Single-cell atlases link macrophages and CD8+ T-cell subpopulations to disease progression and immunotherapy response in urothelial carcinoma

Abstract

Rationale: Immune checkpoint inhibitors (ICIs) have revolutionized the management of locally advanced or metastatic urothelial carcinoma. Strikingly, compared to urothelial carcinoma of the bladder (UCB), upper tract urothelial carcinoma (UTUC) has a higher response rate to ICIs. The stratification of patients most likely to benefit from ICI therapy remains a major clinical challenge. Methods: In this study, we performed the first single-cell RNA sequencing (scRNA-seq) study of 13 surgical tissue specimens from 12 patients with UTUC. The key results were validated by the analysis of two independent cohorts with bulk RNA-seq data for UCB (n = 404) and UTUC (n = 158) and one cohort of patients with metastatic urothelial carcinoma (mUC) who were treated with atezolizumab (n = 348). Results: Using scRNA-seq, we observed a higher proportion of tumor-infiltrating immune cells in locally advanced UTUC. Similar prognostically relevant intrinsic basal and luminal-like epithelial subtypes were found in both UTUC and UCB, although UTUC is predominantly of the luminal subtype. We also discovered that immunosuppressive macrophages and exhausted T-cell subpopulations were enriched in the basal subtype and showed enhanced interactions. Furthermore, we developed a gene expression signature (Macro-C3 score) capturing the immunosuppressive macrophages that better predicts outcomes than the currently established subtypes. We also developed a computational method to model immune evasion, and the Macro-C3 score predicted therapeutic response of mUC treated with first-line anti-PD-L1 inhibitors in patients with lower basal scores. Conclusions: Overall, the distinct microenvironment and Macro-C3 score provide an explanation for ICI efficacy in urothelial carcinoma and reveal new candidate regulators of immune evasion, suggesting potential therapeutic targets for improving antitumor immunity in the basal subtype.

Keywords: Immunosuppression; Immunotherapy; Single-cell RNA sequencing; Tumor microenvironment; Urothelial carcinoma.

Figure 1

Single-cell RNA sequencing reveals intertumor heterogeneity of UTUC. (A) A schematic description of the overall experimental design and data exploration workflow. (B) Clinical features of the patients and the composition of cells in the tumor samples. The samples are ordered according to the T category. LVI, lymphovascular invasion. (C and D) UMAP of 67,392 cells post-QC and filtering grouped by the sample (C) and major cell type (D). (E) Violin plots showing the distribution of expression levels of canonical cell type markers. (F) Fractions of epithelial and immune cells in each of the NMI and MI samples. (G) Fractions of epithelial and immune cells in each of the NMI and MI samples in the Japanese cohort. In (F) and (G), error bars represent the mean ± standard error of the mean. P values were calculated with a t test.

Figure 2

Classification of MI UTUC into luminal and basal subtypes based on BASE47 gene expression. (A) Clustering analysis of the MI UTUC samples based on the average expression levels of the BASE47 gene in epithelial cells. The dendrogram is separated into two groups to reflect the distinct expression patterns of the luminal and basal subtypes. (B) Representative IHC images of luminal and basal UTUC tissue sections stained for SPINK1. Scale bars correspond to 200 µm. (C) Forest plots showing the hazard ratios associated with the luminal score, basal score, and clinical information in univariate Cox proportional hazard models for DSS in the Japanese UTUC cohort. (D) UMAP of epithelial cells in luminal and basal UTUC. (E) Heatmap showing DEGs between the luminal and basal subtypes based on the average gene expression in epithelial cells. Representative genes are indicated (left). (F) Bar plot displaying the statistically significant overrepresentation of GO BP terms in DEGs defined by the UTUC subtypes. (G) Violin plots displaying the evasion, myeloid migration, and macrophage chemotaxis scores of tumor cells in the luminal and basal subtypes. The P value was calculated by the Wilcoxon test.

Figure 3

Accumulation of an immunosuppressive macrophage population in the basal subtype. (A) UMAP projections of 6,292 subclustered myeloid cells. (B) Heatmap of the scaled normalized expression of subcluster-defining genes as determined by the “MAST” method. (C) Violin plots showing the expression of M1, M2, and TAM markers in the macrophage populations. (D) Relative percentage of Macro-C3 cells (immunosuppressive macrophages) among the NMI, luminal, and basal subtypes. P3 was excluded due to the presence of <100 myeloid cells. The relative percentage of Macro-C3 cells is defined as the proportion of Macro-C3 in myeloid populations. Error bars represent the mean ± standard error of the mean. P values were calculated with a t test. (E) GSEA revealed the enrichment of the GO BP term positive response to tolerance induction in the Macro-C3 population compared with the other macrophage populations. NES, normalized enrichment score. (F) Dot plot of genes defining the Macro-C3 signature in macrophage subclusters. Colors indicate the scaled mean expression of a gene. The size of the circles indicates the percentage of cells expressing the gene. (G) The Macro-C3 signature score for the NMI, luminal, and basal subtypes in the Japanese UTUC cohort. P values were calculated with a t test. (H) Kaplan-Meier plot of the relationship of patient DSS and PFS with the Macro-C3 signature score for the Japanese UTUC dataset. Patients were stratified by score quartile, in which Q1 and Q4 had the lowest and highest scores, respectively. P values were calculated using the log-rank test.

Figure 4

Expansion of exhausted T cells in the basal subtype. (A) UMAP projections of 20,862 subclustered T cells. (B) Heatmap indicating the expression of selected functionally relevant genes in the T/NK subtypes. (C) Cumulative distribution function showing the distribution of naïve, cytotoxic, and exhausted state scores in each T/NK subtype. A rightward shift of the curve indicates increased state scores. (D) Differences in the composition of the T/NK population among subtypes. Fractions are visualized as cell density based on the UMAP embedding. (E) Relative percentage of CD8-C1 cells (exhausted CD8⁺ T cells) among the NMI, luminal, and basal subtypes. P10 was excluded due to the presence of <100 T/NK cells. The relative percentage of CD8-C1 cells is defined as the proportion of CD8-C1 cells in the T/NK populations. Error bars represent mean ± standard error of the mean. P values were calculated with a t test. (F) Kaplan-Meier curves of OS and progression free interval (PFI) with the decomposed CD8-C1 (exhausted CD8⁺ T cells) proportions in the TCGA-BLCA cohort, and DSS in the Japanese UTUC cohort. Samples were categorized as high (red, top 50%) or low (blue, bottom 50%) CD8-C1 abundance. P values were calculated using the log-rank test.

Figure 5

Recruitment and exhaustion of CD8⁺ T cells by TAMs define the ecosystem of basal UTUC. (A) Heatmaps showing the number of statistically significant ligand-receptor pairs between myeloid lineage cells and other cell types. The bar plots (top) summarize the overall number of interactions with myeloid cells for the indicated cell types in the UTUC subtypes. B/plasma cells were re-subclustered using a similar approach applied in T cell and myeloid cell subclustering. (B) Summary of selected ligand-receptor interactions between the tumor cells, CD8⁺ T cells, and macrophages. Circle size indicates the P value (permutation test). The color gradient represents the interaction strength. (C) Scatterplot showing the correlation between the expression of CXCL10 in macrophages and the proportion of CD8⁺ T cells in the scRNA-seq dataset. (D) Representative immunofluorescence images illustrating the interaction between CD8⁺ T cells and macrophages in one UTUC sample (P4). The small panels show the magnification of the selected region highlighted in red. Yellow arrows indicate the colocalization of CD8⁺ T cells and macrophages. Scale bars correspond to 50 µm and 5 µm in the large and small panels, respectively. (E and F) Summary of selected ligand-receptor interactions between macrophages and CD8-C1 cells (exhausted CD8⁺ T cells) in the NMI, luminal (E), and basal subtypes (F). (G) Scatterplot showing the correlation between the relative ratio of Macro-C3 (immunosuppressive macrophages) and CD8-C1 cells (exhausted CD8⁺ T cells) in the scRNA-seq dataset. (H) Scatterplot showing the correlation between the signature score of Macro-C3 (immunosuppressive macrophages) and the overall exhaustion score in the Japanese UTUC cohort. Samples are colored according to subtype. (I) Schematic showing the crosstalk among CD8⁺ T cells, macrophages, and tumor-derived epithelial cells involved in the recruitment of immune cells and the formation of an immunosuppressive microenvironment in the basal subtype. In (C), (G), and (H), the Pearson coefficient (R) and associated P value are reported.

Figure 6

The Macro-C3 score predicts the immunotherapy response in mUC. (A) Bar plots showing the classification of subtypes associated with the immune phenotype in the IMvigor210 cohort. Sample sizes are given in the figure. (B) Representative CD8 staining shows the infiltration of CD8⁺ T cells (brown) in basal (P5) samples. Scale bars correspond to 1 mm, 500 µm, and 100 µm in the large and small panels, respectively. (C) PD-L1 on immune cells is associated with response in the NMI subtype (two-sided Fisher's exact test, NMI: p = 0.0009046), while the Macro-C3 score is associated with response in the NMI and luminal subtypes (Q1 versus Q4, NMI: p = 0.01466, luminal: p = 0.01806, basal: p = 0.3308) in the IMvigor210 cohort. Sample sizes are given in the figure. Patients were stratified by IC level or score quartile, wherein IC level stands for PD-L1 expression on tumor-infiltrating immune cells assessed by SP142 immunohistochemistry assay and scored as level 0 (< 1%), level 1 (≥ 1% and < 5%), or level 2+ (≥ 5%). Q1 and Q4 had the lowest and highest Macro-C3 scores, respectively. (D) Scatterplot showing the correlation of the interactions between macrophages and CD8⁺ T cells with the percentage of CD8-C1 cells (exhausted CD8⁺ T cells) in T/NK populations. The statistically significant interactions are highlighted in red and blue indicating positive and negative correlations, respectively. Top 10 positive or negative interactions are labeled. (E) Heatmap indicating the scaled expression of selected genes from different categories in the IMvigor210 cohort. The samples were ordered by log(basal/luminal) score. (F) Linear regression of the z score of overall cytotoxicity or Macro-C3 score on log(basal/luminal) score. The regression equations and the intersection point produced by the two regression lines are indicated in the figure. (G) Boxplot showing Macro-C3 signature score was significantly associated with response to ICIs for patients with log(basal/luminal) scores below the threshold (0.294) in the IMvigor 210 cohort. (H) ROC curves assessing the performance of the TIDE, Macro-C3 signature and ICI response scores in predicting the ICI response in the IMvigor 210 cohort. In (G), the center line in the boxplots indicates the median, the lower and upper hinges correspond to the first and third quartiles, and the whiskers extend at most 1.5 times the interquartile range past the upper and lower quartiles. The P value indicates a two-sided two group t test without adjustment.