LOXL2⁺ cancer-associated fibroblasts shape WNT signaling to drive chemoresistance and poor outcomes in colorectal cancer: Insights from multi-omics and epidemiological analyses.

Introduction
Colorectal cancer (CRC) is a leading cause of cancer-related mortality, resulting in nearly 90,000 deaths per year [1]. Colon cancers account for ∼70% of CRC, while rectal cancers represent ∼30% [2]. Both can disseminate rapidly without treatment, leading to poor outcomes [3]. Standard therapies include surgery, targeted therapy, neoadjuvant radiotherapy, and adjuvant chemotherapy, yet drug resistance remains a major barrier to improved survival [4]. Early-stage rectal cancer is primarily treated with surgery, often combined with postoperative chemotherapy such as 5-fluorouracil, oxaliplatin, and irinotecan to reduce recurrence [5]. Advanced disease may require preoperative therapy, which can decrease tumor seeding, reduce toxicity, increase radiosensitivity, and facilitate sphincter preservation [6]. However, objective responses to 5-FU are limited; one study reported only 10.6% response rate [7], another 16%, with median progression-free survival of 4 months and overall survival of 9 months [8]. Attempts to reverse fluoropyrimidine resistance using gefitinib were largely ineffective [9]. Immunotherapy benefits only a subset of MSI-H patients (∼5%), whereas most MSS patients are resistant [10]. Anti-EGFR therapy is limited by RAS mutations, explaining only ∼35%–50% of nonresponse [11]. Overall, substantial chemoresistance remains unresolved, highlighting the need to identify additional mechanisms.
Fibroblasts are versatile cells maintaining tissue structure [12] and can repair damaged tissues [13]. In tumors, fibroblasts irreversibly adopt cancer-associated phenotypes, termed cancer-associated fibroblasts (CAFs) [14]. CAFs promote tumor proliferation, invasion, migration, and metastasis, and secrete cytokines, growth factors, and exosomes that enhance drug resistance. In gastric cancer, cisplatin and paclitaxel induce CAFs to secrete miR-522, reducing lipid ROS accumulation and decreasing chemotherapy sensitivity[15]. In lung cancer, CAF-derived IL-6 upregulates ANXA3 via JAK2/STAT3 to promote drug resistance [16]. In CRC, CAFs can prime cancer cells to increase stemness before chemotherapy [17]. secrete mediators that inhibit cell death (18), and induce chemoresistance through exosome-mediated signaling [16,18]. Inflammatory CAFs (iCAFs), activated by IL-1α, drive chemoradiotherapy resistance and disease progression [19]. Despite these insights, the mechanisms by which CAFs confer drug resistance remain incompletely understood. The development of single-cell RNA-seq technology provide the possibility for exploring the underlying mechanisms of CAFs on CRC. Unlike bulk RNA-seq, which averages signals across heterogeneous tumor and stromal cells, masks rare but clinically relevant subpopulations and can misattribute shifts in cell-type composition to transcriptional regulation, single-cell RNA-seq in colorectal cancer resolves diverse epithelial, immune and stromal cell states and trajectories, thereby revealing intratumoral heterogeneity and microenvironmental programs that are tightly linked to prognosis and treatment response [20,21].
Here, we found that CAF proportions vary across therapeutic response groups (NR, PR, CR), with myofibroblastic CAFs (myCAFs) predominant in non-responders through scRNA-seq analysis. MyCAFs showed enrichment of WNT signaling, with pseudotime analysis revealing stepwise WNT activation from CR to NR. We identified LOXL2 as a myCAF-associated gene increasing from CR to NR. CRC cells co-cultured with LOXL2-silenced CAFs exhibited reduced WNT signaling, decreased proliferation, migration, and stemness, and enhanced drug sensitivity. These findings delineate a potential mechanism of CRC chemoresistance and nominate a myCAF-based therapeutic target.

Methods

Data collection
Processed single-cell (SC) and spatial transcriptomic (ST) datasets were obtained from the China National GeneBank Database (CNGBdb, accession ID: CNP0004138). Bulk RNA-sequencing data for colorectal cancer (TCGA-CRC, n = 695) were downloaded from the Pan-Cancer Atlas section of the UCSC Xena portal (https://xenabrowser.net/datapages/?host=https%3A%2F%2Fpancanatlas.xenahubs.net&removeHub=https%3A%2F%2Fxena.treehouse.gi.ucsc.edu%3A443), and gene expression values were normalized to TPM. Additional bulk transcriptome cohorts, including GSE87211 (n = 363), GSE106582 (n = 194), GSE39582 (n = 585), GSE40076 (n = 246), GSE71187 (n = 189), and GSE103512 (n = 280), were collected from the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/).

Clinical sample
Eighty surgically excised colorectal tumor specimens and their paired adjacent normal tissues were obtained from individuals treated at Yangpu Hospital, Tongji University, between November 2018 and November 2019. Ethical approval for human sample use was granted by the Yangpu Hospital Ethics Committee (ID: LL-2023-LW-012), and all participants provided written consent prior to specimen collection, in accordance with the principles of the Declaration of Helsinki and relevant laws.

Raw data preprocessing
Gene expression data were processed and analyzed using the Seurat package in R (version 3.1.1). Cells with fewer than 500 or >6000 detected genes, or with mitochondrial transcript content exceeding 10%, were excluded. Genes detected in fewer than three cells were also discarded. Following these quality-control steps, a total of 325,096 cells of high quality were retained for subsequent analyses.

Seurat analysis
All single-cell data were processed within the Seurat environment. The SCTransform method was employed for data normalization, detection of highly variable genes, and scaling. Principal component analysis (PCA) was performed through the RunPCA function, and the informative components were subsequently used for unsupervised clustering and visualization with UMAP. To remove inter-patient batch effects, datasets from different samples were integrated using the SCTransform-based integration pipeline. Cell clusters were defined with the FindClusters function, and their identities were manually assigned based on canonical markers. Differentially expressed genes (DEGs) were determined using FindAllMarkers, retaining those detected in at least 25% of cells within each cluster, showing a log2 fold change ≥ 0.5, and meeting an adjusted P < 0.05 after Bonferroni correction.

Cell-type annotation and cell scoring
Cell clusters were annotated according to their gene expression signatures: endothelial cells (PECAM1 and VWF); Mast cells (TPSB2 and TPAB1); T cells (CD3D and CD3E); B cells (CD79A and MS4A1); Fibroblasts (DCN, LUM, TAGLN); Epithelial cells (KRT19, KRT18, EPCAM, CHGB); Myeloid cells (CD74, PTPRC, HLA-DRA); Stellate cells (RGS5). Genes with logFC > 1 and pct.1 > 0.3 were submitted to ChatGPT for reference-based annotation [22]. WNT pathway activity and stemness scores were computed using AddModuleScore from the Seurat package. The WNT signaling gene set was obtained from MsigDB (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp). Stemness scoring incorporated canonical stem cell markers, including SOX2, NANOG, POU5F1, KLF4, MYC, LIN28A, SALL4, DPPA4, UTF1, DNMT3B, ZFP4, TDGF1, GDF3, and PODXL.

Functional enrichment analysis
For each cluster, DEGs were collected to enrichment analysis using Gene Ontology (BP: biological progression) and KEGG gene sets. Functional terms and pathways enriched in specific subtypes were identified using enrichGO and enrichKEGG functions from the clusterProfiler R package.

Spatial transcriptomics analysis
Both signature gene expression and cell-type proportions were considered to define the cell identity of each spatial transcriptomics (ST) bin. A gene score, representing the sum of signature gene expression for each cell type, was calculated per bin. Bins with top scores were assigned to the corresponding cell type, with thresholds adjusted according to proportions observed in matched single-cell data. Signature gene sets were derived from previously published studies [23].

Cell-cell interaction analysis
Intercellular communication patterns were inferred using the CellChat package in R. Signaling networks between distinct cell populations were reconstructed on the basis of curated ligand–receptor interactions provided in the CellChatDB database. Independent analyses were carried out for various cell types and experimental conditions to delineate the overall intercellular signaling landscape.

Pseudotime trajectory analysis
To investigate variations in CAF developmental trajectories across CR, PR, and NR groups, pseudotime analysis was carried out with the Monocle2 toolkit. Gene expression profiles were projected into a low-dimensional manifold through reversed graph embedding, enabling the reconstruction of continuous cellular transitions over time. The resulting trajectories were visualized using the plot_cell_trajectory function to display the inferred progression order of CAFs.

Analysis of expression programs in bulk-RNA seq
Gene-level expression profiles from the TCGA-CRC dataset were matched to corresponding clinical information. Statistical comparisons between two continuous variables were conducted using Student’s t-test, whereas logistic regression models were employed to examine relationships between gene expression and categorical clinical factors. Drug response prediction was performed with the pRRophetic package in R.

Survival analysis
Patient survival analyses were performed with the survival package in R (v3.1-8). Kaplan–Meier curves were generated using the survminer package (v0.4.6), and differences in survival distributions were evaluated through the log-rank test.

Cell culture and transfection
HCT116 and SW480 colorectal cancer (CRC) cell lines were obtained from Pricella (Wuhan, China). Primary cancer-associated fibroblasts (CAFs) were isolated from human colorectal tumor tissues and authenticated by the expression of canonical CAF markers, including COL1A1 and α-SMA. CRC cells were maintained in RPMI-1640 medium, whereas CAFs were cultured in DMEM, both supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin, and incubated at 37 °C in a humidified atmosphere containing 5% CO₂.
For co-culture experiments, CRC cells and CAFs were co-seeded in DMEM containing 10% FBS either in direct co-culture or using Transwell inserts (0.4 μm pore size, Corning), as specified in individual assays. In selected experiments, co-cultures involving LOXL2-overexpressing CAFs were further treated with the WNT pathway inhibitor LGK974 (1 μM) or the selective LOXL2 inhibitor PXS-5153A (1.5 μM) in order to evaluate the involvement of the LOXL2–WNT signaling axis in regulating tumor cell behavior.
CAFs at 40–50% confluence were transfected with either LOXL2-targeting short hairpin RNAs, as well as with LOXL2 overexpression plasmids, using Lipofectamine 8000 (Beyotime, China) according to the manufacturer’s instructions. The sequences of all shRNAs and plasmids used in this study are listed in Supplementary Table S1. Lentiviral particles containing the LOXL2-targeting shRNA constructs were generated and purchased based on the validated targeting sequences.

Primary fibroblast isolation from CRC and western blot
The detailed process was performed in line with previous research[24]. Purity of fibroblast was confirmed by immunofluorescence for α-SMA, Ki-67, and COL1A1. The specific antibody information is provided in Table S2.

Cell proliferation assay
For the CCK-8 proliferation assay, co-cultured cells (2 × 103 per well) were seeded into 96-well plates in 100 μL of medium. At 24-h intervals (days 1, 2, and 3), 10 μL of CCK-8 reagent was added to each well and incubated for 2–4 h at 37°C. Cell viability was determined by measuring absorbance at 450 nm.
For the EdU incorporation assay, cells grown in 24-well plates were processed using the Click-iT EdU Imaging Kit (Beyotime, China) according to the manufacturer’s protocol to evaluate DNA synthesis activity.

Wound healing assay
Co-cultured cells were seeded in 6-well plates and scratched using a 200 µL pipette tip. After washing with PBS, cells were incubated in DMEM with 1% FBS. Wound closure was imaged under an inverted microscope at indicated intervals.

Immunohistochemistry and immunofluorescence analysis
Immunohistochemical detection of LOXL2 was performed on 80 tumor samples. Primary antibodies against COL1A1, pan-CK, Ki-67, WNT5A, and α-SMA (all at 1:800; Sigma-Aldrich). Other detailed process was performed in accordance with previous research [24].

In vitro tumorsphere formation
For tumorsphere formation assays, 500 colorectal cancer (CRC) cells were co-cultured with an equal number of cancer-associated fibroblasts (CAFs) in 96-well plates containing 200 μL of complete Neurobasal medium. After 10 days of suspension culture, the resulting spheres were imaged under a microscope, enumerated, and quantitatively evaluated.

Transwell
Tumor cell migration was assessed using Boyden chambers with 8-μm pore filter inserts in 24-well plates (BD Biosciences, USA). After the indicated treatments, colorectal cancer cells were harvested, and 2 × 104 cells in serum-free medium were seeded into the upper chamber; the lower chamber contained medium with 20% FBS as a chemoattractant. After incubation, cells that migrated to the underside of the membrane were fixed, stained, and counted at ×200 magnification in 10 random fields per insert.

RT-qPCR
Quantitative PCR was carried out with TB Green Premix Ex Taq II (Tli RNaseH Plus) (RR820A, Takara) on a LightCycler 96 instrument. Relative transcript abundance was determined by the 2^–ΔΔCt method after normalization to 18S rRNA (or GAPDH where indicated). The following primers were synthesized by BioAsia (Hyderabad, India) in Table S3.

Drug dose–response assay
Cells were seeded in 96-well plates in sextuplicate. After 24 h, FOLFOXIRI was applied at preset concentrations for 48 h. The medium was then replaced with a CCK-8 working solution (medium:CCK-8 = 10:1; APE×BIO, USA) for 2–4 h, and viability was read at 450 nm. IC50 values for SW480 and HCT116 were obtained by nonlinear regression in GraphPad Prism 10. These IC50 doses were subsequently used to treat SW480 and HCT116 co-cultured with LOXL2+CAFs or LOXL2-silenced CAFs.

Xenograft model in nude mice
Male BALB/c nude mice (4–6 weeks old, 24–26 g; HFK Bioscience Co., Ltd., Beijing, China) were used for xenograft tumor experiments. A suspension containing 2 × 106 pretreated SW480 cells was injected subcutaneously into the axillary region of each mouse to establish tumors. Once palpable masses developed, drug administration was initiated. FOLFOXIRI was delivered via tail vein injection at a dosage of 10 mg/kg per mouse, administered one to two times per week. Tumor dimensions—length (L) and width (W)—were measured weekly or biweekly, and tumor volume was calculated using the formula V = 0.52 × L × W2. After approximately 3–4 weeks, mice were euthanized, tumors were excised, and subsequent analyses were performed. All animal procedures were approved by and conducted in accordance with the ethical guidelines of the Animal Center of Tongji University.

Statistical analysis
All statistical analyses were performed using R software (version 4.5.1) and GraphPad Prism (version 10). Differences between two groups were assessed using two-sided Student’s t-tests. Associations between gene expression levels and binary clinical parameters were evaluated through logistic regression analysis based on odds ratio (OR) estimation. A P value < 0.05 was regarded as statistically significant.

Result

CRC single-cell RNA-Seq shows distinct distribution in different treatment groups
The flowchart of this work is shown in Fig. 1A. To explore the mechanism of drug resistance in CRC, we first downloaded the CRC single-cell RNA-seq data from a public database and then conducted quality control (Fig. S1). Based on different markers, the cells were annotated as fibroblasts, endothelial cells, T cells, epithelial cells, B cells, myeloid cells, and mast cells (Fig. 1B and C). The treatment timepoints, including pre-treatment (PRET) and post-treatment (POSTN and POSTT) groups, showed distinct distributions in the UMAP plot (Fig. 1D). The therapeutic responses were distributed into different clusters (Fig. 1E). We observed the distribution of different cell subtypes across NR, PR, and CR groups (Fig. 1F). We found that the proportion of epithelial cells gradually decreased across NR, PR, and CR groups, whereas fibroblasts gradually increased. Other cell types, such as B cells, T cells, mast cells, myeloid cells, and endothelial cells, did not show significant changes across NR, PR, and CR groups. It is well known that CAFs are correlated with drug resistance; however, we observed reversed results in the scRNA-seq drug treatment data, so we extracted CAFs for further analysis.

Mycaf are correlated with drug resistance of CRC
To explore the distinct distribution of CAFs across NR, PR, and CR groups, we conducted further annotation. The CAFs were classified into myCAF, pericyte, iCAF, and unknown groups (Fig. 2A). We observed the distribution of NR, PR, and CR groups within CAF subtypes and found that NR groups showed overlapping distributions with myCAF (Fig. 2B). However, the pre- and post-treatment groups did not show significant differences in myCAF (Fig. 2C). Through enrichment analysis, we found cytoplasmic translation, morphogenesis of a branching structure, morphogenesis of a branching epithelium, mammary gland morphogenesis, cell–cell signaling by WNT, mesenchyme development, regulation of the WNT signaling pathway, renal system development, and muscle cell differentiation highly enriched in the NR group (Fig. 2D). Many studies have reported that the WNT pathway is highly correlated with drug resistance. Interestingly, we found that the WNT score was higher in myCAF than in other subtypes and gradually decreased across NR, PR, and CR (Fig. 2E and F). Furthermore, we found that biological processes (BP), including extracellular matrix organization, extracellular structure organization, external encapsulating structure organization, collagen fibril organization, and wound healing, were highly enriched in myCAF (Fig. 2G). Overall, we found that myCAF might be correlated with drug resistance and that the WNT pathway is activated in myCAF.

LOXL2 dynamics along pseudotime reveal stemness, WNT programs, and chemosensitivity in CAFs and epithelial cells
Based on the schematic diagram of the pseudotime sequence provided by the Monocle2 algorithm, branch state one was used as the starting point (Fig. 3A). For the treatment groups, the PRET and NR groups mainly distributed at the starting point, while POSTT and CR groups distributed at the ending point (Fig. 3B and C). WNT scores were higher at the starting point but lower at the ending point (Fig. 3D). The genes showing significant differences in expression along the pseudotime axis in CAFs were then clustered into three modules based on their expression patterns in a heatmap. According to the change from the CR to NR groups, some genes gradually increased, such as CXCL1, DIO2, UBD, COMP, SFRP4, COL10A1, and SFRP2. Enrichment analysis revealed that biological processes such as WNT signaling, extracellular matrix organization, and extracellular structure organization were altered (Fig. 3E). However, some genes gradually decreased from the CR to NR groups, including NPNT, ACTG2, CRYAB, DES, FLNC, KCNMA1, MYH11, and SYNM. These genes were enriched in muscle contraction, muscle system processes, regulation of amyloid fibril formation, and smooth muscle contraction. Furthermore, some genes gradually increased from CR to NR and were enriched in the IL-17 signaling pathway and rheumatoid arthritis KEGG pathways (Fig. S2A). Based on the Venn diagram, we found that LOXL2 was located among the overlapping genes between pseudotime genes and DEGs between NR and CR (Fig. 3F). In addition, LOXL2 and WNT5A expression gradually decreased from the beginning to the end and had strong association (Fig. S2B-C). We further calculated the stemness score and found that it gradually decreased from NR to CR groups, both before and after treatment (Fig. 3G). Furthermore, we found that LOXL2 and WNT5A gradually decreased from the pseudotime beginning to the end in NR, PR, and CR groups, as well as POSTN, POSTT, and PRET groups (Fig. 3H and I). And the proportion of LOXL2+CAFs is inversely proportional to the therapeutic efficacy (Fig. 3J). Drug sensitivity for four agents in LOXL2-high and LOXL2-low groups within the TCGA-CRC cohort found that LOXL2-low tumors were more sensitive to camptothecin, cisplatin, cytarabine (Cytarabine), and gemcitabine (Fig. 3K).
We further conducted the analysis in epithelial cells. The distributions of response groups and timepoint groups are shown in the UMAP (Fig. S3). The epithelial cells were stratified into high and low LOXL2 groups (Fig. 3L). We then calculated the CytoTRACE score and found that high CytoTRACE scores were mainly located in high LOXL2 groups, indicating that LOXL2+ epithelial cells have low differentiation potential (Fig. 3M). Regardless of the response group (NR, PR, and CR), high LOXL2 epithelial groups still showed higher CytoTRACE scores (Fig. 3N). We further conducted GSEA enrichment analysis and found that WNT pathways, stem cell proliferation, epithelial-to-mesenchymal transition, and epithelial cell migration were highly enriched in high LOXL2 expression cells, which further validated that LOXL2 is positively correlated with WNT signaling, stemness, and migration (Fig. 3O).

Spatial transcriptome analysis
To explore LOXL2 expression distribution, we used ST data and conducted cell annotation. The ST data included ST-CR1, ST-CR2, ST-NR1, and ST-NR2, which were annotated as B cells, CAF, endothelial, enteric glial, epithelial, mast cells, myeloid, plasma, and T cells (Fig. 4A; Fig. S4A and B). We observed that the stemness score was highly enriched in ST-NR1 and ST-NR2 samples (Fig. 4B). In addition, we found that LOXL2 and WNT5A were also highly expressed in these two samples rather than CR samples (Fig. 4C). Meanwhile, immunofluorescence confirmed the increased expression and colocalization of LOXL2 and WNT5A in tumor tissues (Fig. 4D; Fig. S5). We further conducted cell–cell communication analysis, which indicated that myCAF and iCAF interacted with other cell types and tissue samples (Fig. 4E). Furthermore, we found that myCAF exhibited the highest outgoing WNT signaling, and NR groups received the highest WNT signaling (Fig. 4F). These findings indicate that the mechanism by which myCAF causes resistance is mediated through the WNT pathway.

LOXL2 highly expresses in tumor tissues and correlated with prognosis
To validate the expression of LOXL2, we first analyzed the TCGA-CRC datasets. The unpaired and paired tumor–normal samples both showed upregulation of LOXL2 in tumor tissues compared with normal tissues (Fig. 5A). We collected 26 tumor samples with corresponding normal tissues and detected protein expression; the results showed that LOXL2 was highly expressed in tumor tissues (Fig. 5B–C). Furthermore, immunohistochemistry revealed that LOXL2 protein expression was higher in tumor tissues than in normal tissues (Fig. 5D). In the TCGA clinical cohort, LOXL2 expression correlated with pathologic T stage (OR = 1.709), pathologic N stage (OR = 1.809), pathologic stage (OR = 1.559), lymphatic invasion (OR = 1.816), and perineural invasion (OR = 2.471) (Fig. 5E). Patients younger than 65 showed higher LOXL2 expression. With the advancement of pathologic T stage, pathologic stage, and pathologic N stage, LOXL2 expression increased. Patients diagnosed with perineural or lymphatic invasion also showed higher LOXL2 expression (Fig. 5F). Furthermore, ROC analysis revealed that LOXL2 expression had high predictive performance in TCGA-CRC (AUC = 0.670), GSE87211 (AUC = 0.957), GSE106582 (AUC = 0.870), GSE39582 (AUC = 0.769), GSE40076 (AUC = 0.820), GSE71187 (AUC = 0.936), and GSE103512 (AUC = 0.706) datasets (Fig. 5G). Furthermore, in the TCGA cohort, high LOXL2 expression was associated with poorer OS, RFS, and PPS than low LOXL2 expression (Fig. 5H-J). Besides, we found that high LOXL2+ CAF signature scores indicated poorer outcomes than low scores (Fig. 5I). Stratified survival analyses were conducted and it deserves to be mentioned that in patients with early-stage disease, higher LOXL2 expression was consistently associated with poorer overall survival, indicating robust prognostic value of LOXL2 in the early phase of colorectal cancer progression. In contrast, in the advanced-stage group, the prognostic significance of LOXL2 expression was markedly attenuated, even exhibited an inverse trend (Fig. 5K). We further conducted ORA and GSEA analyses and found that the high LOXL2 expression group was highly enriched in WNT signaling, focal adhesion, cell migration, and invasion pathways (Fig. S7A–C). Moreover, the high LOXL2 group showed higher fibroblast infiltration based on MCPcounter and xCell analyses (Fig. S7D and E).

LOXL2+ CAF promote CRC cell progression
To validate the function of LOXL2 in CAFs, we extracted CAFs from CRC tissues. The CAF cells were validated through immunofluorescence staining using α-SMA, COL1A1, and Ki67 antibodies (Fig. S6). We constructed LOXL2-overexpression and LOXL2-knockdown CAFs. Transfection efficiency was confirmed by western blot and we found silencing LOXL2 decreased WNT5A expression, whereas LOXL2 overexpression increased it (Fig. 6A). We then established a co-culture system by merging CAFs and CRC cells (Fig. 6B). We found that LOXL2+ CAFs promoted HCT116 and SW480 proliferation and migration, whereas SW480 co-cultured with LOXL2-silenced CAFs showed lower proliferation and migration (Fig. 6C–F, Fig. S8A-D). In addition, LOXL2+ CAFs increased the expression of ZEB2, N-cadherin, vimentin, α-SMA, and Twist1, while decreasing E-cadherin expression. Conversely, LOXL2-silenced CAFs inhibited ZEB2, N-cadherin, vimentin, α-SMA, and Twist1 expression but upregulated E-cadherin (Fig. 6G, Fig. S8E). Moreover, HCT116 co-cultured with LOXL2+ CAFs showed an increased IC50 of FOLFOXIRI, while SW480 co-cultured with LOXL2(–) CAFs showed a decreased IC50 (Fig. 6H). Sphere formation assays revealed that HCT116 co-cultured with LOXL2+ CAFs formed larger spheres (Fig. 6I, Fig. S8F). We also detected the correlation between LOXL2 and several Stemness markers (CD44, ALCAM, SOX2, and ITGB1) were examined. LOXL2 showed strong positive correlations with CD44, ITGB1, and ALCAM (Fig. S8G). Consistently, HCT116 cells co-cultured with LOXL2+ CAFs exhibited significant mRNA and protein upregulation of CD44, ITGB1, ALCAM, and SOX2 (Fig. 6J-K, Fig. S8H). Notably, pharmacological inhibition of LOXL2 or WNT signaling effectively abrogated these pro-tumorigenic effects. When the co-culture system containing LOXL2-overexpressing CAFs and HCT116 cells was treated with the WNT pathway inhibitor LGK974 or the selective LOXL2 inhibitor PXS-5153A, a pronounced attenuation of malignant phenotypes was observed (Fig. 6C-6 K). Overall, these results reveal that LOXL2+ CAFs promote CRC cell proliferation, migration, stemness, and drug resistance by increasing WNT pathway activity. In vivo assays showed that tumor growth was significantly inhibited in SW480 co-cultured with LOXL2-silenced CAFs treated with FOLFOXIRI (Fig. 6L). These results indicate that SW480 co-cultured with LOXL2-silenced CAFs is more sensitive to FOLFOXIRI compared other groups.

Discussion
Preoperative chemoradiotherapy (CRT) is increasingly recognized as the preferred approach for locally advanced rectal cancer, offering better local tumor control and reduced treatment-related toxicity compared with postoperative therapy [25]. In KRAS wild-type colorectal cancer with resectable liver metastases, adding cetuximab to preoperative chemotherapy improves R0 resection rates and 3-year overall survival relative to chemotherapy alone [26]. Among dMMR patients receiving neoadjuvant chemotherapy (NAC), 32.5% achieved tumor regression grade 1, indicating meaningful clinical benefit [27]. However, despite these advantages, most neoadjuvant radiotherapy trials show no significant improvement in overall survival [28]. High-volume disease patients receiving NAC had shorter median survival than non-NAC patients (14.4 vs 23.8 months, p = 0.046), and low-volume patients also trended similarly (36.5 vs 46.4 months, p = 0.17) [29], raising concerns that NAC might accelerate progression or induce drug resistance. Adjuvant chemotherapy also confers little survival benefit, particularly after NAC [30].
Drug resistance is a major cause of treatment failure in rectal cancer, driven by complex mechanisms including tumor-intrinsic features, dysregulated pathways, microenvironmental factors, and immune status. Cancer stem cells (CSCs) are closely linked to 5-FU resistance via ABC transporter–mediated drug efflux [31], enhanced DNA repair reducing oxaliplatin efficacy [32], and additional mechanisms such as anti-apoptosis, quiescence, and autophagy [33]. METTL3 promotes 5-FU resistance by upregulating LDHA-mediated glycolysis and enhancing homologous recombination repair [34]. Noncoding RNAs regulate drug efflux, signaling, apoptosis, autophagy, and DNA repair [35], while the KRAS/ERK/ADAM17 axis and LAMP2A-mediated pathways further contribute to chemoresistance [36,37]. Chemotherapy often fails to eradicate dormant CSCs, and strategies targeting such populations remain insufficiently validated [38]. Exosome-mediated lncRNA transfer and ABC transporter inhibitors have shown limited clinical efficacy [39]. Therefore, exploring new mechanisms of drug resistance is necessary to develop strategies that improve tumor control and enhance drug sensitivity.
Cancer-associated fibroblasts (CAFs) have been linked to drug resistance in multiple cancers, though their role in CRC remains incompletely understood. Using single-cell RNA-seq, we quantified CAF proportions across therapeutic response categories. While overall CAFs increased from non-responders (NR) to partial (PR) and complete responders (CR), myofibroblastic CAFs (myCAFs) were enriched in NR samples, both pre- and post-treatment, suggesting a subtype-specific association with poor response. Prior studies indicate myCAFs promote resistance through extracellular matrix (ECM) remodeling and TGF-β signaling [40,41]; and specific myCAF subsets associate with chemoresistance and immune therapy resistance [42,43].
Functionally, myCAFs were enriched for IL-17 signaling. Chemotherapy-activated CAFs can secrete IL-17A to support cancer-initiating cell survival and drug resistance via NF-κB [44]. The IL-17–CXCR2 axis recruits neutrophils and promotes tumor progression, with CXCR2 blockade suppressing this pro-tumor process [45]. myCAFs were also enriched for ECM organization, structure formation, and wound healing. ECM remodeling increases stiffness, drives epithelial–mesenchymal transition (EMT), and enhances CSC traits, contributing to chemoresistance [46]. WNT signaling was elevated in NR samples, with higher WNT scores in myCAFs than other CAF subtypes, showing a NR > PR > CR gradient. CAF-expressed CCAL, CAF-derived exosomes, and chemotherapy-induced stromal factors such as WNT16B/SFRP2 have been implicated in WNT-mediated chemoresistance [47,18,48].
LOXL2, highly expressed in myCAFs, catalyzes collagen cross-linking, stiffens ECM, and promotes EMT, invasion, and therapy resistance [49]. Lysyl oxidase activity produces H₂O₂, activating PI3K/Akt signaling and HIF-1α synthesis, forming a feedback loop enhancing proliferation [50]. c-Fos/AP-1 upregulates WNT7B/WNT9A, inducing LOXL2 and promoting matrix remodeling and aggressiveness [51]. Experimentally, CRC cells co-cultured with LOXL2-silenced CAFs showed reduced EMT, WNT signaling, proliferation, migration, and stemness, with increased drug sensitivity. Accumulating evidence supports that LOXL2+ myofibroblastic CAFs activate WNT signaling in CRC cells through both ligand-dependent and matrix-dependent mechanisms. For instance, stromal subsets with high WNT5A expression, such as CTHRC1+ CAFs, have been found to secrete non-canonical WNT ligands, WNT5A, that promote EMT, stemness, and metastasis of CRC cells via paracrine WNT/MSLN signaling [24]. In addition, CAF-derived exosomes can transfer oncogenic cargos to tumor cells, enhancing stemness, EMT and chemoresistance in CRC[18]. These findings are conceptually consistent with our observation that LOXL2+ CAF–conditioned media increase WNT target gene expression and drug resistance in tumor cells. In addition, LOXL2 is a secreted amine oxidase for which small-molecule inhibitors and monoclonal antibodies have already entered early-phase clinical testing[52], while multiple classes of WNT pathway inhibitors (including porcupine inhibitors, Frizzled receptor antagonists and β-catenin/CBP or tankyrase inhibitors) are being evaluated in colorectal and other solid tumors [[53], [54], [55]]. In parallel, preclinical studies have shown that reprogramming or functionally targeting CAFs can deplete the cancer stem cell pool and restore sensitivity to chemotherapy, and that WNT/β-catenin blockade can reverse stemness-associated resistance [[56], [57], [58]]. Therefore, combining LOXL2 inhibition with WNT pathway inhibitors represents a rational strategy to simultaneously reduce CAF-derived WNT ligand production, decrease ECM stiffness and weaken downstream β-catenin signaling in tumor cells. Given that WNT-driven stemness and a desmoplastic stroma are also linked to immune exclusion [59], such combinations may further synergize with immune checkpoint blockade by rendering the tumor microenvironment more permissive to cytotoxic T-cell infiltration. However, LOXL2 inhibition could produce both biological “off-target” consequences and compensatory resistance because LOXL2 participates broadly in ECM cross-linking, EMT programs, and tissue remodeling. Systemic blockade of LOXL2 may impair physiological matrix homeostasis and repair, and other family members (LOXL1/3/4) might be compensatorily increased, thereby preserving mechanotransduction-driven EMT and stemness even when LOXL2 is suppressed [[60], [61], [62]]. So it is necessary to consider the off-target and compensatory resistance when using LOXL2 inhibitor with other drugs. Of note, in-vivo experiment reveals that silencing LOXL2 could enhance the sensitivity of CRC to FOLFOXIRI. Given that the high LOXL2 expression patients show worse outcomes compared to low LOXL2 expression patients, it is possible to use the LOXL2 expression as a predictive biomarkers for response to chemotherapeutic agents. Taken together, our data support a model in which LOXL2⁺ cancer-associated fibroblasts promote WNT pathway activation and ECM-related signaling in colorectal cancer cells, whereas the exact receptor usage (including potential FZD receptors and co-receptors) remains to be defined in future functional studies

Conclusion
In conclusion, our study identifies a myCAF-centered mechanism of drug resistance, highlighting LOXL2-associated myCAF programs as actionable therapeutic targets in CRC (Fig. S9).

Limitation
Our study supports a role for CAFs in drug resistance, but several limitations should be acknowledged. First, CAF subsets were defined from a single CRC scRNA-seq cohort using marker-based clustering, which may not capture CAF heterogeneity across sites, treatments, and stages. Second, LOXL2 and ECM-remodeling programs are also expressed in normal fibroblasts and other stromal cells, raising concerns about off-tumor toxicity. Third, we did not perform phenotype-rescue experiments to directly confirm that LOXL2+ myCAFs drive resistance via WNT activation. In addition, we did not stratify colon versus rectum, and our finding of LOXL2+ myCAF enrichment in non-responders relies mainly on public datasets without independent validation; thus, its predictive value for standard therapies still need to explore.

Declarations

Ethics approval and consent to participate
The study was conducted in accordance with the Declaration of Helsinki, as revised in 2013. The study protocol was approved by the Ethics Committee of Yangpu Hospital (Approval No LL-2023-LW-012), and written informed consent was obtained from all participants.

Consent for publication
All authors gave their consent for publication.

Data availability
All data could be downloaded at GEO (http://www.ncbi.nlm.nih.gov/geo), TCGA (https://portal.gdc.cancer.gov/), MsigDB (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp).

Conflict of interests
The authors have declared that no competing interest exists.

Funding
This work was funded by Shanghai Clinical Pharmacy Key Specialty Construction Project Support project (Shanghai Health and Pharmaceutical Administration, 201809).

CRediT authorship contribution statement
Chengyuan Xu: Funding acquisition, Formal analysis, Data curation, Conceptualization. Tengfei Li: Writing – review & editing, Writing – original draft, Visualization. Lin Zhang: Supervision, Software. Qin Zhang: Visualization, Resources. Shanshan Cai: Visualization, Software, Resources. Qiangqiang Fu: Software, Formal analysis. Siqi Zhang: Validation, Supervision, Software.

Declaration of competing interest
The authors have declared that no competing interest exists.