Unveiling the influence of tumor and immune signatures on immune checkpoint therapy in advanced lung cancer

Reviewer #1 (Public Review):

Summary:

The authors study the variability of patient response of NSCLC patients on immune checkpoint inhibitors using single-cell RNA sequencing in a cohort of 26 patients and 33 samples (primary and metastatic sites), mainly focusing on 11 patients and 14 samples for association analyses, to understand the variability of patient response based on immune cell fractions and tumor cell expression patterns. The authors find immune cell fraction, clonal expansion differences, and tumor expression differences between responders and non-responders. Integrating immune and tumor sources of signal the authors claim to improve prediction of response markedly, albeit in a small cohort.

Strengths:

- The problem of studying the tumor microenvironment, as well as the interplay between tumor and immune features is important and interesting and needed to explain the heterogeneity of patient response and be able to predict it.

- Extensive analysis of the scRNAseq data with respect to immune and tumor features on different axes of hypothesis relating to immune response and tumor immune evasion using state-of-the-art methods.

- The authors provide an interesting scRNAseq data set linked to outcomes data.

- Integration of TCRseq to confirm subtype of T-cell annotation and clonality analysis.

- Interesting analysis of cell programs/states of the (predicted) tumor cells and characterization thereof.

Weaknesses:

- Generally, a very heterogeneous and small cohort where adjustments for confounding are hard. Additionally, there are many tests for association with outcome, where necessary multiple testing adjustments would negate signal and confirmation bias likely, so biological takeaways have to be questioned.

- RNAseq is heavily influenced by the tissue of origin (both cell type and expression), so the association with the outcome can be confounded. The authors try to argue that lymph node T-cell and NK content are similar, but a quantitative test on that would be helpful.

- The authors claim a very high "accuracy" performance, however, given the small cohort and lack of information on the exact evaluation it is not clear if this just amounts to overfitting the data.

- Especially for tumor cell program/state analysis the specificity to the setting of ICIs is not clear and could be prognostic.

- Due to the small cohort with a lot of variability, more external validation is needed to be convincingly reproducible, especially when talking about AUC/accuracy of a predictor.

Reviewer #2 (Public Review):

Summary:

The authors have utilised deep profiling methods to generate deeper insights into the features of the TME that drive responsiveness to PD-1 therapy in NSCLC.

Strengths:

The main strengths of this work lie in the methodology of integrating single-cell sequencing, genetic data, and TCRseq data to generate hypotheses regarding determinants of IO responsiveness.

Some of the findings in this study are not surprising and well precedented eg. association of Treg, STAT3, and NFkB with ICI resistance and CD8+ activation in ICI responders and thus act as an additional dataset to add weight to this prior body of evidence. Whilst the role of Th17 in PD-1 resistance has been previously reported (eg. Cancer Immunol Immunother 2023 Apr;72(4):1047-1058, Cancer Immunol Immunother 2024 Feb 13;73(3):47, Nat Commun. 2021; 12: 2606 ) these studies have used non-clinical models or peripheral blood readouts. Here the authors have supplemented current knowledge by characterization of the TME of the tumor itself.

Weaknesses:

Unfortunately, the study is hampered by the small sample size and heterogeneous population and whilst the authors have attempted to bring in an additional dataset to demonstrate the robustness of their approach, the small sample size has limited their ability to draw statistically supported conclusions. There is also limited validation of signatures/methods in independent cohorts, no functional characterisation of the findings, and the discussion section does not include discussion around the relevance/interpretation of key findings that were highlighted in the abstract (eg. role of Th17, TRM, STAT3, and NFKb). Because of these factors, this work (as it stands) does have value to the field but will likely have a relatively low overall impact.

Related to the absence of discussion around prior TRM findings, the association between TRM involvement in response to IO therapy in this manuscript is counter to what has been previously demonstrated (Cell Rep Med. 2020;1(7):100127, Nat Immunol. 2017;18(8):940-950., J Immunol. 2015;194(7):3475-3486.). However, it should be noted that the authors in this manuscript chose to employ alternative markers of TRM characterisation when defining their clusters and this could indicate a potential rationale for differences in these findings. TRM population is generally characterised through the inclusion of the classical TRM markers CD69 (tissue retention marker) and CD103 (TCR experienced integrin that supports epithelial adhesion), which are both absent from the TRM definition in this study. Additional markers often used are CD44, CXCR6, and CD49a, of which only CXCR6 has been included by the authors. Conversely, the majority of markers used by the authors in the cell type clustering are not specific to TRM (eg. CD6, which is included in the TRM cluster but is expressed at its lowest in cluster 3 which the authors have highlighted as the CD8+ TRM population). Therefore, whilst there is an interesting finding of this particular cell cluster being associated with resistance to ICI, its annotation as a TRM cluster should be interpreted with caution.

Author response:

We appreciate the comprehensive reviews and would like to address the critiques and suggestions provided by both reviewers. We will make significant revisions to the manuscript to address these concerns. These include a more cautious interpretation of our results, an expanded discussion on key findings, additional analyses for TRM characterization, and a clearer outline of future validation efforts. We believe these changes will enhance the clarity and robustness of our study, and we hope they meet the reviewer’s expectations.

Reviewer 1:

Weaknesses:

(1) Heterogeneous and small cohort:

Increasing the cohort size is not feasible due to resource constraints. We acknowledge the challenges posed by the heterogeneous and small cohort, which complicate adjustments for confounding. We will apply multiple testing corrections to transparently assess and accurately report the robustness of our findings in the revision.

(2) Influence of tissue of origin on RNAseq:

We agree that RNAseq results can be heavily influenced by the tissue of origin. While immune cell composition in the normal lung tissues and lymph nodes is quite different, we found that in tumor tissues and metastatic lymph nodes, these differences diminish and common features dominate. Although we depicted this data in the supplementary figure 1, we did not provide a quantitative test in the original submission. In the revision, we will perform additional quantitative tests to compare immune cell composition across different tissue origins. These tests will provide a more precise understanding of the cellular composition and support our argument regarding the similarity of tumor-sculpted microenvironment. We will include these results and detailed methodologies in the revision.

(3) Accuracy performance and overfitting:

We acknowledge the concern regarding the high “accuracy” performance potentially indicating overfitting. We will clarify the evaluation methods used and moderate our claims regarding accuracy in the revision.

(4) Specificity of the tumor cell program/state analysis to the setting of ICIs:

The comment suggests that the tumor programs in our study may not be specific to the ICI group but rather prognostic in lung cancer. We acknowledge this possibility as we performed comparisons between responders and non-responders (with different cut-offs) to find common trends and interpreted them in terms of their association with ICI. In the revision, we will test the prognostic association of the tumor programs using public lung cancer data.

(5) More external validation needed:

We recognize the importance of external validation for reproducibility. While increasing the cohort size is not feasible, we will propose future directions for validation using larger, independent cohorts and potential experimental validations.

Reviewer 2:

Weaknesses:

(1) Small sample size and heterogeneous populations:

Increasing the cohort size is not feasible due to resource constraints. We acknowledge the challenges posed by the heterogeneous and small cohort, which complicate adjustments for confounding. We will apply multiple testing corrections to transparently assess and accurately report the robustness of our findings in the revision.

(2) Limited validation of signatures/ methods in independent cohorts:

We recognize the importance of external validation for reproducibility. While increasing the cohort size is not feasible, we will propose future directions for validation using larger, independent cohorts and potential experimental validations.

(3) Lack of functional characterization and discussion on key findings:

We appreciate the feedback regarding the need for functional characterization and a more thorough discussion of key findings on the roles of specific cell populations and genes. In the revised manuscript, we will expand the discussion section to include in-depth analysis of these findings and their relevance to the study. This includes a detailed interpretation of how these factors contribute to the immune response and potential implications for therapy.

(4) TRM findings and marker selection:

We understand the concern regarding the association between TRM involvement in response to IO therapy, which appears counter to previous demonstrations. It is indeed important to note that we employed alternative markers for TRM characterization. Our choice of markers was based on transcriptional references relevant to our study. However, we agree that classical TRM markers such as CD69 and CD103, which were absent in our definition, are critical for accurate TRM identification. To address this, we will include a detailed rationale for our marker selection and acknowledge the limitations of our TRM characterization. We will include additional analyses using classical TRM markers where possible and incorporate these findings into the revision. This will provide a clearer understanding of our TRM population and its role in the immune response to IO therapy.