Programmed cell death 1 ligand 1 (PD-L1) has been used as a biomarker for immune checkpoint inhibitors (ICIs) which exert durable efficacy in non-small cell lung cancer (NSCLC).1,2 However, many PD-L1-high patients only marginally respond to, and PD-L1-low patients still benefit from, ICIs.3,4 Transcription factor aryl hydrocarbon receptor (AhR) plays critical roles in development and function of both innate and adaptive immune cells,5,6 controls the expression of PD-L1 in lung epithelial cells and is associated with patient response to ICIs.7 Here, we tested whether or not AhR and AhR nuclear translocator (ARNT) could enhance PD-L1 prognosis predicting value in NSCLCs upon ICI treatment.

Between July 2016 and November 2018, we retrospectively acquired 65 pre-ICI treatment formalin-fixed, paraffin-embedded (FFPE) specimens from NSCLC patients (Supplementary Table 1) who were then treated with pembrolizumab. In this training cohort, the proportion of cells with multiple positive staining, i.e., nucleus-localized AhR (AhRN), cytoplasmic AhR (AhRC), nucleus-localized ARNT (ARNTN), membrane-localized PD-L1 (PD-L1M) and nucleus-localized PD-L1N (Fig. 1a), was evaluated by multiplex immunohistochemistry (mIHC). The optimum density cutoff values of PD-L1M, AhRN, AhRNARNTN, AhRNPD-L1M (Supplementary Fig. 1a–d), AhRC, PD-L1N, ARNTN, AhRNPD-L1N, AhRCPD-L1M (Supplementary Fig. 2a–e), AhRCPD-L1N, AhRCARNTN, ARNTNPD-L1M, and ARNTNPD-L1N (Supplementary Fig. 3a–d) were determined based on progression-free survival (PFS) of the patients. A high proportion of PD-L1M+ cells indicated a better response to immunotherapy (Fig. 1b; HR: 0.226, 95% CI: 0.0759-0.675), consistent with previous studies.1 Notably, the densities of AhRN (HR: 0.139, 95% CI: 0.0319–0.605), AhRNARNTN (HR: 0.109, 95% CI: 0.0218–0.547) and AhRNPD-L1M (HR: 0.0831, 95% CI: 0.0166–0.417) (Supplementary Fig. 1b; Fig. 1b) effectively predicted patient PFS. Considering the ability to effectively distinguish responders, PD-L1M, AhRNPD-L1M, AhRNARNTN, AhRN, PD-L1N, ARNTNPD-L1N and AhRCPD-L1N were selected for further model construction. Due to the strong multicollinearity among the 7 variables (Supplementary Fig. 4a), we implemented LASSO logistic regression with λ = 0.04 (Supplementary Fig. 4b) to determine the variables most relevant to PFS. If the density of one variable exceeded its cutoff, the status of this variable was equivalent to 1; otherwise, the status was equivalent to 0. Five robust immunotherapeutic markers were identified (Supplementary Fig. 4c), and the formula that reflects the risk of NSCLC progression was risk score = -(1.919 × AhRNPD-L1M status)—(0.119 × AhRNARNTN status)—(0.696 × PD-L1M status) + (0.689 × ARNTNPD-L1N status) + (0.343 × AhRCPD-L1N status). The risk score plots suggested that AhRNPD-L1M, AhRNARNTN and PD-L1M were favorable prognostic factors, while ARNTNPD-L1N and AhRCPD-L1N were negative prognostic factors (Supplementary Fig. 4d).

AhRNPD-L1M is more precise than PD-L1M in predicting the prognosis of NSCLC patients treated with immune checkpoint inhibitors. a Representative mIHC images of NSCLC patients with differential responses to immunotherapy in the training cohort. PD, progression of disease. SD, stable disease. PR, partial response. CR, complete response. Scale bars = 100 μm. b The Kaplan-Meier survival curve of patient PFS corresponding the optimum density cutoff of PD-L1M, AhRNARNTN and AhRNPD-L1M. M, membrane-localized; N, nucleus-localized. c Representative mIHC images of NSCLC patients with differential responses to immunotherapy in the validation cohort 1. d The Kaplan-Meier survival curve of patient PFS in the validation cohort 1. e ROC curve based on PFS of all the 168 patients according to the status of AhRNPD-L1M, AhRNARNTN, and PD-L1M. f The Kaplan-Meier survival curve of PFS of all the 168 patients. g The Kaplan-Meier survival curve of OS of 37 patients who developed progression of disease to PD-1 blocker. h Multivariable Cox regression analyses of the significant variables selected by univariate Cox regression analyses of AhRNPD-L1M, AhRNARNTN, PD-L1M, clinicopathological characteristics, and patient survival. The error bars indicate 95% confidence interval (CI). i The expression level of AhR and PD-L1 in nuclear, cytoplasmic, and membranous compartments of tumor samples of NSCLC patients receiving Keytruda treatment, tested by western blot. j Immunohistochemistry (IHC) assays of AhR expression in tumor samples harvested from mice inoculated with indicated murine cancer cells. Scale bar = 50 μm. k The survival curve of LLC cells-bearing C57BL/6 mice, estimated by the Kaplan-Meier analysis and Log-rank test. l Tumor volume of MC38 cells-bearing C57BL/6 mice that were treated with or without anti-PD-L1 antibody. m Tumor volume of Ag104Ld cells-bearing B6C3F1 mice that were treated with or without anti-PD-L1 antibody. n IHC analysis of Ki67 in tumor samples of mice that were inoculated with indicated murine cancer cells and treated with or without anti-PD-1/anti-PD-L1 antibody. Scale bar = 50 μm. o Representative mIHC images of tumor tissues from mice that were inoculated with indicated murine cancer cells. The statistical significance was assessed by two-sided Student’s t test and P values of Kaplan-Meier survival analysis were calculated by log-rank test

After confirming that AhRNPD-L1M-, AhRNARNTN- and PD-L1M-positive cells were concentrated in responders (Supplementary Fig. 5a), the 3 factors were selected for further comparison. The distribution of all clinicopathological characteristics except smoking status showed no significant differences between the high- and low-proportion groups (Supplementary Table 2 and Supplementary Fig. 5b, c). Similarly, the OS probabilities of NSCLC patients receiving immunotherapy showed salient discrepancies between groups divided according to the cutoff values of AhRNPD-L1M (HR: 0.160, 95% CI: 0.0375–0.680), AhRNARNTN (HR: 0.182, 95% CI: 0.0418–0.794) and PD-L1M (HR: 0.401, 95% CI: 0.176–0.909) (Supplementary Fig. 6a). ROC analysis was carried out and the area under the receiver operating characteristic curve (AUC) indicated that AhRNPD-L1M, AhRNARNTN and PD-L1M could validly predict patient PFS (Supplementary Fig. 6b), and Delong test showed that AhRNPD-L1M was the most powerful one (Z = 2.444, P = 0.0145).

To verify our findings in training group, two validation cohorts were employed (Supplementary Tables 1, 3, 4). Similar to that of training cohort, the distribution of AhRNPD-L1M-, AhRNARNTN- and PD-L1M-positive cells (Fig. 1c) was related to smoking status (Supplementary Table 3 and 4, Supplementary Fig. 7a, b and Supplementary Fig. 8a, b). The factors divided patients into two group with significantly different PFS and OS, and AhRNPD-L1M was the most accurate one (Fig. 1d, Supplementary Fig. 7c and Supplementary Fig. 8c, d).

The three cohorts were pooled to test the predictive ability of these factors and the results showed that compared to AhRNARNTN (AUC: 0.764; 95% CI: 0.692–0.837) and PD-L1M (AUC: 0.702; 95% CI: 0.624-0.78), AhRNPD-L1M had the largest AUC (0.857; 95% CI: 0.799–0.915; Fig. 1e), and Delong test confirmed AhRNPD-L1M as the most accurate predictor (Z = 3.116, P = 0.00183). AhRNPD-L1M possessed the highest accuracy in predicting PFS (Fig. 1f) and OS (Supplementary Fig. 9), in short-term and long-term observations (Supplementary Fig. 10), and in the vast majority of cases when stratified by clinicopathological factors (Supplementary Fig. 11a). Net reclassification index (NRI) for PFS was calculated and the result was 0.297 (Z = 3.094, P = 0.00197), suggesting that the proportion of exact classification increased from PD-L1M to AhRNPD-L1M by 29.7%. NRI for OS was 0.245 (Z = 2.880, P = 0.00398), indicating that the proportion of exact classification increased by 24.5%. Among 37 patients who developed PD upon immunotherapy, higher level of AhRNPD-L1M was associated with longer OS (Fig. 1g). Univariate analysis showed that AhRNPD-L1M, AhRNARNTN and PD-L1M were associated with patient PFS and OS (Supplementary Fig. 11b). Multivariable Cox regression analysis showed that AhRNPD-L1M and PD-L1M were independent prognostic indicators for NSCLCs upon PD-1 blockade, and AhRNPD-L1M was more powerful (Fig. 1h).

To verify the findings by mIHC, we tested the expression of AhR and PD-L1 by western blot. In 15 patients who had enough frozen tumor samples, AhR was detected in nuclear and cytoplasmic compartments, while PD-L1 was detected in membrane, cytoplasm, and nucleus (Fig. 1i, Supplementary Fig. 12a), and the higher expression levels of AhRN, the more likely the patients would benefit from immunotherapy (P = 0.012). However, PD-L1M could not distinguish good response from poor response in these patients (Supplementary Fig. 12b).

We established xenograft murine models to validate the findings in patient samples, and found that tumor tissues of mice injected with murine lung cancer LLC and colon cancer MC38 cells exhibited high level AhRN, whereas murine fibrosarcoma Ag104Ld cells-injected mice had low level AhRN (Fig. 1j). The anti-PD-1/anti-PD-L1 antibody exerted an obvious antitumor effect on mice injected with LLC (Fig. 1k, Supplementary Fig. 13a) and MC38 cells (Fig. 1l, Supplementary Fig. 13b), but not on Ag104Ld-bearing mice (Fig. 1m, Supplementary Fig. 13c, d). The effects of PD-1/PD-L1 inhibition were confirmed by IHC analysis of cell proliferation marker Ki67 (Fig. 1n). mIHC staining of tumor specimens showed that LLC and MC38 cells expressed high level AhRN and moderate level PD-L1M, while Ag104Ld cells expressed high level PD-L1M and low level AhRN (Fig. 1o). Notably, the number of effector CD8+T cells was the highest in MC38 model, yet the lowest in Ag104Ld model (Fig. 1o). Western blot assays using nuclear, cytoplasmic and membranous proteins showed that AhR expression was high in the nucleus of MC38 but almost undetectable in Ag104Ld cells (Supplementary Fig. 14). PD-L1 was expressed on the membrane of the three cancer cells with the highest level in Ag104Ld cells (Supplementary Fig. 14).

To further dissect the underlying mechanisms, RNA-seq data of 255 lung adenocarcinoma and 251 lung squamous cell carcinoma samples were downloaded from The Cancer Genome Atlas (TCGA) database and were divided into AhR-high and AhR-low groups, using the median expression level as a cutoff value. ESTIMATE (https://bioinformatics.mdanderson.org/estimate/) was applied to assess the infiltration of immune cells in tumor samples. We found that AhRHigh samples had higher immune scores than AhRLow samples (Supplementary Fig. 15a). The Cancer Immunome Database (TCIA) (https://tcia.at/home) that provides comprehensive immunogenomic analysis results was applied, and we found that AhRHigh patients were more likely to benefit from inhibitors of CTLA-4, PD-1/PD-L1/PD-L2 (Supplementary Fig. 15b). We previously showed that AhR could bind PD-L1 promoter in cells exposed to tobacco carcinogen benzo(a)pyrene (BaP).7 In the absence of BaP, AhR overexpression increased while AhR suppression inhibited, PD-L1 promoter-driven luciferase activity (Supplementary Fig. 16a, b). By chromatin immunoprecipitation (ChIP) and real-time reverse transcription-polymerase chain reaction (RT-PCR), we found that AhR directly bound the promoter of PD-L1 at −700 to −100 bp upstream of its transcription start size (Supplementary Fig. 16c), confirming that AhR is able to control PD-L1 transcription. Together, the roles of AhR in regulating PD-L1 expression and immune cell infiltration may contribute to its significance in predicting response of NSCLCs to ICIs.

Efforts have been made to identify accurate biomarkers for prognostication of ICIs in treating cancers, but PD-L1,3,4 tumor mutation burden (TMB),8 or mismatch-repair status9 remain unsatisfactory, and the integrative method10 needs further investigation. Our findings here indicate that AhRNPD-L1M is an accurate, simple, and cheap biomarker for immunotherapy in treating NSCLCs.