Senescent fibroblasts drive CD8⁺ T cell dysfunction in colorectal cancer via CD36-mediated lipid transfer and peroxidation

Abstract

Background

Functional exhaustion of tumor-infiltrating CD8⁺ T cells represents a hallmark of colorectal cancer (CRC) immunosuppression, though its mechanistic drivers remain elusive. Given the established correlation between CRC progression and stromal senescence characterized by pathological lipid accumulation and impaired immunity, we investigated whether and how senescent fibroblasts actively regulate CD8⁺ T cell dysfunction.

Methods

Single-cell RNA sequencing (scRNA-seq) analysis was conducted to unveil the diverse fibroblast populations and the significant lipid metabolism changes between senescent fibroblasts and non-senescent fibroblasts in human CRC specimens and adjacent normal mucosa. Machine-learning identified senescent fibroblasts with a distinct gene signature. Cell-cell communication analysis was used to evaluate the interactions between senescent fibroblasts and CD8⁺ T cells in colorectal cancer. Co-culture experiments were conducted among senescent fibroblasts, CD8⁺ T cells and patient-derived organoids of CRC (CRC-PDOs), with the results evaluated with high-content imaging and propidium iodide/Hoechst 33,342 staining. Flow cytometry, ELISA and lipid pulse-chase with BODIPY FL C16 were performed to detect the alterations of CD8⁺ T cell cytotoxic function and metabolic status. AOM/DSS-induced CRC mouse model was used to conduct in vivo validation to evaluate whether senolytics could suppress CRC progression. Patients from the Cancer Genome Atlas colorectal cancer cohort were stratified into CD36-high and CD36-low groups by median expression, and drug sensitivity for GDSC2 compounds was predicted computationally using the oncoPredict R package.

Results

ScRNA-seq demonstrated the specific cell population presence and divergence of senescent fibroblasts between neoplastic and histologically normal adjacent cell clusters in CRC. Random Forest was employed for cell senescence classification. Feature importance analysis identified five genes as key contributors to the model’s decision process. Cell-cell communication analysis revealed enhanced interactions between senescent fibroblasts and CD8⁺ T cells in CRC. Co-culture of senescent fibroblasts significantly impaired the cytotoxic functions of CD8⁺ T cells on CRC-PDOs, which was reflected by the declined proportions of granzyme B (GZMB) ⁺ and interferon gamma (IFNγ) ⁺ CD8⁺ T cells and enhanced viability of CRC-PDOs. Mechanistically, the co-culture with senescent fibroblasts promoted the lipid shuttling into CD8⁺ T cells to induce lipid peroxidation and downstream impairment of cytotoxicity. Furthermore, the inhibition of CD36, the specific scavenger receptor for lipid uptake of CD8⁺ T cells, effectively suppressed lipid transfer and peroxidation thereby preserving the effector functions of CD8⁺ T cells and ultimately promoting tumor apoptosis. Complementarily, in vivo senolytic treatment significantly suppressed CRC progression in AOM-DSS CRC mouse models. Top 12 therapeutic agents were identified significantly enhanced predicted efficacy in CD36-high tumors.

Conclusions

Our study identified a substantial population of senescent fibroblasts in human CRC through single cell transcriptomics, machine-learning and clinical biopsies. These senescent fibroblasts impair CD8⁺ T cell-mediated killing of CRC-PDOs via CD36-dependent lipid transfer, suggesting senolytic targeting of stromal cells as a promising immunotherapeutic strategy for CRC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07636-3.

Introduction

CRC is the most common gastrointestinal malignancy and one of the leading causes of cancer-related mortality worldwide [1, 2]. The tumor microenvironment (TME) is recognized as a fundamental regulator in CRC pathogenesis, characterized by significant heterogeneity [3]. This complex microsystem comprises diverse cellular components including immune cells, fibroblasts and endothelial cells that engage in dynamic interactions to orchestrate both pro- or anti-tumor responses [4]. Deciphering the cellular and molecular intricacies of the CRC immune microenvironment provides critical insights for the development of novel immunomodulatory strategies.

The presence of cytotoxic lymphocytes CD8⁺ T cells, a key component of the tumor immune system, serves as a significant prognostic factor in CRC, associated with improved overall survival [5]. As central mediators of adaptive immunity, CD8⁺ T lymphocytes play a crucial role in tumor immunology due to their potent cytotoxic activity, which enables direct elimination of epithelial-derived malignant cells [6]. These cells also contribute substantially to maintaining mucosal homeostasis within the gastrointestinal tract, positioning them as a major focus for both basic research and therapeutic development [5]. However, under specific conditions such as the prolonged chronic inflammatory environments, CD8⁺ T cells can enter a hyporeactive or dysfunctional state [7]. Consequently, reversing this state of CD8⁺ T cells is a key strategic goal in anti-tumor research.

Notably, fibroblasts constitute the predominant stromal population in many malignancies. Recent research has increasingly focused on delineating the functional heterogeneity among fibroblast subtypes and their distinct contributions to tumorigenesis [8]. Accumulating evidence reveals stromal cell senescence as a hallmark feature of the TME across multiple cancer types [9, 10]. Cellular senescence, a state of irreversible growth arrest triggered by endogenous or exogenous stimuli, is characterized by marked upregulation of cell cycle inhibitors, including P16 and P21 [11]. Critically, the accumulation of senescent cells in tumor tissues not only contributes to protective anti-tumorigenic barriers but also drives therapy resistance and mediates adverse effects [12, 13]. This functional duality underscores the importance of investigating both the roles and mechanisms of action of senescent cells in cancer.

Accumulating clinical evidence indicates that age-related immune senescence compromises tumor surveillance, contributing to poorer prognosis and increased treatment resistance in elderly cancer, with distinct patterns observed across tumor types [14]. Within the TME, senescent cells engage in complex crosstalk that actively sculpts immune responses [15]. Specifically, senescence drives immune dysfunction through sustained pro-inflammatory cytokine secretion and recruitment of immunosuppressive cells, thereby fostering tumor progression [16]. Preclinical evidence robustly supports senolytic therapies for cancer intervention. In PARP inhibitor-induced senescent mouse xenograft models, Navitoclax as a senolytic agent effectively eliminated ovarian and breast cancer cells thereby inhibiting tumor growth [17]. Similarly, CD87-targeting CAR-T cells selectively eliminated senescent cells, which significantly delayed tumor progression in lung adenocarcinoma mice treated receiving combined therapy of MEK and CDK4/6 inhibitors [18].

Accumulating evidences indicate that stromal senescence is a hallmark of tumor microenvironments across malignancies [19]. Fibroblasts constitute the predominant senescent population, with their SASP exhibiting conserved immunosuppressive functions. Notably in pancreatic ductal adenocarcinoma, the senescence rate of fibroblasts demonstrated negative correlation with the number and activation of CD8⁺ T cells, which revealed that senescent stromal fibroblasts limit the activation of cytotoxic CD8⁺ T cells [9]. Substantial accumulation of senescent fibroblasts has been documented in CRC [10], however, their functional interplay with CD8⁺ T cells remains incompletely characterized in the literature.

Concurrently, there is a pressing need to develop senolytic drugs to eliminate senescent cells or senoblockers to suppress their activity. Despite these advances, the potential role of senescent fibroblasts within the stromal compartment, particularly in CRC initiation and progression, remains poorly understood. The functional complexity and phenotypic diversity of senescent cells highlight the need to dissect the specific roles of senescent cell subsets, the context-dependent nature of their activities, and their interaction network within the tumor ecosystem.

In this study, we identified a substantial population of senescent fibroblasts in human CRC samples through single cell transcriptomics datasets and clinical biopsies. We systematically investigated how fibroblast senescence within the immune microenvironment influences CRC pathogenesis and progression. Our results demonstrate that senescent fibroblasts restricted cytotoxic CD8⁺ T cell activation and impaired their tumor-killing capacity. These findings were validated in both patient-derived organoid-immune co-culture models and AOM-DSS induced CRC mouse models, collectively established senescent fibroblasts as promising targets for restoring CD8⁺ T cell-mediated antitumor immunity in CRC.

Results

Senescent fibroblasts are enriched in human colorectal cancer

Given the functional heterogeneity of fibroblasts across malignancies and their tissue-dependent roles in tumorigenesis [20, 21], particularly the established link between cancer-associated fibroblast subpopulations and antitumor immunity in CRC [4], we investigated the role of senescent fibroblasts within the TME of CRC. Using scRNA-seq data from human CRC specimens and matched normal mucosa (GEO database GSE166555), we performed integrated analysis. Principal component analysis (PCA) of highly variably expressed genes revealed global transcriptional divergence between adjacent normal tissue and tumor tissue samples (Fig. 1A). The t-SNE visualization identified major cell clusters (Fig. 1B). The expression of canonical lineage-specific marker genes resolved eight cell clusters including epithelial cells (EPCAM), T cells (CD3E), Mast cells (KIT), B cells (CD79A), plasma B (MZB1), macrophage (CD68), fibroblasts (COL1A1), endothelial cells (VWF) (Fig. 1C). Focusing on fibroblasts in TME, we further performed subclustering and identified six transcriptionally distinct subsets using established marker genes: senescence-associated fibroblasts, vascular fibroblasts, myofibroblast-like fibroblasts, matrix-degrading fibroblasts, cancer-associated fibroblasts, and antigen-presenting fibroblasts (Fig. 1D, E). Comparative analyses of fibroblast subpopulations revealed significant compositional shifts between CRC tumors and adjacent normal mucosa, with senescent fibroblasts markedly enriched in tumor tissues (Fig. 1F). Subsequently, we performed differential analysis on tumor-derived senescent versus non-senescent fibroblasts using FindMarkers. KEGG pathway enrichment analysis identified multiple lipid metabolism pathways as the predominant functional alteration in senescent fibroblasts (Fig. 1G).

Fig. 1.

Senescent fibroblasts are enriched in colorectal cancer tissues. (A-G) The GSE166555 scRNA-seq data were utilized for analyses. (A) Principal component analysis plot of scRNA-seq results from CRC tissues and adjacent normal tissues. (B) UMAPs showing a scRNA-seq cluster of all cell types from CRC tissues and adjacent normal tissues. (C) Dotplots of marker genes of all cell types. (D) UMAPs showing a scRNA-seq cluster of fibroblasts from CRC tissues and adjacent normal tissues. (E). Dotplots marker genes of all types of fibroblasts. (F). The differences in composition of all types of fibroblasts between CRC tissues and adjacent normal tissues. (G) KEGG analysis of lipid metabolism related differentially expressed genes in the selected senescent and non-senescent fibroblast cell types from CRC tissues. (H) Multiple immunofluorescence staining of p16 (red), panCK (green), Vimentin (yellow) and DAPI (blue) marking cells with human colorectal cancer and adjacent normal tissues. White arrows indicated the specific p16⁺ Vimentin⁺ panCK⁻ senescent fibroblasts. Scale bars, 500 μm. (I) Statistical analysis of the proportions of senescent fibroblasts (n = 3). Data represent the mean ± SEM and subjected to Student’s t-test.***, p < 0.001

To clinically validate these findings, we performed multiplex immunohistochemistry (mIHC) on matched human CRC specimens and adjacent normal tissues. While p16-positive cells were detected in both epithelial and stromal compartments, p16⁺Vimentin⁺panCK⁻ cells exhibiting characteristic fibroblastic morphology were significantly enriched in the CRC specimens than in the adjacent normal mucosa (Fig. 1H, I). These cells displayed diffuse vimentin expression, confirming their mensenchymal identity. This integrated approach, combining high-resolution scRNA-seq with spatial mIHC, establishes p16-expressing fibroblasts as a biologically significant senescent population within the CRC TME. Critically, senescent fibroblast infiltration correlates with CRC progression, suggesting stromal senescence may serve as both a novel biomarker for tumor aggressiveness and a clinically potential therapeutic target.

Machine learning-identified senescent fibroblasts license CD8⁺ T cell crosstalk in CRC

To identify senescent fibroblasts, we established a gene signature comprising CDKN2A, CDKN1A, and key senescence-associated secretory phenotype (SASP) factors. Cellular senescence was defined as a composite score based on the expression levels of CDKN2A and CDKN1A. For classification, we trained diverse machine learning models for senescence state prediction including Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), ElasticNet, XGBoost, Gradient Boosting Machine (GBM), K-Nearest Neighbors (KNN). Evaluation across independent datasets revealed potent classification efficacy for all models. The ROC curves on the test set exhibited high AUC values across the board (Fig. 2A). A comprehensive comparison of performance metrics on the test set identified RF as the top-performing classifier (Fig. 2B). The robust performance of this leading model was successfully replicated on a separate, independent validation cohort underscoring its excellent generalizability (Fig. 2C). The comparative analysis of training and validation performance confirmed that the leading models verified the reliability of our selected model (Fig. 2D). Therefore, RF was adopted for final cell senescence classification. Feature importance analysis revealed five genes, SERPINE1, CDKN1A, TGFBI, CXCL2, and CXCL1, as the key contributors to the model’s decision process (Fig. 2E). The UMAP visualization of cell states based on the model predictions exhibited clear agreement with the expression distribution of canonical senescence markers, further validating the methodological reliability and biological relevance of our machine learning framework (Fig. 2F).

Fig. 2.

Identification and tumor-promoting crosstalk of senescent fibroblasts in CRC. (A) ROC Curves of Classifiers on an Independent Test Set. (B) Comparative Performance of Machine Learning Models on the Test Set. (C) Validation of Model Generalizability on an Independent Validation Set. (D) Assessment of Model Overfitting: Training vs. Validation Performance. (E) Top five feature genes were ranked in the Random Forest model. The importance score reflects the contribution of each gene to the model’s classification accuracy. (F) UMAP projection of single-cell transcriptomes, colored by the senescence labels predicted by the Random Forest model. (G) Circle plots depicting the overall cell-cell communication networks between major cell types in adjacent normal (left) and colorectal cancer (right) tissues. The width of the chords represents the interaction number. (H) UMAP visualization of the cellular heterogeneity, showing the subclustering of fibroblasts and T cells. (I) Circle plots depicting the overall cell-cell communication networks between fibroblast subtypes and T cell subsets in adjacent normal (left) and colorectal cancer (right) tissues. The width of the chords represents the interaction number. (J) Network diagram focusing on the specific interactions emanating from the senescent fibroblast cluster. The width of the chords represents the interaction number

To systematically characterize the cell-cell communication in the tumor microenvironment, we performed comparative analysis on paired CRC and adjacent normal tissues. We observed a markedly enhanced interaction network between fibroblasts and T cells in CRC compared to normal tissue (Fig. 2G, S3A). Following subclustering of fibroblasts and T cells, their heterogeneity was visualized using UMAP projection (Fig. 2H). A focused analysis on the senescent fibroblast subpopulation revealed a specific and strong interaction with CD8⁺ T cells in the tumor microenvironment (Fig. 2I, J, S3B, S3C). This interaction was significantly more potent in CRC than in normal tissue and was notably stronger than the interactions between senescent fibroblasts and either CD4⁺ T cells or NKT cells.

In summary, machine learning identifies senescent fibroblasts with a distinct gene signature. Detailed cell-cell communication analysis revealed a specific and strong interaction between senescent fibroblasts and CD8⁺ T cells in TME of CRC. This interaction is significantly enhanced in tumor tissues, suggesting that it may play an important role in tumor immune regulation. The strong interaction between senescent fibroblasts and CD8⁺ T cells provides new insights into how TME influences immune cell function, particularly the activity of CD8⁺ T cells.

Senescent fibroblasts impair CD8⁺ T cell cytotoxic functions

To investigated the functional consequences of fibroblast senescence on T cell immunity, we first generated a senescent fibroblast model using CCD-18Co colon fibroblasts, with senescence induction confirmed through established biomarkers (Figure S4A, B). Subsequent co-culture experiments of senescent fibroblast with CD8⁺ T cells from human peripheral blood mononuclear cells (PBMC) revealed significant functional impairment. Flow cytometry analysis demonstrated that senescent fibroblasts markedly reduced the proportions of GZMB⁺ (Fig. 3A, B) and IFNγ⁺ (Fig. 3C, D) CD8⁺ T cells, indicating compromised cytotoxic function. Notably, this suppressive was specific to effector molecules, as PD-1⁺ T cell populations remained quantitatively unchanged (Fig. 3E, F). Complementary ELISA analysis of culture supernatants further confirmed substantial reductions in secreted GZMB (Fig. 3G) and IFNγ (Fig. 3H) protein levels, providing biochemical validation that senescent fibroblasts actively diminish CD8⁺ T cell effector capacity.

Fig. 3.

Senescent fibroblasts impair CD8⁺ T cell cytotoxic functions. (A) The expression of GZMB was measured by flow cytometry in CD8⁺ T cells after the co-culture with or without senescent fibroblasts. (B) The percentage of GZMB-expressing cells in the CD8⁺ T cell subsets was shown (n = 3). (C) The expression of IFNγ was measured by flow cytometry in CD8⁺ T cells after the co-culture with or without senescent fibroblasts. (D) The percentage of IFNγ-expressing cells in the CD8⁺ T cell subsets was shown (n = 3). (E) The expression of PD-1 was measured by flow cytometry in CD8⁺ T cells after the co-culture with or without senescent fibroblasts. (F) The percentage of PD-1-expressing cells in the CD8⁺ T cell subsets was shown (n = 3). (G-H) The concentrations of GZMB (G) and IFNγ (H) in the CD8⁺ T cell culture medium after the co-culture with or without senescent fibroblasts were detected with ELISA (n = 6). Data represent the mean ± SEM and subjected to Student’s t-test. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ns, not significant

Senescent fibroblasts suppress CD8⁺ T cell cytotoxicity on patient-derived CRC organoids

To evaluate the functional impact of senescent fibroblasts on CD8⁺ T cell-mediated tumor killing, we established two patient-derived CRC organoids (CRC-PDOs) hereafter designated CRC-PDO-I and CRC-PDO-II, which were derived from two separate CRC patients (Fig. 4A). Initial co-culture experiments confirmed that senescent fibroblasts did not alter CRC-PDO-I (Fig. 4B, C; Figure S5A) and CRC-PDO-II (Figure S6A, B) growth kinetics or viability after 4 days of exposure. We then developed a tri-culture system incorporating senescent fibroblasts, CD8⁺ T cells and CRC-PDOs to model TME interactions (Figure S5B). Strikingly, senescent fibroblasts substantially compromised CD8⁺ T cell cytotoxic function. Z-stack imaging demonstrated significantly from day 0 and day 4, the existence of senescent fibroblasts consistently led to a reduction of the transparency and viability in both CRC-PDOs as compared to the control group (Fig. 4D, E). Meanwhile, the propidium iodide (PI) staining results revealed that co-culture of senescent fibroblasts led to significantly decreased apoptosis in CRC-PDO-I (Fig. 4F, G) and CRC-PDO-II (Fig. 4H, I). Collectively, these data establish that senescent fibroblasts functionally impair CD8⁺ T-mediated tumor elimination by cells through suppression of apoptosis pathways in CRC-PDOs.

Fig. 4.

Senescent fibroblasts suppress CD8⁺ T cell cytotoxicity on patient-derived CRC organoids (CRC-PDOs). (A) The schematic overview of primary culture of CRC-PDOs (upper panel). The representative bright field images of CRC-PDO-I/II. Scale bar, 500 μm. (B-C) Representative bright field images (B) and statistical analysis of normalized viability (C, n = 4) of CRC-PDO-I with or without senescent fibroblast co-cultured at day 4. Scale bars, 500 μm. (D-E) Representative bright field images CRC-PDO-I (D) and CRC-PDO-II (E) with or without CD8⁺ T cells and senescent fibroblast co-cultured from day 0 to day 4. Scale bars, 500 μm. (F-I) Representative multifluorescent staining images of propidium iodide and Hoechst 33,342 of CRC-PDO-I (F) and CRC-PDO-II (H) with or without CD8⁺ T cells and senescent fibroblast co-cultured, evaluating the degree of apoptosis. Statistical analysis of normalized mean fluorescence intensity of propidium iodide/ Hoechst 33,342 in CRC-PDOs (G and I, n = 6). Scale bars, 500 μm. Data represent the mean ± SEM and subjected to Studen’s t-test. **, p < 0.01; ***, p < 0.001; ns, not significant

Senescent fibroblasts drive CD8⁺ T cell dysfunction via CD36-mediated lipid transfer and peroxidation

Given the critical immunomodulatory role of senescent fibroblasts in CRC, we continued to explore the mechanisms underlying their suppression of CD8⁺ T cell function. Prior studies reported altered lipid metabolic profile in tumor-infiltrating versus circulating CD8⁺ T cells, characterized by aberrant lipid accumulation [22, 23]. We first confirmed elevated reactive oxygen species (ROS) in senescent compared with normal fibroblasts (Fig. 5A). To investigate whether lipid species were directly shuttled from senescent fibroblasts to CD8⁺ T cells, we performed a lipid pulse-chase testing. The senescent fibroblasts were labeled with BODIPY-C16 and then co-cultured with CD8⁺ T cells (Fig. 5B). Z-stack imaging revealed near-universal BODIPY-C16 incorporation into CD8⁺ T cells within 24 h (Fig. 5C), while flow cytometry quantification further demonstrated the significant increases in both BODIPY-C16⁺ cell percentage (Fig. 5D, E) and fluorescence intensity (Fig. 5F) in CD8⁺ T cells. Crucially, this lipid transfer could be mediated by CD36. CD36 is a ubiquitously expressed transmembrane receptor facilitating cellular fatty acid internalization through selective ligand binding [24]. CD36 serves as a critical node interfacing signal transduction and metabolic regulation, thereby modulating immune cell phenotypic determination and functional activation [24, 25]. Strikingly, we demonstrated that pharmacological inhibition of CD36 with sulfo-N-succinimidyl oleate (SSO) substantially attenuated the lipid transfer (Fig. 5D, E, F). CD36-mediated lipid uptake in CD8⁺ T cells promoted transferrin receptor 1 expression and intracellular iron influx, culminating in lethal lipid peroxidation that cripples anti-tumor functionality [26]. Consistent with this mechanism, flow cytometric using fluorescein isothiocyanate (FITC) staining confirmed elevated levels of lipid peroxidation in CD8⁺ T cells after co-culture with senescent fibroblasts, while CD36 inhibitor SSO abrogated this phenotype (Fig. 5G, H).

Fig. 5.

Senescent fibroblasts drive CD36-dependent lipid transfer and peroxidation in CD8⁺ T cells. (A) Representative fluorescent images of lipid peroxidation detection in normal and senescent fibroblasts. Scale bars, 50 μm. (B-F) The workflow of detection of BODIPY-C16 transfer from senescent fibroblasts to CD8⁺ T cells (B). CD8⁺ T cells were co-cultured with unlabeled senescent fibroblasts (CTL group), or senescent fibroblasts labeled with BODIPY-C16 (SF group) with or without CD36 inhibitor SSO (SSO group) for one day. Lipid transfer and peroxidation of CD8⁺ T cells were evaluated by microscopy and flow cytometry. (C) Representative bright field and fluorescent images of CD8⁺ T cells. Scale bars, 50 μm. (D) Representative flow cytometric plots of FITC-Bodipy in CD8⁺ T cells. (E) Statistical analysis of the percentage of BODIPY-C16⁺ CD8⁺ T cells (n = 6). (F) Statistical analysis of normalized mean fluorescence intensity of BODIPY-C16 in CD8⁺ T cells (n = 6). (G) Representative flow cytometric plots of FITC-Lipid ROS in CD8⁺ T cells. (H) Statistical analysis of the relative lipid ROS of CD8⁺ T cells (n = 5). Data represent the mean ± SEM and subjected to one-way ANOVA followed by the Tukey post hoc test. *, p < 0.05; **, p < 0.01; ***, p < 0.001

The abrogation of CD36 eliminated the disruptive effects of senescent fibroblasts on CD8⁺ T cells

To establish functional causality, we employed SSO to disrupt CD36-mediated lipid trafficking in the senescent fibroblast and CD8⁺ T cell co-culture system. The intervention significantly rescued the cytotoxic capacity of CD8⁺ T cells, evidenced by the restored frequencies of GZMB⁺ (Fig. 6A, B) and IFNγ⁺ (Fig. 6C, D) CD8⁺ T cells, paralleled by the increased secretion of both effectors into the culure supernatants assayed by ELISA testing (Fig. 6E, F). Critically, with the tri-culture system of CRC-PDOs, CD8⁺ T cells and fibroblasts (Figure S5C), we demonstrated that SSO treatment did not impair the killing capacity, but significantly reversed the senescent fibroblast-indued cytotoxicity impairment of CD8⁺ T cells according to the bright field (Fig. 6G) and PI/Hoechst fluorescent staining images (Fig. 6H, I).

Fig. 6.

Senescent fibroblast led to CD8⁺ T cell dysfunction in a CD36-dependent manner. (A-F) CD8⁺ T cells were co-cultured with unlabeled senescent fibroblasts (CTL group), or senescent fibroblasts labeled with BODIPY-C16 (SF group) with or without CD36 inhibitor SSO (SSO group) for one day. The CD8⁺ T cells were collected for flow cytometry to detect the GZMB (A, B) and IFNγ (C, D) expression (n = 4). The culture medium was collected to measure the concentration of GZMB (E) and IFNγ (F) by ELISA (n = 5). (G-I) CRC-PDOs, CD8⁺ T cells and control/senescent fibroblasts were co-cultured with or without SSO treatment from day 0 to day 4. (G) Representative bright field images of CRC-PDOs. Scale bars, 500 μm. (H) Representative multifluorescent staining images of propidium iodide and Hoechst 33,342 of CRC-PDOs. Scale bars, 500 μm. (I) Statistical analysis of normalized mean fluorescence intensity of propidium iodide/Hoechst 33,342 in CRC-PDOs (n = 5-6). Scale bars, 500 μm. Data represent the mean ± SEM and subjected to one-way ANOVA followed by the Tukey post hoc test. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ns, not significant

Collectively, these data demonstrated that senescent fibroblasts led to the cytotoxicity dysfunction of CD8⁺ T cells via CD36-dependent lipid shuttling, triggering peroxidation that ablates anti-tumor activity.

Senolytic treatment suppresses colorectal carcinogenesis in AOM-DSS induced mouse model

Given the important role of senescent fibroblasts on CRC, we evaluated the therapeutic potential of senolytic-mediated clearance in anti-tumor treatment. The combination of quercetin (Q) and dasatinib (D), commonly referred to as the D + Q senolytic cocktail, represents the current gold-standard therapeutic approach for targeted elimination of senescent cells [27]. The colitis-associated CRC mouse model has significant advantages over traditional subcutaneous tumor implantation approaches in cancer pathogenesis modeling. This methodology uniquely captures the dynamic emergence of senescent cell subpopulations that develop spontaneously during tumor evolution in native tissue contexts [28]. Therefore, the colitis-associated AOM-DSS CRC mouse model was generated accordingly, with D + Q treatment was applied to explore whether the decreased senescent fibroblasts in colon influence CRC initiation and progression (Fig. 7A). Strikingly, the colonoscopy results revealed markedly reduced tumor dimensions in D + Q-treated mice versus controls (Fig. 7B). At the experimental endpoint, no statistically significant differences were observed in body weight (Figure S7A) or colon length (Figure S7B, C) between control and D + Q groups. Post-sacrifice analysis confirmed the significantly attenuated tumor burden (Fig. 7C) in D + Q group than in the control group, which is further reflected by the decreased total number (Fig. 7D) as well as the tumor incidence across all size strata (0–2 mm, 2–4 mm, > 4 mm) (Fig. 7E). The proportion of the tumors of both sizes (0–2 mm, > 2 mm) was statistically analyzed, which demonstrated that D + Q treatment led to a lower incidence of tumors at the size of lager than 2 mm than control group (Fig. 7F). H&E staining results showed that D + Q treatment significantly attenuated the tumor size and malignancy grade as compared to the control group (Fig. 6G). Furthermore, immunohistochemical analysis demonstrated a significant reduction in proliferating cell nuclear antigen (PCNA)-positive cells within tumors from D + Q-treated mice compared to controls (Figure S7D, E). Collectively, senolytic intervention suppresses AOM-DSS-induced colon carcinogenesis.

Fig. 7.

Senolytic treatment suppresses colorectal carcinogenesis in AOM-DSS induced mouse model. (A) Schematic overview of generation of AOM/DSS-induced CRC model and treatment with CTL (normal water) or D + Q (5 mg/kg dasatinib + 50 mg/kg quercetin). (B) High-resolution endoscopic images of mouse colon before sacrifice. (C) Representative macroscopic images of colorectal tumors after sacrifice. (D) The total number of colorectal tumors per mouse was counted for comparison (n = 6). (E) Colorectal tumors were classified according to the diameter (x ≤ 2 mm, 2 < x ≤ 4 mm, x > 4 mm) and counted for comparison (n = 6). (F) The proportion of the tumors of different sizes (0–2 mm, > 2 mm) were counted and statistically analyzed (n = 6). (G) Representative images of H&E staining in CTL and D + Q groups. Scale bars, 500 μm. (H) The mechanistic model demonstrating how senescent fibroblasts subvert CD8⁺ T cell function dependent of CD36. Data represent the mean ± SEM and subjected to two-way ANOVA or Student’s t-test. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ns, not significant

To integrate our finding, we present a mechanistic model delineating how senescent fibroblasts subvert CD8⁺ T cell function (Fig. 7H). Under physiological conditions, CD8⁺ T cells preserve homeostatic lipid metabolism, enabling robust release GZMB/IFNγ secretion and effective tumor cell elimination. Upon exposure to senescent fibroblasts, CD8⁺ T cells undergo CD36-mediated exogenous lipid influx and present a state of lipid peroxidation, which cripples the production and secretion of GZMB and IFNγ, ultimately ablating tumoricidal capacity.

Screening for potential targeted therapeutic drugs for CD36-high colorectal cancer

To identify potential targeted therapies for CD36-high CRC, we performed an in silico prediction of drug sensitivity in the Cancer Genome Atlas colorectal cancer cohort using the oncoPredict algorithm (Fig. 8A). The drug sensitivity profiling revealed 12 compounds with the highest potential therapeutic activity against CD36-high colorectal malignancies (Fig. 8A, B). Notably, these agents demonstrated significantly lower predicted IC50 values in CD36-high patients than in CD36-low counterparts, suggesting increased sensitivity in the CD36-high subgroup. Collectively, these results highlight the potential of CD36 as a biomarker for stratifying patient subsets that may benefit more substantially from specific chemotherapeutic agents.

Fig. 8.

Analysis of differential drug sensitivity according to CD36 expression levels. (A) Drug Sensitivity Volcano Plot (CD36-high vs. CD36-low). (B) Top12 sensitive agents in CD36-high tumors. Analysis of IC50 values for each drug, compared between CD36-high and CD36-low groups

Discussion

Aging and dietary patterns constitute the predominant factors for CRC development [29]. Epidemiologically, CRC exhibits a pronounced age-dependent incidence pattern [30–32]. It is characterized by both increasing occurrence and declining survival outcomes in elderly populations, independent of tumor grade or stage at diagnosis, consistent with most malignancies [30–32]. Our scRNA-seq analysis results quantitatively revealed significant accumulations of senescent fibroblasts within CRC patient specimens, though the precise mechanistic drivers require further elucidation. Crucially, we demonstrate that CRC-PDOs critically depend on lipids secreted by senescent fibroblasts to evade CD8⁺ T cell-mediated cytotoxicity. While cancer-associated fibroblasts represent the major senescent stromal population, the potential contributions of other senescent stromal cells cannot be excluded. Notably, recent studies have identified pro-tumorigenic functions of senescent macrophages in lung adenocarcinoma and senescent neutrophils in prostate cancer [33–35]. Further investigations should therefore determine whether non-fibroblastic senescent stromal cells similarly compromise anti-tumor immunity in CRC, potentially through distinct mechanisms.

Extensive research has established that multiple components within the TME actively suppress cytotoxic CD8⁺ T cell function. Tumor-infiltrating regulatory T cells (Tregs) adapt to the immunosuppressive TME milieu, directly inhibiting CD8⁺ T cell activation [36]. Concurrently, IL-15/IL-15Rα complexes expressed by tumor-associated macrophages induce downregulation of CX3CL1 in tumor cells, thereby limiting recruitment of CX3CR1⁺CD8⁺ T cells [37]. These macrophages further impair T cell priming by secreting IL-10 to suppress dendritic cell production of IL-12 [38]. Building on this foundation, our study now identifies senescent fibroblasts as pivotal and previously overlooked contributors of CD8⁺ T cell dysregulation.

Senescent cells primarily drive disease progression through the SASP, which encompasses growth factors, cytokines, chemokines, and bioactive lipids [14]. While cytokines and their interactions with lymphocytes have been extensively studied, the immunoregulatory potential of SASP-derived bioactive lipids remains largely unexplored despite their established roles in modulating immune cell behavior and responses [39]. Our study directly addresses this gap by demonstrating that CD8⁺T cells internalize labeled lipids from senescent fibroblasts, resulting in significant impairment of cytotoxic function (Figs. 5 and 6). This lipid-mediated mechanism operates alongside other reported documented SASP effects. For instance, colorectal senescent fibroblasts secrete GDF 15 to establish chronic inflammatory microenvironments that promote CRC susceptibility [40]. In addition, matrix metalloproteinase and IL6 secreted by senescent fibroblasts accelerate breast and pancreatic cancer progression [41, 42]. Although our findings establish lipids as key mediators of CD8⁺ T cell dysfunction, we cannot exclude synergistic effects from other SASP components that may concurrently modulate T cell activity.

Substantial evidence indicates that senescent cells undergo significant alterations in lipid metabolic composition and flux dynamics when compared to normal proliferative cells. Emerging studies reveal critical metabolic crosstalk between senescent cells and their microenvironment. For example, melanoma cells were demonstrated to develop pronounced dependency on fatty acids secreted by senescent fibroblasts [43]. This metabolic reprogramming extends across cell types. Senescent hepatocytes exhibit disrupted lipid homeostasis through impaired fatty acid oxidation, driving pathogenic lipid deposition and β-cell dysfunction [44]. Dopaminergic neurons show perilipin-2-mediated lipid droplet accumulation, implicating senescence in neurodegenerative pathogenesis [45]. Moreover, although senescent fibroblasts demonstrate distinct lipid metabolic profiles, the specific lipid species exerting biological effects remain poorly characterized.

In parallel, tumor-infiltrating CD8⁺ T cells display elevated lipid accumulation mediated by the transmembrane receptor CD36, which facilitates uptake of microenvironmental lipids including arachidonic acid and oxidized species [26, 46]. This CD36-dependent lipid internalization induces cytotoxic impairment through lipid peroxidation cascades, with excessive peroxidation potentially triggering ferroptosis of CD8⁺ T cells [47, 48]. Notably, preclinical evidence revealed that CD36 blockade with PD-1 inhibition could effectively rescue the T cell-mediated antitumor immunity [47, 48]. Our findings mechanistically extend this paradigm by demonstrating that senescent fibroblasts directly transfer lipids to CD8⁺ T cells via CD36-dependent uptake, which drives metabolic exhaustion and cytotoxicity loss. CD36 blockade can preserve T cell effector functions as well as tumor killing capacity. Additional studies will be performed to validate the role of CD36-mediated lipid transfer, specially its contribution to senescent fibroblast-induced CD8⁺ T cell dysfunction in tumor progression and immune invasion with its antagonist.

While senolytic therapies such as the first-generation combination D+Q have demonstrated efficacy across diverse preclinical disease models, their clinical translation faces significant challenges [49]. Over 20 clinical trials of senolytic therapies are currently completed, ongoing or planned, yet critical limitations persist [50]. For example, the human toxicity profiles are incompletely understood, and the suboptimal senescent cell clearance efficiency is hard to be evaluated [49]. To circumvent these constraints, emerging strategies employ precision-targeted delivery systems. For instance, the engineered vesicles loaded with D + Q that selectively bind senescent cell surface markers in mouse colon cancer models were developed [51]. Although pending trial data will refine senolytic applications, fundamental hurdles in specificity and safety must be resolved before broad clinical application.

The AOM-DSS induced CRC model achieves superior biological fidelity by recapitulating the complex crosstalk between transformed epithelial cells and stromal networks within the intestinal microenvironment. At the molecular level, integrated inflammatory stimuli trigger ROS-mediated oxidative damage while activating cell cycle checkpoint regulators (p16/p21 pathways), driving growth arrest and senescence phenotype acquisition in intestinal cells [52]. Critically, our senolytic cocktail experiments in this model demonstrate the essential role of senescent fibroblast in CRC progression, positioning senolytic targeting of stromal compartments as a strategic approach to enhance antitumor immunity.

Conclusion

In summary, our study defines the role of senescent fibroblasts in regulating CD8⁺ T cells associated with cytotoxicity alterations in CRC. Our results revealed that senescent fibroblast mediated lipid transfer facilitated lipid deposition and peroxidation in CD8⁺ T cells hampering their killing ability on CRC-PDOs and that senolytic elimination of senescent cells could prevent the occurence and progression of CRC. Our findings support the feasibility of targeting fibroblast senescence or lipid metabolic pathways as a promising option for immunotherapy.

Methods

Collection of clinical samples

Human CRC specimens and normal colon specimens were obtained from Huashan Hospital of Fudan University. The study was approved by the Research Ethics Committee of Medical Research, Huashan Hospital of Fudan University (KY2022-913). Human CRC specimens were rinsed in PBS (C0221A, Beyotime, Shanghai, China) with 1% penicillin-streptomycin (C0222, Beyotime, Shanghai, China) three times. Then the specimens were cut into pieces (1–3 mm3 each). Tissue pieces were digested in tumor tissue digestion solution (K601003, bioGenous, Suzhou, China) in the shaker at 37 ℃ for 5 min. The 100 μm cell strainer was used to collect the isolated crypts. After centrifuged at 500 x g for 5 min at 4 ℃, the precipitation was then mixed in Matrigel (356231, Corning, USA) and cultured at 37 ℃ with 5% CO₂. CD8⁺ T cells were isolated from PBMC (isolated via Ficoll gradient) of healthy donors with Human CD8⁺ T Cell Isolation Kit (100–0710, Stemcell, Canada), according to manufacturer’s instructions. CD8⁺ T cells were cultured in RPMI medium supplemented with 10% FBS, 1% penicillin-streptomycin, 2mM L-glutamine, 10 mM HEPES and 50 µM β-mercaptoethanol, and were activated for a week with human IL2 (200-02, Thermo Fisher Scientific, USA) at the concentration of 300 IU/mL.

Animal studies

C57BL/6J mice were obtained from GemPharmatech Co. and housed at the room under controlled conditions of 22–24 ℃ and a 12-h light/dark cycle, with free access to water and standard chow. CRC tumor formation was monitored with mouse colonoscopy. At the end of the experiment, the mice were sacrificed with the colorectal tract excised following longitudinally cutting. The colorectal tract was washed with PBS for three times, and the number and size of visible tumors throughout the colon were recorded.

AOM/DSS-induced colorectal cancer mouse model

For the AOM-DSS CRC mouse model, 8-week-old mice were intraperitoneally injected with AOM (10 mg/kg body weight) (A5486, Sigma-Aldrich, St. Louis, USA) on the first day, followed by 2% DSS (MFCD00081551, MP Biomedicals, USA) in drinking water for one week and next assigned normal drinking water to recover for two weeks. The mice were treated with or without the senolytic cocktail dasatinib (5 mg/kg) (S1021, Selleck, Houston, USA) and quercetin (50 mg/kg ) (S2391, Selleck, Houston, USA) via oral gavage for three consecutive days to clear the senescent cells during the second and third recovery period. After three rounds of DSS treatment with intermittent rest periods, the mice were additionally administered D + Q for three days and were sacrificed for sample collection at the end of week 11.

Cell experiments

Human colon fibroblast cell line CCD-18Co was obtained from the QuiCell Biotechnology Company (Shanghai, China). Cells were cultured in Dulbecco’s modified Eagle’s medium (11966025, Thermo Fisher Scientific, USA) supplemented with 20% fetal bovine serum (12484010, Thermo Fisher Scientific, USA) and 1% penicillin-streptomycin at 37 ℃ with 5% CO₂.

CCD-18Co cells were cultured in medium supplemented with 400 µM hydrogen peroxide (30% w/v ) (H1009, Sigma-Aldrich, St. Louis, USA) for 2 h and recoveried in growth media for 3 days as a cycle. Senescence induction of colon fibroblasts was completed after 5 cycles of treatments. For SA-β-Gal analysis, cells were incubated with SA-β-Gal solution (SG02, Dojindo, Japan) at 1 µM for 15 min at 37 ℃. Cells were washed with PBS followed by DAPI staining. Then, the stained cells were photographed under the microscope.

Cell and organoid counting

To count the CD8⁺ T cell concentration, the cell suspension was gently pipetting with 20 µL removed out and mixed with 20 µL 0.4% trypan blue solution. After incubation at room temperature for 2 min, 10 µL of the mixture was loaded into a hemocytometer until the counting chamber is filled. Count the number of viable cells in five large squares and calculate the average total number (N) of cells per square. The cell concentration is calculated as 2 N × 10⁴/ml. To count the number of organoids, the organoids were gently dissociated from the Matrigel and the total volume after re-suspension was measured. Then a certain volume of the suspension was transferred into a culture dish and the total number of organoids was counted under a microscope. The total number of organoids was calculated according to the proportion. Centrifuge the organoids and resuspend them at a final concentration of 1 × 10⁴/ml.

ELISA assay of GZMB and IFNγ

The concentration of GZMB or IFNγ in the culture medium was detected using Human GZMB ELISA Assay Kit (EH10206M, Weiaobio, Shanghai, China) or Human IFNγ ELISA Assay Kit (EH10230M, Weiaobio, Shanghai, China) according to the manufacturer’s recommended protocol. First, 50 µL standard solution and 50 µL sample supernatant was added into the wells, and mixed at 37 °C to react for 50 min. Wash each well thoroughly for three times with 300 µL washing buffer. 100 µL of biotinylated antibody working solution was further added into each well to react at 37 °C for 50 min. Wash each well thoroughly for three times with 300 µL washing buffer. Add 100 µL SABC complex working solution to each well and react at 37 °C for 30 min. After washing for three times, add 100 µL substrate solution and react at 37 °C for 15 min. Add 50 µL stop solution to each well and measure the absorbance at 450 nm using a microplate reader. Create a standard curve of concentration-OD values and calculate the final concentration of each sample.

Histological staining, multiple immunofluorescence staining and analysis

Tissues were fixed in 4% paraformaldehyde (P0099, Beyotime, Shanghai, China) and then embedded in paraffin. The tissue sections were used for hematoxylin and eosin (H & E) staining. For multiple immunofluorescence staining, sections were stained with primary antibodies including p16 (GB111143, Servicebio, Wuhan, China), Vimentin (GB11192, Servicebio, Wuhan, China), and panCK (GB122053, Servicebio, Wuhan, China). After re-probed with corresponding secondary antibodies, the stained sections were photographed under the microscope.

Primary culture and analyses of patient-derived CRC organoids

The patient-derived CRC organoids were primarily cultured as previously described [19] and supported by a serum-free medium kit (K2103-CR, Biogenous, Suzhou, China). During the culture, CRC organoids culture medium was refreshed every two days. When the confluency reached about 70–80%, the CRC organoids were digested by TrypLE Express Enzyme (12605028, Gibco, USA) for passaging at a 1:3 or 1:4 ratio within 6 days. The organoids were photographed under the microscope and analyzed.

Cell and CRC organoid co-culture models

The organoids were dissociated by TrypLE Express Enzyme for 15 min, washed for three times by PBS, and then resuspended with Matrigel at a ratio of 1000 organoids per 100 uL into the plate. Resuspend in 20 µL fibroblast medium per 5000 fibroblasts and add 20 uL suspension into the plate. After the co-culture of 4 days, add the suspension mixture of CD8⁺ T cells combined with senescent fibroblasts into the plate. Incubate for 4 days in the incubator. Organoids were collected for detection. 5 μm SSO was added into the CD8⁺ T cells to eliminate the effects of senescent fibroblasts.

Organoid viability ATP assay

The organoid viability was measured using an Organoid Viability ATP Assay Kit (E238003, bioGenous, Suzhou, China) according to the manufacturer’s recommended protocol. An equal volume of cell viability detection reagent as the culture medium was added to each well of the culture plate. The plate was then subjected to orbital shaking to ensure homogeneous mixing. Following a 20-minute incubation at room temperature to facilitate complete organoid lysis, chemiluminescence signals were measured with a microplate reader. To normalize and analyze the relative organoid viability, the chemiluminescence value of each well is subtracted by the chemiluminescence value of the blank well to obtain a relative value. The average organoid viability of the control group is designated as a baseline value of 1, and the values of the experimental group are proportionally converted to obtain the value as the normalized viability of organoids.

Lipid pulse-chase assay

Label the lipids in senescent fibroblasts by 1 µM BODIPY FL C16 (D3821, Invitrogen, California, USA) for 6 h at 37 ℃. Cells was then collected and washed three times by PBS supplemented with 0.2% fatty acid-free BSA (ST025, Beyotime, Shanghai, China) to completely remove the dye. Then seed the labeled cells into the chambers of the transwell together with CD8⁺ T cells. After the incubation of 24 h, the T cells were digested and the single cell suspension was then collected for the detection of the BODIPY FL C16 signal via flow cytometry.

ScRNA-seq analysis

ScRNA-seq data was downloaded from GEO database (https://www.ncbi.nlm.nih.gov/geo/), GSE166555 [53], analyzed using Seurat (v5.2.1) after downloading, with quality control standards nFeature_RNA > 200 & < 5000 (Figure S1A), nCount_RNA < 40,000 (Figure S1B), percent.mt < 10 (Figure S1C). DoubletFinder (v2.0.6) was used to remove doublets. Identify highly variable genes in the dataset (Figure S2A). Normalization and dimensionality reduction clustering were performed using default parameters, with dims chosen as 1:20, and classic markers were used for cell annotation. Principal component analysis plot was used to visually demonstrate the global similarities and differences among various samples (Figure S2B). UMAP was used to display single-cell dimensionality reduction clustering and cell annotation results (Figure S2C). A dot plot showed the expression of marker genes (Figure S2D). Genes were defined as differential genes if the fold change was greater than or equal to 2 and the adjusted p-value was less than 0.05. Enrichment analysis was performed using clusterProfiler (v4.14.6), with an entry being considered enriched if the p-value was less than 0.05.

Machine learning-based identification of senescent fibroblasts

A senescence gene signature was defined including CDKN2A, CDKN1A, and SASP factors (IL6, IL8, MMP3, GDF15). Senescence labels were assigned using a composite score of CDKN2A and CDKN1A expression, with cells above the 85th percentile classified as senescent. RF, SVM, LR, ElasticNet, XGBoost, GBM, and KNN classifiers were trained using the caret package with 5-fold cross-validation. Model performance was evaluated by AUC-ROC, accuracy, and feature importance. Expression data were extracted using GetAssayData with layer = “data”.

Cell-cell communication analysis

Cell-cell communication was systematically inferred using the CellChat R package, applied separately to single-cell RNA-seq data from colorectal cancer and matched normal tissues. The normalized expression matrix and pre-defined cell type labels served as input. Differential analysis between normal and cancer conditions was performed using integrated CellChat objects. Specific interactions between fibroblast subsets (particularly senescent fibroblasts) and T cell subsets were extracted for comparative analysis. All visualizations were generated using built-in package functions.

Flow cytometry

CD8⁺ T cells were stimulated by leukocyte activation cocktail (550583, BD Pharmingen, USA) for 5 h and first incubated by the antibody mixture of surface staining for 30 min at 4 ℃. For intracellular staining, the cells were washed, then fixed and permeabilized using a Foxp3 Transcription Factor Staining Buffer Set (562574, BD Pharmingen, USA), followed by intracellular antibody incubation for 30 min at 4 ℃. The antibodies were obtained from BD Pharmingen: APC-Cy7 Mouse Anti-Human CD3, BB515 Mouse Anti-Human CD4, BB700 Mouse Anti-Human CD8, BV711 Mouse Anti-Human IFN-γ, PE-CF594 Mouse Anti-Human GZMB, BV-510 L/D. Flow cytometric analysis was performed on a Cytek^® Aurora flow cytometer (Cytek) and analyzed using FlowJo_v.10.9.0. software.

Propidium iodide staining

Collect the PDOs and incubate them with 1.5µM propidium iodide staining solution (40710ES03, Yeasen, Shanghai, China) for 20 min at 37 ℃. Cells were washed with PBS followed by Hoechst 33,342 staining (40731ES10, Yeasen, Shanghai, China)(0.5 µg/ml). Then visualize the stained cells under the microscope and analyze the necrosis ratio by Image J software.

Lipid peroxidation testing

The Lipid Peroxidation Assay Kit with BODIPY 581/591 C11 (S0043M, Beyotime, Shanghai, China) was used to test the degree of intracellular lipid peroxidation. Incubate target cells by 2 µM BODIPY 581/591 C11 for 20 min at 37 ℃. Cells were then washed with PBS followed by DAPI staining. The stained cells were detected via flow cytometry or the microscope.

Drug sensitivity prediction via oncopredict

Gene expression data from The Cancer Genome Atlas colorectal cancer cohort were analyzed. Patients were stratified into CD36-high and CD36-low groups based on median expression. Using the R package oncoPredict, drug sensitivity (IC50) for compounds in the GDSC2 database was predicted in silico. Differential drug response between the two groups was assessed to identify therapeutics with significantly enhanced predicted efficacy in CD36-high tumors.

Statistical analysis

Statistical analyses were performed using GraphPad Software (Prism 9.0). Data are presented as the mean ± SEM. Unless otherwise stated, one-way ANOVA followed by the Tukey post hoc test was utilized to assess statistical significance among multiple groups. Unpaired Student’s t-test was used for comparisons between two groups. The results are obtained from at least three independent experiments. A p value of less than 0.05 was considered significant. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ns, not significant.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

TME

Tumor microenvironment

CRC

Colorectal cancer

scRNA-seq

Single-cell RNA sequencing

CRA-PDOs

Patient-derived organoids of CRC

GZMB

Granzyme B

IFNγ

Interferon gamma

mIHC

multiplex immunohistochemical

SASP

Senescence-associated secretory phenotype

AUC-ROC

Area under the receiver operating characteristic curve

RF

Random Forest

SVM

Support Vector Machine

LR

Logistic Regression

GBM

Gradient Boosting Machine

KNN

K-Nearest Neighbors

PBMC

Peripheral blood mononuclear cells

FMO

Fluorescence minus one

Q

Quercetin

D

Dasatinib

PCNA

Proliferating cell nuclear antigen

SASP

Senescence-associated secretory phenotype

Author contributions

M.G., S.S., W.C., and Z.X. contributed acquisition, analysis, and interpretation of data. M.G., S.S., W.C., and D.K. contributed to the acquisition of data and analysis. M.G. and Y.G. contributed to the study design. M.G., S.S., W.C., and Z.W. contributed to the article revision. M.G. and Y.G. contributed to the manuscript writing.

Funding

This study was funded by the National Key Research and Development Program of China (Grant Number: 2024YFA1306000).

Data availability

Public data is available in a GEO database (https://www.ncbi.nlm.nih.gov/geo/) and RNA-Sequencing data in the study was uploaded into GEO database with the number GSE166555.

Declarations

Ethics approval and consent to participate

Two human colorecral tumor biopsy samples were collected for research purposes, with approval from the Ethics Committee of Huashan Hospital, Fudan University (Approval No. KY2022-913). All animal experiments were conducted in accordance with relevant guidelines and were approved by the Institutional Animal Care and Use Committee (IACUC) of Fudan University (2021JS-398).

Consent for publication

All authors of the manuscript have read and agreed to its content and are accountable for all aspects of the accuracy and integrity of the manuscript in accordance with ICMJE criteria.

Conflict of interest

The authors have no conflicts to disclose.