Purpose:

Less than 50% of patients with melanoma respond to anti–programmed cell death protein 1 (anti–PD-1), and this treatment can induce severe toxicity. Predictive markers are thus needed to improve the benefit/risk ratio of immune checkpoint inhibitors (ICI). Baseline tumor parameters such as programmed death ligand 1 (PD-L1) expression, CD8+ T-cell infiltration, mutational burden, and various transcriptomic signatures are associated with response to ICI, but their predictive values are not sufficient. Interaction between PD-1 and its main ligand, PD-L1, appears as a valuable target of anti–PD-1 therapy. Thus, instead of looking at PD-L1 expression only, we evaluated the predictive value of the proximity between PD-1 and its neighboring PD-L1 molecules in terms of response to anti–PD-1 therapy.

Experimental Design:

PD-1/PD-L1 proximity was assessed by proximity ligation assay (PLA) on 137 samples from two cohorts (exploratory n = 66 and validation n = 71) of samples from patients with melanoma treated with anti–PD-1±anti–CTLA-4. Additional predictive biomarkers, such as PD-L1 expression (MELscore), CD8+ cells density, and NanoString RNA signature, were also evaluated.

Results:

A PD-1/PD-L1 PLA model was developed to predict tumor response in an exploratory cohort and further evaluated in an independent validation cohort. This score showed higher predictive ability (AUC = 0.85 and 0.79 in the two cohorts, respectively) for PD-1/PD-L1 PLA as compared with other parameters (AUC = 0.71–0.77). Progression-free and overall survival were significantly longer in patients with high PLA values (P = 0.00019 and P < 0.0001, respectively).

Conclusions:

The proximity between PD-1 and PD-L1, easily assessed by this PLA on one formalin-fixed paraffin-embedded section, appears as a new biomarker of anti–PD-1 efficacy.

Translational Relevance

The proximity between programmed cell death protein 1 (PD-1) and programmed death ligand 1 (PD-L1) evaluated by a proximity ligation assay (PLA) can easily be performed on a fixed tumor sample. When assayed on tumor samples from patients with metastatic melanoma harvested prior to immune checkpoint inhibitor (ICI) treatment, PD-1/PD-L1 PLA is associated with response to ICI. The predictive value of PD-1/PD-L1 is stronger than the one of CD8+ cells' infiltration, PD-L1 expression, and a transcriptomic inflammatory signature (TIS). Progression-free and overall survival are significantly longer in patients with high PLA values (P = 0.00019 and P < 0.0001, respectively) than in patients with low PLA values. Thus, PD-1/PD-L1 PLA appears as a predictive biomarker that could potentially help physicians in their therapeutic choices for treating patients with advanced melanoma.

In melanoma, the pioneer tumor in terms of immune checkpoint inhibitor (ICI) development, response rates vary between 35% and 45% for anti– programmed cell death protein 1 (PD-1) monotherapy (pembrolizumab or nivolumab) and is close to 60% for the combination anti–PD-1 + anti–CTLA-4 (nivolumab + ipilimumab; refs. 1, 2). ICI can induce potentially severe immune-related toxicity and it would be critical to optimize the benefit/risk ratio of these drugs with reliable predictive markers of response. Several parameters reflecting the existence of an actionable tumor microenvironment (TME) have been evaluated. ICI benefit is associated, for example, with high tumor mutational burden (TMB) and with various transcriptomic signatures revealing an inflamed TME. A tumor inflammation signature (TIS), in particular, is associated with pembrolizumab response in metastatic melanoma (3–5). But the predictive values of these biomarkers are not strong enough to be of clinical use. Although programmed death ligand 1 (PD-L1) expression is also associated with anti–PD-1 response, no reliable prediction threshold could be identified for patients with melanoma (6). An obvious reason why PD-L1 staining alone is a poor predictive marker is that the efficacy of anti–PD-1 inhibitors relies on their capacity to impair the immunosuppressive interaction between PD-1–expressing lymphocytes and PD-1 ligands. But PD-L1 can also be expressed by tumor cells in the absence of CD8+ T-cell infiltration through cell-intrinsic activation mechanisms (7). In the latter situation, there is no reason to expect an association between PD-L1 expression and response to ICI.

Altogether, predictive biomarkers of ICI benefit remain elusive for patients with metastatic melanoma.

To overcome this issue, we hypothesized that the engagement of PD-L1 with PD-1 in the TME could constitute a reliable predictive marker of response to ICI. We previously used a proximity ligation assay (PLA) procedure to assess the proximity between PD-1 and PD-L1 on formalin-fixed paraffin embedded (FFPE) tumor tissue, and found that it was associated with ICI benefit in a small cohort of patients (8). Here, we used a discovery (exploratory) cohort, and a validation cohort to confirm the predictive value of PD-1/PD-L1 proximity in comparison with PD-L1 expression alone (MELscore), CD8+ T-cells infiltration (IHC), and T-cell inflammatory RNA signature (TIS; NanoString) in a total of 137 patients.

Patients and samples

An exploratory cohort of 66 patients and a validation cohort of 71 patients treated with pembrolizumab or nivolumab ± ipilimumab and evaluable for their response to ICI were studied. Among the patients biopsied during the same period, we did not include in our study, 8 patients, who had a mixed response and for whom a local treatment was associated with ICI because they could not be evaluated for their response status to ICI. The study was conducted in accordance with the Declaration of Helsinki. All patients received appropriate information and signed an informed consent form authorizing tumor biopsies and molecular studies in the context of an institutional review board and Center of Biological Resources–approved protocol (MSN-08–027 CPP Ile de France, registration number: 2008-A00373–52). Patients were classified as responders (partial or complete responses) or nonresponders (stability or progression of the disease) using RECIST1.1 criteria. Characteristics of the patients are shown in Supplementary Table S1. FFPE tumor samples collected within 4 months prior to treatment and containing at least 100 tumor cells were used. Consents were obtained from all patients prior to the study (NCT 02105168).

PLA

PLAs were performed on FFPE samples on 3-μm human melanoma tissue sections and the principle is summarized in Fig. 1A. Following dewaxing and rehydrating the tissue sections, antigen retrieval was performed by heating the slides at 95°C for 45 minutes in buffer citrate pH 6 (catalog number T0050, Diapath).

Figure 1.

PLA. A, Schematic representation of the PLA principle. B, PLA staining and image analysis. Cropped RGB images of areas of interest are open in MacBiophotonics ImageJ software (a). A composite image is created with the macro XsRGB to increase the signal of dots (b). This composite image is split into three channels: red, green, and blue. The red channel is used to calculate the number of dots (c), and the green channel is used to calculate the number of nuclei (d). C, Serial sections of a melanoma sample (tumor-invasive margin) stained for PD-1/PD-L1 proximity ligation assay [a, PLA, dot-like staining; b, SOX10 (staining of melanoma cell nuclei); c, CD8 (purple chromogen) and PD-L1 (clone 22C3)]. Scale bar, 50 μm.

Figure 1.

PLA. A, Schematic representation of the PLA principle. B, PLA staining and image analysis. Cropped RGB images of areas of interest are open in MacBiophotonics ImageJ software (a). A composite image is created with the macro XsRGB to increase the signal of dots (b). This composite image is split into three channels: red, green, and blue. The red channel is used to calculate the number of dots (c), and the green channel is used to calculate the number of nuclei (d). C, Serial sections of a melanoma sample (tumor-invasive margin) stained for PD-1/PD-L1 proximity ligation assay [a, PLA, dot-like staining; b, SOX10 (staining of melanoma cell nuclei); c, CD8 (purple chromogen) and PD-L1 (clone 22C3)]. Scale bar, 50 μm.

Close modal

After blocking, the antibodies were used at the following concentrations: anti–PD-L1 (rabbit, clone SP142 1:50 dilution, Spring Bioscience or clone E1L3N 1:200 dilution, Cell Signaling Technology) and anti–PD-1 (mouse, clone NAT105, 1:200 dilution, Abcam). The PLA minus (catalog number DUO92005–100RXN, Sigma Aldrich) and PLA plus (catalog number DUO92001–100RXN, Sigma Aldrich) probes (containing secondary antibodies conjugated to oligonucleotides) were added and incubated for 1 hour at 37°C. More oligonucleotides were then added and allowed to hybridize to the PLA probes. Ligase was used to anneal the two hybridized oligonucleotides into a closed circle. The DNA was then amplified (with rolling circle amplification), and the amplicons were detected using the Brightfield Detection Kit (catalog number DUO92012–100RXN, Sigma Aldrich) for chromogenic development.

For PD-1/PD-L1 PLAs on tissue sections, a purple chromogen (Discovery purple, catalog number 7053983001, Roche) with hematoxylin counterstaining (catalog number MHS32–1L, Sigma Aldrich) was performed manually. Regions of interest (ROI) for image analysis were selected in tumor areas containing both tumor cells and tumor-infiltrating lymphocytes. The number of ROIs per sample depended on the size of the specimen: for small specimens, the whole peri- or intratumoral areas could be included in 2 ROIs, whereas in larger specimens, and especially in those where the infiltration was multifocal, a higher number of ROIs (up to 6) was selected to encompass tumor heterogeneity. In the peritumoral area, the immune response was evaluated within 500 μm of the connective tissue surrounding the last row of tumor cells.

In lymph nodes, only the immune cells located in the closest vicinity of tumor cells (2 rows) were evaluated, similarly to methods described for PD-L1 IHC assessment in most of the predictive tests used in the clinical setting. Using this methodology, the total number of ROIs by sample ranged from 2 to 6 with a mean of 2.8 in the peritumoral areas and 3.5 in intratumoral areas. Cropped RGB images were opened in the MacBiophotonics ImageJ software. A composite image was created with the XsRGB macro to increase the signal of dots (Fig. 1B). This composite image was split into red, green, and blue channels. The green channel was used to detect and count the number of nuclei and the red channel to detect and quantify the number of dots. The results were reported as the ratio of the number of dots to the number of detected cell nuclei.

Figure 1C shows the invasive margin of a melanoma samples stained for PD-1/PD-L1 proximity ligation assay (a: PLA, dot-like staining).

The highest PLA value was considered for the analysis. When one of the PLA score (the peritumoral or the intratumoral one) was not visible (NA) the other score was considered.

IHC

PD-L1 IHC was performed using a Ventana Benchmark Ultra platform (Ventana Medical Systems) as previously described using the clone E1L3N, from Cell Signaling Technology and Diaminobenzidine (DAB) was used as a chromogen. PD-L1 was evaluated in both tumor and immune cells using the MELscore scale by a trained pathologist blinded from clinical data (Supplementary Fig. S1; Fig. 1C).

CD8+ IHC was performed using a Ventana Discovery Ultra platform with the anti-CD8 antibody (clone SP16, Spring Bioscience) as previously described. Detection steps were carried with an anti-rabbit Ultra-MAP kit coupled with HRP. Discovery purple was used as a chromogen. The density of CD8+ cells infiltrate was scored by a trained pathologist, both in the tumor center and at the invasive margin of the tumor, using a 4-grade semiquantitative scale (Supplementary Fig. S2; Fig. 1C). The highest score among the peritumoral or intratumoral ones was considered for the study.

SOX10 IHC used clone EP268 (dilution 1:100).

TIS score determination with the NanoString technology

We were able to determine the TIS score in a subset of 77 patients (36 in the experimental cohort and 41 in the validation cohort) from the total RNA isolated from 5-mm–thick FFPE sections of tumor tissue fixed on positively charged slides as previously described using an RUO version of the assay. The TIS scores were calculated as a weighted sum of the expression values for the 18 genes (HLA-E, NKG7, CD8A, PSMB10, HLA-DQA1, HLA-DRB1, CMKLR1, CCL5, CXCL9, CD27, CXCR6, IDO1, STAT1, TIGIT, LAG3, PD-L1/2, CD276) normalized to 10 housekeeping genes.

Statistical analyses

PD-L1 staining was evaluated as the MELscore category because this score includes both tumor and immune cells and has shown the highest correlation with clinical outcome of patients in clinical trials. Comparisons between groups of patients were performed using an unpaired nonparametric test, and P value of 0.05 or less was considered as significant. To create a model that predicts patient response, a leave one-out cross-validation was used for each model with one parameter, and a logistic regression was calculated. We repeated the main analysis in the context of a multivariable model including patients' characteristics, treatment history, and localization of the biopsies. We used R studio software for all analyses (0.3 Version (2020–10–10) and an html report of the entire analysis is provided as a Supplementary Data.

The exploratory cohort (N = 66) comprised 33 responders with complete or partial response and 33 nonresponders to ICI. Supplementary Table S1 shows patients' characteristics, and Supplementary Table S2 shows the results of PLA, CD8, MELscores for each sample.

The proximity between PD-1 and PD-L1 was more significantly associated (P = 2×10−7) with the clinical outcome than CD8+ T-cell infiltration (P = 0.00046) or PD-L1 expression (P = 0.00032; Fig. 2A). Comparison of ROC curves also showed a better performance for the PD-1/PD-L1 proximity (AUC = 0.85) as compared with PD-L1 expression (AUC = 0.77) or CD8+ T-cell infiltration (AUC = 0.71; Fig. 2B). Patients with a PD-1/PD-L1 proximity score above the median value (PLAhighn = 33) had a longer PFS (median PFS not reached, P = 0.00019; Fig. 2C, left) and OS (median = 2393 days, P < 0.0001; Fig. 2C, right) than the ones with a PLA score below the median value (PLAlow tumors, n = 33; median PFS = 113 days and median OS = 668 days; see also Supplementary Fig. S3). Combination of the three parameters did not improve the predictive value of the biomarkers (data not shown).

Figure 2.

The PLA PD-1/PD-L1 is correlated with the response to immune checkpoint blockade in metastatic melanomas, in the exploratory cohort. A, Boxplot representation of the CD8+ T-cell infiltration, PD-L1 immunostaining using the MELscore and PD-1/PD-L1 PLA variables. NR, nonresponders n = 33; R, responders n = 33. P values were calculated using a nonparametric Wilcoxon test. B, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA variables. C, Kaplan–Meier representation of the progression-free survival (PFS) and overall survival (OS) for patients with a high PLA value (≥median of the PLA PD-1/PD-L1 variable; yellow curve) and those with a low PLA value (<median of the PLA PD-1/PD-L1 variable; blue curve).

Figure 2.

The PLA PD-1/PD-L1 is correlated with the response to immune checkpoint blockade in metastatic melanomas, in the exploratory cohort. A, Boxplot representation of the CD8+ T-cell infiltration, PD-L1 immunostaining using the MELscore and PD-1/PD-L1 PLA variables. NR, nonresponders n = 33; R, responders n = 33. P values were calculated using a nonparametric Wilcoxon test. B, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA variables. C, Kaplan–Meier representation of the progression-free survival (PFS) and overall survival (OS) for patients with a high PLA value (≥median of the PLA PD-1/PD-L1 variable; yellow curve) and those with a low PLA value (<median of the PLA PD-1/PD-L1 variable; blue curve).

Close modal

The same analyses were performed on an independent validation cohort of 71 patients with 42 responding patients and 29 nonresponding patients. Here again, the performance of the PD-1/PD-L1 proximity was higher than the other tested parameters in terms of association with clinical outcome (P = 0.00014; Fig. 3A), and ROC analysis with an AUC of 0.79 above the AUC obtained with CD8+ T-cells infiltration (0.65) or PD-L1 expression (0.60; Fig. 3B). Patients with a high PD-1/PD-L1 proximity score had a longer PFS (median = 2,200 days, P = 0.0023; Fig. 3C, left) and OS (median = 50% survival not reached, P < 0.013; Fig. 3C, right) than the ones with a low PLA score (median PFS = 102 days and OS = 663 days; Supplementary Fig. S3).

Figure 3.

The PLA PD-1/PD-L1 is correlated with response to ICB in metastatic melanomas in the validation cohort. A, Boxplot representation of the CD8+ T-cell infiltration, PD-L1 immunostaining using the MELscore and PD-1/PD-L1 PLA variables. NR, nonresponders n = 29; R, responders n = 42. P values were calculated using a nonparametric Wilcoxon test. B, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA variables. C, Kaplan–Meier representation of the progression-free survival (PFS) and overall survival (OS) for patients with a high PLA value (≥median of the PLA PD-1/PD-L1 variable; yellow curve) and those with a low PLA value (<median of the PLA PD-1/PD-L1 variable; blue curve).

Figure 3.

The PLA PD-1/PD-L1 is correlated with response to ICB in metastatic melanomas in the validation cohort. A, Boxplot representation of the CD8+ T-cell infiltration, PD-L1 immunostaining using the MELscore and PD-1/PD-L1 PLA variables. NR, nonresponders n = 29; R, responders n = 42. P values were calculated using a nonparametric Wilcoxon test. B, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA variables. C, Kaplan–Meier representation of the progression-free survival (PFS) and overall survival (OS) for patients with a high PLA value (≥median of the PLA PD-1/PD-L1 variable; yellow curve) and those with a low PLA value (<median of the PLA PD-1/PD-L1 variable; blue curve).

Close modal

The TIS NanoString RNA signature could be evaluated together with the three previously described parameters on a subgroup of patients including 36 patients (17 responders and 19 nonresponders) from the exploratory cohort, and 41 patients (22 responders and 19 nonresponders) from the validation cohort. In both cohorts, the PD-1/PD-L1 proximity was the most reliable predictive biomarker (AUC = 0.85 for the first cohort and AUC = 0.79 for the second cohort) compared with other parameters (AUC = 0.69 and 0.63 for TIS, 0.67 and 0.69 for CD8+ T-cell infiltration, 0.72 and 0.62 for PD-L1 expression; Fig. 4). We could not improve the predictability of the test by using a composite biomarker combining these parameters. Of note, some patients received anti–PD-1 monotherapy or the combination of ICI but the number of patients was too low in each category to find significant differences in clinical outcome for the tested parameters. For the same reason, we could not analyze the difference in predictability of the PD-1/PD-L1 PLA in patients with a complete or a partial response.

Figure 4.

Subanalyses in the exploratory (n = 36/66) and validation (41/71) cohorts using TIS NanoString signature: A, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA and TIS NanoString signature variables on the exploratory cohort subpopulation. B, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA and TIS NanoString signature variables on the validation cohort subpopulation.

Figure 4.

Subanalyses in the exploratory (n = 36/66) and validation (41/71) cohorts using TIS NanoString signature: A, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA and TIS NanoString signature variables on the exploratory cohort subpopulation. B, ROC representation of each logistic regression model calculated for the CD8+ T-cell infiltration, MELscore, and PD-1/PD-L1 PLA and TIS NanoString signature variables on the validation cohort subpopulation.

Close modal

We assessed potential interactions between PLA and other covariates (age of patient, melanoma mutation status, LDH level, tumor stage, prior therapies) to perform a multivariable analysis. We found no association when we considered patients and disease characteristics as well as treatment history (age of the patients, melanoma mutation status, lactate dehydrogenase (LDH) level, tumor stage, prior therapies; Supplementary Table S3). Although we found a significant association between localization of biopsy and PLA score (P = 0.007; Kruskal–Wallis test), when this variable (localization of biopsies) was included in the model used to explore the impact of the PLA on the response, we found that the strength of the association between PLA values and clinical outcomes was not modified by this variable. We therefore did not include this variable in the final model (data shown on the html report).

PD-L1 expression remains a mediocre predictive marker for patients with melanoma treated with ICI even if it was improved by designing more appropriate scores such as the MELScore that takes into account PD-L1 expression on both tumor and immune cells (6). In our cohorts, PD-L1 expression has significant but limited predictive value, quite similar to the one of CD8+ T-cell infiltration, a biomarker also previously shown to correlate with clinical response after ICI (9).

Because anti–PD-1 and anti–PD-L1 agents' efficacy relies on the disruption of the PD-1/PD-L1 interaction, we aimed at directly evaluating this interaction in the TME using a PLA. This technology has been widely used as a marker for protein–protein proximity and interaction in in vitro cell experiments (10). Its use on FFPE samples is challenging because of the risk of epitopes alteration induced by formaldehyde fixation and denaturation during the process of paraffin embedding. However, even if the standardization of the assay for clinical use remains to be done, the dynamic range of the test was sufficient to classify patients according to the PLA score results.

Our data clearly showed in two independent cohorts that the proximity between PD-1 and PD-L1 allows a better discrimination of patients with respect to response to ICI as compared with other potential biomarkers such as PD-L1 expression, CD8+ T-cell infiltration, and TIS signature. The high predictability of this PLA as a single biomarker of ICI efficacy might result from its close relationship with the biological activity of the PD-1/PD-L1 engagement, making of it almost a functional marker of the TME immune suppression. It is also possible that the high predictability of this test might be due to the fact that it is not limited to a specific subset of cells that are supposed to express PD-L1 (tumor cells or antigen-presenting cells) and PD-1 (activated T cells) but also to other cells in the microenvironment, such as B cells that can express both PD-1 and PD-L1 and that can suppress CD4+ and CD8+ T cells via PD-1/PD-L1–dependent pathway (11). A colocalization of PD-1 and PD-L1 was previously evaluated with a specific immunostaining analysis software. However, it was not as effective when used a single marker (AUC = 0.58), and had to be combined with the analysis of HLADR/IDO colocalization staining (AUC = 0.66; ref. 12).

Over the last recent years, several studies using highly sophisticated methods including bulk tumor or single-cell RNA sequencing, multiplex immunostainings or cytometry by time of flight, have identified various predictors of ICI efficacy with AUC around 0.8 when used alone or in combination (13–17). In a recent whole exome and transcriptome meta-analysis performed on more than 1,000 patients treated with ICI across seven tumor types, it was found that the clonal TMB was the best predictive biomarker (18). This meta-analysis established a multivariable prediction model based on all biomarkers that achieved significance in the largest cohorts of patients: clonal TMB, indel TMB, nonsense-mediated decay escape TMB, tobacco, AOBEC, T-cell inflamed transcriptomic and UV signatures, sex, gene expression values of PD-L1, CD8A, and CXCL9. The predictive performance of this multivariable biomarker varied across tumor types, but always largely outperformed TMB-only models. The most predictive score (AUC: 0.86 vs. 0.68 for TMB, P = 0.0049) was obtained for a pan-cancer cohort of 76 patients. In a recently published melanoma cohort of 121 patients, the multivariable biomarker achieved a predictive AUC of 0.66, significantly better than the one of TMB (AUC = 0.58; P = 0.025; ref. 14). In another recent meta-analysis addressing the same question of predictive biomarkers to anti–PD-1/PD-L1 therapy in more than 8,000 patients across ten solid tumor types, multiplex IHC/immunofluorescence assays were found more predictive (AUC = 0.79) compared with PD-L1 staining (AUC = 0.65), TMB (AUC = 069), and transcriptomic signatures (AUC = 0.65; ref. 15). With an AUC of 0.85 on the exploratory cohort and 0.79 on the validation cohort, the PD-1/PD-L1 PLA that is performed on one single FFPE tumor section, compares favorably with the other potential biomarkers cited above.

Recently another team used a similar PLA-based PD-1/PD-L1 procedure in negative sentinel lymph nodes from patients with stage IIB-C melanoma (thick primary melanoma without lymph node involvement; ref. 19). Interestingly and in accordance with the immunosuppressive effect of PD-1/PD-L1 interaction, a high PLA score was associated with a higher risk of recurrence in patients who did not receive any ICI adjuvant treatment. Identification of recurrence-associated biomarkers is of major interest in this particular population of patients. Indeed, we want to avoid a useless and potentially toxic adjuvant ICI treatment for patients with low relapse risk and target only the high-risk population for adjuvant therapy.

Strikingly, PD-1/PD-L1 PLA appears as a potentially useful biomarker, being at the same time a pejorative prognostic parameter that can identify patients at risk, and a highly predictive marker for anti–PD-1 efficacy. This score could thus be of paramount interest in patients with melanoma from stage II to stage IV.

The majority of the samples analyzed in our cohorts are from skin (74/137) or lymph node (30/137) metastases, which could be a potential pitfall. However, the localization of the metastases did not impact the predictive value of the PD-1/PD-L1. We had to exclude from our study 8 patients that could not be evaluated for their response to ICI because they were also treated by physical therapy (radiotherapy, cryotherapy, radiofrequency, surgery). This is also a potential limitation but it concerned only a very small proportion of our overall patients' population.

The main limitation of our study is that PLA has not yet been standardized for clinical use by pathologists. However, the PLA method is widely available and can be performed on automated IHC platforms (Ventana Discovery, Leica Bond RX). We presently work to develop standardized, in vitro diagnostic–labeled techniques for PD-1/PD-L1 PLA staining and for subsequent image analysis.

In conclusion, a PLA-based detection of PD-1/PD-L1 proximity on FFPE tissues, easily accessible in clinical pathology laboratories, is highly predictive of response to anti–PD-1 treatment in patients with melanoma treated with anti–PD-1 inhibitors ± anti–CTLA-4. This practical predictive biomarker should be evaluated in other tumor types treated with ICI.

J. Adam reports grants from MSD and Bayer and personal fees from AstraZeneca outside the submitted work. S. Warren reports other support from NanoString Technologies during the conduct of the study. K. Sorg is an employee and stockholder of NanoString Technologies, Inc. S. Ong reports personal fees from NanoString Technologies during the conduct of the study as well as personal fees from Bristol Myers Squibb outside the submitted work. C. Robert reports personal fees from BMS, MSD, Novartis, Roche, Sanofi, and Pierre Fabre and personal fees from AstraZeneca during the conduct of the study. No disclosures were reported by the other authors.

I. Girault: Conceptualization, software, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. J. Adam: Validation, visualization, methodology, writing–original draft, writing–review and editing. S. Shen: Validation, methodology, writing–review and editing. S. Roy: Resources, visualization, project administration, writing–review and editing. C. Brard: Methodology, writing–review and editing. S. Faouzi: Investigation, visualization, writing–review and editing. E. Routier: Resources, writing–review and editing. J. Lupu: Resources, writing–review and editing. S. Warren: Resources, software, formal analysis, methodology, writing–review and editing. K. Sorg: Software, formal analysis, writing–review and editing. S. Ong: Software, formal analysis, writing–review and editing. P. Morel: Resources, writing–review and editing. J.-Y. Scoazec: Conceptualization, formal analysis, validation, visualization, writing–review and editing. S. Vagner: Conceptualization, formal analysis, writing–original draft, writing–review and editing. C. Robert: Conceptualization, resources, formal analysis, supervision, validation, writing–original draft, writing–review and editing.

The authors wish to acknowledge the constant support of the Collectif Ensemble contre le Mélanome, the Association Vaincre le Mélanome, the Foundation Carrefour, and the persons working for the Petra Platform. This work was supported by the SIRIC SOCRATE (grant INCa-DGOS-INSERM 6043) by Ensemble contre le Mélanome Sébastien Bazin and the Foundation Carrefour.

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.

1.
Robert
C
,
Ribas
A
,
Schachter
J
,
Arance
A
,
Grob
J-J
,
Mortier
L
, et al
Pembrolizumab versus ipilimumab in advanced melanoma (KEYNOTE-006): post hoc 5-year results from an open-label, multicenter, randomized, controlled, phase III study
.
Lancet Oncol
2019
;
20
:
1239
51
.
2.
Larkin
J
,
Chiarion-Sileni
V
,
Gonzalez
R
,
Grob
J-J
,
Rutkowski
P
,
Lao
CD
, et al
Five-year survival with combined nivolumab and ipilimumab in advanced melanoma
.
N Engl J Med
2019
;
381
:
1535
46
.
3.
Ayers
M
,
Lunceford
J
,
Nebozhyn
M
,
Murphy
E
,
Loboda
A
,
Kaufman
DR
, et al
IFNγ-related mRNA profile predicts clinical response to PD-1 blockade
.
J Clin Invest
2017
;
127
:
2930
40
.
4.
Cristescu
R
,
Mogg
R
,
Ayers
M
,
Albright
A
,
Murphy
E
,
Yearley
J
, et al
Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy
.
Science
2018
;
362
:
eaar3593
.
5.
Danaher
P
,
Warren
S
,
Lu
R
,
Samayoa
J
,
Sullivan
A
,
Pekker
I
, et al
Pan-cancer adaptive immune resistance as defined by the tumor inflammation signature (TIS): results from The Cancer Genome Atlas (TCGA)
.
J Immunother Cancer
2018
;
6
:
63
.
6.
Daud
AI
,
Wolchok
JD
,
Robert
C
,
Hwu
W-J
,
Weber
JS
,
Ribas
A
, et al
Programmed death–ligand 1 expression and response to the antiprogrammed death 1 antibody pembrolizumab in melanoma
.
J Clin Oncol
2016
;
34
:
4102
9
.
7.
Mezzadra
R
,
Sun
C
,
Jae
LT
,
Gomez-Eerland
R
,
de Vries
E
,
Wu
W
, et al
Identification of CMTM6 and CMTM4 as PD-L1 protein regulators
.
Nature
2017
;
549
:
106
10
.
8.
Cerezo
M
,
Guemiri
R
,
Druillennec
S
,
Girault
I
,
Malka-Mahieu
H
,
Shen
S
, et al
Translational control of tumor immune escape via the eIF4F-STAT1-PD-L1 axis in melanoma
.
Nat Med
2018
;
24
:
1877
86
.
9.
Tumeh
PC
,
Harview
CL
,
Yearley
JH
,
Shintaku
IP
,
Taylor
EJM
,
Robert
L
, et al
PD-1 blockade induces responses by inhibiting adaptive immune resistance
.
Nature
2014
;
515
:
568
71
.
10.
Boussemart
L
,
Malka-Mahieu
H
,
Girault
I
,
Allard
D
,
Hemmingsson
O
,
Tomasic
G
, et al
eIF4F is a nexus of resistance to anti-BRAF and anti-MEK cancer therapies
.
Nature
2014
;
513
:
105
9
.
11.
Wang
X
,
Wang
G
,
Wang
Z
,
Liu
B
,
Han
N
,
Li
J
, et al
PD-1–expressing B cells suppress CD4+ and CD8+ T cells via PD-1/PD-L1–dependent pathway
.
Mol Immunol
2019
;
109
:
20
6
.
12.
Johnson
DB
,
Bordeaux
J
,
Kim
JY
,
Vaupel
C
,
Rimm
DL
,
Ho
TH
, et al
Quantitative spatial profiling of PD-1/PD-L1 interaction and HLA-DR/IDO-1 predicts improved outcomes of anti–PD-1 therapies in metastatic melanoma
.
Clin Cancer Res
2018
;
24
:
5250
60
.
13.
Gide
TN
,
Quek
C
,
Menzies
AM
,
Tasker
AT
,
Shang
P
,
Holst
J
, et al
Distinct immune cell populations define response to anti–PD-1 monotherapy and anti–PD-1/anti–CTLA-4 combined therapy
.
Cancer Cell
2019
;
35
:
238
55
.
14.
Liu
D
,
Schilling
B
,
Liu
D
,
Sucker
A
,
Livingstone
E
,
Jerby-Arnon
L
, et al
Integrative molecular and clinical modeling of clinical outcomes to PD-1 blockade in patients with metastatic melanoma
.
Nat Med
2019
;
25
:
1916
27
.
15.
Lu
S
,
Stein
JE
,
Rimm
DL
,
Wang
DW
,
Bell
JM
,
Johnson
DB
, et al
Comparison of biomarker modalities for predicting response to PD-1/PD-L1 checkpoint blockade: a systematic review and meta-analysis
.
JAMA Oncol
2019
;
5
:
1195
204
.
16.
Sade-Feldman
M
,
Yizhak
K
,
Bjorgaard
SL
,
Ray
JP
,
de Boer
CG
,
Jenkins
RW
, et al
Defining T-cell states associated with response to checkpoint immunotherapy in melanoma
.
Cell
2018
;
175
:
998
1013
.
17.
Auslander
N
,
Zhang
G
,
Lee
JS
,
Frederick
DT
,
Miao
B
,
Moll
T
, et al
Robust prediction of response to immune checkpoint blockade therapy in metastatic melanoma
.
Nat Med
2018
;
24
:
1545
9
.
18.
Litchfield
K
,
Reading
JL
,
Puttick
C
,
Thakkar
K
,
Abbosh
C
,
Bentham
R
, et al
Meta-analysis of tumor- and T-cell–intrinsic mechanisms of sensitization to checkpoint inhibition
.
Cell
2021
;
184
:
596
614
.
19.
Dammeijer
F
,
van Gulijk
M
,
Mulder
EE
,
Lukkes
M
,
Klaase
L
,
van den Bosch
T
, et al
The PD-1/PD-L1-checkpoint restrains T-cell immunity in tumor-draining lymph nodes
.
Cancer Cell
2020
;
38
:
685
700
.