Functional Engagement of the PD-1/PD-L1 Complex But Not PD-L1 Expression Is Highly Predictive of Patient Response to Immunotherapy in Non-Small-Cell Lung Cancer

Abstract

Purpose: In many cancers, the expression of immunomodulatory ligands leads to immunoevasion, as exemplified by the interaction of PD-L1 with PD-1 on tumor-infiltrating lymphocytes. Profound advances in cancer treatments have come with the advent of immunotherapies directed at blocking these immuno-suppressive ligand-receptor interactions. However, although there has been success in the use of these immune checkpoint interventions, correct patient stratification for these therapies has been challenging.

Materials and methods: To address this issue of patient stratification, we have quantified the intercellular PD-1/PD-L1 interaction in formalin-fixed paraffin-embedded tumor samples from patients with non-small cell lung carcinoma, using a high-throughput automated quantitative imaging platform (quantitative functional proteomics [QF-Pro]).

Results: The multisite blinded analysis across a cohort of 188 immune checkpoint inhibitor-treated patients demonstrated the intra- and intertumoral heterogeneity of PD-1/PD-L1 immune checkpoint engagement and notably showed no correlation between the extent of PD-1/PD-L1 interaction and PD-L1 expression. Importantly, PD-L1 expression scores used clinically to stratify patients correlated poorly with overall survival; by contrast, patients showing a high PD-1/PD-L1 interaction had significantly better responses to anti-PD-1/PD-L1 treatments, as evidenced by increased overall survival. This relationship was particularly strong in the setting of first-line treatments.

Conclusion: The functional readout of PD-1/PD-L1 interaction as a predictive biomarker for the stratification of patients with non-small-cell lung carcinoma, combined with PD-L1 expression, should significantly improve the response rates to immunotherapy. This would both capture patients excluded from checkpoint immunotherapy (high PD-1/PD-L1 interaction but low PD-L1 expression, 24% of patients) and additionally avoid treating patients who despite their high PD-L1 expression do not respond and suffer from side effects.

FIG 1.

QF-Pro detects a high degree of inter- and intratumoral heterogeneity in PD-1/PDL1 interaction. (A) Schematics of the QF-Pro platform. QF-Pro is a FRET/FLIM platform that is able to quantify PD-1/PD-L1 interactions over a distance of 1-10 nm. It uses an inverted epifluorescence microscope coupled to a 30-MHz modulated diode laser. The detector is a two-tap CMOS detector that is also modulated at 30 MHz in a homodyne manner. (B) Principle of the QF-Pro assay. The assay uses a cell–cell–compatible amplified FRET method, detected by FRET/FLIM. It is a two-site assay that determines the interactive states of the immune checkpoint ligands and receptors engaged between cells. Both the receptor (PD-1) and the ligand (PD-L1) are labeled with a primary antibody of different species. The primary antibodies are then labeled with a F(ab′)₂ fragment conjugated to ATTO488, for the donor chromophore, and the other with a F(ab′)₂ fragment conjugated to HRP. Using Tyramide signal amplification, the HRP labels the sample with the acceptor chromophore Alexa594. (C) Clinical PD-L1 images show the PD-L1 TPS expression, determined by IHC (SP263 Ventana Roche), on three patient samples. Below, the pseudocolored FLIM images are presented. The left panels show the lifetime maps of the FRET donor alone (no interaction). The right panels show the lifetime map of the FRET donor in the presence of the FRET acceptor. A reduction of donor lifetime in the presence of the acceptor (indicated by a change of pseudocolor from blue to green, yellow, and red depending on the FRET efficiency) represents the functional interaction of PD-1/PD-L1 within a patient sample. In these three examples, a discrepancy between FRET efficiency and PD-L1 expression is visible. (D) Correlation plot shows no correlation between PD-L1 scores and FRET efficiency (each blue dot represents a patient's sample—Spearman r = 0.05; P = .55). (E) Mean FRET efficiency measurements within a selection of patient detected a large degree of inter- and intratumoral heterogeneity. Each box plotted here represents one patient, with the dots representing different tumoral areas analyzed per patient. The NSCLC samples analyzed here have a high degree of intratumoral heterogeneity and also interpatient heterogeneity. Patients classified in order of increasing clinical PD-L1 score showed no correlation with PD-1/PD-L1 interaction state. CMOS, Complementary Metal Oxide Semiconductor; FLIM, fluorescence lifetime imaging microscopy; FRET, Förster resonance energy transfer; HRP, horseradish peroxidase; IHC, immunohistochemistry; n.s., not significant; NSCLC, non–small-cell lung carcinoma, QF-Pro, quantitative functional proteomics; TPS, tumor proportion score.

FIG 2.

High PD-1/PD-L1 interaction state correlates with a significantly enhanced OS. (A) Patients were analyzed for PD-1/PD-L1 interaction states (mean FRET Efficiency). Patients were stratified into two groups: those with the 40% highest interaction states and those with the lowest 60% interaction states. The 40% population with a higher FRET efficiency (higher interaction state) shows a highly significant improvement of the OS compared with the 60% population with a lower FRET (median 31 v 10 months, P < .0001). (B) Patients were analyzed for PD-1/PD-L1 interaction states (mean FRET efficiency). The 40% population with a higher FRET efficiency (higher interaction state) shows a highly significant improvement of PFS compared with the 60% population with a lower FRET (median 17 v 7 months, P = .0001). (C) Kaplan-Meier survival curves of patients stratified by their clinical PD-L1 expression. The patients were stratified as PD-L1 high (50% TPS or higher) or PD-L1 low (< 50% TPS). The PD-L1 score stratification is not predictive of a change in overall survival (P = .162). (D) Kaplan-Meier survival curves of patients stratified by their clinical PD-L1 expression. The patients were stratified as PD-L1 high (50% TPS or higher) or PD-L1 low (< 50% TPS). The PD-L1 score stratification is not predictive of a change in PFS (P = .173). FRET, Förster resonance energy transfer; n.s., not significant; OS, overall survival; PFS, progression-free survival; TPS, tumor proportion score.

FIG 3.

Comparison of the relative influence of PD-L1 expression and PD-1/PD-L1 interaction on OS. (A) The patients' population was categorized according to PD-L1 expression and associated PD-1/PD-L1 interaction states. Patients (each dot represents one patient) were plotted on a matrix on the basis of their mean FRET efficiency (interaction state) and clinical PD-L1 scores and divided into four quadrants. The vertical red line represents the threshold classically applied in clinics for the stratification of patients who should receive or should not receive first-line immunotherapy (≥ 50% PD-L1 score v low < 50% PD-L1, respectively). The horizontal black line represents the 40/60 percent high FRET versus low FRET efficiency cutoff used in our assay stratification. (B-F) Kaplan-Meier analysis comparisons of the patients' OS from each of the matrix categories. (F) shows the very significant difference in OS of patients with a high FRET compared with patients with a low FRET efficiency independent of their PD-L1 expression levels (median OS 27 months versus 17 months; P = .003), indicating that the response to treatment does not depend on PD-L1 expression score. FRET, Förster resonance energy transfer; n.s., not significant; OS, overall survival; TPS, tumor proportion score.

FIG 4.

Correlation between FRET efficiency and lines of treatment on OS. (A) Comparison of the OS of patients who have received first-line ICI immunotherapy and stratified by high versus low PD-L1 scores. (B) Comparison of the OS of patients who have received first-line ICI immunotherapy and stratified according to high versus low FRET efficiency (40/60 percent FRET). (C) Patients treated with immunotherapy in first line were analyzed for PD-1/PD-L1 interaction states (mean FRET efficiency). The 40% population with a higher FRET efficiency (higher interaction state) shows a significant improvement of PFS compared with the 60% population with the lowest FRET (median 27 v 9 months, P = .011). (D) Comparison of the OS of patients who have received ICI immunotherapy as second line of treatment either stratified by low versus high PD-L1 scores or (E) stratified by high versus low FRET efficiency. The statistical analysis shows a highly significant prediction on OS using FRET efficiency as a criterion for stratification in first line of treatment (median still undefined at 50 months v 11 months, P < .0001) and in the second line (median 18 v 10 months, P = .002). (F) Patients treated with immunotherapy in the second line were analyzed for PD-1/PD-L1 interaction states (mean FRET efficiency). A higher FRET efficiency (higher interaction state) shows a significant improvement of PFS compared with the 60% population with the lowest FRET (median 18 v 10 months, P = .003). FRET, Förster resonance energy transfer; ICI, immune checkpoint inhibitor; n.s., not significant; OS, overall survival; PFS, progression-free survival.

FIG A1.

Correlation between OS and FRET efficiency but not with PD-L1 score. (A) The plots from 135 patients show that a highly significant linear correlation exists between the FRET efficiency and the OS (Spearman r = 0.343, P < .0001). However, in (B), there is no correlation between OS and PD-L1 scores (Spearman r = 0.049, P = .571). FRET, Förster resonance energy transfer; n.s., not significant; OS, overall survival; TPS, tumor proportion score.

FIG A2.

High PD-1/PD-L1 interaction state correlates with a significantly enhanced OS in 188 patients. (A) Patients were analyzed for PD-1/PD-L1 interaction states (mean FRET Efficiency). Patients were stratified into two groups: the 40% highest interaction states population and the lowest 60% interaction states population. The 40% population with a higher FRET efficiency (higher interaction state) shows a highly significant improvement of the OS compared with the 60% population with a lowest FRET (median still undefined at 50 months v 11 months, P < .0001). (B) FRET efficiency correlation plot shows a highly significant correlation between FRET efficiency (PD-1/PD-L1 interaction state) and OS (Spearman r = 0.338; P < .0001). FRET, Förster resonance energy transfer; OS, overall survival.

FIG A3.

PD-1/PD-L1 engagement is not predictive of response to chemotherapy treatment. Patients who were treated with chemotherapy in first line were analyzed for PD-1/PD-L1 interaction states (mean FRET efficiency). Patients were stratified into two groups: the 40% highest interaction states population and the lowest 60% interaction states population. The 40% population with a higher FRET efficiency (higher interaction state) did not show any significant improvement of PFS compared with the 60% population with a lowest FRET (median 8 v 6 months, P = .542). FRET, Förster resonance energy transfer; PFS, progression-free survival.