Phenotypic plasticity and reduced tissue retention of exhausted tumor-infiltrating T cells following neoadjuvant immunotherapy in head and neck cancer

Abstract

Neoadjuvant immunotherapies (NITs) have led to clinical benefits in several cancers. Characterization of the molecular mechanisms underlying responses to NIT may lead to improved treatment strategies. Here we show that exhausted, tumor-infiltrating CD8⁺ T (Tex) cells display local and systemic responses to concurrent neoadjuvant TGF-β and PD-L1 blockade. NIT induces a significant and selective increase in circulating Tex cells associated with reduced intratumoral expression of the tissue-retention marker CD103. TGF-β-driven CD103 expression on CD8⁺ T cells is reversed following TGF-β neutralization in vitro, implicating TGF-β in T cell tissue retention and impaired systemic immunity. Transcriptional changes implicate T cell receptor signaling and glutamine metabolism as important determinants of enhanced or reduced Tex treatment response, respectively. Our analysis illustrates physiological and metabolic changes underlying T cell responses to NIT, highlighting the interplay between immunosuppression, tissue retention, and systemic anti-tumor immunity and suggest antagonism of T cell tissue retention as a promising neoadjuvant treatment strategy.

Keywords: TGF-β; glutamine metabolism; neoadjuvant immunotherapy; neoepitope-specific T cells; single-cell transcriptomics; tissue-resident T cells.

Figure 1.. Phenotypic heterogeneity within the tumor-infiltrating CD4⁺ and CD8⁺ T cell compartment.

A, schematic illustrating the data generating process. B, scatter plot showing uniform manifold approximation and projection (UMAP) embedding of tumor-infiltrating lymphocytes color by TCR sequence availability. C, scatter plots showing UMAP embedding of CD4⁺ (top row) or CD8⁺ (bottom row) T cells colored by expression levels of indicated marker genes. D,E, scatter plot showing UMAP embedding of CD4⁺ T cells (c; n = 4799) and CD8⁺ T cells (d; n = 14852) colored by cluster ID. T cell subsets were distinguished based on functional properties abbreviated as: reg = regulatory; prol = proliferating; ex = exhausted; prog = progenitor; cm = central memory; ctl = cytotoxic T lymphocyte; em = effector memory; eff = effector; unk = unknown. See also Figures S1, S2 and Tables S1, S2.

Figure 2.. NIT leads to enrichment of tumor-infiltrating exhausted CD8⁺ T cells.

A, bar graph showing the fraction of non-exhausted cells for 20 CD8 clonotypes selected from patient 5 for TCR specificity analysis. B, heatmap showing log10-transformed ELISpot counts measured from peripheral blood T cells expressing reconstructed TCRs co-cultured with autologous antigen presenting cells pulsed with either in silico predicted neoepitopes or common viral peptides (CEF). Peptide sequences and corresponding gene symbols are indicated in row names; underscore indicates altered amino acid. HPV16 E7 TCR and the corresponding minimal epitope (HPV16 E7_11-19) served as a positive control for the assay. Non-reactive clonotypes were omitted. Representative data from one of two independent experiments with similar results is shown. C, bar graph showing phenotype composition of clonotypes in (A). Clonotypes highlighted in blue and red display virus- and tumor neoantigen–reactivity, respectively. D, scatter plot showing UMAP embedding of CD8⁺ T cells colored by association with clonotypes evaluated in (B). E, scatter plot showing UMAP embedding of CD4⁺ T cells (left) and CD8⁺ T cells (right) colored by sampling time with respect to treatment. F, scatter plot showing log2 fold-change in relative abundance of each CD8⁺ T cell cluster from pre- to post-treatment over cluster cell number. G, bar graph showing cluster-level correlation of select marker gene expression and log2 fold change in relative CD8⁺ T cell abundance. Positive or negative values indicate associations with post-treatment increased or decreased clusters, respectively. H, bar graph showing proportion of CD8⁺ T cells isolated pre- or post-treatment classified as exhausted or non-exhausted. See also Figures S3, S4 and Table S3.

Figure 3.. NIT leads to selective increase of exhausted, tumor-infiltrating T cell clonotypes in peripheral circulation.

A,B, scatter plots showing UMAP embedding of tumor-infiltrating CD8⁺ T cells colored by peripheral TCRβ frequency measured in blood samples obtained pre-treatment (A) or post-treatment (B). C, box plot showing distributions of log10-transformed peripheral TCRβ frequencies of individual CD8 T cells arranged by cluster ID and treatment. Frequencies of undetected peripheral TCRβ were mapped to the smallest measured frequency greater than zero before log-transformation. Wilcoxon test; ns: p > 0.05; *: p <= 0.05; **: p <= 0.01; ***: p <= 0.001; ****: p <= 0.0001. For the box plots, the horizontal line inside each box indicates the median, the top and bottom of the box indicate the interquartile range, and I bars indicate 5^th and 95^th percentiles. D, line graph showing the fraction of CD8⁺ T cells within indicated clusters exhibiting TCRβ sequences that are detectable, i.e. display a frequency greater than 0, in either pre - or post-treatment blood samples. E, scatter plot showing log2 fold changes of intratumoral and peripheral TCRβ frequencies of shared CD8 clonotypes comparing post- to pre-treatment frequencies. Positive values indicate greater frequencies post-treatment. F, line graphs show (left) the pre- and post-treatment frequency of TCRβ sequences associated with validated neoepitope-specific TCRs from the tumor and (right) the IFNγ spot counts (ELISpot assay) of peripheral blood T cells exposed to validated neoepitopes following in vitro stimulation. G, box plot showing distributions of ITGAE/CD103 expression in individual CD8⁺ T cells arranged by cluster ID and treatment. P value was obtained using two-way ANOVA. For the box plots, the horizontal line inside each box indicates the median, the top and bottom of the box indicate the interquartile range, and vertical bars indicate 5^th and 95^th percentiles. H, scatter plot showing the treatment-induced fold change in average peripheral TCRβ frequency over fold change in average ITGAE expression associated with the corresponding tumor-infiltrating CD8⁺ T clusters. I, representative scatter plots and histograms show pre- and post-treatment tumor infiltrating T cells and CD8⁺ T cell marker expression, respectively. FMO, fluorescence minus one. The right dot plot quantifies the fold change (post-treatment/pre-treatment, log2 transformed) of the percentage of CD8⁺ cells within the total live CD45⁺ population or the ΔMFI (experimental mean fluorescence intensive minus FMO intensity) of CD8⁺ cell surface markers CD103, LAG3 and CD39 for five of the six patients included in this study (cohort 1) and four additional trial patients (cohort 2). One sample t-test; *: p <= 0.05. J, representative histograms showing expression of cell surface CD103 on activated T cells in the presence of TGF-β with or without bintrafusp alfa. Percentage of positive cells is shown. TCA, T cell activation with anti-CD3/28 antibodies. K, bar plot shows percentage of CD103 positive activated T cells over five independent experiments. p-values determined with Wilcoxon rank sum test between two pairs of data as indicated; ***, P<0.001. Data shown as means +/− standard deviation. See also Figures S5, S6 and Table S4.

Figure 4.. Treatment-induced clonotypic cluster associations in the CD8⁺ compartment are associated with changes in metabolism and antigen-mediated signaling.

A,B, scatter plots showing UMAP embedding of CD8⁺ T cells associated with shared clonotypes (n = 5361) colored by cluster ID (A) or sampling time with respect to treatment (B). T cell subsets were distinguished based on functional properties abbreviated as: prol = proliferating; mem = memory; ex = exhausted; eff = effector. C, Alluvial diagram displaying CD8 clonotype-mediated cluster associations. Vertical axes represent the pre- and post-treatment cluster states, where clusters are vertically arranged as strata. A flow/association between a pre- (left) and a post-treatment (right) cluster represents all pairs of cells such that the first cell belongs to the pre-treatment cluster and the second cell belongs to the post-treatment cluster and the pair of cells shares the same TCR clonotype. D, box plot showing distributions of Reactome TCR signaling signature expression in individual CD8 T cells arranged by cluster ID and treatment. Wilcoxon test; ns: p > 0.05; *: p <= 0.05; **: p <= 0.01; ***: p <= 0.001; ****: p <= 0.0001. For the box plots, the horizontal line inside each box indicates the median, the top and bottom of the box indicate the interquartile range, and vertical bars indicate 5^th and 95^th percentiles. E, heatmap showing differences in mean expression of indicated gene sets of cells associated with the indicated flows shown in (c). For each flow and gene set the mean expression from pre-treatment cells was subtracted from mean expression of post-treatment cells. The mean difference was further divided by the standard error. Only flows involving more than 25 pre- and post-treatment cells each are shown. Columns were centered at zero and divided by standard deviation. Bar graph (left) shows the average clonotype expansion of post-treatment cells associated with the corresponding flow. F, heatmap showing pairwise correlations of indicated gene sets considering the mean differences shown in (E). G, H, I, scatter plots showing mean clonotype expansion over mean expression difference for each flow in (e) in gene signatures related to TCR signaling (G), glutamine metabolism (H) and TGF-β target genes (I). Point colors represent post-treatment cluster identity using encoding shown in (E). Pearson correlations and p-values are shown. See also Figures S7, S8.

Figure 5.. Treatment-induced clonotype plasticity in the CD4 compartment is associated with Treg phenotypic instability.

A,B, scatter plots showing UMAP embedding of CD4⁺ T cells associated with shared clonotypes (n = 947) colored by cluster ID (A) or sampling time with respect to treatment (B). T cell subsets were distinguished based on functional marker gene expression and abbreviated as: reg = regulatory; prol = proliferating; mem = memory; ex = exhausted. C, Alluvial diagram displaying CD4 clonotype-mediated cluster associations. Vertical axes represent the pre- and post-treatment cluster states, where clusters are vertically arranged as strata. A flow between a pre- and a post-treatment cluster represents all pairs of cells such that the first cell belongs to the pre-treatment cluster (left) and the second cell belongs to the post-treatment cluster (right) and the pair of cells shares the same clonotype. D, box plot showing distributions of TGF-β target gene signature expression in individual CD4 T cells arranged by cluster ID and treatment. Wilcoxon test; ns: p > 0.05; *: p <= 0.05; **: p <= 0.01; ***: p <= 0.001; ****: p <= 0.0001. For the box plots, the horizontal line inside each box indicates the median, the top and bottom of the box indicate the interquartile range, and vertical bars indicate 5^th and 95^th percentiles. E, Alluvial diagram displaying CD4 clonotype-mediated cluster associations, as described in (C), showing only flows between pre-treatment Treg clusters to post-treatment Tex clusters (left) or the reverse (right). F, scatter plots showing UMAP embedding of CD4⁺ T cells associated with shared clonotypes colored by FOXP3 expression; arrows connect pre-treatment Tregs to post-treatment Tex cells that share a clonotype (left) or the reverse (right). G, scatter plots showing UMAP embedding of CD4⁺ T cells associated with shared clonotypes highlighting pre-treatment Tregs and post-treatment Tex cells that share a clonotype (left) or the reverse (right). H, scatter plots showing fold change in average gene expression over mean expression of indicated genes comparing average expression within post-treatment CD4⁺ Tex cells to average expression within pre-treatment Tregs that share a clonotype. Corresponding cells are highlighted in (G, left). See also Figure S8.