Purpose: The success of immune checkpoint inhibitors strengthens the notion that tumor growth and regression are immune regulated. To determine whether distinct tissue immune microenvironments differentially affect clinical outcome in non–small cell lung cancer (NSCLC), an extended analysis of PD-L1 and tumor-infiltrating lymphocytes (TIL) was performed.
Experimental Design: Samples from resected adenocarcinoma (ADC 42), squamous cell carcinoma (SCC 58), and 26 advanced diseases (13 ADC and 13 SCC) treated with nivolumab were analyzed. PD-L1 expression and the incidence of CD3, CD8, CD4, PD-1, CD57, FOXP3, CD25, and Granzyme B TILs were immunohistochemically assessed.
Results: PD-L1 levels inversely correlated with N involvement, although they did not show a statistically significant prognostic value in resected patients. The incidence and phenotype of TILs differed in SCC versus ADC, in which EGFR and KRAS mutations conditioned a different frequency and tissue localization of lymphocytes. NSCLC resected patients with high CD8pos lymphocytes lacking PD-1 inhibitory receptor had a longer overall survival (OS: HR = 2.268; 95% CI, 1.056–4.871, P = 0.03). PD-1-to-CD8 ratio resulted in a prognostic factor both on univariate (HR = 1.952; 95% CI, 1.34–3.12, P = 0.001) and multivariate (HR = 1.943; 95% CI, 1.38–2.86, P = 0.009) analysis. Moreover, low PD-1 incidence among CD8pos cells was a distinctive feature of nivolumab-treated patients, showing clinical benefit with a prolonged progression-free survival (PFS: HR = 4.51; 95% CI, 1.45–13.94, P = 0.004).
Conclusions: In the presence of intrinsic variability in PD-L1 expression, the reservoir of PD-1–negative effector T lymphocytes provides an immune-privileged microenvironment with a positive impact on survival of patients with resected disease and response to immunotherapy in advanced NSCLC. Clin Cancer Res; 24(2); 407–19. ©2017 AACR.
The wave of enthusiasm brought about by the success of PD-1/PD-L1 immune checkpoint inhibitors (ICI) in NSCLC points toward the emerging need to determine the clinical impact of distinct immune microenvironments. Tumor-infiltrating lymphocytes (TIL) critically contribute to tumor biology and response to ICI. Thus, in addition to the evaluation of PD-L1, predictive biomarkers likely require the simultaneous assessment of phenotypes relevant to PD-1/PD-L1 immune checkpoint. The inhibitory receptor PD-1, being also a therapeutic target, was a candidate to provide significant insights in NSCLC immune landscape and response to immunotherapy. Our data suggest that, in the presence of a variable degree of PD-L1 expression, an intrinsic reduction of PD-1pos TILs is associated with delayed NSCLC progression, improved OS, and response to ICI. Based on our findings, the population of cancer patients who benefit from actual and innovative immunotherapeutic approaches may be expanded following a detailed characterization of the tumor immune microenvironment.
The central role of the immune system in the control of tumor growth and progression has been repeatedly demonstrated (1, 2). Accordingly, recently introduced molecules targeting immune checkpoints have received ground-breaking achievements in the treatment of several solid tumors, including non–small cell lung cancer (NSCLC; refs. 3–6). At variance with therapeutic approaches directed against neoplastic cells, such as chemotherapy or small molecule tyrosine kinase inhibitors (7, 8), targeting the interactions between programmed cell death-1 (PD-1), a T-cell coinhibitory receptor, and one of its ligands, programmed cell death-ligand 1 (PD-L1), has shown unprecedented success (9, 10).
Although the expression of PD-L1 by tumor cells potentially identifies NSCLC patients who would benefit from immune checkpoint inhibitors (ICI), it does not consistently represent a reproducible predictive biomarker of the clinical response (11–13). The attempt to correlate PD-L1 expression and patient prognosis has shown conflicting results (14–16). Extensively debated issues on antibody specificity and different methodologies (17, 18) have been partially overcome by several initiatives of harmonization (19–21). However, the variability of PD-L1 expression among and within NSCLC (17) is intrinsic to several biologic and genetic events implicated in tumor immunogenic properties (4, 14). In this regard, neoantigens generation (22, 23) and timing of the natural immune history largely vary among patients and markedly impact on PD-L1 expression.
In addition, other components of the tumor microenvironment, such as tumor-infiltrating lymphocytes (TIL), critically contribute to tumor biology and therapeutic response to ICI (24), suggesting that predictive biomarkers aimed at personalized therapy can be obtained by a coordinated multiparametric definition of the immune contexture (24–27).
Specifically, the balance between PD-L1–mediated tumor immune escape and the efficient immune reaction against cancer by TILs may provide more relevant prognostic and therapeutic tools.
According to PD-L1 status and number of TILs, a classification of tumors into four categories has been recently proposed (28, 29). This includes type I adaptive immune resistance (PD-L1 positive and high TILs), type II immune ignorance (PD-L1 negative and low TILs), type III intrinsic induction (PD-L1 positive and low TILs), and type IV immune tolerance (PD-L1 negative and high TILs). However, a detailed tissue characterization of these different aspects of the immune mediated tumor growth and regression has been only partially investigated in NSCLC.
The present work was addressed to determine whether immunologically defined NSCLC tissue microenvironments might influence the clinical outcome and response to ICI. Thus, 100 newly diagnosed and surgically removed tumors and 26 samples from patients with advanced NSCLC candidate to receive anti–PD-1 drug (nivolumab) were morphometrically analyzed to assess the number and distribution of TILs subpopulations together with the immunohistochemical evaluation of PD-L1 expression. The hypothesis was advanced on whether and to which extent the expression of PD-1 inhibitory receptor on immune relevant cells exerts a prognostic and predictive effect. Thus, the potential association with clinical data of individual and integrated tissue parameters involved in PD-1/PD-L1 immune checkpoint was investigated in both groups of patients.
Materials and Methods
Patient population and tissue sampling
This study involved 100 NSCLC patients (58 SCC and 42 ADC, stage I/II 72%) undergoing lung resections with curative intent at the Unit of Thoracic Surgery, University-Hospital of Parma, and 26 patients (13 ADC and 13 SCC) with stage IIIb to IV treated with nivolumab in second or third-line therapy. The investigation was approved by the institutional review board for human studies (Ethical Committee) of the University-Hospital of Parma and in accord with principles listed in the Helsinki declaration. Patients were enrolled after informed consent to the employment of biologic samples for research purpose. The surgically resected patient population (Table 1) consisted of 74 males and 26 females, ages between 45 and 84 years. The population of patients treated with nivolumab included 25 males and 1 female (49–85 years old). Tumor stage was scored using the staging system from the 7th American Joint Committee on Cancer (AJCC; ref. 30). Smoking history, histological diagnosis, pathological staging, and mutational status are reported in Table 1. Eight patients from the surgically resected cohort and 3 from nivolumab treated group were excluded from the statistical analysis due to the lack of complete clinical records. Additionally, eight surgically resected patients were excluded due the inadequacy of tissue samples precluding a complete immunohistochemical analysis. Thus, clinicopathologic correlations were allowed in 84 surgically resected and 23 advanced NSCLC cases.
The quantification of PD-L1 expression was assessed after a comparative evaluation between immunoperoxidase and immunofluorescence on control tissues (placenta) and on serial sections from the same case using different anti–PD-L1 antibodies (Extended data in Supplementary Table S1). After careful evaluation and consultation with expert pathologists, reliable and consistent results were obtained using clone 28-8 that was used for further quantitative analysis.
Tissue sections (5 μm thick) were cut from formalin-fixed, paraffin-embedded blocks containing representative tumors and processed for immunohistochemistry (IHC).
Immunoperoxidase: Following the incubation with anti–PD-L1 antibody (clone 28-8), the reaction was revealed by DAB staining and nuclei were counterstained by light hematoxylin. In accordance with a reported approach (31), an algorithm was used to obtain the PD-L1 score (H-score: 0–300), which is computed on the basis of both extent and intensity of PD-L1 staining.
Confocal Immunofluorescence: Sections were double labeled by anti–PD-L1 (clone 28-8) and Pan-Cytokeratin antibody (clone AE1/AE3) followed by FITC- and TRITC-conjugated specific secondary antibody, respectively. Images were digitally captured by confocal microscopy at 200× final magnification using a Zeiss LSM 700 system with Axio Observer. The fractional area occupied by the fluorescent signal and its intensity, expressed as integrated optical density (IOD) per unit area, was then evaluated using a software for image analysis (Image Pro Plus 4.0). Scores were normalized to the exposure time and bit depth at which the images were captured, allowing scores collected at different exposure times to be comparable. Nuclei were counterstained with 4′,6-Diamidine-2′-phenylindole (DAPI).
The antibodies used and magnitude of sampling is reported in Supplementary Tables S1 and S2.
Tumor-infiltrating lymphocytes (TIL) were analyzed by the immunohistochemical detection of CD3, CD8, CD4, PD-1, CD25, CD57, and Granzyme B. Immunoperoxidase was performed using an automated staining system (OptiView DAB IHC Detection Kit—Ventana Medical Systems). T lymphocyte subpopulations were also studied by immunofluorescence in order to simultaneously evaluate multiple epitopes. Sections of NSCLC were incubated with primary antibodies followed by specific secondary antibodies. Nuclei were counterstained with DAPI. The list of antibodies used for TILs immunophenotyping is provided in Supplementary Table S1.
TILs were quantified according to specific recommendations (32), and their number and distribution within the neoplastic tissue were assessed following morphometric criteria detailed in Supplementary Materials. Based on their relative incidence, implemented by double immunofluorescence confocal analysis, the ratio of CD8pos cells among the whole population of T (CD3pos) lymphocytes (CD8-to-CD3) and the fraction of PD-1 expressing cells over the cytotoxic (CD8pos) lymphocytes (PD-1-to-CD8) was computed.
The magnitude of sampling for TILs analysis is reported in Supplementary Table S2.
This analysis involved 27 surgically resected and 11 advanced ADC. In each case, 4-μm-thick sections were obtained from a representative block. Clone D5F3 (Ventana; prediluted) was used to evaluate ALK expression, as previously suggested (33). More details are reported in Supplementary Materials and Methods.
EGFR and KRAS mutational analysis
EGFR and KRAS mutations were tested in 36 and 27 surgically resected and 13 and 13 advanced ADC cases, respectively. The methodology has been previously described (33) and is detailed in Supplementary Materials and Methods.
The χ2 test was used to test the differences in categorical variables, and the Wilcoxon rank-sum test to detect differences in continuous variables between groups of patients, given that the distribution of data was not normal (Kolmogorov–Smirnov test). The disease-free survival (DFS), overall survival (OS), and progression-free survival (PFS) were estimated by means of the Kaplan–Meier method. DFS was defined as the interval from surgery to the evidence of recurrence disease, or death, or the last date the patient was known to be recurrence-free or alive; OS was defined as the interval from surgery or start of treatment to death from any cause, or the last date the patient was known to be alive; PFS was defined as the interval from start of treatment to the evidence of progressive disease, or death, or the last date the patient was known to be progression-free or alive. Both DFS and OS data were censored at 5 years. Patients treated with nivolumab were categorized in clinical benefit group, including patients with complete (CR) or partial response (PR) or stable disease (SD) lasting at least 6 months according to RECIST criteria version 1.1 (34) and nonresponders (patients with disease progression and stable disease lasting less than 6 months as best response according to RECIST criteria version 1.1, ref. 34).
Log-rank test was performed to determine the difference in survival between groups. DFS and OS data were then analyzed through multivariate models of Cox regression.
Classification and regression tree (CART) analysis identified specific cutoff values that segregated patients by clinical outcomes. P value of 0.05 was always considered significant. IBM SPSS Statistics v 24.0 (IBM) and Stata 13 with Cart module (Statacorp) were used to perform the computations for all analyses.
Immune microenvironment in resected patients
Immunohistochemical assessment of PD-L1 expression.
This analysis was carried out on sections obtained from 92 resected NSCLC. The variability among and within cases in both membrane and cytoplasmic PD-L1 expression (Fig. 1A and B) required an extended sampling to assess the validity of our measurements (Supplementary Table S2).
Based on the extent and intensity of the immune reactivity of PD-L1 in tumors, the H-score was computed on immunoperoxidase-stained sections and expressed as IOD score for quantitative immunofluorescence (QIF; Fig. 1C,i, Cii). A comparative evaluation of PD-L1 expression obtained by IF and IP was performed on serial sections of the same case, documenting that the H-score gradient of intensity measured by IP correlated with IOD values of PD-L1 immunofluorescent signals (Fig. 1C iii). Thus, H-score or QIF and their average values were interchangeably used to document the impact of PD-L1 expression on other tested parameters. QIF values were preferentially used when PD-L1 and TILs subpopulations were concurrently considered.
Although a great heterogeneity among and within lung cancer samples (Fig. 1E–G) was observed, average H-score was higher in SCC than in ADC samples (Fig. 1C,ii) and more than 60% of ADC displayed values below the overall average compared with only 33% of SCC. A significant difference in PD-L1 expression between the two histotypes was confirmed by QIF. SCC specimens exhibited a 2.5-fold higher average PD-L1 value than ADC cases (P < 0.05; Fig. 1C i).
Specific PD-L1 immunostaining in stromal compartments was also present and involved nearly 10% and 40% of ADC and SCC samples, respectively. However, the significance of PD-L1 in nonneoplastic cells compartments of NSCLC requires a separate study and can only be assessed by an extensive multiple immunofluorescence staining in order to attribute its expression to specific phenotypes.
PD-L1 expression and clinicopathologic features.
PD-L1 levels were significantly higher in smokers than in never-smokers and ex-smokers (P < 0.001; Supplementary Fig. S1A). In addition, we observed an inverse correlation between PD-L1 expression and N status, as N0 tumors had a 2-fold higher (P < 0.001) PD-L1 score compared with N2 (Supplementary Fig. S1B).
At variance with the literature (26, 28), we did not detect a strong association of PD-L1 expression with EGFR mutation (n 7/36), while KRAS mutated cases (n 5/27) compared with WT displayed lower PD-L1 levels (P < 0.001) as measured both by H-score and QIF (Supplementary Fig. S1C).
In our patient population, ALK rearrangement (n = 27) was undetectable.
We did not observe statistically significant correlations between PD-L1 expression and other clinical and pathologic features in both NSCLC histotypes.
Thus, tumor PD-L1 values above the average correlated with smoking history and nodal involvement in resected NSCLC patients.
Immunohistochemical analysis of TIL subpopulations.
An extensive IHC examination of TILs was carried out in the attempt to assess the contribution of relevant cells in cancer immune landscape, including cytotoxic, helper, and regulatory T cells, and natural killer cells.
The density of lymphocyte subpopulations at tissue level largely reflected the expected incidence according to their phenotype in both NSCLC subsets (Fig. 2A–C); however, a nearly 2-fold increase in CD3pos and a 3-fold reduction in CD4pos TILs were measured in ADC compared with SCC (Fig. 2D, P < 0.05). Conversely, no difference was observed in CD8pos lymphocytes density between the two NSCLC histotypes. CD57pos cytotoxic and CD25/FOXP3pos T regulatory cells constituted the lowest subpopulation (Fig. 2E–G) reaching, at most, 2% of TILs. However, compared with ADC, we documented a significant 3-fold increase in CD57pos cells (P < 0.01) and a slight increase in CD25/FOXP3pos cells (P < 0.05) in SCC cases (Fig. 2H).
According to the different location within the tumor microenvironment, distal PD-1pos TILs, largely represented at the invasive margin and within germinal centers of tertiary lymphoid structures (TLS; Fig. 2J), appeared to be prominent in SCC cases. Indeed, a significant 1.5-fold increase in the overall number of PD-1pos lymphocytes was detected in SCC samples compared with ADC (Fig. 2D, P < 0.05). In contrast, the quantity of PD-1pos lymphocytes distributed in intratumor and peritumor (proximal) location, which is expected to be more directly engaged by PD-L1 expressing neoplastic cells, did not significantly vary between the two histotypes (Fig. 2K). Additionally, increasing levels of PD-L1 were coupled by higher frequency of cytotoxic TILs, although this correlation did not reach statistical significance (Supplementary Fig. S2).
Thus, the magnitude and immunophenotype of TILs diverge among NSCLC histotypes, although the fraction of lymphocytes carrying the inhibitory receptor PD-1 located in close contact with neoplastic cells was similar in ADC and SCC samples.
TILs and clinicopathologic features.
A statistically significant increase (P < 0.001) in CD8pos lymphocytes was measured in T1 compared with T2 NSCLC patients (Supplementary Fig. S3A). In addition, T1 samples showed a consistent reduction in PD-1pos lymphocytes (Supplementary Fig. S3B P < 0.001) with respect to T2 and T3. Accordingly, we documented a significant decrease of PD-1pos lymphocytes in N0 cases compared with N1 (P < 0.001) and N2 (P < 0.001; Supplementary Fig. S3C). As a result of these findings, PD-1pos lymphocytes were 2-fold lower in stage I NSCLC compared with stage II (P < 0.001) and III (P < 0.001; Supplementary Fig. S3D), and ADC samples displayed an increasing PD-1 gradient (P < 0.001) according to the severity of the disease (Supplementary Fig. S3E). When the smoking history was considered, in SCC cases, smokers showed higher average values (P < 0.01) of PD-1pos lymphocytes (mean ± SE: 53.7 ± 8.8) compared with ex-smokers (mean ± SE: 38.1 ± 11.35).
No significant association of TILs values with demographic parameters was observed. However, EGFR and KRAS mutations appeared to condition a significant reduction of CD8pos and PD-1pos cells and a different tissue localization of PD-1pos lymphocytes resulting in lower intratumor density compared with WT cases (P < 0.05; Supplementary Fig. S3F). No differences were observed in the amount of PD-1pos cells located at the invasive margin according to the mutational status.
Thus, a strong association of TILs number and PD-1 distribution with clinicopathologic characteristics was documented in our cohort of resected NSCLC, supporting their clinical relevance.
The landscape of NSCLC immune microenvironment.
We reasoned that a comprehensive definition of NSCLC immune microenvironment would have enhanced the biologic and clinical significance of our findings.
Thus, a more complex analysis of the tumor microenvironment was conducted considering the PD-L1 status together with the magnitude and phenotype of TILs involvement. Based on a well-described literature (29) and a recently proposed classification (28), we divided our cases into four categories: type I adaptive immune resistance, type II immune ignorance, type III intrinsic induction, and type IV immune tolerance. By this approach, we documented that more than one third of NSCLC samples displayed a type II contexture, likely reflecting immune exhaustion at the time of diagnosis, while a similar proportion of type III and IV was observed (Supplementary Fig. S4A and S4B). The impact on clinical outcome of this more coordinate contribution of both tumor and TILs was apparent. NSCLC with high PD-L1 score and low TILs (type III) had the longest OS (75th percentile of survival time = 37 ± 16.22 months) and DFS (75th percentile of survival time = 11 ± 3.45 months) while type II immune ignorant cases were associated with worst prognosis (OS: 75th percentile of survival time = 16 ± 6.21 months; DFS: 75th percentile of survival time = 8 ± 2.03 months).
Thus, the overall estimation of PD-L1 in cancer cells and TILs allows the identification of immune tissue characteristics with impact on clinical parameters. However, the low statistical power documented by the Kaplan–Meier curve (Supplementary Fig. S4C), of such an unrefined classification of NSCLC environments prompted us to test whether individual or, more likely, a combination of multiple tissue parameters would have significant prognostic and predictive value in surgically resected patients.
Tissue immunophenotypic signatures with clinical impact
PD-L1 expression did not show a statistical significant impact on survival at univariate analysis (Table 2), although the percentage of survivors was higher (OS: 77%; DFS: 27%) in NSCLC patients with PD-L1 values above the average compared with PD-L1 low group (OS: 58%; DFS: 18%; Supplementary Table S4).
On univariate analysis also, TIL density did not show a statistical significant impact on survival (Table 2). However, higher percentage of survivors were documented in ADC samples rich in CD3pos (OS: 65%; DFS: 37%) and in CD8pos (OS: 65%; DFS: 21%) lymphocytes with respect to CD3, poor (OS: 45%; DFS: 3%) and CD8, poor (OS: 43%; DFS: 9%) cases (Supplementary Table S4).
Similarly, in the overall population of NSCLC patients low tissue density of PD-1pos lymphocytes was associated to a higher fraction of disease free survivors (DFS: 22%) compared with the PD-1 high counterpart (DFS: 0%; Supplementary Table S4).
Conversely, more defined immunophenotypic features appeared to strongly predict the clinical outcome on both univariate and multivariate analysis. Specifically, the proportion of CD8pos cells among the overall population of CD3pos TILs and the fraction of cytotoxic lymphocytes lacking the inhibitory receptor PD-1 were candidate to better represent the population of antitumor effector lymphocytes.
As shown in Table 2, high CD8-to-CD3 ratio was positively correlated with OS and DFS (OS: HR = 0.965, 95% CI, 0.931–1.001, P = 0.057; DSF: HR = 0.954, 95% CI, 0.965–0.983, P = 0.002) on univariate analysis, although not reaching statistical significance on multivariate analysis. The Kaplan–Meier curve documented longer OS in patients with high CD8-to-CD3 ratio (Fig. 3C; HR: 0.542, 95% CI, 0.260–1.133, P = 0.096).
The impact of PD-1-to-CD8 ratio on clinical outcome was striking. This parameter resulted an independent predictor of OS in both univariate (OS: HR = 1.952, 95% CI, 1.34–3.12, P = 0.001) and multivariate analysis (OS: HR = 1.943, 95% CI, 1.38–2.86, P = 0.009) while its impact on DFS was seen only on univariate test (DSF: HR = 1.537, 95% CI, 1.12–2.10, P = 0.007). Importantly, the Kaplan–Meier curve showed longer OS (HR: 2.268, 95% CI, 1.056–4.871, P = 0.03) in patients with low PD-1-to-CD8 ratio (Fig. 3D).
Clinicopathologic features such as age and histotype also had, to some extent, statistical significant impact on survival (Table 2). On univariate analysis, only the N status significantly correlated with OS and DFS (OS: HR = 1.87, 95% CI, 1.2–2.91, P = 0.004; DFS: HR = 1.524, 95% CI, 1.057–2.195, P = 0.02) while on multivariate model statistical significance was maintained only on OS (OS: HR = 1.86, 95% CI, 1.31–3.96, P = 0.015; Table 2). Based on this finding, patients were clustered according to the combination of the two most powerful statistical parameters detected in our patient population of NSCLC. As shown by the Kaplan–Meier curve (Fig. 3F), cases carrying N 0 tumors and low tissue PD-1-to-CD8 ratio had a statistically significant longer OS (HR: 1.624, 95% CI, 1.148–2.296, P = 0.006) compared with other parametric combinations. Importantly, worst prognosis was seen in patients whose samples displayed high number of CD8pos lymphocytes expressing PD-1, independently from the N status. Thus, the tissue content of cytotoxic TILs carrying the inhibitory receptor PD-1 appeared to limit the prognostic value of N status (Fig. 3F).
Immune microenvironment in patients treated with nivolumab
In the attempt to define not only the prognostic role, but also the potential predictive value of the immune microenvironment on the response to ICI, a similar analysis was performed on a selected group of 26 NSCLC patients receiving anti PD-1 (nivolumab) treatment (Fig. 4A–C). Nine patients (35%) were included in the clinical benefit group (CR and PR, an SD lasting 6 months at least) according to RECIST criteria (34), while the remaining 17 patients had progression of disease at first tumor assessment.
PD-L1 expression was high in 55% of patients within the clinical benefit group versus 42.5% in nonresponders (P = 0.225).
In strong agreement with our previous findings, 100% of nivolumab treated patients with clinical benefit displayed low PD-1-to-CD8 ratio compared with 21.4% of nonresponders (P < 0.001; Fig. 4D). High CD8-to-CD3 ratio was present in 55.6% of patients within the clinical benefit group and in 42.8% of nonresponders (P = 0.098). In addition, 66.6% of cases belonging to the clinical benefit group showed low number of PD-1pos lymphocytes compared with 64.2% of nonresponders (P = 0.74). Importantly, although not reaching a statistically significant impact on OS (Fig. 4G and H), we documented a longer PFS (HR = 2.728, 95% CI, 0.967–7.697, P = 0.047) in cases with high CD8-to-CD3 ratio (median PFS = 4.24 months) versus low (median PFS = 1.81 months) and a significant prolonged PFS (HR = 4.51, 95% CI, 1.459–13.944, P = 0.004) in patients with low (median PFS = 12.96 months) versus high (median PFS = 1.84 months) PD-1-to-CD8 ratio (Fig. 4E and F).
Our data suggest that the amount of residual PD-1–negative cytotoxic effectors TILs may be considered a potential predictive factor in NSCLC patients treated with nivolumab.
Blockade of inhibitory immune checkpoints is currently arising as a promising option of anticancer strategies. Targeting the PD-1/PD-L1 pathway has shown encouraging results on various malignancies (9, 35–37) and has changed the standard of care in first- and second-line therapy of advanced NSCLC (5, 6, 10–12). In this clinical setting, reliable predictive biomarkers are the object of significant financial investments although are difficult to obtain due to the complex tumor–host relationship. The actual assessment of risk stratification according to clinicopathologic (38) and biomolecular (39) parameters although showing relevant prognostic impact in NSCLC could be integrated with immunologically defined factors. It is also increasingly clear that the development of biomarkers for anti–PD-1/PD-L1 therapy requires a better understanding of PD-L1 expression profile and its interaction with TILs. It cannot be underestimated that these biomarker-oriented efforts will provide important insights into the pathogenesis and progression of NSCLC.
In order to prospectively evaluate the prognostic impact of different subpopulations of TILs and PD-L1 expression, we performed a retrospective systematic tissue analysis on a well-defined cohort of 100 stage I to IIIa NSCLC patients undergoing primary tumor resection without previous neoadjuvant therapy and 26 advanced NSCLC patients treated with nivolumab.
Although the original aim of our study was not to deliver a novel translatable biomarker, the simultaneous assessment of the prominent actors of NSCLC immune editing, surveillance, and escape allowed us to identify immune tissue contexts with differential impact on clinical outcome. Specifically, we documented that the reservoir of an efficient population of cytotoxic cells lacking PD-1 receptor is a potential independent prognostic and predictive variable in NSCLC. Thus, low PD-1-to-CD8 ratio resulted in longer OS and DFS in resected patients as was able to predict tumor response to nivolumab. These findings are in agreement with the expected positive clinical impact of a locally active TIL population tackling on cancer growth and progression. Interestingly, our data are in accordance with the recent identification in human and experimental NSCLC (40), and in melanoma patients (41) of a tissue and circulating subpopulation of exhausted lymphocytes that can be rescued by anti–PD-1-targeted drugs. These CD8pos cells expressing the costimulatory molecule CD28 undergo active replication (Ki67pos) following anti–PD-1 blockade. Intriguingly, the immunophenotype of these cells also involves PD-1, pointing to the dual activating and inhibitory function of the receptor (40). Importantly, the presence of a similar lymphocyte subpopulation within the neoplastic tissue was documented in NSCLC and melanoma patients by immunophenotyping (41) and clonal analysis of T-cell receptor (TCR) repertoire (42), respectively. Our immunohistochemical approach makes it difficult to dissect whether PD-1 carrying lymphocytes predominantly undergo anergy or shift toward an activating pathway. However, based on our multiparametric clinicopathologic correlation, it is likely that PD-1 expressing TILs were mostly directed to a PD-L1–mediated inhibitory fate. Overall, these findings underline the need to define which subpopulation of TILs expresses PD-1 and whether and to which extent PD-1 expression reflects activation or inhibition. Finally, we are encouraged to support PD-1-to-CD8 ratio as a potential biomarker by the first approved clinical trial applying the CRISPRcas technology in lung cancer, in which PD-1 inhibitory receptor has been genetically disrupted in adoptive cell transfer (43).
In addition, specific subpopulations of TILs strongly correlated with clinicopathologic parameters. Thus, T1 tumors had high CD8 and low PD-1 content and N0 cases were characterized by low number of tissue PD-1pos cells. Taken together, an inverse correlation between the number of PD- 1pos TILs and NSCLC clinical staging was observed (Supplementary Fig. S3). In our cohort of patients, the mutational status, including EGFR and KRAS mutation, had an impact on TILs. Specifically, EGFR mutation appeared to condition a significant reduction in CD8pos and PD-1pos cells and a different tissue distribution of PD-1pos lymphocytes resulting in lower intratumor density compared with WT cases. In agreement with the literature (44), we documented a reduction in the numerical incidence of TILs carrying PD-1 receptor in patients with KRAS mutation compared with WT. The lack of ALK rearrangement in our patient population does not allow to discuss potential correlations with the immunophenotypic changes of TILs.
Our data are in line with the general view that high PD-L1 expression conditions a better outcome in NSCLC (18). In this regard, although PD-L1 did not possess a strong statistical power as an individual prognostic factor on univariate analysis, higher levels of expression were associated with prolonged OS and DFS. Moreover, a significant correlation between high PD-L1 tumor content and low N status was documented in our cohort of patients.
A great inter- and intrapatient variability of PD-L1 levels was confirmed by our immunohistochemical analysis, suggesting the involvement of several factors in shaping the expression of this protein. Inflammatory (45) and genetic (23, 24) backgrounds are known to condition tumor PD-L1 expression. In this regard, in our NSCLC samples, higher levels of PD-L1 were detected in the presence of an active cytotoxic TILs population as documented by a predominant CD8 phenotype (Supplementary Fig. S2). These cells are likely responsible for IFNγ secretion, the most powerful PD-L1 stimulating factor (45). Secondly, KRAS mutation was associated with a reduced expression of PD-L1, as previously documented (44), while, at variance with the literature (26, 46), EGFR mutation did not statistically impact on this parameter. It has been extensively reported that tumors with a greater mutational load could generate more neoantigens leading to a larger repertoire of tumor-specific T cells (41). However, epigenetic mechanisms leading to elevated expression of nonmutated self-antigens could confer high tumor antigenicity despite a low number of mutations (4). Accordingly, recent experimental observations and ongoing clinical trials have been addressed to determine whether the combination of two ICI or the association of ICI with chemotherapy or vaccination (47, 48) are able to induce a specific reinvigoration of T CD8pos lymphocytes. Finally, we observed morphologically recognizable neoplastic areas showing low or absent epithelial markers in the presence of high level of PD-L1 expression (Fig. 1E and G). These findings suggest additional roles of PD-L1 in tumor biology, including epithelial to mesenchymal transition (EMT; ref. 49).
The present investigation highlights the notion that ADC and SCC have a different biologic and tissue background (28, 50), including the immune context. In this regard, we documented that 36% of SCC samples displayed type III intrinsic induction, while in ADC the proportion of type IV and II was consistently higher (32% and 36%, respectively). Additionally, ADC displayed a lower expression of PD-L1 while, in line with other observations (26), the microenvironment seemed to be more chemoattractive for the overall population of T lymphocytes than in SCC. Accordingly, the clinical outcome of our cohort of surgically resected patients consistently varied between the two histotypes in terms of DFS that was considerably higher in SCC compared with ADC (28 months vs. 13 months, P < 0.001).
Some limitations of our study have to be acknowledged. First, the early stage of the disease of our patient population may restrain the wide application of our findings to the prevalent cohort of unresectable and more advanced NSCLC. Although the overall number of cases studied is relatively limited, potentially attenuating our conclusive remarks, patients had an adequate follow up (60 months) since only 10% were collected in 2015. We have to point out that the lack of functional test and an incomplete analysis of other relevant immune cells, as tumor associated macrophages and dendritic cells, may reduce the claim of fully defined immune contexts.
Thus, our efforts are presently addressed to extend the functional characterization of the tumor microenvironment and the role of myeloid derived suppressors cells (MDSC) whose analysis, conversely, represents a challenge due to their complex phenotype. Additionally, to strengthen the predictive value of our proposed biomarker, further analysis in a larger population of patients treated with ICI is required.
As a final consideration, we believe that an accurate pathological evaluation is still a fundamental tool to interpret molecular, cellular and tissue analysis of the immune reaction against tumor. Therefore, an extended and integrated approach as described here to define multiple variables contributing to cancer microenvironment is necessary in the current stage of our understanding of PD-1/PD-L1 immune checkpoint and its therapeutic targeting.
In the context of PD-L1 activation, a high fraction of cytotoxic CD8 lymphocytes lacking the inhibitory receptor PD-1 is a potential positive prognostic and predictive factor in NSCLC. Importantly, this population of effector cells can be fostered by PD-1 targeting drugs. Our study also raises the question whether the current indications for ICI treatment may be expanded, opening the possibility to include earlier stages of the disease. Indeed, neoadjuvant and adjuvant ICI trials are undergoing with the expectation that precocious interventions may restore an active surveillance against cancer, preventing the natural switch of NSCLC immune microenvironment to a silent unfavorable fate.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: G. Mazzaschi, G. Bocchialini, F. Aversa, F. Quaini, M. Tiseo
Development of methodology: D. Madeddu, A. Falco, C.A. Lagrasta, K. Urbanek
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Mazzaschi, D. Madeddu, A. Falco, G. Bocchialini, F. Sogni, G. Armani, C.A. Lagrasta, B. Lorusso, C. Mangiaracina, R. Vilella, C. Frati, L. Ampollini, M. Veneziani, K. Urbanek, F. Quaini, M. Tiseo
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Mazzaschi, M. Goldoni, C.A. Lagrasta, L. Ampollini, F. Quaini, M. Tiseo
Writing, review, and/or revision of the manuscript: G. Mazzaschi, D. Madeddu, R. Alfieri, F. Aversa, F. Quaini, M. Tiseo
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Mazzaschi, D. Madeddu, C.A. Lagrasta, E.M. Silini
Study supervision: G. Mazzaschi, C.A. Lagrasta, E.M. Silini, A. Ardizzoni, F. Aversa, F. Quaini
The authors gratefully thank Emilia Corradini, Gabriella Becchi, and Nicoletta Campanini for their invaluable technical assistance. This work was supported by Associazione Italiana per la Ricerca sul Cancro (AIRC) grants IG 14214 to M. Tiseo and IG 2016 Id 19026 to A. Ardizzoni.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.