Tumor-infiltrating T-cell subsets show a similar distribution across various cancer histological types

We conducted flow cytometry phenotyping of T-cells in freshly resected human tumors from 72 cancer patients eligible for surgery (Fig. 1A). The cohort included a range of cancer types, including Metastatic Melanoma (MM; n = 7), Non-Small Cell Lung Carcinoma (NSCLC; n = 7), Renal Cell Carcinoma (RCC; n = 12), Head and Neck Squamous Cell Carcinoma (HNSCC; n = 10), Epithelial Ovarian Carcinoma (EOC; n = 21), Urothelial Carcinoma (UC; n = 12), Hepato-Cellular Carcinoma (HCC; n = 1), Neuro-Endocrine Tumor (NET; n = 1), and Thyroid Carcinoma (Thyr; n = 1). Most patients (71%) had not received systemic treatment before surgery (see patients characteristics in Supplementary Data 3). The percentage of immune cells expressing CD45 ranged from 0.7% to 95.7% among live cells (median 28.1%), with no significant difference observed across histological types (Fig. 1B). CD3 + T-cells varied across patients independently of tumor types, ranging from 10.8% to 88.1% (median 53.2%) among live CD45 + cells (Fig. 1C). The levels of CD8 + , CD4 + FoxP3-, and CD4 + FoxP3 + T-cell subsets varied between patients and among cancer histological subtypes (Fig. 1D-F). Overall, CD8 + , CD4 + FoxP3-, and CD4 + FoxP3 + T-cell levels were similar across cancer histological types, except for RCC, which showed significantly fewer CD4 + FoxP3 + T-cells than HNSCC (Fig. 1F). CD8 + T-cells (median 36.5%, range [1.9–87.3]) and CD4 + FoxP3- T-cells (38.6% [6.6–74.2]) were more abundant in the tumor microenvironment than CD4 + FoxP3 + T-cells (9.5% [0.1–40.6]), except for MM where no significant differences were observed between CD8 + and CD4 + FoxP3 + T-cells (Fig. 2A).

Fig. 1

Proportions of T-cell subsets in the tumor microenvironment are independent of tumor histological types. A Freshly resected tumors from various histologies (n = 72) were collected and dissociated into a cell suspension and stained for T cell subset identification. Immune checkpoints (ICPs) expression was assessed in CD8+, CD4+FoxP3− and CD4+FoxP3+ T cells at the protein level using flow cytometry (n = 35) and at the transcriptomic level using single-cell RNA sequencing, including TCR sequencing (n = 5). Created with BioRender.com. Flow cytometry analysis from 72 fresh tumor specimens. B Percentage of CD45+ among live cells in the different histologies. C Percentage of CD3+ T-cells among CD45+ cells according to the different histologies. Percentage of CD8+ (D), CD4+FoxP3− (E) and CD4+FoxP3+ (F) among CD3+ cells in the different histologies. The red dotted line delineates the median of the whole cohort. Dunn’s multiple comparison test, *p value ≤ 0.05; **p value ≤ 0.01; ***p value ≤ 0.001; ****p value ≤ 0.0001. MM: Metastatic Melanoma; NSCLC: Non-Small Cell Lung Carcinoma; RCC: Renal Cell Carcinoma; HNSCC: Head and Neck Squamous Cell Carcinoma; EOC: Epithelial Ovarian Cancer; UC: Urothelial Carcinoma

Fig. 2

Intratumoral CD4+FoxP3+ cells make up a small subset but display the highest levels of immune checkpoint protein expression. A Proportions of T cell subsets, i.e., CD8+, CD4+FoxP3− and CD4+FoxP3+ in MM, NSCLC, RCC, HNSCC, EOC, and UC obtained by flow cytometry analysis of 72 freshly resected tumor specimens. B Percentage of immune checkpoint protein (ICP) positive cells in CD8+, CD4+FoxP3− and CD4 + FoxP3 + T cells from 35 tumor specimens. C Mean fluorescence intensity of ICPs in CD8+, CD4+FoxP3− and CD4+FoxP3+ T cells from 35 tumor specimens. Dunn’s multiple comparison test was performed independently for each ICP. D Heat map displaying the ratio of the ICP median MFI of CD4+FoxP3+ cells over CD4+FoxP3.−. Dunn’s multiple comparison test, *p value ≤ 0.05; **p value ≤ 0.01; ***p value ≤ 0.001; ****p value ≤ 0.0001. MM: Metastatic Melanoma; NSCLC: Non-Small Cell Lung Carcinoma; RCC: Renal Cell Carcinoma; HNSCC: Head and Neck Squamous Cell Carcinoma; EOC: Epithelial Ovarian Cancer; UC: Urothelial Carcinoma; ICPs: Immune Checkpoints: (CD25, CD28, CD39, 4-1BB, CTLA-4, ICOS, OX40, PD-1, PD-L1, and TIGIT)

Immune checkpoint protein expression is predominant on FoxP3+ CD4+ T-cells

Our primary analysis centred on the surface expression of 10 immune checkpoints (CD25, CD28, CD39, 4-1BB, CTLA-4, ICOS, OX40, PD-1, PD-L1, and TIGIT) on CD8 + , CD4 + FoxP3-, and CD4 + FoxP3 + T-cell subsets in 35 tumor specimens. Inter-individual variability was found for immune checkpoint proteins (ICPs), independently of cancer histological type. However, the ICP expression profile was homogeneous within T-cell subsets. CD4 + FoxP3 + T-cells were significantly more positive for ICPs than both CD4 + FoxP3- and CD8 + T-cells, with the exception of PD-1, which was more prevalent on CD4 + FoxP3 + than CD4 + FoxP3- T-cells, but not more than CD8 + cells (Fig. 2B). The ICPs CD25, CD28, CTLA-4, ICOS and OX40 were more frequently found on CD4 + FoxP3- than on CD8 + T-cells, although to a lesser extent than CD4 + FoxP3 + T-cells (Fig. 2B). Similar results were obtained by analyzing the level of ICP expression using mean fluorescence intensities (MFI), which correlates with protein abundance on cell membranes (Fig. 2C and Supplementary Data 4). CD39 and CD25 were the most differentially expressed ICPs between CD4 + FoxP3 + and CD4 + FoxP3- T-cell subsets (Fig. 2D). LAG3 was not part of our initial flow cytometry panel, but meanwhile, an anti-LAG3 checkpoint antibody (relatlimab) has been FDA/EMA approved. Therefore we decided to assess LAG3 expression in 3 additional tumor specimens (one HNSCC, one Hodgkin and one NSCLC). Overall we found a poor expression (< 4%) of LAG3 on T-cells membranes with a trend toward more expression on Tregs (median percentage expression: 0.87% of LAG3 on CD8 + , 0.49% on CD4 + Foxp3- and 3.8% on CD4 + Foxp3 +) (Supplementary Data 5).

Although the pattern of ICP expression was not associated with a specific tumor histology, we observed some trends. For example, in HNSCC, more CD8 + and CD4 + FoxP3- T-cells were CD25-positive than in RCC, and in NSCLC, more CD4 + FoxP3 + and CD4 + FoxP3- T-cells were CTLA-4-positive than in RCC (Supplementary Data 6).

Immune checkpoint co-expression is predominant on FoxP3+ CD4+ T-cells

We next analyzed our ICP phenotyping using the PhenoGraph algorithm [20] to characterize tumor-infiltrating T-cell subpopulations according to ICP co-expression. Nine clusters of CD8 + T-cells, seven clusters for each CD4 + FoxP3- and CD4 + FoxP3 + T-cells, one cluster of CD4 + CD8 + and one cluster of CD4-CD8- T-cells were identified (Fig. 3A). CD4 + FoxP3 + T-cell clusters exhibited patterns of multiple ICP co-expressions compared to CD8 + and CD4 + FoxP3- T-cell clusters (Fig. 3B). The distribution of clusters within the T-cell subsets were heterogeneous (Fig. 3C). In CD8 + T-cells, the most abundant cluster (#4) represented an average of 24.4% [1.9–48.1] and expressed none of the tested ICPs except for CD28, while the cluster expressing the most ICPs (#17) represented 19.9% [0.2–84.4] of CD8 + T-cells. In CD4 + FoxP3- T-cells, the most abundant cluster (#20) representing 29.5% [3.5–61.8], expressed none of the tested ICPs, while the most activated cluster (#14) accounted for only 12.9% [0.2–75.0] of CD4 + FoxP3- T-cells. In CD4 + FoxP3 + T-cells, the most abundant cluster (#25, 43.9% [8.2–88.0]) co-expressed all tested ICPs. We found a large interpatient variability, with none of the clusters showing a particular association with any histopathological tumor type (Fig. 3D). ICPs were more highly expressed in CD4 + FoxP3 + clusters, with cluster #25 (co-expressing all 9 ICPs), being the predominant CD4 + FoxP3 + cluster in 58.8% of tumors (Supplementary Data 7A and B, Fig. 3C and D). Interestingly, clusters #25 and #19 (CD4 + FoxP3 +) as well as clusters #5 and #22 (CD8 +) were significantly more abundant in HNSCC compared to RCC (Supplementary Data 8A). Comparison between tumors that were primary resected and those that were resected at relapse revealed a lower frequency of clusters co-expressing CD25, CD39, CTLA-4, OX40, and TIGIT (#6, #19) (Supplementary Data 8B). A classical supervised analysis, examining the proportion of cells double-positive for CD25 + and another checkpoint within CD4 + FoxP3 + cells (Supplementary Data 9A), or the proportion of FoxP3 + cells within CD4+ T-cells double-positive for CD25 + and another checkpoint (Supplementary Data 9B) confirmed the higher ICP co-expression in CD4 + FoxP3 + T-cells compared to CD4 + FoxP3- T-cells. Overall, this analysis showed that tumor infiltrating CD4 + FoxP3 + T-cells co-expressed most of our ICPs of interest.

Fig. 3

Unsupervised clustering of T-cell subsets according to the level of membrane protein expression. Unsupervised clustering analysis of flow cytometric dataset using PhenoGraph algorithm (n = 34). A UMAP displaying the 25 clusters defined based on the fluorescence intensity of each marker tested, including ICPs. B Heatmap showing the protein expression patterns in each cluster. Fluorescence intensity of each marker has been normalized independently. C Pie charts representing the relative abundance (mean) of each cluster in the whole cohort, in CD8+, CD4+FoxP3− and CD4+FoxP3.+ T cells (left panels); stacked bar chart displaying the relative abundance of each cluster in each tumor specimen in the 3 T-cell subsets (right panels). ICPs: immune checkpoints: (CD25, CD28, CD39, 4-1BB, CTLA-4, ICOS, OX40, PD-1, PD-L1, and TIGIT)

We then attempted to determine whether the phenotypic profile of T-cells was associated with disease outcomes and prognostic factors such as lymph node metastasis, neutrophil-to-lymphocyte ratio (NLR), or derived neutrophil-to-lymphocyte ratio (dNLR), but no significant impact was found. However, a trend toward better overall survival was evident with a higher CD8 + /FoxP3 + ratio (Supplementary Data 10). It’s worth noting that this analysis was limited by the size of our cohort and the differences in prognosis across the different cancer histotypes.

Intratumoral T-cell assessment using scRNA-Seq

To understand the ICP expressions in T-cell subsets and their relationship with intratumoral clonality, we analyzed transcriptomes, including TCR sequences, in enriched CD45 + cells from five freshly resected tumors (2 NSCLCs, 1 HCC, 1 HNSCC, and 1 EOC) using droplet-based single-cell RNA sequencing (scRNA-Seq, 10X Genomics). We obtained the transcriptomic profile of 28,555 CD45 + cells, including 24,105 T-cells. After sequence alignment and quality control, we performed dimensional reduction analysis using UMAP (Fig. 4A) and assigned identified clusters to specific T-cell lineages according to the expression of canonical gene markers (Fig. 4B, Supplementary Data 11) [34,35,36]. We identified 9 clusters of CD8 + , 3 clusters of CD4 + FoxP3- and 1 cluster of CD4 + FoxP3 + T-cells (Fig. 4A and B).

Fig. 4

Unsupervised clustering of T-cell subsets according to the level of intracellular gene expression. Single-cell RNA sequencing of five fresh tumor specimens. A UMAP displaying clusters defined based on their gene expression profile. Created with Cerebro (R-studio©). The list of genes expressed by each cluster is provided in Supplementary Data 11. B Stacked violin plot displaying the expression distribution of selected cell markers in the T cell clusters. C Pie chart showing the relative abundance (mean) of each cluster in the whole cohort. D Stacked bar chart showing the relative abundance of each cluster in each tumor specimen. E Volcano plot displaying differential gene expression between CD4+FOXP3− vs CD8+ T-cells (upper left panel); CD4+FOXP3+ vs CD4+FOXP3−T cells (bottom panel) and CD4+FOXP3+ vs CD8.+ T cells (upper right panel). Genes are plotted as log2 fold change versus the − log10 of the adjusted p value. Genes in red are significantly differentially expressed with a fold change > 1.5 compared to the reference population. F. Stacked violin plot displaying the expression distribution of ICPs in the T cell clusters. ICPs: immune checkpoints: (CD25, CD28, CD39, 4-1BB, CTLA-4, ICOS, OX40, PD-1, PD-L1, and TIGIT)

The most abundant CD8 + clusters were effector memory CD8 + T-cells (#1: 17.5% [3.6–40.7]) as defined by the predominant co-expression of IL7R, STAT4 and GZMK, and exhausted CD8 + T-cells (#3: 14.6% [0.7–40.9]) as defined by the predominant co-expression of PD1, CTLA4, TIGIT, HAVCR2 (TIM3) and LAG3 (see Supplementary Data 11). Other CD8 + clusters represented less than 10% each of the CD8 + cell lineage, such as cytotoxic CD8 + T-cells (#4: 8.5% [0.2–58.5]), CD8 + cytotoxic resident T-cells (#9: 3.5% [0.1–26.9]), exhausted TNFRSF18 + (GITR)/LAG3 + CD8 + T-cells (#11: 2.5% [0.1–7.4]), CD8 + MAIT (#14: 1.9% [0.0–6.7]), cytotoxic GNLY + CD8 + T-cells (#15: 1.9% [0.4–4.0]), and proliferating CD8 + T-cells (#21: 0.7% [0.1–1.9]). Cluster #7 consisted mainly of γδ T-cells or natural killer T-cells (NKT) cells (4.0% [0.9–5.3]) (Fig. 4C).

The most abundant CD4 + clusters were effector memory T-cells (#0: 18.6% [7.0–39.2]) as defined by the predominant expression of IL7R and regulatory T-cells (#2: 16.7% [6.0–45.9]) as defined by the predominant expression of FOXP3. The two other clusters of CD4 + cell lineage included follicular helper T-cells (TFH) and effector memory re-expressing CD45RA (TEMRA) (#5: 5.7% [2.2–20.8]) and stem cell like memory CD4 + T-cells (#8: 3.9% [0.1–14.9]) (Fig. 4C).

High heterogeneity was observed across tumors, as noted previously with flow cytometry staining (Fig. 4D). While some clusters were highly prevalent in all samples (> 5% in at least 4/5 samples), such as clusters #0, #1, #2, #5; #7; #15, others were over-represented in one tumor (Fig. 4D).

Next, we compared ICP gene expression between CD8 + , CD4 + FOXP3-, and CD4 + FOXP3 + clusters (Fig. 4E). Significantly differentially expressed ICPs were determined with an average Log2 fold-change (avg_log2FC) of > 0.58 or < -0.58 and a corresponding adjusted p-value < 0.05. IL2RA (CD25), ENTPD1 (CD39), CTLA4, ICOS, TNFRSF4 (OX40), and TIGIT expression was significantly higher in CD4 + FOXP3 + T-cells compared to both CD8 + and CD4 + FOXP3- T-cells. CD28 was more expressed by both subsets of CD4 + T-cells compared to CD8 + T-cells. PDCD1 (PD-1) was less expressed by both subsets of CD4 + T-cells compared to CD8 + T-cells. CD274 (PD-L1) mRNA was not detected in any T-cell subset. CD8 + T-cell subsets over-represented in one tumor expressed higher levels of ICPs (#3, #4, #9, #11, #21) (Fig. 4F). CD4 + cluster #5 had similar levels of CD28, ENTPD1 (CD39), CTLA4, ICOS, and CD274 (PD-L1) expression as CD4 + FOXP3 + cluster #2. Notably, cluster #2, the only cluster co-expressing CD4 and FOXP3 (Fig. 4B), co-expressed all ICP genes except PD-1 and PD-L1 (Fig. 4F).

High expression of ICPs in clonally expanded CD4 +FOXP3 + T-cells

We investigated the TCRαβ repertoire in five tumor specimens using scRNA-Seq. Excluding cluster #7 composed mainly of NKT-cells and γδ T-cells, 84.0% of cells assigned as T-cells had productive TCR sequences. The percentage ranged from 55.3% in cluster #14 to 97.3% in cluster #9 (Fig. 5A). We assessed TCR diversity, i.e., the number of unique TCRs, detecting 9309 different clonotypes in T-cell clusters. CD4 + T-cell clusters showed higher TCR diversity with 6993 clonotypes distributed among 10,833 CD4 + cells, compared to CD8 + clusters with 2524 clonotypes among 13,272 CD8 + cells (Fig. 5A). The highest diversity was found in cluster #0 (Effector Memory CD4 + T-cells), #1 (Effector Memory CD8 + T-cells), and #2 (Regulatory CD4 + T-cells).

Fig. 5

Immune checkpoints are more expressed by expanded clonotypes of intratumoral CD4+Foxp3+ T-cells. TCR repertoire analysis from the single-cell RNA sequencing dataset of five fresh tumor specimen. A TCR diversity showing the number of clonotypes per patient in each T cell cluster. Dunn’s multiple comparison test, *p value ≤ 0.05; **p value ≤ 0.01. B Stacked bar chart displaying the distribution of clonotype frequency in each cluster. C Sankey diagram showing clonotype sharing between clusters and according to clonality (LC ≤ 2 cells; HC > 2 cells). D Heatmap displaying differential ICP expression (median Log2 fold-change) between LC and HC T cells in each CD4+ (left panel) and CD8+ (right panel) clusters. E Stacked bar chart showing the distribution of T-cells from CD4+ clusters according to the level of ICP expression (above the median expression level (HE) or below the median expression level (LE)) and the expansion status (LC or HC). Median expression level was calculated independently for each sample and for each ICP. F Graph displaying the average number with standard deviation of ICP expressed per cells in LC and HC T cells for each cluster; Mann–Whitney test, *p value ≤ 0.05; **p value ≤ 0.01; ***p value ≤ 0.001; ****p value ≤ 0.0001.TCR: T cell receptor; LC: low clonality; HC: high clonality. ICPs: immune checkpoints: (CD25, CD28, CD39, 4-1BB, CTLA-4, ICOS, OX40, PD-1, PD-L1, and TIGIT)

TCR clonality analysis, which measures the number of T-cells expressing the same TCR or “clones”, revealed higher representation of expanded T-cell clones (≥ 3 cells expressing the same TCR subsequently to T-cell proliferation) in CD8 + clusters, with 72.2% of T-cells showing expanded clones, compared to 23.7% in CD4 + clusters (Fig. 5B). The most expanded clones, consisting of more than 100 T-cells expressing the same TCR, were found in CD8 + clusters (#1, #3, #4, #9, #11), while the highest rate of expanded clones among CD4 + clusters were observed in the CD4 + FOXP3 + cluster (#2). Although no TCRs were shared between patients, most likely due to MHC restrictions, 5.4% of detected clonotypes were distributed among clusters (Fig. 5C).

We then examined the relationship between ICP expression and TCR clonality. ICP gene expression was analyzed for each cluster, comparing high clonality (HC) to low clonality (LC) clonotypes. CD8 + clusters showed a wide range of ICP expression levels relative to clonality. All our checkpoints of interest, except ICOS, were expressed at higher levels in clonally expanding TCRs of CD4 + FOXP3 + (cluster #2). ENTPD1 (CD39), TNFRSF9 (4-1BB), and CTLA4 were also expressed at higher levels in expanded clonotypes from the CD4 + TFH/TEMRA (cluster #5), but not in the CD4 + Effector Memory (cluster #0). TNFRSF4 (OX40) was upregulated in clonally expanding CD4 + FOXP3 + T-cells (cluster #2), but not in other clonally expanding CD4 + T-cells (cluster #0 & #5).

We classified T-cells into four compartments based on clonality and ICP expression levels: high clonality/high ICP expression (HC_HE), high clonality/low ICP expression (HC_LE), low clonality/high ICP expression (LC_HE), and low clonality/low ICP expression (LC_LE). Once again, high clonality was defined as having 3 or more T-cells with the same TCR, and high ICP expression was determined when exceeding the median ICP expression for a given patient. The HC_LE compartment contained the largest number of cells from CD8 + clusters, while the LC_LE compartment contained the largest number from CD4 + clusters (Fig. 5E and Supplementary Data 12). In the HC_HE compartment, an average of 88.5% of cells in CD8 + clusters belonged to overrepresented clusters within a given tumor (#3, #4, #9, #11, #21 (Supplementary Data 12)), while an average of 81.4% of CD4 + T-cells belonged to CD4 + FOXP3 + (cluster #2) (Fig. 5E).

We also investigated whether high clonality was associated with ICP co-expression and compared the expanded clones to the non-expanded ones. The highest average levels of ICP co-expression were observed in cluster #2 (Fig. 5F). Clusters predominantly found in one tumor (i.e. #3, #4, #9, #11, #21) tended to display higher levels of ICP co-expression compared to clusters found in all tumors except cluster #5 (i.e. #0, #1, #15). Significantly higher ICP co-expression was evident in expanded clonotypes from clusters #2 (mean 5.7 vs. 4.8), #5 (4.5 vs. 3.9), #3 (3.0 vs. 2.4), #4 (3.7 vs. 3.0), #9 (4.6 vs. 3.5) compared to non-expanded clonotypes. Conversely, expanded clonotypes from clusters #0 (1.0 vs. 1.5) and #1 (1.0 vs. 1.3) displayed lower levels of ICP co-expression.