Heterogeneity of Exhausted T Cell Subsets in Responders and Non-Responders Following Checkpoint Inhibition Therapy

Heterogeneity of Exhausted T Cell Subsets in Responders and Non-Responders Following Checkpoint Inhibition Therapy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Heterogeneity of Exhausted T Cell Subsets in Responders and Non-Responders Following Checkpoint Inhibition Therapy Irina Kareva, Clara Pavillet This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5663382/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The emerging recognition of multiple states of T cell exhaustion, of which only some are targetable by checkpoint inhibitors, has provided new insights into the variability in patient responses to immunotherapy. We hypothesized that non-responders to therapy have a higher proportion of non-targetable, terminally exhausted T cells compared to responders. To investigate this, we analyzed single-cell RNA sequencing data from 27 patients with head and neck squamous cell carcinoma (HNSCC) treated with neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4 therapy. We identified gene signatures for T cells across different states, ranging from naïve to terminally exhausted, and evaluated their distribution post-treatment. Non-responders exhibited a more inflammatory profile, while responders showed a more balanced immune profile with higher proportions of both helper and regulatory T cells, suggesting that a balanced inflammatory environment may be crucial for therapeutic success. Our analysis further revealed differences between responders and non-responders in the composition of predicted T cell states, particularly in the exhausted T cell subsets, with non-responders showing a higher proportion of terminally exhausted T cells. We therefore propose existence of tumors that may be “too hot”, with resulting loss of efficacy and emergence of therapeutic resistance through a pathway that is different from that of “cold” tumors. Despite limitations, including the small sample size and the lack of well-established transcriptomic signatures of exhaustion subsets, our findings offer a starting point to encourage further investigation into the relationship between inflammation, T cell exhaustion, and therapy efficacy towards improving patient outcomes. Biological sciences/Computational biology and bioinformatics Biological sciences/Cancer/Tumour immunology Biological sciences/Cancer/Cancer therapy Biological sciences/Cancer/Cancer therapy/Cancer immunotherapy Biological sciences/Cancer/Cancer therapy/Cancer therapeutic resistance T Cell Exhaustion Tumor Immunology scRNA-Seq Therapeutic Resistance Immune Checkpoint Therapy Cancer Immunotherapy Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction The history of immunotherapy reflects a long journey from skepticism to recognition of the immune system's potential to fight cancer (Dobosz & Dziecikatkowski 2019 ). Early observations, such as those by William Coley in the late 19th century, hinted at this when he noted that bacterial infections could sometimes lead to spontaneous tumor regressions (Coley 1893 ). This led to the development of “Coley’s toxins,” a pioneering but ultimately unstandardized immune-based approach to cancer treatment (Carlson et al. 2020 ). It wasn’t until decades later that immunotherapy made significant strides, with Steven Rosenberg’s work in the 1980s on cytokine therapy, particularly IL-2, marking the next critical milestone (Rosenberg et al. 1994 ). Although IL-2-based therapies were hampered by toxicity (Dutcher et al. 2014 ), they demonstrated that the immune system could be mobilized against cancer, providing a foundation for future advances. The discovery of immune checkpoints like CTLA-4 and PD-1 in the early 1990s was a pivotal breakthrough that finally unlocked the power of immunotherapy (Leach et al. 1996 ; Ishida et al. 1992 ). These molecules, which act as brakes on the immune system to prevent autoimmunity, became the targets of drugs known as checkpoint inhibitors. The approval of ipilimumab in 2011 and pembrolizumab in 2016 marked the beginning of a new era in cancer treatment, where the immune system could be harnessed more effectively (Robert 2020 ). While immune checkpoint inhibition (ICI) has been nothing short of revolutionary, only a fraction of patients respond (Schoenfeld & Hellmann 2020 ; Lee et al. 2022 ), and the factors that separate responders from non-responders remain to be elucidated. Checkpoint inhibitors target T cell exhaustion, a state of T cell dysfunction characterized by a loss of effector function and proliferative capacity, along with the upregulation of inhibitory receptors such as PD-1 (programmed cell death protein 1), CTLA-4 (cytotoxic T-lymphocyte-associated protein 4), TIM-3 (T-cell immunoglobulin and mucin-domain containing-3), and LAG-3 (lymphocyte activation gene-3). It is becoming increasingly understood that checkpoint expression is a mechanism that has evolved to protect against autoimmunity, particularly in chronic viral infections (Kahan et al. 2015 ; McLane et al. 2019 ), since in the context of a chronic infection that cannot be fully eradicated, maintaining a state of exhaustion may be less detrimental to the host’s overall fitness than continuous active cytotoxicity (Kareva & Brown 2024 ). McKinney et al. (McKinney et al. 2015 ) support this hypothesis by showing that the transcriptional profiles associated with CD8 + T cell exhaustion during chronic infection correlate with both poorer infection clearance and a reduced risk of developing autoimmune diseases. In recent years, it has become evident that T cell exhaustion is not a binary condition but rather a spectrum of states, with only certain stages being responsive to reinvigoration through checkpoint inhibition therapy. These distinct stages of exhaustion represent varied degrees of dysfunction. Initially, exhausted T cells may exhibit impaired cytokine production and reduced cytotoxic activity yet retain some functionality and proliferative capability. As T cells progress toward terminal exhaustion, there is a significant increase in the expression of inhibitory receptors and a decrease in co-stimulatory molecules. It appears that the mechanisms that evolved to protect against autoimmunity in chronic viral infections are like the mechanisms that can prevent effective tumor elimination by the immune system. In (Miller et al. 2019 ), the authors reported that analogous subsets of exhausted CD8 + T cells are observed both in tumors and in chronic viral infections, including progenitor exhausted cells that are more responsive to checkpoint inhibitor therapy (Blackburn et al. 2008 ; Huang et al. 2017 ), and terminally exhausted T cells, which are not. Interestingly, in the earlier stages of T cell exhaustion, these cells retain the ability to self-renew and may give rise to more dysfunctional cells. In subsequent work, Beltra et al. (Beltra et al. 2020 ) proposed an expanded four-cell-stage developmental framework for exhausted CD8 + T cells (Tex): Ly108 + CD69+ (progenitor 1, Texprog1), Ly108 + CD69- (progenitor 2, Texprog2), Ly108-CD69- (intermediate, Texint), and Ly108-CD69+ (terminal, Texterm), with each subset defined by its own unique transcriptional and epigenetic profile. Based on these observations, we hypothesized that one possible distinction between responders and non-responders to ICIs is the immune cell composition with respect to different states of T cell exhaustion. To evaluate this, we compiled gene signatures characteristic of various T cell states, including substates of exhaustion from Beltra et al. (Beltra et al. 2020 ). We then applied these signatures to analyze a published single-cell RNA sequencing (scRNA-seq) dataset of samples collected during a phase 2 clinical trial of head and neck squamous cell carcinoma (HNSCC) patients treated with neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4 (Luoma et al. 2022 ). We evaluate differences in T cell states and inflammatory markers between responders and non-responders, both in tumor infiltrating lymphocytes (TILs) and peripheral blood mononuclear cells (PBMCs). We conclude with a discussion of the implications of the observed differences, both for monotherapy and combination therapy. Results Assessing Local Inflammatory Responses The initial processed blood dataset from Luoma et al. ( 2022 ) comprised 68,183 cells across 27 patients, whilst the tissue dataset consisted of 56,537 cells from 19 of these 27 patients. Further filtering was applied to obtain post-treatment tissue samples, resulting in a dataset of 40,856 cells. Based on volumetric response measurements, 10 patients were classified as responders and 9 as non-responders. We conducted a detailed analysis of the cellular landscape by subdividing it into major cell types and further into more granular categories based on clusters and differentially expressed genes (Figs. 1 A and 1 B). This expanded the scope of the initial study by including non-CD45 + cells. Contributing to the inflammatory tumorigenic environment was a population of collagen-rich (COL3A1, COL1A2, COL6A1, COL6A2) cancer-associated fibroblasts (CAFs) characterized by high expression of DCN, LUM, PRRX1, MMP2, FSTL1 and SPARC. CAFs in non-responders were found to be associated with a heightened inflammatory profile, a denser extracellular matrix, and the expression of immunosuppressive factors, collectively contributing to a microenvironment that may be less conducive to effective immune responses. Significantly upregulated genes in non-responder CAFs included CD34, CFD, PI16, TNXB, MFAP5, CLEC3B, PCOLCE2, ADH1B, CD70, IL33, and TGFBR2. Whilst the upregulation of genes like VIT, CD34, and CLEC3B may further contribute to a more inflammatory environment, the higher expression of collagen-related genes, along with markers of enhanced cell adhesion such as MFAP5 and TNXB, indicates a physical barrier to immune cell infiltration. Additionally, the expression of TGFBR3 and IL33 points to an IL-33-TGF-β niche signaling pathway that has been shown to suppress T cell activity. Myeloid cells could be further subdivided into pro-inflammatory subpopulations, including FCN1 + tumor-associated macrophages (TAMs), anti-inflammatory subpopulations such as plasmacytoid dendritic cells (pDCs) and M2-like tumor-associated macrophages (TAMs), and mixed inflammatory subpopulations that expressed a balance of immune activation and suppression molecules. Mast cells expressed high levels of tryptases (TPSAB1, TPSB2, TPSD1, TPSG1), lipid mediators (COX-1, COX-2) and other potent molecules that contribute to inflammatory responses including LTC4S. In tissue, non-responders exhibited a significantly higher proportion of the Cycling T/NK group post-treatment compared to responders, underscoring the need to investigate this population further and the subgroups it comprises (Figs. 1 C and 1 D). Whilst cycling T cells typically indicate activation, this paradox may stem from cells with impaired effector functions that have yet to lose their proliferative ability. Additionally, some cycling cells might be regulatory T cells, that work by suppressing anti-tumour responses. At the sample level, no responders also showed an increase in IFNγ expression compared to responders, as measured at both the single-cell and pseudobulk levels (Figs. 1 E and 1 F). Additionally, a pro-inflammatory gene signature was found to be upregulated in the samples from non-responders, while an anti-inflammatory gene signature showed a slight upregulation in responders (Fig. 1 G). Dissecting the Heterogeneous Landscape of Exhausted Tumour-Infiltrating T Cells Based on the results presented in the previous section, we hypothesized that T cells contained subsets of exhaustion that retained proliferative capacity and effective functions which could explain why non-responders had more Cycling T/NK cells. The dataset from Luoma et al. ( 2022 ) was reexamined in light of the publication by Beltra et al. ( 2020 ), which provided signatures of T cell exhaustion subsets. This reanalysis aimed to enhance our understanding of T cell states by integrating the findings from Beltra et al. into the analysis of tumor-infiltrating T cells in response to neoadjuvant ICB. By doing so, we sought to identify potential markers and pathways that could better delineate the spectrum of T cell responses beyond the classical definitions of exhaustion. Using the combined pre- and post-treatment dataset, T cells were isolated from the rest of the population, and further refined using an ensemble approach. Beyond T cell subtypes, T cells can exist along a gradient of states, each with unique functional properties and implications for immune responses. Our detailed analysis of T cells enabled us to break down T cell populations at various levels of granularity and into distinct groups based on their functional states, phenotypic characteristics, and molecular signatures. By dissecting these states, we aimed to understand the specific components and characteristics that define each group. This multi-level analysis provides deeper insights into the heterogeneity of T cell responses and helps us identify the underlying mechanisms that drive their behavior in different contexts. The reasoning behind this approach is that there is no definitive way to ascertain the 'true' label of T cells. The method used therein allows us to integrate multiple models or perspectives, enhancing the robustness and reliability of our T cell labelling. By leveraging the strengths of various cell identification methods, as depicted in the alluvial plot in Fig. 2 D, we aim to capture a more comprehensive understanding of T cell subsets and states, despite the inherent uncertainties in labeling. For the prediction of T cell subtypes, we harmonized results across the different algorithms before using a majority voting approach in order to classify major subtypes in the dataset and most accurately evaluate proportions. Using the gene signatures described in the methods section, eight distinct T cell states were predicted, specifically TN/TSCM, TCM, TEM, TEFF, TExProg1, TExProg2, TExInt and TExTerm. The proportion of each predicted state was evaluated across each predicted T cell functional group. As can be seen in Fig. 2 , in this dataset, TCM are the most prevalent. These cells are characterized by their ability to rapidly proliferate and differentiate into effector cells upon re-exposure to antigen, making them a crucial component of the adaptive immune response both for immune surveillance and memory formation. TExTerm were mainly observed in cytotoxic T cells from the cluster characterized by high GZMB and CTSW expression. This distribution highlights the diverse functional roles of T cell subsets and their potential implications for immune responses. The TExProg2 cells were found to be associated with proliferating T cells. CD4 T cells are generally less prone to exhaustion compared to CD8 T cells due to their distinct roles and biology. CD4 T cells have a lower activation threshold and primarily help other immune cells by secreting cytokines. As a result, they experience less direct antigen exposure, reducing the likelihood of chronic stimulation and thus exhaustion. In contrast, CD8 T cells are directly involved in the killing of tumour cells, requiring stronger and more sustained activation, making them more susceptible to exhaustion. Accordingly, in our dataset, exhausted states were mainly found in CD8 T cells, but different subsets of cytotoxic T cells exhibited different predicted substates of exhaustion (Fig. 2 A). In Beltra et al. ( 2020 ), TExProg1, defined as quiescent resident, had restricted ex vivo proliferative potential but exhibited robust proliferation upon antigen stimulation. They were characterized by low in vivo proliferation or ongoing cell cycle activity, with little evidence of ex vivo cell death. Genes upregulated in TExProg1 were involved in progenitor biology (TCF7, SELL), T follicular biology (CXCR5, ICOS), and positive co-stimulation (CD28). In our dataset, TExProg 1 cells were few and sparsely identified across cytotoxic subgroups. Beltra et al. ( 2020 ) characterized TExProg 2 cells as proliferative circulating cells that gained access to the blood and exhibited a decline in cytokine coproduction. Like TExProg 1, there was little evidence of ex vivo cell death in this subset, which was preferentially amplified via PD-1 blockade. Interestingly, their accumulation was promoted by PD-L1 blockade. In accordance with their high proliferative potential, we identified TExProg 2 in proliferating cells characterized by high expression of Ki67 and other proliferating markers. In Beltra et al. ( 2020 ), TExInt cells were labeled as circulating and mildly cytotoxic. They also experienced a decline in cytokine coproduction and accumulated in response to PD-L1 blockade, like TExProg 2. These cells were almost incapable of undergoing cell division upon re-stimulation and were readily detectable as apoptotic. In our dataset, TExInt cells mapped to CD8 T cells but were not specific to any particular subset of CD8 T cells. In Beltra et al. ( 2020 ), TExTerm cells were associated with tissue residency and were incapable of undergoing cell division upon re-stimulation in vitro. These cells were readily detectable as apoptotic. They exhibited enriched pathways for the negative regulation of cell activation but showed signs of recent TCR signaling, including markers like ZAP70 and NFATC1, and pathways related to calcium influx, in addition to enrichment in interferon-related transcription factor motifs. We found that TExTerm cells were highly represented across granzyme-expressing subsets of T cells. Naive T cells on the other hand were present in lower numbers, with a predominant representation in the CD4 subset expressing CCR7, IL7R and KLF2 (Fig. 2 B). Central memory T cells constitute the majority across all groups of T cells. The TME presents tumor-associated antigens that lead to the differentiation of naïve T cells into central memory, which have the unique ability to survive long-term in tissues and help establish long-lasting immune memory. This persistence allows rapid reactivation upon antigen re-encounter, crucial for immune surveillance and tumor response. Moreover, memory T cells, including stem cell memory and central memory T cells, show greater persistence and antitumor immunity than effector memory and effector T cells. The TCM/TEFF ratio has also been identified as a predictive biomarker for immune responses in certain tumors (Liu et al. 2020 ). Our dataset indicates that effector (TEFF) and effector memory (TEM) T cells are predominantly found within the cytotoxic T cell population in the TME. This suggests that it is these T cells that have been activated in response to tumor antigens and are involved in executing immune responses. Effector memory cells provide long-term immunity and are ready to respond quickly upon re-exposure to antigens. Treatment Effect on T Cell Compartments Next, to understand the impact of therapy on T cell composition, pre- and post-treatment conditions, which were only available for 6 of the 19 patients, were analyzed separately (Fig. 3 ). Firstly, as can be seen in Fig. 3 A and 2 B, there is a notable difference in immune cell composition before and after treatment, with increases in helper, Treg and inconclusive cells, and a notable decrease in cytotoxic cells (note the logscale in Fig. 3 B). Specifically, the most significant change occurs in cytotoxic T cells, with a 9% decrease post-treatment; helper T cells show a 4.8% increase in the post-treatment data set and Tregs exhibit 8.8% increase post-treatment. The increase in helper T cells post-treatment could indicate an adaptive response attempting to support immune functions. Helper T cells might also be transitioning towards regulatory phenotypes, as suggested by the increase in Tregs. Increase in Tregs post-treatment may account for decrease in the proportion of cytotoxic T cells as an immune regulatory mechanism. Furthermore, there is a notable change in the distribution of T cells with respect to cell state (Fig. 3 C and 3 D), with a 54.4% increase in terminally exhausted (TexTerm), as well as 92.9% increase in TExProg2 and 45.2% increase in TEM, and a decrease in TEFF (-52.2%), TExProg1 (-69.2%) and TExInt cells (-24.7%). The decrease in TCM (-5.3%) and the moderate increase in TN (+ 17.6%) suggest a subtle shift away from central memory towards naïve-like states. This might be part of a replenishment mechanism or a response to the treatment-induced changes in the immune landscape. The decrease in TExInt (-24.7%) could imply that fewer cells are in the transitional phase between progenitor and terminal exhaustion, potentially due to progression towards terminal exhaustion or apoptosis. Overall, decrease in the effector state and increase in cell states associated with exhaustion might suggest that treatment may be triggering transitions towards the multiple states of exhaustion, which in turn suggests that it is possible that pre- vs post-treatment differences could be at least partially attributed to the impact of treatment rather than being an intrinsic property of the patients. The observed increase in CD4 T cells alongside a decrease in CD8 T cells post-treatment could be due to differential response to immune checkpoint blockade. CD4 T cells, particularly helped T cells, may respond more robustly to treatment. The cytokine milieu post-treatment may favor expansion of certain T cell subpopulations over others. PD-1 blockade and anti-CTLA4 can alter the TME differently, increasing the levels of certain cytokines such as IL-2 and IL-7, which promote the proliferation and survival of T cell subsets. This shift in cytokine balance may create a more supportive environment for CD4 T cells over CD8 T cells, leading to their increased presence in the tumor. Meanwhile, the same cytokine changes may not equally support CD8 T cell expansion, contributing to their relative decrease. In addition, CD8 T cells are more prone to exhaustion than CD4 T cells as described earlier. CD8 T cells might be undergoing functional changes or facing other inhibitory pathways that are not fully reversed by immune checkpoint blockade alone. This may result in a slower or less pronounced recovery compared to CD4 T cells. The increase in TEM cells suggests that some T cells are being reactivated and are transitioning into a memory state. PD-1 blockade can enhance the function and survival of memory T cells, leading to their accumulation, whilst a decrease in effector T cells could be due to exhaustion, apoptosis or differentiation into other states such as TEM or exhaustive subsets. Fewer T cells find themselves at the initial stage of exhaustion (TExProg1) as they may be progressing towards more advanced stages of exhaustion due to continuous antigen exposure and insufficient recovery. Alternatively, they might actually be benefiting from the immune checkpoint blockade provided by the therapy. Treatment can partially reverse the exhausted state, allowing these cells to regain functionality and potentially transition out of the exhausted state into more active or memory states. Thus, the observed decrease in TExProg1 cells may be reflecting a combination of progression towards deeper exhaustion and successful reactivation by the therapy. More specifically, when inhibitory pathways are blocked, some exhausted T cells can recover their effector functions and proliferative capacity. This reactivation can lead to their differentiation into effector memory T cells, which are characterised by their ability to respond rapidly to antigens and provide long-term immune surveillance. The increase in TExProg2 observed in our dataset could be attributed to their retained proliferative capacity. TExProg2 cells, whilst still in an exhausted state, have not fully lost their ability to proliferate and can respond to stimulation to some extent. PD-1/CTLA4 blockade therapy could partially reverse the exhaustion in these cells, enabling them to proliferate more effectively. This would result in an accumulation of TExProg2 cells within the TME and thereby contribute to the observed increase. T Cell States are Key Determinants of Treatment Response Next, we wanted to evaluate whether there exists a difference in T cell composition with respect to exhaustion states between patients who responded to treatment compared to those who did not (responders versus non-responders). For this, two types of clinical readouts were evaluated, pathological response and volumetric response. The former evaluates the extent of tumor regression based on histological examination of tissue samples, whilst the latter evaluates changes in the size or volume of the tumor based on imaging studies. For this part of the analysis, post-treatment TIL data were available for 19 out of 27 patients. Let us first analyze the volumetric data (Fig. 4 A). Responders have a higher proportion of helper cells (35.4%) compared to non-responders (26.7%), suggesting that a stronger helper T cell presence might be associated with better outcomes, possibly aiding in the activation and maintenance of other immune cells. There is an increase in Tregs among responders (22.0% vs. 16.2%). While Tregs are typically immunosuppressive, their role here might involve maintaining immune homeostasis and keeping inflammation within a certain threshold to prevent further exhaustion of treatment-responsive T cells and the tapering of their functions. Non-responders exhibit a higher proportion of cytotoxic cells (48.1%) compared to responders (33.6%). This finding aligns with the earlier analysis, which indicated that most cytotoxic T cells are concentrated in a terminally exhausted state (Fig. 2 C). Consequently, in non-responders, the transition further along the exhaustion pathway results in cytotoxic cells being less effective in their function, despite their higher numbers. Next, we examine T cell states and the differences between responders and non-responders as illustrated in Figs. 4 C; relative T cell state proportion differences assessed using scProportionTest can also be found in Supplementary Figure A2. Responders demonstrate a significantly higher proportion of central memory T cells (TCM) at 92.4%, compared to 83.0% in non-responders. This indicates that responders may maintain a more robust central memory pool, potentially contributing to a more sustained immune response. On the other hand, both TEM and TEFF are slightly higher in non-responders, but their overall proportions are very low, indicating a shift away from effector function in both groups. TCM usually demonstrate superior persistence and antitumor immunity compared to TEM, with greater proliferative capacity and longevity, enabling them to maintain a sustained immune response and provide long-term protection against tumors. As described above, non-responders were found to have a significantly higher proportion of TExTerm cells (10.0%) compared to responders (3.7%). This is consistent with the idea that terminally exhausted cells are less functional and may contribute to non-responsiveness. Furthermore, non-responders have a higher proportion of TExProg2 cells (4.1%) than responders (2.0%), which could indicate that cells further down the exhaustion pathway are more prevalent in non-responders, possibly limiting the potential for reactivation. Next, let us conduct the same analysis for pathological data. As one can see in Figs. 4 A and 4 B, both the volumetric and pathological data show a trend, where higher helper T cell proportions are associated with better responses (> 50% pathological response and volumetric response). Across both datasets, non-responders (< 10% pathological response and volumetric non-response) have higher proportions of cytotoxic T cells. This consistency suggests that the mere presence of cytotoxic cells is not sufficient for a positive response—functionality and exhaustion state might be more critical. Tregs are elevated in better responders in both datasets, though the increase is subtle. This might indicate that a controlled regulatory environment is needed for an effective response, preventing excessive inflammation or autoimmune-like damage. Increasing evidence suggests that regulatory populations are highly heterogeneous and subpopulations of Tregs can differentially regulate immune responses, with some subsets having less suppressive capacity or even promoting anti-tumor immunity under certain conditions. This variability may also help explain the differences between responders and non-responders to immune checkpoint treatments, as the balance of these regulatory subsets can significantly influence the overall effectiveness of the immune response against tumors. Finally, the central memory (TCM) subset consistently dominates across all groups in both datasets, providing a reservoir for sustained responses over time. On one hand, TCM cells have higher proliferative potential, allowing them to renew and expand more effectively to sustain their numbers; on the other, they tend to have more robust mitochondrial function and maintain a more stable epigenetic landscape than their effector counterparts Interestingly, the pathological response data shows a more consistent increase in helper T cells and Tregs with better outcomes, whereas the volumetric response data had more variability in these subsets between responders and non-responders. This might suggest that helper and Treg balance could be more predictive of pathological rather than volumetric response. Furthermore, the proportion of TExProg2 cells slightly increases in the higher pathological response group (> 50%), which contrasts with a decrease in this subset in volumetric responders. This discrepancy could point to different roles for TExProg2 in pathological versus volumetric response contexts, or it could reflect a sampling or timing difference. Dynamic Fluctuations in Interferon Signalling Observed on Path to Terminal Exhaustion Next, we wanted to evaluate the relationship between immune activation and T cell states for responders versus non-responders. In this part of the analysis, we evaluated the change in IFN signature scores to assess how they change as T cells go through the progression TN → TCM → TEM → TEFF → TExProg1 → TExProg2 → TExInt → TExTerm. As can be seen in Fig. 5 A, the median IFNg signature score is the lowest in TN, increasing in TCM and reaching the highest level in TEM. In the progression through multiple states of exhaustion, the IFNg score that starts at a relatively high level in TEFF cells decreases as the cells transition to the first stage of exhaustion, TExProg1. It then increases as the cells transition to the TExProg2 state, decreases again in TExInt and finally once again reachest the highest value in TExTerm cells. This is consistent with the current understanding that terminally exhausted cells are short lived but can be very cytotoxic, as well as the observation that T cells in the intermediate stages of exhaustion have very low cytotoxicity. Specifically, in (Miller et al. 2019 ), the authors note that “progenitor exhausted cells killed target cells no more efficiently than did naive CD8 + T cells, indicating that they contribute little or no direct cytotoxicity in the TME”. While the median IFNg signature scores for TExProg1 cells are not as low as for TN, they are second lowest out of all T cell states. Next, we analyzed the differences in IFNg signature scores between responders and non-responders. We subdivided the scores into low, medium, and high IFN activity, calculated as one standard deviation from the median (Fig. 5 B). We then assigned these classifications to the T cells in responders versus non-responders in both volumetric (Fig. 5 C,D) and pathological (Fig. 5 E,F,G) response data. (C) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell identity in volumetric response non-responders. (D) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in volumetric response responders. (E) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with < 10%. (F) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with 10–49% response. (G) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with > 50% response. In the volumetric data set, several differences between responders and non-responders can be highlighted with respect to differences in IFNg signature scores. In non-responders, TEFF cells are predominantly low in IFNg (blue), whereas responders show a higher proportion of medium and high IFNg signatures (Fig. 5 C). Furthermore, non-responders have more TEFF cells with low IFNg signatures compared to responders, who show a shift towards medium and high IFNg signatures. Both groups show medium IFNg signatures as dominant in TExProg1 cells, but responders maintain a slight edge with more medium IFNg cells, potentially indicating better reactivation potential. Non-responders have 50% more TExProg1 cells with low IFNg signatures, whereas responders have a modest increase in medium IFNg signatures. In TExProg2 cells, in both groups, high IFNg signatures dominate, but responders also show an increase in low IFNg signatures. Responders have dramatically fewer low IFNg TExProg2 cells, suggesting a pathway away from deep exhaustion. High IFNg is less prevalent in responders compared to non-responders. In TExInt cells, high IFNg is dominant in both groups, but with a larger proportion in responders. Responders show a significant increase in high IFNg signatures compared to non-responders, suggesting active engagement, while non-responders have a higher proportion of cells with low IFNg. Finally, responders have 25% more of the high IFNg TExTerm cells compared to non-responders, who have slightly more of the low IFNg cells. Interestingly, high IFNg is prevalent in both groups, but responders manage to sustain a higher proportion of medium IFNg signatures. This suggests that high IFNg in TExTerm cells might not be universally detrimental; in responders, these cells may still play a role in controlling the tumor, while in non-responders, low IFNg suggests deeper dysfunction. The patterns observed in the pathological data (Fig. 5 E,F,G) align with those seen in the volumetric response data, particularly in the balance between high and medium IFNg signatures across different T cell states. Both data sets suggest that extreme IFNg signaling (either too high or too low) can be detrimental, and that a balanced IFNg profile is associated with better therapeutic outcomes. Contrasting Response Patterns Between Peripheral and Tumor Compartments Next, we want to evaluate whether the patterns that were observed in TIL data with respect to differentiation between responders and non-responders might also be observed in the peripheral compartment. The initial PBMC dataset comprised 68 183 cells across 3 timepoints for 27 patients. T cells were isolated from the rest of the population, and further refined using an ensemble approach, as done previously. Clusters of cells were generated using the shared nearest neighbour (SNN) algorithm and different populations of cells were identified based on differentially expressed (DE) genes. Due to the overlapping gene profiles between natural killer (NK) and T cells, these two cell types often cluster together. As such, the cell clusters likely to be either NK or T cells were labelled as T/NK, with the intention of separating them downstream. After isolating the T/NK clusters, **38 539** cells remained in the analysis. Three distinct cell type identification methods were subsequently employed. These methods included SingleR, which uses the non-parametric Spearman rank correlation to score each cell in the dataset against each cell type in a reference (used: X-OMICS HNSCC reference); Azimuth, which uses a weighted-nearest neighbour (WNN) model-based approach to map cells to a reference (used: internal PBMC reference); and a random forest classifier by Bogac Aybey. It was believed that each of these algorithms contribute unique strengths to the identification process, collectively enhancing the accuracy of T cell identification. A majority voting approach was used to isolate T cells, selecting those with a score of 4 for a total of 27 922 T cells. This score indicates agreement across clustering results and three distinct tools, providing a high confidence identification of T cells. As one can see in Fig. 6 , the distribution of T cell identities in PBMCs is quite different from TILs (Fig. 3 ). Most helper cells are clustered in TN, although in contrast to TILs, where TNs are composed of both helper and cytotoxic cells, in PBMCs most cytotoxic cells are in the TEFF and TExInt classes. In contrast, in the TME, most of the cytotoxic cells were in the terminally exhausted state. Next, we assessed whether there exist differences in T cell states and composition in PBMCs over time; data were collected at three time points, B1, B2 and B3. In Fig. 7 , we assess the change in proportions of different cell states and types (similarly to the analysis done for TILs in Fig. 4 ). In Fig. 8 , we assess the differences between responders and non-responders as reflected in PBMCs, similarly to the analysis done for TILs in Fig. 4 . In non-responders, we observe 26.7% helper cells, 16.2% Tregs, 48.1% cytotoxic cells in the TME, while in PBMCs, we see higher helper cells (67.4–75%), very low Tregs (0.7–0.9%), and lower cytotoxic cells (21–27.5%). In responders, we observe 35.4% helper cells, 22% Tregs, 33.6% cytotoxic cells in the TME, while in PBMCs, helper cells dominate (71–76.2%), with low Tregs (0.4–0.9%) and lower cytotoxic cells (19.5–23%). Together, these results show that in TILs, non-responders have a much higher proportion of cytotoxic T cells and Tregs compared to PBMCs. The helper T cells are dominant in PBMCs, whereas cytotoxic T cells are more prevalent in TILs, indicating that in non-responders, the cytotoxic response may be more localized to the tumor environment. Furthermore, responders also show a much higher proportion of cytotoxic T cells and Tregs in TILs compared to PBMCs. With respect to T cell exhaustion states, in non-responders, we observe low TN (1.5%), high TCM (83%), moderate TExTerm (10%) in TILs; in PBMCs, we see higher TN (46.7–58.1%), moderate TCM (30.7–39.9%), and very low TExTerm (0.1%). That is, TILs of non-responders show a strong central memory (TCM) phenotype and higher terminal exhaustion (TExTerm), whereas in PBMCs, the profile is more naïve (TN) and lacks significant terminal exhaustion. This suggests the exhaustion observed in non-responders is mainly restricted to the tumor environment. In responders, we see low TN (1.0%), very high TCM (92.4%), low TExTerm (3.7%) in TILs, while in the PBMCs we observe higher TN levels (48.7–57.6%), moderate TCM (32.6–38.3%), and very low TExTerm (0.1%). That is, responders have a higher proportion of central memory T cells (TCM) in TILs, with low terminal exhaustion. In PBMCs, there is a mix of TN and TCM with minimal exhaustion. This indicates that effective immunotherapy might maintain central memory cells, while reducing terminally exhausted cells primarily in the tumor microenvironment. Together, these results show that the key patterns observed in TILs, such as high central memory (TCM), the presence of terminally exhausted T cells (TExTerm), and an increased proportion of cytotoxic T cells and Tregs, are not mirrored in PBMCs. Instead, PBMCs are characterized by a higher proportion of naïve and central memory cells with very little terminal exhaustion. Thus, while T cell exhaustion and memory are critical for the tumor microenvironment, these patterns are not reflected in circulating PBMCs, suggesting that exhaustion-related changes are more localized to the tumor. Therefore, no significant patterns that differentiate responders from non-responders are observed in PBMCs, implying that T cell exhaustion and differentiation dynamics relevant to immunotherapy response are more specific to the tumor site. Discussion Emerging understanding of the existence of multiple states of T cell exhaustion, only some of which are targetable by checkpoint inhibitors, has created a framework to improve our understanding of the differences between patients who respond to therapy and those who do not. We hypothesized that patients who do not respond to therapy would have a higher proportion of non-targetable terminally exhausted T cells compared to responders. In this work we used single-cell RNA sequencing data of samples of 27 HNSCC patients who received neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4. We collated gene signatures for T cells based on lineage and progression through multiple states: TN → TCM → TEM → TEFF → TExProg1 → TExProg2 → TExInt → TExTerm. We assessed both the distribution of cell types and states in pre- vs post-treatment samples for the entire population, as well as stratified responders and non-responders in post-treatment sample analysis. Additionally, we evaluated the change in IFNg signature scores to assess whether the level of IFN signaling in T cells may contribute to differences between responders and non-responders. In our analysis, we found that there indeed exist differences between responders and non-responders with respect to distribution of different T cell types and states. We found that there exist differences between pre- and post-treatment T cell distributions, with the overall decrease in the effector cell types, and increase in cell states associated with the different states of exhaustion. These results suggest that changes in T cell composition may at least partially be attributed to treatment rather than being an intrinsic patient characteristic. This hypothesis remains to be more rigorously evaluated on a larger dataset than was available for this analysis and should include more extensive pre- and post-treatment data for both responders and non-responders. Next, we analyzed T cell identifies in pathological and volumetric responses and found that while there appear to be more cytotoxic T cells in non-responder samples, these cells tend to be primarily terminally exhausted. Indeed, looking more closely at the cell states revealed that the proportion of terminally exhausted cells in non-responders is significantly greater than in patients who responded to therapy. This suggests that while it is typically expected that a higher proportion of CD8 cells in the TME would be associated with a better response, a more informative approach may involve assessing the functional state of these cells rather than just their lineage. Furthermore, responders had higher proportions of both helper and regulatory T cells compared to non-responders. This suggested that responders may have a more balanced immune response compared to non-responders, and that even though Tregs are typically associated with immune regulation, their role here might involve maintaining immune homeostasis post-treatment (Li et al. 2024 ). Similarly, the pathological response results (50%) observed in comparison to volumetric analysis of non-responders versus responders showed similar relationships in T cell states, but with slight discrepancies that may be attributed to variations in the underlying methodologies. The analysis of volumetric response, which captures an overall change in tumor volume along two axes, while largely consistent with pathological responses, which assess changes on a cellular level by histological analysis, was nevertheless not identical. This discrepancy may be due to several factors. First, cell viability is commonly a measure of cellular activity and health , which is independent of, though related to, proliferation, so we might be observing response to treatment of cells that remain viable (i.e., alive or even secreting growth-promoting factors for example, but not necessarily dividing and contributing to measurable growth at the time of assessment). It is also possible that the differences could be attributed to temporal discrepancy, i.e., the volumetric response assesses changes in tumor size/volume using imaging techniques, while pathological response measurement involves examining tissue samples. As a result, there could be a time lag between the measurement of these volumetric changes and observable pathological alterations. It is also possible that a change in size may occur not due to an increase in the number of cancer cells themselves but due to fibrotic responses, a change that would not necessarily indicate a change in pathological status. Moreover, tumors can be highly heterogeneous, so it is also feasible that a pathological examination could reveal areas of non-uniform response, i.e., containing viable and active cells that were either more resistant to treatment or found themselves in regions that received a suboptimal dose of treatment. Finally, the way the data is binned - into two categories for volumetric versus three for pathological - affects the granularity and distribution of the data, which may contribute to the observed discrepancies. Next, we evaluated the relationship between immune activation and T cell states for responders vs non-responders by measuring IFNg signaling scores in different T cell subsets. The results suggested that successful immune responses in responders are associated with a balanced IFNg signaling profile across various T cell states. While high IFNg is present in later stages of exhaustion (TExInt, TExTerm), responders maintain lower IFNg levels in critical progenitor states (TExProg1, TExProg2). This balance may prevent cells from fully committing to terminal exhaustion, preserving their functionality. Non-responders, on the other hand, showed a skewed profile, where either low or excessively high IFNg signatures dominated across T cell states. This could reflect a scenario where T cells are either too exhausted to contribute effectively or are driven into deeper dysfunction by excessive IFNg signaling, particularly in the TEFF and TExInt states. Finally, this analysis was repeated for the PBMC samples to evaluate whether such differences in T cell composition between responders and non-responders could be detected in peripheral samples. Unfortunately, the results suggest that PBMC data is insufficient to discern differences between responders and non-responders with respect to different states of T cell exhaustion, and TILs need to be analyzed. Our analysis suggests that a highly inflammatory environment may trigger transitions between different stages of exhaustion, consistent with the hypothesis that this mechanism evolved in chronic viral infections for protection from autoimmunity (Kareva & Brown 2024 ). An alternative hypothesis is that it is chronic antigen stimulation that may trigger these transitions (Wherry 2011 ; McLane et al. 2019 ). The two hypotheses—inflammation-driven vs. chronic antigen stimulation-driven exhaustion—are both plausible explanations for triggers that may push T cells down the exhaustion pathway. Based on the data we’ve analyzed, there are arguments for both mechanisms, and it’s possible that a combination of both factors is at play. Our data show that high IFNg signaling is a consistent feature across various T cell exhaustion states, especially in non-responders. The presence of high IFNg in TEM, TExInt, and TExTerm states suggests that sustained inflammatory signaling may contribute to deeper exhaustion. The correlation between higher IFNg in certain subpopulations of T cells in non-responders and poorer outcomes supports the idea that excessive inflammation might exacerbate exhaustion, driving T cells into a less functional state. The persistence of exhaustion markers in TExProg2 cells, which are early progenitor-like exhausted states, along with their proliferative nature, suggests that these cells could be continuously being stimulated by antigens. If these cells are exposed to chronic antigen stimulation without adequate rest or clearance, they may be driven further down the exhaustion pathway regardless of the inflammatory context. The data showing that lower IFNg signaling in this subpopulation cells correlates with better outcomes could imply that reducing antigen load might help preserve these cells’ functionality. It is possible and even likely, however, that both mechanisms are at play, with both inflammation and chronic antigen stimulation contributing to T cell exhaustion. High IFNg signaling likely exacerbates the exhaustion process, especially in cells that are already under chronic antigen stimulation. Conversely, chronic antigen exposure could be the initial driver of exhaustion, with inflammation, while initially beneficial, acting as a secondary factor that deepens the exhaustion state at higher levels. Notably, one limitation of the scRNAseq data is its sparsity, meaning it often captures only a subset of the transcriptome, which can lead to incomplete or biased representations of cellular states. Additionally, scRNA-seq provides only a snapshot in time, making it challenging to determine whether observed changes are directly caused by therapy, inherent differences in response, or other factors. This temporal limitation complicates the interpretation of dynamic processes and the establishment of causal relationships in the context of T cell subpopulations and states. Ideally, it would be highly informative and valuable to have multiple samples to track changes in T cell subpopulations and state proportions over time, which unfortunately might be unfeasible for financial and logistical reasons. In addition, increasing evidence suggests that T cell exhaustion results in extensive epigenetic remodeling (Belk et al. 2022 ; Scott-Browne et al. 2016 ; Gennert et al. 2021 ; Ford et al. 2022 ). Epigenomics data would enable us to analyze changes in global patterns of chromatin accessibility, providing important insights into the underlying mechanisms of T cell exhaustion in the context of immune checkpoint inhibition. The theoretical implications of the existence of multiple states of exhaustion were extensively analyzed in (Kareva & Gevertz 2024 ). The authors developed a mathematical model of tumor-immune interactions, where cytotoxic immune cells can exist in three classes: effector, reversibly exhausted and terminally exhausted; transitions between states were hypothesized to be driven by the systemic inflammation level but a case of antigen-driven transitions was evaluated as well. The differences in immune cell composition with respect to states of exhaustion were observed in this analysis, with a prediction of a higher proportion of terminally exhausted T cells in non-responders. The importance of a balanced “goldilocks” inflammatory environment in maintaining effective response was also revealed in simulations of multiple dosing scenarios. The authors then identified three main qualitative strategies that can be effective in preserving the effectiveness of checkpoint inhibition therapy as monotherapy, including high dose-low frequency approach (MTD-like therapy); low dose high frequency approach (metronomic-like therapy), and an intermediate strategy. Virtual population analysis, where patients were characterized by an individual inflammatory threshold that triggers transition from reversible to terminal exhaustion, was conducted to identify if either of these therapeutic strategies would be likely to work better on a population level. The authors showed that for a heterogeneous population, while a standard MTD-like protocol of administration of checkpoint inhibitors as monotherapy may work, the metronomic-like (low dose high frequency) strategy is more likely to work for a larger fraction of the population. Some recent retrospective analyses support the prediction that lower dosing can be as beneficial as the currently set dosing levels. In Chang et al. (Chang et al. 2022 ), the authors found that both median overall survival (OS) and rate of all classes of immune-related adverse events (irAEs) were similar in both the standard-dose and low-dose pembrolizumab in a cohort of 147 patients with non-small cell lung cancer (NSCLC). In another retrospective analysis, Low et al. (Low et al. 2021 ) showed that reducing the dose of pembrolizumab from a 200 to 100 mg flat dose resulted in no significant difference in response rate or grade 3–4 (severe or life-threatening) irAEs. Similarly for nivolumab, in (Zhao et al. 2021 ), the authors showed that the overall response rate (ORR) in renal cell carcinoma patients was similar in both high and low dose cohorts, and among the patients in the low dose cohort, one patient had complete response (CR); no patients had CR in the high-dose cohort. These studies have numerous limitations, of course, like the small sample size and the nature of conducting a retrospective rather than prospective analysis, but they do suggest that current dosing strategies in ICI administration as monotherapy may be suboptimal. Most recently, interim analysis was presented for DEDICATION-1 trial (NCT04909684), an open label randomized non-inferiority study that assessed the impact of reduced pembrolizumab dose (of approximately 75% depending on treatment schedule) vs standard of care in 750 NSCLC patients (Buma et al. 2023 ). Interim analysis for 256 patients (Heuvel et al. 2024 ) revealed minimal differences in the outcomes, suggesting in the first prospective study that lower doses can be non-inferior. While lower doses may be feasible, lower frequency administration is unlikely to take hold, including for logistical reasons. Well-selected combination therapy approaches may therefore present an attractive alternative to mitigate emergence of terminal exhaustion. In fact, it’s feasible that over-stimulating immune cells and thereby pushing them further down the irreversible exhaustion pathway could theoretically lead to tumor hyperprogression, which, for instance, has been observed in approximately 10% of patients with advanced gastric cancer, treated with anti-PD-1 therapy (Kamada et al. 2019 ). If the unbalanced inflammation that occurs as a result of initially successful tumor elimination accelerates T cell transition towards terminally exhausted state, it could effectively diminish the overall efficacy of the anti-tumor immune response compared to pre-treatment levels. This in turn could result in faster tumor growth due to a paradoxically incapacitated immune response. This hypothesis remains to be evaluated on a mechanistic level. A good combination partner therefore might work to maintain the inflammatory level in the TME at a balanced state, without letting the tumor become “too cold” or “too hot”, both of which are expected to lead to insufficient effector function and consequently worse outcomes. In (Kareva & Gevertz 2024 ), the authors conducted sensitivity analysis to predict, which mechanisms could be targeted as combination partners to “flip” the predicted response from a non-responder to a responder. While most combination therapy approaches tend to focus on augmenting different cytotoxic pathways (Schmidt et al. 2020 ; Palmer et al. 2022 ; Meric-Bernstam et al. 2021 ), within the framework of the existence of multiple states of exhaustion, it was predicted that it may be more beneficial to also mitigate inflammatory responses. Excitingly, two recent studies, one in lung cancer and another in Hodgkin lymphoma, have independently applied this approach by administering Janus kinase (JAK) inhibitors, a class of drugs that help reduce inflammation (Shawky et al. 2022 ), as combination partners with anti-PD-1 therapy. In (Mathew et al. 2024 ), the authors showed that administration of JAK1 inhibitor itacitinib after pembrolizumab resulted in improved therapeutic response both in mouse models and in a Phase 2 non-small cell lung cancer trial. In (Zak et al. 2024 ), in a report of a phase I clinical trial the authors showed that a combination of the JAK inhibitor ruxolitinib with nivolumab, administered to patients with relapsed or refractory Hodgkin lymphoma, yielded an overall response of 53% (10/19 patients), with 6/19 patients achieving a complete response. These results highlight not only the viability of the initially counterintuitive approach of using properly selected immunosuppressive medications to augment ICIs, but also the ability to rescue a response in a patient who previously failed ICIs, converting a non-responder to a responder. Notably, the goal of co-administering anti-inflammatory therapy is not to mitigate the adverse events (an extreme example of this approach was the groundbreaking case of the first pediatric CAR-T patient Emily Whitehead, who received anti-IL-6 drug tocilizumab to mitigate the severity of her cytokine release syndrome following her treatment (Newitt 2022 )). Instead, administration of an anti-inflammatory drug in combination with ICI would serve mechanistically to keep T cells in a more balanced state and prevent transition to terminal exhaustion. The presented analysis has several limitations that need to be addressed in future work. Data for only 27 patients was available, with only 19 post-treatment biopsy samples, and only 6 samples to compare T cell states before and after treatment. It would be particularly informative to expand this analysis to a data set, where a more detailed understanding of T cell state composition can be made in pre- vs post-treatment samples for both responders and non-responders. Obtaining longitudinal TIL samples to gain a deeper insight into the progression of treatment response would additionally improve our understanding of the impact of treatment on T cell states. Further investigations and better understanding of the multiple steps of T cell exhaustion and the triggers for state transitions will help devise better monotherapy and combination treatments to increase the number of responders to checkpoint inhibition therapy, and to even potentially rescue those who did not initially respond. Methods Dataset A single-cell RNA sequencing (scRNA-seq) dataset (GSE200996) published by Luoma et al. (Luoma et al. 2022) was analyzed to investigate the relationship between T cell subsets and response to immune checkpoint inhibitors (ICIs). This dataset consists of samples collected during a phase 2 clinical trial of 27 head and neck squamous cell carcinoma (HNSCC) patients treated with neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4. Both oral tissue samples and peripheral blood mononuclear cells (PBMCs) were analyzed. Two clinical readouts are provided: pathological response (50%) and volumetric response (Response | No Response). Statistical Analysis All analysis presented therein was conducted in R (v4.3.1). Data Integration and Normalization To correct for batch effects and technical variations that may be obscuring biological signal, datasets were harmonized using the standard Seurat (v5.0.1) integration workflow. Individual gene expression matrices were normalized using regularized negative binomial regression, as described in Hafemeister and Satija (Hafemeister & Satija 2019). Features were ranked by residual variance, with the top 3000 selected for each dataset. These selected features were used to identify integration features across datasets, facilitating the discovery of cross-dataset pairs of cells predicted to share a common biological state, termed anchors, through an iterative integration process detailed in (Stuart et al. 2019). Cell-Type Identification To distinguish between different cell types in the absence of a gold standard, we adopted an ensemble approach, combining both supervised and unsupervised learning methods. This approach was favorable as it leveraged the strengths of various techniques to achieve a more robust and reliable classification of cell types. Dimensionality Reduction, Clustering and Differential Expression Analysis We performed Principal Component Analysis (PCA) on the integrated dataset to capture the most significant sources of variation across cells. Utilizing the first 30 principal components from PCA, we then employed Uniform Manifold Approximation and Projection (UMAP) to visualize the data in a lower-dimensional space (Figure A1). Shared nearest neighbor (SNN) graph-based clustering was conducted on the PCA space using a resolution parameter of 0.5 to identify clusters of cells with similar molecular profiles. Highly variable features within each identified cell cluster were determined using the Wilcoxon Rank Sum test from Seurat's FindAllMarkers function. By integrating biological knowledge and considering the tissue context, we assigned specific cellular populations based on the calculated highly variable genes (HVGs). Reference-Based Approaches We used three reference-based approaches that leverage existing annotated gene expression profiles. These included SingleR (Aran et al. 2019), Azimuth (Hao et al. 2021), and a random forest-based cell typing approach (Aybey et al. 2023). For further details on each method, please refer to the original publications. This ensemble approach allowed us to combine the strengths of both supervised and unsupervised methods, enhancing the confidence in labeling cell types in our dataset. T Cell Isolation A majority voting approach was used to isolate T cells, selecting those with a score of 4. This score indicates agreement across all four methods described above, collectively enhancing the accuracy of T cell identification. A nuanced scoring method was subsequently employed for assigning T cells to distinct subgroups, as computational methods are less proficient at distinguishing closely related cells and discerning subtle differences. Treg label score was calculated as follows: +1 for Treg prediction, +0.5 for a CD4 subtype prediction. Non-lineage specific labels such as Tn, Tprof, IFN-resp and HSPhi added a score of 1 for all groups. Unconventional T cells (NKT, MAIT, gdT) were classified as Cytotoxic along with CD8 lineage cells. Cells were classified into the following major subsets based on their final score: Cytotoxic, Helper and Treg. Cells that lacked clear assignment were marked as Inconclusive. Gene Signatures Distinct gene signatures were collated to form a comprehensive collection, incorporating four exhausted T cell subsets from Beltra et al. (Beltra et al. 2020), five T cell states (TN, TSCM, TEM, TCM, TEFF) from Chen et al. (Chen et al. 2021), and the IFN-γ signature from Gao et al. (Gao et al. 2016). TN and TSCM share a similar transcriptomic profile making them indistinguishable at that level, thus the two states were considered as one. Gene Set Enrichment and Differential Composition Analysis To assess differences in cell composition with respect to subtype and state across treatment response, we measured the proportion of each subtype within each group by calculating percentages based on total cell counts. We further assessed relative proportions and significance using scProportionTest (Miller et al. 2021). Gene signature scores were measured using UCell by Andreattaa et al. (Andreatta & Carmona 2021), based on the Mann-Whitney U Test, which calculates scores based on gene expression ranks within individual cells. Additionally, Sargent from Nouri et al. (Nouri et al. 2023) was used to annotate individual cells based on gene signature scores where applicable. Sargent sorts non-zero expressed genes in descending order for each cell before transforming an input gene-by-cell expression matrix into a corresponding gene-set-by-cell assignment score matrix. A gini-index is then computed among assignment scores per cell, transforming the gene-set-by-cell assignment score matrix to a distribution of indexes. Interferon scores were stratified into three categories (low, medium, and high) according to their distribution and position relative to thresholds determined by the median-absolute-deviation (MAD), set at one deviation away from the median. Defining Lineage-Agnostic Transcriptomic Signatures of T Cell Functional States To comprehensively explore the T cell space, it was imperative to first identify the specific signatures not just representing the diverse lineages T cells can differentiate into, but also the functional states they can undertake. T cells predominantly exist in a naive (TN) state until they encounter their cognate antigen, wherein they undergo activation, proliferate, and differentiate into various subsets, including effector (TEFF) and memory cells. The memory cell subsets include stem cell memory T cells (TSCM), central memory T cells (TCM), effector memory T cells (TEM), and tissue-resident memory T cells (TRM). Effector T cells, though transient, execute the immediate T cell response, whilst a subset will persist as memory cells, primed to mount rapid secondary immune responses upon re-exposure to the antigen. The suppressive tumour microenvironment also gives rise to dysfunctional states, including exhaustion, senescence, and anergy, which are characterised by, among other phenotypic features, defective effector functions, decreased proliferative capacity, altered metabolism, and reduced cytotoxicity. These dysfunctional states pose obstacles to effective cancer immunotherapy, such as immune checkpoint inhibitors (ICIs), which aim to prevent or reverse them by alleviating immunosuppressive pathways. Within exhaustive states, T cells have been found to undergo progressive stages of exhaustion, transitioning between substates that exhibit varying degrees of dysfunction. To investigate these exhausted states in the context of ICIs, we leveraged gene signatures from a developmental framework for exhausted T cells by Beltra et al. (2020) that classifies progression from exhausted progenitor (TExProg 1 and 2) to intermediate (TExInt) and terminal (TExTerm) subsets. Whilst TExTerm cells do not respond to PD-1 blockade, the authors found that PD-1 blockade partially re-balances subset distribution by fostering the formation of TExProg2 and TExInt cells. Moreover, their study highlights the therapeutic potential of sustaining the TExInt population whilst preventing transition into TExTerm through modulation of TOX and T-Bet. T cell states are characterized by distinct phenotypic markers, some of which are shared across states at varying levels, underscoring the complexity of T cell biology. Therein, transcriptomic signatures of T cell states were compiled through a comprehensive literature search and analysis of reference datasets. We hypothesized that we could generate gene signatures that are independent of T cell lineage by separating genes related to CD4 and CD8 lineages from those associated with T cell states. In addition, inflammatory signatures were obtained to investigate the relationship between inflammation levels and T cell exhaustion. Figure A1 summarizes the specific gene signatures associated with different T cell states, including substates of T cell exhaustion. The key characteristics of the different T cell states are summarized in Table 1. Declarations Conflicts of Interest The authors are employees of Merck KGaA, which operates as EMD Serono in the U.S. Views presented in this manuscript are the authors’ own and do not necessarily represent the views of Merck KGaA/EMD Serono. Author Contribution IK and CP designed the research and wrote the manuscript; CP performed the analysis. Acknowledgements CP started this work during her first rotation in Oncology Data Science and continued with a more comprehensive and targeted analysis during her second rotation with IK in the Quantitative Pharmacology Department at Merck KGaA from January to May 2024. She would like to thank Dr. Eike Staub and Dr. Michail Yekelchyk for valuable discussions and for directing her to the dataset used for the analysis, as well as Bogac Aybey for technical assistance in running his published cell-type predictor and Dr. Sven-Eric Schelhorn for computational optimization help on the cluster. Special thanks to Dr. Olga Bogatyrova for connecting CP and IK and highlighting their shared interests. Thank you to Dr. Katarzyna Wreczycka for granting me access to her GitHub repository, in which she compiled publicly available gene signatures of interferon signaling. Data Availability The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request. References Andreatta, M. & Carmona, S. J. UCell: Robust and scalable single-cell gene signature scoring. Comput. Struct. Biotechnol. J. 19 , 3796–3798 (2021). Belk, J. A., Daniel, B. & Satpathy, A. T. Epigenetic regulation of T cell exhaustion. Nat. Immunol. 23 (6), 848–860 (2022). Beltra, J. C. et al. Developmental relationships of four exhausted CD8 + T cell subsets reveals underlying transcriptional and epigenetic landscape control mechanisms. Immunity 52 (5), 825–841 (2020). Blackburn, S. D. et al. Selective expansion of a subset of exhausted CD8 T cells by αPD-L1 blockade. Proceedings of the National Academy of Sciences , 105(39), pp.15016–15021. (2008). Buma, A. I. et al. Integrating treatment cost reduction strategies and biomarker research to reduce costs and personalize expensive treatments: an example of a self-funding trial in non-small cell lung cancer. Frontiers in Pharmacology, 14, p.1274532. (2023). Carlson, R. D., Flickinger, J. C. Jr & Snook, A. E. Talkin’toxins: from Coley’s to modern cancer immunotherapy. Toxins, 12(4), p.241. (2020). Chang, K. C. et al. Comparative Effectiveness and Safety of Standard-Dose and Low-Dose Pembrolizumab in Patients with Non-Small-Cell Lung Cancer: A Multi-Institutional Cohort Study in Taiwan. Cancers, 14(5), p.1157. (2022). Chen, G. M. et al. Integrative bulk and single-cell profiling of premanufacture T-cell populations reveals factors mediating long-term persistence of CAR T-cell therapy. Cancer Discov. 11 (9), 2186–2199 (2021). Coley, W. B. The treatment of malignant tumors by repeated inoculations of erysipelas: With a report of ten original cases. 1. The American Journal of the Medical Sciences (1827–1924), 105(6), p.487. (1893). Dobosz, P. & Dziecikatkowski, T. The intriguing history of cancer immunotherapy. Front. Immunol. 10 , 2965 (2019). Dutcher, J. P. et al. High dose interleukin-2 (Aldesleukin)-expert consensus on best management practices-2014. J. Immunother. Cancer . 2 , 1–23 (2014). Ford, B. R. et al. Tumor microenvironmental signals reshape chromatin landscapes to limit the functional potential of exhausted T cells. Sci. Immunol. 7 (74), eabj9123 (2022). Gao, J. et al. Loss of IFN-γ pathway genes in tumor cells as a mechanism of resistance to anti-CTLA-4 therapy. Cell 167 (2), 397–404 (2016). Gennert, D. G. et al. Dynamic chromatin regulatory landscape of human CAR T cell exhaustion. Proceedings of the National Academy of Sciences, 118(30), p.e2104758118. (2021). Hafemeister, C. & Satija, R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome biology , 20(1), p.296. (2019). van den Heuvel, M., van der Noort, V., Heine, T. & R 1258MO Low dose versus standard dose pembrolizumab for treatment of stage IV stage non-small cell lung carcinoma: Results of the pre-planned interim analysis of the NVALT-30 clinical trial. Ann. Oncol. 35 , S806 (2024). Huang, A. C. et al. T-cell invigoration to tumour burden ratio associated with anti-PD-1 response. Nature 545 (7652), 60–65 (2017). Ishida, Y. et al. Induced expression of PD-1, a novel member of the immunoglobulin gene superfamily, upon programmed cell death. EMBO J. 11 (11), 3887–3895 (1992). Kahan, S. M., Wherry, E. J. & Zajac, A. J. T cell exhaustion during persistent viral infections 479pp.180–193 (Virology, 2015). Kamada, T. et al. PD-1 + regulatory T cells amplified by PD-1 blockade promote hyperprogression of cancer. Proceedings of the National Academy of Sciences, 116(20), pp.9999–10008. (2019). Kareva, I. & Brown, J. S. Evolutionary and ecological perspective on the multiple states of T cell exhaustion. In Cancer through the Lens of Evolution and Ecology. CRC, 92–104. (2024). Kareva, I. & Gevertz, J. L. Mitigating non-genetic resistance to checkpoint inhibition based on multiple states of immune exhaustion. npj Systems Biology and Applications , 10(1), p.14. (2024). Leach, D. R., Krummel, M. F. & Allison, J. P. Enhancement of antitumor immunity by CTLA-4 blockade. Science 271 (5256), 1734–1736 (1996). Lee, J. B., Kim, H. R. & Ha, S. J. Immune checkpoint inhibitors in 10 years: contribution of basic research and clinical application in cancer immunotherapy. Immune Netw. , 22 (1). (2022). Li, Y. et al. Potential anti-tumor effects of regulatory T cells in the tumor microenvironment: a review. J. Translational Med. 22 (1), 293 (2024). Liu, Q., Sun, Z. & Chen, L. Memory T cells: strategies for optimizing tumor immunotherapy. Protein cell. 11 (8), 549–564 (2020). Low, J. L. et al. Low-dose pembrolizumab in the treatment of advanced non-small cell lung cancer. Int. J. Cancer . 149 (1), 169–176 (2021). Luoma, A. M. et al. Tissue-resident memory and circulating T cells are early responders to pre-surgical cancer immunotherapy. Cell 185 (16), 2918–2935 (2022). Mathew, D. et al. Combined JAK inhibition and PD-1 immunotherapy for non–small cell lung cancer patients. Science 384 (6702), eadf1329 (2024). McKinney, E. F. et al. T-cell exhaustion, co-stimulation and clinical outcome in autoimmunity and infection. Nature 523 (7562), 612–616 (2015). McLane, L. M., Abdel-Hakeem, M. S. & Wherry, E. J. CD8 T cell exhaustion during chronic viral infection and cancer. Annu. Rev. Immunol. 37 , 457–495 (2019). Meric-Bernstam, F. et al. Enhancing anti-tumour efficacy with immunotherapy combinations. Lancet 397 (10278), 1010–1022 (2021). Miller, B. C. et al. Subsets of exhausted CD8 + T cells differentially mediate tumor control and respond to checkpoint blockade. Nat. Immunol. 20 (3), 326–336 (2019). Miller, S. A. et al. LSD1 and aberrant DNA methylation mediate persistence of enteroendocrine progenitors that support BRAF-mutant colorectal cancer. Cancer Res. 81 (14), 3791–3805 (2021). Newitt, V. N. The incredible story of Emily Whitehead & CAR T-cell therapy. (2022). Nouri, N. et al. A marker gene-based method for identifying the cell-type of origin from single-cell RNA sequencing data. MethodsX 10 , 102196 (2023). Palmer, A. C. et al. Predictable clinical benefits without evidence of synergy in trials of combination therapies with immune-checkpoint inhibitors. Clin. Cancer Res. 28 (2), 368–377 (2022). Robert, C. A decade of immune-checkpoint inhibitors in cancer therapy. Nat. Commun. 11 (1), 3801 (2020). Rosenberg, S. A. et al. Treatment of 283 consecutive patients with metastatic melanoma or renal cell cancer using high-dose bolus interleukin 2. Jama 271 (12), 907–913 (1994). Schmidt, E. V. et al. Assessment of clinical activity of PD-1 checkpoint inhibitor combination therapies reported in clinical trials. JAMA Netw. open. 3 (2), e1920833–e1920833 (2020). Schoenfeld, A. J. & Hellmann, M. D. Acquired resistance to immune checkpoint inhibitors. Cancer cell. 37 (4), 443–455 (2020). Scott-Browne, J. P. et al. Dynamic changes in chromatin accessibility occur in CD8 + T cells responding to viral infection. Immunity 45 (6), 1327–1340 (2016). Shawky, A. M. et al. A comprehensive overview of globally approved JAK inhibitors. Pharmaceutics, 14(5), p.1001. (2022). Stuart, T. et al. Comprehensive integration of single-cell data. cell 177 (7), 1888–1902 (2019). Wherry, E. J. T cell exhaustion. Nat. Immunol. 12 (6), 492–499 (2011). Zak, J. et al. JAK inhibition enhances checkpoint blockade immunotherapy in patients with Hodgkin lymphoma. Science 384 (6702), eade8520 (2024). Zhao, J. J. et al. Low-dose nivolumab in renal cell carcinoma: a real-world experience. Oncology 99 (3), 192–202 (2021). Tables Table 1 is available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Tables.docx SupplementaryFigures.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5663382","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":423456319,"identity":"63e0b3bf-1303-4c17-b958-9562194f924a","order_by":0,"name":"Irina Kareva","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAsklEQVRIiWNgGAWjYBACNh4ILQflkqDFmHgtDFAtiQ1Ea+HjOfzsw8c2u/T+/jMGDB/KDhPhMN4245kz25JzZ9zIMWCccY4YLfwMxsw8Z5hzG27wGDDzthGlhf0z858z9eny588YMP8lSgtvjzEzQ8XhBIMDOQbMjERp4TlTzNhTcdxw4420goM959IJa5HvSd/M8MOgWl7u/OGND36UWRPWggIOkKh+FIyCUTAKRgEuAADY8jUtaz+HrAAAAABJRU5ErkJggg==","orcid":"","institution":"EMD Group (United States)","correspondingAuthor":true,"prefix":"","firstName":"Irina","middleName":"","lastName":"Kareva","suffix":""},{"id":423456320,"identity":"7c6c2342-0e10-4c79-a95f-361fb3a149fd","order_by":1,"name":"Clara Pavillet","email":"","orcid":"","institution":"Merck KGaA","correspondingAuthor":false,"prefix":"","firstName":"Clara","middleName":"","lastName":"Pavillet","suffix":""}],"badges":[],"createdAt":"2024-12-17 16:08:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5663382/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5663382/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":78246481,"identity":"53774a83-ab82-4e97-a196-fc4cd47ecadb","added_by":"auto","created_at":"2025-03-11 09:29:52","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":352959,"visible":true,"origin":"","legend":"\u003cp\u003ePost-treatment tumor Microenvironment.\u003cstrong\u003e \u003c/strong\u003eA. Weighted kernel density estimation UMAP visualization of selected cell type markers. B. Cell type proportion in tumor tissue across patients, categorized by volumetric response. C. Bar plots illustrating the mean proportion of the Cyclic T/NK cell population in volumetric responders versus non-responders. D. Relative cell type proportion differences assessed using scProportionTest; red dots indicate statistically significant results (FDR \u0026lt; 0.05 and absolute log2 fold change \u0026gt; 0.58), with larger log2 fold changes reflecting a higher proportion of cells in responders and lower in non-responders. E. Average pseudobulk expression of inflammatory mediators comparing non-responders and responders. F. Single-cell violin plots depicting the expression level of IFNG across volumetric responses, analyzed using a Wilcoxon rank sum test (p \u0026lt; 0.001). G. Differential enrichment levels of two gene signatures from (Borcherding et al. 2021)\u003cstrong\u003e, \u003c/strong\u003emeasured by UCell and normalized by the number of genes expressed in each gene set and cell.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/b2642d657b27a7088db0117d.png"},{"id":78246490,"identity":"79bc0d16-7a17-43fa-b33e-4361444d689c","added_by":"auto","created_at":"2025-03-11 09:29:52","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":448604,"visible":true,"origin":"","legend":"\u003cp\u003ePredicted Tumor-Infiltrating T Cell Identities. (A) UMAPs of subsets of tumor-infiltrating lymphocytes (TILs) identified through a clustering-based method for major populations (B) A more granular subgrouping of subsets of tumor-infiltrating lymphocytes (TILs) based on characteristic gene markers (bottom). (C) UMAPs labeling predicted states projected onto the TIL subset space. (D) Alluvial plot displaying T cell classification predictions from different methods, with 'Elected' predictions representing the final vote based on majority voting across the other methods.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/75fe839bfb8b883cb80cba7b.png"},{"id":78246479,"identity":"81fd6db8-9a2c-462d-9273-09a7de95a4b6","added_by":"auto","created_at":"2025-03-11 09:29:52","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":133566,"visible":true,"origin":"","legend":"\u003cp\u003eTreatment Effect on T Cell Subtypes and States. (A) Proportion of major T cell subtypes pre- and post-treatment. (B) Percent change in different cell types pre- and post-treatment; note the log scale on y-axis. (C) Proportion of T cell states, including various states of exhaustion, pre- and post-treatment. (D) Percent change of the change in different T cell states pre- and post-treatment, with the y-axis in log scale.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/c2250b4d93600dc1a0610914.png"},{"id":78247150,"identity":"1fc6dd9e-aee3-4836-bf3e-5840845615d2","added_by":"auto","created_at":"2025-03-11 09:37:52","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":169521,"visible":true,"origin":"","legend":"\u003cp\u003eT Cell Identities on Pathological and Volumetric Responses. (A) Bar plots showing proportions of different elected T cell subtypes in volumetric response, with non-responders (left) and responders (right). (B) Bar plots showing proportions of different elected T cell subtypes in pathological response, with levels subdivided as \u0026lt;10% (left), between 10% and 50% (middle), and \u0026gt;50% (right). (C) Post-treatment change in proportion of T cell states, including various states of exhaustion, for non-responders (left) and responders (right) in volumetric response. (D) Post-treatment change in proportion of T cell states, including different states of exhaustion, for pathological response.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/4725867b838db9e228b46bd1.png"},{"id":78249136,"identity":"f7aae27d-2ef2-4090-aeb7-a9c37c215eda","added_by":"auto","created_at":"2025-03-11 09:46:21","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":440862,"visible":true,"origin":"","legend":"\u003cp\u003eType II Interferon Signaling Levels Across T Cell Identities. (A) Violin plots of IFNg signature Mann-Whitney U statistic derived UCell signature scores across different T cell states. (B) Distribution of IFNg signature scores with thresholds to classify scores as high, medium, or low based on one median-absolute-deviation away from the median.\u003c/p\u003e\n\u003cp\u003e(C) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell identity in volumetric response non-responders. (D) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in volumetric response responders. (E) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with \u0026lt;10%. (F) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with 10-49% response. (G) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with \u0026gt;50% response.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/2de2839f9c7ce82e3618fbee.png"},{"id":78246491,"identity":"3d28ee0f-2663-45fe-9f96-c06b45b53c35","added_by":"auto","created_at":"2025-03-11 09:29:52","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":170584,"visible":true,"origin":"","legend":"\u003cp\u003ePredicted Peripheral T Cell Identities. (A) UMAPs of subsets of peripheral blood mononuclear cells (PBMCs) identified through a clustering-based method for major populations (top) and more granular subgrouping based on characteristic gene markers (bottom). (B) UMAPs labeling predicted states projected onto the PBMC T cell subset space.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/371a15ebcdf3d7baa84fbd83.png"},{"id":78250585,"identity":"199e6be5-8824-4abd-bab5-72b0b621e0d0","added_by":"auto","created_at":"2025-03-11 09:53:55","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":134725,"visible":true,"origin":"","legend":"\u003cp\u003eTreatment Effect on Peripheral T Cell Subtypes and States. (A) Proportion of major T cell subtypes across three time points (B1, B2, B3) over the course of treatment. (B) Percent change in different cell types across the three time points (B1, B2, and B3), with the y-axis in log scale.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/45b3bfc002db19325a473d11.png"},{"id":78246486,"identity":"aff0ca76-4e2c-46fa-afe3-d44ab916739f","added_by":"auto","created_at":"2025-03-11 09:29:52","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":170324,"visible":true,"origin":"","legend":"\u003cp\u003ePeripheral T Cell Identities and States Across Time Points. (A) Difference in proportions of major T cell subtypes (elected) in PBMCs for non-responders (left) and responders (right) for volumetric response during the first time point B1. (B) Difference during the second time point B2. (C) Difference during the third time point B3. (D) Post-treatment change in proportion of T cell states in PBMCs, including different states of exhaustion, for non-responders (left) and responders (right) for volumetric response during the first time point B1. (E) Difference during the second time point B2. (F) Difference during the third time point B3.\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/90d690937b2ac6e5b863ce87.png"},{"id":92133884,"identity":"b1c33e38-07e2-4924-bc61-c165cc76e4aa","added_by":"auto","created_at":"2025-09-25 03:53:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2768826,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/3f569333-f87e-4219-9722-543e33a235df.pdf"},{"id":78246478,"identity":"2f5997e6-c025-4595-bd40-249fd9a0fbfd","added_by":"auto","created_at":"2025-03-11 09:29:52","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":16434,"visible":true,"origin":"","legend":"","description":"","filename":"Tables.docx","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/65a441b884a63fddad4a54e8.docx"},{"id":78249092,"identity":"63df2f1d-0159-46d7-b02a-82c0a62ab8af","added_by":"auto","created_at":"2025-03-11 09:45:52","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":550789,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFigures.docx","url":"https://assets-eu.researchsquare.com/files/rs-5663382/v1/b1aa76fafe1fbb89f6e5d755.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Heterogeneity of Exhausted T Cell Subsets in Responders and Non-Responders Following Checkpoint Inhibition Therapy","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe history of immunotherapy reflects a long journey from skepticism to recognition of the immune system's potential to fight cancer (Dobosz \u0026amp; Dziecikatkowski \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Early observations, such as those by William Coley in the late 19th century, hinted at this when he noted that bacterial infections could sometimes lead to spontaneous tumor regressions (Coley \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e1893\u003c/span\u003e). This led to the development of \u0026ldquo;Coley\u0026rsquo;s toxins,\u0026rdquo; a pioneering but ultimately unstandardized immune-based approach to cancer treatment (Carlson et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). It wasn\u0026rsquo;t until decades later that immunotherapy made significant strides, with Steven Rosenberg\u0026rsquo;s work in the 1980s on cytokine therapy, particularly IL-2, marking the next critical milestone (Rosenberg et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e1994\u003c/span\u003e). Although IL-2-based therapies were hampered by toxicity (Dutcher et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), they demonstrated that the immune system could be mobilized against cancer, providing a foundation for future advances.\u003c/p\u003e \u003cp\u003eThe discovery of immune checkpoints like CTLA-4 and PD-1 in the early 1990s was a pivotal breakthrough that finally unlocked the power of immunotherapy (Leach et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e1996\u003c/span\u003e; Ishida et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1992\u003c/span\u003e). These molecules, which act as brakes on the immune system to prevent autoimmunity, became the targets of drugs known as checkpoint inhibitors. The approval of ipilimumab in 2011 and pembrolizumab in 2016 marked the beginning of a new era in cancer treatment, where the immune system could be harnessed more effectively (Robert \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). While immune checkpoint inhibition (ICI) has been nothing short of revolutionary, only a fraction of patients respond (Schoenfeld \u0026amp; Hellmann \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Lee et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and the factors that separate responders from non-responders remain to be elucidated.\u003c/p\u003e \u003cp\u003eCheckpoint inhibitors target T cell exhaustion, a state of T cell dysfunction characterized by a loss of effector function and proliferative capacity, along with the upregulation of inhibitory receptors such as PD-1 (programmed cell death protein 1), CTLA-4 (cytotoxic T-lymphocyte-associated protein 4), TIM-3 (T-cell immunoglobulin and mucin-domain containing-3), and LAG-3 (lymphocyte activation gene-3). It is becoming increasingly understood that checkpoint expression is a mechanism that has evolved to protect against autoimmunity, particularly in chronic viral infections (Kahan et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; McLane et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), since in the context of a chronic infection that cannot be fully eradicated, maintaining a state of exhaustion may be less detrimental to the host\u0026rsquo;s overall fitness than continuous active cytotoxicity (Kareva \u0026amp; Brown \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). McKinney et al. (McKinney et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2015\u003c/span\u003e) support this hypothesis by showing that the transcriptional profiles associated with CD8\u0026thinsp;+\u0026thinsp;T cell exhaustion during chronic infection correlate with both poorer infection clearance and a reduced risk of developing autoimmune diseases.\u003c/p\u003e \u003cp\u003eIn recent years, it has become evident that T cell exhaustion is not a binary condition but rather a spectrum of states, with only certain stages being responsive to reinvigoration through checkpoint inhibition therapy. These distinct stages of exhaustion represent varied degrees of dysfunction. Initially, exhausted T cells may exhibit impaired cytokine production and reduced cytotoxic activity yet retain some functionality and proliferative capability. As T cells progress toward terminal exhaustion, there is a significant increase in the expression of inhibitory receptors and a decrease in co-stimulatory molecules. It appears that the mechanisms that evolved to protect against autoimmunity in chronic viral infections are like the mechanisms that can prevent effective tumor elimination by the immune system. In (Miller et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), the authors reported that analogous subsets of exhausted CD8\u0026thinsp;+\u0026thinsp;T cells are observed both in tumors and in chronic viral infections, including progenitor exhausted cells that are more responsive to checkpoint inhibitor therapy (Blackburn et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Huang et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), and terminally exhausted T cells, which are not. Interestingly, in the earlier stages of T cell exhaustion, these cells retain the ability to self-renew and may give rise to more dysfunctional cells. In subsequent work, Beltra et al. (Beltra et al. \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) proposed an expanded four-cell-stage developmental framework for exhausted CD8\u0026thinsp;+\u0026thinsp;T cells (Tex): Ly108\u0026thinsp;+\u0026thinsp;CD69+ (progenitor 1, Texprog1), Ly108\u0026thinsp;+\u0026thinsp;CD69- (progenitor 2, Texprog2), Ly108-CD69- (intermediate, Texint), and Ly108-CD69+ (terminal, Texterm), with each subset defined by its own unique transcriptional and epigenetic profile.\u003c/p\u003e \u003cp\u003eBased on these observations, we hypothesized that one possible distinction between responders and non-responders to ICIs is the immune cell composition with respect to different states of T cell exhaustion. To evaluate this, we compiled gene signatures characteristic of various T cell states, including substates of exhaustion from Beltra et al. (Beltra et al. \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). We then applied these signatures to analyze a published single-cell RNA sequencing (scRNA-seq) dataset of samples collected during a phase 2 clinical trial of head and neck squamous cell carcinoma (HNSCC) patients treated with neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4 (Luoma et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). We evaluate differences in T cell states and inflammatory markers between responders and non-responders, both in tumor infiltrating lymphocytes (TILs) and peripheral blood mononuclear cells (PBMCs). We conclude with a discussion of the implications of the observed differences, both for monotherapy and combination therapy.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eAssessing Local Inflammatory Responses\u003c/h2\u003e \u003cp\u003eThe initial processed blood dataset from Luoma et al. (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) comprised 68,183 cells across 27 patients, whilst the tissue dataset consisted of 56,537 cells from 19 of these 27 patients. Further filtering was applied to obtain post-treatment tissue samples, resulting in a dataset of 40,856 cells. Based on volumetric response measurements, 10 patients were classified as responders and 9 as non-responders. We conducted a detailed analysis of the cellular landscape by subdividing it into major cell types and further into more granular categories based on clusters and differentially expressed genes (Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA and \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB). This expanded the scope of the initial study by including non-CD45\u0026thinsp;+\u0026thinsp;cells. Contributing to the inflammatory tumorigenic environment was a population of collagen-rich (COL3A1, COL1A2, COL6A1, COL6A2) cancer-associated fibroblasts (CAFs) characterized by high expression of DCN, LUM, PRRX1, MMP2, FSTL1 and SPARC. CAFs in non-responders were found to be associated with a heightened inflammatory profile, a denser extracellular matrix, and the expression of immunosuppressive factors, collectively contributing to a microenvironment that may be less conducive to effective immune responses. Significantly upregulated genes in non-responder CAFs included CD34, CFD, PI16, TNXB, MFAP5, CLEC3B, PCOLCE2, ADH1B, CD70, IL33, and TGFBR2. Whilst the upregulation of genes like VIT, CD34, and CLEC3B may further contribute to a more inflammatory environment, the higher expression of collagen-related genes, along with markers of enhanced cell adhesion such as MFAP5 and TNXB, indicates a physical barrier to immune cell infiltration. Additionally, the expression of TGFBR3 and IL33 points to an IL-33-TGF-β niche signaling pathway that has been shown to suppress T cell activity. Myeloid cells could be further subdivided into pro-inflammatory subpopulations, including FCN1\u0026thinsp;+\u0026thinsp;tumor-associated macrophages (TAMs), anti-inflammatory subpopulations such as plasmacytoid dendritic cells (pDCs) and M2-like tumor-associated macrophages (TAMs), and mixed inflammatory subpopulations that expressed a balance of immune activation and suppression molecules. Mast cells expressed high levels of tryptases (TPSAB1, TPSB2, TPSD1, TPSG1), lipid mediators (COX-1, COX-2) and other potent molecules that contribute to inflammatory responses including LTC4S. In tissue, non-responders exhibited a significantly higher proportion of the Cycling T/NK group post-treatment compared to responders, underscoring the need to investigate this population further and the subgroups it comprises (Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC and \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eD). Whilst cycling T cells typically indicate activation, this paradox may stem from cells with impaired effector functions that have yet to lose their proliferative ability. Additionally, some cycling cells might be regulatory T cells, that work by suppressing anti-tumour responses. At the sample level, no responders also showed an increase in IFNγ expression compared to responders, as measured at both the single-cell and pseudobulk levels (Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eE and \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eF). Additionally, a pro-inflammatory gene signature was found to be upregulated in the samples from non-responders, while an anti-inflammatory gene signature showed a slight upregulation in responders (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eG).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eDissecting the Heterogeneous Landscape of Exhausted Tumour-Infiltrating T Cells\u003c/h3\u003e\n\u003cp\u003eBased on the results presented in the previous section, we hypothesized that T cells contained subsets of exhaustion that retained proliferative capacity and effective functions which could explain why non-responders had more Cycling T/NK cells. The dataset from Luoma et al. (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) was reexamined in light of the publication by Beltra et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), which provided signatures of T cell exhaustion subsets. This reanalysis aimed to enhance our understanding of T cell states by integrating the findings from Beltra et al. into the analysis of tumor-infiltrating T cells in response to neoadjuvant ICB. By doing so, we sought to identify potential markers and pathways that could better delineate the spectrum of T cell responses beyond the classical definitions of exhaustion. Using the combined pre- and post-treatment dataset, T cells were isolated from the rest of the population, and further refined using an ensemble approach. Beyond T cell subtypes, T cells can exist along a gradient of states, each with unique functional properties and implications for immune responses. Our detailed analysis of T cells enabled us to break down T cell populations at various levels of granularity and into distinct groups based on their functional states, phenotypic characteristics, and molecular signatures. By dissecting these states, we aimed to understand the specific components and characteristics that define each group. This multi-level analysis provides deeper insights into the heterogeneity of T cell responses and helps us identify the underlying mechanisms that drive their behavior in different contexts. The reasoning behind this approach is that there is no definitive way to ascertain the 'true' label of T cells. The method used therein allows us to integrate multiple models or perspectives, enhancing the robustness and reliability of our T cell labelling. By leveraging the strengths of various cell identification methods, as depicted in the alluvial plot in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD, we aim to capture a more comprehensive understanding of T cell subsets and states, despite the inherent uncertainties in labeling. For the prediction of T cell subtypes, we harmonized results across the different algorithms before using a majority voting approach in order to classify major subtypes in the dataset and most accurately evaluate proportions.\u003c/p\u003e \u003cp\u003eUsing the gene signatures described in the methods section, eight distinct T cell states were predicted, specifically TN/TSCM, TCM, TEM, TEFF, TExProg1, TExProg2, TExInt and TExTerm. The proportion of each predicted state was evaluated across each predicted T cell functional group. As can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, in this dataset, TCM are the most prevalent. These cells are characterized by their ability to rapidly proliferate and differentiate into effector cells upon re-exposure to antigen, making them a crucial component of the adaptive immune response both for immune surveillance and memory formation. TExTerm were mainly observed in cytotoxic T cells from the cluster characterized by high GZMB and CTSW expression. This distribution highlights the diverse functional roles of T cell subsets and their potential implications for immune responses. The TExProg2 cells were found to be associated with proliferating T cells.\u003c/p\u003e \u003cp\u003eCD4 T cells are generally less prone to exhaustion compared to CD8 T cells due to their distinct roles and biology. CD4 T cells have a lower activation threshold and primarily help other immune cells by secreting cytokines. As a result, they experience less direct antigen exposure, reducing the likelihood of chronic stimulation and thus exhaustion. In contrast, CD8 T cells are directly involved in the killing of tumour cells, requiring stronger and more sustained activation, making them more susceptible to exhaustion. Accordingly, in our dataset, exhausted states were mainly found in CD8 T cells, but different subsets of cytotoxic T cells exhibited different predicted substates of exhaustion (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). In Beltra et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), TExProg1, defined as quiescent resident, had restricted ex vivo proliferative potential but exhibited robust proliferation upon antigen stimulation. They were characterized by low in vivo proliferation or ongoing cell cycle activity, with little evidence of ex vivo cell death. Genes upregulated in TExProg1 were involved in progenitor biology (TCF7, SELL), T follicular biology (CXCR5, ICOS), and positive co-stimulation (CD28). In our dataset, TExProg 1 cells were few and sparsely identified across cytotoxic subgroups. Beltra et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) characterized TExProg 2 cells as proliferative circulating cells that gained access to the blood and exhibited a decline in cytokine coproduction. Like TExProg 1, there was little evidence of ex vivo cell death in this subset, which was preferentially amplified via PD-1 blockade. Interestingly, their accumulation was promoted by PD-L1 blockade. In accordance with their high proliferative potential, we identified TExProg 2 in proliferating cells characterized by high expression of Ki67 and other proliferating markers. In Beltra et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), TExInt cells were labeled as circulating and mildly cytotoxic. They also experienced a decline in cytokine coproduction and accumulated in response to PD-L1 blockade, like TExProg 2. These cells were almost incapable of undergoing cell division upon re-stimulation and were readily detectable as apoptotic. In our dataset, TExInt cells mapped to CD8 T cells but were not specific to any particular subset of CD8 T cells. In Beltra et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), TExTerm cells were associated with tissue residency and were incapable of undergoing cell division upon re-stimulation in vitro. These cells were readily detectable as apoptotic. They exhibited enriched pathways for the negative regulation of cell activation but showed signs of recent TCR signaling, including markers like ZAP70 and NFATC1, and pathways related to calcium influx, in addition to enrichment in interferon-related transcription factor motifs. We found that TExTerm cells were highly represented across granzyme-expressing subsets of T cells. Naive T cells on the other hand were present in lower numbers, with a predominant representation in the CD4 subset expressing CCR7, IL7R and KLF2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). Central memory T cells constitute the majority across all groups of T cells. The TME presents tumor-associated antigens that lead to the differentiation of na\u0026iuml;ve T cells into central memory, which have the unique ability to survive long-term in tissues and help establish long-lasting immune memory. This persistence allows rapid reactivation upon antigen re-encounter, crucial for immune surveillance and tumor response. Moreover, memory T cells, including stem cell memory and central memory T cells, show greater persistence and antitumor immunity than effector memory and effector T cells. The TCM/TEFF ratio has also been identified as a predictive biomarker for immune responses in certain tumors (Liu et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Our dataset indicates that effector (TEFF) and effector memory (TEM) T cells are predominantly found within the cytotoxic T cell population in the TME. This suggests that it is these T cells that have been activated in response to tumor antigens and are involved in executing immune responses. Effector memory cells provide long-term immunity and are ready to respond quickly upon re-exposure to antigens.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eTreatment Effect on T Cell Compartments\u003c/h3\u003e\n\u003cp\u003eNext, to understand the impact of therapy on T cell composition, pre- and post-treatment conditions, which were only available for 6 of the 19 patients, were analyzed separately (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFirstly, as can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB, there is a notable difference in immune cell composition before and after treatment, with increases in helper, Treg and inconclusive cells, and a notable decrease in cytotoxic cells (note the logscale in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Specifically, the most significant change occurs in cytotoxic T cells, with a 9% decrease post-treatment; helper T cells show a 4.8% increase in the post-treatment data set and Tregs exhibit 8.8% increase post-treatment. The increase in helper T cells post-treatment could indicate an adaptive response attempting to support immune functions. Helper T cells might also be transitioning towards regulatory phenotypes, as suggested by the increase in Tregs. Increase in Tregs post-treatment may account for decrease in the proportion of cytotoxic T cells as an immune regulatory mechanism.\u003c/p\u003e \u003cp\u003eFurthermore, there is a notable change in the distribution of T cells with respect to cell state (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD), with a 54.4% increase in terminally exhausted (TexTerm), as well as 92.9% increase in TExProg2 and 45.2% increase in TEM, and a decrease in TEFF (-52.2%), TExProg1 (-69.2%) and TExInt cells (-24.7%). The decrease in TCM (-5.3%) and the moderate increase in TN (+\u0026thinsp;17.6%) suggest a subtle shift away from central memory towards na\u0026iuml;ve-like states. This might be part of a replenishment mechanism or a response to the treatment-induced changes in the immune landscape. The decrease in TExInt (-24.7%) could imply that fewer cells are in the transitional phase between progenitor and terminal exhaustion, potentially due to progression towards terminal exhaustion or apoptosis.\u003c/p\u003e \u003cp\u003eOverall, decrease in the effector state and increase in cell states associated with exhaustion might suggest that treatment may be triggering transitions towards the multiple states of exhaustion, which in turn suggests that it is possible that pre- vs post-treatment differences could be at least partially attributed to the impact of treatment rather than being an intrinsic property of the patients.\u003c/p\u003e \u003cp\u003eThe observed increase in CD4 T cells alongside a decrease in CD8 T cells post-treatment could be due to differential response to immune checkpoint blockade. CD4 T cells, particularly helped T cells, may respond more robustly to treatment. The cytokine milieu post-treatment may favor expansion of certain T cell subpopulations over others. PD-1 blockade and anti-CTLA4 can alter the TME differently, increasing the levels of certain cytokines such as IL-2 and IL-7, which promote the proliferation and survival of T cell subsets. This shift in cytokine balance may create a more supportive environment for CD4 T cells over CD8 T cells, leading to their increased presence in the tumor. Meanwhile, the same cytokine changes may not equally support CD8 T cell expansion, contributing to their relative decrease. In addition, CD8 T cells are more prone to exhaustion than CD4 T cells as described earlier. CD8 T cells might be undergoing functional changes or facing other inhibitory pathways that are not fully reversed by immune checkpoint blockade alone. This may result in a slower or less pronounced recovery compared to CD4 T cells.\u003c/p\u003e \u003cp\u003eThe increase in TEM cells suggests that some T cells are being reactivated and are transitioning into a memory state. PD-1 blockade can enhance the function and survival of memory T cells, leading to their accumulation, whilst a decrease in effector T cells could be due to exhaustion, apoptosis or differentiation into other states such as TEM or exhaustive subsets. Fewer T cells find themselves at the initial stage of exhaustion (TExProg1) as they may be progressing towards more advanced stages of exhaustion due to continuous antigen exposure and insufficient recovery. Alternatively, they might actually be benefiting from the immune checkpoint blockade provided by the therapy. Treatment can partially reverse the exhausted state, allowing these cells to regain functionality and potentially transition out of the exhausted state into more active or memory states. Thus, the observed decrease in TExProg1 cells may be reflecting a combination of progression towards deeper exhaustion and successful reactivation by the therapy. More specifically, when inhibitory pathways are blocked, some exhausted T cells can recover their effector functions and proliferative capacity. This reactivation can lead to their differentiation into effector memory T cells, which are characterised by their ability to respond rapidly to antigens and provide long-term immune surveillance. The increase in TExProg2 observed in our dataset could be attributed to their retained proliferative capacity. TExProg2 cells, whilst still in an exhausted state, have not fully lost their ability to proliferate and can respond to stimulation to some extent. PD-1/CTLA4 blockade therapy could partially reverse the exhaustion in these cells, enabling them to proliferate more effectively. This would result in an accumulation of TExProg2 cells within the TME and thereby contribute to the observed increase.\u003c/p\u003e\n\u003ch3\u003eT Cell States are Key Determinants of Treatment Response\u003c/h3\u003e\n\u003cp\u003eNext, we wanted to evaluate whether there exists a difference in T cell composition with respect to exhaustion states between patients who responded to treatment compared to those who did not (responders versus non-responders). For this, two types of clinical readouts were evaluated, pathological response and volumetric response. The former evaluates the extent of tumor regression based on histological examination of tissue samples, whilst the latter evaluates changes in the size or volume of the tumor based on imaging studies. For this part of the analysis, post-treatment TIL data were available for 19 out of 27 patients.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eLet us first analyze the volumetric data (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). Responders have a higher proportion of helper cells (35.4%) compared to non-responders (26.7%), suggesting that a stronger helper T cell presence might be associated with better outcomes, possibly aiding in the activation and maintenance of other immune cells. There is an increase in Tregs among responders (22.0% vs. 16.2%). While Tregs are typically immunosuppressive, their role here might involve maintaining immune homeostasis and keeping inflammation within a certain threshold to prevent further exhaustion of treatment-responsive T cells and the tapering of their functions. Non-responders exhibit a higher proportion of cytotoxic cells (48.1%) compared to responders (33.6%). This finding aligns with the earlier analysis, which indicated that most cytotoxic T cells are concentrated in a terminally exhausted state (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC). Consequently, in non-responders, the transition further along the exhaustion pathway results in cytotoxic cells being less effective in their function, despite their higher numbers.\u003c/p\u003e \u003cp\u003eNext, we examine T cell states and the differences between responders and non-responders as illustrated in Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC; relative T cell state proportion differences assessed using scProportionTest can also be found in Supplementary Figure A2. Responders demonstrate a significantly higher proportion of central memory T cells (TCM) at 92.4%, compared to 83.0% in non-responders. This indicates that responders may maintain a more robust central memory pool, potentially contributing to a more sustained immune response. On the other hand, both TEM and TEFF are slightly higher in non-responders, but their overall proportions are very low, indicating a shift away from effector function in both groups. TCM usually demonstrate superior persistence and antitumor immunity compared to TEM, with greater proliferative capacity and longevity, enabling them to maintain a sustained immune response and provide long-term protection against tumors.\u003c/p\u003e \u003cp\u003eAs described above, non-responders were found to have a significantly higher proportion of TExTerm cells (10.0%) compared to responders (3.7%). This is consistent with the idea that terminally exhausted cells are less functional and may contribute to non-responsiveness. Furthermore, non-responders have a higher proportion of TExProg2 cells (4.1%) than responders (2.0%), which could indicate that cells further down the exhaustion pathway are more prevalent in non-responders, possibly limiting the potential for reactivation.\u003c/p\u003e \u003cp\u003eNext, let us conduct the same analysis for pathological data. As one can see in Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA and \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB, both the volumetric and pathological data show a trend, where higher helper T cell proportions are associated with better responses (\u0026gt;\u0026thinsp;50% pathological response and volumetric response). Across both datasets, non-responders (\u0026lt;\u0026thinsp;10% pathological response and volumetric non-response) have higher proportions of cytotoxic T cells. This consistency suggests that the mere presence of cytotoxic cells is not sufficient for a positive response\u0026mdash;functionality and exhaustion state might be more critical.\u003c/p\u003e \u003cp\u003eTregs are elevated in better responders in both datasets, though the increase is subtle. This might indicate that a controlled regulatory environment is needed for an effective response, preventing excessive inflammation or autoimmune-like damage. Increasing evidence suggests that regulatory populations are highly heterogeneous and subpopulations of Tregs can differentially regulate immune responses, with some subsets having less suppressive capacity or even promoting anti-tumor immunity under certain conditions. This variability may also help explain the differences between responders and non-responders to immune checkpoint treatments, as the balance of these regulatory subsets can significantly influence the overall effectiveness of the immune response against tumors.\u003c/p\u003e \u003cp\u003eFinally, the central memory (TCM) subset consistently dominates across all groups in both datasets, providing a reservoir for sustained responses over time. On one hand, TCM cells have higher proliferative potential, allowing them to renew and expand more effectively to sustain their numbers; on the other, they tend to have more robust mitochondrial function and maintain a more stable epigenetic landscape than their effector counterparts\u003c/p\u003e \u003cp\u003eInterestingly, the pathological response data shows a more consistent increase in helper T cells and Tregs with better outcomes, whereas the volumetric response data had more variability in these subsets between responders and non-responders. This might suggest that helper and Treg balance could be more predictive of pathological rather than volumetric response. Furthermore, the proportion of TExProg2 cells slightly increases in the higher pathological response group (\u0026gt;\u0026thinsp;50%), which contrasts with a decrease in this subset in volumetric responders. This discrepancy could point to different roles for TExProg2 in pathological versus volumetric response contexts, or it could reflect a sampling or timing difference.\u003c/p\u003e\n\u003ch3\u003eDynamic Fluctuations in Interferon Signalling Observed on Path to Terminal Exhaustion\u003c/h3\u003e\n\u003cp\u003eNext, we wanted to evaluate the relationship between immune activation and T cell states for responders versus non-responders. In this part of the analysis, we evaluated the change in IFN signature scores to assess how they change as T cells go through the progression TN \u0026rarr; TCM \u0026rarr; TEM \u0026rarr; TEFF \u0026rarr; TExProg1 \u0026rarr; TExProg2 \u0026rarr; TExInt \u0026rarr; TExTerm.\u003c/p\u003e \u003cp\u003eAs can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA, the median IFNg signature score is the lowest in TN, increasing in TCM and reaching the highest level in TEM. In the progression through multiple states of exhaustion, the IFNg score that starts at a relatively high level in TEFF cells decreases as the cells transition to the first stage of exhaustion, TExProg1. It then increases as the cells transition to the TExProg2 state, decreases again in TExInt and finally once again reachest the highest value in TExTerm cells. This is consistent with the current understanding that terminally exhausted cells are short lived but can be very cytotoxic, as well as the observation that T cells in the intermediate stages of exhaustion have very low cytotoxicity. Specifically, in (Miller et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), the authors note that \u0026ldquo;progenitor exhausted cells killed target cells no more efficiently than did naive CD8\u0026thinsp;+\u0026thinsp;T cells, indicating that they contribute little or no direct cytotoxicity in the TME\u0026rdquo;. While the median IFNg signature scores for TExProg1 cells are not as low as for TN, they are second lowest out of all T cell states.\u003c/p\u003e \u003cp\u003eNext, we analyzed the differences in IFNg signature scores between responders and non-responders. We subdivided the scores into low, medium, and high IFN activity, calculated as one standard deviation from the median (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB). We then assigned these classifications to the T cells in responders versus non-responders in both volumetric (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC,D) and pathological (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE,F,G) response data.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(C) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell identity in volumetric response non-responders. (D) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in volumetric response responders. (E) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with \u0026lt;\u0026thinsp;10%. (F) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with 10\u0026ndash;49% response. (G) Bar plots showing the proportion of IFNg scores categorized as low, medium, or high for each T cell state in pathological response with \u0026gt;\u0026thinsp;50% response.\u003c/p\u003e \u003cp\u003eIn the volumetric data set, several differences between responders and non-responders can be highlighted with respect to differences in IFNg signature scores. In non-responders, TEFF cells are predominantly low in IFNg (blue), whereas responders show a higher proportion of medium and high IFNg signatures (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC). Furthermore, non-responders have more TEFF cells with low IFNg signatures compared to responders, who show a shift towards medium and high IFNg signatures. Both groups show medium IFNg signatures as dominant in TExProg1 cells, but responders maintain a slight edge with more medium IFNg cells, potentially indicating better reactivation potential. Non-responders have 50% more TExProg1 cells with low IFNg signatures, whereas responders have a modest increase in medium IFNg signatures.\u003c/p\u003e \u003cp\u003eIn TExProg2 cells, in both groups, high IFNg signatures dominate, but responders also show an increase in low IFNg signatures. Responders have dramatically fewer low IFNg TExProg2 cells, suggesting a pathway away from deep exhaustion. High IFNg is less prevalent in responders compared to non-responders.\u003c/p\u003e \u003cp\u003eIn TExInt cells, high IFNg is dominant in both groups, but with a larger proportion in responders. Responders show a significant increase in high IFNg signatures compared to non-responders, suggesting active engagement, while non-responders have a higher proportion of cells with low IFNg.\u003c/p\u003e \u003cp\u003eFinally, responders have 25% more of the high IFNg TExTerm cells compared to non-responders, who have slightly more of the low IFNg cells. Interestingly, high IFNg is prevalent in both groups, but responders manage to sustain a higher proportion of medium IFNg signatures. This suggests that high IFNg in TExTerm cells might not be universally detrimental; in responders, these cells may still play a role in controlling the tumor, while in non-responders, low IFNg suggests deeper dysfunction.\u003c/p\u003e \u003cp\u003eThe patterns observed in the pathological data (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE,F,G) align with those seen in the volumetric response data, particularly in the balance between high and medium IFNg signatures across different T cell states. Both data sets suggest that extreme IFNg signaling (either too high or too low) can be detrimental, and that a balanced IFNg profile is associated with better therapeutic outcomes.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eContrasting Response Patterns Between Peripheral and Tumor Compartments\u003c/h2\u003e \u003cp\u003eNext, we want to evaluate whether the patterns that were observed in TIL data with respect to differentiation between responders and non-responders might also be observed in the peripheral compartment.\u003c/p\u003e \u003cp\u003eThe initial PBMC dataset comprised 68 183 cells across 3 timepoints for 27 patients. T cells were isolated from the rest of the population, and further refined using an ensemble approach, as done previously. Clusters of cells were generated using the shared nearest neighbour (SNN) algorithm and different populations of cells were identified based on differentially expressed (DE) genes. Due to the overlapping gene profiles between natural killer (NK) and T cells, these two cell types often cluster together. As such, the cell clusters likely to be either NK or T cells were labelled as T/NK, with the intention of separating them downstream. After isolating the T/NK clusters, **38 539** cells remained in the analysis. Three distinct cell type identification methods were subsequently employed. These methods included SingleR, which uses the non-parametric Spearman rank correlation to score each cell in the dataset against each cell type in a reference (used: X-OMICS HNSCC reference); Azimuth, which uses a weighted-nearest neighbour (WNN) model-based approach to map cells to a reference (used: internal PBMC reference); and a random forest classifier by Bogac Aybey. It was believed that each of these algorithms contribute unique strengths to the identification process, collectively enhancing the accuracy of T cell identification. A majority voting approach was used to isolate T cells, selecting those with a score of 4 for a total of 27 922 T cells. This score indicates agreement across clustering results and three distinct tools, providing a high confidence identification of T cells.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs one can see in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, the distribution of T cell identities in PBMCs is quite different from TILs (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Most helper cells are clustered in TN, although in contrast to TILs, where TNs are composed of both helper and cytotoxic cells, in PBMCs most cytotoxic cells are in the TEFF and TExInt classes. In contrast, in the TME, most of the cytotoxic cells were in the terminally exhausted state.\u003c/p\u003e \u003cp\u003eNext, we assessed whether there exist differences in T cell states and composition in PBMCs over time; data were collected at three time points, B1, B2 and B3. In Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, we assess the change in proportions of different cell states and types (similarly to the analysis done for TILs in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). In Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e, we assess the differences between responders and non-responders as reflected in PBMCs, similarly to the analysis done for TILs in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn non-responders, we observe 26.7% helper cells, 16.2% Tregs, 48.1% cytotoxic cells in the TME, while in PBMCs, we see higher helper cells (67.4\u0026ndash;75%), very low Tregs (0.7\u0026ndash;0.9%), and lower cytotoxic cells (21\u0026ndash;27.5%). In responders, we observe 35.4% helper cells, 22% Tregs, 33.6% cytotoxic cells in the TME, while in PBMCs, helper cells dominate (71\u0026ndash;76.2%), with low Tregs (0.4\u0026ndash;0.9%) and lower cytotoxic cells (19.5\u0026ndash;23%). Together, these results show that in TILs, non-responders have a much higher proportion of cytotoxic T cells and Tregs compared to PBMCs. The helper T cells are dominant in PBMCs, whereas cytotoxic T cells are more prevalent in TILs, indicating that in non-responders, the cytotoxic response may be more localized to the tumor environment. Furthermore, responders also show a much higher proportion of cytotoxic T cells and Tregs in TILs compared to PBMCs.\u003c/p\u003e \u003cp\u003eWith respect to T cell exhaustion states, in non-responders, we observe low TN (1.5%), high TCM (83%), moderate TExTerm (10%) in TILs; in PBMCs, we see higher TN (46.7\u0026ndash;58.1%), moderate TCM (30.7\u0026ndash;39.9%), and very low TExTerm (0.1%). That is, TILs of non-responders show a strong central memory (TCM) phenotype and higher terminal exhaustion (TExTerm), whereas in PBMCs, the profile is more na\u0026iuml;ve (TN) and lacks significant terminal exhaustion. This suggests the exhaustion observed in non-responders is mainly restricted to the tumor environment. In responders, we see low TN (1.0%), very high TCM (92.4%), low TExTerm (3.7%) in TILs, while in the PBMCs we observe higher TN levels (48.7\u0026ndash;57.6%), moderate TCM (32.6\u0026ndash;38.3%), and very low TExTerm (0.1%). That is, responders have a higher proportion of central memory T cells (TCM) in TILs, with low terminal exhaustion. In PBMCs, there is a mix of TN and TCM with minimal exhaustion. This indicates that effective immunotherapy might maintain central memory cells, while reducing terminally exhausted cells primarily in the tumor microenvironment.\u003c/p\u003e \u003cp\u003eTogether, these results show that the key patterns observed in TILs, such as high central memory (TCM), the presence of terminally exhausted T cells (TExTerm), and an increased proportion of cytotoxic T cells and Tregs, are not mirrored in PBMCs. Instead, PBMCs are characterized by a higher proportion of na\u0026iuml;ve and central memory cells with very little terminal exhaustion. Thus, while T cell exhaustion and memory are critical for the tumor microenvironment, these patterns are not reflected in circulating PBMCs, suggesting that exhaustion-related changes are more localized to the tumor. Therefore, no significant patterns that differentiate responders from non-responders are observed in PBMCs, implying that T cell exhaustion and differentiation dynamics relevant to immunotherapy response are more specific to the tumor site.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eEmerging understanding of the existence of multiple states of T cell exhaustion, only some of which are targetable by checkpoint inhibitors, has created a framework to improve our understanding of the differences between patients who respond to therapy and those who do not. We hypothesized that patients who do not respond to therapy would have a higher proportion of non-targetable terminally exhausted T cells compared to responders.\u003c/p\u003e \u003cp\u003eIn this work we used single-cell RNA sequencing data of samples of 27 HNSCC patients who received neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4. We collated gene signatures for T cells based on lineage and progression through multiple states: TN \u0026rarr; TCM \u0026rarr; TEM \u0026rarr; TEFF \u0026rarr; TExProg1 \u0026rarr; TExProg2 \u0026rarr; TExInt \u0026rarr; TExTerm. We assessed both the distribution of cell types and states in pre- vs post-treatment samples for the entire population, as well as stratified responders and non-responders in post-treatment sample analysis. Additionally, we evaluated the change in IFNg signature scores to assess whether the level of IFN signaling in T cells may contribute to differences between responders and non-responders.\u003c/p\u003e \u003cp\u003eIn our analysis, we found that there indeed exist differences between responders and non-responders with respect to distribution of different T cell types and states. We found that there exist differences between pre- and post-treatment T cell distributions, with the overall decrease in the effector cell types, and increase in cell states associated with the different states of exhaustion. These results suggest that changes in T cell composition may at least partially be attributed to treatment rather than being an intrinsic patient characteristic. This hypothesis remains to be more rigorously evaluated on a larger dataset than was available for this analysis and should include more extensive pre- and post-treatment data for both responders and non-responders.\u003c/p\u003e \u003cp\u003eNext, we analyzed T cell identifies in pathological and volumetric responses and found that while there appear to be more cytotoxic T cells in non-responder samples, these cells tend to be primarily terminally exhausted. Indeed, looking more closely at the cell states revealed that the proportion of terminally exhausted cells in non-responders is significantly greater than in patients who responded to therapy. This suggests that while it is typically expected that a higher proportion of CD8 cells in the TME would be associated with a better response, a more informative approach may involve assessing the functional state of these cells rather than just their lineage. Furthermore, responders had higher proportions of both helper and regulatory T cells compared to non-responders. This suggested that responders may have a more balanced immune response compared to non-responders, and that even though Tregs are typically associated with immune regulation, their role here might involve maintaining immune homeostasis post-treatment (Li et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSimilarly, the pathological response results (\u0026lt;\u0026thinsp;10% vs. \u0026gt;50%) observed in comparison to volumetric analysis of non-responders versus responders showed similar relationships in T cell states, but with slight discrepancies that may be attributed to variations in the underlying methodologies. The analysis of volumetric response, which captures an overall change in tumor volume along two axes, while largely consistent with pathological responses, which assess changes on a cellular level by histological analysis, was nevertheless not identical. This discrepancy may be due to several factors. First, cell viability is commonly a measure of cellular activity and \u003cem\u003ehealth\u003c/em\u003e, which is independent of, though related to, proliferation, so we might be observing response to treatment of cells that remain viable (i.e., alive or even secreting growth-promoting factors for example, but not necessarily dividing and contributing to measurable growth at the time of assessment). It is also possible that the differences could be attributed to temporal discrepancy, i.e., the volumetric response assesses changes in tumor size/volume using imaging techniques, while pathological response measurement involves examining tissue samples. As a result, there could be a time lag between the measurement of these volumetric changes and observable pathological alterations. It is also possible that a change in size may occur not due to an increase in the number of cancer cells themselves but due to fibrotic responses, a change that would not necessarily indicate a change in pathological status. Moreover, tumors can be highly heterogeneous, so it is also feasible that a pathological examination could reveal areas of non-uniform response, i.e., containing viable and active cells that were either more resistant to treatment or found themselves in regions that received a suboptimal dose of treatment. Finally, the way the data is binned - into two categories for volumetric versus three for pathological - affects the granularity and distribution of the data, which may contribute to the observed discrepancies.\u003c/p\u003e \u003cp\u003eNext, we evaluated the relationship between immune activation and T cell states for responders vs non-responders by measuring IFNg signaling scores in different T cell subsets. The results suggested that successful immune responses in responders are associated with a balanced IFNg signaling profile across various T cell states. While high IFNg is present in later stages of exhaustion (TExInt, TExTerm), responders maintain lower IFNg levels in critical progenitor states (TExProg1, TExProg2). This balance may prevent cells from fully committing to terminal exhaustion, preserving their functionality. Non-responders, on the other hand, showed a skewed profile, where either low or excessively high IFNg signatures dominated across T cell states. This could reflect a scenario where T cells are either too exhausted to contribute effectively or are driven into deeper dysfunction by excessive IFNg signaling, particularly in the TEFF and TExInt states.\u003c/p\u003e \u003cp\u003eFinally, this analysis was repeated for the PBMC samples to evaluate whether such differences in T cell composition between responders and non-responders could be detected in peripheral samples. Unfortunately, the results suggest that PBMC data is insufficient to discern differences between responders and non-responders with respect to different states of T cell exhaustion, and TILs need to be analyzed.\u003c/p\u003e \u003cp\u003eOur analysis suggests that a highly inflammatory environment may trigger transitions between different stages of exhaustion, consistent with the hypothesis that this mechanism evolved in chronic viral infections for protection from autoimmunity (Kareva \u0026amp; Brown \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). An alternative hypothesis is that it is chronic antigen stimulation that may trigger these transitions (Wherry \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; McLane et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The two hypotheses\u0026mdash;inflammation-driven vs. chronic antigen stimulation-driven exhaustion\u0026mdash;are both plausible explanations for triggers that may push T cells down the exhaustion pathway. Based on the data we\u0026rsquo;ve analyzed, there are arguments for both mechanisms, and it\u0026rsquo;s possible that a combination of both factors is at play.\u003c/p\u003e \u003cp\u003eOur data show that high IFNg signaling is a consistent feature across various T cell exhaustion states, especially in non-responders. The presence of high IFNg in TEM, TExInt, and TExTerm states suggests that sustained inflammatory signaling may contribute to deeper exhaustion. The correlation between higher IFNg in certain subpopulations of T cells in non-responders and poorer outcomes supports the idea that excessive inflammation might exacerbate exhaustion, driving T cells into a less functional state.\u003c/p\u003e \u003cp\u003eThe persistence of exhaustion markers in TExProg2 cells, which are early progenitor-like exhausted states, along with their proliferative nature, suggests that these cells could be continuously being stimulated by antigens. If these cells are exposed to chronic antigen stimulation without adequate rest or clearance, they may be driven further down the exhaustion pathway regardless of the inflammatory context. The data showing that lower IFNg signaling in this subpopulation cells correlates with better outcomes could imply that reducing antigen load might help preserve these cells\u0026rsquo; functionality.\u003c/p\u003e \u003cp\u003eIt is possible and even likely, however, that both mechanisms are at play, with both inflammation and chronic antigen stimulation contributing to T cell exhaustion. High IFNg signaling likely exacerbates the exhaustion process, especially in cells that are already under chronic antigen stimulation. Conversely, chronic antigen exposure could be the initial driver of exhaustion, with inflammation, while initially beneficial, acting as a secondary factor that deepens the exhaustion state at higher levels.\u003c/p\u003e \u003cp\u003eNotably, one limitation of the scRNAseq data is its sparsity, meaning it often captures only a subset of the transcriptome, which can lead to incomplete or biased representations of cellular states. Additionally, scRNA-seq provides only a snapshot in time, making it challenging to determine whether observed changes are directly caused by therapy, inherent differences in response, or other factors. This temporal limitation complicates the interpretation of dynamic processes and the establishment of causal relationships in the context of T cell subpopulations and states. Ideally, it would be highly informative and valuable to have multiple samples to track changes in T cell subpopulations and state proportions over time, which unfortunately might be unfeasible for financial and logistical reasons. In addition, increasing evidence suggests that T cell exhaustion results in extensive epigenetic remodeling (Belk et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Scott-Browne et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Gennert et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Ford et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Epigenomics data would enable us to analyze changes in global patterns of chromatin accessibility, providing important insights into the underlying mechanisms of T cell exhaustion in the context of immune checkpoint inhibition.\u003c/p\u003e \u003cp\u003eThe theoretical implications of the existence of multiple states of exhaustion were extensively analyzed in (Kareva \u0026amp; Gevertz \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The authors developed a mathematical model of tumor-immune interactions, where cytotoxic immune cells can exist in three classes: effector, reversibly exhausted and terminally exhausted; transitions between states were hypothesized to be driven by the systemic inflammation level but a case of antigen-driven transitions was evaluated as well. The differences in immune cell composition with respect to states of exhaustion were observed in this analysis, with a prediction of a higher proportion of terminally exhausted T cells in non-responders. The importance of a balanced \u0026ldquo;goldilocks\u0026rdquo; inflammatory environment in maintaining effective response was also revealed in simulations of multiple dosing scenarios.\u003c/p\u003e \u003cp\u003eThe authors then identified three main qualitative strategies that can be effective in preserving the effectiveness of checkpoint inhibition therapy as monotherapy, including high dose-low frequency approach (MTD-like therapy); low dose high frequency approach (metronomic-like therapy), and an intermediate strategy. Virtual population analysis, where patients were characterized by an individual inflammatory threshold that triggers transition from reversible to terminal exhaustion, was conducted to identify if either of these therapeutic strategies would be likely to work better on a population level. The authors showed that for a heterogeneous population, while a standard MTD-like protocol of administration of checkpoint inhibitors as monotherapy may work, the metronomic-like (low dose high frequency) strategy is more likely to work for a larger fraction of the population.\u003c/p\u003e \u003cp\u003eSome recent retrospective analyses support the prediction that lower dosing can be as beneficial as the currently set dosing levels. In Chang et al. (Chang et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), the authors found that both median overall survival (OS) and rate of all classes of immune-related adverse events (irAEs) were similar in both the standard-dose and low-dose pembrolizumab in a cohort of 147 patients with non-small cell lung cancer (NSCLC). In another retrospective analysis, Low et al. (Low et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) showed that reducing the dose of pembrolizumab from a 200 to 100 mg flat dose resulted in no significant difference in response rate or grade 3\u0026ndash;4 (severe or life-threatening) irAEs. Similarly for nivolumab, in (Zhao et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), the authors showed that the overall response rate (ORR) in renal cell carcinoma patients was similar in both high and low dose cohorts, and among the patients in the low dose cohort, one patient had complete response (CR); no patients had CR in the high-dose cohort. These studies have numerous limitations, of course, like the small sample size and the nature of conducting a retrospective rather than prospective analysis, but they do suggest that current dosing strategies in ICI administration as monotherapy may be suboptimal.\u003c/p\u003e \u003cp\u003eMost recently, interim analysis was presented for DEDICATION-1 trial (NCT04909684), an open label randomized non-inferiority study that assessed the impact of reduced pembrolizumab dose (of approximately 75% depending on treatment schedule) vs standard of care in 750 NSCLC patients (Buma et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Interim analysis for 256 patients (Heuvel et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) revealed minimal differences in the outcomes, suggesting in the first prospective study that lower doses can be non-inferior.\u003c/p\u003e \u003cp\u003eWhile lower doses may be feasible, lower frequency administration is unlikely to take hold, including for logistical reasons. Well-selected combination therapy approaches may therefore present an attractive alternative to mitigate emergence of terminal exhaustion. In fact, it\u0026rsquo;s feasible that over-stimulating immune cells and thereby pushing them further down the irreversible exhaustion pathway could theoretically lead to tumor hyperprogression, which, for instance, has been observed in approximately 10% of patients with advanced gastric cancer, treated with anti-PD-1 therapy (Kamada et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). If the unbalanced inflammation that occurs as a result of initially successful tumor elimination accelerates T cell transition towards terminally exhausted state, it could effectively diminish the overall efficacy of the anti-tumor immune response compared to pre-treatment levels. This in turn could result in faster tumor growth due to a paradoxically incapacitated immune response. This hypothesis remains to be evaluated on a mechanistic level.\u003c/p\u003e \u003cp\u003eA good combination partner therefore might work to maintain the inflammatory level in the TME at a balanced state, without letting the tumor become \u0026ldquo;too cold\u0026rdquo; or \u0026ldquo;too hot\u0026rdquo;, both of which are expected to lead to insufficient effector function and consequently worse outcomes. In (Kareva \u0026amp; Gevertz \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), the authors conducted sensitivity analysis to predict, which mechanisms could be targeted as combination partners to \u0026ldquo;flip\u0026rdquo; the predicted response from a non-responder to a responder. While most combination therapy approaches tend to focus on augmenting different cytotoxic pathways (Schmidt et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Palmer et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Meric-Bernstam et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), within the framework of the existence of multiple states of exhaustion, it was predicted that it may be more beneficial to also mitigate inflammatory responses. Excitingly, two recent studies, one in lung cancer and another in Hodgkin lymphoma, have independently applied this approach by administering Janus kinase (JAK) inhibitors, a class of drugs that help reduce inflammation (Shawky et al. \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), as combination partners with anti-PD-1 therapy. In (Mathew et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), the authors showed that administration of JAK1 inhibitor itacitinib after pembrolizumab resulted in improved therapeutic response both in mouse models and in a Phase 2 non-small cell lung cancer trial. In (Zak et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), in a report of a phase I clinical trial the authors showed that a combination of the JAK inhibitor ruxolitinib with nivolumab, administered to patients with relapsed or refractory Hodgkin lymphoma, yielded an overall response of 53% (10/19 patients), with 6/19 patients achieving a complete response. These results highlight not only the viability of the initially counterintuitive approach of using properly selected immunosuppressive medications to augment ICIs, but also the ability to rescue a response in a patient who previously failed ICIs, converting a non-responder to a responder. Notably, the goal of co-administering anti-inflammatory therapy is not to mitigate the adverse events (an extreme example of this approach was the groundbreaking case of the first pediatric CAR-T patient Emily Whitehead, who received anti-IL-6 drug tocilizumab to mitigate the severity of her cytokine release syndrome following her treatment (Newitt \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)). Instead, administration of an anti-inflammatory drug in combination with ICI would serve mechanistically to keep T cells in a more balanced state and prevent transition to terminal exhaustion.\u003c/p\u003e \u003cp\u003eThe presented analysis has several limitations that need to be addressed in future work. Data for only 27 patients was available, with only 19 post-treatment biopsy samples, and only 6 samples to compare T cell states before and after treatment. It would be particularly informative to expand this analysis to a data set, where a more detailed understanding of T cell state composition can be made in pre- vs post-treatment samples for both responders and non-responders. Obtaining longitudinal TIL samples to gain a deeper insight into the progression of treatment response would additionally improve our understanding of the impact of treatment on T cell states. Further investigations and better understanding of the multiple steps of T cell exhaustion and the triggers for state transitions will help devise better monotherapy and combination treatments to increase the number of responders to checkpoint inhibition therapy, and to even potentially rescue those who did not initially respond.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cem\u003eDataset\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eA single-cell RNA sequencing (scRNA-seq) dataset (GSE200996) published by Luoma et al. (Luoma et al. 2022) was analyzed to investigate the relationship between T cell subsets and response to immune checkpoint inhibitors (ICIs). This dataset consists of samples collected during a phase 2 clinical trial of 27 head and neck squamous cell carcinoma (HNSCC) patients treated with neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4. Both oral tissue samples and peripheral blood mononuclear cells (PBMCs) were analyzed. Two clinical readouts are provided: pathological response (\u0026lt;10% | 10-49% | \u0026gt;50%) and volumetric response (Response | No Response).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eStatistical Analysis\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAll analysis presented therein was conducted in R (v4.3.1).\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eData Integration and Normalization\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003eTo correct for batch effects and technical variations that may be obscuring biological signal, datasets were harmonized using the standard Seurat (v5.0.1) integration workflow. Individual gene expression matrices were normalized using regularized negative binomial regression, as described in Hafemeister and Satija (Hafemeister \u0026amp; Satija 2019). Features were ranked by residual variance, with the top 3000 selected for each dataset. These selected features were used to identify integration features across datasets, facilitating the discovery of cross-dataset pairs of cells predicted to share a common biological state, termed anchors, through an iterative integration process detailed in (Stuart et al. 2019).\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eCell-Type Identification\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003eTo distinguish between different cell types in the absence of a gold standard, we adopted an ensemble approach, combining both supervised and unsupervised learning methods. This approach was favorable as it leveraged the strengths of various techniques to achieve a more robust and reliable classification of cell types.\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003e\u003cem\u003eDimensionality Reduction, Clustering and Differential Expression Analysis\u003c/em\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eWe performed Principal Component Analysis (PCA) on the integrated dataset to capture the most significant sources of variation across cells. Utilizing the first 30 principal components from PCA, we then employed Uniform Manifold Approximation and Projection (UMAP) to visualize the data in a lower-dimensional space (Figure A1). Shared nearest neighbor (SNN) graph-based clustering was conducted on the PCA space using a resolution parameter of 0.5 to identify clusters of cells with similar molecular profiles. Highly variable features within each identified cell cluster were determined using the Wilcoxon Rank Sum test from Seurat's FindAllMarkers function. By integrating biological knowledge and considering the tissue context, we assigned specific cellular populations based on the calculated highly variable genes (HVGs).\u003c/p\u003e\n\u003col start=\"2\"\u003e\n \u003cli\u003e\u003cem\u003eReference-Based Approaches\u003c/em\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eWe used three reference-based approaches that leverage existing annotated gene expression profiles. These included SingleR (Aran et al. 2019), Azimuth (Hao et al. 2021), and a random forest-based cell typing approach (Aybey et al. 2023). For further details on each method, please refer to the original publications.\u003c/p\u003e\n\u003cp\u003eThis ensemble approach allowed us to combine the strengths of both supervised and unsupervised methods, enhancing the confidence in labeling cell types in our dataset.\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eT Cell Isolation\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003eA majority voting approach was used to isolate T cells, selecting those with a score of 4. This score indicates agreement across all four methods described above, collectively enhancing the accuracy of T cell identification. A nuanced scoring method was subsequently employed for assigning T cells to distinct subgroups, as computational methods are less proficient at distinguishing closely related cells and discerning subtle differences. Treg label score was calculated as follows: +1 for Treg prediction, +0.5 for a CD4 subtype prediction. Non-lineage specific labels such as Tn, Tprof, IFN-resp and HSPhi added a score of 1 for all groups. Unconventional T cells (NKT, MAIT, gdT) were classified as Cytotoxic along with CD8 lineage cells. Cells were classified into the following major subsets based on their final score: Cytotoxic, Helper and Treg. Cells that lacked clear assignment were marked as Inconclusive.\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eGene Signatures\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003eDistinct gene signatures were collated to form a comprehensive collection, incorporating four exhausted T cell subsets from Beltra et al. (Beltra et al. 2020), five T cell states (TN, TSCM, TEM, TCM, TEFF) from Chen et al. (Chen et al. 2021), and the IFN-γ signature from Gao et al. (Gao et al. 2016). TN and TSCM share a similar transcriptomic profile making them indistinguishable at that level, thus the two states were considered as one.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eGene Set Enrichment and Differential Composition Analysis\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003eTo assess differences in cell composition with respect to subtype and state across treatment response, we measured the proportion of each subtype within each group by calculating percentages based on total cell counts. We further assessed relative proportions and significance using scProportionTest (Miller et al. 2021). Gene signature scores were measured using UCell by Andreattaa et al. (Andreatta \u0026amp; Carmona 2021), based on the Mann-Whitney U Test, which calculates scores based on gene expression ranks within individual cells. Additionally, Sargent from Nouri et al. (Nouri et al. 2023) was used to annotate individual cells based on gene signature scores where applicable. Sargent sorts non-zero expressed genes in descending order for each cell before transforming an input gene-by-cell expression matrix into a corresponding gene-set-by-cell assignment score matrix. A gini-index is then computed among assignment scores per cell, transforming the gene-set-by-cell assignment score matrix to a distribution of indexes. Interferon scores were stratified into three categories (low, medium, and high) according to their distribution and position relative to thresholds determined by the median-absolute-deviation (MAD), set at one deviation away from the median.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eDefining Lineage-Agnostic Transcriptomic Signatures of T Cell Functional States\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTo comprehensively explore the T cell space, it was imperative to first identify the specific signatures not just representing the diverse lineages T cells can differentiate into, but also the functional states they can undertake. T cells predominantly exist in a naive (TN) state until they encounter their cognate antigen, wherein they undergo activation, proliferate, and differentiate into various subsets, including effector (TEFF) and memory cells. The memory cell subsets include stem cell memory T cells (TSCM), central memory T cells (TCM), effector memory T cells (TEM), and tissue-resident memory T cells (TRM). Effector T cells, though transient, execute the immediate T cell response, whilst a subset will persist as memory cells, primed to mount rapid secondary immune responses upon re-exposure to the antigen. The suppressive tumour microenvironment also gives rise to dysfunctional states, including exhaustion, senescence, and anergy, which are characterised by, among other phenotypic features, defective effector functions, decreased proliferative capacity, altered metabolism, and reduced cytotoxicity. These dysfunctional states pose obstacles to effective cancer immunotherapy, such as immune checkpoint inhibitors (ICIs), which aim to prevent or reverse them by alleviating immunosuppressive pathways.\u003c/p\u003e\n\u003cp\u003eWithin exhaustive states, T cells have been found to undergo progressive stages of exhaustion, transitioning between substates that exhibit varying degrees of dysfunction. To investigate these exhausted states in the context of ICIs, we leveraged gene signatures from a developmental framework for exhausted T cells by Beltra et al. (2020) that classifies progression from exhausted progenitor (TExProg 1 and 2) to intermediate (TExInt) and terminal (TExTerm) subsets. Whilst TExTerm cells do not respond to PD-1 blockade, the authors found that PD-1 blockade partially re-balances subset distribution by fostering the formation of TExProg2 and TExInt cells. Moreover, their study highlights the therapeutic potential of sustaining the TExInt population whilst preventing transition into TExTerm through modulation of TOX and T-Bet.\u003c/p\u003e\n\u003cp\u003eT cell states are characterized by distinct phenotypic markers, some of which are shared across states at varying levels, underscoring the complexity of T cell biology. Therein, transcriptomic signatures of T cell states were compiled through a comprehensive literature search and analysis of reference datasets. We hypothesized that we could generate gene signatures that are independent of T cell lineage by separating genes related to CD4 and CD8 lineages from those associated with T cell states. In addition, inflammatory signatures were obtained to investigate the relationship between inflammation levels and T cell exhaustion. Figure A1 summarizes the specific gene signatures associated with different T cell states, including substates of T cell exhaustion. The key characteristics of the different T cell states are summarized in Table 1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cbr\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eConflicts of Interest\u003c/h2\u003e\n\u003cp\u003eThe authors are employees of Merck KGaA, which operates as EMD Serono in the U.S. Views presented in this manuscript are the authors\u0026rsquo; own and do not necessarily represent the views of Merck KGaA/EMD Serono.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eIK and CP designed the research and wrote the manuscript; CP performed the analysis.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eCP started this work during her first rotation in Oncology Data Science and continued with a more comprehensive and targeted analysis during her second rotation with IK in the Quantitative Pharmacology Department at Merck KGaA from January to May 2024. She would like to thank Dr. Eike Staub and Dr. Michail Yekelchyk for valuable discussions and for directing her to the dataset used for the analysis, as well as Bogac Aybey for technical assistance in running his published cell-type predictor and Dr. Sven-Eric Schelhorn for computational optimization help on the cluster. Special thanks to Dr. Olga Bogatyrova for connecting CP and IK and highlighting their shared interests. Thank you to Dr. Katarzyna Wreczycka for granting me access to her GitHub repository, in which she compiled publicly available gene signatures of interferon signaling.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAndreatta, M. \u0026amp; Carmona, S. J. UCell: Robust and scalable single-cell gene signature scoring. \u003cem\u003eComput. Struct. Biotechnol. J.\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e, 3796\u0026ndash;3798 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBelk, J. A., Daniel, B. \u0026amp; Satpathy, A. T. Epigenetic regulation of T cell exhaustion. \u003cem\u003eNat. Immunol.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e (6), 848\u0026ndash;860 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeltra, J. C. et al. Developmental relationships of four exhausted CD8\u0026thinsp;+\u0026thinsp;T cell subsets reveals underlying transcriptional and epigenetic landscape control mechanisms. \u003cem\u003eImmunity\u003c/em\u003e \u003cb\u003e52\u003c/b\u003e (5), 825\u0026ndash;841 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlackburn, S. D. et al. Selective expansion of a subset of exhausted CD8 T cells by αPD-L1 blockade. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, 105(39), pp.15016\u0026ndash;15021. (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBuma, A. I. et al. Integrating treatment cost reduction strategies and biomarker research to reduce costs and personalize expensive treatments: an example of a self-funding trial in non-small cell lung cancer. Frontiers in Pharmacology, 14, p.1274532. (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarlson, R. D., Flickinger, J. C. Jr \u0026amp; Snook, A. E. Talkin\u0026rsquo;toxins: from Coley\u0026rsquo;s to modern cancer immunotherapy. Toxins, 12(4), p.241. (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChang, K. C. et al. Comparative Effectiveness and Safety of Standard-Dose and Low-Dose Pembrolizumab in Patients with Non-Small-Cell Lung Cancer: A Multi-Institutional Cohort Study in Taiwan. Cancers, 14(5), p.1157. (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, G. M. et al. Integrative bulk and single-cell profiling of premanufacture T-cell populations reveals factors mediating long-term persistence of CAR T-cell therapy. \u003cem\u003eCancer Discov.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (9), 2186\u0026ndash;2199 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eColey, W. B. The treatment of malignant tumors by repeated inoculations of erysipelas: With a report of ten original cases. 1. The American Journal of the Medical Sciences (1827\u0026ndash;1924), 105(6), p.487. (1893).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDobosz, P. \u0026amp; Dziecikatkowski, T. The intriguing history of cancer immunotherapy. \u003cem\u003eFront. Immunol.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 2965 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDutcher, J. P. et al. High dose interleukin-2 (Aldesleukin)-expert consensus on best management practices-2014. \u003cem\u003eJ. Immunother. Cancer\u003c/em\u003e. \u003cb\u003e2\u003c/b\u003e, 1\u0026ndash;23 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFord, B. R. et al. Tumor microenvironmental signals reshape chromatin landscapes to limit the functional potential of exhausted T cells. \u003cem\u003eSci. Immunol.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e (74), eabj9123 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao, J. et al. Loss of IFN-γ pathway genes in tumor cells as a mechanism of resistance to anti-CTLA-4 therapy. \u003cem\u003eCell\u003c/em\u003e \u003cb\u003e167\u003c/b\u003e (2), 397\u0026ndash;404 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGennert, D. G. et al. Dynamic chromatin regulatory landscape of human CAR T cell exhaustion. Proceedings of the National Academy of Sciences, 118(30), p.e2104758118. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHafemeister, C. \u0026amp; Satija, R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. \u003cem\u003eGenome biology\u003c/em\u003e, 20(1), p.296. (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan den Heuvel, M., van der Noort, V., Heine, T. \u0026amp; R 1258MO Low dose versus standard dose pembrolizumab for treatment of stage IV stage non-small cell lung carcinoma: Results of the pre-planned interim analysis of the NVALT-30 clinical trial. \u003cem\u003eAnn. Oncol.\u003c/em\u003e \u003cb\u003e35\u003c/b\u003e, S806 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, A. C. et al. T-cell invigoration to tumour burden ratio associated with anti-PD-1 response. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e545\u003c/b\u003e (7652), 60\u0026ndash;65 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIshida, Y. et al. Induced expression of PD-1, a novel member of the immunoglobulin gene superfamily, upon programmed cell death. \u003cem\u003eEMBO J.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (11), 3887\u0026ndash;3895 (1992).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKahan, S. M., Wherry, E. J. \u0026amp; Zajac, A. J. \u003cem\u003eT cell exhaustion during persistent viral infections\u003c/em\u003e479pp.180\u0026ndash;193 (Virology, 2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKamada, T. et al. PD-1\u0026thinsp;+\u0026thinsp;regulatory T cells amplified by PD-1 blockade promote hyperprogression of cancer. Proceedings of the National Academy of Sciences, 116(20), pp.9999\u0026ndash;10008. (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKareva, I. \u0026amp; Brown, J. S. Evolutionary and ecological perspective on the multiple states of T cell exhaustion. In Cancer through the Lens of Evolution and Ecology. CRC, 92\u0026ndash;104. (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKareva, I. \u0026amp; Gevertz, J. L. Mitigating non-genetic resistance to checkpoint inhibition based on multiple states of immune exhaustion. \u003cem\u003enpj Systems Biology and Applications\u003c/em\u003e, 10(1), p.14. (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeach, D. R., Krummel, M. F. \u0026amp; Allison, J. P. Enhancement of antitumor immunity by CTLA-4 blockade. \u003cem\u003eScience\u003c/em\u003e \u003cb\u003e271\u003c/b\u003e (5256), 1734\u0026ndash;1736 (1996).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, J. B., Kim, H. R. \u0026amp; Ha, S. J. Immune checkpoint inhibitors in 10 years: contribution of basic research and clinical application in cancer immunotherapy. \u003cem\u003eImmune Netw.\u003c/em\u003e, \u003cb\u003e22\u003c/b\u003e(1). (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, Y. et al. Potential anti-tumor effects of regulatory T cells in the tumor microenvironment: a review. \u003cem\u003eJ. Translational Med.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e (1), 293 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Q., Sun, Z. \u0026amp; Chen, L. Memory T cells: strategies for optimizing tumor immunotherapy. \u003cem\u003eProtein cell.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (8), 549\u0026ndash;564 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLow, J. L. et al. Low-dose pembrolizumab in the treatment of advanced non-small cell lung cancer. \u003cem\u003eInt. J. Cancer\u003c/em\u003e. \u003cb\u003e149\u003c/b\u003e (1), 169\u0026ndash;176 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuoma, A. M. et al. Tissue-resident memory and circulating T cells are early responders to pre-surgical cancer immunotherapy. \u003cem\u003eCell\u003c/em\u003e \u003cb\u003e185\u003c/b\u003e (16), 2918\u0026ndash;2935 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMathew, D. et al. Combined JAK inhibition and PD-1 immunotherapy for non\u0026ndash;small cell lung cancer patients. \u003cem\u003eScience\u003c/em\u003e \u003cb\u003e384\u003c/b\u003e (6702), eadf1329 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcKinney, E. F. et al. T-cell exhaustion, co-stimulation and clinical outcome in autoimmunity and infection. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e523\u003c/b\u003e (7562), 612\u0026ndash;616 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcLane, L. M., Abdel-Hakeem, M. S. \u0026amp; Wherry, E. J. CD8 T cell exhaustion during chronic viral infection and cancer. \u003cem\u003eAnnu. Rev. Immunol.\u003c/em\u003e \u003cb\u003e37\u003c/b\u003e, 457\u0026ndash;495 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeric-Bernstam, F. et al. Enhancing anti-tumour efficacy with immunotherapy combinations. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e397\u003c/b\u003e (10278), 1010\u0026ndash;1022 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiller, B. C. et al. Subsets of exhausted CD8\u0026thinsp;+\u0026thinsp;T cells differentially mediate tumor control and respond to checkpoint blockade. \u003cem\u003eNat. Immunol.\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e (3), 326\u0026ndash;336 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiller, S. A. et al. LSD1 and aberrant DNA methylation mediate persistence of enteroendocrine progenitors that support BRAF-mutant colorectal cancer. \u003cem\u003eCancer Res.\u003c/em\u003e \u003cb\u003e81\u003c/b\u003e (14), 3791\u0026ndash;3805 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNewitt, V. N. The incredible story of Emily Whitehead \u0026amp; CAR T-cell therapy. (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNouri, N. et al. A marker gene-based method for identifying the cell-type of origin from single-cell RNA sequencing data. \u003cem\u003eMethodsX\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 102196 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePalmer, A. C. et al. Predictable clinical benefits without evidence of synergy in trials of combination therapies with immune-checkpoint inhibitors. \u003cem\u003eClin. Cancer Res.\u003c/em\u003e \u003cb\u003e28\u003c/b\u003e (2), 368\u0026ndash;377 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRobert, C. A decade of immune-checkpoint inhibitors in cancer therapy. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (1), 3801 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRosenberg, S. A. et al. Treatment of 283 consecutive patients with metastatic melanoma or renal cell cancer using high-dose bolus interleukin 2. \u003cem\u003eJama\u003c/em\u003e \u003cb\u003e271\u003c/b\u003e (12), 907\u0026ndash;913 (1994).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchmidt, E. V. et al. Assessment of clinical activity of PD-1 checkpoint inhibitor combination therapies reported in clinical trials. \u003cem\u003eJAMA Netw. open.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e (2), e1920833\u0026ndash;e1920833 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchoenfeld, A. J. \u0026amp; Hellmann, M. D. Acquired resistance to immune checkpoint inhibitors. \u003cem\u003eCancer cell.\u003c/em\u003e \u003cb\u003e37\u003c/b\u003e (4), 443\u0026ndash;455 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScott-Browne, J. P. et al. Dynamic changes in chromatin accessibility occur in CD8\u0026thinsp;+\u0026thinsp;T cells responding to viral infection. \u003cem\u003eImmunity\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e (6), 1327\u0026ndash;1340 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShawky, A. M. et al. A comprehensive overview of globally approved JAK inhibitors. Pharmaceutics, 14(5), p.1001. (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStuart, T. et al. Comprehensive integration of single-cell data. \u003cem\u003ecell\u003c/em\u003e \u003cb\u003e177\u003c/b\u003e (7), 1888\u0026ndash;1902 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWherry, E. J. T cell exhaustion. \u003cem\u003eNat. Immunol.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e (6), 492\u0026ndash;499 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZak, J. et al. JAK inhibition enhances checkpoint blockade immunotherapy in patients with Hodgkin lymphoma. \u003cem\u003eScience\u003c/em\u003e \u003cb\u003e384\u003c/b\u003e (6702), eade8520 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, J. J. et al. Low-dose nivolumab in renal cell carcinoma: a real-world experience. \u003cem\u003eOncology\u003c/em\u003e \u003cb\u003e99\u003c/b\u003e (3), 192\u0026ndash;202 (2021).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 is available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"T Cell Exhaustion, Tumor Immunology, scRNA-Seq, Therapeutic Resistance, Immune Checkpoint Therapy, Cancer Immunotherapy","lastPublishedDoi":"10.21203/rs.3.rs-5663382/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5663382/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe emerging recognition of multiple states of T cell exhaustion, of which only some are targetable by checkpoint inhibitors, has provided new insights into the variability in patient responses to immunotherapy. We hypothesized that non-responders to therapy have a higher proportion of non-targetable, terminally exhausted T cells compared to responders. To investigate this, we analyzed single-cell RNA sequencing data from 27 patients with head and neck squamous cell carcinoma (HNSCC) treated with neoadjuvant anti-PD-1 or anti-PD-1/CTLA-4 therapy. We identified gene signatures for T cells across different states, ranging from na\u0026iuml;ve to terminally exhausted, and evaluated their distribution post-treatment. Non-responders exhibited a more inflammatory profile, while responders showed a more balanced immune profile with higher proportions of both helper and regulatory T cells, suggesting that a balanced inflammatory environment may be crucial for therapeutic success. Our analysis further revealed differences between responders and non-responders in the composition of predicted T cell states, particularly in the exhausted T cell subsets, with non-responders showing a higher proportion of terminally exhausted T cells. We therefore propose existence of tumors that may be \u0026ldquo;too hot\u0026rdquo;, with resulting loss of efficacy and emergence of therapeutic resistance through a pathway that is different from that of \u0026ldquo;cold\u0026rdquo; tumors. Despite limitations, including the small sample size and the lack of well-established transcriptomic signatures of exhaustion subsets, our findings offer a starting point to encourage further investigation into the relationship between inflammation, T cell exhaustion, and therapy efficacy towards improving patient outcomes.\u003c/p\u003e","manuscriptTitle":"Heterogeneity of Exhausted T Cell Subsets in Responders and Non-Responders Following Checkpoint Inhibition Therapy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-11 09:29:47","doi":"10.21203/rs.3.rs-5663382/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f8498bbe-e3a5-49af-ac98-47b65647d738","owner":[],"postedDate":"March 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":45126620,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":45126621,"name":"Biological sciences/Cancer/Tumour immunology"},{"id":45126622,"name":"Biological sciences/Cancer/Cancer therapy"},{"id":45126623,"name":"Biological sciences/Cancer/Cancer therapy/Cancer immunotherapy"},{"id":45126624,"name":"Biological sciences/Cancer/Cancer therapy/Cancer therapeutic resistance"}],"tags":[],"updatedAt":"2025-09-25T03:53:36+00:00","versionOfRecord":[],"versionCreatedAt":"2025-03-11 09:29:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5663382","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5663382","identity":"rs-5663382","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}