PD1 expression is associated with exhaustion in CD8 but not CD4 memory TILs
To functionally characterize T cell populations from NSCLC, we dissociated surgically resected tumor and FACS-sorted tumor infiltrating lymphocytes (TILs) and their matching counterparts from adjacent histologically uninvolved lung tissue (NILs) from 18 patients (8 adenocarcinoma, 10 squamous cell carcinoma). Following isolation of CD45RO+ memory CD4+ and CD45RO+ memory CD8+ memory T cells, we further stratified both cell types by expression level of PD1, resulting in four subpopulations (PD1+CD4+, PD1-CD4+, PD1+CD8+ and PD1-CD8+) for each patients tumor and matched normal sample (Figure 1A-B and Figure S1A). We found that memory PD1+ T cells were more abundant in tumors than in uninvolved tissues (50% versus 31% in the CD4 and 45% versus 27% in the CD8 fraction), suggesting an activated and potentially exhausted phenotype of TILs (Figure 1C). In matched normal controls, PD1+CD4+ cells were less abundant than their PD1-CD4+ counterparts (31% versus 42%), while PD1+CD8+ and PD1-CD8+ cells were found with about equal frequency (27% versus 29%) (Figure 1C).
Next, we determined the effector potential of the T cell subpopulations by measuring their ability to expand in vitro and cytotoxic activity against lung cancer cell lines in a three-dimensional (3D) tumor spheroid assay14 (Figure 1E). All four CD4+ memory T cell subpopulations expanded well in vitro (Figure 1D) and effectively eliminated A549 lung adenocarcinoma cells, irrespective of their PD1 status and whether they originated from tumor or uninvolved tissue, indicating that their effector potential was intact (Figure 1F and Figure S1B,D). Conversely, amongst the CD8 subpopulations, only tumor-derived PD1+CD8+ memory TILs showed diminished growth capacity in vitro (Figure 1D) and failed to kill A549 lung adenocarcinoma cells(Figure 1F and Figure S1C). In contrast, tumor killing capacity of CD8+ T cells from the peripheral blood of healthy donors was intact irrespective of PD1 status (Figure 1F and Figure S1D). Similar patterns of tumor cell killing capacity were also confirmed against H157 lung squamous cell carcinoma cells (Figure S1E-F). We conclude that high PD1 expression correlates with an exhausted phenotype in CD8+TILs, but not CD4+ TILs or PD1+ CD8+ or PD1+ CD4+ T cells derived from adjacent normal tissue.
Transcriptional characterization of NSCLC memory T cell populations
Gene expression patterns of T cell populations derived from tumor tissue and adjacent matched uninvolved lung from eight NSCLC patients was then assessed by RNA Seq. Two-dimensional projection of the transcriptomics data using T-distributed Stochastic Neighbor Embedding (t-SNE) demonstrated clear separation of CD4 and CD8 subpopulations into two distinct clusters (Figure 2A). Within those clusters, the projection distinguished samples from tumor and matched uninvolved tissue, indicating a strong effect of the microenvironment on both CD4+ and CD8+ memory T cells. The expression level of PD1 has a small, but visible influence on global gene expression, particularly in sorted TILs.
The difference between CD4+ and CD8+ memory T cells is also reflected in the number of differentially expressed genes. Although we determined 664 genes significantly up- or downregulated (FDR <= 0.05) in both CD4 and CD8 cells sorted from tumor tissue compared to matched normal tissue, we found a roughly equal number of CD4- and CD8-specific genes (727 and 821, respectively) (Figure S2B). Interestingly, high PD1 expression had a much more pronounced effect on the number of significantly changed genes in TILs compared to NILs from matched uninvolved tissue, in particular in CD8+ T cells (820 tumor-specific genes, 74 shared, 33 normal-specific) (Figure S2C). We also noted a significant overlap of differentially expressed genes in PD1hi TILs between CD8+ and CD4+ T cells, compared to PD1lo (259 and 33 genes shared, respectively). PD1 expression is part of the T cell activation transcriptional program, thus it is conceivable that this observation relates to T cells encountering more antigen in the tumor but less so in matched uninvolved lung tissue.15
Gene set enrichment confirmed that downregulation of GPCR signaling is associated with tumor residence and high PD1 expression in CD4+ and CD8+ memory T cells, suggesting reduced T cell activation in chronically activated TILs (Figure 2B). CD4 and CD8 cells differed with respect to genes involved in cytokine receptor interaction, with CD4 TILs generally showing upregulation of such genes, while CD8+ TILs, particularly PD1+ cells, showing downregulation (Figure 2B and Figure S2A). A possible explanation is an increased number of helper or regulatory T cells within the CD4+ TIL population.
To provide a concise overview of the gene expression patterns found in the different T cell populations, we curated lists of T cell co-stimulatory and co-inhibitory genes, transcription factors playing a part in T cell function, as well as genes involved in cell cycle, mitosis and DNA repair, and plotted their expression as a heatmap (Figure 2C). PD1+CD4+ and PD1+CD8+ memory TILs displayed high expression of known exhaustion-associated genes, CTLA4, ENTPD1 (CD39), CD200, LAG3, TIGIT, HAVCR2 (TIM3)16, as well as CD38, and LAYN 17. In addition, PD1+CD4+ memory TILs showed elevated expression of PDCD1LG1 (CD274/PDL1) and PDCD1LG2 (CD273/PDL2), which negatively regulate T cell effector function by engaging with PD11. Moreover, we observed upregulation of MAGEH1 and CCR8, two genes associated with CD4 regulatory T cell function in breast18 and lung cancer patients19. Unlike PD1+CD4+ TILs, PD1+CD8+ memory TILs are characterized by downregulation of co-stimulatory molecules, including CD28, CD40LG, TNFSF8 (CD30LG) and memory-associated genes IL7R (CD127), SELL (CD62L) and LEF1 (Figure 2C). These changes coincided with downregulation of the memory precursor gene TCF7 (TCF-1). Furthermore, key transcription factors associated with T cell exhaustion, including TOX20-22, TOX223, and ID324, were upregulated in PD1+CD8+ memory TILs, whereas RUNX3, which controls expression of cytotoxicity-related genes, was downregulated. Expression of cell cycle and DNA repair genes was generally elevated in both PD1+ TILs, particularly in CD8 TILs (Figure 2C and Figure S2A). This finding is interesting considering the reduced ability of PD1+CD8 TILs to expand in vitro (Figure 1D). Furthermore, WNT signaling was downregulated specifically in tumor-derived PD1+ CD8 cells (Figure S2D). Overexpression of several canonical exhaustion-associated genes (TOX, TOX2, IRF4, CD200, CD38 and CTLA4) were shared between PD1+CD8+ and PD1+CD4+ memory TILs (Figure 2D). Additionally, we confirmed co-expression of multiple inhibitory receptors associated with chronic T cell activation and exhaustion at the protein level by flow cytometric analysis of PD1+CD8+ and PD1+CD4+ memory TILs (Figure S3A-B). We confirmed that a subset of PD1+CD8+ memory TILs that were enriched for CD38 and CD101 (Figure S3C), which were previously shown to be associated with a dysfunctional chromatin state in NSCLC7, exhibited diminished proliferation (Figure S3D).
We obtained gene signatures corresponding to functional T cell clusters from a recent single cell sequencing analysis in NSCLC17 and used them to determine the relative abundance of those clusters in our FACS sorted T cell populations with single sample gene set enrichment (ssGSEA). We found that in both CD4+ and CD8+ memory TILs, tumor residence and PD1 expression correlated with an exhausted state (C7-CXCL13 and C6-LAYN clusters, respectively) (Figure 2E and Figure S2E). Conversely, effector cell clusters (C3-CX3CR1 and C3-GNLY) showed a higher correlation in matched uninvolved lung tissue, compared with tumor, indicating an immunosuppressive tumor microenvironment. Naïve clusters tracked with both PD1- TILs in both CD4 and CD8 compartments. Notably, naïve regulatory T cells were most enriched in matched normal CD4 populations, while the suppressive regulatory T cell signature (C9-CTLA4) strongly correlated with tumor PD1+ status in CD4+ memory TILs (Figure 2E and Figure S2E).
Knock-down of ID3 in CD8 T cells increases their tumor killing capacity
Re-invigoration of exhausted T cells is a key goal of cancer immunotherapy, so we searched our gene expression data for candidates that could potentially regulate T cell exhaustion. Recent work has shown that the transcriptional regulator ID3 is enriched in dysfunctional TILs in human melanoma and NSCLC24. We found increased expression of ID3 in CD4+ and CD8+ TIL populations, compared to their matched controls (Figure 2C). Furthermore, ID3 expressing cells were most prominently represented in the exhausted CD8 cluster (C6-LAYN) at the single cell level. Notably expression of ID3 in some cells in the naïve cluster (C1-LEF1) (Figure 3A), may reflect the described role of ID3 in T cell differentiation25. Conversely, the effector cluster (C3-CX3CR1) did not show any expression of ID3 (Figure 3B). To test if ID3 had a functional role in T cell exhaustion, we knocked-down ID3 in peripheral blood-derived CD8 T cells using siRNA and evaluated their capacity to kill A549 lung adenocarcinoma cells (Figure 1E). We found that reduction of ID3 mRNA levels led to significantly increased killing compared to scrambled siRNAs and no siRNA controls (Figure 3C-D). Similar results were obtained in Jurkat (a leukemic T cell line) following knockdown of ID3 (Figure S4A-B). We conclude that ID3 suppresses the effector potential of CD8 cells. Therefore, strategies aimed at decreasing ID3 expression in CD8 TILs could help to re-invigorate them from their exhausted state.
PD1+CD4+ memory TILs facilitate B cell activation and expansion
Our transcriptional analysis suggested that PD1+CD4 memory TILs displayed features of both helper and regulatory T cells (Figure 2C). Zappasodi et al.26 recently identified a population of CD4 PD1+ cells (4PD-1Hi) that were transcriptionally similar to T follicular helper (TFH)-like cells, but were functionally immunosuppressive and correlated with poor patient survival in NSCLC. Visual inspection of the expression of the signature genes showed enrichment in CD4+ memory TILs (Figure 4A) and ssGSEA analysis demonstrated a relative enrichment of the T follicular regulatory (TFR) signature in the PD1+CD4+ memory TIL population (Figure 4B). Many of the genes most enriched in PD1+CD4+ memory TILs, such as CXCL13, CD200, TIGIT and SH2D1A, are also part of a TFH cell gene signature27 (Figure 4A). Moreover, PD1+CD4+ memory TILs were enriched in other CD4 helper-related genes, TNFRSF18 (GITR), TNFRSF4 (OX40) and TOX2 28. Using ssGSEA, we found equally strong enrichment of the TFH and the TFR signatures in PD1+CD4 TILs (Figure 4B).
Due to the challenges in differentiating TFR from TFH cells solely based on gene expression data, we utilized a more functional approach. Following in situ staining, PD1+CD4+ TILs were observed in close proximity to CD19 positive B cells (Figure 4C and Figure S4C), suggesting functional interaction. FACS-purified PD1+CD4 memory TILs, co-expressing the tumor-specific antigen receptor CD3929, as well as CD25 (IL2RA) and CD200 exhibited robust expansion in culture (Figure S4D). In vitro expanded PD1+CD4+ memory TILs were co-cultured with carboxyfluorescein succinimidyl ester (CFSE)-labeled CD19+ B cells, derived from dissected mediastinal lymph nodes of lung cancer patients and stimulated with the TCR activator staphylococcal enterotoxin B (SEB) (Figure S4E). As compared to B-cells alone, culture with activated CD4+ TILs induced B cell proliferation (38±13 versus 2.6±0.6%, respectively) (Figure 4D). Proliferating B cells were observed to downregulate expression of CD27 and upregulate CD38 (Figure 4E). This coincided with an upregulation of homing molecules CXCR5 and CCR7 (Figure 4F). Importantly, the majority of B cells did not upregulate CD24 (Figure 4G) or IL10 (Figure S4F), markers of B regulatory cells. Activated CD4+ TILs co-cultured with CD19 B cells re-express high levels of PD1 (Figure 4H), along with CD200, CD25 and CTLA4 (Figure 4I,J and Figure S4H,I) and showed an intact ability to degranulate, based on increased expression of CD107a (Figure 4K). In addition, within the PD1+ population, we observed an increase in the fraction of CD4 TILs co-expressing CXCR5 and CCR7, suggestive of an elevated potential to migrate to B cell zones (Figure 4L). Despite the fact that the sorted PD1+CD4+ memory TIL population likely represents a mixture of both TFH and TFR cells, we were able to show these cells are competent to engage B cells and thus functionally behave more like TFH rather than TFR cells.
Patient T cells expanded in vitro retain their cytotoxic and helper functions
Adoptive cell therapy (ACT) using autologous TILs expanded from patient tumors is a promising approach to treat solid tumors, including metastatic melanoma30,31 and NSCLC32. Presently, it is unclear if and to what extent T cells change their phenotype under persistent TCR stimulation. To address this, we generated primary T cell cultures from resected tissue of NSCLC patients and expanded T cells via repetitive rounds of T cell receptor (TCR) ligation, using anti CD3/CD28/CD2 beads in the presence of the gamma chain (γc) cytokines IL-2, IL-7, and IL-15 (Figure 5A). T cells from single cell suspensions of uninvolved tissues proliferated more readily than their matched tumor counterparts (38x106±12.8 versus 26x106±7 T cells, respectively) (Figure S5A). Bulk primary T cell cultures were then FACS sorted into populations, as previously described (Figure 1A). Compared with the native tumor, in vitro expansion altered the percentage of CD4 and CD8 subsets, with the extent and direction depending on the donor (Figure 5B). To quantify shifts in the composition within the CD4 and CD8 subpopulations, we performed transcriptomics analysis on each of the in vitro expanded subpopulation within the CD45RO memory compartment, calculated the enrichment of the T cell cluster signatures used previously (Figure 2E) and compared the in vitro patterns with the ex vivo prepared samples for which there were matching donors. Patterns of cell type composition were preserved well upon expansion in vitro, although the sharp distinctions between populations observed in vivo were less pronounced. Importantly, the dysfunctional clusters in both CD4 (C7-CXCL13) and CD8 (C6-LAYN) remain prominent in the PD1+ TIL population (Figure 5D). The PD1+CD4+ memory TIL subset re-isolated from expanded bulk cultures retained their ability to help B cells (Figure S5F-H). To test the composition of the TCR clonotype between matched ex vivo prepared and in vitro expanded populations, we in silico re-constructed the TCR repertoire from samples with overlapping donors (four for CD4 and three for CD8 populations). Quantification of the overlap of TCR alpha (Figure S5E) and beta chains (Figure 5E) demonstrated that the donor-specific repertoire was stable in vitro.
Surprisingly we observed a general decrease of PD1 transcript levels in all in vitro expanded subpopulations following persistent TCR stimulation (Figure 5C), a finding that was confirmed at the protein level (Figure S5A-B). This trend towards a less “exhausted” gene expression profile was further exemplified by downregulation of the co-inhibitory molecules CD200, CD38, CD160, LAYN, and CTLA4 and concomitant upregulation of co-stimulatory molecules, such as IL2RA (CD25), TNFRSF4 (CD134) and CD40LG, in in vitro expanded PD1+CD8+ memory TILs (Figure 5C). We confirmed these changes for several additional cell surface molecules at the protein level within the TIL fraction (Figure S5C,D). Furthermore, all T cell populations downregulated the transcriptional regulators IRF8, TOX, TOX2, and ID3, which are associated with an exhausted phenotype. PD1+CD8+ TILs upregulate memory/quiescence-associated genes (SELL, LEF1, CCR8) as well as cell cycle and DNA repair genes (PLK1, BRCA1) indicating a more proliferative phenotype. Both in vitro expanded PD1- and PD1+ populations displayed robust killing ability (Figure 5F), indicating that in vitro expansion of CD8+ memory TILs produces functionally intact cells.
Downregulation of components of the WNT pathway had already been observed in PD1+ CD8+ memory TILs in vivo (Figure S2D) and we observed a further downregulation of WNT1, WNT10A, and DKK3 upon in vitro expansion in all T cell populations (Figure 5C). As WNT signaling is known to arrest the development of effector CD8+ T cells33, we examined whether restoring WNT signaling via inhibition of GSK3β could further improve T cell function. We found that treating TILs during expansion with the GSK3 β inhibitor lithium chloride (LiCl) significantly increased the number of primary patient-derived primary lung tumor cells undergoing apoptosis, as judged by Annexin V staining, suggestive of enhanced TIL killing capacity (Figure 5G). Despite no difference in upregulation in PD1 (Figure S6D), activated TILs pretreated with LiCl displayed downregulation of Annexin V (Figure 5H). During the expansion period in vitro a reduction in the CD8 compartment occurred in the presence of LiCl (Figure S6A), despite this, when challenged with a pool of MHC Class I peptides both CD4 and CD8 TILs increased IFNγ secretion (Figure S6B) and upregulated expression of PD1 and CD80 (Figure S6C). We conclude that expansion of T cells in vitro could increase the number of autologous tumor-specific T cells for adoptive transfer and potentially also relieve some of the tumor-induced dysfunction, partially via restoring WNT signaling.