Pinpointing the tumor-specific T cells via TCR clusters

Adoptive cell transfer (ACT) is a promising approach to cancer immunotherapy, but its efficiency fundamentally depends on the extent of tumor-specific T cell enrichment within the graft. This can be estimated via activation with identifiable neoantigens, tumor-associated antigens (TAAs), or living or lysed tumor cells, but these approaches remain laborious, time-consuming, and functionally limited, hampering clinical development of ACT. Here, we demonstrate that homology cluster analysis of T cell receptor (TCR) repertoires efficiently identifies tumor-reactive TCRs allowing to: (1) detect their presence within the pool of tumor-infiltrating lymphocytes (TILs); (2) optimize TIL culturing conditions, with IL-2low/IL-21/anti-PD-1 combination showing increased efficiency; (3) investigate surface marker-based enrichment for tumor-targeting T cells in freshly isolated TILs (enrichment confirmed for CD4+ and CD8+ PD-1+/CD39+ subsets), or re-stimulated TILs (informs on enrichment in 4-1BB-sorted cells). We believe that this approach to the rapid assessment of tumor-specific TCR enrichment should accelerate T cell therapy development.

conditions 1 , inflammation 2 , establishment of an immunosuppressive microenvironment 3-6 , 38 downregulation of antigen presentation 7-9 , promotion of regulatory T cell (T reg ) infiltration 10 and 39 outgrowth 11 , and induction of T cell dysfunction 12 . The infusion of large numbers of expanded 40 autologous tumor-reactive T cells-typically after the implementation of lymphodepleting 41 regimens-represents a powerful therapeutic option that may override these immunosuppressive 42 mechanisms. Clinical protocols for such adoptive cell transfer (ACT) therapeutic strategies 13 , 43 inspired by the pioneering work of Steven Rosenberg's group [14][15][16] , are now being actively 44 developed and used to treat patients 17-21 . 45 There is accordingly great demand for methods for the enrichment of autologous tumor antigen-46 specific T cells for use in ACT protocols. Current techniques rely on the identification of patient-47 specific peptide neoantigens, which are then used for the functional characterization and 48 selection of cultured tumor-infiltrating lymphocytes (TILs) 22,23 . Alternatively, since the identification 49 of unique neoantigens is costly, time-consuming, and functionally limited in terms of the spectra 50 of identifiable antigens, cultured autologous tumor tissue can be used as a source of antigen-51 specific stimulus 24 . Certain cell-surface markers of ongoing and chronic activation such as PD-1, 52 CD39, CD69, CD103, or CD137 may also help to delineate T cell subpopulations (typically CD8 + ) 53 that are enriched for clonally-expanded tumor-reactive T cells 25-27 , therefore culturing selected 54 TIL subsets, such as PD-1 + T cells 28 , is a feasible option. In all of these scenarios, however, there 55 is the need for a robust method that enables estimation of enrichment of the transplanted cells 56 with tumor-specific T cells based on the T cell receptor (TCR) repertoire without prior knowledge 57 of TCR specificities. 58 The ongoing adaptive immune response is often driven by groups of T cell clones with highly 59 homologous TCR sequences (clusters) that recognize the same epitopes 29-32 . However, clusters 60 of highly similar TCRs also abundantly arise from the so-called "public" variants having high 61 probability of being generated in the course of V(D)J recombination 33,34 . ALICE approach 62 evaluates the number of "neighbors" relative to the baseline expectation from V(D)J 63 recombination statistics, and retains the node clonotypes of the TCR clusters with higher 64 numbers of neighbors than expected by a null model of recombination [29][30][31][32] . This approach is 65 highly efficient to capture clonotypes involved in the current immune response from a single 66 repertoire snapshot, and does not require longitudinal data collection [29][30][31][32]35 . 67 Here, we have employed ALICE-based cluster analysis to identify groups of TCR clonotypes 68 involved in the anti-tumor immune response. We demonstrate that this approach successfully 69 pinpoints known tumor-associated antigens (TAA)-specific TCRs among TIL repertoires in HLA-70 A*02 melanoma patients. Furthermore, we find that the number of cluster-related clonotypes and 71 the proportion of the bulk TIL repertoire that they occupy grows significantly after anti-PD-1 72 immunotherapy. We next investigate the TCR content in sorted CD4 + CD39 + PD1 + and 73 CD8 + CD39 + PD1 + TILs, and show that these subsets are prominently enriched for TCR clusters, a 74 substantial fraction of which consists of tumor-specific TCRs. These results provide a rationale for 75 focusing on CD39 + PD1 + TILs in adoptive cancer therapy. Finally, we show that repertoire analysis 76 facilitates optimization of TIL culturing conditions, and allows estimation of the extent of tumor-77 specific T cell enrichment in cultured donor cells and sorted TAA-activated T-cells. Altogether, our 78 findings strongly support the use of cluster TCR analysis as a powerful tool with practical 79 applications in clinical ACT. 80 81 Results 82 83 TIL clusters include TAA-specific TCRs and grow after immunotherapy 84 We first analyzed published TIL TCR beta chain (TCRꞵ) repertoires obtained before and after 85 anti-PD-1 immunotherapy for two cohorts comprising 21 and 8 patients with cutaneous 86 melanoma 36,37 . Using the ALICE algorithm 29 , we were able to identify clusters of convergent 87 TCRꞵ clonotypes in all patients. The number of cluster-related clonotypes significantly increased 88 after therapy in both cohorts (p = 0.019 and 0.038, respectively; Fig. 1a anti-tumor immune response (Fig. 1c,d). Fig. 1d shows identified TCR clusters for one of the 99 patients after immunotherapy. Summary for the count and size of TCR clusters before and after 100 immunotherapy for each patient is shown on Fig. 1e,f. 101 Notably, the HLA genotypes of the patients for these two cohorts were unknown. Since VDJdb 102 currently includes limited diversity of HLA contexts, we believe that a much higher proportion of 103 cluster-related clonotypes will be assigned to TAA-specificities with the accumulation of TCR 104 specificity data from more diverse HLA contexts. 105 106 CD39 + PD1 + TILs are enriched with clonal, convergent, and tumor-specific TCRs 107 It was previously reported that CD39 + and PD-1 + TIL subsets can be enriched for tumor-specific T  108  cells 25,28,40-45 . To verify whether there is concurrent enrichment with convergent TCR clusters, we  109 performed fluorescence-activated cell sorting (FACS) of CD39 + PD-1 + (double-positive, DP) and 110 non-DP CD4 + and CD8 + T cells from TILs freshly isolated from lymph node metastases from eight 111 melanoma patients (Fig. 2a, Fig. 2-figure supplement 1a). Overall, TIL composition was skewed 112 towards greater prevalence of CD4 + T-cells, although cells with the DP phenotype were more 113 prevalent among CD8 + TILs compared to CD4 + cells ( Fig. 2-figure supplement 2a,b). TCRꞵ 114 repertoire analysis revealed increased clonality and cluster enrichment for both CD4 + and CD8 + 115 DP TILs compared to the corresponding non-DP subsets ( Fig. 2b-g), with greater clonality 116 amongst CD8 + TILs than CD4 + TILs regardless of immune checkpoint expression status ( Fig. 2-117  figure supplement 2c,d). 118 For TILs obtained from HLA-A*02-positive patient mp26, with BRAF wt melanoma, a VDJdb search 119 identified three TCR clusters that included A*02-Melan-A aa26-35 -specific clonotypes (Fig. 2h). 120 Clonotypes matching to Melan-A-specific VDJdb entries were prominently enriched within the 121 CD8 + DP subset (about 1.5% of repertoire), compared to fresh-frozen tumor tissue (FFT) and to 122 the CD8 + non-DP subset (Fig. 2i). Most of these clonotypes belonged to TCR clusters, and such 123 Melan-A-specific cluster-related clonotypes constituted about 1% of CD8 + DP subset repertoire 124 (Fig. 2j). 125 Similar results were obtained for another BRAF wt HLA-A*02-positive patient, pt41 ( Fig. 2-figure  126 supplement 1b,c) where TCR repertoires of CD8 + DP and CD8 + non-DP subsets were 127 compared. We concluded that the CD39 + PD1 + fraction is enriched for large and convergent T cell 128 clones that are involved in an ongoing tumor-specific immune response, a substantial portion of 129 which are detectable via cluster TCR analysis. 130 To functionally confirm our findings, we used the CD137 (4-1BB) upregulation assay. Sorted DP 131 and non-DP TIL subsets from patient mp26 were cultured for two weeks and stimulated with 132 autologous monocyte-derived dendritic cells loaded with a TAA peptide mix. CD8 + CD137 high 133 subsets were subsequently quantified with flow cytometry and sorted for TCRꞵ library 134 preparation. 135 As shown in Figure 2k, the proportion of CD137 high cells was higher in cultured DP cells, but we 136 found no difference between T cells stimulated by TAA-loaded or control dendritic cells. At the 137 same time, TCRꞵ repertoire analysis revealed that the CD137 high fraction of TAA-stimulated CD8 + 138 DP-but not non-DP or control DP-cells was enriched with known Melan-A-specific clonotypes 139 (Fig. 2l). These clonotypes included a TAA-reactive TCRꞵ variant, CSARVGNQPQHF-TRBV20-140 TRBJ1-5, which was previously detected in cluster analysis of non-cultured CD8 + DP cells, and 141 variant CASSGGMGQPQHF-TRBV19-TRBJ1-5, which is homologous to another cluster. TAA-142 specific clonotypes cumulatively occupied ~13% of the CD137 high fraction of the cultured and 143 TAA-activated DP TILs. These results underscore the importance of TCR repertoire analysis of 144 responding cells, even in the apparent absence of a quantifiable difference between antigen and 145 control conditions, and demonstrate that CD137 marker analysis on its own is insufficiently 146 informative. 147 148 TCR cluster analysis facilitates optimization of TIL culturing conditions 149 We next investigated the effect of distinct TIL culture conditions on the expansion of tumor-150 reactive T cells, including concentration of IL-2 (stimulates expansion of both conventional and 151 regulatory T cells), and presence of IL-21 (plays a key role in the development and maintenance 152 of memory CD8 + T cells, inhibits T reg proliferation), anti-PD-1 antibody (blocks inhibitory 153 interaction of T cells' PD-1 with PD-L1), and IFNɣ (activates antigen presentation and supports 154 type 1 immune response). Four distinct TIL culture conditions included: IL-2 high , IL-2 low /IL-21, IL-155 2 low /IL-21/anti-PD-1, and IL-2 low /IL-21/anti-PD-1/IFNɣ ( Fig. 3-figure supplement 2a), where the 156 anti-PD-1 agent employed was nivolumab and the concentration of IL-2 in the low and high 157 conditions was 100 IU/mL and 3,000 IU/mL, respectively. For each condition, we analyzed TCRꞵ 158 repertoires of TILs independently cultured from 12 tumor fragments collected from patient mp26. 159 TCR clusters identified from all samples were joined and visualized along with the non-cultured 160 FFT, DP CD8 + , and non-DP CD8 + , repertoires (Fig. 3a). As a readout, we used the following: 161 i) normalized count of cluster-related clonotypes (Fig. 3b), 162 ii) cumulative proportion of the repertoire occupied by Melan-A-specific TCRꞵ clusters 163 (clusters predominantly comprising VDJdb-defined Melan-A-specific clonotypes) ( Fig.  164 3c), and 165 iii) the number of differentially-expanded clonotypes compared to pan-activating IL-2 high 166 culture conditions, and the proportion of such expanded clonotypes that were also 167 initially detected among CD8 + DP TILs (Fig. 3d). 168 The IL-2 low /IL-21/anti-PD-1 combination yielded the greatest number of cluster-related clonotypes 169 and the highest cumulative proportion of Melan-A-specific clusters out of all culture conditions, as 170 well as compared to initial non-cultured FFT samples (Fig. 3b,c). Addition of IFN did not further 171 enhance the expansion of potentially tumor-specific clones. 172 We utilized edgeR 46 software, which was initially designed for differential gene expression 173 analysis, to identify clonotypes that were significantly expanded in tumor fragment cultures in the 174 presence of IL-21 and IL-2 low (either with or without nivolumab and IFNɣ) compared to classical 175 pan-activating IL-2 high culture conditions. The IL-2 low /IL-21/anti-PD-1 combination yielded the 176 highest number of reproducibly expanded clonotypes, 60% of which were detected among initial 177 non-cultured CD8 + DP TILs (Fig. 3d, Fig. 3-figure supplement 1b). The overall count of such 178 CD8 + DP-matched clonotypes was also highest for this combination (Fig. 3e). These results show 179 the positive influence of PD-1 inhibition on the proliferative potential of CD8 + CD39 + PD-1 + T cells. 180 We also noted a slight increase in the number of CD8 + non-DP matched clonotypes which were 181 expanded in these same conditions-10 clonotypes, compared to three clonotypes in IL-2 low /IL-21 182 without nivolumab (Fig. 3d) phenotype 51 . We also noted a significant increase in the number of memory phenotype 52 196 CD127 high cells among CD4 + lymphocytes in IL-21 + conditions regardless of the presence or 197 absence of nivolumab ( Fig. 3-figure supplement 1e). Remarkably, CD8 + TILs displayed 198 significant CD127 high enrichment only upon simultaneous introduction of IL-21 and nivolumab 199 ( Fig. 3-figure supplement 1f). These results reveal the synergistic action of IL-21 and disruption 200 of PD-1-dependent signaling by nivolumab on expansion of CD8 + memory T cells. 201 Regarding overall proliferative potential, the highest T cell count was evident for IL-2 high 202 conditions, although we observed comparable numbers in the IL-2 low cultures (Fig. 3g). The 203 presence of IL-21 stifled IL-2-dependent TIL proliferation, as previously reported for human CD4 + 204 T cells 53 . On the other hand, IL-21 was favorable for expansion of CD8 + TILs, whereas IL-2 alone 205 favored CD4 + cell growth (Fig. 3h). 206 207 Discussion 208 Here we demonstrate that rational TCR clustering can be used to identify tumor-specific T cell 209 clones and estimate their relative enrichment among TILs. Starting with published TCR repertoire 210 data from melanoma tumors, we identified clusters of convergent TCR clonotypes, which 211 increased in numbers and total frequency after anti-PD-1 immunotherapy. We next found 212 significant enrichment of convergent TCR clusters in PD1 + CD39 + subpopulations of both CD4 + 213 and CD8 + TILs, which were previously shown to be enriched in terms of tumor reactivity 41,43,54 . 214 These findings are further supported by the data we obtained for cases where we know the HLA 215 context, as well as some TCR clonotypes of interest and their cognate antigens. A VDJdb 216 database search successfully identified TCR variants specific to TAA antigens in HLA-A*02-217 positive patients, where approximately half of the clusters could be matched to known Melan-A-218 specific sequences. 219 The cumulative frequency of such Melan-A specific clusters within the CD8 + DP population was 220 only slightly lower than the total frequency of all identified Melan-A specific TCRs, indicating that 221 our approach was able to identify most of the high-frequency tumor-specific clonotypes. 222 It should be noted that Melan A-specific T cells have unusually high frequency in melanoma 223 patients, and even in healthy individuals with HLA-A*02 allele 55-57 . Also, Melan A-specific T cell 224 clones in some cases represent predominant population in melanoma TILs 58 and are present at 225 high frequency in melanoma-infiltrated lymph nodes 59 . This may mean that using Melan A-226 specific clonotypes as a model is overestimating the sensitivity of our approach. On the other 227 hand, in our data, as well as in public datasets we used in this paper, there are other clusters of 228 comparable size that are not identified as Melan A-specific, which means that there are other 229 specificities with similar behavior. 230 It should be also noted that TCR clusters are an essential feature of a convergent immune 231 response that involves several homologous clonotypes. For those cases where a single T cell 232 clone dominates in response to a particular antigen and/or homologous "neighbors" are absent 233 due to the very low probability of convergent TCR generation 30 , cluster analysis may miss some 234 of the tumor-reactive TCRs. Nevertheless, our results demonstrate the overall power of this 235 approach, which is applicable in those situations where specific antigens are unknown. 236 We describe one potential implementation of our approach by using it to optimize conditions for 237 ex vivo TIL expansion. In particular, TILs cultured in IL-21 + conditions demonstrated the highest 238 number of cluster-related TCRs, which is indicative of a more prominent influence of antigen-239 driven TCR selection. We speculate that IL-21 exerts its influence in our culture system both at 240 the antigen-presentation stage (as MHCI-and, to a lesser extent, MHC II-restricted antigen 241 presentation by tumor cells occurs while tumor fragments are cultured ex vivo) 60                Pathomorphology Department. From each patient enrolled in this study, we also obtained 20-30 410 mL of peripheral blood before the surgery. Patient information is provided in Table S1. 411 412

Brief TIL culture 413
Freshly-resected tumor specimens were dissected into fragments measuring 1-3 mm in each 414 dimension. Several fragments were frozen in liquid nitrogen for further cDNA library preparation 415 and HLA-typing. Individual fragments were seeded into the wells of a 24-well tissue culture plate 416 with 2 mL of complete T cell cultivation medium (CM) supplemented with 5% heat-inactivated 417 human AB serum (PanBiotech, Germany) and 1,000 IU/mL IL-2 (Ronkoleukine, BiotechSpb, 418 Russia) . CM consisted of RPMI-1640 (PanEco, Russia), 25 µM/mL HEPES, pH 7.2 (PanEco, 419 Russia), 100 IU/mL penicillin, 100 µg/mL streptomycin, 10 µg/mL gentamicin, 1x non-essential 420 amino acids mix, 1x GlutaMAX , β-mercaptoethanol (0.55 µM) (all from Gibco, Thermo Fisher 421 Scientific, US) and 110 μg/m sodium pyruvate. On the fourth day of cultivation, TILs were 422 harvested, filtered through a 70-μm mesh, stained with fluorophore-labeled antibodies, and FACS 423 sorted. For the generation of TCR repertoire libraries, T cells were sorted directly into the RLT cell 424 lysis buffer (QIAGEN, Netherlands) and stored at -80ºC until used for RNA isolation. For 425 functional assays, TILs were sorted into 1.5 mL Eppendorf tubes with 0.5 mL RPMI-1640. Live-426 sorted T cells were seeded into wells of 96-well cell culture plate at 10 6 cells/mL in CM and 427 cultured for at least five days. Here, CM was supplemented with 1,000 IU/mL IL-2, 50 ng/mL IL-428 21 (SCI-STORE, Russia), 20 μg/mL nivolumab (Bristol-Myers Squibb, USA), and 10% autologous 429 patient-derived serum. Half of the media was replaced three times a week. One day before the 430 functional assays, all the media was replaced with fresh CM without interleukins or nivolumab. 431 432

Expansion of sorted T-cells 447
FACS-sorted T-cells were expanded using non-specific TCR-dependent stimulation with anti-448 CD3/CD28 Dynabeads (Thermofisher Scientific, USA). Beads were added into the cultivation 449 medium on the next day after seeding the cells, with 2 µl of bead solution per 10 5 cells. Sorted 450 TILs were expanded for 2 weeks. Beads were magnetically removed upon achieving desired cell 451 numbers. Before co-cultivation experiments, cells were allowed to "rest" for one day in interleukin-452 free CM supplemented with 10% autologous patient-derived serum (i.e., "resting" medium). 453 454 Peripheral blood mononuclear cells (PBMCs) were derived from patients' blood samples using 456 gradient centrifugation with Ficoll-Paque Plus (GE Healthcare). Briefly, 18 mL of whole blood 457 were diluted to 50 mL volume with sterile 1x PBS. Diluted blood was layered over the  Paque solution in 50 mL SepMate tubes (StemCell Technologies, USA), with 25 mL of diluted 459 blood per tube. SepMate tubes were centrifuged for 20 min at 1200 x g with brake off. Afterward, 460 buffy coats were collected and washed two times with 50 mL of sterile PBS. 461 462

Monocyte-derived dendritic cells cultivation 463
Autologous dendritic cells were generated as described in Ref. 62 . Briefly, CD14 + cells 464 (monocytes) were isolated from patients' PBMCs with a magnetic enrichment procedure using 465 anti-CD14 MicroBeads (Miltenyi Biotec, Germany). Then, monocytes were seeded into the wells 466 of 24-well tissue culture plates at 5 x 10 5 cells/well. X-Vivo-15 medium (Lonza, Switzerland) with 467 400 U/mL IL-4 and 800 U/mL GM-CSF was used for the differentiation of monocytes. On the 468 fourth day of cultivation, the medium was renewed, and dendritic cells were loaded with the mix of 469 melanoma TAA peptides (PepTivator Melan-A/MART-1, gp100/Pmel, and MAGE-A3 human; 470 Miltenyi Biotec) at a concentration of 600 nM each. The next day, loaded DCs were matured 471 using 1 µg/mL PGE, 10 ng/mL IL-1β, and 25 ng/mL TNF-α. Following 24 hours of maturation, 472 DCs were harvested and used for co-cultivation with T cells. 473 474 CD137 antigen-specific activation assay 475 FACS-sorted PD1 + CD39 + (DP) and non-DP populations after expansion and two days "rest" 476 without IL-2 were co-cultured with antigen-loaded autologous DCs at a ratio of 10:1 T cells:DCs. 477 The co-cultivation medium consisted of 1:1 CM plus AIM-V serum-free medium (Gibco, USA) 478 supplemented with 50 ng/mL IL-21. Both CD137 high and CD137 low T-cells were lysed with RLT 479 buffer for further RNA isolation and TCR library construction. The frequency of CD137 high cells 480 was measured for both CD4 + and CD8 + TILs. Antigen-specific activation was measured as a ratio 481 of CD137 high T cell frequencies in TILs co-cultured with antigen-loaded vs unloaded DCs. functional assays, we identified T reg s as CD4 + CD25 + CD127cells 63 and sorted them separately. 493 For functional assays, we sorted T cells into CD39 + PD-1 + double-positive (DP) and non-DP 494 (CD39 or PD-1 single-positive and double-negative) subpopulations, with CD4 + and CD8 + cells 495 together. For TCR library construction, CD4 + DP and CD8 + DP cells, as well as corresponding 496 non-DP populations, were sorted separately in order to evaluate individual properties of their TCR 497 repertoires. For the CD137 activation assay, T cells were stained with fluorescently-labeled CD4-498 BV510 (RPA-T4), CD8-Alexa-647 (SK1), CD137-PE (4B4-1) (BioLegend, Germany) antibodies. 499  , we selected columns with amino acid CDR3 sequence, TRBV, TRBJ, clonotype  545 frequency and read count. These pre-processed TCR repertoire tables were used as an input for 546 the ALICE algorithm. P gen of amino acid sequences was estimated using Monte Carlo simulation. 547 For each VJ pair, 5 million TCR sequences were simulated, and 20 iterations of the algorithm 548 were performed. Based on P gen , ALICE infers the number of highly similar CDR3 sequences (up 549 to 1 amino-acid mismatch) for each clonotype. Under the expectation that antigen-driven clonal 550 selection doesn't occur, TCRs with high probability of generation are expected to have a high 551 number of neighbors, while for low-probability TCRs there should be low number or no neighbors. 552 Based on this assumption, clonotypes with significantly higher numbers of neighbors than 553 expected from the recombination model (Benjamini-Hochberg-adjusted p < 0.001) are expected 554 to have come through convergent antigen-specific selection and expansion. The algorithm output 555 is formed as a list of these significant results referred to as "ALICE hits". 556 The number of ALICE hits strongly depends on the initial variability of the TCR repertoire ( Figure  557  2-figure supplement 1d). To account for this, "Normalized ALICE hits" metric was calculated by 558 dividing the number of ALICE hits by the number of clonotypes in initial input. 559 ALICE code can be found at https://github.com/pogorely/ALICE. 560 To visualize the resultant clusters of convergently selected TCRs we used the igraph function 65 . 561 This function utilizes the de Bruijn graph method to calculate the distance between amino acid 562 sequences of CDR3 regions. It creates a graph file in GML format where each node represents 563 an individual TCR clonotype and the distance between nodes is proportional to the difference 564 between CDR3 amino acid sequences. Edges connected nodes representing TCR clonotypes 565 with Hamming distance ≤1. Graphs were visualized using Gephi 0.9.2 network analysis 566 platform 66 . The size of the node represents the frequency of the corresponding TCR clonotype. 567 Upon construction of composite graphs including clonotypes from multiple TCR repertoires, 568 identical clonotypes from different TCR repertoires were displayed by separate nodes. 569 570

Matching cluster-related TCR clonotypes to VDJdb 571
We annotated TCR repertoire data using the VDJdb database of T cell receptors with known 572 specificity 38 . We assumed that TCRs of interest had the same specificity as TCRs from the 573 database if: i) CDR3 regions of compared TCRs differed no more than by one central amino acid 574 substitution, ii) substituted amino acids belonged to the same group based on their R properties 575 (polar, aliphatic, aromatic, positively/negatively charged), and iii) HLA-restriction of TCR 576 clonotypes from the database matched one of the patient's HLA alleles, if known. 577 TCR clusters consisting predominantly of TAA-specific clonotypes, but not clonotypes of other 578 specificities, were considered TAA-specific as a whole. VDJdb-unmatched members of the TAA-579 specific clusters were deemed to possess the same specificity as the whole cluster based on 580 structural similarity to VDJdb-matched clonotypes and were included in the subsequent analysis. 581 582

Analysis of differentially expanded clonotypes with edgeR software 583
We used a statistical approach implemented in the edgeR 46 package to identify TCRβ clonotypes 584 that were significantly expanded in bulk TILs of patients mp26 and mp34. We implemented 585 edgeR for comparison of TCR repertoires of TILs cultivated in experimental conditions 2-4 (described above) and TILs expanded in IL-2 high conditions. Six and four biological replicate 587 samples of each cultivation setting were used for the analysis of TCR repertoires from mp26 and 588 mp34, respectively. TCR clonotypes were deemed expanded if the false discovery rate adjusted 589 p value was <0.01 and the log 2 fold-change was >1. 590 591

Statistical analysis 592
Statistical analysis was performed using Graph Pad Prism 8.0 (GraphPad Software Inc., USA). 593 All data was reported as mean ± SD. The Shapiro-Wilk test was used for normality estimation in 594 all cases. Names of statistical tests and numbers of biological replicas in each comparison group 595 are provided in the figure legends.