WHAT IS ALREADY KNOWN ON THIS TOPIC

A population of resident memory CD8+T lymphocytes (TRM), initially identified by the marker CD103 and present in the tumor microenvironment, may be the target of anti-PD-1. Tumor infiltration by this CD103+CD8+T cell population is associated with a good prognosis. More recently, other TRM subpopulations, expressing or not CD103, have been defined by other markers (CD49a, CD69, CXCR6) whose relationship, function and prognostic impact in the case of intratumoral infiltration are poorly understood.

WHAT THIS STUDY ADDS

This study shows that there are two main populations of TRM in non-small cell lung cancer (NSCLC), one co-expressing CD103 and CD49a and the other expressing CD49a only. They share many T-cell receptors, suggesting a common origin. Only intratumoral infiltration by TRM co-expressing CD103 and CD49a can predict treatment response and survival in patients with NSCLC treated with an anti-PD-1 agent in first or second line therapy. In a multivariate analysis including infiltration by total CD8+T cells, TCF1+CD8+T cells and the PD-L1 marker, the contribution of this subpopulation of resident T cells remained statistically significant.

HOW THIS STUDY MIGHT IMPACT RESEARCH, PRACTICE OR POLICY

TRM are targets for anti-PD-1 therapy and are involved in the mechanism of action of cancer vaccines. This work shows that not all TRM populations are equivalent and need to be better characterized for these purposes. Only one TRM subpopulation is able to predict response to anti-PD-1 immunotherapy in patients with NSCLC. Its predictive impact is statistically robust after multivariate analysis including the reference marker PD-L1, which is already used in clinical practice in the choice of first-line treatment of these patients. The inclusion of this TRM subpopulation in the initial work-up of these patients could improve their stratification and better personalize the different therapeutic options.

Introduction

More than two decades ago, pioneering work on vesicular stomatitis virus and Listeria monocytogenes revealed the presence and persistence of non-circulating resident memory T cells (TRM) in non-lymphoid organs after the resolution of the primary infection.1 It rapidly became apparent that these TRM constituted a specific lineage associated with a profile of transcription factors including Blimp1, Runx3, and Notch family proteins.2 In terms of their phenotype, these TRM express core markers such as CD69, CD103, and CD49a, together with the loss of expression of other markers such as CD62L, CCR7, S1PR1, and KLF2, favoring the persistence of these cells within tissues.3 4 TRM have innate-like “sensing and alarming” properties that enable them to recruit other immune cells to control microbial infections.5–7 As they are located at the site of inflammation in the tissues, TRM respond much more rapidly to reinfection and provide superior protection compared with circulating memory cells, including central memory and effector memory T cells.5 6 8

In a range of preclinical cancer models, we have shown that TRM are required for the efficacy of antitumor vaccines against mucosal tumors such as lung and head and neck cancer.9 10 In humans, high levels of intratumoral TRM infiltration have been associated with better clinical outcomes in multiple solid tumors including lung, melanoma, bladder, breast, cervical, ovarian, endometrial, gastric, and colorectal cancers receiving standard-of-care treatments.10–12 More recently in humans, correlative studies in non-small cell lung cancer (NSCLC), bladder cancer, and melanoma have revealed an association between tumor infiltration by CD8+ T cells with a resident phenotype before immunotherapy and responses to immune checkpoint blockade.13–16

To explain this predictive role of TRM during immunotherapy, various groups have shown that during neoadjuvant treatment in breast and head and neck cancers, CD8+ tumor-infiltrating lymphocytes (TILs) with a tissue-resident phenotype expand and are characterized by a gene expression program related to activation, cytotoxicity, and effector functionality.17 The same expansion of CD8+ TRM has been observed after anti-PD-1/PD-L1 monotherapy or combined anti-CTLA-4 treatment in melanoma, lung, breast, and esophageal cancers.13 18 19 The role of TRM as immunotherapeutic targets raises the possibility that other effectors recruited secondarily or present in the blood may play a role in the efficacy of immunotherapy.20

Given the antitumor role of TRM and their prognostic and predictive value in the context of patient responses to immunotherapy, the issue of optimal strategies for inducing or increasing this population is becoming a major challenge in immuno-oncology. In preclinical models, we have shown that the nasal route, but not the muscular route, induces TRM with a CD103+CD49a+CD69+ phenotype.9 10 This induction was associated with the inhibition of the growth of lung or head and neck tumors. In infectious and oncological models, other groups have also documented a correlation between the preferential ability of mucosal immunization to induce TRM expansion and protection against the development of cancers and viral infections.21 However, other studies based on vaccinations using systemically administered recombinant viral vectors have shown that TRM can be induced in the lungs.22 This result may be explained by the use of viruses that enable vaccine dissemination in the pulmonary or head and neck compartments. However, messenger RNA-based vaccinations have also been shown to induce TRM when administered via the intramuscular (i.m.) route, but the cells induced in this context generally express CD69 or CD49a without any concomitant CD103 expression.23 The fact that different subpopulations of TRM exist with differing core marker (CD103, CD49a, CD69) expression patterns has been reported in various tissues, and this effect has sometimes been linked to the different functions of these cells.24 Although CD103 is considered a hallmark of TRM, persistent CD103-negative (CD103neg) TRM have also been described in tissues. In contrast with CD103+ TRM, these cells were able to develop in a transforming growth factor beta (TGF-β)-independent manner.25

In this work, we aimed to better characterize these different TRM subpopulations in mice and humans and to determine whether they play distinct roles as predictors of responses to immunotherapy in patients with NSCLC.

ResultsDifferent immunization routes give rise to distinct subpopulations of resident memory CD8+T lymphocytes

Our previous work had shown that only the intranasal (i.n) immunization route induced resident memory CD103-expressing CD8+T cells in bronchoalveolar lavage fluid (BAL), and this mucosal immunization route was associated with tumor rejection.10 In recent years, it has emerged that there are different TRM subpopulations defined by the markers CD103, CD49a, and CD69.4 In the present study, only i.n vaccination with the STxB-E7 vaccine combined with alpha-galactosylceramide (αGalCer)-induced Db-restricted E749–57 peptide-specific CD8+T cells co-expressing CD103 and CD49a in the BAL (figure 1A). In contrast, the i.m. route also induced E7-specific CD8+ T cells expressing CD49a but not CD103 (figure 1A). Both CD103+CD49a+ and CD103negCD49a+CD8+ T-cell populations induced by the i.n route expressed high levels of CD69 (for CD103+CD49a+: mean 94.82%±1.86% and for CD103negCD49a+ mean: 95.5%±1.49%) (figure 1A, lower right), whereas the frequency of CD8+ T cells specific for E7 and expressing neither CD103 nor CD49a, which are considered to be effector T cells, exhibited weaker CD69 expression (mean 49.24%±10.24%) (figure 1A, lower right). Immunization via the i.n route induced a marginal CD103+CD49a− T-cell population (<5%) (figure 1).

Figure 1

Various subpopulation of specific TRM with different phenotypes and functionality are induced depending on the route of immunization. C57BL/6J mice were immunized with STxB-E7 and alpha-galactosylceramide by intranasal (i.n) or intramuscular (i.m.) route at day 0 and 14, then sacrificed at day 21, CD8a APCeFluo 780 (5 μg) were injected intravenous 5 min before sacrifice to discriminate circulating CD8+T cells and resident CD8+T cells. (A) Representative flow cytometry plots in BAL (broncho-alveolar lavage) of specific E7-tetramer, CD103+CD49a+TRM, CD103−CD49a+ TRM and CD103−CD49a− Teff, and CD69 frequency among these populations. (B) Absolute number of (left) E7-tetramer CD8+ and (right) CD103+CD49a+TRM, CD103−CD49a+ TRM and Teff CD103−CD49a− in BAL. Datas are expressed as mean±SEM. One representative experiment with three to five mice from two independent experiments is shown. Analysis of difference within groups were performed with two-side Mann-Whitney t-test. (C) Percentage of PD-1 among E7-specific CD103+CD49a+ TRM, CD103−CD49a+ TRM and CD103−CD49a− Teff in the BAL. (D) E7-specific CD103+CD49a+ TRM, CD103−CD49a+ and CD103−CD49a− Teff were sorted from BAL at D21, and stimulated (10,000 cells/well) with E749–57 peptide (10 μg/mL) for 18 hours. Then supernatant were harvested and cytokines multiplex assay was performed. Datas are expressed as mean±SEM. One representative experiment with three to five mice from two independent experiments is shown. Analysis of difference within groups were performed with one-way analysis of variance paired-test with Tukey multiple comparison. *p<0,05 **p<0,01 ***p<0001 ****p<0,0001. IFN, interferon; PD-1: programmed cell death 1; TNF: tumor necrosis factor; TRM, resident memory T cells.

It should be noted that i.m.-induced CD49a+CD103−CD8+T cells were less likely to express CD69 as compared with i.n-induced ones (mean: 68.4%±11.6%) (figure 1A, upper right), but their numbers were equivalent in the BAL irrespective of the immunization route (figure 1B). These results were replicated by analyzing the same resident memory CD8+ T-cell subpopulations in lung parenchyma after i.m. or i.n vaccination (online supplemental figure S2A,B).

Similarly, using another vaccine system consisting of the protein ovalbumin mixed with the adjuvant c-di-GMP, only immunization via the i.n route induced ovalbumin-specific CD103-expressing CD8+T cells in the BAL (online supplemental figure 2C,D). The subcutaneous route, like the i.m. route above, induced only CD103negCD49a+CD8+T cells (online supplemental figure S2C,D). Single-cell transcriptomic analyses in lung tumor-bearing mice showed that the CD49a+CD103neg T-cell population expressed CD69 as well as other resident markers (Zfp683, Runx3, Blimp1, Fabp5), but no lymph node homing markers such as Sell (online supplemental figure S4A). Characterization of these resident memory E7-specific CD8+ T-cell subpopulations (CD103+CD49a+ and CD103negCD49a+) revealed that the CD103+CD49a+CD8+T cells expressed more PD-1 (mean: 78.04%±9.65%) as compared with the CD103negCD49a+ population (mean 62.66%±8.14%) and the CD103negCD49aneg effector CD8+ T-cell population (mean: 47.46%±7.03%) (figure 1C). Interestingly, the resident memory CD8+ T-cell population co-expressing CD103 and CD49a appeared to be more functional after antigenic stimulation with a peptide derived from the E7 protein, secreting more interferon (IFN)-γ, tumor necrosis α (TNF-α), CCL4, and CCL5 (figure 1D) highlighting the more functional and protective phenotype of these cells.

Figure 2

Phenotypic analyses of subpopulations of TRM and effector T cells among TILs derived from patients with lung cancer. Fresh biopsies from patients with lung cancer (n=20) were dissociated and digested, and flow cytometry was used to analyze TILs. The number of TILs tested per marker is shown below each figure. (A) The percentages of TRM subpopulations (CD103+CD49a+ and CD49a+CD103+) among CD8+T cells, as well as non-effectors TRM (CD49a− CD103−) among CD8+T cells are shown. (B) The percentages of different markers defining TRM (CD69, CXCR6), exhausted T cells (PD-1, Tim-3, CD39), cytotoxicity (GZMB), and proliferation (Ki67) are shown among the two populations of TRM and non-TRM effectors (CD103−CD49a−). Significance was determined using paired t-tests. p<0.05 was regarded as statistically significant. *p<0.05, **p<0.01, ***p<0.001; n=4–20. TIL, tumor-infiltrating lymphocyte; TIM-3: T-cell immunoglobulin mucin 3; GZMB: granzyme B; PD-1: programmed cell death 1; TRM, resident memory T cells.

Distinct subpopulations of TRM-type CD8+T lymphocytes coexist among lung TILs

To determine whether the same TRM populations are present in human lung tumors and to compare their phenotype, TILs were obtained from 20 dissociated tumors from patients with NSCLC. The same TRM subpopulations identified in mice (CD103+CD49a+ and CD103−CD49a+) were detected in humans, with the CD103+CD49a+ population being present at a higher frequency (mean: 47%±22.52% of total CD8) (figure 2A). Nearly all of these TRM exhibited an effector memory T-cell phenotype (CCR7−CD45RA−) (online supplemental figure S3). The two TRM subpopulations expressed CD69 at a higher frequency (mean %CD69+: 87.2%±6.9% in CD103+CD49a+ and 66.6%±12.8% in CD103negCD49a+) as compared with the effector CD8+T cell population (CD103negCD49aneg; 33.8%±14.12%) (figure 2B). The CD103+CD49a+ cells expressed exhaustion markers (PD-1, T-cell immunoglobulin mucin 3 (Tim-3), CD39) and the proliferation marker Ki67 significantly more frequently (figure 2B). These data were confirmed by single-cell transcriptomic analyses of intratumoral CD8+T cells, which showed that the resident memory CD103+CD49a+CD8+ T cells exhibited higher expression levels of exhaustion, cytotoxicity and proliferation markers compared with the CD103negCD49a+CD8+T cell population (online supplemental figure S4B). Based on our own cohort data set and the re-analyzed data set from Clarke et al,26 we also showed that the CD103+CD49a+CD8+ T cells expressed higher levels of cytokines (IFN-γ, interleukin (IL)-2) and chemokines (CCL3, CXCL13) (online supplemental figure S4B,C).

We regard the CD103negCD49a+ population as a TRM population, as most of these cells express CD69, which is considered to be a TRM marker.3 Interestingly, as opposed to the non-resident CD103negCD49aneg CD8+ T cells, they do not express the circulating and lymph node homing markers (SELL, S1PR1) or the KLF2 transcription factor (online supplemental figure S4D).

In contrast, they express the transcription factors Hobit (ZNF683) and Runx3, which are hallmarks of TRM. However, unlike to mouse TRM, these transcription factors are not enriched in human TRM (online supplemental figures S4D and S5).2

Relationship between the TRM subpopulations in terms of differentiation

To define the relationship or not between these two main TRM populations, a comparative analysis of the T-cell receptor (TCR) repertoire of these cells and differentiation markers was carried out.

Four fresh tumors were used to conduct single-cell transcriptomic analyses, revealing that, among all the patients, the five most frequent clonotypes (TRA or TRB) of the double positive TRM CD8+T cell subpopulation (CD103+CD49a+) were also present in the single positive TRM CD8+ T-cell subpopulation (CD103negCD49a+) as well as in the CD103negCD49aneg effector CD8+ T-cell population (figure 3A). These clonotypes were most amplified in the CD103+CD49a+CD8+ T cells, followed by the CD103negCD49a+CD8+T cell population and finally the CD103negCD49aneg CD8+T cell population (figure 3B), suggesting possible differentiation and proliferation of effector CD8+ T lymphocytes into the resident memory CD103negCD49a+CD8+ T cells and then into resident memory CD103+CD49a+CD8+ T cells. Based on the Jaccard and the Morisita-horn dissimilarity indices (figure 3C,D, respectively), we found that for three out of four patients analyzed, closer repertoire composition and less dissimilarity (index close to 0) was observed between the CD103+CD49a+ and the CD103negCD49a+ CD8+T cell populations as compared with the CD103negCD49aneg population, reinforcing the relationship between the two TRM populations. With respect to the trajectories of these three subpopulations, another argument supports a more terminal differentiation of the CD103+CD49a+ TRM population since they express less TCF1, a progenitor marker (online supplemental figure S6). Indeed, in mice, the % of cells expressing TCF1, a stemness marker, is lower (mean 57.03%±SD 6.37%) in the CD103+CD49a+ CD8+T cell population, than in the CD103negCD49a+CD8+T cells (mean 73.1%±5.45%) and in the CD103negCD49aneg effector CD8+T populations (mean 71.78%+9.45%) (online supplemental figure S6). In humans, multiplex in situ immunofluorescence imaging revealed that the CD103+CD49a+CD8+T cell population does not express TCF1 (results not shown), as described previously.27

Figure 3

T-cell receptor sharing among the subpopulations of resident memory CD8+T cells. (A) Tracking of the most predominant clonotypes within the ITGA1+(CD49a)/ITGAE+ (CD103) population. Alluvial plots represent the relationships between the frequencies of the five most predominant T-cell clonotypes detected within the ITGA1+/ITGAE+ population (right barplot), in the ITGA1neg/ITGAEneg (left barplot) and ITGA1+/ITGAEneg (middle barplot) populations for each patient. Each square represents the frequency of a clonotype in the corresponding population. (B) Fold change of the most predominant clonotypes within the ITGA1+/ITGAE+ population. Dots represent the five most predominant T-cell clonotypes detected in the ITGA1+/ITGAE+ population observed using the log2 fold change in ITGA1+/ITGAE+ (right) and ITGA1+/ITGAEneg (middle) compared with the ITGA1neg/ITGAEneg (left) cell populations by patient. Each dot is linked across cell populations by a line colored by patient. Statistical analysis was performed using paired Student’s t-tests (****p<0.0001, **p<0.01). (C) Jaccard overlap among repertoires was analyzed by generating a heatmap of the Jaccard dissimilarity index calculated across the three cell populations: ITGA1neg/ITGAEneg (white), ITGA1+/ITGAEneg (gray), and ITGA1+/ITGAE+ (black). The Euclidean distance was used for hierarchical clustering as a color-coded matrix ranging from 0 (minimum dissimilarity) to 1 (maximum dissimilarity). (D) Morisita horn overlap among repertoires was analyzed by generating a heatmap of the Morisita horn dissimilarity index calculated across the three cell populations: ITGA1neg/ITGAEneg(white), ITGA1+/ITGAEneg (gray), and ITGA1+/ITGAE+ (black). The Euclidean distance was used for hierarchical clustering as a color-coded matrix ranging from 0 (minimum dissimilarity) to 1 (maximum dissimilarity). Patients included in this figure are color-coded as patient 1 (orange), patient 2 (green), patient 3 (blue), and patient 4 (red).

To confirm this relationship and the possible difference in the stage of differentiation between these T-cell populations, we performed pseudotime trajectory inference analysis on the transcriptome single cell RNA-seq (scRNA-seq) data. We can observe a differentiation from the double negative (DN) population (CD103negCD49aneg) to the DP (double positive) population (CD103+CD49a+) via the SP population (CD103negCD49a+) (online supplemental figure S7).

Distribution and infiltration of lung tumors by resident memory CD8+ T-cell subpopulations

The distribution and location of the different TRM subpopulations were then determined using a multiplex in situ immunofluorescence.

This technique has enabled us to compare infiltration by different TRM subpopulations either in the tumorous or stromal zone (figure 4A). We also included the TCF1 marker when conducting this staining (figure 4A), as it is often considered a marker of stemness potentially associated with response to immunotherapy.28 We observed a higher density of total CD8+T cells, as well as all TRM subpopulations, in the stroma compared with the tumor zone (figure 4B). In the tumor, the CD103+CD49a+ CD8+ TRM population was more frequently detected (mean: 20±62 cells/mm2) as compared with the CD103−CD49a+CD8+ TRM population (mean: 9±11 cells/mm2), but with no statistical difference. Both populations were also present in the stroma at higher density (mean: 119±179 cells/mm2 for CD103+CD49a+ CD8+ T cells; mean: 92±84 cells/mm2 for CD103−CD49a+CD8+T cells) (figure 4B). The CD49a marker has also been reported to be expressed by endothelial cells29 (figure 4A). Since E-cadherin interacts with the CD103 molecule, we investigated whether the CD103+CD8+T cell population interacted more with E-cadherin+ cells than the CD103negCD8+T cells.

Figure 4

Infiltration of non-small cell lung cancers by subpopulations of resident memory CD8+ T lymphocytes. (A) Representative image of the infiltration of non-small cell lung cancer. Multiplexed immunostaining was performed on paraffin-embedded tissues with antibodies to detect CD8, CD103, TCF1, CD49a and E-cadherin. inForm software enabled cell phenotyping and tissue segmentation that was performed with E-cadherin staining to discriminate tumor and stromal areas. Automated counting and mapping enabled the phenotyping of T cells: subpopulations of non-TRM tumor-infiltrating lymphocyte (defined as CD8+CD49a−CD103−TCF1±) and of CD8+ TRM lymphocytes defined as CD8+CD49a+CD103+TCF1− (white arrow), CD8+CD49a+CD103−TCF1+, CD8+CD49a+ CD103− TCF1−. Original magnification ×200. Cell number (B) and percentage (C) of non-TRM and CD8+ TRM lymphocytes tumor-infiltrating lymphocytes were determined by in situ immunofluorescence. Each dot represents one patient. The average number of fields counted per patient is 16. Isotype control antibodies were done for each experiment. Significance was determined by a Wilcoxon test. Values of p<0.05 were considered statistically significant. **p<0. 01; ****p<0. 0001. TRM, resident memory T cells.

We measured the % interaction between CD103+CD8+T cells or CD103negCD8+T cells with E-cadherin using Akoya phenoptrReport software. We considered an interaction between E-cadherin and CD8+T cells when the distance between these cells was <15 µm. We observed that CD103+CD8+T cells interacted more strongly with E-cadherin than the CD103negCD8+T cells (online supplemental figure S8).

The TCF1+CD8+T cell population also infiltrated the tumor microenvironment in these patients, but these cells were only found in the stroma (mean: 11±15 cells/mm2) and not in the tumor nest. CD103+CD49a+CD8+ T cells were not found to express TCF1, but 13% of the CD103−CD49a+CD8+ T cells expressed TCF1 (figure 4B,C). TCF1 expression was mainly observed in the stromally localized populations but not in the epithelial tumor islets (figure 4B,C).

The resident memory CD103+CD49a+CD8+T cell population is the strongest predictor of clinical responses to anti-PD-1 immunotherapy

The prognostic impact of these different TRM subpopulations was then evaluated, taking into account their location within the tumor microenvironment. We also analyzed the TCF1-expressing T-cell population, which is considered to be the target of anti-PD-1 therapy, and whose pretreatment infiltration is the best predictor of anti-PD-1 response.28 30–32

Cox proportional-hazards univariate analysis showed that PD-L1 remained the most predictive biomarker of clinical response (HR=3.06 (0.002–6.31), p=0.002) in patients with NSCLC undergoing second-line treatment with anti-PD-1 (figure 5A). Interestingly, the population of TRM CD8+ T cells localized in the tumor co-expressing CD103 and CD49a markers was also a pretreatment feature correlated (HR=2.41 (1.26–4.62), p=0.008) with clinical response (figure 5A), but this association did not remain true in the stroma (HR=1.76 (0.092–3.39), p=0.09). Intratumoral infiltration by CD103+CD49a+CD8+T cells did not predict clinical response as defined by the response evaluation criteria in solid tumors (RECIST) criteria (data not shown).

Figure 5

Correlation between the infiltration of various subpopulations of CD8+T cells in the NSCLC tumor microenvironment on second line therapeutic and clinical outcomes. (A) Forest plot showing the HRs and 95% CIs computed using a univariate Cox model. The infiltration of several subsets of CD8+T cells and PD-L1 expression were quantified, using the median as a cut-off for dichotomization. Variables are ordered according to decreasing Wald statistic values. The sublocalization of these subpopulations in the stroma or the tumor or not (total) was taken into account. P value<0.05 was considered significant (in red). The HRs are calculated using the high group as a reference. A positive HR means that a high level of a measure is protective. (B) Kaplan-Meier curves corresponding to the overall survival of patients with NSCLC grouped according to tumorous or stromal infiltration by subpopulations of resident memory CD8+T cells or total TCF1+CD8+T cells and the expression of PD-L1 on tumor cells. Each variable was dichotomized separately based on the median value in order to define low and high groups. Log-rank test values are first displayed together with HRs, 95% CIs, and p values from the Wald test computed using a univariate Cox model. (C) Time-dependent receiver operating characteristic curves were used to analyze the true positive (TP) rate (sensitivity) and false positive (FP) rate (1-specificity) of the two subpopulations of resident memory CD8+ T cells, total CD8+T cells, TCF1+CD8+T cells, and PD-L1 when predicting 2-year overall survival. For each variable, only patients whose variable values were located in the extreme tertiles of the corresponding distribution were included. The resulting area under the curve values are shown. NSCLC, non-small cell lung cancer.

In the same analysis, total CD8+ T cells (HR=2.00 (1.04–3.84), p=0.037) and TCF1-expressing CD8+ T cells (HR=2.11 (1.09–4.06), p=0.025) also served as biomarkers associated with clinical response (figure 5A). In contrast, the resident memory CD103−CD49a+CD8+ T-cell population, did not predict clinical response to immunotherapy (figure 5A) irrespective of its stromal or tumorous location (HR=1.66 (0.87–3.17), p=0.12 and HR=1.70 (0.89–3.22), p=0.106, respectively)

These results were confirmed through Kaplan-Meier survival curve analyses and log-rank tests, confirming the relationships between survival and infiltration by CD103+CD49a+ CD8+T cells, total CD8 and stromal TCF1+CD8+T cells, well as the expression of PD-L1 by tumor cells (TCs) (figure 5B). In contrast, only PD-L1 expression (>1%) by TCs and infiltration in the stroma by TCF1+CD8+ T cells were correlated with progression-free survival (PFS) (HR=2.89 (1.52–5.5), p=0.01 and HR=1.84 (1.02–3.31), p=0.043) in these immunotherapy-treated patients (online supplemental figure S9A). This observation was made when comparing two groups of patients dichotomized using the median values for the parameters of interest. When we focused on extreme values using tertiles as cut-offs, the intratumoral CD103+CD49a+ CD8+ cell population appeared to be also prognostic variable for PFS (HR=2.22 (1.05–4.67), p=0.036) (online supplemental figure S9B).

Receiver operating characteristic (ROC) curve analyses of CD103+CD49a+CD8+T cell infiltration yielded an area under the curve (AUC) of 0.88 when predicting overall survival (OS) at 2 years, with this being the most robust predictor as compared with the other analyzed parameters (figure 5C). In a multivariate model, we found that the CD103+CD49a+CD8+ T-cell population remained predictive of survival when the model was adjusted for potential cofounders including PD-L1 expression and the infiltration of TCF1+CD8+T cells (table 1).

Table 1

Multivariable model adjusted for PD-L1 level and total stromal TCF-1+CD8+ T cells

To confirm these results in a second cohort, we selected patients with NSCLC with PD-L1 expression on >50% of TCs who underwent first-line anti-PD-1 treatment (n=30) or second-line treatment (n=6). We found that only intratumoral infiltration by CD8+ TRM cells co-expressing CD103 and CD49a was correlated with patient survival (HR=2.77 (1.13–6.75), p=0.025) (figure 6A), confirming the results obtained in the discovery cohort. Kaplan-Meier curves and log-rank tests (p=0.025) were used to analyze patients dichotomized based on median values (figure 6B). Tumor infiltration by the CD8+ TRM cell population co-expressing CD103 and CD49a was also the only population correlated with PFS when using this same approach to patient dichotomization (HR=2.62 (1.18–5.81), p=0.018) (figure 6C).

Figure 6

The clinical impact of the infiltration of various subpopulations of CD8+ T cells in the NSCLC tumor microenvironment in a validation cohort. (A) Forest plot representing Cox overall survival regression in patients with NSCLC (n=36). The infiltration of several subsets of CD8+T cells was quantified using the median as a cut-off. The sublocalization of these subpopulations in the stroma or the tumor or not (total) was taken into account. P value<0.05 was considered significant. (B) Kaplan-Meier analyses of the overall survival of patients with NSCLC depending on their level of intratumoral CD49a+CD103+CD8+ T cells dichotomized with the median. Statistical analyses were performed with the log-rank test. (C) Forest plot representing Cox progression-free survival regression in patients with NSCLC (n=36). The infiltration of several subsets of CD8+T cells was quantified using the median as a cut-off. The sublocalization of these subpopulations in the stroma or the tumor or not (total) was taken into account. P value<0.05 was considered significant. The HRs are calculated using the high group as a reference. A positive HR means that a high level of a measure is protective. P value is indicated in red. NSCLC, non-small cell lung cancer.