Immunologic Characterization and T cell Receptor Repertoires of Expanded Tumor-infiltrating Lymphocytes in Patients with Renal Cell Carcinoma

The successful use of expanded tumor-infiltrating lymphocytes (TIL) in adoptive TIL therapies has been reported, but the effects of the TIL expansion, immunophenotype, function, and T cell receptor (TCR) repertoire of the infused products relative to the tumor microenvironment (TME) are not well understood. In this study, we analyzed the tumor samples (n = 58) from treatment-naïve patients with renal cell carcinoma (RCC), “pre-rapidly expanded” TILs (pre-REP TIL, n = 15) and “rapidly expanded” TILs (REP TIL, n = 25) according to a clinical-grade TIL production protocol, with single-cell RNA (scRNA)+TCRαβ-seq (TCRαβ sequencing), TCRβ-sequencing (TCRβ-seq), and flow cytometry. REP TILs encompassed a greater abundance of CD4+ than CD8+ T cells, with increased LAG-3 and low PD-1 expressions in both CD4+ and CD8+ T cell compartments compared with the pre-REP TIL and tumor T cells. The REP protocol preferentially expanded small clones of the CD4+ phenotype (CD4, IL7R, KLRB1) in the TME, indicating that the largest exhausted T cell clones in the tumor do not expand during the expansion protocol. In addition, by generating a catalog of RCC-associated TCR motifs from >1,000 scRNA+TCRαβ-seq and TCRβ-seq RCC, healthy and other cancer sample cohorts, we quantified the RCC-associated TCRs from the expansion protocol. Unlike the low-remaining amount of anti-viral TCRs throughout the expansion, the quantity of the RCC-associated TCRs was high in the tumors and pre-REP TILs but decreased in the REP TILs. Our results provide an in-depth understanding of the origin, phenotype, and TCR specificity of RCC TIL products, paving the way for a more rationalized production of TILs. Significance: TILs are a heterogenous group of immune cells that recognize and attack the tumor, thus are utilized in various clinical trials. In our study, we explored the TILs in patients with kidney cancer by expanding the TILs using a clinical-grade protocol, as well as observed their characteristics and ability to recognize the tumor using in-depth experimental and computational tools.


Introduction
Renal cell carcinoma (RCC) is considered as a highly immunogenic cancer with an abundance of tumor-infiltrating lymphocytes (TIL; refs. 1-3). However, contrastingly to other cancers, the link between increased number of tumor CD8 + The reinfusion of TILs in the clinical setting has resulted in striking clinical responses in patients with metastatic melanoma with the combination of lymphodepletion and high-dose IL2 (13,14). Adoptive TIL therapies have resulted in objective responses in almost half of the patients with metastatic melanoma (13,15), including those refractory to anti-PD-1 therapy (16)(17)(18), encouraging the need for optimization in patients who fail other types of immunotherapies. Although long-lasting responses have been observed in patients with metastatic melanoma (17,19), only one ongoing clinical trial using adoptive TIL therapy currently exists for RCC (NCT02926053; ref. 20), despite both tumors being known to respond to immunotherapies (21,22). Past studies have reported weaker immune responses in RCC TILs than in metastatic melanoma (20,23). However, a comprehensive understanding of the immunophenotype and function of the cells in the ex vivo expanded TIL product compared with the original tumor sample is still lacking.
In this study, we characterized the non-expanded and expanded TILs in patients with RCC and explored differences in the immune phenotype, tumor reactivity, cytotoxic ability, and T cell receptor (TCR) repertoires. With the use of the clinical-grade TIL expansion protocol that is currently applied in various clinical trials involving adoptive TIL therapies for patients with cancer (24)(25)(26), we compared the immunological differences between the minimally cultured "pre-rapid expansion" TILs (pre-REP TILs) and "rapidly expanded" TILs (REP TILs). We show our REP TIL protocol favors the expansion of CD4 + T cells, which originate from small T cell clones in the tumor microenvironment (TME). In contrast, large, expanded tumor CD8 + T cell clones mostly vanish during the TIL expansion. Furthermore, we identified RCC tumor-associated TCR motifs that were validated in multiple TCRβ-sequencing (TCRβ-seq) and single-cell RNA (scRNA)+TCRαβ-seq (TCRαβ sequencing) datasets. As a result, T cells carrying the RCC-associated TCR motifs were enriched in the pre-REP TIL samples, whereas the frequency was reduced in the REP TILs, suggesting that some tumor-reactive T cell clones are lost during the expansion.

Patient Cohort and Study Approval
The study included 58 patients with RCC undergoing radical nephrectomy between 2016 and 2020 (Fig. 1A). Primary tumor tissue (n = 58), adjacent healthy kidney tissue (n = 30), and matched peripheral blood (PB; n = 40) samples were obtained. All procedures were conducted in compliance with the Declaration of Helsinki and the studies were approved by the Helsinki University Hospital Ethical Committee (Dnro 115/13/03/02/15) and. All samples were obtained after a written informed consent.

Clinical Data
In total, 18 clinical parameters including tumor size, World Health Organization International Society of Urologic Pathologists (ISUP) 2016 grade, tumor-nodemetastasis (TNM) staging, presence of necrosis, perirenal and peripelvic fat infiltration, rhabdoid histology, and other medical histories were assessed (Supplementary Table S1).

Sample Processing
Briefly, the tumor and healthy kidney tissue samples were stored in MACS tissue storage solution (Miltenyi Biotec 130-100-008) at 4°C upon harvest and processed immediately upon arrival using Miltenyi's Tumor Dissociation kit (Miltenyi Biotec 130-095-929). The remaining dissociated cells were viably frozen in 10% FBS-DMSO freezing solution and kept at −150°C until further use. Peripheral blood mononuclear cells (PBMC) were separated from PB samples with density gradient centrifugation using Ficoll-Paque (GE Healthcare) and viably frozen in 10% FBS-DMSO at −150°C.
PBMCs for the feeder cells were isolated by density gradient centrifugation using Ficoll-Paque (GE Healthcare). The complete pre-REP TIL and REP TIL protocols have been previously described in detail (27).

Multi-parameter Flow Cytometry and Immunophenotyping
Freshly dissociated tumor, healthy kidney, and TIL samples were used to examine the immune cell numbers and immunophenotypes. The samples were stained for 15 minutes with a comprehensive antibody staining panel as described previously (28) and were washed twice with PBS before phenotyping.
A total of 50,000 lymphocytes were acquired per tube with FACS Verse (BD Biosciences) and analyzed with FlowJo (Version 10.0.8rl, Treestar). A full list of markers is provided in Supplementary Table S2. All antibodies were purchased from BD Biosciences unless mentioned otherwise. In addition, the availability of samples used in each analysis is provided in Supplementary Table S3.

REP TIL and Tumor Co-cultures
Autologous in vitro cultured two-dimensional tumor cells (derived from the same lesions as the TIL cultures) were established from live frozen tumor dissociated cells using the Primary Cancer Culture System kit (PromoCell C-28081) in 12-well plates. The morphology and growth patterns of the tumor cells were examined using light microscopy. Mycoplasma tests (MycoAlert PLUS, Lonza LT07-710) were performed and were negative. After 4-7 weeks of cell culture maintenance, cells were checked for 90%-100% confluency and used for the co-cultures. The cocultures included a 6-hour and 48-hour incubation period, whereby 1 million REP TILs were incubated with the corresponding tumor cells for both timepoints. GolgiStop was added 6 hours before the end of the experiment to measure intracellular cytokine production. The TILs were gently pipetted from the wells and phenotyped using the FACS Verse (BD Biosciences). The data were analyzed with FlowJo (v. 10 was used for all analyses. ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Bulk TCRβ-seq Data Acquisition and Analysis
Bulk TCRβ-seq was carried out from genomic DNA using the Adaptive Biotechnologies ImmunoSEQ Assay "Survey" resolution (29). Only produc-

scRNA+TCRαβ-seq Data Acquisition and Analysis
Viable frozen tumor dissociated cells from 3 patients with RCC were thawed in PBS, 2 mmol/L Ethylenediaminetetraacetic acid (EDTA), and the CD45 + cells were selected with the Sony SH800 cell sorter (Sony Biotechnology Inc.) for sequencing. Next, single cells were partitioned into the Chromium Controller (10X Genomics

scRNA+TCRαβ-seq Data Analysis
The scRNA-seq+TCRαβ-seq data were analyzed to assess the phenotype of the expanded pre-REP TIL and REP TIL clonotypes from the tumor (n = 3). Quality control metrics were assessed individually for each sample by visual inspection and low-quality cells were additionally removed in further stages of the analysis when identified. We combined the samples and utilized scvi-tools (ref. 35; v.0.13.0) for batch correction using default parameters, treating each sample as one batch and correcting for cell-cycle heterogeneity by using scores calculated with the Seurat "CellCycleScoring" function as covariates. The latent embeddings were used for graph-based clustering and uniform manifold approximation and projection (UMAP) dimensionality reduction in Seurat (refs. 36, 37; v.4.1.0) with default parameters. In addition, we used scRepertoire (38) to combine the TCRαβ-seq data with the Seurat object, where individual clonotypes were defined with the "CTstrict" criteria, meaning that a clonotype should share exactly the same nucleotide sequence of the TCR.
For clustering, we visually inspected the results by varying the "resolution" parameter in the "FindClusters" function between 0.2 and 0.5. This allowed us to find the optimal number of clusters without overclustering or underclustering, based on the agreement between UMAPs and the chosen clustering resolution, together with whether we could find biological phenotypes for each cluster. Clusters were named and annotated by analysis of canonical markers (39,40), differentially expressed genes (DEGs), relationship to other clusters, and TCR repertoire clonalities. The default parameters in the "RunUMAP" function were used for all UMAP dimensionality reductions. The "AddModuleScore" functions, as defined by Tirosh and colleagues (41), were used to calculate the different scores based on the expression of specific genes. The cytotoxicity score was calculated using the genes GZMB, GZMA, GZMH, PRF, GNLY, and FGFBP. The exhaustion score was calculated on the basis of PDCD, LAG, CTLA, TIGIT, TOX, and HAVCR. We used ITGA, ITGAE, ZNF, CD, IFNG, and CCR to calculate the tissue resident score. scRepertoire (38) was used to combine the TCRαβ-seq data with the Seurat objects. DEG analyses were performed with Wilcoxon rank-sum test and P values were Bonferroni adjusted and corrected. Genes with adjusted P values <0.05 were considered significant. Fisher exact test with a two-sided alternative was used to study the overlapping TCRs between the pre-REP TILs and REP TILs.

Identification and Validation of RCC-associated Motifs
The GLIPH2 (42) algorithm (v.1.0.0) was run separately on the discovery cohorts of the pooled tumor (n = 42), healthy kidney (n = 24), and PB (n = 32) TCRβ-seq samples using default parameters. The RCC-associated motifs consisted of those that were exclusive and enriched to the tumor. Tumor-exclusive motifs were found only in the tumor samples after filtering. Motifs that were shared between the tumor-healthy kidney and tumor-PB were analyzed using the two-sided Wilcoxon test; all TCRs with a motif were pooled together, and TCRs without motifs were noted as 0. P values were corrected with Benjamini-Hochberg adjustment and only motifs with an adjusted P value <0.05 and log 2 fc >1 were annotated as tumor-enriched. The RCC-associated motifs are listed in Supplementary Table S4.

Analysis of RCC-associated Motifs in Different TCRβ-seq and scRNA+TCRαβ-seq Datasets
Bulk TCRβ-seq RCC-associated motifs were analyzed in different cohorts, including the discovery cohort (

Immunophenotype of RCC-associated Motifs in scRNA+TCRαβ-seq Data
In the validation cohorts, the phenotypes were obtained from the original publications (49,50). For each pattern with at least five supporting cells, the odds ratio (OR) for a pattern with a certain phenotype was calculated, and the phenotype with the largest OR was assumed to be the dominant phenotype. ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

RCC Tumors, Healthy Tissue, Pre-REP TILs, and REP TILs Display Immunophenotypic Differences
In total, we collected 58 primary tumor samples (Supplementary Fig. S1; Supplementary  Table S3). Representative flow cytometry gating strategies for the various sample types, marker expressions, and paired co-culture assays described below are presented in Supplementary Fig. S2-S5. The initial median proportion of lymphocytes from all dissociated tumor cells was 1.6% compared with 0.69% in the healthy kidney samples. During the clinical grade TIL expansion protocol, T cells were massively expanded; over 80% of the lymphocytes were CD3 + T cells in the REP TILs and only few natural killer (NK) cells (0.03%) were observed (Fig. 1A). The overall median proportion of lymphocytes was 89.2% in the pre-REP TILs and 94.9% in the REP TILs (Supplementary Fig. S6A). In contrast, an increased proportion of cells in the tumor (14.4%), healthy kidney (14.4%), and pre-REP TIL (13.5%) samples were NK cells (Fig. 1A). In the tumor samples, CD4 + T cells were already more prevalent than CD8 + T cells (34% vs. 19.5%), and in pre-REP TILs and REP TILs, the CD4 + T cell dominance further increased (46.8% and 68.4%, respectively; We next analyzed the T cell immunophenotypes between the different samples using the expression of CCR7 and CD45RO markers. Both tumor and healthy CD4 + and CD8 + T cells predominantly displayed an effector memory (T EM ) phenotype ( Fig. 1B; Supplementary Fig. S3B). Among the pre-REP TILs, CD4 + T cells mostly displayed a central memory (T CM ) phenotype (64%), whereas CD8 + T cells had both T CM (37.7%) and T EM (40.8%) cells in similar quantities.
In the REP TILs, 37% of CD4 + T cells had a T CM phenotype and approximately half were T EM cells (55.8%), whereas more than half the CD8 + T cells were predominantly T EM cells (61.5%; Fig. 1B).
From the limited number of cases, further analyses with paired sample types were available (Supplementary Fig. S6C and S6D). Because of the small sample size, differences between the groups were not significant, but similar trends (increased LAG-3 and HLA-DR, decreased PD-1 expression) were observed (Supplementary Fig. S6C and S6D).

The Expansion Potential of TILs Correlates with CD4 + T CM Phenotype
To understand whether the expansion potential was correlated with a distinct cell type in the product, we calculated the fold change between the number of REP TILs at the end of the expansion protocol and the starting number of pre-REP TILs (Supplementary Table S3 Supplementary Fig. S6E), suggesting that a low amount of CD4 + T CM cells initially in the pre-REP TILs led to a better yield of REP TILs. Accordingly, a higher quantity of CD4 + T EM cells in the pre-REP TILs led to a better REP TIL expansion although no statistical significance was observed (Spearman, R = 0.67, P = 0.059). A similar negative trend was observed between the pre-REP TIL CD8 + T CM cell phenotype and REP TIL/pre-REP TIL fold-change ( Supplementary Fig. S6F). In addition, a borderline negative correlation between the pre-REP TIL CD4 + T cells and PD-1 expression and REP TIL/pre-REP TIL fold-change was observed (Spear-man, R = −0.62, P = 0.06), suggesting that putatively exhausted CD4 + T cells expressing the marker do not expand well in the REP protocol ( Supplementary   Fig. S6G).

REP TIL CD4 + and CD8 + T cells Respond to Tumor Cells in Coculture Models
To explore how REP TILs react when exposed to tumor cells, we set up a coculture system by incubating (6 and (Fig. 2B), but similar trends were not observed at 48 hours of co-culture ( Supplementary Fig. S7D). The changes in PD-1 expression upon co-culture varied between individuals, and in half of the patients, PD-1 expression increased (Fig. 2B).
We also stimulated T cells with anti-CD3 and the costimulatory antibodies, anti-CD28 and anti-CD49d to activate T cell cytokine production and degranulation. REP TILs were able to respond to activation in the presence of tumor cells; increases in the CD3 + (median 0.72% vs. 1.33%, P = 0.098) and CD4 + (0.46% vs. 1.08%, P = 0.012) T cell IFNγ and TNFα expressions were observed after 6 hours of co-culture (Fig. 2C). However, after prolonged co-culture conditions (48 hours), anti-CD3 + , anti-CD28 + , and anti-CD49d + stimulation of the T cells no longer led to increased cytokine production, suggesting either that the immunosuppressive role of the tumor cells, or cytokine production had already reached its maximum, with only tumor cells as a stimulus to the T cells ( Fig. 2D). Furthermore, a moderate decrease in CD8 + T cell GzB expression (P = 0.039) at 6 hours was observed, with a borderline increase in the expression of CD107a/b in CD8 + T cells at 6 and 48 hours of T cell stimulation (P = 0.063; Supplementary Fig. S7E and S7F).

Bulk TCRβ-seq Shows Differences Between Tumor and TIL T cell Repertoires
First, we analyzed the clonality between the TCR repertoires of the healthy kidney (n = 24), tumor (n = 36), pre-REP TIL (n = 7), and REP TIL (n = 7) samples using the Gini index, a measure of clonal inequality (values closer to 1 denote higher clonality). The pre-REP TILs and REP TILs were significantly more clonal than the tumor and healthy kidney samples, suggesting that the TCR repertoire diversity was lost during the expansion protocol (Fig. 3A). When we further explored the differences between the TCR repertoires of the various samples, we observed that the pre-REP TIL TCR repertoire was mostly dominated by the top 10 most abundant clones, whereas in the tumors, the repertoire was more variable, comprised of smaller clones ( Fig. 3B; Supplementary Fig. S8AI-S8V  (Continued) co-cultured REP TILs without any stimulation at 6 hours. C, T cell activation potential was assessed by stimulating the cells with anti-CD3 (OKT3), anti-CD28, anti-CD49d antibodies, and IFNγ and TNFα cytokine secretion was measured as described above. When REP TILs were stimulated and co-cultured for 6 hours, an increase in the CD3 + T cell IFNγ and TNFα expressions was observed between unstimulated and stimulated co-cultures (median 0.72% vs. 1.33%, P = 0.098). A moderate increase in CD4 + T cell IFNγ and TNFα expressions was also observed (0.46% 1.089%, P = 0.012), but not in the CD8 + T cells. CO-CULTURE_6h = cocultured REP TILs without any stimulation at 6h, TSTIM_6h = cocultured REP TILs that were T cell stimulated at 6 hours. D, Prolonged co-culture conditions (48 hours) did not lead to increased IFNγ and TNFα production in the CD3 + , CD4 + , and CD8 + T cell compartments. CO-CULTURE_48h = cocultured REP TILs without any stimulation at 48h, TSTIM_48h = cocultured REP TILs that were T cell stimulated at 48 hours. Non-parametric Wilcoxon matched-pairs signed rank test was used for all co-culture analyses. ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.
comprised approximately 5% of the entire TCR repertoire in the tumor sample ( Fig. 3C; Supplementary Fig. S8BI-S8V). Overall, the clonotypes enriched in the REP TILs were found to be very small clones in the matching samples.
Our results were confirmed when we analyzed the total clonal overlap between REP TILs and pre-REP TILs, as well as between the tumor and pre-REP TILs in terms of the total number of clones in each sample repertoire ( Fig. 3D; Supplementary Fig. S9 and S10). To understand whether the expanded clonotypes target common viral antigens, we matched the TCRs to VDJdb (45), a publicly curated TCR database. Individual matches to numerous viral epitopes (such as CMV, EBV, Influenza-A, and HIV-1) were discovered; however, these clonotypes were few and comprised only a small proportion of the entire TCR repertoire, suggesting that the expanded clones either in the tumor, pre-REP TILs, and REP TILs did not target common viruses ( Fig. 3E; Supplementary   Fig. S11).

Identification and Validation of RCC-associated TCR Motifs in Large TCRβ-seq and scRNA+TCRαβ-seq Sample Cohorts
After exploring the TCR repertoires between the tumors and the various sample types, we sought to further explore the shared TCR clonotypes using machine learning tools, such as TCRGP (32) and GLIPH2 (42) to discover antigenspecific TCR clusters enriched in the tumor. First, we ran GLIPH2 for each of the pooled tumor (n = 42), healthy kidney (n = 24), and PB (n = 32) TCRβ-seq samples. Similar to previous studies (47), we noticed that although the TCRs were highly conserved between individuals, the antigen-specific motifs were shared between the tumor samples. Next, to prune out the TCRs that were unlikely to target RCC-associated antigens, we filtered out motifs that were also found in the PB samples of healthy donor CD4 + (n = 37), CD8 + (n = 37) sorted T cells (34) Supplementary Fig. S12A). In a scRNA+TCRαβ-seq cohort (50) of patients with RCC (n = 6), higher levels of RCC-associated TCRs were found in patients who received nivolumab and ipilimumab combination therapy (n = 4) than in untreated patients (n = 2, P value not counted due to low n; Supplementary Fig. S12B). The phenotype of the T cells carrying the motifs varied significantly; different motifs were linked to various phenotypes, but mostly to exhausted CD8 + T cells, tissue-resident memory T cells, and regulatory T cells ( Supplementary Fig. S12C and S12D).
In TCGA cohort of patients with at least 100 recovered TCR reads from the bulk RNA-sequencing data, the number of TCRs (and the overall number of T cells) was associated with worse overall survival (OS) (P = 0.033, log-rank test; Supplementary Fig. S12E-S12G). When the number of TCRs with RCC-associated motifs was normalized to the number of TCRs in total, the proportion of TCRs with RCC-associated TCR motifs was not associated with OS (P = 0.3, log-rank test between patients with above median proportion of TCRs with RCC motifs compared with those below median proportion of TCRs with RCC motifs; Supplementary Fig. S12H).

Pre-REP TILs Carry the Most TCRs with RCC-associated Motifs
We further utilized the validated RCC-associated TCR motifs to estimate the proportion of RCC-associated TCRs in our samples using the TCR-epitope recognition machine learning classifier, TCRGP (32). To estimate the frequency of other antigen-specific clones, we predicted the specificity of the TCRs against common endemic viruses such as CMV, EBV, Influenza A, and SARS-CoV-2.  Fig. 4C, left). The proportion of TCRs with RCCassociated motifs was the highest in pre-REP TILs, second in tumors, and third in the REP TILs. Anti-viral TCRs in the pre-REP TIL and REP TIL samples were not enriched (Fig. 4C, right). In addition, we noted a patient (pt240) that underwent two subsequent nephrectomies possessed a higher number of TCRs with RCC-associated motifs in the right kidney (26.5%) than in the left (11.1%), reflective of the different histopathologic tumor (TNM) stages (pT3aNX vs. pT1a; Fig. 4D). The variation in the proportion of TCRs with RCC-associated motifs was higher in the tumor (mean 23.4%, SD 0.1) than in the final REP TIL samples (mean 16.5%, SD 0.08), and the baseline amount of TCRs with RCC-associated motifs was not related to the quantity in the REP TIL products ( Fig. 4D and E)  (Continued) Comparisons between the rest of the samples are found in Supplementary Fig. S4A. C, The top 20 most abundant REP TIL clonotypes were tracked to the corresponding tumor and pre-REP TIL samples in the same representative cases (pt081 and pt115). Overall, the clonotypes ending up in the REP TILs were very small clonotypes in the other sample types. Comparisons between the rest of the samples are found in Supplementary  Fig. S4B. D, The frequencies of the overlapping T cell clonotypes in one representative case (pt115) between the REP TIL and pre-REP TIL samples (left), as well as between the tumor and pre-REP TILs (right). Spearman correlation was used to compare the frequencies of the clonotypes in the samples. The xand y-axes are logarithmically transformed. R and P values refer to the log-transformed frequencies of the clonotypes. The frequencies for the rest of the samples are found in Supplementary Fig. S5 and S6). E, Treemaps showing the size of the T cell clonotypes in each sample type (tumor, pre-REP TIL, and REP TIL) in one representative case (pt081). The size of each box indicates the size of one clonotype in the TCR repertoire.
Colored boxes indicate TCRs matched to viral-specific TCRs found in VDJdb (45). Few matches were discovered and represented as small T cell clonotypes, indicating that expanded clones in the tumor, pre-REP TILs and REP TILs do not target common viruses. Treemaps for all the samples are found in Supplementary Fig. S7.

Single-cell Transcriptomics Reveals Distinct Origins of pre-REP TIL and REP TIL Clonotypes in RCC Tumors
To further explore the pre-REP TIL and REP TIL clonotypes in greater reso-  Table S5).
Most of the CD4 + T cells were of T CYTOTOXIC phenotype, with high expression of the cytotoxic marker gene, KLRB, as well as tissue resident marker genes (ref. 55; MYADM, ITGAE, CD; Fig. 5D and E; Supplementary Fig. S13C and S13D).
The second largest CD4 + T cell cluster corresponded to T REGS (cluster 4), with high expression of T REG marker genes (FOXP, ILRA/CD, CTLA, TIGIT), but also the expression of activation genes (TNFRSF/OX, TNFRSF/GITR) and CCR, which has been proposed as a marker of clonally expanded T REGS that recognize tumor antigens (56)(57)(58). Cluster 7, that was almost exclusive  Supplementary Fig. S13C and S13D). This cluster was named pre-exhausted as in other publications, due to its expression of exhaustion-related markers such as PDCD (encoding PD-1), but the lack of expression of other immune checkpoint molecules associated with terminal exhaustion, such as HAVCR (encoding TIM-3) and TOX (65)(66)(67)(68)(69)(70). These terminal exhaustion markers were expressed by cluster 1 (Fig. 5D and E; Supplementary Fig. S13C and S13D). On the basis of the TCR-repertoire clonality analysis, the terminally exhausted cluster 1 had the most clonally restricted population (Gini index 0.8), making it plausible that the T PRE-EXH cluster 2 indeed had stem-like properties, and thus gave rise to the cells in the terminally exhausted cluster 1 ( Supplementary   Fig. S14A and S14B).
We also analyzed the size of the clonotypes in the scRNA+TCRαβ-seq and TCRβ-seq data. The largest clonotypes found in the scRNA+TCRαβ-seq data were mainly found in the TCRβ-seq pre-REP TILs, whereas the REP TILs encompassed smaller clonotypes in pt115 and pt616 (Fig. 6A). We further analyzed the size of the clonotypes in the tumor scRNA+TCRαβ-seq data for pt115 and pt616, as well as from the TCRβ-seq data in the corresponding samples (healthy, tumor, pre-REP TIL, and REP TIL) and found that the expanded clonotypes sequestered to distinct parts of the UMAP (Fig. 6B; Supplementary Fig. S14C).
In the third patient (pt423), we did not find matching clones between the tumor scRNA+TCRαβ-seq, pre-REP TIL and REP TIL samples ( Supplementary  Fig. S14C), suggesting the result of either sampling bias, or that the dissociated tumor cells and REP TILs were derived from physically distant regions of the tumor tissue.
Subsequently, we visualized the shared and expanding clonotypes between the pre-REP TIL and REP TIL samples as well as their corresponding phenotypes.
We defined the statistically significant expanded clonotypes as those at least ≥0.1% in either one of the pre-REP TIL or REP TIL samples, using Benjamini-Hochberg-corrected two-sided Fisher's exact test. In both patients, the largest clonotypes in the pre-REP TILs resulted from large tumor-infiltrating CD8 + AACRJournals.org Cancer Res Commun; 3(7) July 2023  T cell clones (Fig. 6B). However, the CD8 + T cell phenotypes in the patients were heterogeneous. In pt115, the expanded pre-REP TIL clonotypes were more of the CD8 + T PRE-EXH phenotype (cluster 2), whereas in pt616, most of the expanded pre-REP TIL clonotypes were of the CD8 + T EXH (cluster 1) phenotype (Figs. 5A and 6B). The clonotypes that were expanded in the REP TIL samples displayed a CD4 + T CYTOTOXIC phenotype (cluster 0) from the original tumor sample for both cases (Figs. 5A and 6B).
Next, we analyzed the DEGs between the statistically significant clonotypes expanded in the REP TILs for pt115 and pt616. In pt115, the genes that were most upregulated in the expanded REP TIL clonotypes in the original tumor were NKG, CCL, and CDA, but also those related to cytotoxicity, such as GZMB/H/K (Fig. 6C). In pt616, genes related to activation such as HLA-DRB, HLA-DRA, chemokines such as CCL and CCLL, as well as LAG- were observed, in line with our previous observations (Fig. 6C). Furthermore, gene ontology (GO) pathway analysis using clusterProfiler (71) showed T cell activation and response to IFNγ among the top GO pathways enriched in pt115 pre-REP TIL clonotypes in the original tumor, whereas pathways related to cytokine production and T cell differentiation were found in the original REP TIL clonotypes ( Supplementary Fig. S15A). In pt616, IFNγ signaling, T cell activation, and regulation were among the most enriched pathways in the pre-REP TIL clonotypes, whereas cellular pathways related to cotranslational protein targeting were enriched in the REP TIL clonotypes ( Supplementary Fig. S15B).
Finally, we explored the phenotype of T cells with RCC-associated motifs found in the three tumor samples (pt115, pt423, pt616) and observed that the phenotypes of the T cells with RCC-associated motifs were largely distinct (Fig. 6D). T cells with RCC-associated motifs in pt115 were concentrated in the CD4 + T FH compartment (cluster 7), pt423 displayed a CD8 + T EXH cell phenotype (cluster 1), and pt616 displayed the CD8 + T PRE-EXH phenotype (cluster 2; Fig. 6D; Supplementary Fig. S15C). The same clonotypes that carried RCC-associated motifs were also observed in the pre-REP TIL and REP TIL products for pt115 and pt616 and were among the most expanded clonotypes (Fig. 6B and D). Of note, all the patients were in remission during the study follow-up after the surgery, thus the phenotypes could not be linked to clinical outcomes.

Discussion
Adoptive TIL therapies have been successfully used in melanoma patients, even in anti-PD-1 therapy refractory cases (16,18); thus, the interest in applying TILs to other tumor types has increased. Previous studies have demonstrated that TILs can be expanded from RCC tumor tissues (20,(72)(73)(74)(75). However, the composition, possible tumor reactivity, and clinical efficacy of RCC TILs remain unclear. Some studies have shown that the infiltration of T cells is associated with good prognosis in RCC (7-9), whereas others have reported contradictory results in which T cell infiltration has been associated with poor prognosis (4)(5)(6)76). In addition, although the REP TILs are mostly used as the final infusion product in the adoptive TIL therapies, there are only a few studies that have explored the differences in the phenotype between the original, minimally cultured pre-REP, and REP TILs (27,77,78). Therefore, we performed in-depth analyses of the pre-REP TILs and REP TILs that were expanded using a clinical expansion protocol from patients with treatment-naïve RCC. We also compared the TIL phenotype, function, and TCR repertoires with those originating from the autologous tumor tissue to shed light on the changes that occur during the ex vivo culture conditions. Our results show that the immunophenotype and TCR repertoire drastically change between the tumors, pre-REP TILs, and REP TILs, suggesting that initially, only a proportion of T cells expand from the tumor tissue (pre-REP TILs). In addition, from the pre-REP TILs, a smaller proportion of T cells further expand during the REP protocol.
Phenotypically, the REP TILs displayed a CD4 + T CYTOTOXIC -cell phenotype, which was drastically different from the T cells found in the TME, where a significant proportion of T cells exhibited CD8 + T PRE-EXH and T EXH phenotypes. The elevated expression of PD-1 in the tumor CD4 + and CD8 + T cells indicates the exhausted nature of T cells in the TME, which may ultimately hinder their expansion. These exhausted PD-1-expressing CD8 + T cells have traditionally been known as tumor-reactive and targets of ICI therapies. The exhausted clonotypes were usually large, and by comparing the TCR repertoires of different sample types, we discovered that the tumor and pre-REP TIL TCR repertoires overlapped, whereas the REP TILs possessed distinct repertoires. The most abundant clonotypes in the REP TILs were observed to be small, and/or unique in the pre-REP TIL and tumor samples, suggesting that many of the larger clonotypes present in the tumor and pre-REP TILs did not proliferate during the REP protocol. Accordingly, the scRNA+TCRαβ-seq data confirmed that the expanded pre-REP TIL clonotypes mostly encompassed a CD8 + T EXHcell phenotype in the original tumor samples, whereas the most expanded REP TIL clonotypes originate from tumor CD4 + T CYTOTOXIC cells.
Many previous studies have not been able to assess the proportion of tumorreactive or bystander (such as anti-viral) T cell clonotypes following the REP expansion. In contrast to metastatic melanoma (47), there is not yet a database of epitope-specific TCRs against RCC-associated antigens. However, with modern bioinformatics tools that assess antigen specificities based on TCR similarities (42), we were able to cluster TCRs with shared motifs into putative antigen-specific groups. We discovered RCC-associated TCR motifs and were able to validate them in multiple TCRβ-seq and scRNA+TCRαβ-seq datasets, , with distinct T cell phenotypes. The CD8 + T PRE-EXH and CD4 + T FH phenotypes were dominant in pt115, pt423 encompassed a more CD8 + T EFF and CD4 + T CYTOTOXIC phenotype, and pt616 was characterized by CD8 + T EXH and CD8 + T TRM phenotypes. C, Mapped expanded and singlet TCRαβ clonotypes on the tumor UMAP show that the clonotypes lie on distinct parts of the UMAP. Most of the non-expanded (singlet) clonotypes have a CD4 + T CYTOTOXIC phenotype (cluster 0), whereas expanded clonotypes are dominant in the CD8 + T EFF , T PRE-EXH and T EXH phenotypes (cluster 3, 2, and 1, respectively). The relative proportion of the different clonotypes in each of the UMAP clusters is found in Supplementary Fig. S9B. D, Scaled expression of selected canonical markers related to stemness, tissue residency, cytotoxicity, and exhaustion in the T cell UMAP representation as in A. E, Dot plot showing the scaled average (scale) and percentage (dot size) of expression between the selected DEGs for each UMAP cluster. The full list of the top 10 DEGs and their scaled expressions are found in Supplementary Fig. S9C and S9D).

FIGURE 6
Expanded clonotypes between the pre-REP TILs and REP TILs show distinct phenotypic differences within the tumor. A, UMAPs comparing the size of the T cell clonotypes found in the sc-RNAseq+TCRαβ-seq and TCRβ-seq data from 2 patients (pt115 and pt616). Each dot represents a clone in which the color corresponds to the size of the clone in the sc-RNAseq+TCRαβ-seq (left column), TCRβ-seq data from the pre-REP TILs (middle column), and TCRβ-seq data from the REP TILs (right column). Representations for all samples, including the available tumor and healthy kidney samples are found in Supplementary Fig. S10C). B, UMAPs showing the expanded pre-REP TIL and REP TIL clones in pt115 and pt616 (Benjamini-Hochberg-corrected two-sided Fisher's exact test, P adjusted < 0.05). In both patients, more pre-REP TIL clonotypes were expanded in the CD8 + T PRE-EXH (pt115) and CD8 + T EXH (pt616) phenotypes, whereas expanded REP TIL clonotypes were mainly characterized by the CD4 + T CYTOTOXIC cell phenotype. C, Volcano plots showing the DEGs in the expanded REP TIL samples (right) compared with those in the pre-REP TILs (left) in pt115 and pt616. In pt115, genes that were most upregulated (right) in the expanded REP TIL clonotypes were NKG7, CCL5, and CD8A, together with those related to cytotoxicity, such as GZMB, GZMH, and GZMK. In pt616, genes related to activation such as HLA-DRB1, HLA-DRA, chemokines such as CCL3 and CCL4L2, as well as LAG-3 were upregulated in the expanded REP TIL clonotypes. In both patients, IL7R was among the DEGs in the expanded pre-REP TIL clonotypes. D, UMAP showing T cells carrying the RCC-associated motifs found in the sc-RNAseq+TCRαβ-seq tumor patients (n = 3). T cells carrying the RCC-associated motifs showed distinct T cell phenotypes in each individual sample.
confirming their enrichment specifically in RCC tumors and not in the healthy or other cancer types. We further utilized the RCC-associated motifs to estimate the proportion of tumor-reactive TCRs, which were 4.3-84.4 times greater than the anti-viral (CMV, Influenza A, EBV, SARS-CoV2) TCR targets. Our analysis reiterates that REP TILs are specific to targets other than the large, exhausted clones, and without clinical follow-up, it remains unknown whether the expanded TCRs in the REP TIL products are optimal for efficient tumor killing.
Studies on pancreatic ductal adenocarcinoma (PDA) tumors have also shown that the TCR repertoires in TILs vastly change during in vitro expansions, with the loss of tumor-dominant T cell clones and emergence of new T cell clones that are barely detectable in the tumor (79). These changes are thought to be driven by differences in the in vitro expansion capacity of the T cell clones. In our RCC samples, the expansion potential was inversely correlated with the CD4 + T cell phenotype in the tumor. In PDA, the heterogeneity of the TILs resulted in TCR repertoires that were greatly divergent between TIL cultures derived from distant tumor samples of the same patient, suggesting that cultureinduced changes in clonal composition are likely to affect the tumor reactivity of TIL preparation (79).
Further studies are necessary to explore how to improve TIL expansion protocols, and whether TIL-based therapies for other T cell-rich tumors in addition to melanoma, will be effective. When tumor-targeting T cells are better recognized, it would be possible to selectively isolate the cytotoxic or tumor-reactive CD8 + T cells, and expand them separately, such as with autologous T cells, to increase the tumor killing potential. In addition, although not analyzed in our study, the use of NK and NKT cells as well as other immune cell types may hold potential for combinatorial use with TILs. In conclusion, our study highlights the differences between pre-REP TILs and REP TILs from primary RCC tumor samples and provides tools to analyze RCC-associated T cells in other datasets, such as during ICI therapies.