Graphical abstract graphic file with name fx1.jpg [81]Open in a new tab Highlights * • Circulating CD8^+ T[EM]s are more abundant in chemoimmunotherapy responders * • These CD8^+ T[EM]s show a progenitor-exhausted phenotype and stem-like properties * • Stem-like CD8^+ T[EM] expansion links to permissive pre-treatment tumor environment * • Combining tumor and blood profiles improves chemoimmunotherapy response prediction __________________________________________________________________ Chin et al. analyze tumor and serial blood immune profiles in mesothelioma patients receiving chemoimmunotherapy. They identify stem-like CD8^+ T effector memory cells associated with treatment response and demonstrate that combining peripheral blood and tumor data improves prediction of clinical outcome. Introduction Pleural mesothelioma is a rare malignancy linked to asbestos exposure. Mesothelioma is often advanced at diagnosis and carries a poor prognosis, with patients surviving approximately 12–15 months despite treatment, which for decades has been platinum-based chemotherapy.[82]^1 However, the treatment paradigm is changing. More recently, several pivotal clinical studies have established the role of immune checkpoint therapy (ICT) in first-line treatment, which includes single agent PD-(L)1 inhibitors, dual inhibition of CTLA-4 and PD-(L)1, and PD-(L)1 inhibitors in combination with chemotherapy.[83]^2^,[84]^3^,[85]^4^,[86]^5^,[87]^6 Despite these new therapies becoming available, treatment outcomes are variable. No clinically validated biomarker exists to predict response to immunotherapy-based treatments in mesothelioma, which makes treatment decisions complex and confronts patients with the prospect of uncertain efficacy against the potential for substantial immune-related side effects.[88]^7 Ideally, predictive biomarkers would be sampled from easily accessible biospecimens before or early during treatment, including from archival pre-treatment tumor tissue samples or sequential on-treatment samples from peripheral blood,[89]^8 to predict treatment response. The latter is clinically more accessible for measurement compared to tumor tissue. Many efforts are underway to identify such predictive biomarkers for response, but it is unclear whether a single biomarker will explain the variance in clinical outcomes.[90]^9 This is particularly relevant for platinum chemotherapy in combination with anti-PD-(L)1, as the interaction can at least statistically be explained as additive and not synergistic,[91]^10^,[92]^11 suggesting that biological correlates for response may not overlap between the chemo- and immunotherapy components of the combination.[93]^12 Here, we explore dynamic changes in peripheral blood in combination with the pre-treatment tumor state to correlate with response to chemoimmunotherapy in mesothelioma. Results Study design, cohort characteristics, and molecular assays In the phase 2, single-arm, multi-center, durvalumab with chemotherapy as first-line treatment in advanced pleural mesothelioma (DREAM) trial, we enrolled 54 participants with previously untreated, unresectable pleural mesothelioma.[94]^3 Participants received the anti-PD-L1 antibody durvalumab, given once every 3 weeks in combination with pemetrexed and cisplatin for up to 6 cycles, followed by a maintenance phase of durvalumab every 4 weeks for a maximum duration of 12 months ([95]Figure 1A).[96]^3 Participants were classified as responders (n = 31) or non-responders (n = 23), with responders defined as those achieving progression-free survival at 6 months after commencing treatment, which was the primary endpoint of the DREAM study ([97]Figures 1A–1C).[98]^3 To dissect out possible molecular signatures predictive of response in the DREAM cohort, we collected transcriptome data from both peripheral blood and tumor biopsies. This included bulk RNA sequencing (RNA-seq) on peripheral whole blood from 40 participants at three time points (pre-treatment and prior to the second and third cycles of chemoimmunotherapy) and single-cell RNA-seq and single-cell T cell receptor sequencing (TCR-seq) on peripheral blood mononuclear cells (PBMCs) at the same time points for 35 participants ([99]Figure 1B). Additionally, gene expression profiling was performed on archival formalin-fixed paraffin-embedded (FFPE) pre-treatment tumor biopsies from 46 participants using the NanoString nCounter platform with the PanCancer IO 360 panel, which measures the expression levels of 770 immune- and cancer-related genes. Figure 1. [100]Figure 1 [101]Open in a new tab Peripheral blood bulk transcriptomics reveals immune pathways that differentiate responders from non-responders to chemoimmunotherapy (A) DREAM chemoimmunotherapy protocol. (B) Participant characteristics and biospecimen availability for each participant, with the first three rows describing biospecimens available for each participant. (C) Participant response by histology. (D) Differentially expressed genes across pairwise comparisons, subdivided by log[2]FC direction. (E–G) (E) Enrichment scores, raw p values, and FDR values for gene set enrichment analysis (GSEA) using the GO:0043374 (CD8-positive alpha beta T cell differentiation) gene set for response contrasts with enrichment plots displayed for statistically significant contrasts (F, G). BOR, best overall response; R, responder; NR, non-responder; NES, normalized enrichment score; NOM p value, nominal p value. Peripheral blood bulk transcriptomics reveals immune pathways that differentiate responders from non-responders to chemoimmunotherapy Differential expression analysis of the peripheral blood bulk transcriptomics data showed that most differences (abs(log[2]FC) > 0.5, false discovery rate [FDR] ≤ 0.05) were detected between treatment initiation (time point 0) and week 3 (time point 1) ([102]Figures 1D, [103]S1A, and S1B). We were unable to detect any differentially expressed genes between responders and non-responders at individual time points, nor any differences in temporal gene expression changes when comparing the two groups. Principal-component analysis (PCA) indicated that the peripheral blood samples exhibited high heterogeneity ([104]Figure S1C). By week 6 after commencing therapy (time point 2), we did not uncover any differentially expressed genes between responder and non-responder samples ([105]Figures 1D and [106]S1B). When comparing responder and non-responder samples at each time point individually using gene set enrichment analysis, we discovered only a single term enriched in responder samples: CD8-positive, alpha beta T cell differentiation, which was upregulated in responders at time points 1 and 2 (FDR ≤ 0.1) ([107]Figures 1E–1G). No terms were enriched in non-responder samples. Together, our results demonstrate substantial heterogeneity in patients’ responses to chemoimmunotherapy but also point to a subtle yet consistent difference in CD8 T cell differentiation between responders and non-responders. Circulating activated CD8^+ T effector memory cells are more abundant in chemoimmunotherapy responders before and shortly after treatment Given that the peripheral blood bulk transcriptomics data revealed increased T cell activity and antigen presentation between time point 0 and time point 1 in responders, we sought to identify specific immune cell populations responsible for this signature. We performed droplet-based 5′-end single-cell transcriptomics on PBMCs at all 3 time points from 35 participants and obtained 390,373 cells expressing 21,603 genes after data preprocessing and quality control. We identified 31 cell subtypes by leveraging reference mapping from publicly available cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) data[108]^13 and further confirmed cell identity via expression of canonical markers ([109]Figures S2A–S2E). Dimensionality reduction ([110]Figures 2A–2C) and compositional analysis of immune cell populations grouped by time point and response ([111]Figure S2F) showed time- and response-dependent changes in cell proportions in multiple immune cell compartments. Figure 2. [112]Figure 2 [113]Open in a new tab Activated CD8^+ T[EM]s are more abundant in chemoimmunotherapy responders before and shortly after treatment (A–C) (A) Uniform manifold approximation and projection (UMAP) of all cells in single-cell transcriptomics dataset, grouped by (B) time point and (C) response. (D) Forest plot depicting compositional analysis using scCODA, repeated using two chain lengths for the No-U-Turn (NUTS) sampler, with positive log-fold changes indicating differential abundance with respect to responders. (E) Bar chart of Milo analysis showing the proportions of neighbourhoods across cell types, with statistically significant positive (LFC+) and negative (LFC−) log-fold changes, normalized against the total number of neighbourhoods, at an FDR threshold of <0.05. (F) Proportion of expanded clones in CD8 T[EM]s across time points grouped by response. Bar plots show the frequency of clonally expanded T cells at different time points and PFS groups. Expanded T cell clones were defined as >1 cell that expressed the same TCRαβ sequence, and non-expanded cells (singlets) were individual cells each with a unique TCRαβ sequence. Asterisks indicate significance after multiple-testing correction, comparing distribution of expanded and singlet cells for all T cell subsets, responder versus non-responders (logistic regression using two-sided t test with Benjamini-Hochberg FDR correction to compare between all T cell subsets). (G) Proportion of expanded clones in CD8^+ T[EM]s across time points grouped by response. Mean ± SEM CD8 T[EM] proportion of T cell repertoire per participant depicted. Each sample was downsampled to 500 cells. Blue dots depict responding participants, and red open squares depict non-responding participants. A mixed-model ANOVA corrected for multiple comparisons with a Bonferroni’s test was used to compare differences between time and response groups. (∗p < 0.05). We reasoned that a reliable biomarker of treatment response from peripheral blood would be derived from the most consistently enriched immune cell populations between responders and non-responders over time, and not driven by only a few participants. We employed scCODA[114]^14 (single-cell compositional data analysis) to analyze the changes in proportions of immune cells across time. Comparative analysis between responders and non-responders revealed that only CD8^+ T[EM] cells exhibited higher abundance in chemoimmunotherapy responders across time points ([115]Figure 2D). This finding was further corroborated by visualizing temporal changes in individual patient CD8^+ T[EM] proportions ([116]Figure S2G). To further explore the CD8^+ TEM immune compartment, we used Milo,[117]^15 a single-cell differential abundance tool to identify more specific immune cell subpopulations expanded in chemoimmunotherapy responders. We confirmed that CD8^+ T effector memory (T[EM]) cells were more abundant across responders across all time points ([118]Figures 2E and [119]S2H). We also mapped differentially abundant cell subpopulations to individual participants, confirming that CD8^+ T[EM] cells were more consistently abundant in responders ([120]Figure S2H). We further confirmed that these CD8^+ T[EM] cells in responders were transcriptionally distinct from their non-responder counterparts. In responders, pathway enrichment analysis of CD8^+ T[EM] showed increased responsiveness to interferon-γ, with earlier time points (time point 0, time point 1) evidencing larger differences in CD8^+ T[EM] activation ([121]Figures S2I–S2N). Having observed enrichment of CD8^+ T[EM] in responders, we next interrogated single-cell T cell receptor (TCR)αβ sequencing data to determine whether the clonal distribution of CD4^+ and CD8^+ T cell subsets was different between responders and non-responders. There were no systemic differences between the total number of TCRαβ clones recovered between responders and non-responders ([122]Figure S2O). Comparing the relative proportions of expanded to singlet clones for each T cell subset between these two groups, CD8^+ T[EM] was the most clonally expanded subset at each time point ([123]Figure 2F). Overall, proportions of CD8^+ T[EM] expanded clones were also significantly increased in responders ([124]Figure 2G), in line with the single-cell transcriptomics analysis. Chemoimmunotherapy responders are enriched for circulating CD8^+ T[EM]s with stem-like properties Having shown that the CD8^+ T[EM] cells were consistently more abundant in responders across time points, we next further characterized these cells in detail. To ensure our observations are representative and generalizable, we focused on subclusters of cells derived from at least 5 participants. On this criterion, we used a Leiden-based clustering on the top 2,000 highly variable genes to delineate 7 clusters ([125]Figure 3A), composed of 1,200–12,400 cells per cluster ([126]Figure 3B). Clusters 1, 2, and 3 were more abundant in responding participants at all time points ([127]Figure 3C) and were abundant across all participants ([128]Figure S3A). Although we observed enrichment of clusters 4 and 5 in responding participants, and cluster 6 enrichment in non-responding participants, these clusters were enriched in fewer than 5 participants ([129]Figure S3A). We therefore focused downstream analysis on clusters 1, 2, and 3. Figure 3. [130]Figure 3 [131]Open in a new tab Chemoimmunotherapy responders are enriched for CD8^+ T[EM]s with stem-like properties (A) Subclusters of CD8^+ T[EM] using Leiden-based clustering. (B) Cluster composition, grouped by both treatment response and time point. (C) Differential abundance analysis across CD8^+ T[EM] clusters. (D) Average (gene-scaled) expression of CD8^+ T cell activation, differentiation, and exhaustion markers. (E) Average (gene-scaled) expression of genes associated with T cell activation (GO:0042110), by time point and CD8^+ T[EM] cluster. (F) GSEA of stem-like signature in CD8_TEM_2 cluster. (G) Proportion of persistent clones in CD8^+ T[EM] clusters across time and response. Mean ± SEM proportion of cluster CD8_TEM_1, CD8_TEM_2, and CD8_TEM_3 cells persistent clones per participant depicted. Blue dots depict responding participants, and red open squares depict non-responding participants. Repeated measures two-way ANOVA with Sidak’s multiple comparisons test was used to compare differences between time and response groups. Data presented as mean ± SEM; ∗p < 0.05, ∗∗p < 0.01. (H) T[pex] and T[stem] enrichment per single-cell cluster across time points. Positive NES is with respect to responders. (I) T[pex] and T[stem] enrichment in bulk transcriptomics data across time points. We next examined markers for these clusters ([132]Figures 3D and [133]S3B). Cluster 1 (CD8_TEM_1) cells highly expressed ITGB1, GZMH, GZMB, FGFBP2, and LGALS1, all genes associated with cytotoxic function.[134]^16 Cluster 2 (CD8_TEM_2) cells demonstrated expression of TCF7, MYB, SLAMF6, and PD1 in the absence of TOX ([135]Figure 3D), suggesting a progenitor-exhausted memory T cell population.[136]^17 Cluster 3 (CD8_TEM_3) cells did not express MYB or TCF7 but instead expressed thymocyte selection associated high mobility group box (TOX) and killer cell lectin-like receptor (KLR) genes such as KLRC1, KLCR2, and KLCR3[137]^18^,[138]^19 and genes coding for cytotoxic molecules GNLY, PRF1, and GZMB, consistent with an exhausted T cell effector phenotype. In these three subclusters, responders showed higher expression of genes associated with T cell activation including IRF1, EOMES, IFNG, major histocompatibility complex class I and II, and CD5, with larger differences in phenotype-specific gene expression at both time point 0 and time point 1 compared to time point 2 ([139]Figure 3E). Peripheral blood CD8^+ memory T cells have stem-like features that enable them to self-renew, proliferate, and differentiate into various effector and memory progeny. Additionally, progenitor-exhausted CD8^+ cells (T[pex]) also possess such stem-like features, which are important for durable antigen-specific T cell responses in chronic infection and tumor responses to immune checkpoint blockade.[140]^20^,[141]^21^,[142]^22 We verified that only cluster 2 was enriched for stem-like gene signatures found from other large single T cell studies in mice, HIV patients, and melanoma patients[143]^23^,[144]^24^,[145]^25^,[146]^26 ([147]Figures 3F and [148]S3C). Across time, enrichment analysis supported expansion of stem-like T cells in responders, both in cluster 2 cells and in bulk transcriptomics data ([149]Figure S3D). We examined the clonal properties of cluster 2 cells to determine if they displayed stem-like features. Cluster 2 cells had more unique single clones compared to clusters 1 and 3 ([150]Figure S3E). This is reflected by increased TCRαβ diversity in cluster 2 ([151]Figure S3F), suggesting a less clonally expanded and less differentiated population. Responders and non-responders were similar in TCRαβ diversity for cluster 2 ([152]Figure S3G). Responding participants had more cluster 2 clones that were present at all time points, but this was not observed with cluster 1 and 3 clones ([153]Figure 3G), suggesting that cluster 2 consisted of CD8^+ T cell clones that were persisting and self-renewed over time in responders. This is consistent with previous studies showing persistence of neoantigen-specific CD8^+ T cells and superior therapeutic outcomes after adoptive cell transfer when the CD8^+ T cells had a stem-like phenotype.[154]^27 A recent landmark study in human peripheral blood identified two subsets of peripheral stem-like memory CD8^+ T cells with distinct fate commitments.[155]^23 One subset is the precursor to a lineage of exhausted CD8^+ T cells (termed progenitor-exhausted-like T cells, or T[pex]), and another is the precursor to peripheral memory cells (stem-like T cells, or T[stem]).[156]^23 Understanding the lineage of blood CD8^+ memory T cells is important because intratumoural T[pex] cells have previously been shown to be reinvigorated by ICT in melanoma patients.[157]^25 We investigated if cells from clusters 1, 2, and 3 shared a transcription program similar to T[stem] and T[pex] in responding versus non-responding participants following chemoimmunotherapy. Gene set enrichment analysis demonstrated that responders showed enrichment of the T[pex] signature while non-responders had enrichment of the T[stem] signature in cluster 2 cells across all time points ([158]Figure 3H). Additionally, we confirmed these observations in bulk transcriptomics data, with responders showing enrichment of the T[pex] signature in peripheral blood at all time points and downregulation of T[stem] ([159]Figure 3I). Taken together, our results suggest that a distinguishing characteristic of responding patients following chemoimmunotherapy is the high numbers of stem-like CD8^+ T cells with a T[pex] phenotype that persist over time. Stem-like CD8^+ T[EM] expansion in responders is linked to a permissive pre-treatment tumor microenvironment Next, we explored whether the tumor microenvironment was associated with CD8^+ T[EM] expansion in peripheral blood in chemoimmunotherapy responders. We reasoned that the observation of more stem-like CD8^+ T[EM] cells in peripheral blood could be combined with signatures found in the tumor transcriptome to improve response prediction to chemoimmunotherapy treatment. We used data integration analysis for biomarker discovery using latent components (DIABLO)[160]^28 as a multi-omics feature selector to allow us to simultaneously identify the most discriminative genes for chemoimmunotherapy response in both NanoString tumor data and bulk transcriptomics data from peripheral blood samples. We observed that a feature selector trained on a combination of tumor data and peripheral blood transcriptomics data from the first two time points provided the best discrimination between responders and non-responders ([161]Figure S4A). We therefore focused downstream analysis on discriminative genes from this assay combination. In this combination, genes most strongly correlated with responders originated from bulk peripheral blood data, while features most strongly correlated with non-responders were found within the tumor transcriptome ([162]Figure 4A). Given the relatively small number of participants and high number of genes in each assay, we selected only robust, discriminative genes, defined as those displaying high feature importance across repeated, diverse subsamples of participants ([163]Figures S4B and S4C). Discriminative genes for responders in peripheral blood were more highly expressed in T cells rather than in myeloid cells ([164]Figure 4B). Conversely, discriminative genes for responders identified in tumor were not highly expressed in the most common cell populations in peripheral blood ([165]Figure S4D). These genes were predictive in most but not all participants, as shown by a subset of responder participants (n = 7) who displayed some but not all the identified immunological correlates ([166]Figure 4A). Figure 4. [167]Figure 4 [168]Open in a new tab Stem-like CD8^+ T[EM] expansion in responders is linked to a permissive pre-treatment tumor microenvironment (A) Feature correlation matrix visualizing correlation strength between DIABLO-extracted genes (x axis) and samples (y axis) using tumor and peripheral bulk transcriptomics data from the first two time points. (B) Combined factor loading plot and expression dot plot showing the top genes selected by DIABLO in peripheral blood in responders (red) and non-responders (blue). Displayed genes were expressed in at least 25% of cells in the single-cell dataset. (C) Enrichment analysis of predictive genes in peripheral blood (bulk transcriptomics) assay. (D–F) (D) Enrichment analysis of predictive genes in tumor (NanoString) assay. Kaplan-Meier analysis for (E) PFS and (F) OS for participants, stratified by stem-like CD8^+ T cell signature enrichment in peripheral blood. (G) Average gene set enrichment of the DIABLO signature (at N = 50 features) in each of the top 10 most common cell populations in single-cell data. Next, we examined biological pathways enriched from discriminative genes within peripheral blood and the tumor transcriptome. Using pathway analysis, discriminative genes in peripheral blood ([169]Figure 4C) were associated with lymphocyte and T cell differentiation, consistent with our bulk and single-cell analysis. Examining feature importance from loading plots, we observed that DIABLO prioritized key T cell regulators including PTPN22 and IKZF3, implicated in T cell differentiation and activation[170]^29^,[171]^30^,[172]^31 ([173]Figure 4B and [174]Table S1). Genes associated with poor response identified in tumor were linked to cellular and nuclear division ([175]Figure 4D). Of interest, DIABLO prioritized CEP55,[176]^32 a key regulator of cytokinesis, and both IL1R and TGFB1 ([177]Table S1), genes implicated in poorer patient outcomes to immunotherapy-containing regimens.[178]^33^,[179]^34^,[180]^35 Since we had observed in the single-cell data that subsets of stem-like CD8^+ T[EM] differentiated responders from non-responders, we asked whether DIABLO uses these subsets of CD8^+ T[EM] cells for predicting response. For peripheral blood gene expression data at each time point, DIABLO selects a set of principal components, which capture discriminative variance between responders and non-responders. We tested the statistical association between gene signatures for stem-like CD8^+ T cells and DIABLO-selected principal components using principal-component gene set enrichment (PCGSE).[181]^36 The stem-like T cell signature,[182]^24 which we previously showed to be enriched in the cluster 2 CD8^+ T[EM] cells (CD8_TEM_2) in single-cell data ([183]Figure S3D) in responders across time, was also the most consistently enriched signature in DIABLO principal components by PCGSE ([184]Table S1). Indeed, participants with higher enrichment of this stem-like T cell signature in peripheral blood demonstrated significantly longer progression-free survival (PFS) and overall survival (OS) ([185]Figures 4E and 4F). Orthogonally, we also showed that discriminative genes for responders in peripheral blood were more highly expressed in CD8^+ T[EM]s ([186]Figure 4G), specifically cluster 2 of CD8^+ T[EM]s ([187]Figure S4E), as compared to other CD8^+ T[EM] subclusters. These results together indicate that increased stem-like cluster 2 CD8^+ T[EM]s (CD8_TEM_2) were critical for differentiating responders and non-responders to chemoimmunotherapy. We conclude that expansion of the stem-like CD8 T[EM] compartment in responders was linked to a more permissive tumor microenvironment, with both gene expression in pre-treatment tumor and pre-treatment peripheral blood serving as surrogates to predict treatment response to chemoimmunotherapy in mesothelioma. Discussion Platinum-based chemotherapy is one of the most successful treatments to be used in combination with ICT.[188]^10 In mesothelioma, deep and prolonged partial responses are observed in some patients treated with chemoimmunotherapy.[189]^3^,[190]^6 However, previous studies have shown that the interaction between chemotherapy and ICT can be explained by some patients responding to chemotherapy, and others to ICT, with a minority only responding due to the combination.[191]^11^,[192]^12 Similarly, in this study, we observe that we can correlate the response to chemotherapy and ICT with CD8^+ T[EM] gene expression profiles in peripheral blood and tumor tissue, either before or early on during treatment, but not in all patients. For example, although we see more expanded CD8^+ T[EM] clones before and after treatment in responders as a group, there are individual responders who do not display this expansion. Conversely, some non-responders had expanded CD8^+ T[EM] clones.[193]^37^,[194]^38 Previous studies also demonstrate that clonal expansion of peripheral blood CD8^+ effector T cell subsets prior to or after the first cycle correlated with ICT response. It is possible that some of the response can be explained by relative high chemotherapy sensitivity in patients who did not have T cell expansion, or who responded through different effector mechanisms, such as, for example natural killer cell-mediated killing, which has been shown to be crucial for the response to ICT in mouse models of mesothelioma.[195]^39 Recent preclinical studies indeed showed that different immune effector cells can mediate the therapeutic response to ICT.[196]^40 Even within the biomarker-positive patients, it is possible that some patients responded to the chemotherapy and others to the immunotherapy, as there may be an overlap in the underlying biology. For example, recent studies showed that a T cell-driven inflamed tumor microenvironment predicted response to chemotherapy.[197]^41 A recent meta-analysis found that high levels of intratumoural but not circulating CD8^+ T cells prior to treatment are associated with improved treatment outcomes following ICT.[198]^42 CD8^+ memory cells, both intratumoural[199]^43 and in peripheral blood,[200]^44 have been linked with response to ICT. CD8^+ memory T cells with a stem-like, progenitor-exhausted phenotype have been found via single-cell RNA-seq of tumors in three independent melanoma studies.[201]^43^,[202]^45^,[203]^46 These cells have been known to mediate immunotherapy responses following ICT[204]^47^,[205]^48^,[206]^49 or adoptive cell transfer.[207]^27 The presence of these cytotoxic stem-like CD8^+ T cells, which we found in peripheral blood in responders, combined with previous studies indicating their presence within tumor tissue, suggests a model for explaining the coupling of tumor and peripheral blood transcriptome, whereby persisting circulating CD8^+ memory T cells with a progenitor-exhausted lineage form a responder-specific niche, which migrates and forms a CD8^+ T[pex] population within the tumor microenvironment. The phase 2 Pre0505 chemoimmunotherapy mesothelioma trial showed that patients with favorable outcomes to treatment possessed a more diverse tumor T cell repertoire.[208]^6 Taken together with our data, this is consistent with our model that stem-like CD8^+ T cells possessing higher TCR diversity enter the tumor microenvironment. Our study provides proof of concept that a parsimonious model that predicts treatment response can be formulated using biospecimens collected prior to treatment or early after initiation of treatment. Intriguingly, this study demonstrates the predictive utility of archival tissue, despite reservations in the clinical community about sampling potentially far removed from the start of treatment. In non-responding tumors, we demonstrate enrichment of genes related to proliferation and cell division, consistent with a recent study showing that such genes are associated with poorer survival in mesothelioma patients.[209]^50 However, when used on its own, the predictive value of tumor tissue is suboptimal. This is consistent with a recent study employing multi-omics analysis of tumor tissue in melanoma patients treated with ICT alone, which was also unable to find one common predictive biomarker, indicating heterogeneity in response.[210]^9 In our study, single-cell analysis allowed us to identify responder-specific, stem-like CD8^+ T[EM] subpopulations associated with response to chemoimmunotherapy. Having identified these subpopulations, we were able to recover gene signatures corresponding to these subpopulations from our bulk transcriptomics data. Of note, such an analysis would not have been possible using bulk RNA-seq alone. Our study illustrates the added utility and the enhanced resolution provided by a multi-platform approach for analyzing peripheral blood-based biomarkers for ICT response. In conclusion, our study demonstrates the potential utility of integrating peripheral blood and tumor tissue analyses for predicting response to chemoimmunotherapy in mesothelioma patients. Importantly, we demonstrate here that we could improve the predictive power of our analyses by including peripheral blood results from bulk transcriptomics data in combination with tumor RNA data. The current study provides a case in point that easily accessible peripheral blood samples, which are part of routine safety monitoring before initiating treatment, and archival tumor tissue, readily accessible from the patient’s diagnostic biopsy, could improve the prediction of response to chemoimmunotherapy treatment. Limitations of the study This study has several limitations that should be acknowledged. Distinguishing the chemotherapy response from the immunotherapy/chemotherapy combination response requires a randomized comparison between these treatments. The DREAM3R trial, which recently completed accrual, addresses this limitation with its randomized design and similar blood sampling scheme.[211]^4 As a larger study, DREAM3R will provide greater statistical power to identify small differences in immune cell subsets between responders and non-responders and to validate the hypothesis-generating computational studies presented in this paper. Cisplatin/pemetrexed treatment requires supportive care with dexamethasone before and after chemotherapy administration, to prevent allergic/hypersensitivity reactions to pemetrexed and nausea and vomiting from cisplatin.[212]^1 A short course of dexamethasone substantially reduces peripheral blood lymphocyte numbers, which could interfere with the biomarker readout.[213]^51 For that reason, the baseline sample was taken prior to dexamethasone. However, to not burden participants with frequent hospital visits, the second and third blood samples were taken on the day of cisplatin/pemetrexed treatment, after participants had received three doses of 4 mg of dexamethasone in the 24 h prior to each sampling. It is possible that this reduced the lymphocyte numbers at the later time points and that therefore the baseline blood sample was the most predictive. Although it would be attractive to omit dexamethasone from the treatment regime, as it potentially interferes with the mechanism of action of the ICT, there currently are no good alternatives to prevent these chemotherapy-associated toxicities.[214]^52 However, even with a potential effect of dexamethasone, we observed differences in CD8^+ T[EM]s between responders and non-responders at later time points. Our findings suggest that peripheral blood is an appropriate compartment for identifying potential early predictive markers during treatment, consistent with recent studies.[215]^53^,[216]^54^,[217]^55^,[218]^56 However, combining peripheral blood data with tumor data may further improve predictive capabilities. Additionally, while our study suggested enrichment of certain CD8^+ T[EM] populations in responders ([219]Figure S2G), the sample size did not permit a definitive analysis of specific subsets within this population. These limitations highlight the need for larger, randomized studies to validate and extend our findings. The ongoing DREAM3R trial is expected to address many of these limitations and provide more definitive insights into the predictive value of immune cell subsets in response to combination immunotherapy and chemotherapy. Resource availability Lead contact Further information and requests for resources should be directed to Dr. W. Joost Lesterhuis (willem.lesterhuis@uwa.edu.au). Materials availability This study did not generate new reagents or materials. Data and code availability * • The data generated in this study are available through NCBI’s Gene Expression Omnibus (GEO) repository under the following accession numbers: bulk RNA-seq data, GEO: [220]GSE252990 ; single-cell RNA-seq data, GEO: [221]GSE253173; NanoString data, GEO: [222]GSE248514; TCR data, GEO: [223]GSE252432. * • All original code used in this study has been deposited in a public repository and is available as of the date of publication. Computational analyses were performed using Snakemake (v6.3). Code to reproduce these analyses is available at [224]https://doi.org/10.5281/zenodo.8357270. * • Any additional information required to reanalyze the data reported in this paper is available from the [225]lead contact upon request. Acknowledgments