SDC1+ CAFs secreting CTGF drive tumour metastasis via FGFR3 signalling in cancers

Abstract

Background

Cancer-associated fibroblasts (CAFs) are key stromal components of the tumour microenvironment (TME) that profoundly influence tumour progression. However, CAFs exhibit pronounced phenotypic and functional heterogeneity, and whether conserved CAF subtypes with shared functional hallmarks exist across different cancer types remains unclear.

Objective

We sought to uncover universal CAF subtypes that transcend tumour origins, defining their core molecular signatures and pro-tumorigenic functions within the TME.

Design

We constructed a pan-cancer CAF atlas through single-cell transcriptomic analysis of 554 specimens across 14 cancer types. To validate the findings, we performed further functional analyses, including in vitro migration and invasion assays, in vivo lymphatic metastasis models and mechanistic studies focusing on candidate signalling pathways.

Results

We identified a conserved syndecan 1 (SDC1) ⁺ CAF subset associated with advanced tumour stage and poor outcomes. These CAFs enhanced tumour cell migration and invasion in vitro and promoted lymphatic metastasis in vivo. This effect is mediated through connective tissue growth factor (CTGF) secretion, which activates fibroblast growth factor receptor 3 (FGFR3) signalling in tumour cells to induce epithelial-mesenchymal transition (EMT). Blocking CTGF or FGFR3 signalling abrogated these effects. We also found that kruppel like factor 6 (KLF6) directly regulates CTGF in SDC1⁺ CAFs, establishing a complete KLF6-CTGF-FGFR3 metastatic axis.

Conclusions

Our study establishes SDC1⁺ CAFs as a universal, metastasis-promoting CAF subset across multiple cancer types and uncovers a novel KLF6-CTGF-FGFR3 axis that drives EMT and tumour dissemination. These findings provide mechanistic insight into CAF-tumour cell crosstalk and highlight actionable stromal targets for anti-metastatic therapies across diverse malignancies.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Cancer-associated fibroblasts (CAFs) play essential roles in tumour progression.

• CAFs exhibit substantial phenotypic and functional heterogeneity.

• Recent pan-cancer bioinformatic analyses have identified common transcriptional programmes and spatial features of CAFs, with limited experimental validation of their functions and mechanistic pathways.

WHAT THIS STUDY ADDS

• Identifies a conserved, metastasis-promoting CAF subtype (SDC1⁺ CAFs) that is present across multiple cancer types and is linked to poor patient outcomes.

• Uncovers a novel signalling axis (KLF6-CTGF-FGFR3) by which SDC1⁺ CAFs drive epithelial-mesenchymal transition and metastasis in tumour cells.

• Provides functional evidence that targeting this stromal axis (CTGF or FGFR3) can effectively abrogate pro-metastatic effects.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study establishes SDC1⁺ CAFs as a stromal subset that could be targeted for therapy with significant clinical potential. Unlike previous approaches that target tumour-specific CAFs and may pose off-target risks, focusing on selective inhibition of the KLF6-CTGF-FGFR3 pathway in SDC1⁺ CAFs presents an innovative method for disrupting stromal support, which may improve clinical outcomes across multiple solid tumours.

Introduction

Tumour microenvironment (TME) is a dynamic and heterogeneous ecosystem comprising malignant cells, stromal components, immune infiltrates and vascular networks.¹ Among the non-malignant components, cancer-associated fibroblasts (CAFs) have emerged as pivotal orchestrators of tumour progression through their roles in extracellular matrix (ECM) remodelling, angiogenesis and immune modulation.^{2 3} CAFs exhibit both tumour-promoting and tumour-suppressive functions depending on cancer type and context. For instance, pro-tumourigenic roles have been reported for antigen-presenting CAFs (apCAFs) in pancreatic cancer⁴, senescent CAFs in breast cancer⁵ and THBS2⁺ CAFs in colorectal cancer⁶, whereas tumour-suppressive effects have been observed for apCAFs in gastric cancer⁷ and α-SMA⁺ myofibroblastic CAFs (myCAFs) in pancreatic cancer.⁸ Notably, the functional plasticity of CAF subtypes is exemplified by apCAFs, which promote immune evasion in pancreatic cancer by inducing regulatory T cells⁴, yet enhance anti-tumour immunity in gastric cancer by activating T cells and polarising macrophages towards a pro-inflammatory phenotype.⁷ These observations highlight the context-dependent roles, phenotypic plasticity and functional heterogeneity of CAF populations.^9–13

Recent advances in single-cell RNA sequencing (scRNA-seq) technologies have provided the high-resolution insights into CAF heterogeneity, leading to the identification of multiple distinct subtypes, including myCAFs¹⁴, inflammatory CAFs¹⁵ and apCAFs.¹⁶ These subtypes exhibit unique molecular signatures, spatial distributions and differential impacts on tumour growth, immune evasion and therapeutic response.^17–20 Despite these progresses, it remains unclear whether conserved CAF subtypes or functional hallmarks are shared across diverse tumour types. Emerging pan-cancer scRNA-seq studies have begun to address this gap, revealing shared transcriptional programmes and spatial features of CAFs across malignancies.^{21 22} For example, Luo et al. identified endothelial-to-mesenchymal transition as a conserved phenomenon linked to pro-angiogenic CAF phenotypes and poor survival outcomes in multiple cancers.²³ Similarly, Liu et al. leveraged spatial multi-omics to uncover conserved CAFs spatial niches and cellular neighbourhoods across different malignancies.²⁴ However, these findings have largely derived from computational inferences, with limited experimental validation of CAF functions, intercellular interactions and mechanistic pathways.

In this study, we performed a comprehensive pan-cancer single-cell transcriptomic analysis of 554 tumour specimens spanning 14 distinct cancer types and identified SDC1⁺ CAFs as a novel and universal CAF subtype with robust pro-metastatic potential. Using in vitro co-culture assays and in vivo metastasis models, we demonstrated that SDC1⁺ CAFs significantly enhance tumour invasion and lymphatic dissemination. Mechanistic investigations revealed a previously uncharacterised KLF6-CTGF-FGFR3 signalling axis in SDC1⁺ CAFs that drives epithelial-mesenchymal transition (EMT) in adjacent tumour cells, thereby promoting metastatic progression. These findings establish SDC1⁺ CAFs as a clinically relevant and broadly conserved CAF subset across multiple cancers and reveal actionable stromal signalling pathways with therapeutic potential in combating cancer progression.

Materials and methods

See the online supplemental materials for full methodological details.

Results

Pan-cancer single-cell transcriptome analysis identifies pro-tumorigenic SDC1⁺ CAF subtype

To investigate the heterogeneous features of CAFs across TME in diverse cancer types, we integrated 27 single-cell transcriptomic datasets comprising 554 samples from 14 distinct cancer types, including both primary tumours and adjacent normal tissues (figure 1A; online supplemental figure S1A and B). After stringent quality control, we constructed a high-resolution single-cell atlas encompassing 2 546 394 cells, which were annotated into major cell types based on canonical lineage-specific markers (online supplemental figure S1C and S2). Among all major cell types, CAFs exhibited the highest number of differentially expressed genes between tumour and adjacent normal tissues (online supplemental figure S3A). ROGUE index analysis revealed significantly lower scores for CAFs within tumour tissues compared with fibroblasts in adjacent normal tissues and other cell types (online supplemental figure S3B), indicating elevated intratumoral heterogeneity within the CAF compartment. Moreover, ligand-receptor interaction analysis demonstrated that CAFs were involved in the most extensive intercellular communications of any major cell type (online supplemental figure S3C), underscoring their central role in TME regulation.

Figure 1. Pan-cancer single-cell analysis identifies SDC1⁺ CAFs as a tumour-enriched, transcriptionally distinct CAF subpopulation. (A) Schematic diagram of the study workflow (created using https://biogdp.com ⁵⁴). (B) UMAP visualisation of CAF subpopulation across multiple cancer types. The region enclosed by the dashed circle highlights the SDC1⁺ CAFs population. Each dot represents a single CAF; colours indicate cluster identity as indicated on the right. (C) Heatmap displaying homogeneity scores for each CAF subpopulation across cancer types. Higher scores (yellow) denote greater consistency across cancer types. (D) Bar plots showing the proportions of tumour-universal and tumour-specific CAF subpopulations across individual cancer types in tumour versus normal tissues. (E) Bar plots show the proportions of each CAF subpopulation among all CAFs in tumour (left) versus normal (right) tissues. (F) Representative multiplexed immunohistochemistry (mIHC) images of SDC1⁺ CAFs markers in tumour tissues. SDC1 (green), podoplanin (PDPN, orange) and fibroblast activation protein (FAP, red) indicate positive cells for each respective marker. Scale bars: 50 µm (merge) and 25 µm (individual channels). (G) Fluorescence-activated cell sorting (FACS) analysis quantifying SDC1⁺ CAF cell populations in COAD and BLCA clinical samples. (H) RT-qPCR validation of mRNA expressions of the top 10 marker genes in SDC1⁺ CAFs isolated from (G). Statistical analyses: two-sided unpaired Student’s t-test (H). ** p<0.01; *** p<0.001. BLCA, bladder cancer; CAF, cancer-associated fibroblast; COAD, colon adenocarcinoma; scRNA-seq, single-cell RNA sequencing.

Unsupervised clustering of CAFs across individual tumours revealed 31 transcriptionally distinct CAF subpopulations (figure 1B; online supplemental figure S4A). To assess their conservation across cancer types, we calculated homogeneity scores based on the relative abundance of each CAF subtype across tumours. Five subtypes, including ANGPT2⁺ CAFs, TNXB⁺ CAFs, SDC1⁺ CAFs, PLN⁺ CAFs and MMP1⁺ CAFs, were consistently observed across multiple cancer types (figure 1C,D; online supplemental figure S4B). These ‘universal’ CAFs, together, accounted for more than 50% of the total CAF population in most cancer types (figure 1D). In contrast, tumour-specific CAF subtypes exhibited restricted distributions, predominantly confined to one or a few cancer types (figure 1B,E). Functional annotation further stratified the five universal CAF subtypes into two main phenotypic categories: ANGPT2⁺ CAFs and PLN⁺ CAFs were enriched for muscle contraction-related pathways, consistent with a myCAF phenotype, while SDC1⁺ CAFs, MMP1⁺ CAFs and TNXB⁺ CAFs were enriched for ECM remodelling pathways, consistent with an ECM-remodelling CAF phenotype (online supplemental figure S4C and D). Among these, SDC1⁺ CAFs emerged as the most abundant subtype, accounting for approximately 30% of the total CAF population within tumours (figure 1E). In contrast, this subtype was rarely detected in adjacent normal tissues, highlighting their tumour-specific enrichment and suggesting a pivotal role in TME remodelling (figure 1E and online supplemental figure S4E). Additionally, comparing transcriptomic profiles with two earlier pan-cancer studies^{22 23} showed that SDC1⁺ CAFs represent a distinct and more refined population, partially similar to specific established subtypes (online supplemental figure S5 and S6).

To validate the identity and localisation of SDC1⁺ CAFs, we conducted multiplex immunohistochemistry (mIHC) staining for its signature markers (online supplemental figure S4F), which confirmed the specific localisation of SDC1⁺ CAFs in the tumour stroma and its widespread distribution across colon adenocarcinoma (COAD) and bladder cancer (BLCA) tissues (figure 1F). We further isolated SDC1⁺ CAFs using fluorescence-activated cell sorting (FACS) based on subtype-specific surface markers (figure 1G). Subsequent reverse transcription quantitative polymerase chain reaction (RT-qPCR) analysis of sorted populations confirmed significant enrichment of the top ten marker genes in SDC1⁺ CAFs, compared with SDC1⁻ CAFs (figure 1H), validating both the specificity of the marker set and the purity of the isolated cells. Together, these findings establish SDC1⁺ CAFs as a unique, predominant, tumour-enriched and transcriptionally distinct CAF subtype with potential functional significance in TME remodelling and cancer progression.

SDC1⁺ CAFs correlate with tumour progression and poor survival outcomes

To investigate the role of SDC1⁺ CAFs in tumour development, we first quantified their abundance in clinical COAD samples using mIHC. This analysis revealed a significant increase in the proportion of SDC1⁺ CAFs in tumour tissues compared with adjacent normal tissues (figure 2A). Consistently, single-cell transcriptomic data from healthy human tissues showed minimal expression of canonical SDC1⁺ CAF signature genes, such as COL11A1 and SDC1, further confirming the tumour-specific enrichment of this CAF subtype (online supplemental figure S7A and B). In a longitudinal cohort of head and neck cancer patients, SDC1⁺ CAF abundance exhibited a stepwise increase from normal tissue to precancerous lesions and ultimately to malignant tumours (figure 2B). Complementary analysis of bulk RNA-seq data from The Cancer Genome Atlas (TCGA) further revealed a stage-dependent accumulation of SDC1⁺ CAFs across multiple cancer types, including colon, pancreatic and bladder cancers (figure 2C), suggesting an essential role of SDC1⁺ CAFs in early tumourigenesis.

Figure 2. SDC1⁺ CAFs correlate with tumour progression, metastasis and poor patient survival. (A) Representative mIHC images of colon cancer samples comparing the abundance of SDC1⁺ CAFs in primary tumour tissues and adjacent normal tissues. SDC1⁺ CAFs are defined by co-expression of SDC1 (green), PDPN (yellow) and FAP (red). White arrows indicate representative SDC1⁺ CAFs. Scale bars: 50 µm (merge) and 25 µm (individual channels). Statistical analysis is shown on the right (n=9). (B) Line graph depicting the progressive increase in the proportion of SDC1⁺ CAFs (% in CAFs) during stepwise progression from normal tissues to precancerous leukoplakia and malignant lesions in head and neck squamous cell carcinoma. The results were based on scRNA-seq data.⁵⁵ (C) Box plots demonstrating the stage-dependent accumulation of SDC1⁺ CAF scores across three cancer types as indicated, based on bulk RNA-seq data of the TCGA cohort. (D) Box plots showing ORs for lymph node metastasis comparing patients with high versus low proportional signatures of CAF subpopulations. Each point represents the OR derived from logistic regression within an individual cancer type, based on bulk RNA-seq data from the TCGA cohort. (E) Representative mIHC images demonstrating the percentage of SDC1⁺ CAFs in bladder cancer lymph node metastases compared with primary tumour tissues. Quantification is shown on the right (n=8). Scale bars: 50 µm (merge) and 25 µm (individual channels). (F) Dot plots showing HRs and statistical significance (-log₁₀ P) for overall survival (OS) comparing patients with high versus low SDC1⁺ CAF proportional signatures across TCGA cancer types. Each dot represents a tumour type; colours indicate direction and significance of correlation. (G) Kaplan-Meier survival analysis showing the correlation between the SDC1⁺ CAFs proportional signatures and OS across multiple cancer types. Patients were stratified based on median SDC1⁺ CAFs signature score. Statistical analyses: paired t test (A, E); Kruskal-Wallis test (C, D); log-rank test (G). * p<0.05, ** p<0.01, *** p<0.001. BLCA, bladder cancer; CAF, cancer-associated fibroblast; COAD, colon adenocarcinoma; LIHC, liver hepatocellular carcinoma; mIHC, multiplexed immunohistochemistry; PAAD, pancreatic adenocarcinoma; scRNA-seq, single-cell RNA sequencing; OR, odds ratio; STAD, stomach adenocarcinoma.

To explore the clinical significance of SDC1⁺ CAFs, we examined their association with tumour metastasis in TCGA samples, using a signature-based scoring approach (see the Materials and methods section). Among all universal CAF subtypes, SDC1⁺ CAFs exhibited the strongest correlation with lymph node metastasis, demonstrating the highest ORs across multiple cancer subtypes (figure 2D). This observation was further validated in clinical BLCA specimens, where mIHC analysis showed a significantly higher proportion of SDC1⁺ CAFs in metastatic lymph nodes compared with matched primary tumours (figure 2E). Survival analyses across 31 TCGA cancer types revealed that high SDC1⁺ CAFs abundance was significantly associated with reduced overall survival and disease-specific survival, pronounced in gastric, liver, pancreatic and bladder cancers (figure 2F,G; online supplemental figure S7C). Collectively, these findings identify SDC1⁺ CAFs as key contributors to tumour progression and metastasis and underscore their potential utility as prognostic biomarkers and therapeutic targets across a broad range of cancers.

SDC1⁺ CAFs promote tumour metastasis through ECM remodelling

To elucidate the functional role of SDC1⁺ CAFs within TME, we first performed pathway enrichment analysis using the previously described pan-cancer single-cell transcriptomic data. This analysis revealed that SDC1⁺ CAFs are consistently enriched for genes related to ECM remodelling and pro-tumorigenic signalling pathways across diverse cancer types (online supplemental figure S7D). To further characterise their transcriptional signatures, we conducted bulk RNA sequencing on FACS-sorted SDC1⁺ and SDC1⁻ CAF populations isolated from clinical COAD and BLCA samples. Our analysis identified 605 genes significantly upregulated in SDC1⁺ CAFs, with functional enrichment in pathways related to ECM organisation, cell adhesion and regulation of cell migration (figure 3A; online supplemental figure S7E and F). These signatures were corroborated by scRNA-seq analysis, which demonstrated a higher ECM remodelling capacity in SDC1⁺ CAFs compared with other universal CAF subtypes (figure 3B). Consistently, transwell invasion and collagen gel contraction assays confirmed that SDC1⁺ CAFs exhibit significantly greater invasive potential and matrix remodelling activity than SDC1⁻ CAFs (figure 3C, D).

Figure 3. SDC1⁺ CAFs promote tumour metastasis through extracellular matrix remodelling. (A) KEGG pathway enrichment analysis of shared differentially expressed genes between SDC1⁺ and SDC1^- CAF subtypes isolated from COAD or BLCA tumours. (B) Box plots showing ECM disassembly scores among tumour-universal CAF subpopulations based on single-cell transcriptomic data. (C) Representative images (left) and quantitative analysis (right) of invasion capability for SDC1⁺ and SDC1^- CAFs. Scale bars: 100 µm. (D) Representative images (left) and quantification (right) of collagen gel contraction assays for SDC1⁺ and SDC1^- CAFs. Scale bars: 5 mm. (E) KEGG pathway enrichment analysis of shared differentially expressed genes in COAD or BLCA cells co-cultured with SDC1⁺ and SDC1^- CAFs. (F) Representative images (left) and quantification (right) of the migration (upper) and invasion (lower) assays for cancer cells alone (-) or co-cultured SDC1^+/- CAFs. Scale bars: 100 µm. (G) Western blot analysis showing the protein expressions of EMT-related markers in cancer cells described in (F). β-ACTIN was used as a loading control. (H) Schematic of in vivo lymph node metastasis model. COAD or BLCA cells were injected alone or co-injected with SDC1⁺ or SDC1^- CAFs into mouse footpads (created using https://biogdp.com). (I) Representative H&E staining images of tumour metastasis within lymph nodes in mice described in (H) (n=9–10). Scale bars: 625 µm (4×) and 200 µm (10×). (J) Quantification of lymph node metastasis rates across experimental groups. Numbers within columns represent the absolute counts of metastatic and non-metastatic lymph nodes (n, the total number of lymph nodes collected per group). Statistical analyses: two-sided Wilcoxon rank-sum test (B); two-sided unpaired Student’s t-test (C, D); one-way ANOVA (F); χ² test (J). *, p<0.05, **, p<0.01, ***, p<0.001. BLCA, bladder cancer; CAF, cancer-associated fibroblast; COAD, colon adenocarcinoma; ECM, extracellular matrix; EMT, epithelial-mesenchymal transition; KEGG, Kyoto Encyclopedia of Genes and Genomes.

To assess the impact of SDC1⁺ CAFs on tumour cell behaviour, we co-cultured COAD (HCT116) and BLCA (UMUC-3) cancer cells with SDC1⁺ or SDC1^- CAFs, followed by RNA-seq (online supplemental figure S8A and B). Transcriptomic profiling revealed extensive gene expression reprogramming in the co-cultured cancer cells, with prominent activation of pathways associated with ECM remodelling and cell migration, hallmarks of metastatic progression (figure 3E). Transwell assays showed that cancer cells co-cultured with SDC1⁺ CAFs exhibited significantly enhanced migratory and invasive capacities compared with those co-cultured with SDC1⁻ CAFs or cultured alone (figure 3F; online supplemental figure S8C and D). This was accompanied by elevated expression of key EMT-associated proteins (figure 3G). The pro-metastatic function of SDC1⁺ CAFs was further substantiated in an in vivo lymph node metastasis model (figure 3H), where co-injection of cancer cells with SDC1⁺ CAFs into mice significantly increased the incidence and extent of lymph node metastasis compared with co-injection with SDC1⁻ CAFs or cancer cells alone (figure 3I, J). Collectively, these findings establish SDC1⁺ CAFs as a metastasis-promoting fibroblast subset that drives tumour progression through both intrinsic ECM remodelling and extrinsic modulation of cancer cell invasiveness.

SDC1⁺ CAFs secrete CTGF to facilitate tumour metastasis

To elucidate the molecular mechanisms by which SDC1⁺ CAFs promote tumour metastasis, we investigated their intercellular communication with cancer cells. Ligand-receptor interaction analysis using NicheNet identified several candidate signalling pathways involved in pro-metastatic crosstalk between SDC1⁺ CAFs and malignant cells (figure 4A). Among the predicted effector molecules expressed by CAFs, SFRP2, CTGF, INHBA, ADAM12, CD99, FN1 and PLAU emerged as key candidates. RT-qPCR analysis confirmed significantly elevated expression of SFRP2, CTGF and FN1 in SDC1⁺ CAFs compared with SDC1⁻ CAFs in both COAD and BLCA clinical samples (online supplemental figure S8E). To functionally assess their contribution, each gene was individually knocked down in SDC1⁺ CAFs (online supplemental figure S9A), followed by co-culture with cancer cells and measurement of cancer cell migration. These experiments identified CTGF as the primary mediator, of which knockdown in SDC1⁺ CAFs significantly reduced the migration ability of COAD and BLCA cell lines (online supplemental figure S9B and C).

Figure 4. SDC1⁺ CAFs secrete CTGF to facilitate tumour metastasis. (A) NicheNet analysis of ligand-receptor interactions between SDC1⁺ CAFs and tumour cells. Ligands are ranked by predicted relevance to EMT process. Dot colour indicates average gene expression in SDC1⁺ CAFs, and square colour intensity reflects Pearson correlation with EMT signatures as computed by NicheNet. (B) Western blot analysis showing CTGF expression in SDC1⁺ and SDC1^- CAFs isolated from different patient tumours (#, patient IDs). (C) ELISA quantification of secreted CTGF levels in culture supernatants from SDC1⁺ and SDC1^- CAFs from multiple patient samples (#, patient IDs). (D) Western blot validation of CTGF knockdown in SDC1⁺ CAFs following transfection with CTGF-targeting siRNAs (si-CTGF-1, si-CTGF-2) or negative control siRNA (si-NC). SDC1^- CAFs serve as an additional control. (E) Representative images for transwell migration (upper) and invasion (lower) assays of cancer cells following co-culture with CAFs described in (D). Quantitative analysis is shown below. Scale bars: 100 µm. (F) Western blot analysis of EMT-related protein expression in cancer cells described in (E). (G) Schematic of in vivo lymph node metastasis model. COAD or BLCA cells were co-injected with CAFs into mouse footpads (created using https://biogdp.com). (H) Representative H&E staining images of lymph node metastasis in nude mice injected with cancer cells alone or co-injected with CAFs described in (G) (n=10). Scale bars: 625 µm (4×) and 200 µm (10×). (I) Quantification of lymph node metastasis rates across indicated experimental groups. Numbers within columns represent the absolute counts of metastatic and non-metastatic lymph nodes (n, the total number of lymph nodes collected per group). For western blot analysis, β-ACTIN serves as a loading control. Statistical analyses: two-sided unpaired Student’s t-test (C); one-way ANOVA (E); χ² test (I). *, p<0.05, **, p<0.01, ***, p<0.001. BLCA, bladder cancer; CAF, cancer-associated fibroblast; COAD, colon adenocarcinoma;

Western blot analysis confirmed elevated CTGF protein expression in SDC1⁺ CAFs compared with SDC1⁻ CAFs (figure 4B), and ELISA demonstrated significantly higher levels of secreted CTGF in the conditioned medium of SDC1⁺ CAFs (figure 4C), consistent with its role as a secreted factor.²⁵ To systematically investigate the function of CTGF, we conducted a series of gain-of-function and loss-of-function experiments. Co-culture followed by transwell assays revealed that silencing CTGF in SDC1⁺ CAFs markedly reduced cancer cell migration and invasion, along with decreased expression of EMT-associated proteins (figure 4D–F; online supplemental figure S10A and B). Conversely, ectopic overexpression of CTGF in SDC1⁻ CAFs enhanced the migratory and invasive capacities of both COAD and BLCA cells (online supplemental figure S10D–G). Importantly, CTGF knockdown in SDC1⁻ CAFs reduced tumour cell mobility (online supplemental figure S10H and I), but substantially less so than in SDC1⁺ CAFs (online supplemental figure S10J), indicating that CTGF facilitates cancer cell migration and invasion depending on specific engagement of SDC1⁺ CAFs.

In an in vivo metastasis model, CTGF knockdown in SDC1⁺ CAFs significantly reduced lymph node metastases, as evidenced by fewer and smaller metastatic lesions (figure 4G–I; online supplemental figure S10C). Given the secreted nature of CTGF, we further explored its therapeutic potential using neutralisation and supplementation approaches. Treatment with CTGF-neutralising antibodies in SDC1⁺ CAFs-conditioned media impaired cancer cell migration, invasion and EMT marker expression (online supplemental figure S11A–C). In contrast, supplementation with recombinant CTGF (rhCTGF) protein dose-dependently enhanced tumour cell migration and EMT induction (online supplemental figure S11D–G). Together, these findings establish CTGF as a critical secreted effector of SDC1⁺ CAFs-driven metastasis, acting partially through promoting EMT in cancer cells.

SDC1⁺ CAFs-derived CTGF promotes tumour metastasis through FGFR3 signalling

To delineate the molecular mechanism underlying CTGF-mediated metastasis, we first identified potential CTGF receptors in scRNA-seq data (see the Materials and methods). Our results revealed that EGFR, FGFR3, FGFR2, LRP1 and LRP6 exhibited significant positive correlations with EMT scores in tumour cells (figure 5A), suggesting potential roles in mediating CTGF-driven metastatic signalling. To further investigate the downstream pathways, we performed transcriptomic profiling of colon cancer HCT116 cells co-cultured with SDC1⁻ CAFs overexpressing CTGF or vector control. RNA-seq profiling uncovered significant gene expression reprogramming in cancer cells on co-culture, and Gene Set Enrichment Analysis (GSEA) analysis demonstrated an enrichment of FGFR downstream signalling pathways in HCT116 cells when co-cultured with SDC1^- CAFs overexpressing CTGF (online supplemental figure S12A and B). KEGG pathway enrichment analysis highlighted FGFR3 signalling as one of the top activated pathways, implicating FGFR3 as a key effector of CTGF (figure 5B).

Figure 5. CTGF promotes tumour metastasis through FGFR3-dependent signalling. (A) Correlation analysis between the expression levels of CTGF receptors and EMT scores in tumour cells. Dot colour reflects Spearman correlation coefficient. (B) KEGG pathway enrichment analysis of genes upregulated in HCT116 cells co-cultured with SDC1^- CAFs overexpressing CTGF compared with empty vector controls. (C) Representative images (top) and quantification (bottom) of transwell invasion (left) and migration (right) assays of tumour cells transfected with FGFR3 siRNA or control siRNA and concurrently co-cultured with SDC1^- CAFs overexpressing CTGF or empty vector controls. Scale bar, 100 µm. (D) Western blot analysis of EMT-related marker expressions in cancer cells described in (C). β-ACTIN serves as a loading control. (E) Schematic of in vivo lymph node metastasis model. COAD cells with FGFR3 knockdown were co-injected with CAFs into mouse footpads (created using https://biogdp.com). (F) Representative HE staining images of lymph node metastasis in vivo. HCT116 cells (with or without FGFR3 knockdown) were co-injected with SDC1^-CAFs overexpressing CTGF or vector controls into nude mice footpads (n=11). Scale bars: 625 µm (4×) and 200 µm (10×). (G) Quantification of lymph node metastasis rates across experimental groups in (F). Numbers within columns represent the absolute counts of metastatic and non-metastatic lymph nodes (n, the total number of lymph nodes collected per group). Statistical analyses: one-way ANOVA (C); χ² test (G). ns, non-significant; *, p<0.05, **, p<0.01, ***, p<0.001. BLCA, bladder cancer; CAF, cancer-associated fibroblast; COAD, colon adenocarcinoma; ECM, extracellular matrix; EMT, epithelial-mesenchymal transition; si-NC, negative control siRNA.

To functionally validate this hypothesis, FGFR3 was silenced in tumour cells using small interfering RNA (siRNA), followed by co-culture with CTGF-overexpressing or control SDC1⁻ CAFs. Notably, FGFR3 knockdown significantly attenuated CTGF-induced cancer cell migration and invasion (figure 5C; online supplemental figure S12C–E). Consistently, western blot analysis demonstrated that silencing FGFR3 abolished the CTGF-mediated upregulation of EMT-associated markers (figure 5D). Moreover, FGFR3 knockdown also inhibited tumour cell migration and invasion induced by SDC1⁺ CAFs (online supplemental figure S12F and G) or rhCTGF (online supplemental figure S14A and B). In contrast, knockdown of other candidate CTGF receptors, EGFR and FGFR2, only modestly reduced CTGF-driven migratory and invasive phenotypes (online supplemental figure S13A–E), with inhibitory effects that were consistently weaker than those observed on FGFR3 depletion, thereby highlighting FGFR3 as the predominant functional receptor mediating CTGF signalling in SDC1⁺ CAFs.

Additionally, co-culture with SDC1⁺ CAFs upregulated FGFR3 signalling-related genes in tumour cells (online supplemental figure S14C), further supporting the activation of FGFR3 signalling. In an in vivo metastasis model, FGFR3 knockdown in tumour cells significantly reduced lymph node metastases induced by co-injection with CTGF-overexpressing SDC1^- CAFs (figure 5E–G, online supplemental figure S14D). Consistently, correlation analyses across multiple cancer types revealed a positive association between SDC1⁺ CAF abundance and FGFR3 pathway activation, as well as elevated CTGF expression in tumours (online supplemental figure S14E–G). Together, these results demonstrate that CTGF, secreted by SDC1⁺ CAFs, facilitates tumour cell migration, invasion and EMT progression primarily through FGFR3 signalling activation.

KLF6 promotes CTGF transcription in SDC1⁺ CAFs to drive tumour metastasis

To elucidate the upstream regulatory mechanism responsible for CTGF upregulation in SDC1⁺ CAFs, we systematically analysed transcription factors associated with CTGF expression across single-cell transcriptomic data of CAFs from multiple cancer types. We observed that KLF6 expression was significantly upregulated in SDC1⁺ CAFs compared with SDC1⁻ CAFs across eight cancer types (figure 6A). Correlation analysis identified KLF6 as the most consistently correlated transcription factor, showing significant positive correlations with CTGF expression in SDC1⁺ CAFs across nine distinct cancer types, implicating it as a key regulator (figure 6B). Supportively, KLF6 mRNA and protein levels were markedly elevated in SDC1⁺ CAFs compared to SDC1⁻ CAFs derived from COAD and BLCA tissues (figure 6C, D).

Figure 6. KLF6 directly binds to the CTGF promoter to facilitate its transcription. (A) Dot plot analysis of CTGF-associated transcription factors (TF) across pan-cancer single-cell datasets. X-axis represents the number of correlated cancer types; Y-axis shows average TF expression in fold change between SDC1⁺ and SDC1^- CAFs. Colour indicates the percentage difference in expression (pct.diff) between SDC1⁺ and SDC1⁻ CAFs, with red denoting higher expression in SDC1⁺ CAFs and blue denoting higher expression in SDC1⁻ CAFs. Dot size reflects statistical significance based on adjusted p value (larger dots indicate greater significance). (B) Spearman correlation analysis of KLF6 and CTGF expression across cancer types, ranked by correlation coefficient. Dot colour represents correlation strength; dot size indicates p value. (C, D) RT-qPCR (C) and western blot (D) analysis of KLF6 expression in SDC1⁺ and SDC1^- CAFs isolated from different patient tumours (#, patient IDs). (E, F) RT-qPCR (E) and western blot (F) analyses of KLF6 and CTGF mRNA and protein expressions in SDC1⁺ CAFs transfected with KLF6 siRNAs or control siRNA. (G, H) RT-qPCR (G) and western blot (H) analyses of KLF6 and CTGF mRNA and protein expressions in SDC1^- CAFs overexpressing KLF6 or control vector. (I) Schematic of CTGF promoter constructs used in luciferase reporter assays. Constructs include: WT: wild-type construct containing full-length promoter; DEL and MUT: construct with KLF6 binding site deleted or mutated, respectively. (J-K) Dual-luciferase reporter assays evaluating CTGF promoter activity in 293 T cells silencing KLF6 with WT (J) or DEL/MUT (K) promoter constructs. (L, M) Dual-luciferase reporter assays illustrating CTGF promoter activity in 293 T cells with KLF6 overexpression under KLF6 motif variants: WT (L), DEL and MUT (M). (N,O) ChIP-qPCR analysis of KLF6 binding enrichment at the CTGF promoter in SDC1⁺ CAFs from COAD (N) or BLCA (O) tumours. IgG: control antibody; KLF6-HA: anti-HA antibody. (P) ChIP-PCR analysis validating KLF6 occupancy at the CTGF promoter region containing the KLF6 binding site in SDC1⁺ CAFs. For western blot analysis, GAPDH or β-ACTIN is used as a loading control. Statistical analyses: Wilcoxon test (A); two-sided unpaired Student’s t-test (C, G, N, M, O); one-way ANOVA (E, J, K, L). ns, non-significant; *, p<0.05, **, p<0.01, ***, p<0.001. BLCA, bladder cancer; CAF, cancer-associated fibroblast; ChIP, chromatin immunoprecipitation; COAD, colon adenocarcinoma.

Functional validation revealed that KLF6 knockdown led to a significant reduction in CTGF expression, whereas KLF6 overexpression robustly increased both CTGF transcript and protein levels (figure 6E–H). To determine whether KLF6 directly activates CTGF transcription, we performed dual-luciferase reporter assays using a construct containing CTGF promoter (figure 6I). KLF6 overexpression significantly enhanced CTGF promoter activity, while KLF6 silencing markedly suppressed it (figure 6J and L). In silico motif analysis using the JASPAR database²⁶ identified a putative KLF6-binding motif within the CTGF promoter (figure 6I). Site-directed mutagenesis or deletion of this motif abolished KLF6-mediated transcriptional activation of the CTGF promoter (figure 6K and M). Moreover, chromatin immunoprecipitation assays confirmed direct binding of KLF6 to the CTGF promoter in CAFs from both COAD and BLCA tissues (figure 6N–P), collectively establishing KLF6 as a direct transcriptional activator of CTGF in SDC1⁺ CAFs.

To examine the functional consequences of the KLF6-CTGF regulatory axis, we performed rescue experiments. Silencing KLF6 in SDC1⁺ CAFs significantly suppressed tumour cell migration, invasion and expression of EMT markers, whereas these effects were almost completely rescued by forced CTGF overexpression. (figure 7A,B, online supplemental figure S15A–C). In vivo, stable KLF6 knockdown in SDC1⁺ CAFs significantly reduced lymph node metastasis, which was reversed by simultaneous CTGF overexpression (figure 7C–E, online supplemental figure S15D). These data demonstrate that CTGF functions downstream of KLF6 to mediate pro-metastatic phenotypes. Clinically, immunohistochemical analysis of tumour specimens revealed a strong positive correlation between KLF6 and CTGF expression (figure 8A–D). Moreover, survival analyses showed that high expression of KLF6 or CTGF was associated with worse overall survival in COAD and BLCA cohorts, with their concomitant upregulation exhibiting the strongest adverse prognostic impact (figure 8E–G). Taken together, these findings identify the KLF6-CTGF axis as a pivotal regulatory mechanism within SDC1⁺ CAFs that drives tumour metastasis and correlates with adverse clinical outcomes, representing a valuable prognostic biomarker and a potential therapeutic target for disrupting stromal-driven metastatic progression across multiple cancer types.

Figure 7. CTGF mediates KLF6-driven metastasis in SDC1⁺ CAFs. (A) Representative images of transwell migration (upper) and invasion (lower) assays of cancer cells co-cultured with SDC1⁺ CAFs silencing KLF6 and/or overexpressing CTGF. Quantification of migrated/invaded cells is presented on the bottom. Scale bar, 100 µm. (B) Western blot analysis of EMT-related proteins in tumour cells described in (A). β-ACTIN serves as a loading control. (C) Schematic of in vivo lymph node metastasis model. T24 or HCT116 cells were co-injected with SDC1⁺ CAFs silencing KLF6 and PCDH/CTGF overexpression into nude mice footpads (created using https://biogdp.com). (D) Representative H&E staining images of lymph node metastasis in vivo described in (C) (n=10). Scale bars: 625 µm (4×) and 200 µm (10×). (E) Quantification of lymph node metastasis rates across experimental groups as in (D). Numbers within columns represent the absolute counts of metastatic and non-metastatic lymph nodes (n, the total number of lymph nodes collected per group). Statistical analyses: one-way ANOVA (A); χ² test (E); *, p<0.05, **p<0.01, ***p<0.001. CAF, cancer-associated fibroblast; EMT, epithelial-mesenchymal transition; FACS, fluorescence-activated cell sorting.

Figure 8. Stromal CTGF and KLF6 expression correlate with poor overall survival. (A-B) Representative IHC images for KLF6 (top) and CTGF (bottom) expression in human COAD (left) and BLCA (right) tumour sections. Scale bars: 100 µm. (C-D) Correlation analysis of stromal CTGF and KLF6 protein expression levels in patients with COAD (C; n=59) and BLCA (D; n=80). (E, F) Kaplan-Meier survival analysis correlating CTGF (E) or KLF6 (F) protein expression levels in stroma and overall survival of patients with COAD and BLCA. Patients were stratified based on median protein expression levels. (G) Kaplan-Meier survival analysis for COAD and BLCA patients, stratified by co-expression of KLF6 and CTGF in tumour stroma: double-high (KLF6^HighCTGF^High), single-high (either KLF6^High or CTGF^High) or double-low groups. (H) Proposed working model for stromal CAFs and tumour interaction. SDC1⁺ CAFs with KLF6 upregulation transcriptionally activate CTGF secretion, which in turn engages FGFR3 on tumour cells to induce EMT and metastasis via upregulation of SLUG, SNAIL and MMP1. This metastatic signalling cascade can be disrupted by CTGF neutralisation (FG-3019, CTGF-neutralising antibody) or silencing of the KLF6-CTGF-FGFR3 axis. Statistical analyses: log-rank test (E, F, G). BLCA, bladder cancer; CAF, cancer-associated fibroblast; COAD, colon adenocarcinoma; EMT, epithelial-mesenchymal transition; IHC, immunohistochemistry.

Discussion

Given the heterogenous composition and divergent tumourigenic functions of CAFs, we conducted a large-scale integrative pan-cancer analysis and identified 31 transcriptionally distinct CAF subtypes. These subtypes exhibit distinct distribution patterns—some were broadly conserved across multiple cancer types (eg, SDC1⁺ CAFs, ANGPT2⁺ CAFs, PLN⁺ CAFs), while others showed tumour-specific distribution. Importantly, these subtypes also exhibited functional diversity, reflected in their distinct gene expression programmes. For instance, ANGPT2⁺ CAFs and PLN⁺ CAFs clustered within the myCAFs phenotype, which are predominantly associated with pro-angiogenic signatures.²⁷ Recent studies have also shown that specific CAF subsets emerged in response to therapeutic pressure (eg, CD16⁺ CAFs, SLC14A1⁺ CAFs)^{28 29} or contributed to immune evasion by forming physical or molecular barriers that suppress immune infiltration.^{30 31} Together, these findings underscore the multidimensional heterogeneity of CAFs, which dynamically cope with other TME components to facilitate tumour progression and likely contribute to variable treatment responses. These complexities highlight the challenges of developing CAF-targeted therapies, especially in the context of tumour type-dependent CAF behaviour.

Among the universal CAF subtypes, SDC1⁺ CAFs emerged as a robust and clinically relevant pro-metastatic population. While several metastasis-promoting CAF subsets have previously been reported, such as PDGFRα⁺ITGA11⁺ CAFs in BLCA³², myCAFs in pancreatic cancer ³³ and PDPN⁺ CAFs in gastric cancer³⁴, these are predominantly tumour tissue-specific, suggesting their functional specialisation in tumour type context. In contrast, SDC1⁺ CAFs are widely conserved across malignancies and exhibit a unique transcriptional profile that sets them apart from other CAF subsets. Functional validation through in vitro assays and in vivo models confirmed that SDC1⁺ CAFs significantly enhance tumour cell invasiveness and promote lymphatic metastasis across diverse cancer types, establishing them as functionally conserved and metastasis-promoting CAF subsets with potential for universal therapeutic targeting. Additionally, the unique transcriptional and functional characteristics of SDC1⁺ CAFs necessitate a more precise classification of previously described CAF subtypes, such as CAF_LRRC15 and CAF_infla^{22 23}, which share partial signatures across multiple subsets identified in our study.

Mechanistically, our findings further elucidate that SDC1⁺ CAFs exert their pro-metastatic effects predominantly through CTGF secretion into the TME. CTGF, a matricellular protein of the CCN family, is known to regulate multiple oncogenic processes, including cell migration, ECM remodelling and angiogenesis.^35–37 While CTGF is predominantly derived from tumour cells, stromal CTGF production, particularly from CAFs, has been implicated in enhancing ECM deposition and inducing immunosuppressive niches.^{38 39} Our results extend these findings by showing that SDC1⁺ CAFs-derived CTGF directly influences tumour cell behaviour and promotes EMT. Previous studies have suggested that CTGF from undefined CAF populations contributes to tumour progression^{40 41}; however, our study pinpoints SDC1⁺ CAFs as a definitive cellular source. Further, we identify FGFR3 as the key receptor mediating CTGF-induced EMT signalling in tumour cells, which is a well-known pathway to modulate oncogenic signalling across cancer types.^{42 43} Notably, this interaction unveils a previously unrecognised ligand-receptor relationship, as CTGF is not a canonical ligand for FGFR3. Importantly, both CTGF blockade and FGFR3 silencing significantly attenuated metastasis in preclinical models, supporting the therapeutic relevance of this novel paracrine axis.

To better understand the formation and maintenance of SDC1⁺ CAFs, we identified KLF6 as a critical upstream transcription factor of CTGF, thereby activating FGFR3 downstream signalling and sustaining the pro-metastatic phenotype of SDC1⁺ CAFs (online supplemental figure S15E). Notably, while KLF6 has traditionally been characterised as a tumour suppressor in prostate cancer and hepatocellular carcinoma^{44 45}, it has also been shown to act as a tumour promoter in other contexts, such as clear cell renal cell carcinoma⁴⁶, colon cancer⁴⁷ and TH17-driven inflammatory conditions.⁴⁸ Our findings support the concept of cell type-dependent and context-dependent functions of KLF6 and reveal its pro-tumorigenic role in stromal fibroblasts via activation of the KLF6-CTGF axis. Interestingly, KLF6 is essential for maintaining the established transcriptional programme of SDC1⁺ CAFs, but it alone cannot drive the differentiation of CAFs into this specific subtype (online supplemental figure S16A and B). Similarly, tumour cells alone were insufficient to induce the SDC1⁺ CAFs phenotype in normal fibroblasts or SDC1⁻ CAFs (online supplemental figure S16C-G), in contrast to certain myCAF subtypes which can be readily induced by tumour-derived factors such as TGF-β.^{49 50} This suggests that additional TME components, including immune and stromal signals, may be necessary to drive SDC1⁺ CAFs differentiation. This is consistent with previous findings that CAF specification is shaped by context-specific extracellular signals and transcriptional regulators.^{22 51–53}

In summary, our study identifies SDC1⁺ CAFs as a universal pro-metastatic fibroblast population present across diverse human cancers and uncovers a previously unappreciated KLF6-CTGF-FGFR3 signalling axis that mediates stromal-tumour communication to drive EMT and metastatic progression. Mechanistically, KLF6 activation in SDC1⁺ CAFs induces CTGF secretion, which in turn activates FGFR3 signalling in tumour cells to promote EMT and enhance invasive capacity (figure 8H), highlighting a distinct mechanism from previous studies that focus on KLF6 and CTGF derived from malignant cells. These findings not only reveal a conserved mechanism of CAF-driven metastasis, but also establish SDC1⁺ CAFs as a clinically actionable stromal subset with broad therapeutic relevance. Unlike prior approaches targeting tumour-specific CAFs that risk off-target effects, selective inhibition of the KLF6-CTGF-FGFR3 axis in SDC1⁺ CAFs may represent a novel approach to disrupt stromal support and improve clinical outcomes across multiple solid tumours.

Despite these insights, several limitations remain. First, while we establish CTGF as a pivotal effector of SDC1⁺ CAFs-mediated metastasis, contact-dependent mechanisms and other paracrine factors may also contribute. Second, the cellular origin and differentiation pathways of SDC1⁺ CAFs remain undefined. It is plausible that non-tumour components of the TME, such as immune cells or fibroblast-intrinsic programmes, participate in their emergence. Finally, although SDC1⁺ CAFs represent a major pro-metastatic population, other CAF subsets likely play context-specific roles in tumour progression that merit further investigation.

Data availability statement

Data are available upon reasonable request.