An immunometabolism subtyping system identifies S100A9⁺ macrophage as an immune therapeutic target in colorectal cancer based on multiomics analysis

Summary

Immunometabolism in the tumor microenvironment (TME) and its influence on the immunotherapy response remain uncertain in colorectal cancer (CRC). We perform immunometabolism subtyping (IMS) on CRC patients in the training and validation cohorts. Three IMS subtypes of CRC, namely, C1, C2, and C3, are identified with distinct immune phenotypes and metabolic properties. The C3 subtype exhibits the poorest prognosis in both the training cohort and the in-house validation cohort. The single-cell transcriptome reveals that a S100A9⁺ macrophage population contributes to the immunosuppressive TME in C3. The dysfunctional immunotherapy response in the C3 subtype can be reversed by combination treatment with PD-1 blockade and an S100A9 inhibitor tasquinimod. Taken together, we develop an IMS system and identify an immune tolerant C3 subtype that exhibits the poorest prognosis. A multiomics-guided combination strategy by PD-1 blockade and tasquinimod improves responses to immunotherapy by depleting S100A9⁺ macrophages in vivo.

Graphical abstract

Highlights

• •An IMS system is established in colorectal cancer
• •Three IMS molecular subtypes showed distinct metabolic and immunogenic properties
• •A S100A9⁺ macrophage population contributes to the immunosuppressive status
• •PD-1 blockade plus tasquinimod combination improves responses to immunotherapy

Bao et al. establish an immunometabolism subtyping system in colorectal cancer. Three IMS molecular subtypes (C1, C2, C3) show distinct metabolic and immunogenic properties. A S100A9⁺ macrophage population contributes to the immunosuppressive tumor environment in the C3 subtype. PD-1 blockade plus tasquinimod combination improves responses to immunotherapy.

Introduction

Colorectal cancer (CRC) is one of the most common malignancies and ranks the second leading cause of cancer death around the world.^1,2 More than 50% of CRC patients develop distant metastases during the course of disease.³ Despite recent advances in systemic treatment regimens, CRC remains challenging and rarely cured. In 2017, both the PD-1 inhibitors nivolumab and pembrolizumab were approved by the FDA for the treatment of deficient mismatch repair (dMMR)-high microsatellite instable (MSI-H) chemotherapy of refractory CRC,⁴ which is characterized by high tumor mutation burden (TMB) and immunogenicity. According to the encouraging result of KEYNOTE 177,⁵ immune checkpoint inhibitors (ICIs) have gained an indication of first-line treatment in metastatic CRC (mCRC). However, MSI-H mCRC accounted for only 5% of all mCRC patients, and the overall response rate of ICI monotherapy in these patients is only 44%.⁶ Most mCRC patients who belong to the microsatellite stable (MSS) subgroup display primary resistance to immunotherapy. Different combination therapies have been tried to improve the efficacy of immunotherapy for MSS-CRC^6,7; however, no breakthrough results have been achieved. Investigating the potential mechanism for the poor response to immunotherapy in the majority of CRCs is urgent for achieving a better prognosis in CRC patients.

The tumor microenvironment (TME) is a complex, heterogeneous cellular environment composed of different cell types. Several factors associated with dysregulation of the TME may be involved in the resistance of CRC patients to ICI-based immunotherapy, e.g., low T cell infiltration and low T cell cytotoxic activities and high infiltration of immune invasive-related M2 macrophages.⁸ The molecular subtyping of CRC helps to better stratify patients with distinct TME properties and to investigate personalized therapeutics. Several molecular subtyping methods have been proposed for CRC patients.^9,10 For instance, the consensus molecular subtype (CMS) I–IV classification system is one classical subtyping method for CRC. CMS1 (immune subtype) is defined as a tumor with high TMB and strong immune activation, CMS2 (canonical subtype) includes upregulation of WNT and MYC signaling, CMS3 (metabolic subtype) encompasses epithelial tumors with metabolic deregulation, and CMS4 (mesenchymal subtype) shows stromal infiltration, TGF-β activation, and angiogenesis.¹¹ Although potential treatment options are provided by the CMS classification system (e.g., TGF-β inhibitor application in CMS4), the clinical relevance remains to be properly addressed.

Altered metabolism is a hallmark of cancer.¹² Increased metabolic activity leads to the depletion of key nutrients and the accumulation of waste products within the TME, which further leads to the dysfunction of immune cells and immune evasion. Therefore, understanding the immunometabolism properties could help develop therapeutic strategies that leverage metabolism to boost antitumor immunity. In this study, we performed immunometabolism molecular subtyping (IMS) based on a comprehensive analysis of the transcriptome, proteome, and single-cell transcriptome in public training cohorts and in-house validation cohorts. The underlying mechanism is further explored.

Results

Construction of an IMS system in a large CRC cohort

The workflow of this research is shown in Figure 1A. In brief, an integrated public cohort (n = 1,399) and three in-house validation cohorts (MSS-CRC n = 114, MSI-CRC n = 30, and MSI-CRC n = 1) were constructed to perform multiomics analysis (transcriptome, proteome, single-cell transcriptome, and metabolome). In vivo analysis was performed to confirm the validity of the findings from the integrative analysis.

Figure 1.

IMS system of CRC patients with distinct prognoses and therapeutic response

(A) Graphical scheme describing the experimental workflow.

(B) Graphical scheme describing the IMS method.

(C) The volcano plot indicating the metabolism-related genes that were significantly associated with recurrence-free survival (RFS) of CRC as demonstrated by univariate Cox analysis.

(D) The volcano plot indicating the immune-related genes that were significantly associated with RFS of CRC by univariate Cox analysis. Bar plot indicating the enriched biological process from gene ontology analysis.

(F) Heatmap showing distinct metabolic-, immune-, and tumor microenvironment (TME)-related pathway activation in CRC tissues.

(G) Consensus clustering identifying IMS groups in the training cohort (patients n = 1,399).

(H) Bar plot indicating the proportion of each subtype in each cohort (left) and in the integrated training cohort (right).

(I) Principal-component analysis (PCA) analysis showed the difference of the three metabolism subtypes.

(J) Kaplan-Meier estimates of RFS for patients with different immunometabolism subtypes in the integrated training cohort.

(K) Kaplan-Meier estimates of RFS for patients with different immunometabolism subtypes in a testing cohort (GSE28722 n = 125).

The training cohort was constructed with seven independent cohorts. The immune-related and metabolism-related gene expression patterns and pathway activities were applied for consensus clustering. Patient prognosis and predicted immunotherapy were further analyzed (Figure 1B). First, we integrated seven public CRC cohorts after removing the batch effect of transcriptome data by sva method.¹³ Principal-component analysis (PCA) was applied for monitoring, which showed that no batch effects existed (Figure S1A). We then tested the metabolism-related and immune-related genes and pathway activities in the integrative transcriptome data. Firstly, we performed univariate Cox analysis for all metabolism-related and immune-related genes in the integrative cohort. A volcano plot indicated the 1,147 and 1,177 genes that were significantly associated with poor or prolonged survival in CRC, respectively (univariate Cox p < 0.01, Figures 1C and 1D, detailed information in Tables S1 and S2). Gene ontology (GO) analysis revealed that several metabolism- and immune-related GO enrichments (e.g., T cell activation, response to lipid, inflammatory response, glycoprotein metabolic process, carbohydrate metabolic process) were enriched with prognosis-significant immune genes and prognosis-significant metabolic genes, respectively (Figure 1E). The 2,324 genes were therefore used for further analysis. Secondly, the metabolic- and immune-related pathways (pathway information seen in the Gmt file Table S3) were also investigated, showing the distinctly activated pathways in the integrative cohort (Figure 1F). Three major clusters were found by the unsupervised clustering of metabolic- and immune-related pathway activities. We therefore applied the genes and pathways above to construct the IMS in CRC. In brief, we used the matrix of gene expression and pathway actives (calculated by ssGSEA method) as the input of consensus clustering. Three IMS groups were identified by consensus clustering as C1, C2, and C3 (Figure 1G). A similar proportion of each subtype was found among the seven cohorts, implying insignificant batch effects among the seven cohorts and the robustness of the IMS classification system (Figure 1H). PCA plot revealed clear transcriptome differences among the three subtypes (Figure 1I). The immunometabolism genes were then applied to build a CRC gene-metabolite network (Figure S1B). The Kaplan-Meier estimator suggested significantly poorer recurrence-free survival (RFS) in C3 patients than in C1 and C2 patients (p < 0.001) (Figure 1J). We then tested the IMS in a test cohort (GSE28722 n = 125). The tumor tissues were separated into C1, C2, and C3 subtypes and patients with the C3 subtype showed the poorest survival outcome compared with C1 or C2 subtype patients (p = 0.035), confirming the robustness of the IMS system for stratifying patients with poor prognosis (Figure 1K).

Given the existence of several molecular subtyping methods (e.g., the CMS method⁹ and the modified iCMS method¹⁰) in CRC, we then compared the CMS classification method and IMS system. The CMS system also showed the capacity for stratifying patients with poor prognosis (p = 0.0023 compared with p < 0.0001 for the IMS system) (Figure S1C). Importantly, CMS4, the mesenchymal type, which exhibited poorest prognosis, encompassed 18% of C1 subtype cases, 9% C2, and 73% C3 (Figure S1D). Regarding the genomic differences among the three IMS groups, we tested the IMS in TCGA colorectal cancer (TCGA-COADREAD) cohort (Figure S1E). The MSI patients constituted 21.1%, 7%, and 14.3% of the C1, C2, and C3 clusters, respectively (Figure S1F). C2 exhibited a greater altered fraction in the genome than the C1 and C3 clusters (Figure S1G). ACTA1, PDZD2, and several other genes were prominently mutated in the C3 subtype compared with the non-C3 subtype (Figure S1H). Taken together, we revealed three IMS groups in CRC patients with distinct prognoses and immunotherapy responses. The comparison of the IMS system and the CMS classification method showed the link between these systems. The genomic difference showed the genomic landscape for the three IMS groups.

Distinct immune and metabolic properties of the IMS system in the in-house MSS-CRC cohort showed by integrated transcriptomic, proteomic, and metabolomic analysis

To investigate the underlying biological mechanism for the three IMS subtypes, we then performed transcriptomic, proteomic, and metabolomic analyses based on our in-house MSS-CRC cohort. Tumor tissues and plasma samples were collected from 114 patients (sample information is shown in Table S4). Transcriptomics and proteomics were performed using tumor tissues, and metabolomics was performed using plasma samples (Figure 2A). Transcriptomics was used to extract the metabolism-related and immune-related gene expression profile (the same as the training cohort) and was used to calculate the pathway activities using the ssGSEA method. With the matrix of gene expression and pathway activities as the input of consensus clustering, three groups were identified using the established IMS system. In total, 2,864 differentially expressed gene (DEGs) were identified between C3 and C2, 617 between C1 and C2, and 774 between C3 and C1 subtypes (DEGs: |fold change| > 2 and adjusted p < 0.01) (Figures S2A–S2C). The Venn diagram shows the overlapping DEGs from the comparisons (Figure 2B). The most significant DEGs are shown in Figure 2C, suggesting distinct expression patterns between the C3 and non-C3 subtypes. For instance, RGS5, SPARCL1, LXN, FGL2, and GIMAP7 were significantly higher in C3 compared with non-C3 subtypes (Figure 2C). As a distinct immune response was found in the three immunometabolism subtypes of human CRC, we then analyzed the expression of immune pathway-related genes. We found that most genes involved in antigen presentation, cell adhesion, coinhibitory, costimulatory, ligand, and receptor were significantly upregulated in the C3 subtypes compared with the non-C3 subtypes (Figure S2D). Furthermore, most immune checkpoint genes showed upregulation in the C3 subtype, except PDCD1 (PD-1) (Figure S2E). The associations among immune checkpoints, evasion genes, and metabolism pathways were then investigated. For instance, CSF1R, a key gene involved in promoting the M2 macrophage phenotype, showed a negative correlation with most metabolic pathways, whereas PAK6 showed a positive correlation with most metabolic pathways (Figure S2F). To further illustrate the proteomic landscape of the IMS at the protein level, we applied proteomics analysis to the same tumor tissues. A total of 9,469 proteins were identified, and no batch effects were found during proteomics loading (Figure S2G). We performed the IMS system with proteomics data. The confusion matrix suggested the consistency of IMS at the RNA and protein levels with a high accuracy rate (Figure S2H). Furthermore, the differently expressed proteins (DEPs) were investigated to show the proteomic landscape. In total, 3,365 DEPs were noted between C3 and C2, 577 between C1 and C2, and 833 between C3 and C1 subtypes (DEPs: adjusted p < 0.01) (Figures S2I–S2K).

Figure 2.

Multiomics analysis of tumor and plasma samples in an in-house MSS-CRC cohort (n = 114)

(A) Graphical scheme describing the multiomics workflow including RNA transcriptome, proteomics, and metabolomics.

(B) Venn diagrams showing the overlapping genes for a comparison of (B) and (E).

(C) Heatmap showing the most significant DEGs.

(D) Heatmap showing the most significant DEMs.

(E) The enriched pathways in the C3 subtype identified from DEMs.

(F) The metabolite-pathway network in the C3 subtype.

Metabolomics was then applied to reveal the differentially upregulated metabolites from the plasma of the three IMS groups. In total, 343 differently expressed metabolites (DEMs) were noted between C3 and C2, 124 between C1 and C2, and 90 between C3 and C1 subtypes (DEMs: adjusted p < 0.05) (Figures S2L–S2N). N-Lauroyl lysine and 2-[(2-furanylmethyl)thio]-6-methylpyrazine were the featured metabolites in the C1 subtype. 3-Deoxy-D-manno-octulosonate and methyl 3-mercaptobutanoate were highly released in C2, while fasoracetam, L-alpha-amino-1H-pyrrole-1-hexanoic acid, and methyl sorbate were significantly enriched in C3 subtypes. The most differentially altered metabolites are shown in Figure 2D, suggesting distinct enriched metabolites between the C3 and non-C3 subtypes. The metabolite-pathway network analysis further suggested that the cGMP-PKG signaling pathway, renin secretion, primary immunodeficiency, creatine pathway, C4-dicarboxylic acid cycle, arachidonic acid metabolism, folate biosynthesis, and other pathways were enriched in C3 subtypes (Figures 2E, 2F, and S2O).

Taken together, we applied transcriptomic, proteomic, and metabolomic analyses in our in-house MSS-CRC cohort. Transcriptome analysis showed differences in immune pathway-related genes and immune checkpoint genes among the three subtypes. Proteomic and metabolomics further suggested distinct proteomic and metabolite differences from patients with the three subtypes, especially between the C3 and non-C3 subtypes.

The immune landscape of the IMS in the in-house MSS-CRC cohort

Next, we analyzed the immune cell landscape of the three IMS subtypes in the in-house MSS-CRC cohort and the integrated GEO cohort. The C2 subtype showed less immune infiltration than the C1 and C3 subtypes in both cohorts (Figures 3A and S2A). We therefore considered C1 as an immune inflamed phenotype, C2 as an immune dessert phenotype, and C3 as an immune tolerance phenotype. Macrophages were significantly enriched in the C3 subtype (Figures 3B and S3A). Furthermore, the association of immune cell abundance and metabolism pathways showed a positive correlation of macrophages with folate metabolism, glycosaminoglycan biosynthesis and arachidonic acid metabolism in CRC (Figure S3B). As a significant difference was identified for macrophages in the three IMS groups, we then applied mIHC staining to reveal the abundance of macrophage phenotypes. The results showed that the greatest numbers of PD-L1⁺ CD68⁺ cells (p = 0.0029), CD163⁺ CD68⁺ cells (p = 0.003), and HLA-DR⁺ CD68⁺ cells (p = 0.0019) were noted in the C3 subtype (Figures 3D–3F). The results above further confirmed an immune tolerant phenotype with high abundance of immunosuppressive macrophage phenotype (PD-L1⁺ CD68⁺ cells and CD163⁺ CD68⁺ cells) of C3 subtype.

Figure 3.

The immune gene landscape in the three IMS groups in human CRC

(A) Heatmap showing the expression of genes involved in several immune pathways in the three IMS groups in the training cohort.

(B) The network of S100A9 and metabolism pathways. Yellow indicates a positive correlation, while green indicates a negative correlation. The line thickness represents the absolute value of Pearson’s coefficient.

(C) Macrophage multiplex immunohistochemistry (mIHC) panel of the C1, C2, and C3 subtypes. Labeling colors: yellow (CD68), red (CD163), green (HLA-DR), purple (PD-L1), and cyan (PanCK).

(D–F) The statistics of PD-L1⁺ CD68⁺ cells (D), CD163⁺ CD68⁺ cells (E), and HLA-DR⁺ CD68⁺ cells (F) in the C1, C2, and C3 subtypes (technical replicates n = 6).

Distinct recurrence rate of IMS groups in the in-house MSI-CRC cohort

MSI-H CRC patients may benefit from immunotherapy. We therefore investigated the IMS in in-house MSI-CRC patients. An in-house cohort was constructed with 30 MSI patients (sample information is shown in Table S5) (Figure 4A). H&E staining and IHC staining of PMS2, MSH2, MSH6, and MLH1 (the deficiencies of the four proteins are commonly used for detection of MSI status) were performed to confirm the MSI status of the 30 patients (Figure 4B). The RECIST 1.1 criteria were used on computed tomography (CT) scans to test whether the patients relapsed over 5 years (Figure 4C). Most patients who belonged to the C3 IMS group relapsed within 5 years (70%) (chi-square p < 0.01) (Figure 4D). To better illustrate the pathway alterations in C3 versus non-C3 subtypes, FGSEA was performed (Figure 4E), showing that metabolism- and immune-related pathways (e.g., inflammatory response, ribosome biogenesis, and chromatin modification) were altered. The immune cell landscape was performed with the 22 immune cell signatures using the ssGSEA method, which showed distinct cell abundance in the three IMS groups. The results were consistent with the findings from the integrative GEO cohort and the MSS-CRC cohort (Figure 4F). In particular, tumors of the C3 subtype had greater macrophage abundance than those of the C1 and C2 subtypes (Figure 4G). The greater macrophage abundance in C3 was further validated by IHC and IF staining (Figures 4H and 4I). Finally, the immunotherapy-related pathways and the innate PD-1 resistance signature pathways indicated distinct immune pathway activation in the three subtypes (Figure 4J). Particularly, patients in the C3 subtype may benefit more from immunotherapy by PD-1 blockade (Figure 4J, upper panel). Nevertheless, most of the innate PD-1 resistance signature also showed high activities in the C3 subtype (Figure 4J, bottom panel). We then inferred that the combination application of PD-1 blockade and other therapies (e.g., macrophage-targeting therapeutics) may help improve the immunotherapy response. Taken together, we explored the underlying regulation pattern and the immune landscape of IMS subtypes in both in-house MSS-CRC and MSI-CRC cohorts. Especially, macrophage abundance was significantly enriched in the C3 subtype in both public and in-house MSS and MSI cohorts. The abundant macrophages in C3 and the association with poor survival is worthy of further exploration.

Figure 4.

Validation of the IMS system in an in-house human MSI-H CRC cohort

(A) Schematic of the data analysis of the in-house MSI-CRC cohort.

(B) H&E and IHC staining (PSM2, MSH2, MSH6, and MLH1) for validation of the MSI status of CRC.

(C) Typical CT scans for relapse and nonrelapse patients.

(D) The 5-year relapse rate for patients in the three IMS groups.

(E) The FGSEA between C3 and non-C3 subtypes.

(F) Heatmap showing the immune cell abundance in the three IMS groups.

(G) Boxplot showing the macrophage abundance in the three IMS groups (C1 n = 12, C2 n = 8, and C3 n = 10).

(H) CD68 immunohistochemistry (IHC) staining for the three IMS groups. Scale bar, 120 μm.

(I) Quantification of macrophages in the three IMS groups (C1 n = 12, C2 n = 8, and C3 n = 10).

(J) Heatmap showing the immunotherapy-related pathways and innate PD-1 resistance signature (IPRES) in the three IMS groups.

Single-cell RNA sequencing analysis of a PD-1-resistant MSI-H patient revealed that S100A9⁺ macrophages contribute to the C3 phenotype

To investigate the immunosuppressive TME in C3 MSI patients at the single-cell level, one special tumor tissue was used to generate single-cell RNA sequencing (scRNA-seq) data (Figure 5A). The patient was identified as belonging to the C3 subtype based on bulk transcriptome data. PD-1 monotherapy was applied as the first-line treatment. After two cycles of treatment, due to the worsening of clinical symptoms, we preformed CT examination, which indicated that the target lesion increased by 133%. The tSNE plot revealed the cell types (e.g., endothelial cells, fibroblast cells, epithelial cells, stromal cells, CD4⁺ T cells, and CD8⁺ T cells) in this C3 MSI patient (Figure 5B). Distinct expression profiles were identified in several cell types. For instance, GZMA and GNLY were highly expressed in CD8⁺ T cells, whereas SPP1 and TIMP1 were highly expressed in monocytes/macrophages (Figure S4A). A great abundance of monocytes/macrophages (53%) was found in the C3 MSI-H patient (Figure 5D). We then applied the IMS method to the scRNA-seq data. Sixty-two percent of cells from the C3 tumor were identified as the C3 type (Figures 5F–5G). Most monocytes/macrophages and neutrophils were identified as the C3 type cells while other cell types showed non-C3 properties. In detail, 76% of cells in the C3 type were macrophages/monocytes (C3 macrophages/monocytes), whereas only 15% of cells in the non-C3 type were macrophages/monocytes (non-C3 macrophages/monocytes). These findings highlight the importance of macrophages/monocytes in contributing to the C3 phenotype (Figure 5H). Metabolism pathway analysis suggested that folate metabolism, glycolysis, gluconeogenesis, and nitrogen metabolism were significantly upregulated in monocytes/macrophages, which showed distinct metabolic activities among different cell types (Figure S4B). The cell-cell interaction among each cell type was also investigated. C3 macrophages/monocytes showed a stronger interaction with CD8⁺ T cells than non-C3 macrophages/monocytes (Figures 5J, S4C, and S4D). For instance, CCL signaling pathway activities were found between C3 macrophages/monocytes and CD8⁺ T cells but not between non-C3 macrophages/monocytes and CD8⁺ T cells (Figures S4E and S4F).

Figure 5.

Single-cell transcriptome analysis in C3-CRC patients who did not respond to immunotherapy

(A) Scheme for single-cell transcriptome analyses.

(B) Color-coded t-distributed stochastic neighbor embedding (tSNE) plot of cell types in the single-cell transcriptome data.

(C) The proportion of cell types in the single-cell transcriptome data.

(D) Color-coded tSNE plot of cell clusters in the single-cell transcriptome data.

(E) The proportion of cell clusters in the single-cell transcriptome data.

(F) The proportion of cell types in the C3 subtype and non-C3 subtype.

(G) Network plot indicating the ligand-receptor (LR) pairs for cell-cell interactions (CCIs) between each cell population. The size of the line represents the number of cell interactions between the ligand and receptor gene pairs.

(H) Color-coded tSNE plot of monocyte/macrophage subpopulations in the single-cell transcriptome data.

(I) Heatmap showing the feature genes in macrophage/monocyte subpopulations.

(J) The proportion of macrophage/monocyte subpopulations in the C3 and non-C3 subtypes. (K) Violin plot showing S100A9 expression in each cell type.

(L) Boxplot showing S100A9 expression in the C3 subtype and non-C3 subtype.

(M) Color-coded tSNE plot of S100A9 expression in monocyte/macrophage subpopulations.

(N) ROC analysis for the capacity of S100A9 expression to stratify cells belonging to the C3 and non-C3 subtypes.

(S) Violin plot showing the S100A9⁺ Mac score in the three subtypes in the integrative CRC cohort.

(T) ROC analysis for the capacity of the S100A9⁺ Mac score to stratify patients belonging to the C3 and non-C3 subtypes.

(U) Kaplan-Meier estimates of RFS for patients with high or low S100A9⁺ Mac scores in the integrated training cohort.

As macrophages significantly contributed to the C3 phenotype, we investigated macrophage heterogeneity to explore the influence of macrophage/monocyte subpopulations on the IMS system. Five subpopulations (Mac/Mono 1, Mac/Mono 2, Mac/Mono 3, Mac/Mono 4, Mac/Mono 5) were found in monocytes/macrophages (Figure 5H). The distinct gene expression pattern further confirmed the heterogeneity of Mac/Mono subpopulations (Figure 5I). For instance, C1QA, C1QB, and APOE were highly expressed in the Mac/Mono 3 population, which represented the resident macrophage population (Figure 5I). S100A9⁺ Mac/Mono 1, Mac/Mono 2, and Mac/Mono 3 constituted 83.5% of C3 monocytes/macrophages, whereas Mac/Mono 5 constituted 88.4% of non-C3 monocytes/macrophages (Figure 5J). Furthermore, S100A9 expression was significantly greater in monocytes/macrophages compared with other cell populations (Figure 5K). Regarding the Mac/Mono cell population, S100A9 expression was greater in C3 Mac/Mono populations (Mac/Mono 1, Mac/Mono 2, and Mac/Mono 3) compared with non-C3 populations (Mac/Mono 4 and Mac/Mono 5) (p < 0.0001) (Figures 5L and 5M). Furthermore, S100A9 expression can distinguish the C3 and non-C3 populations with an area under the curve (AUC) of 0.87 (Figure 5N). We also tested the IMS system in a public cohort (Zhang et al. cohort)¹⁴ (Figure S4G). Consistent with our findings, most monocytes/macrophages were identified as the C3 type cells while other cell types showed non-C3 properties (Figure S4H). Furthermore, macrophages can be further divided into S100A9^high and S100A9^low clusters with distinct S100A9 expression (Figures S4I and S4J). S100A9 ^high macrophages constituted majority of C3 macrophages (Figure S4K). We therefore identified S100A9⁺ Mac/Mono as a special Mac/Mono type that contributes to the C3 subtype in CRC at the single-cell level.

Finally, we validated the relationship between the S100A9⁺ Mac/Mono and C3 subtypes at the bulk RNA-seq level. S100A9 expression was significantly associated with folate metabolism activity, arachidonic acid metabolism, and glycosaminoglycan biosynthesis, which contribute to the activation of immunosuppressive macrophages in CRC (Figure 5O). The S100A9⁺ Mac/Mono score was calculated with the S100A9⁺ Mac/Mono feature gene using the ssGSEA method. The C3 subtype exhibited a greater S100A9⁺ Mac/Mono score than the C1 and C2 subtypes (p < 0.0001) (Figure 5P). The S100A9⁺ Mac/Mono score exhibited an AUC of 0.79 to distinguish the C3 and non-C3 subtypes in the bulk transcriptome (Figure 5Q). Finally, CRC patients with a higher S100A9⁺ Mac/Mono score exhibited poorer RFS than patients with a lower S100A9⁺ Mac/Mono score (HR = 1.39; 95% CI, 1.15–1.7) (Figure 5R). Taken together, we confirmed that S100A9⁺ Mac/Mono contributed to the immunosuppressive C3 subtype at both the single-cell and bulk transcriptome levels.

Dual inhibition of S100A9⁺ macrophages and PD-1 in vivo showed a combined effect on tumor regression

A mouse model of CRC was used to further test the treatment potential of S100A9⁺ monocytes/macrophages in CRC. Four groups were established: the control group (G1), the tasquinimod treatment group (G2), the anti-PD-1 treatment group (G3), and the tasquinimod+anti-PD-1 dual treatment group (G4). From the 8th day, tasquinimod was administered by oral gavage once a day for 15 days. Treatments with anti-PD-1 were administered every 3 days for a total of four times. The experimental schedule is shown in Figure 6A. According to the tumor growth curve of mice in each group, treatment with anti-PD-1, tasquinimod, and tasquinimod+anti-PD-1 dual treatment resulted in a delay in tumor growth compared with that of the control group (Figures 6B and 6C). Moreover, mice receiving tasquinimod+anti-PD-1 dual treatment showed the highest tumor inhibition efficacy compared with those of all other groups (Figure 6D). We further dissected the tumors and weighed them. As shown in Figure 6E, tasquinimod+anti-PD-1 dual treatment caused noticeable tumor inhibitory effects, which is consistent with the tumor volume results. Treatment with tasquinimod+anti-PD-1 significantly delayed tumor growth compared with treatment with anti-PD-1, as evidenced by the tumor volume and weight, suggesting that tasquinimod+anti-PD-1 antibody dual treatment can enhance checkpoint blockade-based immunotherapy. In addition, no significant variation in the body weights of the mice was observed during the above treatments (Figure S5A). We also tested the metabolic features in an in vivo model. With the DEMs, GO analysis was performed (Figure S5B). The results suggested several key metabolism pathways such as folate biosynthesis and arachidonic acid metabolism were enriched, which was consistent with our findings in human samples. We then performed flow cytometry analysis on the tumor tissues from the four groups. The gating strategy for macrophage populations is shown in Figure 6F. The macrophage proportion in all CD45⁺ cells was decreased in G2, G3, and G4 compared with G1 (Figure 6G). We therefore tested several key markers such as CD86, CD206, PD-L1, Arg1, CD163, iNOS, and MHC-II. The G2 and G4 groups showed significantly higher M1 macrophages in all macrophage populations compared with the G1 group (Figure 6H for CD86). Furthermore, the G2 and G4 groups showed decreased M2 macrophage populations compared with the G1 group (Figure 6I for CD206). PD-L1⁺ macrophage was increased in G2 and G4 (Figures S5C and S5D). The increased PD-L1 expression in the TME may in turn reduce the tasquinimod-induced antitumor response. The increased PD-L1 in the TME therefore provided an explanation that the association of tasquinimod with anti-PD-1 block enhanced the antitumor immunity compared with tasquinimod alone. Arg1 and CD163, two other markers for the M2 phenotype, were also decreased in G2, G3, and G4 (Figures S5E and S5H). In contrast, iNOS and MHC-II, were increased in G2, G3, and G4 (Figures S5I–S5L). Tasquinimod+anti-PD-1 antibody dual treatment exhibited a combined effect in depleting M2 populations compared with tasquinimod or anti-PD-1 antibody treatment alone (Figure 6I). Therefore, the M1/M2 ratio was highest in the G4 group (Figure 6J). We then investigated the influence of tasquinimod+anti-PD-1 antibody dual treatment on T cell activation. Th1 cells (IFN-γ⁺CD4⁺) and CTLs (IFN-γ⁺CD8⁺) were identified. The numbers of CD4⁺ T cells and CD8⁺ T cells were analyzed (Figures S6A and S6B). The number of CD8⁺ T cells were greater in G2 and G4 (Figure S6B). As shown in Figures 6K–6N, S6C, and S6D, the number and portion of Th1 cells and CTLs in G4 group were significantly higher than those in the other groups, respectively. Exhaustion and other important markers were also detected. PD-1 and TIM-3 showed no significance. TCF-7 was increased in the G3 and G4 groups (Figures S6E–S6G). These data are consistent with one previous study, which showed that TCF-7⁺ T cells have expansion, regeneration, and differentiation capacity and are critical for tumor control in response to immunotherapy.¹⁵ Finally, to determine whether tasquinimod-mediated antitumor immune response is dependent on macrophages, we used clodronate liposomes (Clo) to deplete macrophages from mice (Figure S6H). As shown in Figures S6I and S6J, macrophage depletion by Clo significantly attenuated the inhibitory effect of tasquinimod in the mouse model. Therefore, combined treatments using tasquinimod and PD-1 antibody reduce the number of TAMs while reshaping the pro-tumor M2-type TAMs into antitumor M1-type polarization, thus activating T cell-mediated antitumor immune responses. Together, the tasquinimod+anti-PD-1 antibody dual treatment can induce a strong antitumor immunotherapy against CRC.

Figure 6.

In vivo experiments tested the combination effects of the anti-PD-1 and anti-S100A9 regimens

(A) Schedule for various treatments (control group, tasquinimod treatment group, anti-PD-1 treatment group, and tasquinimod+anti-PD-1 dual treatment group) against the subcutaneously implanted MC38 model.

(B) Spider diagrams showing the growth kinetics of individual tumor volumes in the control and treated groups.

(C) Average tumor growth kinetics of each group under different treatments.

(D) Subcutaneous tumors surgically removed from each group.

(E) Average tumor weights of mice in each group monitored over time.

(F) Gating strategy for the detection of macrophage populations by flow cytometry.

(G–J) Relative infiltration of (G) macrophages, (H) M1 macrophages, (I) M2 macrophages, and (J) the M1/M2 ratio in the TME of each group.

(K–L) The proportion of $\mathrm{I}\mathrm{F}\mathrm{N}-{\gamma }^{+}$ cells in CD4⁺ T cells (technical replicates n = 6 for each group).

(M and N) The proportion of $\mathrm{I}\mathrm{F}\mathrm{N}-{\gamma }^{+}$ cells in CD8⁺ T cells (technical replicates n = 6 for each group).

Discussion

Immunotherapy has been recommended as the standard treatment in MSI mCRCs. However, most CRC patients have pMMR/MSS. Several factors impede the success of ICIs in pMRR/MSS-CRC, including a low TMB, fewer neoantigens, and more infiltrating immune suppressive cells (e.g., MDSCs, Tregs, and TAMs) with activation of immunosuppressive pathways.^16,17 Several strategies have been proposed to overcome the obstacle of tumor evasion and immune resistance by combining ICIs with other therapeutics, including agents that increase tumor antigen presentation, promote effector immune cell recognition, and generate an immunogenic TME.^18,19,20,21 In recent years, reversing the suppressive metabolic tumor environment has provided insight for improving ICI-based immunotherapy.^22,23 Here, we performed a comprehensive multiomics analysis of the IMS system to illustrate the underlying mechanism of immunometabolism in CRC.

We found that patients with the C3 subtype showed the poorest prognosis. Next, we explored the immunologic and metabolic characteristics associated with the different treatment outcomes in the three subtypes. C1 and C3 both have abundant immune cell infiltration compared with C2, which has a poor immune infiltration status. Nevertheless, immunosuppressive cell types (e.g., macrophages and MDSCs) were more abundant in the C3 subtype than in the C1 subtype. Therefore, we considered that C1 may exhibit an immune inflamed phenotype, C2 shows an immune dessert phenotype, and C3 represents an immune tolerance phenotype. Besides, the number of featured metabolites in the C3 subtype was significantly greater than featured metabolites in C1 and C2 subtypes. Several C3-featured metabolites, such as 2-octenedioic acid²⁴ and lactic acid,²⁵ have been proved to play key roles in suppressing the anti-cancer immune response or acting as an indicator for the abnormal TME metabolism.

MSI-CRC patients respond well to immunotherapy; therefore, we applied the subtyping method in our in-house MSI-CRC cohort. Strikingly, most of the patients in the C3 subtype relapsed within 5 years, whereas most of the patients who belonged to the other two subtypes did not relapse. Consistent with the training cohort and the testing in-house MSS-CRC cohort, the patients in the C3 subtype had greater macrophage abundance. Targeting macrophages has emerged as a promising intervention to reduce tumor progression and overcome immunotherapy resistance. For instance, a CSF-1R inhibitor (JNJ-40346527) has been evaluated in a phase I clinical test for high-risk localized prostate tumor (NCT03177460). Trastuzumab plus CD47 blockade regimen has shown promising efficacy in breast cancer therapy by improving phagocytosis mediated by macrophages.²⁶ Tasquinimod is a small molecule exhibiting pleiotropic effects on the tumor TME.²⁷ It binds and inhibits the interactions of the damage-associated molecular pattern receptor S100A9, a key cell surface regulator of macrophage function.²⁸ Hence, tasquinimod redirects the phenotype from pro-angiogenic and immunosuppressive TAMs to pro-inflammatory macrophages.²⁷ Tasquinimod has been evaluated in several clinical trials as either a single intervention or in combination with other therapeutics.²⁹ Considering the abundant macrophage populations in C3 patients, combination immunotherapy with a macrophage-targeting drug (e.g., tasquinimod) may help improve the efficacy of systemic treatment in patients with the C3 subtype.

We applied our IMS system at the single-cell level and found that greater than 50% of TME cells belonged to the C3 subtype in the sequenced C3 tumor tissue. Furthermore, the C3 subtype cells were mainly composed of monocytes/macrophages. The findings were consistent with what we found in the bulk transcriptome and staining data, which showed greater macrophage abundance of macrophages in the C3 subtype than in the non-C3 subtype. Furthermore, C3 type macrophages showed strong interactions with CD8⁺ T cells than non-C3 macrophages. Several signaling pathways such as CCL were significantly upregulated between C3 type macrophages and CD8⁺ T cells. We therefore confirmed that monocytes/macrophages were the key players that contributed to the C3 phenotype at the single-cell level. Hence, we investigated which monocyte/macrophage subpopulation could contribute to the poor immunotherapy response in C3 CRC. In our analysis of CRC, S100A9 expression was associated with poor prognosis. Importantly, through scRNA-seq, we did not observe high S100A9 expression in epithelial cells. Most S100A9 expression was observed in monocytes/macrophages. Regarding the heterogeneity of monocytes/macrophages, we found that S100A9⁺ macrophage constituted more than 80% of C3 type macrophages. S100A9 expression exhibited a strong capacity to dedifferentiate the C3 and non-C3 subtype macrophages, implying the potential important roles of S100A9⁺ macrophages in the C3 subtype. Therefore, S100A9⁺ macrophages may represent an immunotherapeutic target for C3 patients as these patients exhibit the worst prognosis and are poor responders to immunotherapy.

Reprogramming TME cell metabolism alters tumor proliferation, metastasis, and other biological processes, whereas the TME is altered in response to hypoxia, acidosis, nutrient depletion, and cellular waste accumulation, promoting tumor immune evasion. The interplay between metabolism and immune cells plays crucial roles in reshaping the degree of phenotypic and molecular heterogeneity within the tumor. For instance, one target in the Trp-Kyn-AhR pathway was indoleamine 2,3-dioxygenase 1 (IDO1),³⁰ a principal enzyme involved in tryptophan catabolism. Moreover, blocking the glycolytic pathway using IDO inhibitors regulates D-2-HG levels in MDSCs, thus hindering TAM-M2 polarization.^31,32 Another study revealed that the combination of the anti-folate agent pemetrexed with ICIs shows direct antitumor effects together by stimulating mitochondrial biogenesis of CTLs, therefore facilitating their antitumor activation.³³ The characteristic metabolic plasticity of these cells above suggests a preferred target for immunotherapy development.

Hence, a regimen combining metabolic inhibitors with ICIs should be developed that targets immune regulatory TME metabolites and takes TME cell metabolism into account to reverse immune escape and favor the antitumor immune response. Through our integrative scRNA-seq and bulk transcriptome analysis, we demonstrated the importance of S100A9⁺ macrophages in regulating the suppressive TME. S100A9 forms a heterodimer with S100A8.³⁴ The S100A8/S100A9 heterodimer plays crucial roles in regulating the metabolic activities of multiple immune cells. For instance, S100A8/S100A9 facilitates leukocyte arachidonic acid trafficking and metabolism,³⁴ modulates the tubulin-dependent cytoskeleton during the migration of phagocytes,³⁵ activates neutrophil NADPH-oxidase,³⁶ and promotes arachidonic acid metabolism in macrophages.³⁷ Our analysis also suggested enriched arachidonic acid metabolism in the C3 subtype, the suppressive TME of which was orchestrated by S100A9⁺ macrophages. We therefore applied a combination regimen including an S100A9 inhibitor tasquinimod and ICI to test the antitumor activity in CRC. The combined regimen showed a better effect than PD-1 alone or S100A9 inhibitor alone, suggesting that targeting S100A9⁺ macrophages may improve the ICI-based immunotherapy response by reversing immune escape.

In conclusion, we introduced an IMS system by integrating transcriptomic, proteomic, and metabolomic data. Three subtypes were identified for stratifying CRC patients with different prognoses. The C3 subtype, which featured high S100A9⁺ macrophage abundance, was associated with the poor outcome. S100A9⁺ macrophages showed the capacity to affect T cells by cell-cell education in the C3 subtype of CRC. The combination strategy of targeting S100A9⁺ macrophages and immune checkpoint might open up a magnificent prospect for reversing immune escape and improving immunotherapy efficacy in certain characteristic immunometabolism subtype of CRC patients.

Limitation of the study

First, we recruited two in-house cohorts. The number of recruited patients is relatively low. Also, the survival data of in-house MSS cohort are not available. Second, the number of patient samples used for scRNA-seq is relatively low. Only one PD-1-resistant MSI-H patient was recruited for generating scRNA-seq data. Third, we illustrated the importance of S100A9⁺ macrophages in the CRC TME. The efficiency of macrophage depletion assay by Clo only reached about 80% in vivo, which may need to be further validated using a conditional gene knockout mouse model. Finally, the exhausted T cell phenotype and the underlying molecular mechanism of the in vivo model needs further exploration.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit monoclonal Anti-CD68	Biolynx	Cat#BX50031
Rabbit monoclonal Anti-PD-L1	Cell Signaling Technology	Cat#13684S
Rabbit monoclonal Anti-CD163	Abcam	Cat#ab182422
Rabbit monoclonal Anti-HLA-DR	Abcam	Cat#ab92511
Rabbit monoclonal Anti-PanCK	GeneTech	Cat#GM351507
Mouse monoclonal Anti-CD45-BV785	BioLegend	Cat#103149
Mouse monoclonal Anti-IFN-γ-PE-cy7	BioLegend	Cat#505826
Mouse monoclonal Anti-CD206-APC	BioLegend	Cat#141708
Mouse monoclonal Anti-F4/80-PE-cy7	BioLegend	Cat#123114
Mouse monoclonal Anti-CD86-PE	BioLegend	Cat#159204
Mouse monoclonal Anti-CD3-AF700	BioLegend	Cat#100216
Mouse monoclonal Anti-CD45-FITC	BioLegend	Cat#157608
Mouse monoclonal Anti-CD4-PerCP-Cy™5.5	BD Biosciences	Cat#561115
Mouse monoclonal Anti-CD8-BV510	BD Biosciences	Cat#563068
Mouse monoclonal Anti-CD11b-Buv395	BD Biosciences	Cat#743983
Fixable Viability Stain 700	BD Biosciences	Cat#564997
Mouse monoclonal Anti-CD45-BV785	BD Biosciences	Cat#563716
Mouse monoclonal Anti-CD11b-Buv395	BD Biosciences	Cat#563553
Mouse monoclonal Anti-F4/80- BV650	BioLegend	Cat#123149
Mouse monoclonal Anti-I-A/I-E− PE-cy7	BioLegend	Cat#107630
Human monoclonal Anti-CD163- BV421	BioLegend	Cat#333612
Mouse monoclonal Anti-ArgI-APC	Thermo	Cat#17-3697-82
Mouse monoclonal Anti-CD3-FITC	BioLegend	Cat#100306
Mouse monoclonal Anti-INOS-PE	BioLegend	Cat#696806
Mouse monoclonal Anti-CD4-PerCP-Cy5.5	BD Biosciences	Cat#100744
Mouse monoclonal Anti-TIM-3-APC	BioLegend	Cat#119706
Mouse monoclonal Anti-CD8-BV605	BioLegend	Cat#552051
Mouse monoclonal Anti-TCF-7-PE	BD Biosciences	Cat#564217
Tasquinimod	Sellect	Cat#S7617
Combo: Clophosome®-A and Control Liposomes (anionic)	FormuMax	Cat#F70101C-AC-10
Biological samples
Male C57BL/6 mice	Zhejiang Center of Laboratory Animals	N/A
Critical commercial assays
Bulk transcriptome	Shanghai Lu-Ming Biotech Co.,Ltd	N/A
Single-cell sequencing	ZHEJIANG PULUOTING HEALTH TECH CO., LTD.	N/A
Tandem mass tag (TMT)-based proteomics	Westlake Omics	N/A
Liquid chromatography-mass spectrometry	Shanghai Lu-Ming Biotech Co.,Ltd	N/A
Deposited data
RNA sequencing data	This paper	HRA003444
Single-cell sequencing data	This paper	HRA003501
Proteomics data	This paper	NGDC: OMIX002456
Metabolomics data	This paper	NGDC: OMIX002340
TCGA-COAD cohort RNA sequencing data	TCGA database	N/A
GEO cohorts	GEO	GSE56699, GSE14333, GSE39582, GSE17536, GSE17537, GSE33113, GSE37892
Experimental models: Cell lines
Mc38 cell line	YBio	Cat# YB-H4117
Software and algorithms
R software (version 4.04)	R Core	https://www.r-project.org/
Seurat (version 4.2.0)	Butler et al., 201838	https://satijalab.org/seurat/
sva R package (version 3.44.0)	Leek et al. 201213	https://bioconductor.org/packages/release/bioc/html/sva.html
ConsensusClusterPlus (version 3.16)	Matt Wilkerson, Peter Waltman	https://git.bioconductor.org/packages/ConsensusClusterPlus
CMScaller R package	Eide et al. 201739	https://github.com/peterawe/CMScaller
FGSEA R package (version 3.16)	Korotkevich et al. 202140	https://bioconductor.org/packages/release/bioc/html/fgsea.html
ClusterProfiler R package (version 4.5.2)	Yu et al. 201241	https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html
GSVA R package (version 1.45.5)	Hänzelmann et al. 201342	https://bioconductor.org/packages/release/bioc/html/GSVA.html
Cellchat R package	Jin et al. 202143	https://github.com/sqjin/CellChat/issues
Limma R package (version 3.52.4)	Smyth et al. 200544	https://bioconductor.org/packages/release/bioc/html/limma.html
Image-Pro Plus software (version 6.0)	Media Cybernetics	https://www.mediacy.com/78-products/image-pro-plus
ImageJ software (version 4.0)	NIH	https://imagej.nih.gov/ij/
Other
Gene sets	The Broad Institute	https://www.gsea-msigdb.org/gsea/msigdb/
Immune cell signatures	Bindea et al. 201345	PMID:24138885

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Weijia Fang (weijiafang@zju.edu.cn).

Materials availability

This study did not generate new unique reagents.

Experimental model and subject details

Cell lines

MC38 cells were purchased from YBio, and were cultured in DMEM supplemented with 10% fetal bovine serum (FBS) at 37°C with a humidity of 5% CO₂.

Animal model

Male C57BL/6 mice (6 weeks old, immunocompetent) were purchased from Zhejiang Center of Laboratory Animals (ZJCLA) and housed in the ZJCLA. Mice were fed regular chow in standard caging and kept under a 12-hour light-dark cycle with no enrichment. All animal husbandry and experimental procedures, including animal housing and diet, were performed under the guidelines, and were approved by the Institutional Animal Care and Use Committee (IACUC) and ZJLA (ethical number: ZJCLA-IACUC-20020072). For animal model, fifty microlitres of PBS containing 2.5 × 10⁵ MC38 cells was subcutaneously injected into the right flanks of mice to establish the subcutaneous tumour-bearing model. One week later, the tumor volume reached approximately 70–100 mm³, and all tumour-bearing mice were divided randomly into 4 groups (n = 6). From the 8th day, tasquinimod (30 mg/kg) dissolved in a solution (5% DMSO and 30% PEG300 in Milli-Q water) and PBS (100 μL) was administered by oral gavage once a day for 15 days. Treatments with anti-PD-1 (5 mg/kg, i.p.) were conducted every 3 days for a total of four times. Tumor volumes were calculated according to the modified ellipsoidal formula: V = 1/2 (length × width²). In another model of macrophage clearance in vivo, tasquinimod (30 mg/kg) was administered by oral gavage once a day for 18 days, clodronate liposomes were applied for macrophage depletion in the dose of 150 μL per mouse for on the first day, followed by 100 μL per mouse every three days for a total of six times (Figure S6H).

Human subject

The information of 1399 CRC patients were collected from The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/) and Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) databases. Two in-house cohorts (n = 114 and n = 30, respectively) were constructed as the in-house validation cohorts. The mean age of the CRC patients is 64. The percentage for male is 55% and the percentage for female is 45%. They are not involved in previous procedures and test naïve. The collection of primary tumor tissue and blood samples from CRC patients was approved by the Ethics Committee of the First Affiliated Hospital, School of Medicine, Zhejiang University (ethical number: IIT0037B-R1). The CRC patient studies were conducted in accordance with International Ethical Guidelines for Biomedical Research Involving Human Subjects (CIOMS). The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Written informed consents were obtained from participants or their immediate families.

Method details

Bulk transcriptome analysis

RNA libraries were created using the TruSeq Stranded mRNA LT Sample Prep Kit (Illumina, San Diego, CA, USA) following the manufacturer’s instructions. The libraries were sequenced at Shanghai OE Biotech using the Illumina sequencing platform (HiSeq X Ten). Trimmomatic was applied to trim adapter sequences and eliminate low-quality reads (v.0.36). Then, using HISAT2 (v.2.1.0), reads were mapped to the human genome (GRCh38), and read counts for each gene were calculated with HTSeq (v.0.6.1). Cufflinks were used to determine the values for fragments per kilobase of exon model per million mapped fragments (FPKM).

Tandem mass tag (TMT)-based proteomic analysis

The CRC tissues were dewaxed and rehydrated, and then acidic hydrolysis with formic acid (FA) was performed. Proteins were denatured with 6 M urea (Sigma–Aldrich, Germany) and 2 M thiourea (Sigma–Aldrich, Germany) before digestion into peptides with trypsin (1:20; Hualishi, Beijing, China) and Lys-C (1:80; Hualishi, Beijing, China) using pressure-cycling technology (PCT).^46,47 TMTproTM 16 plex (Thermo Fisher Scientific, San Jose, USA) was used to label peptides.⁴⁸ In the TMT126 channel, each batch contained 15 experimental samples and one pooled sample for normalization. The fractions (60 per batch) were separated using offline high-pH reversed-phase fractionation with a Thermo Dionex Ultimate 3000 RSLC Nano System and then merged to produce a total of 30 fractions per batch. The fractionated samples were subsequently separated using a Thermo Dionex Ultimate 3000 RSLC Nano System before being examined with an HF mass spectrometer in data-dependent acquisition (DDA) mode (Thermo Fisher Scientific, San Jose, USA). Using Proteome Discoverer (version 2.4, Thermo Fisher Scientific, Waltham, MA), all reviewed human entries from UniProt (downloaded on 14 April 2020, containing 20,365 proteins) were searched. The detailed parameters were previously described without modification.^49,50

Liquid chromatography-mass spectrometry (LC-MS) analysis and identification of metabolic profiles

The plasma samples from 114 CRC patients were applied for LC-MS analysis. Metabolic profiling was performed using a VION IMS QTOF Mass Spectrometer (Waters Corporation, Milford, CT, United States), which was installed on an ACQUITY UPLC I-Class system (Waters Corporation, Milford, CT, United States). All samples were separated using an ACQUITY UPLC BEH C18 column (1.7 μm, 2.1 mm∗100 mm) with a 45°C temperature setting. The mobile phases were acetonitrile and 0.1% acetic acid in water (A) with a flow rate of 0.4 mL/min (B). The compound separation procedure was performed using the gradient program (time, %B) shown below: 0 min, 1% B; 1 min, 30% B; 2.5 min, 60% B; 6.5 min, 90% B; 8.5 min, 100% B; 10.7 min, 100% B; 10.8 min, 1% B and 13 min, 1% B. The electrospray capillary voltage was 2500 V, the injection voltage was 4 eV, the ion source temperature was 115°C, the desolvation gas temperature was 450°C, the desolvation gas flow was 900 L/h, the m/z range was 50–1000, the scan duration was 0.2 s, and the interscan delay was 0.02 s. The same mass spectrum parameters are used for the two modes. To normalize the peaks, Progenesis QI v2.3 (Nonlinear Dynamics, Newcastle, UK) was applied. The data were then qualified using the Human Metabolome Database (HMDB), Lipidmaps (v2.3), and METLIN software.

Molecular subtype identification

The R software package “ConsensusClusterPlus” was used to identify probable molecular subtypes of CRC utilizing the transcriptome and proteomic data with default parameters.⁵¹ The CMS on the integrated training cohort was performed with the “CMScaller” package.³⁹ The gene‒metabolite network was constructed with the preconstructed network from this study.⁵²

Single-sample gene set enrichment analysis (ssGSEA), FGSEA and GO analysis

The gene sets downloaded from The Broad Institute (https://www.gsea-msigdb.org/gsea/msigdb/) were used to analyze the enrichment scores of each sample using ssGSEA. For this investigation, the “GSVA” package in R software was used with the default parameters.⁴² The FGSEA was performed with the “fgsea” package with the default parameters. Data from the Broad Institute (https://www.gsea-msigdb.org/gsea/msigdb/) were accessed to retrieve the hallmark gene sets. Bindea G et al. provided all the gene sets of immune cells.⁴⁵ The "Limma" package was used with R software to perform differentially expressed gene (DEG) and differently expressed protein (DEP) analyses using default parameters.⁴⁴ To estimate the fold-change between each pair of molecular subtypes, an empirical Bayesian method was applied with moderated t tests. The Benjamin-Hochberg adjustment was used to generate the modified p values for multiple tests. The "clusterProfiler" package in R software was used to perform GO analysis based on the DEGs and DEPs.⁴¹

scRNA-seq data processing in human CRC tissues

The human CRC tumor tissues were transported in RPMI 1640 (Gibco, Cat. no. 11875–093, US) with 1 mM protease inhibitor (Solarbio, Cat. no. P6730, China). Tissues were digested for 40 min at 37°C with a dissociation enzyme cocktail prepared by dissolving 2 mg/mL Dispase II (Sigma–Aldrich, Cat.42613-33-2 US), 1 mg/mL Type VIII collagenase (Sigma–Aldrich, Cat. no. C2139, US), and 1 unit/mL DNase I (NEB, Cat. no. M0303S, US) in PBS with 5% fetal bovine serum (FBS; Gibco, Cat. no. 16000–044, US). The cells were manually dissociated and pipetted, collected every 20 min and filtered through a 40-μm nylon cell strainer (Falcon, Cat. no. 352340, US). To eliminate red blood cells, red blood cell lysis buffer (Invitrogen, Cat. US) with 1 unit/mL DNase I was employed. Finally, the cells were rinsed in PBS containing 0.04% BSA (BSA; Sigma–Aldrich, Cat. no. B2064, US). Countess (Thermo) was used to calculate the concentrations of the single-cell suspensions, which were then corrected to 1000 cells/L. To capture 5,000–10,000 cells per chip site, cells were loaded following the Chromium single-cell 3′ kit standard protocol. Library construction and all other processes followed the manufacturer’s standard protocol.

RNA isolation and cDNA library preparation were conducted in accordance with the manufacturer’s instructions. More than ten thousand cells were captured using a restricted dilution method. The surface of supersaturated beads was covered in oligonucleotide barcodes, allowing them to be matched to cells in a microwell. The beads were hybridized by polyadenylated RNA molecules after the cells were lysed in cell lysis buffer. Finally, the barcoded cells and gene expression counts were combined into a gene-barcode matrix. Then, the RNA molecules were reverse transcribed to obtain cDNA. During cDNA synthesis, a special cell label fragment was added to each cDNA to carry information about the cell of origin. The BD Rhapsody platform was used to create scRNA-seq libraries, which were then sequenced on the Illumina Novaseq 6000. FASTQs were aligned to the human reference genome (GRCh38) with the default parameters of the STAR software.⁵³ Gene-barcode matrices were created for each sample by counting the unique molecular identifiers (UMIs) and filtering noncell-associated barcodes. Finally, the barcoded cells and the gene expression counts were combined into a gene-barcode matrix.

Analyses of single-cell transcriptomes were performed using the Seurat package of R software with the default parameters.⁵⁴ The t-distributed stochastic neighbor embedding (t-SNE) method was used to conduct nonlinear dimensional reduction using R software. The highly-expressed genes of each cluster (adjusted-P < 0.001) were identified using “Seurat” package. The “GSVA” package from R software was used to determine the score of each gene set as previously described above. Cell‒cell interaction (CCI) analysis was performed by exploring the ligand and target gene pairs using the “cellchat” package⁴³ from R software.

Haematoxylin and eosin (H&E) staining of human CRC tissue sections

CRC tumor tissue sections (4 μm) were deparaffinized with xylene and subsequently rehydrated with graded alcohol. Tissue sections were rinsed thrice with PBS before being stained with haematoxylin for 30 min at room temperature and washed with PBS three times. After that, the sections were immersed in ammonia water to change the haematoxylin-stained nuclei from reddish to blue–purple. Subsequently, the tissue slices were rinsed with 75% alcohol for 2 min at room temperature. Eosin was used to stain the cytoplasm for 1 h at room temperature. Finally, xylene was used to displace the anhydrous alcohol before mounting with slides. Sections were examined using light microscopy (Leica), and images were analyzed using Image-Pro Plus software (version 6.0).

Immunohistochemistry (IHC) staining of human CRC tissue sections

Human CRC tumor tissues from the training cohort were fixed with 4% paraformaldehyde before being embedded in paraffin. Tissue sections (4 μm) were later deparaffinized with xylene and rehydrated with graded alcohol. Microwave heating was used to promote antigen epitope retrieval. Tissue slices were then blocked for 1 h at room temperature in goat serum blocking solution (Proteintech, B900780, China) and phosphate buffered saline (PBS). The sections were incubated overnight at 4°C with a primary antibody against CD68 (1:200; 76437S, CST). The next day, CRC tissue samples were stained according to the manufacturer’s instructions using an anti-mouse/rabbit universal immunohistochemical detection kit (pk10006, Proteintech) following the manufacturer’s instructions. Light microscopy (Leica) was used to examine mounted sections, and images were processed with Image-Pro Plus (version 6.0).

Multiplex immunofluorescence (mIF) staining of human CRC tissue sections

Before embedding in paraffin, resected tumor tissues from CRC patients were fixed with 4% paraformaldehyde. The prepared tissue sections (4 μm) were first baked for 1 h at 65°C in an oven, dewaxed with xylene and rehydrated with graded alcohol. Following rehydration, the sections were fixed for 20 min at room temperature in 10% neutral buffered formalin (NBF) (Solarbio, G2161, China), placed in appropriate AR buffer and microwaved for 1 min at 100% power followed by an additional 15 min at 20% power. The slides were blocked after the slides were cooled at room temperature and incubated with primary antibody at room temperature for 10 min. The slides were washed thrice with TBST (Solarbio, T1082, China) and then incubated with Opal Polymer HRP Ms + Rb (AKOYA Biosciences, USA, NEL820001KT) at room temperature for 10 min. Before incubation with Opal Signal Generation, the slides were rinsed with TBST thrice. Microwave treatment, blocking, primary antibody incubation, introduction of Opal Polymer HRP, and signal amplification were repeated. After labeling targets (CD163, 1:300, ab182422; CD68, 1:400, BX50031; PD-L1, 1:400, 13684S; HLA-DR, 1:1000, ab92511; PanCK, 1x, GM351507), DAPI working solution was applied in the dark for 5 min at room temperature, and the slides were cleaned with distilled water and TBST before being mounted. Finally, a confocal microscope (Nikon, Japan) was used to take images of tissue samples, and the acquired images were processed using ImageJ (version 4.0).

Flow cytometry

Single-cell suspensions were prepared, and the cells were labeled with related dyes and antibodies according to the manufacturer’s instructions. Antibodies against CD45-BV785 (Clone: 30-F11; 103,149; 1:600), IFN-γ-PE-cy7 (Clone: XMG1.2; 505826; 1:600), CD206-APC (Clone: C068C2; 141708; 1:600), F4/80-PE-cy7 (Clone: BM8; 123114; 1:600), CD86-PE (Clone: A17199A; 159204; 1:600), CD3-AF700 (Clone: 17A2; 100216; 1:600), CD45-FITC (Clone: QA17A26; 157608; 1:600), F4/80- BV650 (Clone: BM8; 123149; 1:600), I-A/I-E− PE-cy7 (Clone: M5/114.15.2; 107630; 1:600), CD163-BV421 (Clone: GHI/61; 333612; 1:600), CD3-FITC (Clone: 145-2C11; 100306; 1:600), INOS-PE (Clone: W16030C; 696806; 1:600), TIM-3-APC (Clone: RMT3-23; 119706; 1:600), and CD8-BV605 (Clone: 53-6.7; 100744; 1:600) were purchased from BioLegend (USA, CA). The LIVE/DEAD Fixable Violet dead cell staining kit (1:1000), CD4-PerCP-Cy™5.5 (Clone: RM4-5; 561115; 1:600), CD8-BV510 (Clone: 53–6.7; 563068; 1:600), CD11b-Buv395 (Clone: OX-42; 743983; 1:600), CD45-BV785(Clone: HI30; 563716; 1:600), CD11b-Buv395(Clone: M1/70; 563553; 1:600), CD4-PerCP-Cy5.5(Clone: GK1.5; 552051; 1:600), and TCF-7-PE (Clone: S33-966; 564217; 1:600) were purchased from BD Bioscience (USA, CA). Antibodies against Arg1-APC (A1exF5; 17-3697-82; 1:600) were purchased from Thermo Fisher Scientific (USA). For the Arg1, CD206 and TCF-7 staining, cells were fixed and permeabilized (Fixable Viability Stain 700, BD Biosciences, Cat#564997). Flow cytometry analysis was performed on a BD Fortessa 5 laser flow cytometer (BD Bioscience).

Quantification and statistical analysis

All statistical analyses in this study were performed with R software (version 4.04). Survival analysis was performed using the “survival” package.⁵⁵ Univariate Cox regression analysis was applied to determine the hazard ratio (HR). Statistical tests were selected based on the specific assumptions relative to the data distribution and its variability. A p value of less than 0.05 was considered significant. The bars represent means ± SEM ns = not significant, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001.

Published: March 28, 2023

Data and code availability

The RNA sequencing data were deposited at Gene Expression Omnibus (GEO) (GSE56699, GSE14333, GSE39582, GSE17536, GSE17537, GSE33113 and GSE37892) and The Cancer Genome Atlas (TCGA), which are publicly available. All raw data generated by this study including RNA sequencing, single-cell RNA sequencing, proteomics and metabolomics data have been deposited in the Chinese national genomics data center (https://ngdc.cncb.ac.cn), under accession number HRA003444, HRA003501, NGDC: OMIX002456 and NGDC: OMIX002340. The software and algorithms for data analyses used in this study are published and referenced throughout the STAR Methods section. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.