Tumor stroma-immune interactions shape the immunosuppressive microenvironment and predict response to neoadjuvant chemoradiotherapy plus immunotherapy in rectal cancer

Abstract

Background

The tumor microenvironment, particularly the tumor stroma, plays a critical role in tumor progression, immune evasion, and therapeutic resistance. However, its interaction with the immune landscape in rectal cancer (RC) remains incompletely understood. This study aimed to comprehensively characterize the stromal-immune ecosystem associated with the tumor stroma ratio (TSR) in RC and to evaluate its clinical and therapeutic relevance.

Methods

We analyzed a multicenter cohort of 498 patients with treatment-naïve RC in whom TSR was assessed on H&E-stained sections. Integrative multi-omics analyses were performed, including bulk RNA sequencing (n=118) and single-cell RNA/T-cell receptor (TCR) sequencing (n=10). Key findings were validated by immunohistochemistry (n=114) and multiplex immunofluorescence (n=20). Survival analyses and statistical comparisons were conducted to evaluate clinical associations and treatment responses.

Results

High TSR was an independent predictor of unfavorable disease-free survival and cancer-specific survival and was associated with aggressive clinicopathological features. Single-cell analyses revealed that TSR-high tumors exhibited a profoundly immunosuppressive microenvironment, characterized by clonally expanded terminally exhausted CD8⁺ T cells (CD8⁺ Tex-CXCL13) and activated CD4⁺ regulatory T cells (CD4⁺ Treg-TNFRSF4). Two LRRC15⁺ cancer-associated fibroblast (CAF) subsets (mCAF-CTHRC1 and mCAF-FAP) were enriched in TSR-high tumors. Among them, mCAF-CTHRC1 was associated with increased Treg abundance and activation features, with predicted interactions with CD4⁺ Treg-TNFRSF4 cells through the LGALS9–CD44 signaling axis. In addition, SPP1-expressing monocytes (Mon-SPP1) and malignant epithelial cells were prominent in TSR-high tumors and showed a predicted SPP1–CD44 interaction with T-cell subsets, suggesting potential involvement in immunosuppressive stromal-immune interactions. In patients receiving neoadjuvant therapy, pretreatment TSR-low tumors showed improved pathological response and survival outcomes compared with TSR-high tumors. In the neoadjuvant chemoradiotherapy plus immunotherapy cohort, TSR-low tumors were associated with a significantly higher major pathological response rate, whereas pathological complete response showed a non-significant trend in the same direction.

Conclusions

High TSR identifies a clinically aggressive subtype of RC characterized by a profoundly immunosuppressive stromal-immune ecosystem enriched for exhausted T cells, immunosuppressive CAF programs, and SPP1-associated stromal-myeloid interactions. These findings highlight LGALS9−, LRRC15−, and SPP1-related stromal-immune pathways as candidate stromal-immune therapeutic vulnerabilities that warrant further mechanistic and preclinical validation.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• The tumor stroma is a key component of the tumor microenvironment and contributes to tumor progression, immune evasion, and therapeutic resistance. The tumor-stroma ratio (TSR) has emerged as a prognostic biomarker in several solid tumors, including colorectal cancer. However, the stromal-immune interactions underlying TSR heterogeneity in rectal cancer remain poorly understood.

WHAT THIS STUDY ADDS

• Using integrative multi-omics analyses, we show that TSR-high rectal cancers harbor a profoundly immunosuppressive microenvironment characterized by terminally exhausted CD8⁺ T cells, activated CD4⁺ Tregs, LRRC15⁺ CAFs, and SPP1⁺ monocytes and tumor cells. These analyses suggest that predicted LGALS9–CD44 signaling from CAFs and SPP1–CD44 signaling from myeloid and epithelial cells may contribute to T-cell suppression.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• TSR may serve as a practical biomarker linking stromal architecture with immune suppression and treatment response in rectal cancer. Targeting stromal-immune interactions, including LRRC15, LGALS9, and SPP1 pathways, may represent a promising strategy to overcome stromal-mediated immune evasion and improve immunotherapy outcomes.

Introduction

Rectal cancer (RC) is one of the most common malignant tumors worldwide and a significant cause of cancer-related mortality.¹ Recent studies have highlighted the critical role of the tumor microenvironment (TME) in tumor initiation, progression, metastasis, and the development of resistance to therapy.² The tumor stroma, a key component of the TME, is characterized by its high dynamicity, heterogeneity, and tumor-type specificity, primarily comprising cancer-associated fibroblasts (CAFs), endothelial cells, and extracellular matrix.³ Interactions between tumor cells and the tumor stroma play a decisive role in tumor progression, metastasis, and therapeutic resistance.^{3 4} Notably, the tumor-stroma ratio (TSR), defined as the proportion of tumor stroma to tumor cell components, has emerged as an important prognostic and predictive biomarker in various solid tumors, including colorectal cancer (CRC).^5–10 However, despite the shared classification under CRC, colon and RCs differ substantially in anatomical context, treatment paradigms, and tumor microenvironmental remodeling. In particular, RC has a distinct therapeutic framework centered on neoadjuvant chemoradiotherapy, with increasing incorporation of immunotherapy-based strategies. These treatment modalities may dynamically reshape the stromal-immune landscape and create a unique clinical setting in which TSR may not only serve as a prognostic biomarker but also help stratify sensitivity to neoadjuvant chemoradiotherapy, with or without immunotherapy. Therefore, investigating TSR and stromal-immune interactions specifically in RC represents a clinically relevant and distinct context that is not fully captured in previous CRC-based studies.

Increasing evidence suggests that the tumor stroma significantly influences tumor biology, particularly in cancer immune evasion mechanisms.³ In CRC and other tumor types, the tumor stroma not only provides structural support for tumor growth and expansion but also regulates the tumor immune microenvironment through interactions with immune cells, thereby impacting antitumor immune responses.¹¹ Studies have shown that the tumor stroma can promote tumor immune evasion by creating an immunosuppressive microenvironment, which in turn affects treatment response and patient prognosis in certain solid tumors.^{5 10 12} Specifically, tumor-associated stromal cells, such as CAFs, may secrete immunosuppressive cytokines, impairing the migratory capacity of immune cells and leading to an “immune-excluded” phenotype, which further suppresses antitumor immunity.¹³ Additionally, the dense stromal architecture poses unique challenges for immune cell infiltration.^{14 15} However, the precise role of the tumor stroma in the TME of RC, particularly its interplay with the tumor immune microenvironment, remains incompletely understood. Therefore, investigating the interaction between the tumor stroma and the tumor immune microenvironment in RC, especially how the stroma regulates the immune microenvironment to influence immune evasion, holds significant clinical relevance.

This study aims to systematically elucidate the complex interplay between tumor stroma and the TME through an integrative analysis of single-cell RNA sequencing (scRNA-seq), single-cell T-cell receptor sequencing (scTCR-seq), bulk RNA sequencing (RNA-seq), immunohistochemistry (IHC), and multiplex immunofluorescence (mIF). We will explore the mechanisms underlying the role of tumor stroma in the RC TME, particularly its potential functions in regulating the immune microenvironment, and identify potential clinical applications in the process.

Methods

Patient data

To investigate the prognostic value of the TSR, 498 patients with treatment-naïve RC who underwent curative surgery at two institutions were included in the survival analysis. Of these, 282 patients treated at Peking Union Medical College Hospital (PUMCH) between January 2017 and October 2019 were assigned to the discovery cohort, while 216 patients from the Chinese PLA General Hospital (PLAGH) between January 2016 and December 2020 comprised the validation cohort. To further characterize stromal-immune interactions within the TME, multiple molecular and histological datasets were included, comprising scRNA-seq (n=10; collected between April 2025 and July 2025 from PUMCH and PLAGH), paired scTCR-seq (n=10), bulk RNA-seq (n=118; collected between September 2023 and November 2024 from PUMCH and PLAGH), IHC (n=114; collected between September 2021 and September 2023 from PLAGH), and mIF (n=20; collected between March 2023 and September 2023 from PLAGH).

For therapeutic response analyses, two neoadjuvant treatment cohorts were additionally included: a neoadjuvant chemoradiotherapy (NCRT) cohort (n=151), comprising patients treated between January 2013 and January 2018, and a neoadjuvant chemoradiotherapy plus immunotherapy (NCRIT) cohort (n=33), comprising patients treated between March 2023 and January 2026 at PLAGH. These cohorts were used to evaluate the association between TSR status in pretreatment tissues and therapeutic response.

In the NCRT cohort, all patients received long-course chemoradiotherapy (CRT) (50 Gy/25 f, 2 Gy/f, 5 days/week plus three 21-day cycles capecitabine 850–1,000 mg/m², two times per day, orally, day 1–14) as neoadjuvant. In the NCRIT cohort, all patients received three 21-day cycles of tislelizumab, an anti-PD-1 antibody, at 200 mg intravenously on day 8, with long-course CRT as neoadjuvant. These patients then received total mesorectal excision (TME) for radical resection 6–12 weeks after the end of radiotherapy.

All patients included in this study were confirmed as proficient mismatch repair (pMMR). Mismatch repair (MMR) status was assessed by immunohistochemistry for MLH1, PMS2, MSH2, and MSH6. Tumors with retained nuclear expression of all four MMR proteins were classified as pMMR. Patients with deficient MMR tumors were not included in the present study.

The study received approval from the ethics committees of PUMCH (I-25PJ0636) and PLAGH (S2025-047-02). The study was conducted in adherence to the principles outlined in the Declaration of Helsinki. All patients included in this study provided informed consent before surgery, permitting the use of clinical profiles and discarded specimens for research purposes.

Clinicopathological characteristics

Clinicopathological features included age, sex, body mass index, tumor location (low, middle and upper), tumor diameter, T stage, N stage, lymph node ratio (LNR), tumor grade (G1‐2: well/moderate; G3: poor), tumor deposits, vascular invasion, perineural invasion, were obtained from pathology reports. The tumor, node, metastases (TNM) stage was re-evaluated by the eighth edition TNM staging system. Tumor markers, including carbohydrate antigen 19–9 (CA19‐9) and carcinoembryonic antigen (CEA), were measured within 1 week before surgery.

The assessed follow‐up outcomes were disease‐free survival (DFS) and cancer‐specific survival (CSS). DFS was defined as the time between radical surgery and recurrence during follow‐up. CSS was defined as the time between radical surgery and death due to RC during follow‐up.

Tumor regression score (TRS) in the resected specimen after neoTx was calculated using the recommendations provided by the College of American Pathologists and American Joint Committee on Cancer guidelines. Patients with TRS 0 and 1 were considered as major pathological response (MPR), and those with TRS 2 and 3 as non-MPR (NMPR). Pathological complete response (pCR) was defined as TRS 0, while patients with TRS 1–3 were considered non-pCR.

Assessment of the tumor-stroma ratio

All available H&E-stained sections were reviewed and the representative sections were selected. These representative sections should include the tumor tissue and normal tissue surrounding the tumor tissue. The corresponding tissues were subsequently resected into 4 µm thick slices. All the representative H&E-stained sections were scanned using a slide scanner (SQS-40P, TEKSQRAY, China) and analyzed using ImageViewer (DPVIEW, TEKSQRAY, China) software.

TSR was identified using the corresponding H&E-stained sections. The most invasive tumor field was selected at high magnification (×100) and tumor cells would have to be present on all field boundaries. TSR was defined as the percentage of tumor stroma in the field. In this study, TSR >50% was considered as TSR-high and TSR ≤50% as TSR-low (online supplemental figure 1A,B), as described by Park et al.⁶ Interobserver agreement for TSR assessment was evaluated using Cohen’s kappa (κ) statistic. The inter-rater agreement was κ=0.814, with an overall agreement accuracy of 90%. When discrepancies occurred between observers, the slides were reviewed jointly and the senior pathologist made the final decision.

Immunohistochemistry, multiplex immunofluorescence and bioinformatic analysis

IHC and mIF were performed on formalin-fixed paraffin-embedded tissue sections following standard protocols. Bioinformatic analyses, including bulk RNA-seq processing, functional enrichment analysis, scRNA-seq analysis, TCR clonotype analysis, copy number variation inference, pseudotime trajectory analysis, and cell preference analysis, were conducted using established pipelines. Detailed experimental procedures and analytical methods are described in the online supplemental methods.

Statistical analysis

The statistical analysis software used in this study included R V.4.2.1, SPSS V.25, and GraphPad Prism V.8. Continuous variables were summarized as median and IQR and compared using the Mann-Whitney U test for two independent groups or the Wilcoxon signed-rank test for paired data. Categorical variables were summarized as counts and percentages and compared using the Pearson χ² test when all expected cell counts were ≥5; otherwise, Fisher’s exact test was used. Survival curves were estimated using the Kaplan-Meier method and compared using the log-rank test. Prognostic factors associated with DFS and CSS were evaluated using univariate and multivariate Cox proportional hazards regression models. Variables with clinical relevance or statistical significance in univariate analysis were included in the multivariate models.

For scRNA-seq analyses, cell-type proportions were calculated for each individual patient before comparison between TSR-high and TSR-low groups using the Mann-Whitney U test. For targeted gene-expression and signature-score comparisons within specific cell subsets, expression values or scores were averaged at the patient level when used to support between-group biological differences, whereas cell-level analyses were used for visualization and exploratory characterization. Bulk RNA-seq differential expression was analyzed using DESeq2, with Benjamini-Hochberg false discovery rate (FDR) correction. Single-cell cluster-specific markers were identified using Seurat-based Wilcoxon rank-sum testing with Benjamini-Hochberg FDR correction. For predefined hypothesis-driven comparisons, including selected cell populations, marker genes, signature scores, pathological response rates, and IHC/mIF quantifications, nominal two-sided p values were reported without additional FDR correction. All statistical tests were two-sided, and p<0.05 was considered statistically significant.

Results

Clinicopathological characteristics of the patients

This study included 498 patients with treatment‐naïve RC who had undergone radical surgery (282 and 216 patients from the training cohort and validation cohort, respectively) (figure 1A). Figure 1B illustrates the distribution of various clinicopathological characteristics in the multicenter cohort. The detailed clinicopathological characteristics of the two independent cohorts (training and validation) are presented in table 1 and online supplemental table S1. The TSR levels were assessed based on H&E slides (online supplemental figure 1A,B). The proportions of patients in the high-TSR and low-TSR groups were similar between the training and validation cohorts: in the training cohort, 101 patients (35.8%) were classified as high TSR, while 181 patients (64.2%) were classified as low TSR; in the validation cohort, 88 patients (40.7%) were in the high-TSR group, and 128 (59.3%) were in the low-TSR group. In the training cohort, high TSR was significantly associated with higher pathological T stage, N stage, LNR, CEA, CA19-9 levels, perineural invasion, and positive tumor deposits. Similarly, in the validation cohort, high TSR was significantly associated with higher pathological T stage, N stage, LNR, CEA, CA19-9 levels, vascular invasion, perineural invasion, and positive tumor deposits.

Figure 1. Clinicopathological characteristics of the RC cohorts and prognostic and therapeutic relevance of TSR. (A) Screening workflow of the multicenter RC cohorts. (B) Distribution of clinicopathological characteristics in the RC cohort. (C–D) Kaplan-Meier curves showing DFS and CSS according to TSR status in the training (C) and validation (D) cohorts. (E–F) Forest plots of multivariate Cox regression analyses for DFS and CSS in the training (E) and validation (F) cohorts, respectively. (G) Association between TSR status and pathological response in the NCRT cohort. P values were calculated using Fisher’s exact test. (H) Kaplan-Meier curves for DFS according to TSR status in the NCRT cohort. (I) Association between TSR status and pathological response in the NCRIT cohort. P values were calculated using Fisher’s exact test. (J) Paired comparison of TSR values before and after NCRIT in patients achieving MPR or NMPR. Each line represents an individual patient. P values were calculated using the paired Wilcoxon signed-rank test. (K) Waterfall plot showing individual changes in TSR after NCRIT. Bars represent the change in TSR values between post-treatment and pretreatment samples (ΔTSR=post − pre) for each patient. CA19-9, carbohydrate antigen 19–9; CEA, carcinoembryonic antigen; CSS, cancer-specific survival; DFS, disease-free survival; LNR, lymph node ratio; MPR, major pathological response; NCRT, neoadjuvant chemoradiotherapy; NCRIT, neoadjuvant chemoradiotherapy plus immunotherapy; NMPR, non-major pathological response; pCR, pathological complete response; RC, rectal cancer; TD, tumor deposits; TSR, tumor-stroma ratio.

Table 1. The associations of TSR group with clinicopathological factors in PUMCH cohort.

Variables	Patients, No. (%)			P value
	All (n=282)	TSR-high (n=101)	TSR-low (n=181)
Sex				0.731
Female	101 (36)	38 (38)	63 (35)
Male	181 (64)	63 (62)	118 (65)
Age, mean±SD, years	64±11.17	64.16±10.67	63.91±11.47	0.856
BMI, mean±SD	24±3.14	24.07±3.13	23.96±3.15	0.779
Tumor location				0.24
Low	70 (25)	29 (29)	41 (23)
Middle	113 (40)	34 (34)	79 (44)
Upper	99 (35)	38 (38)	61 (34)
T stage				0.027
T1	14 (5)	2 (2)	12 (7)
T2	73 (26)	23 (23)	50 (28)
T3	160 (57)	56 (56)	104 (57)
T4	34 (12)	19 (19)	15 (8)
N stage				< 0.001
N0	171 (61)	46 (46)	125 (69)
N1	78 (28)	32 (32)	46 (25)
N2	33 (12)	23 (23)	10 (6)
LNR				< 0.001
High	30 (11)	21 (21)	9 (5)
Low	252 (89)	80 (79)	172 (95)
Tumor deposits				0.002
No	245 (87)	79 (78)	166 (92)
Yes	37 (13)	22 (22)	15 (8)
Tumor grade				0.999
G1-2	270 (96)	97 (96)	173 (96)
G3	12 (4)	4 (4)	8 (4)
Vascular				0.192
Invasion	91 (32)	38 (38)	53 (29)
No-invasion	191 (68)	63 (62)	128 (71)
Perineural				0.047
Invasion	25 (9)	14 (14)	11 (6)
No-invasion	257 (91)	87 (86)	170 (94)
CEA, median (IQR), ng/mL	3.31 (1.94–6.6)	4.36 (2.4–9.14)	3.12 (1.74–5.54)	0.004
CA19-9, median (IQR), U/mL	12.45 (7.6–21)	14.65 (9–27.67)	11.65 (7.1–19.8)	0.018
Tumor diameter, median (IQR), cm	4 (3–5)	4 (2.6–5)	3.8 (3–5)	0.811
Follow-up duration, median (Q1, Q3), month	71.37 (57.56, 79.19)	62.03 (41.63, 78.2)	73.4 (59.87, 79.37)	<0.001

CA19-9, carbohydrate antigen 19–9; CEA, carcinoembryonic antigen; G3, grade poor; G1-2, grade well and moderate; LNR, lymph node ratio; PUMCH, Peking Union Medical College Hospital; RC, rectal cancer; TSR, tumor stroma ratio.

High TSR is associated with decreased survival in RC

The median follow-up duration was 71.37 months (IQR 57.56–79.19) for the training cohort and 58.55 months (IQR 39.76–76.22) for the validation cohort. In both cohorts, high TSR was significantly associated with poorer DFS and CSS (figure 1C–F). Specifically, in the training cohort, the 5-year DFS (92.1% vs 70.3%, p<0.001) and 5-year CSS (95.4% vs 79.4%, p<0.001) were significantly better in the TSR-low group than in the TSR-high group. Similarly, in the validation cohort, the TSR-low group had superior 5-year DFS (85.4% vs 56.5%, p<0.001) and 5-year CSS (90.1% vs 68.6%, p<0.001) compared with the TSR-high group. Furthermore, multivariate Cox analysis in the training cohort identified TSR-high as an independent adverse prognostic factor for DFS (HR: 3.39, 95% CI 1.77 to 6.50, p<0.001) and CSS (HR: 2.93, 95% CI 1.14 to 7.53, p=0.026) (figure 1E, online supplemental tables S2 and S3). Similar results were observed in the validation cohort, where TSR-high was an independent adverse prognostic factor for DFS (HR: 2.01, 95% CI 1.11 to 3.65, p=0.022) and CSS (HR: 2.25, 95% CI 1.08 to 4.67, p=0.030) (figure 1F, online supplemental tables S4 and S5).

To further evaluate the clinical relevance of TSR in predicting treatment response, we analyzed two independent cohorts of patients with RC receiving neoadjuvant therapy. In the NCRT cohort (n=151), 64.90% (98/151) of patients were classified as TSR-low and 35.10% (53/151) as TSR-high based on pretreatment colonoscopic biopsy specimens. Patients with TSR-low tumors exhibited a significantly higher pCR rate compared with those with TSR-high tumors (22.45% (22/98) vs 7.55% (4/53), p=0.0237, Fisher’s exact test). Similarly, the MPR rate was markedly higher in the TSR-low group than in the TSR-high group (47.96% (47/98) vs 15.09% (8/53), p<0.001) (figure 1G). In addition, patients with TSR-low tumors demonstrated significantly improved DFS compared with those with TSR-high tumors (p<0.001) (figure 1H).

We next assessed the association between baseline TSR and treatment response in an independent cohort of 33 patients with RC treated with NCRIT. Among these tumors, 16 were classified as TSR-high and 17 as TSR-low. Consistent with the findings in the NCRT cohort, patients with TSR-low tumors exhibited a higher MPR rate than those with TSR-high tumors (MPR rate: 76.47% vs 37.50%, p=0.0366), whereas the pCR rate showed a non-significant trend in the same direction (47.06% vs 12.50%, p=0.0570) (figure 1I). Dynamic changes in TSR were further evaluated in paired pretreatment and post-treatment samples from the NCRIT cohort. In patients achieving MPR, TSR values decreased substantially after treatment, whereas TSR levels remained largely stable in the NMPR group (figure 1J,K).

Transcriptomic profiling reveals immune features in TSR-low tumors

Tumor stroma is a critical event in RC progression; however, the underlying molecular mechanisms remain unclear. We first performed bulk messenger RNA sequencing to compare differentially expressed genes between TSR-high and TSR-low tumors. A total of 240 genes were upregulated and 1336 genes were downregulated in TSR-low tumors. Since tumor stroma can suppress antitumor immune responses, and T cells and B cells are the main immune cells involved in these responses, we analyzed the differential gene expression in T and B cells (figure 2A). Volcano plot analysis revealed that TSR-high tumors showed higher expression of T cell-related genes, including CD3E, CD3D, CD3G, CD8A, GZMA, GZMB, and IFNG (figure 2B). However, genes associated with B cell and plasma cell biology, including CD79A, JCHAIN, MZB1, and TNFRSF17 did not show significant enrichment (figure 2B). Using XCELL, TIMER, MCP, CIBERSORT and EPIC, we evaluated the association between tumor stroma and immune cell infiltration (figure 2C–G). TSR-high tumors tended to harbor higher levels of CD8⁺ T-cell infiltration, whereas TSR-low tumors were enriched for CD4⁺ Th1 cells. TSR-high tumors displayed markedly greater infiltration of CAFs, macrophages, neutrophils, and monocytes. Similar patterns were reproduced in the PLAGH cohort (online supplemental figure 2A–F).

Figure 2. Immune landscape and stromal features in TSR-high and TSR-low rectal tumors. (A) Hierarchical cluster of the T and B-cell subset signatures. Samples are ordered by TSR-high and TSR-low (n=66). (B) Volcano plot representing differentially expressed genes between tumors with TSR-high and TSR-low. Genes from the T and B-cell subset signatures are highlighted. (C–G) Immune cell infiltration differences between TSR-high and TSR-low tumors were evaluated using computational tools, including TIMER, MCPCOUNTER, EPIC, CIBERSORT, and XCELL. P values were derived from the Mann-Whitney U test. (H) IHC staining for the αSMA, PDGFRα, CD15⁺ neutrophils and CD68⁺ macrophages for representative TSR-high and TSR-low tumors. (I) Distribution of αSMA, PDGFRα, CD15⁺ neutrophils, and CD68⁺ macrophages in TSR-high and TSR-low tumors in the tumor center and invasive margin regions. P values were calculated using Pearson’s χ² test. (J) Distribution of CMS among TSR-high and TSR-low tumors. P values were derived from the χ² test. (K) Heatmap showing the different TSR status progeny pathway scores. P values were derived from the Mann-Whitney U test. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. CMS, consensus molecular subtypes; IHC, immunohistochemistry; PUMCH, Peking Union Medical College Hospital; TSR, tumor-stroma ratio.

TSR-high tumors exhibited increased CAF infiltration, reflected by higher α-SMA and PDGFRα expression, together with increased infiltration of CD15⁺ neutrophils and CD68⁺ macrophages in both CT and IM regions (figure 2H,I). Molecular subtype analysis revealed significant differences between TSR-high and TSR-low tumors. TSR-low tumors exhibited a significantly higher proportion of CMS1, and CMS2 subtypes, while TSR-high tumors were markedly enriched in CMS4 (figure 2J). Pathway scoring using Progeny demonstrated that TSR-high tumor scored higher on the transforming growth factor (TGF)-β, and tumor necrosis factor (TNF)-α pathways (figure 2K). Similar patterns were reproduced in the PLAGH cohort (online supplemental figure 2G,H).

Single-cell transcriptomics reveals immune landscape differences between TSR-high and TSR-low tumors

To further investigate the impact of stromal content on the tumor immune microenvironment in primary RC, we performed scRNA-seq on 10 RC tumor samples, classified into TSR-high (n=4) and TSR-low (n=6) groups based on their TSR status (figure 3A). After quality control, 105,581 cells were retained (34,487 from TSR-high; 71,094 from TSR-low tumors). The Uniform Manifold Approximation and Projection (UMAP) plot shows the identified cell types, including epithelial cells, endothelial cells, fibroblasts, T/natural killer (NK) cells, B cells, plasma cells, mast cells, and myeloid cells (figure 3B). Figure 3C illustrates the distribution of cells from the 10 RC tumor samples, while figure 3D,E compares the cellular composition between the TSR-high and TSR-low groups. UMAP plots showing expression levels of highly expressed genes (including cluster-specific marker genes) in eight major cell clusters using data of patients with RC (figure 3F). Moreover, bar charts were applied to show the percentage of each cell type across different samples, and TSR status groups (figure 3G). TSR-high tumors exhibited a higher proportion of fibroblasts, and endothelial cells, but a significantly lower proportion of epithelial cells compared with TSR-low tumors (figure 3G). Notably, TSR-low tumors had significantly higher proportions of B cells and plasma cells, while TSR-high tumors had increased T/NK cells, and myeloid cells (figure 3G). These results revealed the complex and heterogeneous cell components in TME of both TSR-high and TSR-low RC samples.

Figure 3. scRNA-seq profiling of the landscape of RC. (A) Overview of the experimental design. UMAP plot of 105,581 cells from 10 RC primary tumor samples, colored by cell type (B), patient ID (C) and TSR status (D–E). (F) The accompanying heatmap highlights the expression of characteristic genes within these cell types. (G) Characterization of the proportion of cell types identified in each sample in TSR-high versus TSR-low RC tissue. RC, rectal cancer; scRNA-seq, single-cell RNA sequencing; TSR, tumor-stroma ratio; UMAP, Uniform Manifold Approximation and Projection.

TSR-high tumor exhibits a highly elevated infiltration level of CD8⁺ Tex cells along with an increased cell proportion of CD4⁺ Tregs

Further unsupervised analysis of preprocessed and normalized scRNA-seq data for T/NK cells via UMAP revealed 16 T/NK cell clusters, including CD4⁺ Treg-TNFRSF4, CD4⁺ Treg-FOXP3, CD4⁺ Treg-IL1R1, CD4⁺ Tfh-CXCL13, CD4⁺ Th17-RORA, CD4+Tn-IL7R, CD4⁺ Tn-KLF2, CD4⁺ Tn-CCR7 and CD4⁺ Tem-GZMA, CD8⁺ Tex-CXCL13, CD8⁺ Tex-STMN1, CD8⁺ Trm-ITGA1, CD8⁺ Tem-CCL4, CD8⁺ MAIT-KLRB1, Cycling T-STMN1, and NK_GNLY (figure 4A). To identify each cluster, we constructed a dot plot of key marker genes for these various T-cell clusters (online supplemental figure 3A). The distributions of these T-cell clusters across patients and TSR status were visualized through UMAP (figure 4B,C).

Figure 4. T-cell heterogeneity of distinct TSR status groups. (A) Clustering of T/NK cells based on all enrolled samples (n=10), colored and labeled by T-cell clusters. (B) UMAP plot showing T/NK cells from all the samples annotated by patient origin. (C) UMAP visualization of T/NK cells colored by TSR-high and TSR-low tumors. (D) Fraction of CD4⁺ T cells in T/NK cells and fraction of CD4⁺ Treg-TNFRSF4 in T/NK cells. *p<0.05 of Wilcoxon test. Boxes indicate median±IQR range; whiskers show minima and maxima. (E) Fraction of CD8+T cells in T/NK cells and fraction of CD8⁺ Tex-CXCL13 cells in T/NK cells. *p<0.05 of Wilcoxon test. Boxes indicate median±IQR range; whiskers show minima and maxima. (F) Expression of marker genes for activation of immunosuppressive CD4+Tregs in CD4⁺ Treg-TNFRSF4. P values were calculated using the two-sided Wilcoxon rank-sum test. ns, not significant. Boxes indicate median±IQR range. (G) The difference in progenitor and terminal exhaustion score of CD8⁺ Tex-CXCL13 between TSR-high and TSR-low. *p<0.05 of Mann-Whitney U test. Boxes indicate median±IQR range. (H) The difference in progenitor and terminal exhaustion score among TSR-high and TSR-low from an independent RC cohort. *p<0.05 of Mann-Whitney U test. Boxes indicate median±IQR range. (I) The difference in infiltration score of CD4⁺ Treg-TNFRSF4, and CD8⁺ Tex-CXCL13 among TSR-high and TSR-low from an independent RC cohort. ***p<0.001 of Mann-Whitney U test. Boxes indicate median±IQR range. (J) Representative images of mIF staining for CXCL13⁺CD8⁺ T cells from TSR-high and TSR-low in the independent validation cohort. TSR-high, n=10; TSR-low, n=10. Scale bar: 20 µm. (K) Fraction of CD8⁺ CXCL13⁺ T cells in CD8⁺ T cells. **p<0.01 of Mann-Whitney U test. TSR-high, n=10; TSR-low, n=10. Boxes indicate median±IQR range; whiskers show minima and maxima. mIF, multiplex immunofluorescence; NK, natural killer; RC, rectal cancer; Treg, regulatory T cell; TSR, tumor-stroma ratio; UMAP, Uniform Manifold Approximation and Projection.

We next explored the differences in cellular proportions in all T-cell types between TSR-high and TSR-low tissues. We observed that the fraction of CD4⁺ T cells was higher in TSR-low patients, whereas the fraction of CD8⁺ T cells was higher in TSR-high patients (figure 4D,E). For CD4⁺ T cells, we identified a highly increased cell proportion of CD4⁺ Treg-TNFRSF4 cells in TSR-high (figure 4D). In addition, CD4⁺ Treg-TNFRSF4 cells in TSR-high tumors showed directionally increased but not statistically significant expression trends of canonical regulatory T cell (Treg) activation-associated markers, including FOXP3, TNFRSF4, TNFRSF18, and CD40LG, compared with TSR-low tumors (figure 4F). We observed that the fraction of CD8⁺ Tex-CXCL13 cells was increased in TSR-high patients, although this difference did not reach statistical significance, likely due to the limited number of scRNA-seq samples (figure 4E). Next, we calculated the progenitor and terminally exhausted signature of CD8⁺ Tex cells. Our results showed that CD8⁺ Tex-CXCL13 cells in TSR-high had prominently higher terminal exhaustion score but lower progenitor exhaustion score versus TSR-low (figure 4G). Based on RNA-seq data from 66 cases in an independent retrospective RC cohort, single-sample gene set enrichment analysis (ssGSEA) analysis revealed that TSR-high tumor exhibited a markedly higher terminal exhaustion score but a lower progenitor exhaustion score compared with TSR-low tumor (figure 4H). In addition, we validated increased infiltration of CD4⁺ Treg-TNFRSF4 and CD8⁺ Tex-CXCL13 populations in TSR-high tumor (figure 4I). Multiplex immunofluorescence of CD8, and CXCL13 was conducted in the samples from 20 patients with RC (figure 4J). We observed significantly higher CXCL13/CD8 dual-positive cell abundance in TSR-high versus TSR-low (figure 4K).

TSR-high tumors are characterized by increased clonal expansion of CD4⁺ Tregs and CD8⁺ Tex cells

To delineate the relationship between the characteristics of T cells and the TSR status, we next investigated changes in clonality through scTCR-seq analysis. We recovered the α chains and β chains of TCR from RC samples at single-cell resolution. T cells with TCR data across T cells, TSR status, and clone size were shown in the UMAP plots (figure 5A–C). We examined the clonal expansion patterns of different T-cell subtypes in RC, in which we observed higher percentages of expanded clonotypes (n ≥3 cells) were mainly in CD8⁺ Tex cells (figure 5D,E, online supplemental figure 3B). Importantly, TSR-high tumors showed a higher proportion of cells belonging to expanded clonotypes within the CD4⁺ Treg-TNFRSF4 and CD8⁺ Tex populations (figure 5D,E). These findings suggest preferential enrichment of clonally expanded immunosuppressive Treg and exhaustion-associated CD8⁺ T-cell states in TSR-high tumors, rather than expansion of selected effector-memory-like CD8⁺ T-cell states.^{16 17} Consistently, complementary analysis using scRepertoire-derived clonal frequency categories showed a similar pattern, with large and hyperexpanded clonotypes enriched in exhausted CD8⁺ T-cell states (online supplemental figure 3C). Further, to trace the lineage transitions between T-cell phenotypes, we measured the fraction of clonotypes shared with a secondary phenotype in each primary phenotype. We observed significant clonotype overlaps between CD8⁺ T-cell clusters, and the phenotype transitions were more active in TSR-high (figure 5F), especially the transition between CD8⁺ Tem-CCL4 or CD8⁺ Tex-CXCL13 and CD8⁺ Tex-STMN1. Moreover, CD8⁺ Tex-CXCL13 shared clones with several CD8⁺ T-cell subtypes, especially CD8⁺ Tem (online supplemental figure 3D,E), suggesting the existence of lineage transition between CD8⁺ Tem and CD8⁺ Tex.

Figure 5. Clonality features and dynamics of T-cell receptors in TSR-high and TSR-low rectal tumors. (A–C) UMAP plots of T/NK cells with detected TRB clones from 10 samples, annotated by patient origin, TSR status, and clone size. (D–E) Fractions of TRB clones with different sizes (n<3 and n ≥3) for different T-cell clusters in TSR-high and TSR-low. (F) Heatmap showing the fraction of CD8⁺ T-cell clonotypes shared with a secondary phenotype (column) in each primary phenotype (row). All (left); TSR-high (middle); TSR-low (right). (G) UMAP plot of CD8⁺ Tex-relevant cells in all samples (n = 10). (H) UMAP plot showing CD8+ terminal Tex and progenitor Tex (Texp) cells in all samples (n=10). (I) Pseudotime trajectory of CD8⁺ Tex cells clusters inferred by Monocle 3. Cells are arranged along the trajectory curves, with numbers indicating pseudotime states or progression along the differentiation path. (J) Fraction of CD8⁺ terminal Tex and Texp cells in CD8+Tex-relevant cells for TSR-high and TSR-low. NK, natural killer; Texp, precursor Tex; TRB, T-cell receptor beta; TSR, tumor-stroma ratio; UMAP, Uniform Manifold Approximation and Projection.

We selected all CD8⁺ T cells that shared TCRs with CD8⁺ Tex-CXCL13 (figure 5G) and referred to these T cells as Tex-relevant cells. Among the Tex-relevant cells, those not included in CD8⁺ Tex-CXCL13 cells (terminal Tex) were named as precursor exhausted T cells (Texp) (figure 5H). Texp exhibits a prominently higher expression level of marker genes enhancing antitumor activity of CD8⁺ T cells (online supplemental figure 3F), like IL7R (supporting CD8⁺ T-cell survival and maintenance) and GZMK (marker of cytolytic T-cell activity). In contrast, terminal Tex was mainly marked by increased expression of checkpoint molecules, including LAG3, HAVCR2, CTLA4, PDCD1, ITGAE, and CXCL13 (online supplemental figure 3F). ENTPD1 (CD39), a marker previously associated with tumor-reactive and antigen-experienced CD8⁺ TILs, was also highly expressed in terminal Tex cells.¹⁸ Together with elevated inhibitory receptor expression, these findings support the annotation of this subset as an antigen-experienced, exhaustion-associated CD8⁺ T-cell state rather than a fully functional effector population.¹⁷ Gene Ontology (GO) enrichment analysis revealed that Texp cells had the highest scores for positive regulation of cytokine production involved in immune response, positive regulation of T-cell cytokine production and positive regulation of leukocyte activation (online supplemental figure 3G). In contrast, terminal Tex cells showed the highest scores for negative regulation of T-cell activation, negative regulation of lymphocyte activation, and negative regulation of cytokine production (online supplemental figure 3H). By ordering cells in pseudotime, we observed a clear developmental trajectory originating from the Texp population and progressing toward the terminally exhausted (terminal Tex) state (figure 5I). Texp cells were predominantly positioned at the early pseudotime, whereas terminal Tex cells accumulated at the late pseudotime, indicating a potential transitional route from Texp to terminal Tex and supporting the notion that Texp may serve as progenitors that give rise to terminally exhausted CD8⁺ T cells. Our results showed that TSR-high tumors contained a higher proportion of terminal Tex cells but a lower proportion of Texp cells compared with TSR-low tumors (figure 5J).

TSR-associated fibroblast subtypes and their functional characteristics

Accumulating evidences has shown that CAFs play a key part in TME remodeling. We investigated the distribution of CAFs underlying distinct TSR statuses on the basis of scRNA-seq data. Using previously reported marker genes, CAFs were annotated to 12 cell clusters (figure 6A). To identify each cluster, we constructed a dot plot of key marker genes for these various CAF clusters (online supplemental figure 4A). The distribution of various CAF clusters across different patients and TSR status were studied and displayed (figure 6B,C). The mCAF-CTHRC1, and mCAF-FAP were found to be highly enriched in the TSR-high group (figure 6D). Additionally, our CellChat analysis revealed an extensive and complex cell–cell communication network among various CAFs and diverse T-cell clusters in RC samples of both TSR-high and TSR-low (online supplemental figure 4B,C), suggesting a potential multiaspect role of CAFs. CAFs have been reported to be closely associated with immunosuppressive Treg-enriched microenvironments. Our next focal point was the relationship between these two clusters. Notably, whereas mCAF-CTHRC1 cell proportions were significantly positively correlated with Tregs cell proportions, mCAF-FAP showed no correlation (figure 6E,F). Further ssGSEA analysis revealed consistent associations between mCAF-CTHRC1 and CD4⁺ Treg–TNFRSF4, whereas no such association was observed between mCAF-FAP and CD4⁺ Treg–TNFRSF4 (figure 6G,H). CellChat analysis showed that the number of ligand-receptor pairs between mCAF-CTHRC1 and CD4⁺ Treg-TNFRSF4 was higher in the TSR-high group compared with TSR-low (online supplemental figure 4D), whereas interactions between mCAF-FAP and CD4⁺ Treg-TNFRSF4 were comparable between the two TSR groups (online supplemental figure 4E). Focusing on ligand-receptor pairs, CellChat analysis in TSR-high tumors identified LGALS9 derived from mCAF-CTHRC1 as a predicted ligand potentially interacting with CD44/CD45 on CD4⁺ Treg-TNFRSF4 cells, suggesting a possible role in the immunosuppressive Treg-associated niche (online supplemental figure 4F,G). Importantly, LGALS9 showed a preferential expression pattern in mCAF-CTHRC1 cells (figure 6I,J). LGALS9 has been reported to induce apoptosis in T cells by binding to HAVCR2 and bolster the stability and functionality of immunosuppressive Treg cells via interaction with CD44.^{19 20}

Figure 6. CAF cell heterogeneity of distinct TSR status groups. (A) Clustering of CAF cells based on all enrolled samples (n=10), colored and labeled by cell clusters. (B) UMAP plot showing CAF cells from all the samples annotated by patient origin. (C) UMAP visualization of CAF cells colored by TSR-high and TSR-low tumors. (D) Fraction of mCAF-CTHRC1 cells and mCAF-FAP cells in CAF. P values were derived from the Wilcoxon test. *p<0.05, **p<0.01. Boxes indicate median±IQR range. (E–F) Scatterplot showing a significant correlation between cell proportion of CD4⁺ Treg-TNFRSF4 in T/NK cells with that of mCAF-CTHRC1 (E) and mCAF-FAP (F). Spearman-correlation test. (G–H) Scatterplot of correlation between the infiltration level of CD4⁺ Treg-TNFRSF4 with that of mCAF-CTHRC1 (G) and mCAF-FAP (H). Spearman-correlation test. (I) UMAP plot showing the distribution of LGALS9 in CAFs. (J) The expression level of LGALS9 in various CAF cell clusters. (K) UMAP plot showing the distribution of LRRC15 in CAFs. (L) The expression level of LRRC15 in various CAF cell clusters. (M–N) Heatmap showing the progeny pathway scores of mCAF-FAP and mCAF-CTHRC1 in TSR-high and TSR-low tumor. P values were derived from the Mann-Whitney U test. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (O) Representative images of mIF staining for LRRC15 ⁺α-SMA⁺ CAF cells from TSR-high and TSR-low in the independent validation cohort. TSR-high, n=10; TSR-low, n=10. Scale bar: 20 µm. (P) Fraction of LRRC15 ⁺α-SMA⁺ CAF cells in α-SMA⁺ CAF cells. ***p<0.001 of Mann-Whitney U test. TSR-high, n=10; TSR-low, n=10. Boxes indicate median±IQR range; whiskers show minima and maxima. CAF, cancer-associated fibroblast; mIF, multiplex immunofluorescence; Treg, regulatory T cell; TSR, tumor-stroma ratio; UMAP, Uniform Manifold Approximation and Projection.

Importantly, we also found the molecule LRRC15 showed preferential expression in mCAF-CTHRC1 and mCAF-FAP cells, supporting their LRRC15⁺ CAF-like features (figure 6K,L). Previous studies have shown that activation of the TGF-β pathway can induce LRRC15 expression in CAFs and consequently promote resistance to immunotherapy.²¹ Consistent with this, pathway scoring using Progeny revealed that mCAF-FAP cells in TSR-high tumors exhibited higher TGF-β pathway activity (figure 6M), and a similar increase was observed in mCAF-CTHRC1 cells from TSR-high tumors (figure 6N).

Through performing mIF assay on markers of mCAFs (α-SMA⁺ LRRC15⁺ cells), we further validated the enrichment of LRRC15⁺ CAFs in TSR-high tumors (figure 6O,P). Collectively, our results indicated that high infiltration of LRRC15⁺ CAFs, particularly the mCAF-CTHRC1, is associated with the activation-associated features and enrichment of CD4⁺ Treg-TNFRSF4 cells, suggesting a potential association with impaired antitumor immune responses in RC.

SPP1-high monocytes are enriched in TSR-high rectal cancer

Myeloid cells, representing a key component of TME, have been reported to promote tumor immune evasion via targeting receptors, transcription factors, or secreted proteins involved in downstream/upstream signaling pathways. 12 subclusters of myeloid cells were identified through scRNA-seq data, namely Macro-APOC1, Macro-FMNL2, Macro-STMN1, Mon-EREG, Mon-SPP1, Mon-TIMP1, NE-IFIT3, NE-NEDD9, pDC-GZMB, DC1-NAPSB, DC2-PPA1, and DC3-CCR7 (figure 7A). The distribution of various myeloid clusters across different patients and TSR status (figure 7B,C). To identify each cluster, we constructed a dot plot of key marker genes for these various myeloid clusters (online supplemental figure 5A). Of note, Mon-SPP1 was observed with prominent enrichment in TSR-high compared with TSR-low in RC (figure 7D,E). Accumulating evidences have proved that SPP1 could promote tumor progression, mediate TME remodeling and limit response to programmed death-ligand 1 (PD-L1) blockade.^{22 23} Consistently, upregulated differentially expressed genes (DEGs) of Mon-SPP1 showed a significant enrichment in signaling pathways associated with TME remodeling including cytokine/chemokine mediated signaling pathways and leukocyte migration (figure 7F). Notably, Mon-SPP1 cell proportions were significantly positively correlated with Tregs cell proportions (figure 7G). Cell–cell interaction analysis revealed a diverse communication network between Mon-SPP1 and various T-cell clusters in both TSR-high and TSR-low (figure 7H,I).

Figure 7. Myeloid cell heterogeneity of distinct TSR status groups. (A) Clustering of myeloid cells based on all enrolled samples (n=10), colored and labeled by cell clusters. (B) UMAP plot showing myeloid cells from all the samples annotated by patient origin. (C) UMAP visualization of myeloid cells colored by TSR-high and TSR-low tumors. (D) Fraction of Mon-SPP1 cells in myeloid cells. *p<0.05 of Wilcoxon test. Boxes indicate median±IQR range. (E) UMAP plot showing the distribution of SPP1 in myeloid cells. (F) Gene Ontology enrichment analysis of differentially expressed genes in Mon-SPP1 cells. (G) Scatterplot showing a significant correlation between cell proportion of CD4+Treg-TNFRSF4 cells in T/NK cells and Mon-SPP1 cells in myeloid cells. Spearman-correlation test. (H–I) Number of significant ligand-receptor pairs calculated via CellChat analysis from Mon-SPP1 cells to T-cell subsets in TSR-high (left) and TSR-low (right). (J) Differential expression of SPP1 in Mon-SPP1. **p<0.01 of two-sided Wilcoxon test. (K) The difference in infiltration score of Mon-SPP1 among TSR-high and TSR-low in an independent RC cohort. ***p<0.001 of Mann-Whitney U test. Boxes indicate median±IQR range. (L) Scatterplots showing correlation between the infiltration score of CD4+Treg-TNFRSF4 with that of Mon-SPP1 in the independent RC cohort. (M–N) DFS and OS analysis according to SPP1 expression in TCGA-READ cohort (n=91). Cox regression HRs and 95% CIs are shown with log-rank p values. (O) Representative images of mIF staining for SPP1⁺ CD14⁺ monocyte cells from TSR-high and TSR-low in the independent validation cohort. TSR-high, n=10; TSR-low, n=10. Scale bar: 20 µm. (P) Fraction of SPP1 ⁺ CD14⁺ monocyte cells among CD14⁺ monocyte cells. **p<0.01 of Mann-Whitney U test. TSR-high, n=10; TSR-low, n=10. Boxes indicate median±IQR range; whiskers show minima and maxima. DFS, disease‐free survival; mIF, multiplex immunofluorescence; NK, natural killer; OS, overall survival; RC, rectal cancer; Treg, regulatory T cell; TSR, tumor-stroma ratio; UMAP, Uniform Manifold Approximation and Projection.

CellChat analysis indicated that Mon-SPP1 communicates with T cells via the MIF-(CD74+CD44/CXCR4) pathway in the TSR-low group and TSR-high group (online supplemental figure 5B,C). In addition, the ligand-receptor pair SPP1-CD44, which was associated with impairment of cytotoxic T lymphocyte, was observed as a specific communication pathway from Mon-SPP1 to various T-cell clusters for TSR-high tumor (online supplemental figure 5B,C). Consistent with these findings, SPP1 expression in Mon-SPP1 cells was markedly upregulated in the TSR-high group compared with the TSR-low group (figure 7J). Through comparing the ssGSEA score of Mon-SPP1 between two groups, a similar difference was observed, with the infiltration level of Mon-SPP1 in TSR-low decreasing prominently and remaining significantly lower compared with TSR-high (figure 7K). We also found that the Mon-SPP1 score seemed to be positively associated with CD4⁺ Treg-TNFRSF4 score in RC samples (figure 7L). In the The Cancer Genome Atlas Rectum Adenocarcinoma (TCGA-READ) cohort, high expression level of SPP1 emerged as a significant prognostic risk factor for DFS (HR: 3.92, 95% CI 1.60 to 9.59, p=0.00133) and overall survival (OS; HR: 3.61, 95% CI 1.32 to 9.86, p=0.00765) (figure 7M,N). Through performing mIF assay on markers of SPP1+CD14+ monocyte cells, we further validated the enrichment of SPP1⁺ CD14⁺ monocyte cells in TSR-high tumors (figure 7O,P). These findings suggest that Mon-SPP1 is associated with an immunosuppressive TME in TSR-high RC and may participate in predicted SPP1–CD44-mediated stromal-immune communication.

SPP1-high malignant epithelial cells are enriched in TSR-high rectal cancer

We next performed UMAP-based clustering of epithelial cells and identified 15 epithelial transcriptional states, reflecting substantial epithelial-cell heterogeneity (figure 8A). To assess whether these epithelial states were driven predominantly by individual samples, epithelial cells were additionally visualized according to sample of origin (figure 8B). CopyKAT analysis was then performed to infer large-scale copy-number alterations and distinguish putative malignant epithelial cells from diploid epithelial cells (figure 8C). The CopyKAT-inferred aneuploid epithelial cells were interpreted as putative malignant epithelial cells, whereas diploid epithelial cells were considered non-malignant or less aberrant epithelial populations.

Figure 8. Differences in malignant epithelial cells between distinct TSR status groups. (A) Clustering of epithelial cells based on all enrolled samples (n=10), colored and labeled by cell clusters. (B) UMAP plot showing epithelial cells from all samples annotated by patient origin. (C) CopyKAT-inferred classification of epithelial cells as aneuploid, diploid, or not defined. (D) Number of significant ligand-receptor pairs calculated via CellChat analysis from malignant epithelial cells to CD4⁺ T cell subsets (left) and CD8⁺ T cells (right) in TSR-high. (E) Number of significant ligand-receptor pairs calculated via CellChat analysis from malignant epithelial cells to CD4⁺ T cell subsets (left) and CD8⁺ T cells (right) in TSR-low. (F–G) Bubble plots showing ligand–receptor interactions between malignant epithelial cells and T-cell subsets in TSR-high (F) and TSR-low (G) tumors. TSR, tumor-stroma ratio; UMAP, Uniform Manifold Approximation and Projection.

To further characterize malignant cell behavior under different stromal conditions, malignant cells were stratified into TSR-high and TSR-low groups. Consistent with these pathway features, SPP1 expression was markedly elevated in malignant cells from TSR-high tumors compared with those from TSR-low tumors (online supplemental figure 6A). Pathway scoring using Progeny demonstrated that malignant cells in TSR-high tumor scored higher on the hypoxia, JAK-STAT and VEGF signaling (online supplemental figure 6B), suggesting enhanced stress-response, inflammatory, and angiogenesis-related programs rather than directly proving a more aggressive phenotype.

Cell–cell communication analysis revealed that RC tumor cells from both TSR-high and TSR-low tumors showed stronger predicted interaction potential with CD8⁺ T cells than with CD4⁺ T cells (figure 8D,E). These interactions were largely mediated through the MIF signaling axis, particularly the MIF–CD74+CXCR4 and MIF–CD74+CD44 ligand-receptor pairs in both TSR-high and TSR-low tumors (figure 8F,G).

Discussion

In this study, we systematically investigated the impact of tumor stroma on the TME and its clinical implications in RC. Our findings demonstrate that a high TSR is not only a robust prognostic indicator associated with adverse clinicopathological features and decreased survival but also intricately shapes an immunosuppressive TME that facilitates tumor progression and immune evasion.

The tumor stroma is a critical component of the TME, characterized by its dynamic nature, high heterogeneity, and tumor type-specific composition.³ It primarily consists of CAFs, endothelial cells, and extracellular matrix components.³ The interactions between tumor cells and the surrounding stroma play a pivotal role in tumor initiation, progression, metastasis, and therapeutic resistance.^{3 4} In our multicenter cohort study, we found that TSR-high was significantly associated with poorer DFS and CSS in patients with RC. Moreover, TSR-high was identified as an independent adverse prognostic factor in multivariate Cox regression analysis. These findings are consistent with prior studies.^{6 9 24} Our study also revealed significant associations between high TSR and several pathological features, including advanced tumor stage, lymph node metastasis, perineural invasion, and elevated CEA levels. These data suggest that the tumor stroma may play a crucial role in promoting tumor invasion and metastasis.

Transcriptomic and single-cell analyses revealed profound immunological differences between TSR-high and TSR-low tumors. Although TSR-high tumors are often described as “immune-excluded” tumors characterized by impaired T-cell penetration into tumor nests,^25–27 our data suggest that TSR-high RCs exhibit a more complex stromal-immune phenotype rather than a simple absence of immune infiltration. TSR-high tumors exhibited increased CD8⁺ T-cell infiltration but diminished CD4⁺ T-cell infiltration. At first glance, this observation appears to contrast with the classical view that CD8⁺ cytotoxic T cells mediate antitumor immunity and are generally associated with favorable outcomes in CRC.²⁸ However, the prognostic significance of CD8⁺ T-cell infiltration is highly context-dependent and is determined not only by cell abundance but also by functional state, clonal expansion, spatial localization, and the surrounding stromal-immune microenvironment.²⁹ Single-cell profiling resolved this apparent paradox by showing an accumulation of terminally exhausted CD8⁺ Tex-CXCL13 cells in TSR-high tumors. CD8⁺ Tex cells arise from persistent antigen stimulation and display impaired effector function, high expression of inhibitory receptors, and diminished recall capacity.³⁰ Although progenitor-like Tex cells may respond to PD-1/PD-L1 blockade,³¹ terminal Tex cells—marked by PD-1, LAG-3, CTLA-4, TIM-3, and CXCL13—are largely refractory to reinvigoration.¹⁷ Importantly, recent studies suggest that CXCL13-associated exhausted CD8⁺ T-cell states are highly context dependent and may also reflect tumor-reactive, inflamed, or treatment-responsive immune niches in certain tumor types.^{32 33} For example, CXCL13⁺ CD8⁺ Tex cells have been associated with improved response to neoadjuvant immunochemotherapy in esophageal squamous cell carcinoma and with favorable immune contexture and survival in CRC.^{32 33}

In our study, however, TSR-high tumors were additionally characterized by enrichment of terminal exhaustion signatures, clonally expanded CD8⁺ Tex-CXCL13 cells, activated CD4⁺ Treg-TNFRSF4 cells, immunosuppressive CAF programs, and poor therapeutic response. These findings suggest that the CXCL13⁺ exhausted CD8⁺ T-cell state in TSR-high RC predominantly reflects a chronically activated yet dysfunction-dominant stromal-immune niche. Therefore, increased CD8⁺ T-cell infiltration in TSR-high tumors may reflect an activated but functionally constrained immune response within a stromal-rich immunosuppressive niche, rather than a fully effective cytotoxic antitumor response.

TSR-high tumors demonstrated a significantly elevated terminal exhaustion signature and decreased progenitor exhaustion signature, suggesting a skewed differentiation trajectory toward more terminal exhaustion-associated states. Our scTCR-seq analysis further revealed enrichment of expanded clonotypes within CD8⁺ Tex populations in TSR-high tumors, suggesting that antigen-experienced CD8⁺ T-cell clones may preferentially accumulate in exhaustion-associated states. In addition, TCR clonotype-sharing analysis showed that CD8⁺ Tex-CXCL13 and CD8⁺ Tex-STMN1 cells shared clonotypes with several effector-memory-like CD8⁺ T-cell subsets, supporting potential clonal relationships across the effector-memory-to-exhaustion spectrum. These findings indicate that the TSR-high tumors may not simply prevent CD8⁺ T-cell infiltration, but instead may favor the accumulation of antigen-experienced CD8⁺ T cells that have progressed toward exhaustion-associated transcriptional states.

At the transcriptomic level, terminal Tex cells in TSR-high tumors exhibited enrichment of programs associated with suppressed T-cell activation, impaired lymphocyte activation, and reduced immune cell adhesion. Rather than directly proving loss of cytotoxic function, these enrichment patterns are consistent with a functionally constrained local immune response. This interpretation is supported by previous studies showing that CD8⁺ T-cell exhaustion develops under persistent antigen stimulation and is characterized by increased inhibitory receptor expression, altered transcriptional states, and reduced effector potential.^{30 34}

Importantly, exhausted CD8⁺ T cells are heterogeneous. Progenitor exhausted CD8⁺ T cells can maintain proliferative potential and respond to PD-1 blockade, whereas terminally exhausted CD8⁺ T cells show reduced reinvigoration capacity and weaker tumor control.¹⁷ Accordingly, the enrichment of terminal Tex states together with reduced progenitor exhaustion signatures in TSR-high tumors may contribute to a stromal-associated immune context that is less responsive to checkpoint blockade, although this remains an inference that requires direct functional validation. Correspondingly, both scRNA-seq and bulk RNA-seq demonstrated significantly higher terminal exhaustion scores but diminished progenitor exhaustion signatures among TSR-high tumors. Given that progenitor exhausted T cells are more responsive to immune checkpoint blockade, the depletion of this population may partially explain the poor immunotherapy response observed in stromal-rich tumors and the overall treatment resistance associated with the CMS4 phenotype. Importantly, these observations suggest that TSR-high tumors are not simply immune-cold, but rather represent an immune-infiltrated yet functionally suppressed microenvironment.

Beyond CD8⁺ T-cell dysfunction, TSR-high tumors displayed a marked increase in CD4⁺ Treg-TNFRSF4 cells, a population strongly associated with tumor progression and poor prognosis. TSR-high Tregs showed directionally increased expression patterns of immunosuppressive markers. In addition, TSR-high tumors exhibited an increased abundance of expanded clonotypes within CD4⁺ Treg-TNFRSF4 populations, suggesting preferential enrichment of clonally expanded immunosuppressive Treg states. Together, these features support a profoundly immunosuppressive TME associated with both quantitative expansion and activation-associated features of Tregs, underscoring the therapeutic potential of combining Treg-targeted strategies with immune checkpoint blockade in RC.

CAFs represent a central stromal component influencing immune suppression.¹³ Consistently, we observed a marked infiltration of CAFs in TSR-high tumors. Among CAF subsets, mCAF-CTHRC1 was significantly elevated in TSR-high tumors and showed a positive correlation with the proportion of CD4⁺ Treg-TNFRSF4 cells, whereas mCAF-FAP, although also increased in TSR-high tumors, did not exhibit a significant association with CD4⁺ Treg-TNFRSF4 abundance. These findings suggest that specific CAF subsets may be differentially associated with Treg-enriched microenvironments, with mCAF-CTHRC1—but not mCAF-FAP—showing a stronger association with CD4⁺ Treg-TNFRSF4 abundance and a more immunosuppressive stromal-immune niche. This interpretation is consistent with previous evidence that CAFs can interact with immune cells, regulate T-cell infiltration and function, and shape an immunosuppressive tumor immune microenvironment.^{35 36} More broadly, TGF-β-associated LRRC15⁺ fibroblast programs have been implicated in immune suppression and resistance to immune checkpoint blockade across multiple tumor types.^{21 37 38} In CRC, previous single-cell and spatial transcriptomic analyses have further identified coordinated stromal-myeloid programs involving FAP⁺ fibroblasts and SPP1⁺ macrophages that are associated with immune exclusion and poor prognosis.³⁹ Therefore, our findings extend this stromal-immune suppressive paradigm into RC and further link these stromal programs to TSR-defined stromal richness.

Moreover, compared with other CAF subsets, both mCAF-CTHRC1 and mCAF-FAP were characterized by higher expression of LRRC15, which is predominantly expressed by fibroblasts. LRRC15⁺ CAFs have been reported to markedly suppress antitumor immunity and limit responsiveness to anti-PD-L1 therapy.^{21 37 38} Consistently, mIF validation showed enrichment of LRRC15⁺ CAFs in TSR-high tumors, supporting the association between LRRC15⁺ CAF programs and stromal-rich immunosuppressive microenvironments. Notably, mCAF-CTHRC1 showed preferentially predicted interactions with Tregs and other T-cell subsets via the LGALS9–CD44 ligand-receptor axis, which was more evident in TSR-high than in TSR-low tumors, suggesting a potentially specialized immunoregulatory role of this subset. Furthermore, previous studies have shown that TGF-β signaling can induce LRRC15 expression in CAFs,^{21 38} and our results indicate that both mCAF-CTHRC1 and mCAF-FAP in TSR-high tumors are associated with TGF-β pathway activation. Collectively, these findings suggest that selective targeting of LRRC15⁺ CAFs may represent a candidate therapeutic strategy that warrants further preclinical validation.

Of additional interest, our myeloid analysis identified SPP1⁺ monocytes as a hallmark of TSR-high tumors, characterized not only by increased abundance but also by substantially elevated SPP1 expression. Given the established role of SPP1 in promoting tumor progression and predicting unfavorable outcomes,^{23 40} this subset represents a key immunosuppressive component of the TSR-high microenvironment. Consistent with this, SPP1⁺ monocytes correlated positively with CD4⁺ Treg-TNFRSF4 levels. CellChat analysis further showed that the SPP1–CD44 signaling axis was active in both TSR-low and TSR-high tumors. However, its interaction strength was markedly higher in TSR-high tumors, suggesting a stronger predicted association between SPP1⁺ monocytes and impaired T-cell activation in a stroma-rich microenvironment. In addition to myeloid cells, our epithelial analysis revealed that malignant cells in TSR-high tumors also expressed markedly higher levels of SPP1, accompanied by enhanced activation of oncogenic pathways. Moreover, SPP1 secreted by tumor cells has been recognized as a potent chemokine capable of recruiting immunosuppressive myeloid populations.⁴¹ Together, these findings support the notion that SPP1-associated stromal-myeloid interactions may contribute to immune suppression in TSR-high RC, consistent with previously reported fibroblast-myeloid suppressive programs in CRC.³⁹ These observations further suggest that SPP1-related pathways may represent candidate biomarkers or therapeutic targets warranting future validation in stromal-rich rectal tumors.

Our study has several limitations. First, the scRNA-seq sample size, though comprehensive, remains modest and may underestimate certain rare cell populations. Second, the retrospective design and potential institutional heterogeneity in patient selection, treatment practice, pathological assessment, and follow-up may introduce residual bias. Third, the NCRIT cohort was relatively small, and although TSR was assessed on pretreatment biopsy specimens, the use of biopsy-based TSR as a treatment-stratification biomarker requires prospective validation in larger immunotherapy-treated cohorts. Fourth, while we identified stromal-immune interactions associated with immune exhaustion, these findings were largely inferred from transcriptomic correlation and ligand-receptor modeling, including the predicted LGALS9–CD44 and SPP1–CD44 signaling axes. Therefore, mechanistic validation using functional assays, spatially resolved approaches, and preclinical models is required. Finally, as all patients included in this study were pMMR and tumor mutational burden (TMB) was not uniformly available, future studies integrating TSR with genomic biomarkers may further refine treatment stratification in pMMR RC.

In conclusion, our integrative multi-omics approach reveals that TSR-high tumor is associated with a distinctive immune-evasion phenotype, characterized by terminal T-cell exhaustion, expanded immunosuppressive Treg and CAF subsets, and SPP1-mediated myeloid-lymphoid interactions. These insights provide a potential biological explanation for the poor prognosis of stroma-rich RC and highlight candidate stromal-immune pathways, including LGALS9-related, LRRC15-related, and SPP1-related programs, that warrant further preclinical validation.

The funders had no role in the study design; in the collection, analysis, or interpretation of data; in the writing of the manuscript; or in the decision to submit the article for publication.

Data availability statement

Data are available upon reasonable request.