non-targeting control shRNAs (shNCs), were custom-synthesized by HanBio. Additionally, recombinant AAV‑CNP vectors (including AAV‑CNP‑Gfral, AAV‑CNP‑Pfkm, and AAV‑CNP‑Cre encoding Cre recombinase), the CCN3‑promoter‑controlled DIO‑DTA vector (AAV‑CCN3‑promoter‑DIO‑DTA), and the vector encoding enhanced green fluorescent protein (EGFP) (designated as AAV-CNP-EGFP) were all procured from HanBio. Intratumoral administration of recombinant adeno-associated viral vectors (20 μl per 100 mm^3 tissue volume, with 2-3 discrete injection sites) was initiated during post-induction of a PDAC-mediated chronic pain model. For early therapeutic intervention, key interventions (AAV-CNP-shCcn3, AAV-CNP-shGfral, AAV-CNP-shRunx2, AAV-CNP-shPfkm) were delivered at Week 2 post‑tumor implantation, a time point at which significant pain behaviors had already developed. As a late‑stage therapeutic intervention at the peak of pain, we supplemented key experiments by administering AAV‑CNP‑shGfral or AAV‑CNP‑shPfkm at Week 5, when pain reached its maximum in the model. AAV‑CNP‑Cre and AAV‑CCN3‑promoter‑DIO‑DTA were co‑injected intratumorally at a 1:1 volume ratio. Following the localized viral delivery, the mice were maintained under standard husbandry conditions for continued nourishment and observation. Regarding the transduction efficiency of AAV‑CNP, we calculated the ratio of the EGFP fluorescence signal area to the S100β fluorescence signal area as a metric for viral transduction efficiency. For orthotopic tumors reaching 100-150 mm³, AKTi1/2 (5 mg/kg in DMSO/PBS) (612847-09-3, MedchemExpress) or ML251 (1 mg/kg in DMSO/PBS) (1486482-16-9, MedchemExpress) was injected via needle inserted at a 45° angle into the tumor core. Treatments were administered every 72 h for 2 weeks.

Mouse tissue collection

Mice were subjected to anesthesia using a 1% solution of sodium pentobarbital at a dosage of 100 mg/kg to ensure deep sedation. Subsequently, euthanasia was performed via cervical dislocation. Following euthanasia, the abdominal compartment was surgically opened to gain access to the pancreas and pancreatic neoplasms. The pancreatic tissues, inclusive of tumor samples, were meticulously harvested and subsequently preserved in a 4% paraformaldehyde solution or in liquid nitrogen.

Cell culture and transfection

The K8484 cell line was provided by the laboratory of Prof. Peng Huang. The KPC cell line was provided by the laboratory of Prof. Jian Zheng. The RSC96 cell line was purchased from ATCC (CRL-2765), and the ND7/23 cell line was purchased from Bioesn (BES2091OC). Cells were routinely maintained in Dulbecco’s Modified Eagle Medium (DMEM; Gibco), enriched with 5% equine serum and 5% fetal bovine serum (both from Gibco). Cultures were incubated under a humidified atmosphere containing 5% CO2 at a controlled temperature of 37 °C. AKTi1/2 (15 μg/mL) / ML251 (10 μg/mL) was added for 24 hours under standard conditions (37 °C, 5% CO₂) for subsequent experiments. Cells were serum-starved for 6 hours and then treated with recombinant mouse GDF15 protein (50 ng/mL) in serum-free medium at 37 °C for 30 minutes for subsequent analysis. Small interfering RNAs (siRNAs) targeting Gdf15, Gfral, Akt, Pfkm and Runx2 genes, along with non-targeting control siRNAs (siNC), were custom-synthesized by TsingkeBio. Transient transfection of these siRNAs was executed utilizing the RiboFECT CP Transfection Kit (RiboBio), adhering to the manufacturer’s recommended protocol. For the multiple siRNA sequences targeting the same gene, we validated their knockdown efficiency at both the mRNA and protein levels, and selected the siRNA sequence with the highest knockdown efficiency for subsequent functional experiments. Wild-type (WT) and signaling-deficient mutant (Mut) Ccn3 constructs were synthesized by TsingKe Biological Technology. Plasmid transfection was performed using Lipofectamine™ 3000 according to the manufacturer’s protocol. Lentiviral vector-mediated short hairpin RNAs (shRNAs) specifically designed to target the Gdf15 gene, as well as non-targeting control shRNAs (shNCs), were custom-synthesized by HanBio. The subsequent infection of these shRNAs with lentiviral vectors was conducted employing the Lentiviral Infection Assay Kit (HanBio), strictly adhering to the protocol outlined by the manufacturer.

CCK-8 cell proliferation assay

A CCK-8 assay was performed to assess the effect of GDF15 knockdown on PDAC cell proliferation. Stable shGdf15 or control PDAC cells were seeded in 96‑well plates at a density of 2000 cells per well in 100 µL of complete medium and pre‑cultured for 24 h. At designated time points (0, 24, 48, and 72 h), 10 µL of CCK‑8 reagent was added to each well, followed by incubation for 2 h at 37°C. Absorbance was measured at 450 nm using a microplate reader. Background absorbance from medium‑only wells was subtracted. The experiment was performed with at least three replicates per group, and data were presented as the mean ±SEM of the absorbance values over time.

Quantitative real-time PCR (qPCR)

Total cellular or tissue RNA was isolated employing TRIzol® reagent (Invitrogen), followed by reverse transcription to synthesize complementary DNA (cDNA). qPCR assays were executed on a Real-Time Detection System utilizing Color SYBR Green qPCR Mix (EZBioscience). Primer sequences are provided in Supplementary Table 2. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) served as an endogenous reference gene for data normalization. Relative mRNA expression levels were determined employing the comparative CT (2 − ΔΔCt) method.

Western blot assay

Cellular and tissue samples obtained from culture were subjected to lysis in a chilled lysis buffer comprising 10 mM Tris-HCl, 5 mM EGTA, 2 mM MgCl2, 1 mM dithiothreitol (DTT), 1 mM phenylmethylsulfonyl fluoride (PMSF), and 40 μM leupeptin. Post-centrifugation at 4 °C for 10 minutes at 12,000 g, the supernatant was collected and denatured at 100 °C for 5 minutes. Aliquots of total protein extracts (20 μg per well) were resolved via 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and electroblotted onto a polyvinylidene fluoride (PVDF) membrane. The membrane was blocked with 5% bovine serum albumin (BSA) for 1 hour at room temperature, followed by an overnight incubation at 4 °C with the following respective primary antibodies: GDF15 (1:1000; TD6006, Abmart), CCN3 (1:1000; PS08472, Abmart), S100β (1:1000; MG237235, Abmart), PFKM (1:1000; PA4490, Abmart), GFRAL (1:1000; PU983152, Abmart), AKT (1:1000; T55561, Abmart), Phospho-AKT (Ser473) (1:1000; T40067, Abmart), PI3K-p85 (1:1000; T40115, Abmart), Phospho-PI3K-p85 (1:1000; T40116, Abmart), RUNX2 (1:1000; TA5186, Abmart), Phospho-RUNX2 (1:1000; TA7379, Abmart), Phospho-S6 (1:1000; 80130-2-RR, Proteintech), S6 (1:1000; 66886-1-Ig, Proteintech), CyclinD1 (1:1000; 26939-1-AP, Proteintech), GAPDH (1:5000; 60004-1-Ig, Proteintech), and β-tubulin (1:5000; AC008, ABclonal). On the subsequent day, the membranes underwent a washing protocol followed by incubation with either horseradish peroxidase (HRP)-conjugated goat anti-rabbit immunoglobulin G secondary antibody (dilution 1:10,000; SA00001-2, Proteintech) or HRP-conjugated goat anti-mouse IgG secondary antibody (dilution 1:10,000; 7076S, Cell Signaling). The detection of immune complexes was facilitated through chemiluminescent reactions, and the band intensities were quantified employing ImageJ software. Relative protein expression levels were normalized against the corresponding GAPDH or β-tubulin housekeeping protein loading controls. Uncropped scans of all blots and gels are provided in the Source Data file.

Enzyme-linked immunosorbent assay

Plasma samples from PDAC patients and conditioned medium from SCs co-cultured with PDAC/tumor cells were collected and stored at -80 °C. Commercial ELISA kits for human GDF15 and rat NGF, BDNF, and IGFBP5 were used according to the manufacturer’s instructions. Standards and appropriately diluted samples (plasma typically diluted 1:10, conditioned medium dilution optimized in pilot experiments) were added to pre-coated 96-well plates (100 µL/well) and incubated for 2 hours at room temperature. After washing, biotinylated detection antibodies were added (incubated for 1 hour), followed by streptavidin-HRP (incubated for 30 minutes). Following the final wash, TMB substrate was added, the reaction was stopped with stop solution, and absorbance was measured at 450 nm. Concentrations were calculated based on the standard curve. All samples were run in duplicate, and the operator was blinded to sample information.

Isolating and culturing primary mouse dorsal root ganglion neurons

Following euthanasia, the thoracic and lumbar spinal column is rapidly dissected and placed in ice-cold DMEM/F12. DRGs are removed from the vertebral foramina under a dissection microscope and collected. Ganglia are enzymatically digested in a solution of papain and collagenase at 37 °C for 30 minutes. Digestion is stopped with serum-containing medium. After centrifugation, the tissue is gently triturated in neuronal complete medium to obtain a single-cell suspension. The suspension is plated onto culture dishes pre-coated with poly-D-lysine and laminin. After 4 hours, the medium is fully replaced with fresh neuronal complete medium. Neurons are typically maintained in a 37 °C, 5% CO₂ incubator and are ready for experiments such as electrophysiology after 3 days in culture.

Electrophysiological recording

Electrophysiological recordings were performed on primary DRG neurons from PDAC-bearing mice and control mice, with or without pretreatment with CCN3⁺ Schwann cell-conditioned medium. Whole-cell patch-clamp recordings were conducted under current-clamp mode. Rheobase was determined by applying a series of depolarizing current steps (0 pA start, 10 pA increments, 500 ms duration) and identifying the minimal current required to elicit the first action potential (AP). AP frequency was assessed by applying a fixed suprathreshold current stimulus (1.5×Rheobase or a standardized 600 pA, 500ms duration) and counting the number of APs elicited. Data were acquired using pCLAMP software, and analysis was performed in a blinded manner.

Cell co-culture

Co-culture of K8484 cells and RSC96 cells

K8484 or RSC96 cells (1×10^5 cells/well) were seeded in Dulbecco’s Modified Eagle Medium (DMEM; Gibco), enriched with 5% fetal bovine serum (from Gibco). Cells were allowed to adhere for 24 hours under standard culture conditions (37 °C, 5% CO₂), followed by medium replacement to remove non-adherent cells. Cell supernatant collection occurred at 24 hours intervals for subsequent analysis. RSC96 cells (1 × 10^5 cells/well) were plated in a culture dish in advance and incubated for 24 hours to allow cell attachment. After removing the original culture medium, add the collected cell supernatant to the dish, and continue incubation for another 24 hours. Finally, harvest the cells for subsequent experiments, such as qPCR, western blot assay, flow cytometry assay, EdU proliferation assay (C6044M, UElandy), and glycolytic rate assay. For transwell assays, RSC96 cells were seeded in 8 μm pore inserts placed above the cell supernatant. After 24 hours of additional incubation, RSC96 cells in the Transwell inserts were stained with 1% crystal violet solution, and cell counting was performed under an optical microscope.

Co-culture of K8484 cells, RSC96 cells and ND7/23 cells

After 24 hours of co-culturing RSC96 cells with K8484 cell supernatant, the resulting conditioned medium was collected. ND7/23 cells (1 × 10^5 cells/well) were plated in a culture dish and incubated for 24 hours to allow attachment. The original culture medium was then removed and replaced with conditioned medium, followed by an additional 24-hour incubation period. The co-cultured ND7/23 cells were stained with 1% crystal violet solution, and axonal lengths were quantified under an optical microscope.

Flow cytometry analysis

Cell samples (RSC96 cells after co-culture) were harvested, washed with PBS, and stained at 4 °C for 30 minutes in the dark following respective primary antibodies: S100β (1:200; MG237235, Abmart), CCN3 (1:200; PS08472, Abmart). After fixation (4% paraformaldehyde, 15 minutes) and following the washing steps, cells were incubated with species-specific secondary antibodies conjugated to Alexa Fluor 488 (mouse or rabbit, 1:200; L3016 and L3036, SAB), DyLight 594 (mouse or rabbit, 1:200; A23420 and A23410, Abbkine) at ambient temperature for a duration of 1 hour in the dark. Cells were resuspended in FACS buffer. Data acquisition was performed using a CytoFLEX flow cytometer, with ≥10,000 events recorded per sample. We performed sequential gating steps on the flow cytometer, starting by selecting the main intact cell population from the collected cells (FSC‑A/SSC‑A), then isolating non‑aggregated single cells (FSC‑A/FSC‑H), and finally identifying the target cell populations (S100β⁺CCN3⁺ for CCN3⁺ SCs) from among these. Compensation was applied using single-stained controls, and analysis was conducted with CytoExpert software.

Fluorescence-activated cell sorting

Fresh tumor tissue was dissociated into a single‑cell suspension using a multi‑enzyme digestion cocktail (collagenase IV, dispase, and DNase I). Cells were stained at 4 °C for 30 minutes in the dark following respective primary antibodies. Following the washing steps, cells were resuspended in FACS buffer. We applied a sequential gating strategy, beginning with selection of the main intact cell population (FSC‑A/SSC‑A), progressing to isolation of non‑aggregated single cells (FSC‑A/FSC‑H), then to gating of the broad SC population (CD56⁺ cells), and culminating in identification of the target cell populations (S100β⁺p75NTR⁺ for SCs). Under sterile conditions, the S100β⁺/p75NTR⁺ cell population was isolated using a high‑speed cell sorter (BD FACSAria™ III).

Immunofluorescence

Paraffin-embedded tissue sections, derived from both human and murine samples, were placed in a 65 °C oven for 3 hours. Subsequently, the sections underwent a graded series of deparaffinization steps, involving sequential immersion in xylene (three changes), followed by anhydrous ethanol (two changes), and ethanol solutions of decreasing concentration (90%, 80%, and 70%). The sections were then rinsed in a phosphate-buffered saline (PBS) solution. Post-deparaffinization, the sections were immersed in a citrate buffer and subjected to high-temperature, high-pressure antigen retrieval treatment for 5 minutes, after which they were allowed to equilibrate to room temperature. Tissue sections were pre-incubated in a blocking solution consisting of PBS supplemented with 5% normal goat serum at ambient temperature for a period of 1 hour. Subsequently, the sections were subjected to an overnight incubation at 4 °C with the following respective primary antibodies: GDF15 (1:200; TD6006, Abmart), CCN3 (1:200; PS08472, Abmart), S100β (1:200; MG237235, Abmart), PFKM (1:200; PA4490, Abmart), GFRAL (1:200; PU983152, Abmart), TRPV1 (29223, Signalway Antibody), CGRP (TU397291, Abmart), p-AKT (80455-1-RR, Proteintech), SOX10 (1:200; T55826S, Abmart), GAP43 (1:200; T55634S, Abmart). Following the washing steps, tissue sections were incubated with species-specific secondary antibodies conjugated to Alexa Fluor 488 (mouse or rabbit, 1:200; L3016 and L3036, SAB), DyLight 594 (mouse or rabbit, 1:200; A23420 and A23410, Abbkine) or ABflo 647 (rabbit, 1:200; AS060, ABclonal) at ambient temperature for a duration of 1 hour in the dark. Subsequently, the sections were mounted by a DAPI-containing sealing tablet (S2110, Solarbio) for subsequent fluorescence microscopy analysis. Then, micrographs of the stained tissue sections were acquired utilizing a Zeiss LSM 980 confocal laser scanning microscope. We performed semi-automated, area-based quantification using ImageJ. This method measures the total area occupied by the positive immunofluorescence signal within a defined field of view, which is a standard and objective approach for dense or interconnected cellular networks (e.g., nerve fibers, clusters of stained cells)⁸⁷. Positive cells were identified using a uniform intensity threshold that was set based on negative control (no primary antibody) staining for each experiment and then applied consistently to all experimental images within that batch. A specific threshold was defined as the mean fluorescence intensity of the negative control plus two standard deviations. For analyses involving nerve fibers or cell counts in tissue volumes (mentioned in legends), z-stacks were indeed acquired. Quantification was performed on maximum intensity projections for 2D analyses.

Multiplex immunofluorescence

For multiplex immunofluorescence staining using tyramide signal amplification (TSA) on formalin-fixed paraffin-embedded sections, slides are first baked, deparaffinized, and rehydrated, followed by heat-induced antigen retrieval in citrate buffer. After blocking, the first primary antibody is applied and incubated overnight at 4 °C: PFKM (1:200; PA4490, Abmart), S100β (1:200; MG237235, Abmart), GFRAL (1:200; PU983152, Abmart), CCN3 (1:200; PS08472, Abmart), FOSB (1:200; 49091, Signalway Antibody), ANGPTL7 (1:200; 10396-1-AP, Proteintech), MBP (1:200; 10458-1-AP, Proteintech), p75NTR (1:200; 55014-1-AP, Proteintech). Sections are then incubated with an HRP-conjugated secondary antibody, followed by TSA fluorophore incubation to visualize the target. The antibody–TSA complex is removed by microwave treatment in retrieval buffer before sequentially applying additional primary antibodies with distinct TSA fluorophores. Finally, nuclei are counterstained with DAPI, and slides are mounted for multispectral imaging.

Glycolytic rate assay

RSC96 cells were seeded in XF96 microplates (2×10⁴/well) and cultured overnight at 37 °C/5% CO₂. Prior to assay, cells were washed twice and incubated in substrate-limited assay medium (pH 7.4) for 1 hour at 37 °C (non-CO₂). Real-time extracellular acidification rate (ECAR) was measured using a Seahorse XFe Analyzer with sequential injections of: 0.5 μM Rotenone and Antimycin A (Rot/AA), and 50 mM 2-DG. Real-time extracellular acidification rate (ECAR) and oxygen consumption rate (OCR) were measured using a Seahorse XFe Analyzer with sequential injections of: 0.5 μM Rotenone and Antimycin A (Rot/AA), and 50 mM 2-DG. The glycolytic PER was calculated based on measurements of ECAR and OCR.

Chromatin immunoprecipitation assay

Cell cultures were treated with 1% formaldehyde to fix protein-DNA interactions. Glycine was added to halt the crosslinking process. Subsequently, cells were lysed, and chromatin was fragmented to 200–1000 bp using sonication. The fragmented chromatin was subjected to immunoprecipitation with specific antibodies: RUNX2 (2 μg, TA5186, Abmart), followed by capture using protein A/G magnetic beads. After thorough washing to remove non-specific interactions, the antibody-chromatin complexes were eluted, and crosslinks were reversed. DNA was extracted and purified, then analyzed via qPCR to assess the relative enrichment of target sequences compared to input and a reference region.

Co-immunoprecipitation assay

Cells were lysed in a buffer containing 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, and protease inhibitors. After incubation on ice and centrifugation, the supernatant was incubated overnight at 4 °C with a specific primary antibody: AKT (2 μg, T55561, Abmart). Protein A/G agarose beads were added and incubated for 2 hours to capture the complexes. The beads were washed with lysis buffer, and the complexes were eluted using a low pH buffer. The eluted proteins were resolved by SDS-PAGE, transferred to a PVDF membrane, blocked with 5% BSA for 1 hour at room temperature, followed by an overnight incubation at 4 °C with the following respective primary antibodies: Phospho-RUNX2 (1:1000; TA7379, Abmart), AKT (1:1000; T55561, Abmart), GAPDH (1:5000; 60004-1-Ig, Proteintech), and detected using HRP-conjugated secondary antibodies and chemiluminescence. Uncropped scans of all blots and gels are provided in the Source Data file.

Single-cell sequencing data analysis

The single-cell dataset (GSE212966 and GSE278688) was sourced from the NCBI public database, comprising samples of PDAC tissues and paired adjacent normal tissues. Following stringent quality control, batch correction, and integration of the raw data, we performed unsupervised clustering and annotated cell types using canonical marker genes (S100B, SOX10, GAP43), successfully identifying the Schwann cell population from all cells. Subclustering of this population revealed three functionally distinct Schwann cell subsets. Among them, one subset was specifically defined as CCN3⁺ SCs due to its high expression of CCN3 and a characteristic gene set. Subsequent analyses on these Schwann cell subsets included cellular trajectory inference, cell-cell communication analysis using CellChat, functional enrichment analysis (GO and KEGG), and transcriptional regulatory network reconstruction with SCENIC. The regulon specificity score (RSS) is a metric calculated by the SCENIC workflow to quantify the uniqueness of a transcription factor’s activity pattern across different cell clusters. An RSS value ranges from 0 to 1. A score closer to 1 indicates that the activity of that transcription factor’s target gene set (its “regulon”) is highly specific to a particular cell cluster, meaning it is uniquely active in that cluster compared to all others in the dataset. Subsequent single-cell data analysis was performed using the R programming language (version 4.3.2).

Statistics & Reproducibility

All statistical analyses were conducted using GraphPad Prism software (version 9.0). No data were excluded from the analyses. Data are presented as mean ± standard error of the mean (SEM). Student’s t-test was used when comparing two groups, and one-way analysis of variance ANOVA was used for comparisons involving more than three groups. If significant differences were observed, then Tukey’s post hoc test was applied to pairwise comparisons. For behavioral tests, a two-way repeated measures ANOVA was used. If significant differences were detected, then Šidák’s post hoc test was used to compare values at different time points. The Mann–Whitney U test was used for non‑continuous variables. Detailed statistical information (effect sizes (e.g., Cohen’s d), confidence intervals, and corrections for multiple comparisons) is provided in the figure legends. The “replicates” for in vitro experiments refer to independent experiments, and animal studies now report the number of animals per group. Differences were considered statistically significant at P < 0.05. No statistical method was used to predetermine sample size. While a formal power analysis was not conducted, the sample size was determined based on previous studies and preliminary experiments in this research area. The experiments were randomized. The investigators were blinded to allocation during experiments and outcome assessment.

Reporting summary

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

Author contributions

J.X. and G.C. conceived and designed the study. G.C., W.L., M.L. and Q.Y. performed most of the experiments. Y.X., Y.Z., X.G., Y.Y., X.Y. and X.L. assisted in carrying out the experiments. G.C., W.L., M.L. and Q.Y. contributed to data analysis. G.C. and W.L. prepared and revised the figures and drafted the manuscript. J.X. and W.X. supervised the research. All authors read and approved the final manuscript.

Peer review

Peer review information

Nature Communications thanks Jörg Wischhusen, who co-reviewed with Vincent Thiemann, Moran Amit, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The single-cell datasets (GSE212966 and GSE278688) are sourced from the NCBI public database. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72932-5.