Targeting fibroblast TXNDC5 resolves tumor desmoplasia and PD-1 resistance in colorectal cancer with mesenchymal traits Authors: Kai-Lin Cheng, Chih-I Chen, Shu-Han Yu, Huai-Wen Liang, Yi-Wei Tsai, Chen-Ting Hung, Yu-Shan Lin, Gang-Hui Lee, Po-Yu Chen, Ming-Jer Tang, Yueh-Feng Wu, Sung-Jan Lin, Jen-Kuang Lee, Linda Chia-Hui Yu, Shuei-Liong Lin, Shih-Yu Chen, Wen-Chih Lee, Yung-Ming Jeng, Ruby Yun-Ju Huang, Ruey-Hwa Chen, Ching-Chow Chen, Kai-Chien Yang Published: 2026 ABSTRACT AbstractMesenchymal-type colorectal cancer is frequently refractory to immunotherapy, with resistance largely driven by a cancer-associated fibroblast-mediated fibrotic barrier and immune exclusion. Here, we show that TXNDC5, a protein disulfide isomerase upregulated in both human and mouse colorectal cancer-associated fibroblasts, is a critical regulator of this immunosuppressive niche. Mechanistically, TXNDC5 enhances TGFβ signaling by stabilizing the TGFβ receptor TGFBR1 in fibroblasts, thereby driving fibroblast activation and pro-tumor signaling pathways. Fibroblast-specific Txndc5 deletion attenuates tumor stiffness, decompresses tumor vessels, and reduces intratumoral hypoxia while inhibiting pro-tumorigenic inflammatory fibroblast polarization and augmenting cytotoxic T-cell recruitment. This stromal remodeling restrains tumor growth and markedly sensitizes tumors to PD-1 blockade therapy. Our findings identify the TXNDC5–TGFBR1 axis as a key orchestrator of the fibrotic shield within the malignant microenvironment and suggest that targeting TXNDC5 in fibroblasts could help overcome immunotherapy resistance in patients with mesenchymal-type colorectal cancer.Subject terms: Cancer microenvironment, Colorectal cancer, Cancer immunotherapy, Cancer therapeutic resistance, Targeted therapies Introduction Colorectal cancer (CRC) is a global health burden, ranking among the most prevalent and lethal malignancies1. While conventional treatment decisions are primarily based on tumor stage and location, the outcomes vary due to extensive genomic, cellular and molecular heterogeneity. Advances in the molecular understanding of CRC have transitioned therapeutic strategies from targeting cancerous cells alone to combinational therapies in clonal-stromal-immune perspectives2. Recently, a transcriptome-based molecular classification defined most CRC tumors into four consensus molecular subtypes (CMS)3. The MSI-high CMS1 tumors, featured by hypermutation, hypermethylation and immune activation, respond well to immune checkpoint blockade (ICB). The canonical CMS2 and metabolic-dysregulated CMS3, albeit poorly immunogenic or immune “cold”, are sensitive to the FOLFOX cocktail chemotherapy. CMS4 tumors, linked to chronic inflammation caused by colitis or intestinal microbiota, are characterized by activated TGFβ signaling, abundant cancer-associated fibroblasts (CAF), extracellular matrix (ECM) remodeling and epithelial-mesenchymal transition (EMT). Notably, CMS4 tumors resist to chemo- and immunotherapies and have the least favorable outcomes, highlighting the importance of tumor microenvironment (TME) and CAF in tumor progression and treatment responses. Activated fibroblasts in the tumor stroma or CAFs, exhibit enhanced proliferation, ECM production and resistance to apoptosis, resembling activated myofibroblasts in fibrotic organs. CAFs secrete factors that promote tumor growth and angiogenesis4, thereby contributing critically to tumor initiation, progression, metastasis and therapeutic resistance. The resemblance in the cellular behavior and properties between activated myofibroblasts and CAFs suggests a pathophysiological mechanism shared by these two seemingly unrelated disorders. Recently, our group has identified thioredoxin domain containing 5 (TXNDC5), a fibroblast-enriched endoplasmic reticulum (ER) protein, as a critical mediator of fibrosis5–8 by augmenting TGFβ signaling and fibroblast activation. In addition, fibroblast-specific deletion of Txndc5 protects against post-injury fibrosis and preserves organ function in mice. Given the parallels between fibrosis and tumor desmoplasia, we hypothesized that TXNDC5 also drives CAF activation and stroma formation, contributing to cancer progression and drug resistance. We chose to test this hypothesis directly in CRC, where CAF activation and a stromagenic TME clearly impact tumorigenesis and clinical outcome. We demonstrate that TXNDC5 is significantly upregulated in CAFs from human and mouse CRC, particularly in stroma-rich tumors. Increased TXNDC5 levels drive CAF activation and tumor desmoplasia via TGFβ signaling, creating a hypoxic, immunosuppressive TME, thereby promoting tumor progression. Eliminating Txndc5 in CAFs alleviates tumor desmoplasia, reinstates immune equilibrium within the TME, and potentiates the effectiveness of ICB therapy in CRC. Our findings highlight a critical yet previously unidentified role of TXNDC5 in fostering tumor fibrosis and progression in CRC, and suggest the therapeutic potential of simultaneously targeting TXNDC5 and immune checkpoint pathways. Results TXNDC5 is highly up-regulated in the stroma of human and murine CRC To assess the clinical relevance of TXNDC5 in CRC, a human CRC microarray dataset (GSE21510)9 was re-analyzed. As shown in Supplementary Fig. 1a, TXNDC5 transcript expression was significantly (P < 0.001) upregulated in tumors, compared to normal colonic tissues. Immunohistochemistry (IHC) staining of a human CRC tissue array revealed markedly increased TXNDC5 protein expression in CRC samples, particularly within the tumor stroma (Supplementary Fig. 1b). Furthermore, both TXNDC5 staining intensities (Fig. 1a) and the percentage of TXNDC5-positive areas in the tumor stroma (Supplementary Fig. 1b) increased with advancing tumor grade, suggesting a progressive TXNDC5 upregulation during CRC progression, especially within the tumor stroma. Stromal TXNDC5 upregulation correlates with advanced CRC grades and poor prognosis. a TXNDC5 staining on a CRC tissue array (n = 84 cases; 6 Grade I, 16 Grade I–II, 48 Grade II, 10 Grade II–III, 4 Grade III tumors, and 42 normal controls). Stromal TXNDC5 increases with advancing tumor grades. b Sorted human colorectal tumors (GSE39396, n = 6 patients) show TXNDC5 upregulation in fibroblasts and endothelial cells, but not in tumor epithelial cells. c Co-staining of TXNDC5, α-SMA, and CK19 on colorectal tumors (T, n = 8 patients) and adjacent normal tissues (N, n = 8 patients) shows TXNDC5 and α-SMA upregulation and co-localization in tumors. d Analysis of the GSE103479 cohort shows increased TXNDC5 expression in CMS4 tumors (n = 28), compared with CMS1 (n = 25), CMS2 (n = 60), and CMS3 (n = 22) tumors. e Kaplan-Meier analysis (GSE63216) shows that TXNDC5high CRC tumors were associated with worse survival compared with TXNDC5low tumors (n = 80 patients per group). f TXNDC5 expression in CAFs from CRC tumors positively correlates with the fibrogenic genes COL1A1 and ACTA2. g A schematic illustrates the AOM/DSS-induced and Apc-mutant (Apcmin/+) CRC models. Immunohistochemical staining shows increased stromal TXNDC5 expression (arrows) in AOM/DSS-induced and Apcmin/+ tumors compared with normal colons (Ctrl) (n = 6 mice per group). h TXNDC5 staining in AOM/DSS-induced tumors (Col1a1-GFPTg background, n = 6 mice) shows predominant localization in fibroblasts based on co-staining with lineage markers (COL1A1: CAF; CK19: tumor cells; CD11b: immune lineages; VE-cadherin: endothelial cells; CD11b: myeloid cells; CD4 and CD8: T cells; FOXP3: Tregs; B220: B cells). Data are presented as mean ± SEM. P-values are determined using Welch’s one-way ANOVA followed by Dunnett’s T3 post-hoc test for multiple comparisons or two-sided Student t test for two-group comparisons. For Kaplan-Meier analysis, the two-sided log-rank (Mantel-Cox) test was used. Exact P-values are indicated in figures. Source data are provided in the Source Data file. The stroma-enriched expression pattern of TXNDC5 was also observed in the CRC samples from the Cancer Genome Atlas-Colorectal Cancer (TCGA-CRC) database10 (Supplementary Fig. 1c). In line with this finding, microarray studies on the individual cell types within human CRC revealed significantly (P < 0.05) higher TXNDC5 expression in CAFs compared to tumor cells or leukocytes (GSE3939611, Fig. 1b). Immunofluorescence (IF) staining of additional CRC tumors (T) and adjacent normal colon tissues (N) also revealed a marked TXNDC5 upregulation in tumor stroma, coinciding with increased expression of CAF marker ɑ-SMA (Fig. 1c). Consistent with the observed enrichment of TXNDC5 in CRC CAFs, TXNDC5 expression was particularly elevated in mesenchymal CMS4, compared to CMS1-3, CRC subtypes (GSE10347912, Fig. 1d). In addition, analysis of the survival data from CRC patients (GSE63216)13 revealed that high TXNDC5 expression is associated with reduced overall survival compared to low TXNDC5 expression (Fig. 1e). Notably, the transcript expression levels of TXNDC5 showed a strong positive correlation with that of fibrogenic genes including COL1A1, ACTA2 and COL3A1 in CAFs isolated from human CRC (Fig. 1f and Supplementary Fig. 1d). To investigate the functional role of TXNDC5 in mesenchymal-type CRC, we utilized a mouse CRC model induced by azoxy-methane (AOM)/dextran sodium sulfate (DSS) treatment (scheme shown in Fig. 1g and Supplementary Fig. 2a), which forms inflamed, stroma-rich CRC tumors sharing molecular, immunological and mutational features with human CMS4 mesenchymal-type CRC14. Indeed, analyses of a single cell RNA-Seq (scRNA-Seq) dataset from mouse CRC (GSE134255)15 revealed a marked infiltration of stromal fibroblasts and immune-tolerant myeloid cells in AOM/DSS-induced tumors (Supplementary Fig. 1e), similar to the stromal- and myeloid-enriched cellular signature identified in CMS4 CRC (GSE132465, Supplementary Fig. 1f). This is in striking contrast to the cellular composition of CRC tumors derived from Apcmin/+ mice, where CAFs account for only a small portion of stromal cells (Supplementary Fig. 1e). Furthermore, the intra-tumoral cytokines, CCL-2, IL-17, IL-23 and TGFβ, produced by stromal fibroblasts and contribute to immune tolerance in CMS4 and colitis-associated CRC2, were upregulated in CAFs from AOM/DSS tumors, accompanied by an increase in their corresponding receptors (CCR2, IL17R, IL23R and TGFBR) in immune and cancer cells (Supplementary Fig. 1g). We also performed a comprehensive scRNA-seq analysis of Txndc5 expression across all major cell populations in the AOM/DSS-induced mouse CRC tumors. As shown in Supplementary Fig. 1i, fibroblasts exhibit the highest Txndc5 expression among all major cell types, followed by epithelial cells, endothelial cells, and lastly, immune cells. This expression pattern is similar to that observed in human CMS4 tumors (Supplementary Fig. 1h), in which fibroblasts, endothelial cells, and epithelial cells are the predominant expressors of TXNDC5. In contrast, B cells exhibit higher TXNDC5 expression in non-CMS4 tumors, particularly in CMS2/3 subtypes. These data together suggest that the AOM/DSS-induced mouse CRC model effectively recapitulates the stroma-rich, immune-tolerant features and aberrant TXNDC5 expression pattern observed in human CMS4 mesenchymal-type CRC. Notably, scRNA-Seq analyses revealed a marked upregulation of Txndc5 in CAFs from AOM/DSS-induced, compared to those from Apcmin/+, CRC tumors (Supplementary Fig. 1e, right panel). This finding is in line with the higher expression levels of TXNDC5 observed in CAFs from CMS4, compared to those from CMS1-3, CRC tumors (Supplementary Fig. 1f, right panel, and 1 h). Consistent with the results of scRNA-Seq analyses, IHC staining revealed a marked upregulation of TXNDC5 in AOM/DSS-induced tumors, particularly in the tumor stroma, compared to normal colon tissues (Fig. 1g). This stroma-enriched upregulation of TXNDC5 was also observed in Apcmin/+ tumors, albeit to a much lesser extent compared to AOM/DSS-induced tumors (Fig. 1g and Supplementary Fig. 2d). Using a Col1a1-GFPTg transgenic mouse line that allows visualizing active, collagen-producing CAFs with GFP16, IF staining revealed a high degree of co-localization (68.8%) between TXNDC5 and GFP-positive, collagen-producing CAFs (Fig. 1h and Supplementary Fig. 2e). In contrast, TXNDC5 was significantly lower in other cell types, with 10.8% co-localization in VE-cadherin+ endothelial cells, 7.0% in CK19+ tumor cells, 5.2% in CD11b+ myeloid cells, 5.1% in B220+ B cell, 1.4% in CD4+ T cells, 1.2% in CD8+ T cells and 0.6% in FOXP3+ regulatory T cells (Tregs) (Fig. 1h). In addition, the GFP-positive and/or α-SMA-positive CAF population was clearly expanded in AOM/DSS-induced tumors (Supplementary Fig. 2f), confirming their stromagenic, CAF-enriched phenotype. Moreover, TXNDC5 expression was observed to be highly co-localized with α-SMA+ area (Supplementary Fig. 2g) in the AOM/DSS-induced tumors (51.4%), as well as in the human CRC tumors (44.9%). Taken together, these findings demonstrate significant TXNDC5 upregulation in activated CAFs within both human CMS4 and mouse AOM/DSS-induced CRC. This elevated expression is strongly linked to tumor desmoplasia and worsened clinical outcomes. Txndc5 deletion markedly attenuated CRC progression To determine the in vivo contribution of TXNDC5 to CRC tumorigenesis, control C57BL/6 (WT) and Txndc5 global knockout (Txndc5 -/-) mice5–8 were subjected to AOM/DSS treatment. The severity of AOM/DSS-induced colitis, assessed by body weight loss, diarrhea/bloody stool scores, colon length shortening, and spleen enlargement, was similar in both groups (Supplementary Fig. 3a). The overall tumor load, including tumor number, sum of tumor diameters and the incidence of large tumors ( > 4 mm), was significantly reduced in Txndc5 -/-, compared to WT, mice following AOM/DSS treatment (Fig. 2a). In addition, the abundance of CAFs, quantified by α-SMA/COL1A1-GFP double-positive areas (Fig. 2b), was significantly lower in Txndc5 -/-, compared to WT, tumors. This reduction was accompanied by decreased expression of the reactive CAF marker α-SMA/Acta2 (Supplementary Fig. 3b, c), smaller tumor stroma areas (quantified using Masson’s trichrome staining, picrosirius red staining and second harmonic generation [SHG] imaging, Fig. 2b), and reduced expression of ECM proteins and genes, including COL1A1 and fibronectin (Supplementary Fig. 3b, c). These findings demonstrate that TXNDC5 deficiency limits CAF activation, stroma formation, and tumor progression in CRC. Txndc5 deficiency attenuates fibroblast activation and ECM production, reducing CRC tumor burden. a Representative images of the dissected and Swiss-rolled cancerous colons from AOM/DSS-treated control wild-type and Txndc5-/- mice (n = 8 mice per group). Tumor number and total tumor diameter per colon were significantly reduced in the Txndc5-/- group. Tumor size distribution ( < 2 mm, 2–4 mm, and > 4 mm) was plotted as stacked bars. Notably, tumors > 4 mm are absent in Txndc5-/- mice. b ɑ-SMA staining in WT and Txndc5-/- tumors (on a Col1a1-GFPTg background; n = 6 mice per group) reveals reduced CAF abundance (ɑ-SMA+COL1A1+) in Txndc5-/- tumors. Masson’s trichrome, picrosirius red staining, and second-harmonic generation (SHG) imaging demonstrate reduced tumor fibrosis in Txndc5-/- tumors. c Tumor number and total diameters are reduced in Col1a2-CreERT2;Txndc5fl/fl mice compared with control mice (Col1a2-CreERT2, n = 10 mice per group). Consistent with the findings in Txndc5-/- mice, tumors > 4 mm are observed exclusively in Col1a2-CreERT2 control, but not in Col1a2-CreERT2;Txndc5fl/fl mice. d Fibroblast-specific Txndc5 deletion reduces tumor ɑ-SMA+ areas and collagen deposition (Masson’s trichrome, picrosirius red staining, and SHG imaging) compared with controls (n = 6 mice per group). Data are presented as mean ± SEM. P-values are determined using a two-sided Student t test. Exact P-values are indicated in figures. Source data are provided in the Source Data file. Next, we investigated whether Txndc5 deletion affects TGFβ signaling activity in CRC, a key driver of CAF activation and stromagenesis in gastrointestinal cancers17. Immunoblotting revealed a marked reduction of SMAD3 phosphorylation (at S423/425) in Txndc5 -/-, compared to WT, tumors, indicating lower TGFβ signaling activity in tumors lacking TXNDC5 (Supplementary Fig. 3b). In addition, protein expression levels of TGFBR1, but not TGFBR2, were significantly reduced in AOM/DSS-induced Txndc5 -/-, compared to WT, tumors (Supplementary Fig. 3b). These data indicate that impaired TGFβ signaling underlies the reduced CAF activation and stromagenesis in Txndc5 -/- tumors. CAF-specific Txndc5 deletion diminished tumor fibrosis in CRC To elucidate the impact of TXNDC5 in specific cell lineages on CRC formation, Txndc5 was selectively deleted in tumor epithelial, endothelial, or fibroblast lineages, followed by AOM/DSS induction. A significant reduction of tumor burden (including tumor number, sum of tumor diameters and tumor size) was observed in mice with fibroblast-specific Txndc5 deletion (Col1a2-CreERT2;Txndc5 fl/fl), compared to controls (Col1a2-CreERT2) (Fig. 2c), accompanied by significantly reduced CAF activation (as evidenced by reduced α-SMA/COL1A1 double-positive area) and tumor desmoplasia (as indicated by smaller fibrotic areas on Masson’s trichrome/picrosirius red staining and SHG imaging) in AOM/DSS-induced CRC (Fig. 2d). By contrast, conditional deletion of Txndc5 in either colonic epithelium (Cdx2-Cre;Txndc5 fl/fl, Supplementary Fig. 4a) or endothelial cells (Cdh5-CreERT2;Txndc5 fl/fl, Supplementary Fig. 4c) failed to impact tumor progression or fibrosis (Supplementary Fig. 4b, d). Similar to the results observed in global Txndc5 KO mice, fibroblast-specific Txndc5 deletion reduced intra-tumoral TGFβ signaling activity (as reflected by reduced p-SMAD3/total SMAD3 ratios on immunoblots), CAF marker α-SMA/Acta2, ECM proteins/genes (COL1A1 and fibronectin), and TGFBR1 protein (Supplementary Fig. 4e, f) levels in AOM/DSS tumors. Together, these findings demonstrate that TXNDC5 in colonic fibroblasts, but not in epithelial or endothelial cells, contributes critically to CAF activation, tumor fibrosis, and progression of mesenchymal-type CRC. Furthermore, eliminating Txndc5 in fibroblasts is sufficient to diminish the fibrogenic TGFβ signaling activity in CRC. To test whether TXNDC5 in fibroblasts also contributes to non-mesenchymal-type colorectal tumors, we crossbred Col1a2-CreERT2;Txndc5 fl/fl mice with Apcmin/+, a model known to develop adenomas and carcinomas driven by a loss-of-function Apc mutation18. Intestinal tumor numbers and sizes were markedly reduced in Apcmin/+ mice with fibroblast-specific Txndc5 deletion (Supplementary Fig. 5a), accompanied by significantly lower levels of fibroblast activation, ECM production, and tumor fibrosis (Supplementary Fig. 5b, c), indicating that fibroblast TXNDC5 promotes stromagenesis and the progression of intestinal tumors arising from diverse underlying causes. TXNDC5 reinforces the TGFβ1-TGFBR1 axis to activate fibroblasts The functional role of TXNDC5 in fibroblasts was next investigated using an in vitro system. As the TME of mesenchymal-type CRC was characterized by elevated TGFβ levels and enhanced TGFβ signaling activity3, TGFβ1 (10 ng/ml) was administered in the culture media of human colonic CCD-18co fibroblasts. TGFβ treatment strongly activated the canonical TGFβ signaling, as evidenced by an increased phospho-SMAD3/total SMAD3 ratio (Fig. 3a) and elevated transcript (Supplementary Fig. 6a) and protein (Fig. 3a) levels of CAF marker α-SMA/ACTA2, ECM components COL1A1/fibronectin, and TXNDC5. In addition, TGFBR1 protein expression levels were increased in response to TGFβ1 stimulation, suggesting a positive feedback loop that mediates TGFβ1-induced fibroblast activation. Knocking down TXNDC5 in fibroblasts using shRNA, however, significantly reduced SMAD3 phosphorylation and the mRNA/protein upregulation of α-SMA and ECM proteins (COL1A1 and fibronectin) upon TGFβ1 stimulation (Fig. 3a and Supplementary Fig. 6a). In addition, TXNDC5 knockdown also led to a marked downregulation of TGFBR1 in fibroblasts, both at baseline and after TGFβ1 stimulation (Fig. 3a). Together, these findings suggest that TXNDC5 is essential for TGFβ-induced fibroblast activation and ECM production, likely by maintaining TGFBR1 protein expression and reinforcing the TGFβ1-TGFBR1 positive feedback loop. TXNDC5 is both required and sufficient to induce fibroblast activation by upregulating TGFBR1 and TGFβ signaling. a Immunoblots (n = 6 independent biological replicates per group) show that TGFβ1-induced upregulation of TXNDC5, TGFBR1, p-SMAD3, ECM proteins (fibronectin and COL1A1), and the CAF marker ɑ-SMA in control (shScr) human colonic fibroblasts CCD-18co are abolished by TXNDC5 knockdown (shTXNDC5). b TXNDC5, ɑ-SMA, and ECM proteins (fibronectin and COL1A1) are upregulated in human CRC CAFs compared with adjacent normal fibroblasts (adjNFs) (6 patients per group), alongside increased TGFBR1 expression and SMAD3 phosphorylation (4 patients per group). c TXNDC5 overexpression (OE) increases TGFBR1 levels, TGFβ signaling activity (reflected by SMAD3 phosphorylation), and fibrogenesis (COL1A1, fibronectin, and α-SMA expression), compared to empty vector-transduced controls (Empty), which is completely reversed by TGFBR1 depletion (shTGFBR1) (n = 6 independent biological replicates per group). Data are presented as mean ± SEM. P-values are determined using a two-sided Student t test. Exact P-values are indicated in figures. Source data are provided in the Source Data file. To further understand how TXNDC5 mechanistically modulates CAF activity and stroma formation in CRC, CAFs and adjacent normal fibroblasts (adjNFs) were isolated from human CRC and their adjacent normal colon tissues, respectively. Consistent with the observation made in intact CRC samples, both TXNDC5 protein (Fig. 3b) and transcript (Supplementary Fig. 6b) levels were significantly upregulated in CAFs, compared to adjNFs, accompanied by a marked increase in the expression levels of CAF marker α-SMA, ECM proteins (fibronectin, COL1A1) and p-SMAD3 and TGFBR1 (but not TGFBR2) (Fig. 3b). Importantly, TXNDC5 overexpression (OE) in human colonic fibroblasts was sufficient to drive the upregulation of transcript (Supplementary Fig. 6c) and protein (Fig. 3c) levels of α-SMA, fibronectin and COL1A1, as well as SMAD3 activation and increased TGFBR1 protein levels. In addition, TXNDC5 OE-induced fibroblast activation, ECM production, and SMAD3 phosphorylation were abolished by pharmacological inhibition of TGFBR1 (using LY364947, Supplementary Fig. 6d) or TGFBR1 knockdown (using shTGFBR1, Fig. 3c). Collectively, these findings indicate that elevated TXNDC5 levels drive fibroblast activation and ECM production by increasing TGFBR1 protein expression, thereby enhancing TGFβ signaling activity. TXNDC5 aids TGFBR1 folding and stabilization via its PDI activity Next, we went on to investigate the mechanisms by which TXNDC5 regulates TGFBR1 expression in fibroblasts. A cycloheximide (CHX) chase assay was performed in CCD-18co fibroblasts with TXNDC5 knockdown. As shown in Fig. 4A, knocking down TXNDC5 accelerated TGFBR1 degradation, compared to control (shScr), while TXNDC5 overexpression stabilized TGFBR1 and prolonged its protein half-life in fibroblasts (Fig. 4b). Based on this observation, we hypothesized that TXNDC5, as a protein disulfide isomerase (PDI), promotes the correct folding and stabilization of nascent proteins such as TGFBR1 in the ER19, allowing their subsequent trafficking and expression on cell surface; in the absence of TXNDC5, TGFBR1 may become misfolded and subjected to proteasome-mediated ER-associated degradation (ERAD)(30). Consistently, protein co-immunoprecipitation (co-IP, Fig. 4c) and proximity ligation assays (Supplementary Fig. 6e) revealed physical binding between TXNDC5 and TGFBR1. In addition, treatment with a proteasome inhibitor, MG132, reversed TGFBR1 protein downregulation caused by TXNDC5 knockdown in fibroblasts (Fig. 4d). TXNDC5, regulated by TGFβ1 through ER stress-ATF6 pathway, facilitates TGFBR1 folding and stabilization via its PDI activity. a Cycloheximide (CHX) chase assay (n = 3 independent biological replicates per group) shows that TXNDC5 knockdown (shTXNDC5) in fibroblasts accelerates TGFBR1 protein degradation compared with controls (shScr). b TXNDC5 overexpression (OE) slows TGFBR1 protein degradation compared with controls (Empty) (n = 3 independent biological replicates per group). c Co-immunoprecipitation assay demonstrates physical interaction between TXNDC5 and TGFBR1. Representative images from three independent experiments with similar results are shown. d TXNDC5 knockdown (shTXNDC5)-induced TGFBR1 downregulation was reversed by proteasome inhibition with MG132 (n = 6 independent biological replicates per group). e Overexpression of WT TXNDC5 (OE), but not the AAA mutant (AAA), induced TGFBR1 upregulation and fibrogenic activation (n = 6 independent biological replicates per group). f FRET-based assay shows that WT TXNDC5 overexpression enhances TGFBR1 protein folding in fibroblasts compared with cells transfected with empty or TXNDC5 AAA mutant vectors (n = 4 independent biological replicates per group). g TGFβ1-induced ER stress markers (BiP/ATF6) and TXNDC5 upregulation are reversed by ER stress inhibition with 4-PBA (n = 4 independent biological replicates per group). h Schematic of the human TXNDC5 promoter luciferase reporter construct, which harbors an ATF6-binding site. Luciferase reporter assay (n = 4 independent biological replicates per group) shows that ATF6-binding site deletion reduces the transcriptional activity of TXNDC5 promoter in response to TGFβ1 stimulation. Data are presented as mean ± SEM. P-values are determined using two-way ANOVA or a two-sided Student t test. Exact P-values are indicated in figures. Source data are provided in the Source Data file. To determine if TXNDC5 modulates TGFBR1 protein expression and fibrogenic responses in fibroblasts through its PDI function, we generated a mutant TXNDC5 construct with complete loss of PDI activity by replacing cysteines with alanines (CGHC - > AGHA) in both ends of all three of its thioredoxin domains (TXNDC5 AAA mutant, scheme shown in Fig. 4e, left panel). Indeed, protein expression levels of TGFBR1, ECM proteins (fibronectin and COL1A1), and α-SMA were markedly upregulated in fibroblasts overexpressing WT, but not AAA mutant, TXNDC5 (Fig. 4e, right panel). To test whether TXNDC5 regulates TGFBR1 protein folding in colonic fibroblasts, a fluorescence resonance energy transfer (FRET)-based protein folding assay5 was conducted by transfecting fibroblasts with an expression vector encoding TGFBR1 fused with cyan- (CFP) and yellow- (YFP) fluorescence proteins at its N- and C-termini, respectively (CFP-TGFBR1-YFP). The fluorescence resonance energy transferred from CFP to YFP would become detectable only when the CFP-TGFBR1-YFP fusion protein folds properly and brings CFP and YFP fluorophores in close proximity. Using a standard acceptor photobleaching FRET protocol, we showed that WT TXNDC5 overexpression in colonic fibroblasts significantly increased FRET efficiency, indicating enhanced folding of TGFBR1 ternary fusion protein, while overexpression of the TXNDC5 AAA mutant construct failed to promote TGFBR1 protein folding, as evidenced by low FRET efficiency (Fig. 4f). Taken together, these results demonstrate that TXNDC5, functioning as a PDI, stabilizes TGFBR1 protein by facilitating its proper folding, thereby enhancing TGFβ signaling, fibroblast activation, excessive ECM production and tumor fibrosis. TGFβ1 induces TXNDC5 by ER stress and ATF6 transcriptional regulation Hyperactivated TGFβ signaling is a key driver of the activation and transition of quiescent fibroblasts into CAFs, driven by increased intra-tumoral production and activation of TGFβ20,21. Consistent with this notion, TGFB1 mRNA is markedly upregulated in human (GSE39582, Supplementary Fig. 6f) and mouse AOM/DSS-induced (GSE8629922, Supplementary Fig. 6g) CRC, where Tgfb1 levels showed a strong positive correlation with that of Txndc5 (Supplementary Fig. 6h). Re-analysis of human CRC microarray data revealed TXNDC5 levels in CAFs positively correlated with TGFB1 expression in both cancer cells (Supplementary Fig. 6i) and CAFs (Supplementary Fig. 6j), suggesting that TXNDC5 in CAFs may be regulated by TGFβ1 derived from either tumor or stromal cells. To model the interaction between CRC cells and colonic fibroblasts in vitro, normal colonic fibroblasts were treated with conditioned media (CM) from CRC cell lines. The TGFB1 mRNA expression (Supplementary Fig. 7a) and TGFβ1 protein secretion into the medium (Supplementary Fig. 7b) were significantly higher in CMS4 CRC cell lines (Colo-320-HSR and NCI-H716), compared to CMS1 CRC cell line HCT116. CM from CMS4 CRC cells also induced marked upregulation of TXNDC5, CAF activation marker ACTA2, and ECM protein genes FN1 and COL1A1 in fibroblasts (Supplementary Fig. 7c). These responses were abrogated by TGFβ receptor inhibitor LY346947, indicating that CMS4 CRC cells drive TXNDC5 upregulation and fibrogenic responses in fibroblasts through increased TGFβ1 secretion. Interestingly, CM from CMS4 CRC cells also triggered increased TGFβ1 expression (Supplementary Fig. 7d) and secretion (Supplementary Fig. 7e) in fibroblasts, suggesting that fibroblasts primed by CRC cells can produce additional TGFβ1, thereby further amplifying TGFβ signaling activity. Next, we went on to determine how TGFβ signaling regulates TXNDC5 in fibroblasts. TGFβ1 treatment upregulated TXNDC5 in fibroblasts (Supplementary Fig. 7f), suggesting that TXNDC5 could be controlled by transcriptional regulators downstream of TGFβ signaling. We5–8 and others23,24 have shown that TGFβ1 stimulation can lead to increased ER stress and the activation of ER stress-induced transcriptional program in fibroblasts. Consistently, ER stress markers Bip and ATF6 p50 were markedly increased in AOM/DSS-induced CRC (Supplementary Fig. 7g) as well as in TGFβ1-treated fibroblasts (Fig. 4g). In addition, pharmacological inhibition of ER stress with 4-phenylbutyrate (4-PBA) abolished TGFβ1-induced upregulation of TXNDC5 mRNA (Supplementary Fig. 7f) and protein (Fig. 4g) in fibroblasts, suggesting that TGFβ1 induces TXNDC5 expression in fibroblasts through an ER stress-dependent mechanism. Indeed, promoter analysis identified a binding site of ATF6, one of the major transcription factors driving unfolded protein response upon increased ER stress, in the promoter region of both human and mouse TXNDC5/Txndc5. Knocking down ATF6 in fibroblasts abolished TGFβ1-induced TXNDC5 transcriptional upregulation (Supplementary Fig. 7h). Furthermore, using WT and ATF6 binding site-deleted (ΔATF6) TXNDC5 promoter luciferase reporter constructs, we showed that TGFβ1-induced TXNDC5 transcriptional activity in fibroblasts was completely abrogated upon removing the ATF6 binding site (Fig. 4h). Collectively, these findings indicate that elevated intra-tumoral TGFβ1 levels drive TXNDC5 upregulation in colonic fibroblasts/CAFs of CRC through increased ER stress and ATF6-dependent transcriptional control. CAF-specific Txndc5 deletion restores immune surveillance in CRC Both human CMS4 and mouse AOM/DSS-induced CRC exhibit an “immune-excluded” phenotype characterized by enhanced TGFβ signaling, reactive tumor stroma, abundant myeloid-derived suppressor cells (MDSCs), and a consequently immune suppressive TME, leading to poor responsiveness to immunotherapy25,26. Activated CAF and ECM contribute to this immune-suppressive milieu not only by providing biochemical/biomechanical signals but also by impacting immune cell infiltration and migration27,28. Because TXNDC5 promotes CAF activation and ECM production in CRC, we hypothesized that TXNDC5 in CAFs also contributes to the establishment of an immunosuppressive TME through tumor fibrosis. Indeed, IF staining showed that CD8+ T cells were predominantly excluded from the tumor invasive margin (IM) in control (Col1a2-CreERT2) animals, whereas a marked increase in CD8+ T cell infiltration into CRC tumor center (TC) was observed from Col1a2-CreERT2;Txndc5 fl/fl (Txndc5 cKO) mice (Fig. 5a). Meanwhile, the expression of immunosuppressive molecule PD-L1, which was primarily detected in the tumor epithelium and, to a lesser extent, in CAFs (Supplementary Fig. 2j), was also significantly lower in Txndc5 cKO than in control CRC tumors (Fig. 5a). Fibroblast-specific Txndc5 deletion reprograms the immunosuppressive niche and increases cytotoxic T-cell infiltration. a Increased CD8+ T cell infiltration and reduced PD-L1 expression are shown in Col1a2-CreERT2;Txndc5fl/fl tumors compared with controls (n = 6 mice per group). CD8+ T cell infiltration (%) is defined as infiltrated versus total CD8+ area (TC/TC + IM; TC: tumor center; IM: invasive margin). b Schematic of scRNA-seq analysis in Col1a2-CreERT2 (Cre) and Col1a2-CreERT2;Txndc5fl/fl (cKO) tumors. c Proportions of cancer, immune, fibroblast, and endothelial cells are presented as stacked columns. d Proportions of immune sub-clusters are presented as stacked columns. Fibroblast-specific Txndc5 deletion reduces myeloid cells (MDSCs, TAMs, TANs, DCs, mast cells) and increases T and B lymphocytes. (e) MDSC proportion (orange) decreases from 51.3% to 31.1% in cKO tumors compared with controls. f Proportions of T-cell subtypes are presented as stacked columns. g CD8 Teff (blue) to CD4 Treg (red) ratio increased from 0.4 to 2.1 in cKO tumors compared with controls. h Bubble plots show a shift toward M1-like TAM and N1-like TAN phenotype in cKO tumors. Color scale indicates the fold change in average gene expression between Cre and cKO tumors; bubble size represents the percentage of expressing cells. i, j In cKO tumors, CD8+ T cells exhibited increased stemness/proliferation, enhanced effector function, and reduced exhaustion (i), and heightened metabolic activity, with upregulated expression of glycolysis and oxidative phosphorylation (OXPHOS) marker genes, alongside reduced fatty acid oxidation (FAO) and reactive oxygen species (ROS) signatures (j), compared with controls. Data are presented as mean ± SEM. P-values are determined using a two-sided Student t-test for bulk data (exact P-values in figures) and a two-sided Wilcoxon rank-sum test for single-cell data (‡P < 0.05, *P < 0.001). Source data are provided in the Source Data file. MDSC: myeloid-derived suppressor cell; TAM: tumor-associated macrophage; TAN: tumor-associated neutrophil; DC: dendritic cell. To delineate the impact of CAF TXNDC5 on the immune TME in CRC, scRNA-Seq was performed using cells isolated from AOM/DSS-induced tumors in control and Txndc5 cKO mice (Fig. 5b) with an immune cell-enrichment isolation protocol, which increased the yield of immune cells from 26.1% to 35.2%, without altering the relative proportions of the immune cell subpopulations (Supplementary Figs. 8b and 2c). As shown in Fig. 5c and Supplementary Fig. 8a, four main cell compartments were identified: cancer cells, immune cells, fibroblasts, and endothelial cells. The composition of the immune cell population in CRC was drastically altered after fibroblast-specific Txndc5 deletion. The proportion of myeloid-lineage cells (MDSCs, TAMs, TANs, DCs, and mast cells) decreased from 79.6% to 43.7%, while the proportion of lymphoid-lineage cells increased, with B cells rising from 5.6% to 26.2% and T cells from 14.8% to 30.2% in Txndc5 cKO tumors compared to those of control mice (Fig. 5d). Re-clustering of myeloid lineage cells29 (Supplementary Fig. 8c, d) revealed significantly reduced MDSC infiltration (from 51.3% to 31.1%) in Txndc5 cKO tumors (Fig. 5e). In addition, re-clustering of T cell lineages30 (Fig. 5f and Supplementary Fig. 8e) showed a marked decrease in the proportion of regulatory T (CD4 Treg) and T gamma delta (γδT) cells, alongside an expansion of effector T (CD4 Teff and CD8 Teff) cells in Txndc5 cKO, compared to control, tumors. This led to a substantial increase in the ratio of cytotoxic CD8 Teffs to immunosuppressive Tregs (from 0.4 to 2.1, Fig. 5g). Further analyses revealed that deletion of Txndc5 in CAFs caused a functional polarization of tumor-associated macrophages (TAM) and neutrophils (TAN) from tumor-promoting (M2/N2) to tumor-suppressive (M1/N1) state (Fig. 5h). To exclude the potential systemic immunomodulatory effects resulting from Txndc5 deletion in Col1a2-expressing mesenchymal cells in bone marrow and lymph nodes, we performed flow cytometry-based immune profiling of peripheral blood and splenocytes from Col1a2-CreERT2 and Col1a2-CreERT2;Txndc5 fl/fl mice following tamoxifen induction. As shown in Supplementary Fig. 10a and b, deletion of Txndc5 in Col1a2-expressing cells did not result in discernible changes in the composition of lymphoid or myeloid cells, nor in the activation markers of T lymphocytes (CD69, PD-1 and CD25). These findings indicate that the immunomodulation observed in Col1a2-CreERT2;Txndc5 fl/fl CRC tumors did not arise from off-target effects on mesenchymal cells or disruption of systemic immune homeostasis. Moreover, CD8 Teffs in Txndc5 cKO tumors exhibited enhanced cytotoxic function, evidenced by increased expression of stemness/proliferation markers and genes involved in effector function, alongside reduced expression of exhaustion markers (Fig. 5i). CD8 Teff exhaustion has been linked to metabolic defects, characterized by suppressed glycolytic and OXPHOS activities, elevated fatty acid oxidation (FAO), increased mitochondrial damage and ROS production31. In Txndc5 cKO tumors, CD8 Teffs enhanced glycolysis and OXPHOS signatures, along with reduced expression of FAO and ROS marker genes, suggesting that TXNDC5 loss in CAFs drives mitochondrial reprogramming, facilitating the restoration of T cell effector function (Fig. 5j). Consistently, IF staining revealed a ~ 70% decrease in MDSC (CD11b+Gr1+) and Treg (CD4+FOXP3+), while showing a 3-fold increase in activated cytotoxic CD8+ T cells (CD8+ GzmB+) in Txndc5 cKO, compared to control, tumor sections (Fig. 6a). In addition, multiparametric immune-profiling using cytometry by time of flight (CyTOF) on Col1a2-CreERT2 and Txndc5 cKO tumors (Fig. 6b) confirmed the immune compartment remodeling in mouse CRC caused by Txndc5 deletion in CAFs. Specifically, there was a reduction in CD11b+ myeloid cells and an expansion in CD4+ and CD8+ T lymphocytes (Fig. 6c). Notably, neutrophils (Ly6G+), monocytes (Ly6C+) and MDSCs (Gr1+, S100A8+)32 were significantly decreased in Txndc5 cKO, compared to control, tumors (Fig. 6d). In addition, the proportion of tumor-promoting M2-like macrophages (CD206+F4/80+) decreased, whereas anti-tumor M1-like macrophages (CD80+F4/80+ and CD86+F4/80+) increased in Txndc5 cKO tumors (Fig. 6e). Loss of TXNDC5 in fibroblasts also led to an expansion of total (CD3+), helper (CD4+) and proliferative (Ki67+)/cytotoxic (GzmB+) CD8+ T lymphocytes (Fig. 6f, g) in tumors, along with a reduction in immunosuppressive regulatory (CD4+FOXP3+) and exhausted (PD-1+CD4+ and PD-1+CD8+) T cells (Fig. 6g). Fibroblast-specific Txndc5 deletion mitigates immunotolerance and activates a cytotoxic anti-tumor program. a CD11b+Gr1+ MDSCs, CD4+FOXP3+ Tregs are decreased, whereas CD8+GzmB+ cytotoxic T cells are increased, in Col1a2-CreERT2;Txndc5fl/fl tumors compared with controls (n = 6 mice per group). b Cytometry by time-of-flight (CyTOF) analysis pipeline for Col1a2-CreERT2 (Cre) and Col1a2-CreERT2;Txndc5fl/fl (cKO) tumors. c FlowSOM plots and stacked columns illustrate alterations in total myeloid (CD11b+), B-cell (CD19+), and T-cell (CD4+ and CD8+) populations in Col1a2-CreERT2 and Col1a2-CreERT2;Txndc5fl/fl tumors (pooled from n = 6 mice per group). d–g In cKO tumors, CyTOF profiling (n = 6 mice per group) reveals decreased Ly6G+ neutrophil, Ly6C+ monocyte and Gr1+/S100A8+ MDSC populations within the myeloid compartment (d); reduced CD206+ M2-like and increased CD80+ and CD86+ M1-like macrophage populations (e); increased CD3+, CD4+ and CD8+ T-cell infiltration (f); reduced immunosuppressive Foxp3+ and PD-1+ populations within CD4+ T cells, as well as enhanced proliferative (Ki67+) and cytotoxic (GzmB+) populations, along with reduced exhausted (PD-1+) populations in CD8+ T cells (g), compared with control tumors. Data are presented as mean ± SEM. P-values are determined using a two-sided Student t-test. Exact P-values are indicated in figures. Source data are provided in the Source Data file. Tumor-infiltrating B lymphocytes (TIL-Bs), including naive B cells (Bn), germinal center B cells (Bgc), memory B cells (Bm) and plasma cells (PC)33, play a key role in modulating anti-tumor immune responses in immunologically ‘hot’ tumors. The density of TIL-Bs, especially within the tertiary lymphoid structures (TLS), is positively correlated with a favorable prognosis and improved responsiveness to ICB34. In Txndc5 cKO tumors, the number of TIL-Bs, particularly Bn and Bm, was significantly expanded (Supplementary Fig. 8i), exhibiting a pronounced antitumor phenotype and upregulated B-cell chemokine receptors. In contrast, B cells from Col1a2-CreERT2 tumors displayed a tumor-promoting signature (Supplementary Fig. 8j). Consistent with increased TIL-B, the number and sizes of TLS formations were markedly increased in Txndc5 cKO tumors (Supplementary Fig. 8k). Taken together, these findings indicate that CAF-derived TXNDC5 facilitates tumor stroma formation, creating an immune-suppressive/exclusive TME. Targeted Txndc5 deletion in CAFs mitigated immunosuppression/exclusion, restoring immune surveillance in AOM/DSS-induced CRC. TXNDC5-driven tumor fibrosis creates an immunosuppressive niche Next, we investigated the mechanisms by which TXNDC5 in CAFs shapes the immune TME. First, IF staining for endothelial marker VE-cadherin in tumor sections from AOM/DSS-treated animals, perfused with fluorescence-conjugated lectin (as an indicator of vessel perfusion), revealed a marked increase in perfused vessels (lectin+) in Txndc5 cKO, compared to control, tumors (Fig. 7a). Using atomic force microscopy (AFM), we demonstrated that the tumor stiffness was significantly reduced following Txndc5 deletion in CAFs (Fig. 7b). This reduction in tumor stiffness correlated with decreased tumor fibrosis and increased perfused vessels, suggesting that TXNDC5-mediated desmoplasia increases CRC tumor stiffness, contributing to solid stress-induced hypoperfusion. Indeed, Txndc5 deletion in CAFs reduced fibrosis-induced solid stress, alleviating hypoperfusion and hypoxia in tumors, as evidenced by decreased pimonidazole (a chemical probe for hypoxia, Fig. 7c) staining, less HIF1α expression (Fig. 7d) and lower intra-tumoral lactate concentrations (Fig. 7e) in Txndc5 cKO, compared to control, tumors. Consistent with the notion that hypoxia and local lactate accumulation in the TME impair immune surveillance in tumor cores35, AOM/DSS tumors displayed high levels of the immune checkpoint protein PD-L1 and low numbers of infiltrating CD8+ T cells. These effects were completely reversed with the relief of hypoxia following Txndc5 deletion in CAFs (Fig. 5a). Furthermore, scRNA-Seq analysis of the myeloid compartment in Txndc5 cKO tumor revealed a significant reduction in the population of hypoxic (Hif1a+) cells (from 23.0% to 10.8%, Fig. 7f), accompanied by a decreased Cd274+ (PD-L1) population (from 35.4% to 21.1%, Fig. 7g), compared to controls. These findings are consistent with the notion that a hypoxic TME induces PD-L1 expression in TAMs and MDSCs, leading to Teff suppression36. Taken together, these results suggest that TXNDC5-induced CAF activation and tumor fibrosis drive a solid stress-induced hypoperfusion in CRC, thereby creating a hypoxic and immunosuppressive niche. Fibroblast-specific Txndc5 deletion alleviates fibrotic solid stress-induced hypoperfusion, reversing intra-tumoral hypoxia and lactate accumulation. a Lectin-perfused vessels increase in Col1a2-Cre;Txndc5fl/fl tumors compared to controls (n = 6 mice per group), suggesting relief of vessel compression and hypoperfusion. b Atomic force microscopy (AFM) experiments show reduced rigidity in collagen-enriched areas of Col1a2-CreERT2;Txndc5fl/fl tumors compared to controls ( > 200,000 force curves pooled from three mice per group). Representative images from three independent experiments with similar results are shown. c Pimonidazole (a hypoxia-specific probe) staining in CK19+ cells decreases in Col1a2-CreERT2;Txndc5fl/fl tumors compared with controls (n = 6 mice per group), indicating a reversal of intra-tumoral hypoxia. d The number of HIF1ɑ+ hypoxic tumor cells reduces in Col1a2-CreERT2;Txndc5fl/fl tumors compared with controls (n = 6 mice per group). e Lactate concentrations are lower in Col1a2-CreERT2;Txndc5fl/fl tumor lysates compared with controls (n = 6 mice per group). f–g scRNA-seq analyses show that reduced Hif1a+ (pink; 23.0% to 10.8%) (f) and Cd274+ (blue; 35.4% to 21.1%) (g) myeloid cells in Col1a2-CreERT2;Txndc5fl/fl tumors compared with controls. h Tumor epithelial cell proliferation (indicated by Ki67) decreased in Col1a2-CreERT2;Txndc5fl/fl tumors compared with controls (n = 6 mice per group). Data are presented as mean ± SEM. P-values are determined using a two-sided Student t test; two-sided Welch’s t test was applied to account for unequal variances in AFM data. Exact P-values are indicated in figures. Source data are provided in the Source Data file. TXNDC5-enriched CAFs drive tumor proliferation and immune tolerance Beyond facilitating the formation of a hypoxic, immunosuppressive milieu through tumor stroma remodeling, we hypothesized that TXNDC5-enriched CAFs may also promote CRC progression by impacting tumor cell growth. Indeed, fibroblast-specific Txndc5 deletion resulted in a marked reduction in the number of Ki67+ proliferative tumor cells (labeled with CK19) in AOM/DSS-induced tumors (Fig. 7h). In addition, CM derived from both TXNDC5hi CAF (Supplementary Fig. 8m, n) or TXNDC5 OE colonic fibroblasts (Supplementary Fig. 8o) accelerated CRC cell (HCT116) proliferation. CM derived from TGFβ1-treated colonic fibroblasts, mimicking CAF activation in vivo, also triggered CRC cell proliferation, the effects of which were abolished by TXNDC5 knockdown (Supplementary Fig. 8p). The epithelial compartment of CRC has been shown to recapitulate the multilineage processes of normal colon epithelial differentiation, wherein stem-like markers increase, and mature enterocyte markers decrease along the malignancy continuum37. Consistently, scRNA-Seq analyses of AOM/DSS-induced tumors revealed multilineage differentiation into four cell types: stem-like (Stem), intermediate (Int), enterocyte-like (Ent), and goblet-like (Gob) cells. Stem-like cells, which exhibit strong stemness, proliferative, and mesenchymal features, dominate tumors in control, but not in Txndc5 cKO, mice. Conversely, intermediate and enterocyte-like cell populations, expressing higher levels of epithelial and enterocyte marker genes, respectively, are more abundant in Txndc5 cKO tumors (Supplementary Fig. 8q, r). In addition, ECM and stiff TME are known to promote cancer cell proliferation, EMT, metabolic reprogramming, and chemoresistance by activating mechanosensors such as integrins, syndecans (SDCs), discoidin domain receptors (DDRs), CD44, Yes-associated protein (YAP), transcriptional coactivator with PDZ-binding motif (TAZ), and HIF1α in tumor cells38,39. Indeed, these mechanosensors were found to be lower in the cancer cells from Txndc5 cKO, compared to control, tumors (Supplementary Fig. 8s). Furthermore, Txndc5 deletion in CAFs decreased glycolysis while increasing OXPHOS gene expression in CRC cells (Supplementary Fig. 8t), indicating metabolic rewiring in CRC tumors upon ECM remodeling. As shown in Supplementary Figs. 7d, e and 6c, colonic fibroblasts primed with CMS4 tumors or overexpressed with TXNDC5 produced more TGFβ1, an immunosuppressive factor known to induce the differentiation of naïve CD4+ T cells into Tregs40, repress the anti-tumor effector function of CD8+ T cells41 and drive the immunosuppressive features of MDSCs and M2-like TAMs42. Consistently, the TGFβ pathway activation score (calculated using PROGENy43, Supplementary Fig. 9e) and the ECM/myofibroblastic gene signature (Supplementary Fig. 9f) were high in CAFs from AOM/DSS-induced CRC tumors but significantly reduced in CAFs with Txndc5 deletion. In line with the notion that TXNDC5-expressing CAFs could drive downstream immune tolerance through TGFβ, the TGFβ activity scores in CD4⁺ Tregs and CD8⁺ Teffs were markedly reduced in the mouse CRC tumors following fibroblast-specific Txndc5 deletion (Supplementary Fig. 8f). In addition, a cytokine array assay revealed that multiple immune-modulatory cytokines were increased in the CM from CRC CAF, compared to that from adjNF, including CXCL1, CXCL2, CXCL3, CSF3, CXCL5 and CCL8 (Supplementary Fig. 8u). Importantly, re-analysis of scRNA-Seq data from human CMS4 CRC (GSE132465) revealed a marked upregulation of receptors corresponding to these CAF-derived cytokines in multiple target immune cell populations (Supplementary Fig. 8v). Consistent with the observations made in human CRC CAFs, ligand-receptor analyses44 using scRNA-Seq data from control and Txndc5 cKO mouse CRC tumors identified multiple secreted signaling and ECM-receptor interaction pathways between CAFs and myeloid (Supplementary Fig. 8w), lymphoid (Supplementary Fig. 8x) and tumor cell linages (Supplementary Fig. 8y), most of which were markedly attenuated in the absence of TXNDC5 in CAFs. These TXNDC5-dependent, CAF-derived ligands (such as CXCL12 and galectin-9) and ECM components (such as FN1, COL1A1 and COL1A2) promote the recruitment of tumor-promoting MDSCs, M2-like macrophages and N2-like neutrophils45–49, drive Treg differentiation and activation and suppress anti-tumor Teffs and B cells50–52. Furthermore, CRC CAFs were found to produce factors (eg, BMP2/4, Epiregulin [Ereg], Progranulin [Grn], HGF, FN1, and COL1A1/2)53–59 that drive cancer cell proliferation and invasion, and the expression levels of these factors were reduced in Txndc5 cKO CRC CAFs (Supplementary Fig. 8y). Colorectal CAFs were firstly reported to be divided into two subtypes: CAF-A, expressing ECM-related molecules (MMP2, DCN and COL1A2), and CAF-B, characterized by myofibroblastic features (ACTA2, TAGLN, and PDGFA)60. Recently, phenotypically distinct CAF subtypes were identified in pancreatic ductal adenocarcinoma (PDAC)61,62, including ɑ-SMA+ myofibroblastic CAFs (myCAFs), IL-6+ inflammatory CAFs (iCAFs) and CD74+ antigen-presenting CAFs (apCAFs). These CAF subtypes have also been described across different cancer types, including CRC63,64. Employing published human (GSE132465, Supplementary Fig. 1l) and mouse (GSE134255, Supplementary Fig. 1o) CRC scRNA-seq datasets, we subclustered CAFs into iCAFs, myCAFs, and apCAFs, which revealed that TXNDC5 is expressed across all three major CAF subtypes. In addition, the proportion of TXNDC5-expressing cells remained comparable across these CAF subtypes. These findings were subsequently confirmed using our own scRNA-seq dataset derived from mouse AOM/DSS tumors (Supplementary Fig. 9a–c). Consistent with these scRNA-seq analyses, IF co-staining for TXNDC5 and CAF subtype markers (PDGFRɑ for iCAFs, ɑ-SMA for myCAFs and CD74 for apCAFs) in Col1a1-GFPTg CRC tumors revealed widespread expression of TXNDC5 across all CAF subtypes (Supplementary Fig. 2h). Further analyses revealed that the number of iCAFs, but not myCAFs or apCAFs, decreased upon global (Supplementary Fig. 3h) and fibroblast-specific (Supplementary Fig. 9a) Txndc5 deletion, suggesting that Txndc5 is required for the efficient polarization or maintenance of the iCAF state. Furthermore, this shifted polarization was accompanied by a marked downregulation of Il1r1 in Txndc5-deficient CAFs, observed in both the total population (Supplementary Fig. 9d) and specifically within iCAFs (Supplementary Fig. 9h). Given that the IL1α-IL1R signaling axis drives CAF polarization toward the iCAF phenotype and associated immunosuppressive programs in CRC65, these results suggest that TXNDC5 also plays a role in driving iCAF polarization by maintaining IL1R1 expression and signaling competence. Consistently, in human colonic fibroblasts, IL1R1 protein levels decreased upon TXNDC5 depletion due to accelerated degradation (Supplementary Fig. 6k, m), whereas IL1R1 protein was stabilized and upregulated following TXNDC5 overexpression (Supplementary Fig. 6l). Furthermore, the downregulation of IL1R1 protein induced by TXNDC5 depletion was reversed by proteasome inhibition with MG132 (Supplementary Fig. 6m). Moreover, a FRET-based IL1R1 folding assay showed that TXNDC5 overexpression increased IL1R1 folding, an effect abolished when the TXNDC5 catalytic domain was dysfunctional (AAA mutant, Supplementary Fig. 6n). These results suggest that TXNDC5 maintains IL1R1 stability via a folding-mediated proteostasis mechanism. In parallel, myCAFs from fibroblast-specific Txndc5-deficient tumors exhibit reduced TGFβ pathway activity (Supplementary Fig. 9i) and decreased expression of ECM-related genes (Supplementary Fig. 9j), consistent with the established role of TXNDC5 in regulating fibroblast activation and ECM production by modulating TGFBR1 protein stability and subsequent TGFβ signaling. Collectively, these results suggest a unified model in which TXNDC5 functions as a fibroblast-enriched chaperon, facilitating iCAF polarization via IL1R1 folding while simultaneously supporting myCAF function through TGFBR1 stabilization. In line with the scRNA-Seq data showing that TXNDC5 contributes critically to the fibroblast-immune crosstalk, fibroblast-specific Txndc5 deletion resulted in decreased expression of immunomodulatory genes, including Il11, Lif, Cxcl3, Cxcl5, Cxcl12, Ccl2, Tgfb1, Hgf, Bmp4, and C3 (Supplementary Fig. 9g). Several of these factors were more abundantly produced by TXNDC5-expressing human CAFs (Supplementary Fig. 8u) or predicted to have ligand–receptor interactions within the immunosuppressive microenvironment in Col1a2-CreERT2 control tumors (Supplementary Fig. 8w–y). Furthermore, co-culturing with TXNDC5-depleted colonic fibroblasts promotes, whereas TXNDC5-OE colonic fibroblasts suppress, the differentiation of peripheral blood mononuclear cell (PBMC)-derived CD4+CD25+ T cells into FOXP3+ Tregs (Supplementary Fig. 11a, b). In addition, PBMC-derived CD3+CD8+ T cells were more proliferative (Supplementary Fig. 11c, d), expressed higher levels of GzmB and lower levels of exhaustion marker PD-1 when co-culturing with TXNDC5-depleted colonic fibroblasts, the observations of which were completely reversed when co-culturing with TXNDC5-OE colonic fibroblasts (Supplementary Fig. 11e, f). Together, these data suggest that TXNDC5-enriched CAFs also actively modulate immune cell activities in CRC tumors through paracrine signaling. CAF-specific Txndc5 deletion synergistically boosts ICB therapy Excessive buildup of collagen and other ECM components within CMS4 and AOM/DSS-induced tumors creates a barrier that hinders immune cell infiltration. Furthermore, this dense network maintains an immunosuppressive TME by reducing oxygen supply and dysregulating signaling pathways within immune cells, contributing to resistance against ICB therapy66. The observation that TXNDC5 deletion in CAFs markedly reduced tumor fibrosis, corrected hypoxia, and increased immune cell infiltration suggests that inhibiting TXNDC5 expression or activity in CMS4 CRC could potentiate the anti-tumor effects of ICB. To test this hypothesis, fibroblast-specific Txndc5 deletion was induced by tamoxifen treatment after the final round of DSS treatment (week 16), followed by two weeks of anti-PD-1 therapy, in Col1a2-CreERT2 and Col1a2-CreERT2;Txndc5 fl/fl mice (Fig. 8a). In these experiments, colonic tumors were fully developed in situ by week 16 following AOM administration and three cycles of DSS (Supplementary Fig. 2b), consistent with previously published findings67. Both Txndc5 deletion and anti-PD-1 treatment were initiated after tumor establishment to mimic a clinically relevant therapeutic setting. As shown in Fig. 8a, anti-PD-1 therapy alone did not result in a discernible difference in tumor burden in Col1a2-CreERT2 mice compared to those treated with anti-IgG control, while Txndc5 deletion in CAFs significantly reduced AOM/DSS-induced tumor number and size. Notably, the combination of Txndc5 deletion in CAFs and anti-PD-1 therapy led to a significantly greater reduction in tumor burden compared to either intervention alone, indicating a synergistic anti-tumor effect when targeting both CAF TXNDC5 and immune checkpoint molecule concurrently (Fig. 8a). Indeed, Txndc5 deletion reduced CAF activation as evidenced by a marked decrease in α-SMA expression in tumors (Fig. 8b), which was accompanied by a significant reduction in tumor hypoxia status (Supplementary Fig. 12a), immunosuppressive MDSCs (CD11b+Gr1+, Fig. 8c) and Tregs (FOXP3+CD4+, Fig. 8d), as well as an increase in cytotoxic GzmB+ CD8 T cells (Fig. 8e). The number of Ki67+ proliferative tumor cells (labeled with CK19, Supplementary Fig. 12b) and iCAFs (labeled with PDGFRα, Supplementary Fig. 12c) was also reduced in the fibroblast-specific Txndc5-deleted tumors. Administration of anti-PD-1 therapy following Txndc5 deletion in CAFs further reduced MDSC/Treg infiltration, increased GzmB+ CD8 T cell infiltration, and significantly decreased tumor cell proliferation in AOM/DSS CRC tumors (Fig. 8c–e). Fibroblast-specific Txndc5 deletion synergizes with PD-1 blockade against CRC. a Tumor number and the sum of tumor diameters were assessed in Col1a2-CreERT2 and Col1a2-CreERT2;Txndc5fl/fl tumors treated with either anti-PD-1 or IgG control (n = 8 mice per group). Stacked columns show tumor size distribution ( < 2 mm, 2–4 mm, and > 4 mm). A combination of fibroblast-specific Txndc5 deletion with PD-1 blockade reduced the tumor burden compared with either approach alone (n = 8 mice per group), indicating a synergistic effect. b Fibroblast-specific Txndc5 deletion, but not PD-1 blockade, reduced CAF activation (indicated by ɑ-SMA) (n = 8 mice per group). c, d Fibroblast-specific Txndc5 deletion, but not PD-1 blockade alone, reduces immunosuppressive CD11b+Gr1+ MDSC (c) and CD4+FOXP3+ Treg (d) infiltration. Notably, PD-1 blockade combined with fibroblast-specific Txndc5 deletion synergistically decreased MDSC and Treg infiltration in tumors (n = 8 mice per group). e Fibroblast-specific Txndc5 deletion increases CD8+GzmB+ cytotoxic T-cell infiltration, whereas anti-PD-1 treatment alone shows minimal effects. Moreover, combined therapy leads to a pronounced cytotoxic T-cell infiltration in tumors (n = 8 mice per group). Data are presented as mean ± SEM. P-values are determined using a two-sided Student t test. Exact P-values are indicated in figures. Source data are provided in the Source Data file. Discussion Our findings demonstrate that ER protein TXNDC5 is upregulated in stromal fibroblasts of human and mouse CRC, driving CAF activation, ECM accumulation and tumor desmoplasia through augmented TGFβ1 signaling (Fig. 9a). This accelerates CRC progression by creating an immune-tolerant TME secondary to solid stress-induced hypoxia and CAF-derived paracrine signaling. In addition, TXNDC5 facilitates pro-tumorigenic iCAF differentiation by promoting the folding-mediated stabilization of IL1R1 protein. Consequently, fibroblast-specific deletion of Txndc5 alleviates CAF activation, tumor fibrosis and immune exclusion/evasion, leading to reduced CRC burden and increased ICB responsiveness (Fig. 9b). Schematic illustration of the mechanisms by which TXNDC5 contributes to CRC progression. a TXNDC5 is upregulated in CAFs in response to TGFβ1 through enhanced ER stress and ATF6-dependent transcriptional regulation. Elevated TXNDC5 levels drive CAF activation and intra-tumoral fibrogenesis by stabilizing and upregulating TGFBR1 protein expression, thereby forming a feed-forward loop that amplifies intra-tumoral TGFβ signaling. Illustrations were created using BioRender. Yang, K. (2025) https://BioRender.com/w09c175. b Increased TXNDC5 further enhances CAF activation and tumor fibrosis, resulting in solid stress-induced vessel compression, intra-tumoral hypoxia, and the formation of an immunosuppressive niche characterized by enhanced recruitment of immune-tolerant cells, including MDSCs and Tregs, and the exclusion of cytotoxic T cells. Illustrations were created using BioRender. Yang, K. (2025) https://BioRender.com/i96r570. Consistent with the characterization of mesenchymal-type CRC by increased TGFβ signaling3, CMS4 CRC cells produced more TGFβ1 than CMS1 cells (Supplementary Fig. 7a, b). Tumor-derived TGFβ1 not only triggered TGFβ1 production from neighboring CAFs (Supplementary Fig. 7d, e) but also induced TGFBR1 (Figs. 3, 4) and TGFB1 (Supplementary Fig. 6c) expression through a TXNDC5-dependent mechanism, creating a positive feedback loop that amplified TGFβ signaling activities in CAFs. TXNDC5 plays an essential role in maintaining heightened TGFβ signaling of mesenchymal-type CRC, as targeted deletion of Txndc5 in CAFs reduced TGFBR1 expression and TGFβ signaling activities in AOM/DSS-induced CRC, resulting in decreased tumor fibrosis. This disruption of the self-amplified TGFβ signaling feedback loop resolves tumor desmoplasia and re-establishes immune surveillance in CRC. Although the upregulation of TXNDC5 in CAFs is most prominent in mesenchymal-type CRC, its aberrant expression and enrichment in the tumor stroma are observed across CRC subtypes. For example, CMS1 tumors also contain abundant fibroblasts with elevated TXNDC5 expression compared with CMS2 and CMS3 tumors (Supplementary Fig. 1f). These findings suggest a TXNDC5-dependent, non-cell-autonomous CAF function contributing to CRC progression, regardless of the underlying genetic alterations in tumor cells. Consistent with this hypothesis, fibroblast-specific deletion of Txndc5 in Apcmin/+ mice, a CRC model driven solely by Apc mutation, also significantly reduces CAF activation, tumor fibrosis, and tumor burden. More studies are needed to test directly whether the tumor-promoting effects of TXNDC5 in CAFs are shared by non-CMS4 CRC subtypes or even by other cancer types. As shown in this study, TXNDC5 facilitates TGFBR1 protein upregulation and SMAD phosphorylation, thereby enhancing TGFβ signaling that drives fibroblast activation. Indeed, the TGFβ activation score (Supplementary Fig. 9e) and the ECM/myofibroblastic signature (Supplementary Fig. 9f) were significantly reduced in Txndc5-deficient CAFs. Furthermore, IF co-staining of TGFBR1 and p-SMAD3 with TXNDC5 in Col1a1-GFPTg CRC (in which CAFs were GFP-labeled) demonstrated significantly higher expression of TGFBR1 and p-SMAD3 in TXNDC5⁺ than in TXNDC5⁻ CAFs (Supplementary Fig. 2i). Consistently, scRNA-seq analyses on the Col1a2-CreERT2 control CRC tumors showed significantly higher TGFβ pathway activity (Supplementary Fig. 9k) and ECM gene levels (Supplementary Fig. 9l) in Txndc5⁺, compared to Txndc5⁻, CAFs. Re-analyses of published human (GSE132465) and mouse (GSE134255) CRC scRNA-seq datasets revealed higher TGFβ pathway activity (Supplementary Fig. 1m, p) and ECM gene expression (Supplementary Fig. 1n, q) in Txndc5⁺ compared to Txndc5⁻ CAFs. Beyond its role in driving tumor fibrosis, our findings demonstrate that TXNDC5 also facilitates pro-tumorigenic iCAF differentiation by enhancing the folding-mediated stabilization of IL1R1 protein. Notably, TXNDC5 is expressed across all three major CAF subtypes (myCAF, iCAF and apCAF) in both human (Supplementary Fig. 1l) and mouse (Supplementary Figs. 1o, 2h, and 9c) CRC tumors. The proportion of TXNDC5-expressing cells is comparable across these CAF subtypes, indicating that its expression is not confined to a single CAF population; rather, TXNDC5 likely contributes broadly to CAF function and differentiation through its chaperone function. In collagen-producing CAFs, including myCAFs (Supplementary Fig. 9i, j), TXNDC5 regulates fibroblast activation and ECM production through modulating TGFBR1 protein stability and subsequent TGFβ signaling activity, thereby promoting tumor fibrosis. Within the iCAF subset, however, TXNDC5 shapes an immunosuppressive TME by maintaining IL1R1 protein stability and signaling competence. This dual role reflects the established plasticity of CAF states, where iCAFs can transition toward myCAF programs under dominant TGFβ signaling68. Thus, TXNDC5 expression marks fibroblasts that are biochemically “primed” to respond to microenvironmental cues, while the prevailing signaling context determines the ultimate CAF phenotype. Together, our data support a unified model in which TXNDC5 acts as a central chaperone that maintains receptor folding and signaling capacity in CAFs, with TGFβ and IL1 signaling representing the dominant pathways underlying the fibrotic and immunosuppressive phenotypes observed in vivo. Pericytes are known to be a source of CAFs69 and have been associated with abnormal vasculature in advanced malignancies, contributing to a hypoxic tumor microenvironment, immune tolerance, and the establishment of a pre-metastatic niche70. Because both fibroblasts and pericytes express Col1a2, targeted deletion of Txndc5 in Col1a2-expressing cells may also reduce TXNDC5 expression in pericytes and perivascular fibroblasts, thereby contributing to altered tumor vasculature and intratumoral hypoxia. Further research is needed to test this hypothesis directly. Although a large number of immune cells are present in CMS4 CRC tumors, they are more prone to an adverse inflammatory immunophenotype characterized by the activation of TGFβ signaling, tumor fibrosis, and EMT. This malignant inflammatory TME blocks the anti-tumor effects of immune cells, resulting in immunosuppression and ICB resistance. The observation that Txndc5 deletion in CAFs restored the immune-competent TME and enabled the anti-tumor effects of ICB therapy in CRC suggests that concurrent inhibition of TXNDC5 and immune checkpoint molecules represents a potential strategy to treat mesenchymal-type CRC. Further studies may explore whether TXNDC5 targeting also boosts the anti-tumor efficacy of chemotherapy or emerging immunotherapies, such as bispecific antibodies and adoptive T cell therapies, against mesenchymal-type CRC. Despite being most enriched and upregulated in CAFs, TXNDC5 expression was also observed in immune cell compartments, including myeloid, B and T cell lineages in AOM/DSS tumors (Fig. 1h). Re-analysis of scRNA-Seq data from human CRC tumors and normal colonic tissues (GSE132465) revealed that TXNDC5 was indeed expressed in immune, particularly B, cells in both CRC and normal colonic tissues. However, in contrast to its expression pattern in fibroblasts, where TXNDC5 is particularly upregulated in CMS4 CRC tumors (Supplementary Fig. 1h), the B cell expression levels of TXNDC5 were relatively low in CMS4, compared to CMS1-3 CRC tumors and normal colonic tissues. Consistent with the human data, analysis of the mouse CRC scRNA-seq dataset revealed that, among non-fibroblastic populations, Txndc5 expression was largely restricted to plasma cells (PCs), which represent a minor subset of immune cells (approximately 3.3%) in both control and Txndc5 conditional knockout tumors (Supplementary Fig. 8i). Nevertheless, the overall cellular distribution of Txndc5 expression in the mouse CRC tumors was dominated by stromal fibroblasts (Supplementary Fig. 1i), closely mirroring the expression pattern observed in human CMS4 tumors (Supplementary Fig. 1h). Notably, TXNDC5 was recently reported to be upregulated in a subset of antibody-secreting tumor-infiltrating B lymphocytes (TIL-B)71. In line with this observation, our scRNA-seq analysis showed relatively high expression of Txndc5, together with multiple PDI family members, including P4hb, Pdia3, Pdia4, Pdia5, and Pdia6 (Supplementary Fig. 8l), specifically within the PC compartment among all B-lineage cells. As the primary function of PDIs, including TXNDC5, is to facilitate the proper folding and processing of secretory proteins like antibodies, their high expression levels observed in tumor-infiltrating plasma cells could be attributed to the high demand for PDIs to catalyze disulfide bond isomerization of the massively produced antibodies in these cells. More experiments are needed, however, to test this hypothesis directly. This study demonstrated that the suppression of CD8⁺ T cells and differentiation of Tregs were significantly reversed in both the TXNDC5 knockdown fibroblast co-culture system (Supplementary Fig. 11) and in fibroblast-specific Txndc5 knockout tumors (Fig. 6). Consistent with these results, the TGFβ activity score was decreased in CD4⁺ Treg and CD8⁺ Teff subpopulations within fibroblast-specific Txndc5-deleted tumors (Supplementary Fig. 8f). However, Txndc5 expression revealed no correlation with either Tgfbr1 expression or TGFβ activity in CD4⁺ Tregs and CD8⁺ Teff cells (Supplementary Fig. 8g, h). Although histological analysis revealed fewer Treg and greater cytotoxic T cell infiltration in the AOM/DSS-induced tumors of global Txndc5 knockout mice (Supplementary Fig. 3d, e), as in the fibroblast-specific Txndc5 knockout tumors (Fig. 6a), TGFBR1 expression levels in both FOXP3⁺ Tregs and CD8⁺ T cells remained unchanged (Supplementary Fig. 3f, g). These findings suggest that the immune tolerance within the T cell compartment relies more on secondary TGFβ signaling derived from CAFs rather than on the direct TXNDC5-TGFBR1 axis in tumor-infiltrating T cells. While our study establishes the therapeutic rationale for targeting TXNDC5 through fibroblast-specific genetic deletion, the development of potent and selective pharmacological inhibitors remains an essential next step toward clinical translation. Future investigations focusing on the structure-guided design of TXNDC5 small-molecule inhibitors or the development of neutralizing agents will be required to validate these findings in a pharmacological context. Given that TXNDC5-mediated proteostasis represents a central regulatory node linking fibrotic and immunosuppressive signaling, such agents represent a potential class of “stroma-reprogramming” therapies to overcome PD-1 resistance in mesenchymal-type CRC. In conclusion, our data demonstrate a role of fibroblast TXNDC5 in CRC progression through enhancing CAF activation, stroma formation, and immune evasion. Combining immune checkpoint blockade with TXNDC5 deletion synergistically improved anti-tumor effects, highlighting a potential therapeutic strategy for CRC. Methods Ethics All research activities in this study complied with all relevant ethical regulations. Animal handling and experimental procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of National Taiwan University College of Medicine (Approval Numbers: 20170012 and 20220073). The maximal tumor burden permitted by the IACUC was 20 mm in diameter for a single tumor, or a combined diameter of no more than 20 mm for the two largest tumors in cases of multiple tumor development. We confirm that this maximal permitted tumor burden was strictly adhered to and was not exceeded in any of the experimental animals. Mice were monitored daily for signs of distress, and any animal reaching the ethical endpoint was euthanized via CO2 inhalation followed by cervical dislocation. All studies involving human colonic and CRC tissues were approved by the E-Da Hospital Institutional Review Board (Approval Number: EMRP-108-091). Informed consent was obtained from all participants, and no compensation was provided for their participation. A total of 14 CRC patients, aged 45–78 years (median age: 62), including 8 males and 6 females, were recruited randomly. Clinical information beyond diagnosis was not utilized in the primary analysis of this manuscript. Biological sex was determined based on clinical records. While sex and gender were not specific considerations in the study design—and no sex-disaggregated statistical analysis was performed due to the sample size—all individual-level clinicopathological data (including age, biological sex, and TNM stage) for each participant recruited between October 2019 and July 2020 are documented in Supplementary Table 1 and the Source Data to ensure data transparency. Mice All mouse strains in this study were on a C57BL/6 background and maintained under specific pathogen-free conditions. Txndc5 -/- and Txndc5 fl/fl mice were generated using CRISPR/Cas9 genome editing, targeting exons 2 and 3 of the Txndc5 gene to induce a frameshift mutation and functional protein loss5–8. To generate inducible Txndc5 conditional knockout lines, Txndc5 fl/fl mice were crossed with the following transgenic strains: fibroblast-targeted Col1a2-CreERT2 (Jackson Laboratory, #029567), epithelial cell-targeted Cdx2-Cre (Jackson Laboratory, #009350), and endothelial cell-targeted Cdh5-CreERT2 (kindly provided by Dr. Ralf H. Adams). For activation of the Cre-ERT system, tamoxifen (80 mg/kg, dissolved in olive oil) was injected intraperitoneally for 4 consecutive days in 6-week-old mice. Col1a1-GFPTg mice (kindly provided by Dr. David Brenner) served as a reporter mouse line to visualize collagen-producing CAFs. Apcmin/+ mice (kindly provided by Dr. Linda Chia-Hui Yu) were utilized for spontaneous intestinal tumorigenesis and crossed with Col1a2Cre/ERT2 or Col1a2Cre/ERT2;Txndc5 fl/fl mice to achieve fibroblast-specific Txndc5 deletion. For all experiments, adult male mice were exclusively utilized to eliminate the potential confounding effects of cyclic hormonal variations and to ensure maximum consistency72 in tumor incidence, disease progression, and the derived cellular phenotypes. For the AOM/DSS-induced CRC model, 8-week-old mice received a single intraperitoneal injection of 12.5 mg/kg AOM (Sigma-Aldrich, #A5486), followed by 7 days of regular diet and drinking water. Subsequently, mice underwent 3 cycles of DSS (MP Biomedicals, #9011-18-1) treatment (3.5% DSS in the drinking water for 5 days, followed by 14-days of regular water. Body weight, occult or gross rectal bleeding, and stool consistency were monitored daily. Bleeding was scored as 0 when there was no blood in the Hemoccult test; 1, for a positive Hemoccult result; 2, for slight bleeding; or 3, for gross bleeding. For stool consistency, 0 points were given for well-formed pellets; 1, for semi-formed stools that did not adhere to the anus; 2, for pasty stools; and 3, for liquid stools that adhered to the anus. To evaluate the therapeutic impact of Txndc5 deletion and/or ICB treatment, 80 mg/kg tamoxifen, as well as 200 μg/mouse anti-PD-L1 (Leinco Technologies, #P372) or anti-IgG (Leinco Technologies, # R1367, isotype control) antibodies, were injected intraperitoneally every other day for 2 weeks at the end of the AOM/DSS induction protocol (week 16). To quantify tumor hypoxia and vessel perfusion were quantified by intravenous injection of 60 mg/kg pimonidazole (Hypoxyprobe, # HP2-100Kit) and 100 μg fluorescence-labeled lectin (Vector Laboratories, # DL-1177-1) at 1 h and 5 min before sacrifice. Experimental animals were assigned to groups on a rolling basis using block randomization to ensure an adequate sample size for each condition. Unique identifiers were utilized to blind the experimenters to genotype and treatment group during data collection and analysis. Human samples Multiple 0.5-1 cm biopsies were collected from CRC patients (n = 14, including tumors and adjacent normal tissues). Samples were either paraffin-embedded for immunofluorescence and immunohistochemical staining or flash-frozen in liquid nitrogen and stored at − 80 °C for RNA and protein isolation. Protein lysates were generated from the colonic tissues through homogenization in deionized urea lysis buffer (8 M urea, 1% SDS, 10% glycerol, 60 mM Tris, pH 6.8, 0.002% pyronin Y, 5% β-mercaptoethanol), while RNA was isolated using TRIzol reagent (Thermo Fisher Scientific, #15596018). Human cancer-associated fibroblasts (CAFs) and normal fibroblasts (adjNFs) were isolated from collected CRC samples with adjacent normal tissues, respectively. Tissues were digested with 10 mg type 4 collagenase (Worthington Biochemical, # LS004189) at 37 °C for 1 h, and the resulting cells were cultured in FM medium (ScienCell, #2301) containing 2% FBS, 1% FGS and 1% penicillin/streptomycin solution; cells within three passages were used in experiments. For large-scale clinicopathological analysis, human CRC tissue microarrays (COC1262; Pantomics, CA, USA) were employed, encompassing 84 tumor cases (Grade I, n = 6; Grade I–II, n = 16; Grade II, n = 48; Grade II–III, n = 10; Grade III, n = 4) and 42 normal colon controls. For quantification of TXNDC5 intensity, 3–5 regions of interest (ROIs) were selected per case, encompassing both tumor and stromal compartments while specifically excluding necrotic or tissue-disrupted areas. The mean staining intensity for each case was calculated and normalized to the average value of the normal control group. Cell lines Normal colonic fibroblast CCD-18co (American Type Culture Collection, CRL-1459™) was grown in MEM α medium with 10% heat-inactivated FBS and 1% P/S and stimulated by 10 ng/μL TGF-β1 (PeproTech, #100-21 C) for 48 hours. HCT116, COLO-320-HSR and NCI-H716 (Bioresource Collection and Research Center, #60349, #60109, and #60517) cells were grown in RPMI 1640 medium (Thermo Fisher Scientific, #31800022) with 10% heat-inactivated FBS and 1% P/S. For experiments using conditioned medium, 2 × 105 CRC cells or fibroblasts were seeded in a 6-well plate for 3 days. Supernatants were collected, centrifuged, and filtered before use. For the co-culture assay, 1 × 104 fibroblast-seeded cell culture inserts were placed on 24-well plates (Corning, #353095) with isolated CD4+CD25+ or CD8+ T cells cultured for 3 days. All cells were cultured in an incubator with humidified air supplemented with 5% CO2 at 37 °C. Bioinformatics and data Re-analysis Transcriptomic data from publicly available CRC datasets were retrieved from the Gene Expression Omnibus (GEO). Analysis of TXNDC5 expression in sorted cell populations (fibroblasts, endothelial cells, and tumor epithelial cells) was performed using the GSE39396 dataset (n = 6 patients). Subtype-specific expression analysis was conducted using the GSE103479 dataset according to the Consensus Molecular Subtypes (CMS1–4) classification. For survival analysis, patients from the GSE63216 dataset were stratified into TXNDC5high and TXNDC5low groups based on median expression; the hazard ratio (HR) and P-values were determined using the Kaplan-Meier method. Correlation analyses between TXNDC5 and fibrogenic genes (COL1A1 and ACTA2) were performed using linear regression. Histology, immunohistochemistry (IHC), and immunofluorescence (IF) Human and mouse colonic normal/tumor samples were fixed in 4% paraformaldehyde for 2 h, embedded in paraffin, and sectioned in 5 μm thickness for Masson’s trichrome (Sigma-Aldrich, # HT15) and Picrosirius red (Abcam, # ab246832) staining. For IHC staining (using Leica Biosystems, #RE7140-CE), sections were deparaffinized, followed by antigen retrieval, treated with Peroxidase Block, and incubated with primary antibody (anti-TXNDC5: Proteintech, #19834-1-AP, polyclonal, 1:5000) overnight at 4 °C. Tissue sections were then washed, incubated with Post Primary for 1 h at room temperature, and treated with Novolink Polymer for 15 min. The sections were then developed with DAB solution, counterstained with hematoxylin, and mounted with mounting medium. For frozen section, samples after fixation were dehydrated in 30% sucrose, and embedded in Tissue-Tek O.C.T. (SAKURA, #4583). These 5 μm-thick sections were permeabilized and blocked with 0.3% Tween 20 in 1% BSA (Sigma-Aldrich, # A7906) for 1 hour, incubated with primary antibody (anti-CK19: Abcam, #ab52625, EP1580Y, 1:500; anti-αSMA: Abcam, #ab5694, polyclonal, 1:500; anti-CD11b: Abcam, #ab133357, EPR1344, 1:1000; anti-VE-cadherin: Abcam, #ab205336, EPR19138, 1:500; anti-Vimentin, Abcam, #ab8979, RV203, 1:100; anti-PD-L1: Cell Signaling Technology, #64988, D8W4Z, 1:200; anti-CD8: Abcam, #ab17147, YTS169.4, 1:500; anti-Gr1: Abcam, #ab25377, RB6-8C5, 1:100; anti-CD4: Abcam, #ab183685, EPR19514, 1:500; anti-FOXP3: Abcam, #ab215206, EPR22102-37, 1:1000; anti-GzmB: Abcam, #ab255598, EPR22645-177, 1:1000; anti-HIF1a: Novus Biologicals, #NB100-479, polyclonal, 1:100; anti-Ki67: Genetex, #GTX16667, polyclonal, 1:100; anti-TGFBR1: Thermo Fisher Scientific, #PA5-32631, polyclonal, 1:100; anti-CD45R/B220: Biolegend, #103202, RA3-6B2, 1:100) overnight at 4 °C. After washing, the sections were treated with fluorescent secondary antibodies (Biolegend, #405319, #405418, #406404, #407409, #405326, #406418, #405416, #406618, and #A-21244, 1:500) at room temperature for 1 h. The sections were then washed and mounted with DAPI Fluoromount-G (SouthernBiotech, #0100-20). High-resolution images were acquired using the TCS SP8 X STED 3X confocal microscope system (Leica Microsystems). Wide-field images were obtained using the EVOS M7000 Imaging System (Thermo Fisher). Fluorescence signal was quantified using ImageJ v1.54 f (National Institutes Health). Multiphoton microscopy and second-harmonic generation (SHG) imaging The SHG imaging of the colon tissue sections was performed using a multiphoton microscopic system73. In short, the excitation source was set up through a wavelength-tunable Ti:Sapphire laser (Spectra Physics, Mai-Tai® DeepSee Tsunami) pumped by a diode-pumped solid-state laser (Spectra Physics, Millennia X). A Galvanometer-driver x-y mirror scanning system (Southport Corp.) was used to scan the 950 nm output of the Ti:Sapphire laser in the focal plane. The laser power was set to be about 61 mW on the tissue section to optimize image quality without laser ablation. A high-numerical-aperture water immersion 20x/NA1.0 objective lens was used for high-resolution imaging. In addition, the SHG signals were spectrally separated by band-pass filters of 473/10 nm (FF01-473/10-25). The scan resolution was set to 1024 × 1024 pixels. For each of the sections scanned for SHG, 2-photon-excited fluorescence (TPEF) imaging was obtained to show the profile of the scanned tissue. Immunoblot analysis Cells and tissues were homogenized in 1x Cell Lysis Buffer (Cell Signaling Technology, #9803) supplemented with protease and phosphatase inhibitors (Omics, # C0001 and # C0004). After centrifugation at 10,000 g for 15 minutes at 4 °C, the supernatant was collected, and the protein concentrations were quantified using Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, #23225) according to the manufacturer’s recommendation. Protein lysates were mixed with 4 x protein loading buffer (Omics, # JB06-F002) and boiled for 5 min at 95 °C. Protein samples were fractionated on 8% or 10% SDS-PAGE gel, transferred to PVDF membranes, and incubated in blocking buffer (5% BSA, 0.1% Tween 20 in PBS). Membranes were incubated with primary antibodies (anti-Fibronectin: BD Biosciences, #610077, clone 10, 1:1000; anti-COL1A1: OriGene, #TA309096, polyclonal, 1:1000; anti-BiP: Cell Signaling Technology, #3177, C50B12, 1:1000; anti-TGFBR1: Thermo Fisher Scientific, #PA5-32631, polyclonal, 1:1000; anti-TGFBR2: OriGene, #TA311643, polyclonal, 1:1000; anti-t-SMAD3: Abcam, #ab40854, EP568Y, 1:1000; anti-p-SMAD3: Abcam, #ab52903, EP823Y, 1:1000; anti-ATF6: Cosmo Bio, #BAM-73-505-EX, clone 1-7, 1:500; anti-IL1R1: Abcam, #ab106278, EPR4116, 1:1000; anti-β-actin: OriGene, #TA811000, OTI1B12, 1:5000; anti-GAPDH: Thermo Fisher Scientific, #MA5-15738, GA1R, 1:5000) overnight at 4 °C. After washing, blots were incubated with HRP-conjugated anti-mouse or anti-rabbit IgG secondary antibodies (Thermo Fisher Scientific, #31430 and #31460, 1:5000), washed with PBS-T, and developed using WesternBright ECL HRP substrate (Advansta, #K-12045-D50). Protein bands were imaged using the ChemiDoc MP system (BioRad Laboratories). The intensity of protein bands was quantified with ImageLab v5.1 (BioRad Laboratories). RNA extraction and quantitative real-time PCR (qRT-PCR) Total RNA was extracted from cells and tissues using TRIzol (Thermo Fisher Scientific, #15596018) according to the manufacturer’s protocols. Maxima First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, #K1641) was used for RNA reverse transcription, followed by SYBR Green-based qRT-PCR6. Transcript levels of each gene were normalized to the endogenous control gene HPRT and expressed relative to the mean value of the control samples. The primers are listed in Supplementary Table 2. Protein stability assay TGFBR1 protein stability was measured using a cycloheximide (CHX) pulse chase assay6. In short, control and TXNDC5 knockdown or overexpressed human colonic fibroblasts were treated with 100 μg/mL cycloheximide (Cayman Chemical, #14126) to block protein translation. TGFBR1 protein levels were quantified with immunoblotting at 0, 4, and 8 h following cycloheximide treatment and expressed as percentages relative to the 0 h time point. Co-immunoprecipitation (Co-IP) The human TXNDC5 (pLAS2w-p.puro-TXNDC5) plasmid was transfected into human colonic fibroblasts CCD-18co for 48 hours. Protein lysates were incubated with rabbit anti-TXNDC5 (Proteintech, #19834-1-AP, polyclonal, 5 μg per 500 μg lysate) or mouse IgG (Cell Signaling Technology, #5415, 5 μg per 500 μg lysate) isotype control. Immunoprecipitation of TXNDC5 was conducted using the Pierce Magnetic Protein A/G IP Kit (Thermo Fisher Scientific, #88804) according to the manufacturer’s protocol. The co-immunoprecipitated proteins were eluted in 1x SDS loading buffer, boiled for 5 min at 95 °C, and subjected to immunoblotting using the antibodies and procedures described above. In situ proximity ligation assay (PLA) To determine the physical interaction between proteins (TXNDC5 and TGFBR1), Duolink in situ PLA (Sigma Aldrich, #DUO92002-30RXN, #DUO92004-30RXN, and #DUO92008-30RXN) was conducted according to the manufacturer’s instructions. Briefly, 4% PFA–fixed human colonic fibroblasts CCD-18co were permeabilized and blocked with 0.3% Tween 20 in 1% BSA for 1 h, followed by incubation with anti-TXNDC5 (Proteintech, #19834-1-AP, polyclonal, 1:200) and anti-TGFBR1 (Abnova, #H00007046-A01, 2B1, 1:100) overnight at 4 °C. After washing, cells were incubated with PLA probes for 1 hour at 37 °C, followed by ligation and amplification. PLA signals (red fluorescence), indicating physical proximity between TXNDC5 and TGFBR1, were captured using the EVOS M7000 Imaging System (Thermo Fisher). Negative controls included cells incubated without primary antibodies or with a combination of anti-TXNDC5 and anti-GAPDH (Thermo Fisher Scientific, # MA5-15738, GA1R, 1:500) to confirm the specificity of the interaction signals. FRET-based protein folding assay Plasmids encoding human TGFBR1 fused with a CFP at the N-terminus and a YFP at the C-terminus were constructed (CFP-hTGFBR1-YFP) and transfected into human colonic fibroblasts. When TGFBR1 is correctly folded, fluorescence resonance energy will be transferred from donor (CFP) to acceptor (YFP) fluorophore. In these experiments, a plasmid encoding a fusion protein that connects CFP and YFP with a short connecting peptide of 15 amino acid residues (CFP-YFP) was used as a positive control, whereas a plasmid with a di-cistronic operon for co-expression of isolated CFP and YFP (CFP-IRES-YFP) was used as a negative control. WT and AAA-mutant TXNDC5- or empty vector-expressed human colonic fibroblasts were transfected with CFP-hTGFBR1-YFP for 48 h before image acquisition. A standard photobleaching FRET protocol was conducted using a Carl Zeiss Confocal Laser Scanning Microscope 780 (LSM 780) to determine FRET efficiency. Photobleaching was performed by repeated laser scanning at a selected target cell using 514 nm at maximal intensity. For CFP, an excitation wavelength of 458 nm and an emission wavelength of 470–500 nm were used. For YFP, an excitation wavelength of 514 nm and an emission wavelength of 525–545 nm were used. ZEN Black 2.3 software (Carl Zeiss) was used to quantify and analyze the FRET signals. The difference between total fluorescence before and after photobleaching was divided to postbleaching total fluorescence of the entire target cell as FRET efficiency. For each experimental condition, a nonbleached cell was used as the control. Lentiviral transduction for gene knockdown and overexpression Lentivirus particles packed with shRNAs pLKO-shTXNDC5, pLKO-shATF6, and pLKO-shTGFBR1 were applied to knockdown TXNDC5, ATF6, and TGFBR1, respectively. Lentiviral particles containing pLKO-shScr were used as a nontargeting control. To overexpress TXNDC5, lentiviral particles containing full-length human TXNDC5 cDNA (pLAS2w.pPuro-TXNDC5) were used, while lentiviral particles expressing empty pLAS2w.pPuro vector was used as a control. Cell proliferation assay Fibroblast and tumor cell proliferation rates were assessed using a Cell Proliferation ELISA BrdU kit (Roche, #11647229001) according to the manufacturer’s recommendations. In short, cells were seeded and labeled with 10 μmol/L bromodeoxyuridine (BrdU) in a 96-well plate for 48 h at 37 °C, followed by a colorimetric ELISA assay using a spectrophotometer (Bio-Tek, Synergy HT, absorbance at 370 nm, reference wavelength 492 nm). TXNDC5 promoter luciferase reporter activity assay TXNDC5 promoter luciferase reporter constructs with (pGL3-hTXNDC5 ∆ATF6, + 642 to + 653 deleted) or without (pGL3-hTXNDC5 WT) deletion of its ATF6 binding site were transfected into human colonic fibroblasts (1 μg of each plasmid) for 24 h, followed by TGFβ1 (PeproTech, #100-21 C, 10 ng/ml) stimulation for 48 h in a 12-well plate. Meanwhile, a Renilla luciferase expression vector was co-transfected and used as a control. A Luc-PairTM Duo-Luciferase Assay Kit 2.0 (GeneCopoeia, # LF001) was used to quantify luciferase activity according to the manufacturer’s instructions on a plate reader. Flow cytometry analysis For surface marker staining, the cell suspension was stained with fluorescence-conjugated antibodies (FITC anti-CD3: Biolegend, #300405, UCHT1, 1:100; PerCP anti-CD8: Biolegend, #344707, SK1, 1:100; FITC anti-CD4: Biolegend, #317408, OKT4, 1:100; PerCP anti-CD25: Biolegend, #356132, BC96, 1:100; APC/Cy7 anti-PD-1: Biolegend, #329922, EH12.2H7, 1:100; FITC CD45: BD Biosciences, #553079, 30-F11, 1:200; BV650 CD11b: BD Biosciences, #563402, M1/70, 1:200; BV605 Ly6G: BioLegend, #127639, 1A8, 1:200; BV785 Ly6C: BioLegend, #128041, HK1.4, 1:200; PE/Cy7 CD64: BioLegend, #139314, X54-5/7.1, 1:100; BV785 CD19: BioLegend, #115543, 1D3, 1:200; BV605 anti-CD3: BD Biosciences, #563004, 17A2, 1:100; PE anti-CD4: BioLegend, #100408, GK1.5, 1:200; PerCP anti-CD8: BioLegend, #100732, 53-6.7, 1:200; BUV395 anti-CD25: BD Biosciences, #564022, PC61, 1:100; PE anti-CD69: BioLegend, #104507, H1.2F3, 1:100; PE anti-PD-1: BD Biosciences, #551892, J43, 1:100) for 30 mins at 4 °C. For intracellular staining, cells were fixed and permeabilized using Cytofix/Cytoperm kit (BD Biosciences, #554714) according to the manufacturer’s instructions before staining with fluorescence-conjugated antibody (PE/Cy7 anti-GzmB: Biolegend, #372214, QA16A02, 1:100) for 30 mins at 4 °C. To perform nuclear FOXP3 staining, the FOXP3 staining buffer set (Thermo Fisher Scientific, #00-5523-00) was used to fix and permeabilize the cells before staining with BV421 FOXP3 antibody (Biolegend, #320124, 206D, 1:100) or BV421 anti-FOXP3 (BioLegend, #126419, MF-14, 1:100) for 30 mins at 4 °C. Dead cells were stained with fixable viability kit (Biolegend, #423107 and #423105). The cell suspension was then subjected to flow cytometry analysis using LSRFortessa (BD Biosciences). Analysis of the acquired data was performed using FlowJo v10.8 (BD Life Sciences). 10x library preparation and sequencing Tumors were minced thoroughly and digested using the mouse tumor dissociation kit (Miltenyi Biotec, #130-096-730) according to the manufacturer’s recommendations. Dead cells were stained with fixable viability kit (Biolegend, #423105) and removed by sorting. scRNA-seq libraries were generated with the use of the Chromium Single Cell 3’ Reagent Kit v2 (10x Genomics, #120267). 2000 cells/μL tumor-dissociated cells were loaded onto the 10X Chromium Single Cell Platform (10X Genomics, #1000414, #1000496 and #1000422). Generation of gel beads in emulsion (GEMs), barcoding, GEM-RT clean-up, complementary DNA amplification and library construction were all performed according to the manufacturer’s protocol. The final library pool was sequenced on an Illumina NovaSeq 6000 instrument using an S1 flow cell. Total cell recovery for Col1a2-CreERT2 and Col1a2-CreERT2; Txndc5 fl/fl tumors was 21,439 cells captured at a mean depth of 62,466 reads per cell. Analysis of scRNA sequencing data For scRNA-seq, raw sequence reads were quality-checked using FastQC software. The Cell Ranger version 3.0 software (10X Genomics) was used to process, align, and summarize unique molecular identifier (UMI) counts against the mouse mm10 assembly reference genome analysis set, obtained from the University of California, Santa Cruz (UCSC). Downstream bioinformatic analyses were performed using R version 4.0 (RStudio). Raw UMI count matrices were filtered with the following criteria: barcodes with fewer than 1,024 and more than 131,072 UMIs; with more than 15% mitochondrial UMIs; and with fewer than 500 genes expressed. Following normalization using the SCTransform function in Seurat v474, the principal components were computed. The top principal components were used for the t-SNE dimensionality reduction. The graph-based clustered cells were annotated based on the biomarker analysis and the canonical markers from the literature. Differential gene expression of reported markers was illustrated as a bubble plot according to the mean expression normalized by the z-score and the percentage of cells whose value was greater than 0.0. MAGIC75 imputation (t = 3, knn = 15) was applied to facilitate visualization of TXNDC5 expression trends by mitigating dropout effects while preserving local transcriptomic structure. Pathway activation scores were calculated using PROGENy43. With manifold learning and quantitative contrasts, ligand-receptor analysis was performed using the CellChat DB package44 for network analysis and pattern recognition approaches, to predict major signaling between CAFs in Col1a2-CreERT2 and Col1a2-CreERT2; Txndc5 fl/fl tumors with different cell subpopulations from scRNA-Seq data. Mass cytometry by time-of-flight immunoprofiling Single-cell suspensions from Col1a2-CreERT2 and Col1a2-CreERT2; Txndc5 fl/fl tumors were sorted using fluorescence-activated cell sorter (FACS) Aria III (BD Biosciences) to remove noncellular contaminants. Samples were resuspended, and viability staining was performed using 25 μmol/L cisplatin (Sigma-Aldrich, #P4394) working solution. Fixation was then performed using 1.6% formaldehyde solution for storage. After re-suspension, cell barcoding using diluted dye was performed, followed by FcR blocking, surface and intracellular staining using metal-tagged antibodies (CD11b: BioLegend, #101202, M1/70, 1:100; Ly6C: BioLegend, #128002, HK1.4, 1:100; Ly6G: BioLegend, #127602, 1A8, 1:100; Gr-1: BioLegend, #108402, RB6-8C5, 1:100; S100A8: Novus Biologicals, #NBP2-25273, 63N13G5, 1:100; F4/80: BioLegend, #123143, BM8, 1:100; CD80: BioLegend, #104735, 16-10A1, 1:100; CD86: BioLegend, #105002, GL-1, 1:100; CD206: BioLegend, #141702, C068C2, 1:100; CD3: BioLegend, #100201, 17A2, 1:100; CD4: BioLegend, #100506, RM4-5, 1:100; CD8: BioLegend, #100702, 53-6.7, 1:100; PD-1: BioLegend, #135202, 29 F.1A12, 1:100; Granzyme B: BioLegend, #396402, QA16A02, 1:100; FOXP3: Thermo Fisher Scientific, #14-5773-82, FJK-16s, 1:100; Ki67: BD Biosciences, #556003, B56, 1:100; E-Cadherin: BioLegend, #147302, DECMA-1, 1:100; CD45: BD Biosciences, #103102, 30-F11, 1:100; CD19: BD Biosciences, #115502, 1D3, 1:100). Cells were washed and resuspended in 1.5% PFA/PBS before acquisition. Data were acquired by Helios Mass Cytometer (Fluidigm) and analyzed using FlowJo 10 (BD Life Sciences) and Cytobank Premium (Beckman Colter). To generate viSNE plots, individual files from each treatment group were concatenated into a single file, and then a subset of equal events was selected at random through the FlowJo DownSample plugin. The total number of CD45+ cells from each group was used to create viSNE through Cytobank with 1000 iterations, perplexity of 30, and q of 0.5. Co-axial system of confocal microscopy and atomic force microscopy (AFM) Rigidity measurement was performed with the co-axis system76. Briefly, tumor sections Col1a2-creERT2 and Col1a2-creERT2; TXNDC5fl/fl were analyzed following cryopreservation and immunofluorescence staining with anti-Collagen I (Novus Biologicals, # NB600-408, polyclonal, 1:200) to identify regions of interest (ROIs). Tumor stiffness was measured immediately after frozen tissue slicing. In this experiment, 10000 points were intended for each ROI (10 µm2), and at least 3 ROIs were taken for each mouse. AFM (Bruker) was operated in Fast-Force volume mode at a 20 Hz scanning rate. Cantilevers with sharp tips (Tip radius: 65 nm, Spring constant: 0.1 N/m) were calibrated using No-Touch QNM calibration. Force-distance curves were calculated on the Sneddon model by NanoScope Analysis software v1.90 (Bruker). The unit of Young’s elastic modulus is calculated as N/m2, and expressed as kilopascal (kPa). Quantification and statistical analysis All experimental data are presented as means ± SEM. For bulk experimental data, statistical significance was evaluated using GraphPad Prism 9, with test statistics (t, F, or W), degrees of freedom (df), 95% confidence intervals (CIs), and P-values derived from the software. To quantify biological impacts, effective sizes were calculated as follows: Cohen’s d for pairwise comparisons (by multiplying the t-statistic by the square root of the sum of the reciprocals of the group sample sizes) and ω2 multi-group comparisons (defined as the ratio of the adjusted between-group sum of squares to the total variance). Comparisons between two groups were performed using a two-sided Student’s t-test, while Welch’s one-way ANOVA followed by Dunnett’s T3 post-hoc test was employed for clinical tissue arrays to account for unequal variances. Categorical variables were assessed using the chi-square test or Fisher’s exact test. For these experiments, exact P-values are indicated in the figures. Single-cell transcriptomic processing was performed using R version 4.0 (RStudio). Statistical significance for single-cell data was determined using a two-sided Wilcoxon rank-sum test, with significance levels indicated in the figures as ‡P < 0.05 and ∗P < 0.001. Detailed statistical outputs and effect sizes are documented in the Source Data files. All data shown were obtained from at least 3 biologically independent experiments. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgements We would like to thank Dr. David Brenner for his kind gifts of the Col1a1-GFPTg mice, and Dr. Ralf H. Adams for kindly providing the Cdh5-Cre/ERT2 mice. We thank the technical services provided by the Transgenic Mouse Models Core Facility of the National Core Facility for Biopharmaceuticals, National Science and Technology Council, Taiwan and the Gene Knockout Mouse Core Laboratory of the National Taiwan University Center of Genomic and Precision Medicine. We thank the staff of the Biomedical Resource Core and the imaging core at the First Core Labs, National Taiwan University College of Medicine, for technical assistance. We also thank the technical services provided by the 2nd, 3rd,7th and 8th Core Labs of the Department of Medical Research at National Taiwan University Hospital. Author contributions K.L.C. conceptualized the study, performed experiments, analyzed data, and prepared the manuscript. Y.W.T., C.T.H., Y.S.L., Y.S.T., Y.F.W., G.H.L., P.Y.C. and W.C.L. guided the experiments and data analysis. S.H.Y., S.Y.C., M.J.T., S.J.L., L.C.H.Y. and S.L.L. developed the methodology and supplied experimental materials. C.I.C., H.W.L., J.K.L. and Y.M.J. provided the human sample resources. C.C.C., R.H.C. and R.Y.J.H. conceptualized the study and guided the preparation of the manuscript. K.C.Y. conceptualized the study, guided the experiment and data analyses, and edited the manuscript. Peer review Peer review information Nature Communications thanks anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Funding This work was funded by Taiwan National Science and Technology Council Grants 111-2314-B-002-069 MY3, 112-2314-B-002-277 MY3, 112-2918-I-002-002, and 112-2926-I-002-511-G (K.C.Y), Innovative Research Grants from Taiwan National Health Research Institute NHRI-EX113-11213BI, NHRI-EX115-11521BI (K.C.Y), CRC Translational Research Grants IBMS-CRC111-P01, IBMS-CRC114-P03 (KCY), Translational Project Grants AS-TM-113-01-02, AS-TM-115-01-02 (K.C.Y), and a Grand Challenge Program Grant AS-GC-110-L06 (K.C.Y, S.Y.C) from Academia Sinica, Taiwan, grants from National Taiwan University Hospital NTUH. VN112-06, VN-113-03, 112-S0307, 112-S0311, 113-S0196, 113-IF0002, 113-E0008 (K.C.Y), Collaborative Research Projects of National Taiwan University College of Medicine, National Taiwan University Hospital and Min-Sheng General Hospital 110F005-112-M2 (K.C.Y), grants from the Excellent Translation Medicine Research Projects of National Taiwan University College of Medicine and National Taiwan University Hospital, NSCCMOH-131-41, 111C101-051, 112C101-031 (K.C.Y), Career Development Grants from National Taiwan University 112L7849, 113L7832 (K.C.Y), and NTU Advanced Biomedical Research Center, Taiwan Centers of Excellence (NTU ABRC TCE) 114M7070-3 (K.C.Y). Data availability The authors declare that all data supporting the findings of this study are available within the article and its Supplementary Information files. Source data are provided with this paper. scRNA-seq datasets generated during this study have been deposited in the Gene Expression Omnibus (GEO) repository on the NCBI website under accession code GSE261776. This paper reports data derived from published and publicly available GEO datasets: GSE103479, GSE63216, GSE39396, GSE21510, GSE132465, GSE134255, GSE39582, GSE86299. Protein expression and localization data were referenced from the Human Protein Atlas (https://www.proteinatlas.org/). Cell and mouse lines generated in this study are available from the lead contact upon request with a completed Materials Transfer Agreement (MTA). Source data are provided in this paper. Competing interests The authors declare no competing financial or non-financial interests directly related to the work submitted for publication. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-74073-1. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source Data Data Availability Statement The authors declare that all data supporting the findings of this study are available within the article and its Supplementary Information files. Source data are provided with this paper. scRNA-seq datasets generated during this study have been deposited in the Gene Expression Omnibus (GEO) repository on the NCBI website under accession code GSE261776. This paper reports data derived from published and publicly available GEO datasets: GSE103479, GSE63216, GSE39396, GSE21510, GSE132465, GSE134255, GSE39582, GSE86299. Protein expression and localization data were referenced from the Human Protein Atlas (https://www.proteinatlas.org/). Cell and mouse lines generated in this study are available from the lead contact upon request with a completed Materials Transfer Agreement (MTA). Source data are provided in this paper.