Diversity of the intratumoral TCR repertoire
In contrast to TCRα alleles, which in approximately 20% of all T cells are both functionally recombined , TCRβ alleles are subjected to strict allelic exclusion. Because of this, TCR β-chain usage can be used as a straightforward means to analyze the diversity of TCR repertoires. Traditionally, such analyses have been performed by flow cytometry using TCR Vβ-segment specific antibodies , or by CDR3 size spectratyping . However, the resolution of these approaches is only modest, as TCR clonotypes using the same TCR Vβ-segment or with the same CDR3 length cannot be distinguished. With the development of next generation sequencing (NGS), techniques have been developed that can reveal the nucleotide sequences of all TCRβ CDR3 sequences present within a given T-cell population. Because of the immense read depth that can be achieved, NGS of TCR repertoires allows the quantitative detection of even low-frequency TCR sequences [note that proper filtering to exclude sequence errors  is required]. Furthermore, because of the high diversity of the TCRβ CDR3 repertoire, the sequences obtained will in most cases represent individual TCR clonotypes (a noted exception are TCRs sharing a common/public TCR β-chain but distinct TCR α-chains). NGS sequencing of TCRβ CDR3 has first been used in a study that analyzed the TCR distribution among various T-cell compartments in a healthy individual . More recently, the technology has also been implemented to analyze TCR repertoires in disease settings, such as TCR reconstitution upon allogeneic hematopoietic stem cell transplantation .
The TCR repertoire diversity of different T-cell subsets (e.g. CD8+ effector T cells) found within different human tumors has not been studied systematically thus far. As exceptions, Sherwood et al.  assessed the TCR diversity of tumor-infiltrating lymphocytes (TILs) derived from colorectal cancer and compared it to the TCR repertoire of mucosa-infiltrating T cells. This study revealed that the TCR repertoire diversity in colorectal cancer TILs is more restricted compared with TCR repertoires found among mucosal T cells and variable between patients . In another study, the same group analyzed the TCR repertoire of TILs in ovarian cancer, showing them to be largely distinct from circulating T cells . It will be interesting to assess whether the steady-state diversity of intratumoral TCR repertoires correlates with clinical prognosis, and, more importantly, whether (changes in) TCR repertoire diversity among intratumoral T cells may be a predictive marker for a clinical response following immunotherapeutic interventions, such as TIL therapy or T-cell checkpoint blockade. Given the rapid progress in the development of high-throughput sequencing technologies, the use of TCRβ CDR3-sequencing for patient selection would seem a realistic option with regards to time and financial requirements.
Tumor reactivity of the intratumoral TCR repertoire
The analysis of TCR repertoire diversity by bulk TCR gene sequencing reveals the total number of TCRs present in a T-cell population and the extent of clonal dominance within such populations. As outlined above, such data may prove valuable in the context of biomarker identification. Furthermore, because the TCRβ sequences obtained function as genetic barcodes, such information may also be used to describe kinship between different intratumoral T-cell subsets . As a downside, this method does not identify the TCRαβ pairs of individual T cells, implying that the information cannot be utilized to analyze or reconstruct the tumor reactivity or antigen specificity of the intratumoral TCR pool.
Understanding which fraction of intratumoral T cells is reactive to tumor cells and which determinants these cells recognize is of obvious interest. Thus far, the tumor-reactivity of the intratumoral T-cell pool has primarily been assessed in two ways. First, the ability of bulk tumor-resident T cells to recognize HLA-matched allogeneic or (preferably) autologous tumor has been studied in functional assays from the 1980s until today, primarily for melanoma [17-19]. These assays give a straightforward overview of the functional capacity of the T cells on a population level and this type of analysis has inspired the development of TIL therapy [20, 21]. Second, in several studies, T-cell clones generated from intratumoral T cells have been utilized to obtain TCR genes that could subsequently be shown to confer tumor reactivity after TCR gene transfer [22-24]. Furthermore, TCRs obtained in this fashion have been utilized in the first clinical studies of TCR gene therapy [25, 26].
As a downside to these approaches, both these strategies will not capture the activity of T cells that do carry a tumor-reactive TCR but have been rendered anergic. Furthermore, since these studies commonly utilized in vitro expanded T-cell material, they are restricted to those T cells that can expand in vitro, thereby likely resulting in a significant bias or even precluding analysis altogether (e.g. for tumor types for which such T-cell expansion cannot be achieved). Therefore, while these studies have provided ample evidence for the presence of tumor-reactivity within the intratumoral pool of T cells, by their nature they cannot provide an unbiased enumeration of the ‘true’ fraction of tumor-reactive T cells within human tumors.
To obtain a better understanding of the tumor reactivity within intratumoral TCR repertoires, it would be of value to determine the frequency of tumor-reactive TCRs among different intratumoral T-cell subsets without a requirement for (prolonged) in vitro expansion. Toward this goal, we propose to isolate large libraries of TCRαβ gene pairs directly from intratumoral T cells. Identified TCR pairs can then be introduced in peripheral blood lymphocytes by gene transfer, to allow assessment of autologous tumor recognition independent of the parental T-cell phenotype. Such analyses could be focused on intratumoral CD8+ T cells but may also be of interest for the heterogeneous population of intratumoral CD4+ T-cell subsets, for which studies on tumor-reactivity are particularly limited at this moment. This type of autologous TCR gene transfer experiments would provide insights into several aspects of the intratumoral TCR repertoire. First, in analogy to the analysis of TCR repertoire diversity (see above), the frequency of tumor-reactive TCRs may represent a prognostic or predictive clinical marker. Second and somewhat related, these analyses would establish whether the previously described correlation between tumor infiltrating lymphocyte numbers and clinical prognosis can be explained by the tumor-recognition capacity of the tumor-resident TCR repertoire. Finally, such studies would make it is feasible to explore whether the frequency or diversity of tumor-reactive TCRs found in intratumoral TCR repertoires correlates with the size of the cancer anti-genome, as for instance reflected by the mutational load of tumors.
The isolation of the large TCR libraries that are required for such experiments has become a realistic option following the development of a number of strategies to identify TCRαβ gene pairs [27-29]. First, the sequencing of cDNA generated from single T cells has proven a viable strategy to identify TCRαβ pairs [27, 28]. To date, these strategies have not been utilized to dissect intratumoral TCR repertoires, but this will likely prove feasible. In addition, our laboratory has developed an independent method to infer TCRαβ sequences directly from the genome (rather than RNA) of T cells that has already been utilized to recreate TCR pairs from intratumoral T cells. This approach, called ‘TCR gene capture’ utilizes an RNA-bait library targeting the TCRαβ and TCRβ loci to specifically select and sequence genomic fragments encoding the TCR sequences of T cells  (Fig. 1).
Figure 1. Schematic representation of T-cell receptor (TCR) gene capture approach. Genomic DNA is extracted from (oligo)clonal T-cell populations of interest, the DNA is sheared into small fragments (average length 500 basepairs). Using an RNA-bait library targeting all functional TCR V- and J-segments on both the TCRα and TCRβ loci, all DNA fragments encoding TCR sequences are selected and subsequently analyzed by paired-end Illumina sequencing. Using bioinformatic tools , TCR CDR3 sequences are identified in the resulting sequencing data.
Download figure to PowerPoint
TCR gene capture has proven a versatile tool to analyze TCR repertoires within intratumoral CD8+ T cells. First, with the goal to identify TCRs that may be used clinically, we have utilized the technique to assemble TCR libraries from single T cells expanded in vitro for a short period. Using MHC-multimer selected T cells as input, this resulted in the identification of a large panel of TCRs against shared tumor-antigens (we isolated 21 different TCRs against nine distinct Cancer/Germline antigens). In addition, we isolated a library of 19 different tumor-reactive TCRs from a library of clonal tumor-reactive T-cell populations from a melanoma patient without further knowledge of their antigen specificity. The latter data illustrate that the tumor-reactive TCR repertoire of intratumoral CD8+ T cells can be broad and that it is feasible to rapidly assemble a library of patient-specific tumor-reactive TCRs. Second, since TCR gene capture supplies quantitative read counts for all TCR α- and β-sequences within the sample, TCRαβ pairs can be directly identified within oligoclonal T-cell populations through matching of TCR α- and β-sequences that occur with similar frequency. For polyclonal T-cell populations, such ‘frequency-based matching’ will be precluded by the occurrence of multiple TCR clonotypes at the same frequency. However, we have successfully unraveled the TCR repertoire of several intratumoral CD8+ T-cell populations specific for shared melanoma antigens (Meloe-1, MAGE-A1, MAGE-10, TAG-1, LAGE-1) by this approach. In these experiments, in which bulk MHC-multimer positive T-cell populations from TILs were used as input for TCR gene capture, the TCR repertoire of the tumor antigen-specific T-cell populations analyzed commonly was restricted to only 1–5 TCRs. Comparison of the TCR diversity within individual tumor antigen-specific T-cell populations and within the entire tumor-reactive T-cell pool may in future studies perhaps be used to provide a first estimate of the breadth of the antigen repertoire that is recognized.
Using TCR gene capture, it has also proven feasible to identify dominant TCRs within the intratumoral tumor-reactive T-cell population without knowledge of their antigen specificity. Specifically, profiling of the TCR repertoire of three TCRVβ subpopulations among intratumoral, tumor-reactive CD8+ T cells showed that the TCR repertoire in each of these subpopulations was markedly restricted: the two most abundant TCRs comprised at least 75% of all functional TCRαβ CDR3 sequences. This recognition allowed the straightforward identification of tumor-reactive TCRαβ pairs.
Both the recently developed single cell-based approaches and the TCR gene capture strategy described above will be of substantially value to increase our understanding of the intratumoral TCR repertoire in human cancers. Nevertheless, the development of technologies that can sample the repertoire of intratumoral TCRαβ pairs at even greater depth (e.g. revealing the identity of many thousands of TCRαβ pairs) remains an important goal. The recent description of an emulsion-PCR based approach to identify TCRαβ pairs from single cells that are contained within droplets is of interest . While further technical developments will clearly be required to allow an unbiased analysis of TCR repertoires with this type of technology, it does offers the potential to advance capacities for (intratumoral) TCR profiling well beyond the currently available medium-throughput approaches.
Antigen specificity of the intratumoral TCR repertoire
In particular for intratumoral T cells in melanoma, convincing evidence for tumor reactivity is provided by the clinical responses seen in patients that are treated with autologous TIL products. Furthermore, the fact that such clinical responses have also been observed in patients treated with CD8+ enriched TIL cell products  maps at least part of the clinically relevant tumor reactivity toward the cytotoxic T-cell subset. At present, our understanding of the critical cancer regression antigens in TIL therapy is limited. A better understanding of the role of different antigens in the clinical responses in patients receiving TIL therapy (or other non-antigen directed immunotherapies, such as blockade of CTLA-4 or PD-1) would possibly enable the ‘engineering’ of anti-tumor immunity in the large group of patients who do not benefit from current immunotherapies. Because of this, a large effort has been made over past few years to reveal the antigen-specificity of tumor-specific CD8+ T-cell response in patients receiving immunotherapy.
As demonstrated by pioneering work by Altman and Davis , fluorescently labeled multimeric pMHC-complexes can be used to detect antigen-specific T cells with high sensitivity and independent of their functional capacities. However, the large-scale screens that are required to assess the antigen specificity of the intratumoral T-cell pool by MHC multimer-based analyses have only become feasible with two subsequent developments: (i) the generation of a high-throughput pipeline to obtain the pMHC collections required for such monitoring and (ii) the design of experimental approaches for multiplexed analysis that allows comprehensive screens to be performed with clinically realistic amounts of biological material.
To obtain very large collections of pMHC-complexes in a high throughput fashion, we have developed (and now routinely use) a peptide exchange technology in which collections of pMHC complexes of interest can be generated in a 1 h procedure [33, 34]. This approach is based on the use of pMHC-complexes that carry a peptide ligand that cleaves itself upon UV exposure, and by exposing such conditional pMHC complexes to UV light in the presence of peptide ligands of interest. Peptide exchange technology is now available for around two dozen HLA class I alleles ([33-35], M. Toebes and L. van Dijk personal communication), also through work from the Grotenbreg lab . As a side note, recent work demonstrates that MHC multimer-based detection of antigen-specific T cells is highly sensitive to minor sequence variation between HLA subtypes (M. M. van Buuren, F. E. Dijkgraaf, C. Linnemann, M. Toebes, C. X. L. Chang, J. Y. Mok, M. Nguyen, W. J. E. van Esch, P. Kvistborg, G. M. Grotenbreg and T. N. M. Schumacher, manuscript submitted). The requirement for proper matching between patient HLA alleles and HLA tools used for immunomonitoring that is revealed by these data will make it essential to further expand the HLA-based toolkit in the years to come.
To allow the analysis of T-cell reactivity against many (potential) tumor antigens in MHC-multimer-based screens, both our group and the Davis laboratory have developed the concept of combinatorial MHC-multimer staining [37, 38]. In this approach, each pMHC complex is coupled to a unique combination of fluorochromes or more recently lanthanides (see below). The use of such coding schemes then allows one to define the pMHC specificity of individual T cells/TCRs by the combinatorial code that they bind. In addition to offering the possibility of multiplexed analysis, this approach also markedly increases the sensitivity of pMHC-based T-cell detection, enabling the unambiguous detection of low-frequency antigen-specific T-cell populations . The fluorochrome-based combinatorial coding schemes that to date have been used to analyze tumor-specific T-cell responses allow analysis of around 30 T-cell responses within a single sample. Furthermore, as an interesting extension of this approach, Newell and Davis recently demonstrated that the use of combinatorial coding schemes in MHC multimer mass cytometry can allow multiplexed analysis at an even higher complexity .
In recent studies, the combination of UV-mediated peptide exchange and combinatorial coding has been used to study the T cell/TCR repertoire in TIL products used for adoptive T-cell therapy of melanoma patients [40, 41]. In work that analyzed T-cell reactivity against a panel of around 150 shared melanoma-associated tumor antigens (TAAs), reactivity patterns in more than 50 TIL products were analyzed. While this work was primarily restricted to the HLA-A2 allele – few shared TAA are known for most other HLA class I alleles – several important conclusions can be drawn from these studies. First, in every TIL culture, an essentially unique pattern of antigen reactivities was found. Furthermore, for the few antigen-TIL product combinations for which this was examined, expression of a given melanocyte differentiation antigen/cancer-germline antigen was generally accompanied by the presence of T cells specific for this epitope, suggesting that tumor antigen expression may in most cases be noted by the immune system (note that a larger data set will certainly be required to test this notion in a rigorous manner). These T-cell monitoring data extend the concept of ‘tumor heterogeneity’ – described above for tumor genomes and for T-cell infiltrates – to the TCR specificities of the intratumoral CD8+ T-cell pool. A second important observation made in these studies has been that the frequency of antigen-specific T cells that were detected for the shared TAA used in these studies was very low (median of less than 1% for all HLA-A2-restricted T-cell responses detected per TIL product). Even taking into account the fact that T-cell reactivity against the five other possible HLA alleles was not measured, these data suggest that reactivity against shared TAA may only explain part of the composition of clinically used TIL products. As a first explanation for this discrepancy, the fraction of non-tumor reactive ‘bystander’ TCR specificities in TIL products may in many cases be high. As a second explanation, a large fraction of the tumor-specific TCRs within the intratumoral repertoire may recognize highly patient-specific antigens. Early evidence for a (perhaps small) contribution by bystander cells has been obtained by the detection of low frequencies of CMV and EBV-specific T cells in TIL products. However, the contribution of bystander cells may be addressed in a more definitive manner by the isolation of large TCR libraries from the intratumoral TCR repertoire and their subsequent characterization with regards to tumor-reactivity (see section ‘Tumor-reactivity of the intratumoral TCR repertoire’). With respect to the second possibility, the particularly high mutational load of melanoma and other tumors, such as lung cancer, raises the question whether the intratumoral T-cell repertoire could contain a variety of TCRs that recognize neo-antigens derived from tumor-specific mutations, thereby forming a highly personalized, tumor-reactive TCR repertoire.
Recent and exciting studies in animal models by the Sahin and Schreiber groups [42, 43] have demonstrated that cancer exome sequencing data may be utilized to analyze T-cell reactivities against neo-antigens formed by tumor-specific mutations. The ability to dissect T-cell reactivity against neo-antigens on the basis of human cancer exome data has now also been demonstrated by others and us. Rosenberg et al.  have utilized cancer exome sequencing data and recognition of target cells loaded with putative neo-antigens by autologous CD8+ TILs to uncover neo-antigen-specific T-cell reactivity within TIL products. On average, two T-cell responses against neo-antigens were identified in the three patients analyzed. As neo-antigen-specific T-cell reactivity was only analyzed for some of the HLA alleles expressed by these patients, these data suggest that TILs may potentially recognize a series of patient-specific antigens. However, as this study focused on patients that experienced a particularly strong clinical response upon TIL therapy, it is possible that broad neo-antigen-specific T-cell reactivity may not be invariably present in melanoma, and it will be important to extend these studies to address this issue. In parallel work, our group has provided proof of concept for the combination of cancer exome sequencing and MHC-multimer technologies to reveal T-cell responses against patient-specific neo-antigens arising from genomic mutations . Comparison of whole exome sequencing data of a melanoma tumor with that of autologous healthy tissue revealed more than 1000 non-synonymous changes resulting in altered open reading frames. These data were then combined with RNA-expression data to predict potential neo-epitopes for four of the HLA class I alleles expressed by this patient. When TILs from this patient were then screened with a library of MHC-multimers containing these putative T-cell epitopes, two neo-antigen-specific T-cell responses were identified: a low-frequency response against a mutated peptide of ZNF462 gene (0.003% of CD8+ T cells) and a dominant response (3.3% of CD8+ T cells) against a neo-antigen derived from the ATR DNA damage response gene. The magnitude of the T-cell response against the ATR neo-antigen was of a considerably higher magnitude than virtually all of the T-cell responses against shared antigens seen in our previous analyses of TILs [40, 41]. While more data are certainly required, these data suggest that (some) T-cell responses against neo-antigens may perhaps be of a higher magnitude than T-cell responses against shared (self) antigens, due to their foreign nature. Comparison of the tumor recognition potential of neo-antigen and shared antigen-specific T cells will also be of importance to address the relative importance of the two antigen classes.
Analysis of peripheral blood samples pre- and post-treatment with anti-CTLA-4 showed an approximately fivefold increase in the frequency of ATR-specific CD8+ T cells upon treatment. This increase in frequency, which coincided with a partial tumor regression within this patient, is consistent with the possibility that the T-cell response against the mutated ATR peptide may have been therapeutically meaningful, but the evidence is obviously indirect. By the same token, Lu et al.  used cDNA library screening to reveal a T-cell response against a mutated peptide of PPP1R3B in a metastatic melanoma patient, and could show that a long-term T-cell response against this epitope was present in this patient who experienced a durable complete response after TIL therapy.
The recent studies that have started to link cancer exome data to tumor-specific T-cell responses will likely still only sketch a fraction of the patient-specific intratumoral TCR repertoire. As a first issue, imperfection in epitope predictions will lead investigators to miss neo-antigens, in particular for the less well-studied HLA alleles. Furthermore, in addition to neo-antigens that arise as a consequence of single nucleotide variants or insertions/deletions in known open reading frames, it is highly likely that T-cell epitopes arising from alternative translation events [47, 48] will also form part of the patient-specific cancer anti-genome. As precedent for the latter, minor histocompatibility antigens, which bear some similarity with neo-antigens in solid tumors (both generally originate from single nucleotide differences), can also be derived from alternative open reading frames [49, 50].
A number of developments can be foreseen that will make the identification of patient-specific antigens recognized by the intratumoral TCR repertoire more efficient. First, an increase in the quality/coverage of sequence data and in particular the quality of epitope predictions will facilitate the identification of neo-antigens. With respect to the latter, especially for the less common HLA-A and -B alleles and for the HLA-C alleles, the quality of prediction algorithms may readily be increased by the generation of more input data. As a less biased approach (that altogether avoids the need to predict T-cell epitopes), systems that allow the efficient expression of the entire set of tumor-specific mutations could be valuable, in particular to identify T-cell epitopes from non-canonical sources (e.g. alternative translation events). Finally, the moment it becomes feasible to (roughly) predict the epitope recognized by a TCR on the basis of TCR sequence data (at present a far removed goal), the combination of intratumoral TCR sequence data and cancer exome data could form a strategy to identify patient-specific antigens in a manner that is altogether independent of immunological analyses.