Automated single-cell pathology and clustering
First, we sought to apply an automated single-cell count for immunolabeled LAG-3, TIM-3, and TIGIT to 105 primary ccRCC tumour samples (i.e., COHORT 1, Table 1). A quantitative immunohistological assessment provides a robust research platform worldwide. In this automated analysis, positively stained immune cells in tumours were distinguished together with nuclei and counted as illustrated for LAG-3 (Fig. 1a), TIM-3 (Fig. 1b), and TIGIT (Fig. 1c). The mean number of cells was 102.3 ± 16.8/mm2 for LAG-3, 473.4 ± 51.5/mm2 for TIM-3, and 6.0 ± 1.8/mm2 for TIGIT (Fig. 1d). Spearman rank correlations among continuous variables of the three new IRs were 0.063 for LAG-3 and TIM-3 (p = 0.526), 0.130 for LAG-3 and TIGIT (p = 0.186), and 0.044 for TIM-3 and TIGIT (p = 0.657) (Supplementary Fig. 1). No correlations among these IRs seemed obvious; rather, exclusive relationships were observed, allowing small fractions to overlap (Fig. 1e). Thus, we hypothesize that primary ccRCC profiles can be divided robustly with our automated platform based on the signatures of three IRs.
By applying hierarchical clustering of 105 ccRCC tumour samples based on standardized positive cell densities immunolabeled for the IRs, we successfully identified three groups with distinct IR levels (Fig. 1f). The phenotypic signatures of TIM-3 (53%), LAG-3 (35%) and TIGIT (11%) were individually dominant in independent clusters (Fig. 1g). Patients in the LAG-3 cluster had worse pathological outcomes, a higher nuclear glade and more venous invasion, than those in the remaining clusters (Table 1). A prognostic analysis of our 105 primary ccRCC tumours showed that the recurrence-free survival (Fig. 1h) and overall survival (Fig. 1i) rates were different among the three clusters. The LAG-3 cluster was associated with the worst recurrence-free survival and overall survival rates, with significant differences in patients with favourable TIGIT clusters. Among the 105 ccRCC patients, 14 (13%) received PD-1 blockade therapy with nivolumab after disease relapse. A waterfall plot of changes in target lesion size revealed a certain tendency (Fig. 1j). While patients in the TIM-3 cluster demonstrated a certain effect, patients in the TIGIT or LAG-3 cluster seemed to show resistance to nivolumab treatment.
Immunogenomic differences
Second, we sought to examine the genomic alterations underlying the signatures of the three new IRs in ccRCC since recent advances in sequencing have revealed a subset of genes that correlate with the response to anti-angiogenic therapy as well as immuno-oncology therapy (15, 16). Herein, we analysed forty-three ccRCC tumour samples from COHORT 1 comprising the LAG-3 (n = 13), TIM-3 (n = 23), and TIGIT (n = 7) clusters for alterations in 160 cancer-associated genes (Fig. 2a). The frequently altered genes across the three IR spectra (> 5%) were VHL, PBRM1, SETD2, MTOR, and ATM, indicating that the incidence of VHL mutations in our cohort is slightly less than that in previous studies (17). However, there seemed to be no specific alteration pattern among individual clusters in our population (Fig. 2b).
We next investigated the association between the three new IR clusters and the tumour immune microenvironment. In total, fifteen immunolabeled molecules, including 6 for acquired immunity, 3 for innate immunity, 4 for cancer metabolism, and 2 for cancer stroma, were assessed by automated signal segmentation in 105 ccRCC tumour samples from COHORT 1. Cell-by-cell immunohistological analysis for acquired immunity revealed that the LAG-3 cluster had significantly higher levels of CD39 than the two other clusters, revealing substantial cell exhaustion in tumour-infiltrating CD8 T cells in tumours belonging to the LAG-3 cluster (Fig. 2c) (18). The levels of CD3 and CD8 in the LAG-3 cluster were also higher than those in the TIM-3 clusters, while interestingly, the CTLA-4 level was high in the TIGIT cluster.
Then, our immunohistological analysis for innate immunity revealed that the LAG-3 cluster had significantly higher levels of CD163 than the two other clusters (Fig. 2d), demonstrating that many infiltrating tumour-associated macrophages (TAMs) in tumours belong to the LAG-3 cluster. Our assessments also extended to the state of tumour metabolism in ccRCC samples labelled for the proliferative Ki-67, IDO-1, GLUT-1, and CD73 markers, but no difference was noted in inhibitory tumour metabolism among the three groups (Fig. 2e). Furthermore, two molecules associated with the cancer stroma, i.e., CD34 (to label blood vessels) and D2-40 (to label lymph vessels), were assessed, and we found that tumours belonging to the LAG-3 cluster were less dependent on the development of angiogenic blood and lymph vessels than the other clusters (Fig. 2f).
In summary, the immunosuppressive microenvironment appears in tumours belonging to the LAG-3 cluster (i.e., ranging from acquired immunity to innate immunity in primary ccRCC), revealing high levels of cell exhaustion in tumour-infiltrating CD8 T cells and infiltrating TAMs.
Metastasis-specific differences
We next investigated metastasis-specific differences in the three new IR clusters by applying ccRCC tumour samples harbouring metastases in the lung, bone, viscera, brain, and other sites (i.e., COHORT 2, Supplementary Table 1). Automated single-cell counting was applied, and LAG-3/TIM-3/TIGIT-positive cells were counted in a total of 47 ccRCC metastases. The mean number of cells was 146.0 ± 31.5/mm2 for LAG-3, 974.3 ± 130.9/mm2 for TIM-3, and 25.2 ± 4.4/mm2 for TIGIT. Compared with primary ccRCC tumours in the kidney, increased numbers of TIM-3+ (p < 0.001) and TIGIT+ (p < 0.001) cells were obvious in ccRCC metastases. Spearman rank correlations among continuous variables of the IRs were 0.172 for LAG-3 and TIM-3 (p = 0.248), 0.136 for LAG-3 and TIGIT (p = 0.363), and − 0.087 for TIM-3 and TIGIT (p = 0.560) (Supplementary Fig. 1). No correlations among these IRs seemed obvious in metastatic ccRCCs; rather, exclusive relationships were observed, allowing small fractions to overlap.
Hierarchical clustering of 47 metastatic ccRCC tumour samples based on positive cell densities for the IRs identified three groups with distinct IR levels. TIM-3 (68%), LAG-3 (9%) and TIGIT (23%) were individually dominant in independent clusters (Fig. 3a). Interestingly, hierarchical clustering revealed differences between metastatic ccRCC tumours and primary ccRCC tumours, and the TIGIT cluster accounted for more than the LAG-3 cluster (Fig. 3b). However, varying patterns of these IR signatures across metastatic regions were obvious, indicating that regional heterogeneity has added another layer of complexity to the second series of IRs (LAG-3, TIM-3, and TIGIT) in RCC. Taken together, these results suggest that only lung metastasis has a unique characteristic in metastatic ccRCC and show that the TIGIT cluster occupies the majority.
Next, cell-by-cell immunohistological analysis for acquired immunity was performed using metastatic ccRCC samples. The levels of tumour-infiltrating CD8 T cells were higher in metastatic lesions from the LAG-3 cluster than in those from the two other clusters (Fig. 3c). Tumours in the LAG-3 cluster were also associated with high levels of CD39, PD-1 and CTLA-4 (Fig. 3c). Our immunohistological analysis for innate immunity revealed high levels of infiltrating TAMs in metastatic tumours belonging to the LAG-3 cluster (Fig. 3d). Therefore, the immunosuppressive microenvironment also appeared in tumours belonging to the LAG-3 cluster in metastatic ccRCC, ranging from acquired immunity to innate immunity. Our assessments also extended to the state of tumour metabolism in metastatic ccRCC samples labelled with the proliferative Ki-67, IDO-1, GLUT-1, and CD73 markers, but no difference in inhibitory tumour metabolism was noted among the three groups (Fig. 3e). The cancer stroma molecules, CD34 and D2-40, were also not different among the three groups (Fig. 3f).
Histological subtype-specific differences
We further examined subtype-specific differences in the three IR clusters by applying primary non-ccRCC tumour samples of the papillary, chromophobe, sarcomatoid, Xp11.2 translocation, and collecting duct subtypes (i.e., COHORT 3, Supplementary Table 2). Automated single-cell counting was applied, and LAG-3/TIM-3/TIGIT-positive cells were counted in a total of 41 non-ccRCC individuals. The mean number of cells was 93.7 ± 34.0/mm2 for LAG-3, 444.3 ± 82.0/mm2 for TIM-3, and 23.7 ± 6.1/mm2 for TIGIT. Compared with primary ccRCC tumours in the kidney, increased numbers of TIGIT+ (p < 0.001) cells were obvious in non-ccRCC tumours. Spearman rank correlations among continuous variables of the three IRs were 0.069 for LAG-3 and TIM-3 (p = 0.670), 0.327 for LAG-3 and TIGIT (p = 0.037), and 0.215 for TIM-3 and TIGIT (p = 0.178) (Supplementary Fig. 1). In non-ccRCC, a weak correlation was confirmed only among LAG-3 and TIGIT; the others were not exclusive.
Hierarchical clustering of 41 non-ccRCC tumour samples based on positive cell densities for the three IRs identified three groups with distinct IR levels. TIM-3 (39%), LAG-3 (12%) and TIGIT (48%) were individually dominant in independent clusters (Fig. 4a). Interestingly, hierarchical clustering of non-ccRCC tumours yielded the opposite results, and the TIGIT cluster occupied the majority (Fig. 4b). However, varying patterns of the three IR signatures across histological subtypes were obvious even in non-ccRCCs, indicating that inter-subtype heterogeneity has added another layer of complexity to the second series of IRs (LAG-3, TIM-3, and TIGIT) in RCC. Taken together, these results show that both chromophobe and Xp11.2 translocation subtypes had slightly different characteristic in non-ccRCC and that the TIM-3 cluster occupied the majority (Fig. 4b). Cell-by-cell immunohistological analyses for acquired immunity, innate immunity, inhibitory tumour metabolism, and cancer stroma were performed in the same fashion, but no difference was noted in any field (Supplementary Fig. 2).
Validation and discrimination of the LAG-3, TIM-3, and TIGIT signatures
The profiling of ccRCC patients with regard to the three new IR signatures may provide robust screening when combined with the second series of IRs and anti-PD-1/PD-L1 therapies. However, is the phenomenon where each IR exists exclusively across individual clusters universal in RCCs? To answer this question, we sought to validate our hypothesis using two publicly available confirmation cohorts: the TCGA (Fig. 5a) and Sato (Fig. 5b) ccRCC datasets (17, 19). Interestingly, similar results were obtained from the two large-scale RNA-sequencing datasets, revealing that each transcription dataset from the two ccRCC cohorts was successfully distributed into three different clusters with the individual dominations of each IR level.
We next asked how we should judge the three IR signatures in each tumour in clinical practice. In summary, our automated platform for immunohistologically discriminating LAG-3, TIM-3, and TIGIT signatures at the single-cell level may provide a quantitative and high-throughput pathological assessment. We herein propose an optimal workflow and biomarker cut-off for the three IRs to translate in future practice (Fig. 5c). First, COHORT 1 consisted of FFPE tumour samples, which are stored in vast numbers in biobanks worldwide. We referred to ROC curves (Supplementary Fig. 3) to analyse automated single-cell pathology data from COHORT 1 to test our workflows and determine potential biomarker cut-offs for the three IRs. The results revealed that patients in this training group were successfully divided into three groups with distinct IR levels by immunohistochemistry (Fig. 5d). Our current model was further applied to the external validation dataset containing in-house primary ccRCC samples, namely, COHORT 4 (n = 96) (stored in alcohol-based fixative; see Methods) and achieved similarly good discrimination to screen matched IR clusters by immunohistochemistry (Fig. 5e). Prognostic analysis of the validation cohort, COHORT 4, showed that the recurrence-free survival (Fig. 5f) and overall survival (Fig. 5g) rates were different among the three IR clusters, in which the LAG-3 cluster was associated with the worst recurrence-free survival and overall survival rates.