The T cell receptor repertoire of tumor infiltrating T cells is predictive and prognostic for cancer survival

Tumor infiltration by T cells is paramount for effective anti-cancer immune responses. We hypothesized that the T cell receptor (TCR) repertoire of tumor infiltrating T lymphocytes could therefore be indicative of the functional state of these cells and determine disease course at different stages in cancer progression. Here we show that the diversity of the TCR of tumor infiltrating T cell at baseline is prognostic in various cancers, whereas the TCR clonality of T cell infiltrating metastatic melanoma pre-treatment is predictive for activity and efficacy of PD1 blockade immunotherapy.

interest and could provide some insight.
Reviewer #3: Remarks to the Author: Valpione and colleagues investigate the TCR repertoire in melanoma biopsies from newly diagnosed patients and attempt to find correlates based on the hypothesis that the TCR repertoire of TIL could determine the disease course (clinical outcome). 16 patients recruited from 3 centers (Manchester, Padova, Meldola) received initial systemic therapy with anti-PD1 agents. The authors state that there were no differences in TIL numbers, TIL clones/clonotypes, TCR diversity, or TCR clonality (Extended data Fig 1). In addition, no correlation was found between OS and serum LDH, TIL numbers, TIL clones, TCR diversity, #SNVs, or PD-L1 expression levels (Extended data table 1). They did, however, find an association between OS and TCR clonality (Figure 1). Validation was assessed using datasets from Riaz (2017) and Yusko (2019) with melanoma patients (n=106) treated with either anti-PD1 or the combination anti-CTLA4/anti-PD1. The investigators then examined TCGA data base and found that TCR diversity (not clonality) as a prognostic marker correlated with OS for several cancers including BRCA, LUAD, TGCT, KIRK, THYM (Fig 2). The authors conclude: "high TIL/Tc clonality identifies which patients will mount an effective anti-PD1induced immune response" and that this observation has important clinical implications. This is an interesting observation; however, it is based on modest data with no mechanistic insight. Several major concerns are listed below.
1. The authors studied 16 patients with metastatic melanoma. Please describe the clinical, pathologic, radiographic, and demographic features of these individuals. At first glance, the sample size appears exceedingly small and if this association (OS and high TCR clonality) is true, it should be broadened to be truly representative of newly diagnosed patients to convince one that others can build upon these results. The limited information in the data summary is inadequate. Figure 1 panel B reports the "total unique TCR sequences" for 6 data points from extranodal sites and 10 data points from nodal sites. The median number of unique TCR sequences in each group is 10 which is dramatically lower than expected based on the literature (and our own experience). For example, most tumor biopsies contain thousands or tens of thousands of unique clonotypes based on TCR VB CDR3 sequences obtained from melanoma patients (Pasetto CIR 2016;Riaz Cell 2017) or other histologies (Wu Nature 2020) using Adaptive Biotechnologies method of bulk sequencing gDNA or 10x Genomics sc mRNA platform. The low number of unique TCR clonotypes for each patient probably reflects a sampling bias or alternatively, a sequencing artifact due to their methodology. Finally, the authors should be more precise in their definition of a TCR clonotype and the primary sequencing data for all 16 subjects should be provided.

Data shown in Extended Data
3. The rationale to include TCGA data ( Figure 2) from other cancer types is puzzling and some would argue it is extraneous to the observation made in figure 1. Namely, TCR clonality at baseline is predictive for clinical benefit after anti-PD1 therapy as first-line therapy. It would appear logical to investigate non-melanoma patients with newly diagnosed cancer administered anti-PD1 (either on study or as standard of care) to provide confirmatory evidence. The most pressing question for investigators in the field concerns the specificity of the high frequency TCRs evident in some patients and this was not addressed in the current study.
Minor Points 1. A point worth mentioning relates to an assumption implicit in the authors conclusion that "an effective anti-PD1-induced immune response" somehow equates with improved OS. The data in Figure 1 concerns OS as a clinical endpoint. The authors show no data related to the quality, composition or magnitude of the T cell response after anti-PD1 treatment.
conclusion as many stage II (and a small percentage of stage I) patients will not develop nodal regional disease but rather proceed to stage IV metastatic disease that can be biopsied to assess TCR clonality prior to first line anti-PD-1.

NCOMMS-20-30697-T: Responses to reviewers' comments.
We are grateful to the reviewers for the time spent on our manuscript. Their comments were both insightful and helpful, and in responding, the quality of the manuscript is much improved. Below we provide our point-by-point responses to each comment (for ease the page and line information refers to the tracked version of the revised manuscript). Overall, I believe this work is convincing and scientifically sound. It advances knowledge in the field of tumour immunology, it is clinically relevant and will be informative for scientists in the field even though it lacks a more comprehensive insight into the underlying immune response mechanisms.

REVIEWER COMMENTS
Response. We thank Reviewer #1 for their appreciation of our work. We are delighted that the reviewer found our work to be convincing and scientifically sound and are particularly pleased that they concur that it advances knowledge in tumour immunology, is clinically relevant and informative. Response. We thank the reviewer for this and have now changed the format from a brief communication to that of an article, which has allowed us to discuss our data more comprehensively.

Comment 2.
A more detailed presentation of the TCR analysis should be included i.e. average clonotype sizes, spectratypes, VJ usage profiles, read counts etc.
Response. Thank you for the suggestion. In response we have added new descriptive and comparative analyses of the TCR repertoire in the cohorts studied and include these in new  Response. Our apologies for the lack of detail. In response, we explain in the Methods (page 22 line 395) the methodology applied to identify the grouping cut-off for the survival curves and we do however wish to highlight that, although the categorisation of TIL/TC TCR clonality and diversity was necessary for the Kaplan-Meier curves, to avoid the potential bias introduced by cut-offs, all regression analyses were performed retaining the TCR metrics as continuous variables: this is now better explained in the Methods (22 line 396). Figure 1G, the nomogram needs more detail on its construction and interpretation in the main text.

Comment 4. In
Response. We thank the reviewer for the suggestion, and in response we have added details about the nomogram construction and interpretation in the Methods section (page 21 line 381 and page 22 line 385) and have additionally added an example of its use in the Results (page 6 line 118, Extended Data Fig 6).

Methods.
Response. Apologies for the omission; more details have now been added to the Methods (page 19 line 327 and 20 line 351).

Reviewer #2 (Remarks to the Author):
General comment. In  Response. We are delighted that the reviewer found our results on TCR clonality to be convincing and that they agree that our data can predict patient response to anti-PD1 in melanoma. We are pleased that they agree that we show that TCR diversity can predict overall survival in numerous cancers and that our findings could impact which patients will receive PD1-based immunotherapy.

Comment 1. Line 44. Define LDH when first used.
Response. We thank the reviewer for spotting this, and have added the definition (page 4 line 71).

Comment 2.
Line 58-60: Since clonality is discussed In the statement, "Notably, whereas TIL/Tc… with better OS (Fig. 1e,f).", either plot clonality in Figure 1e or include Extended Data Response. We thank Reviewer #2 for this observation. In response, we have made it clearer in the text that although the categorisation of TIL/TC TCR clonality and diversity was necessary to generate the Kaplan-Meier curves, to avoid potential bias introduced by cutoffs all the regression analyses were performed retaining the TCR metrics as continuous covariates (page 22 line 396). The reviewer's observation also prompted us to comment in the Introduction that although anti-PD1 drugs have revolutionised the treatment of melanoma, only about 20-55% of treated patients benefit from these drugs (page 3 line 39). Figure 1f: Please reconcile the color indicated in the figure legend with the color in the graph since the graph appears pink and not purple.

Comment 4.
Response. Apologies for the mistake and thank you for highlighting it. It is now corrected (now in Figure 2b and 3b-g legends).

Comment 5. A comparative analysis of the pre-PD1 versus post-PD1 TCR repertoire (if available) would be of interest and could provide some insight.
Response. Thank you for this interesting suggestion. In response, we performed the analysis for the neoadjuvant therapy cohort, because sequential samples were available.  1). In addition, no correlation was found between OS and serum LDH, TIL numbers,  -j and note that, despite the overall survival data for these cohorts not being available, this new data confirms that that high TIL/Tc clonality in the pre-treatment biopsies correlated with benefit from therapy. Figure 1  about this conclusion as many stage II (and a small percentage of stage I) patients will not develop nodal regional disease but rather proceed to stage IV metastatic disease that can be biopsied to assess TCR clonality prior to first line anti-PD-1.

Comment 2. Data shown in Extended Data
Response. We thank Reviewer #3 for this comment, and in response we have rephrased this section and also commented on the fact that adjuvant anti-PD1 clinical trials are currently ongoing in stage II melanoma (page 7 line 151).

Reviewers' Comments:
Reviewer #1: Remarks to the Author: In this revised version of the manuscript the authors have adequately addressed my previously raised concerns. I support the publication of this work.
Reviewer #2: Remarks to the Author: The authors have satisfactorily addressed this reviewer's concerns in their revised manuscript.
Reviewer #3: Remarks to the Author: I continue to have reservations about the use of RNA-Seq to infer TCR repertoire especially when targeted methods such as Adaptive Biotechnologies platform are readily available to reliably capture the near-complete BV repertoire (see P. Barennes et. al. Sept 2020 Nature Biotech). In addition, the lack of mechanistic insight is a weakness.
Comment 1 requested additional details regarding the study population of 16 patients. In response, the authors provided Table 1 which provides limited clinical details. In addition, Figure  1e-j is added to the revised manuscript which contains data from 3 published melanoma cohorts that reported ORR yet lacks OS endpoint.
Comment 2 addressed the small number (median 10.5) of clonotypes inferred using ImReP (developed for Ig clonotypes) and the authors responded by adding new Figures 3 and 4 which compares RNA-seq (n=16 samples) to targeted BV sequencing (Adaptive Bio platform) of samples from ref 3,4,6,7. A better comparison would have been to compare RNA-seq (ImReP) and a targeted BV assessment using the same 16 samples.
Comment 3 addressed the claim that TCR diversity is prognostic for OS in several other solid tumors. It is unclear how this observation made using TCGA samples is relevant to the melanoma work.

NCOMMS-20-30697-B; Valpione et al. Responses to reviewers
1 We are grateful to the reviewers for the time spent on our manuscript. Their comments from the first review were both insightful and helpful, and in responding, the quality of the manuscript improved. We are delighted that Reviewers #1 and #2 have recommended publication of the revised manuscript. We are however somewhat surprised by the new requests from Reviewer #3, which arose from the second round of review. Specifically, addressing these new requests would not affect the main ideas or data quality in our manuscript and below, we outline our point-by-point responses to the specific requests. We ask your forbearance in this matter, because if we did address the new requests, it would not strengthen the manuscript any further, but it would cause us unnecessary delay and significant expense.

Reviewer #1 (Remarks to the Author): Comment:
In this revised version of the manuscript the authors have adequately addressed my previously raised concerns. I support the publication of this work. Response: We thank Reviewer #1 for suggesting the improvements in the first round of review, and we are delighted that the reviewer is satisfied with our responses and has recommended publication of the manuscript.

Reviewer #2 (Remarks to the Author): Comment: The authors have satisfactorily addressed this reviewer's concerns in their revised manuscript.
Response: Again, we are grateful to Reviewer #2 for the suggestions in the first round of review and are delighted that this reviewer too is satisfied with our responses and has also recommended publication.

Reviewer #3 (Remarks to the Author):
Comment: I continue to have reservations about the use of RNA-Seq to infer TCR repertoire especially when targeted methods such as Adaptive Biotechnologies platform are readily available to reliably capture the near-complete BV repertoire (see P. Barennes et. al. Sept 2020 Nature Biotech). In addition, the lack of mechanistic insight is a weakness. Response: We thank Reviewer #3 for their continued consideration of our manuscript. However, we wish to emphasise that the point of our study is to show that the clonality of tumour infiltrating T cells is predictive of benefit from anti-PD1 therapy. We do not seek to endorse RNA-Seq-based algorithms as an approach to reconstruct TCR sequences of tumour infiltrating T cells, as this has already been done elsewhere by those who developed the pipeline. Moreover, an in-depth mechanistic analysis is beyond the scope of our manuscript, the purpose of which is to explore how personalised immunotherapy could be delivered.
More pertinent, please note that 170 of of 186 samples we present were analysed on the Adaptive Biotechnology platform that Reviewer #3 favours. The results consistently indicate that tumour infiltrating T cell clonality predicts outcome to immune-checkpoint blockade. Specifically, using two independent cohorts, we show that high tumour infiltrating T cell clonality is predictive for better OS in 122 patients, and moreover that, in 3 additional independent cohorts it is predictive of better radiological response in 61 patients. We wish also to highlight that in the revised manuscript, and in response to the requests of the other reviewers, we provide a comparison of the T cell repertoire metrics inferred using the two TCR analysis platforms and that the distribution of the predictive biomarker -the tumour infiltrating T cell clonality -is not skewed in the two approaches (Extended Data Fig 3a). We fail therefore to understand how repeating the Adaptive Biotechnologies analysis on 16 of the 186 samples could affect the key concepts of our manuscript or, indeed, increase the quality of the data.

Comment: Comment 1 requested additional details regarding the study population of 16 patients.
In response, the authors provided Table 1 which provides limited clinical details. In addition, Figure 1e-j is added to the revised manuscript which contains data from 3 published melanoma cohorts that reported ORR yet lacks OS endpoint. Response: Please note that the clinical details provided in Table 1 are presented in accordance with commonly reported descriptive variables for melanoma patient clinical data. We are happy to add further variables if we receive specific and justified requests, but they would need to be within the limits of patient anonymisation requirements (GDPR). Response: This is a new request for us to re-analyse our original 16 test samples with the Adaptive Biotechnology platform. However, we cannot see a compelling justification to do so, because 4 out of the 5 cohorts, representing 170 of the 186 patient samples we present were analysed using the Adaptive Biotechnology platform that Reviewer #3 favours. Note also that the results across the 5 cohorts consistently indicate that high T cell clonality in pre-treatment biopsies predicts better outcomes to immune-checkpoint blockade. In particular, the OS findings on the 16 patients analysed by RNA-seq have been fully validated by the external cohort of 106 patients analysed by the Adaptive Biotechnology platform (Figure 1c,d).
We understand of course that with evolving technologies, more sensitive and accurate solutions will eventually replace existing state-of-the-art platforms. However, an Adaptive Biotechnology re-analysis of only 16 of our 186 samples (less than 10%) will not affect the main concept of the study or the data quality of our manuscript. It is therefore unclear to us what the additional analysis would add, apart from causing us unnecessary very high costs and time delays in this highly competitive field. Moreover, performing this analysis is made all the more complicated by the current COVID lockdown situation.

Comment: Comment 3 addressed the claim that TCR diversity is prognostic for OS in several other solid tumors. It is unclear how this observation made using TCGA samples is relevant to the melanoma work.
Response: We wish to emphasise that the purpose of our analysis of TCGA cohorts is to support the idea that tumour infiltrating T cell diversity is prognostic in tumours not treated with immunotherapy, thus endorsing the idea that tumour infiltrating T cell clonality is a predictive biomarker specific for immune-checkpoint blockade. This analysis demonstrates that our observation has broad ramifications, not only in the clinical implications for treatment decisions for immune-checkpoint blockade in the adjuvant setting, but also because it is confirmed across multiple cancer cohorts and thus supports the hypothesis that the mechanisms of immune-surveillance that determine cancer prognosis are different from the determinants of response to immune-checkpoint blockade. The broader ramifications of our study will therefore be of interest to the wide readership of Nature Communications, and