Deciphering immune predictors of immunotherapy response: A multiomics approach at the pan-cancer level

Abstract

Immune checkpoint blockade (ICB) therapy has transformed cancer treatment, yet many patients fail to respond. Employing single-cell multiomics, we unveil T cell dynamics influencing ICB response across 480 pan-cancer and 27 normal tissue samples. We identify four immunotherapy response-associated T cells (IRATs) linked to responsiveness or resistance and analyze their pseudotemporal patterns, regulatory mechanisms, and T cell receptor clonal expansion profiles specific to each response. Notably, transforming growth factor β1 (TGF-β1)+ CD4⁺ and Temra CD8⁺ T cells negatively correlate with therapy response, in stark contrast to the positive response associated with CXCL13+ CD4⁺ and CD8⁺ T cells. Validation with a cohort of 23 colorectal cancer (CRC) samples confirms the significant impact of TGF-β1+ CD4⁺ and CXCL13+ CD4⁺ and CD8⁺ T cells on ICB efficacy. Our study highlights the effectiveness of single-cell multiomics in pinpointing immune markers predictive of immunotherapy outcomes, providing an important resource for crafting targeted immunotherapies for successful ICB treatment across cancers.

Keywords: immunotherapy; multiomics; pan-cancer; single cell.

Graphical abstract

Figure 1

Demographic overview of the pan-cancer T cell atlas In the donut diagram, numbers in the brackets represent sample number/patient number. The atlas comprised three major cohorts, namely, the pan-cancer transcriptomics cohort, the pan-cancer immune cohort, and the CRC cohort.

Figure 2

Pan-cancer T cell landscape in the transcriptomics cohort (A) Tomographic images of a CRC responder taken before and 6 weeks after ICB treatment. (B) Uniform manifold approximation and projection (UMAP) representations of CD4⁺ and CD8⁺ T cells. (C) The proportion of cell types across cancer types in ICB-R and ICD-NR groups. Cancer types included BCC, SCC, ccRCC, CRC, HNSCC, and NSCLC. (D) Summary statistics, including FDR and fold change (FC) of CD4⁺ and CD8⁺ T cells (in log scale), for the proportion test between the ICB-R and ICB-NR groups in both CD4⁺/CD8⁺ T cells. Two-proportion z-test was used to compare between the groups. Only comparisons with FDR < 0.05 were shown. (E) Heatmap showing standardized expression of the signature markers in CD4⁺ and CD8⁺ T cells. (F) UMAP representations of the expression of signature markers of CD4⁺ and CD8⁺ T cells across all cell types. (G) Clustering of TFs based on their transcriptional activities across cancer types in ICB-R and ICB-NR groups. (H) TF signatures of the IRATs in each response group.

Figure 3

T cell landscapes of the responders and non-responders in the transcriptomics cohort (A) Pseudotime of the IRATs in the response groups. (B and C) Enriched pathways in each of the pseudotime branches for the IRATs. (D) Normalized enrichment scores (NESs) of the enriched hallmark pathways in each pseudotime branch of the IRATs in (B) and (C). (E) Cell-cell interaction frequencies of the IRATs with all CD4⁺/CD8⁺ T cells. y axis displays the log₁₀FC (represented in −log₁₀FC scale) in the number of ligands or receptors facilitating cell-cell interactions between the specified cell types (noted in the plot subtitles) and the cell types shown on the x axis. The numerator of the fold change represents cells from responders, while the denominator represents cells from non-responders.

Figure 4

Pan-cancer immune cohort demonstrated an expansion in the IRATs (A) Workflow for obtaining response-specific TCRs. (B) Fold change of TCR clonal frequency between the R (numerator) and the NR (denominator) groups. Fisher exact test was used to assess the proportional difference between the groups. (C) Violin plot displaying the comparison between the response groups of the IRATs based on t test comparison. (D) Clonal expansion mapped onto the UMAP of CD4⁺/CD8⁺ T cells. Color represented the level of clonal frequency. The small UMAP diagrams depicted the location of the IRATs (indicated by their respective cell type colors) among all cells. (E) Mean receiver operating characteristic (ROC) curve for post-bootstrapping (n = 100) prediction results. (F) Confusion matrix of the final prediction model. (G) Final response index across cancer types.

Figure 5

Phenotypic landscape of the validation in CRC cohort and survival analysis on external datasets (A) UMAP representations of the CD4/CD8 T cells found in the CRC CyTOF dataset. (B) The abundance of cell types in each response group. (C and D) Comparison of the abundance between the response groups for the IRATs. (E) Volcano plots showing Cox hazard ratio (HR) in log scale of the survival analysis based on the expression group of the signature genes in IRATs. Log (HR) > 1 indicates that there is higher risk/worse survival in the cell type expressing these signature genes, while log (HR) < 1 indicates that the cell type presents a protective measure in terms of patient survival.