The tumor inflammation signature (TIS) is associated with anti-PD-1 treatment benefit in the CERTIM pan-cancer cohort

Fifty-eight patients from the CERTIM (Immunomodulatory Therapies Multidisciplinary Study group) were included in the study. The CERTIM cohort was initiated in our hospital in July 2015 and prospectively includes all the patients treated with a PD-1 checkpoint inhibitor (either nivolumab or pembrolizumab) administered per label as a single agent, for advanced solid tumor. Treatment was continued until disease progression or intolerable toxicity, physician or patient decision. Tumor imaging by CT scan was performed at baseline, every 8 weeks for patients receiving nivolumab and every 6 weeks for patients receiving pembrolizumab through the first 6 months and every 12 weeks for both thereafter. Response was assessed per Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. Clinically stable patients considered to be deriving clinical benefit could continue therapy until disease progression was confirmed on imaging done at least 4 weeks after the first assessment of stable disease. The toxicities observed in the patient cohort were monthly reviewed by the dedicated CERTIM multidisciplinary board. Clinical status was assessed monthly during treatment and continued when the treatment was stopped. Database lock-up was 9 February 2018. Median patients follow-up was 19.2 months.

This study was approved by the ethics committee (CPP Ile de France II, no. 2008-133, 2012 06-12, 2018 MS1) in agreement with article L.1121-1 of French law.

For each tumor, we performed PD-L1 immunostaining on fresh-cut slides from representative FFPE blocks using an anti-PD-L1 antibody (E1L3N, Cell signaling) on Bond automat (Leica) as previously described and validated by the PATTERN French thoracic pathologists group [10]. Staining was blinded analyzed (DD, ALM, BB) on tumoral cells.

Genomic DNA from 22 tumors was isolated from formalin-fixed paraffin-embedded (FFPE) blocks using Maxwell 16 FFPE Tissue LEV DNA Purification Kit (Promega), according to the manufacturer’s instructions. Whole-exome was sequenced as previously reported [11]. DNA were sequenced on the Novaseq 6000 platform, using at least 100 ng of double stranded tumoral DNA (Qubit dsDNA HS kit). After shearing, 11 randomly selected samples were run on the Agilent Bioanalyzer to confirm successful shearing prior to library construction The SureSelect Human All Exon V6 capture kit was used to capture coding regions of genes included in the major genomic databases. Paired end FASTQ files of 101mer sequence reads were generated. All sequence data was quality controlled for read counts, quality values, kmer usage, GC-content, and all other relevant parameters with FastQC (v0.10.1). The DNA read sequences were aligned to the genome (UCSC hg19; Feb 2009 release) using BWA (v0.7.15) and reads sorting and PCR duplicate removal were conducted using Picard (v2.8.3). VarScan2 (v2.4.2) with Samtools mpileup (v1.3) was used to call SNVs/Indels against human reference genome. Germline polymorphisms were removed by retaining only variants with MAF in all races of < 1% or unknown MAF within the 1000 genomes and NHLBI-ESP project with 6500 exome database. The retained SNVs/indels were further filtered by removing SNPs in dbSNP129. Depth of sequencing coverage was 60× on average across all samples in the cohort.

Archival biopsies of patients included in the CERTIM cohort, sampled before anti-PD-1 treatment and with adequate tumor tissue left in FFPE blocks were selected for RNA extraction. RNA was extracted with High Pure FFPE RNA Isolation Kit (Roche) from FFPE tumor samples according to the recommendations of the manufacturer and quantified using fluorimetry with Qubit™ RNA XR Assay Kit (Invitrogen). 100 ng RNA (46 samples) or 30 to 85 ng RNA (12 samples) were successfully hybridized to a beta version of the NanoString^® PanCancer IO 360 Panel code set, according to the recommendations of the manufacturer.

Raw data for each sample and gene were normalized to internal ERCC controls to eliminate technical variability of the assay, and then counts were normalized to the geometric mean of endogenous housekeeping genes followed by log2 transformation. Gene expression signatures, including the TIS, were calculated as a weighted linear average of the constituent genes [2, 12]. The weighted scores used for calculation of the TIS and other signatures are NanoString intellectual property. For the correlation analysis with clinical response to treatment, clinical benefit was defined as complete or partial RECIST response while stable and progressive disease were defined as lack of clinical benefit. Normalized gene counts and signature scores were compared to the response category using a linear model. The log2 fold change, Wald-type confidence interval and p-value were calculated for each gene and signature (Additional file 1: Table S1 and Additional file 2: Table S2). To assess predictive performance of the additional signatures above and beyond the predictive performance of TIS, a logistic regression model was used to assess performance conditional upon TIS score.

For the analysis on the survival time, the genes and scores were dichotomized into high and low groups based on median, with the exception of TIS which was divided into tertiles. The survival time was fit to the binary group with Cox proportional hazard model. The hazard ratio between the high and low group, the Wald-type confidence interval and the log-rank test p-value are reported for each gene and signature (Additional file 1: Table S1 and Additional file 2: Table S2).

Consecutive metastatic cancer patients treated with anti-PD-1 monoclonal antibodies in the outpatient monocentric CERTIM cohort with available FFPE tumor specimen were selected for analysis of gene expression and immune signatures using NanoString PanCancer IO 360 Panel (beta version), which contains an RUO version of the TIS. The clinical characteristics of the 58 patients with different cancer types included in this study are described in Table 1. Several genes showed a differential expression in responders to anti-PD-1 [patients with complete response (CR) or partial response (PR) according to RECIST criteria] compared to non-responders (stable or progressive disease) (Additional file 1: Table S1). After correction for multiple testing, 5 genes related to IFNγ signaling and antigen processing remained significantly higher in responders: CXCL9, CXL10, CXCL11, TAP1 and PSMB9 (Fig. 1a). Since a number of the genes with greatest association with clinical benefit are contained within or closely related to genes in the TIS, we evaluated the TIS as a predictive biomarker in this cohort. In this study, a high TIS score was significantly associated with response to anti-PD-1 treatment (odds ratio = 2.64, 95% CI [1.4; 6.0], p = 0.008, Logistic regression) (Fig. 1b). Furthermore, patients with a high TIS score (upper tertile) had a prolonged overall survival compared to patients with lower scores (hazard ratio = 0.37, 95% CI [0.18, 0.76], p-value = 0.005, Cox regression) (Fig. 1c). The expression of 16 genes included in the TIS score were closely correlated, whereas CD276 and HLA-DQA1 expression appeared more variable across TIS scores (Fig. 1d). The normalized gene expression data, TIS score, as well as response to ICI and survival for each of the samples included in this study are provided in Additional file 3: Table S3. Altogether, these data indicate that the TIS is significantly associated with clinical benefit of anti-PD-1 (pembrolizumab or nivolumab) in a « real life » cohort of patients.

We then focused our analysis on NSCLC which represented the majority of the cases that were studied in this cohort. All 37 patients had received nivolumab, and the clinical characteristics of the patients, including the tumor subtype and smoking status, are indicated in Table 2. Overall, 7/37 (19%) patients responded to treatment. As in the whole cohort, we observed that TIS enriched for tumor response in NSCLC (odds ratio = 3.27, 95% CI [1.2; 11.6], p = 0.03, Logistic regression) (Fig. 2a). Furthermore, patients with a high TIS score (upper tertile) had a prolonged survival compared to those with lower TIS scores (hazard ratio = 0.36, 95% CI [0.14, 0.90], p-value = 0.02, Cox regression) (Fig. 2b). In order to assess the predictive value of TIS score within the context of one of other biomarkers of clinical interest, we also scored the NSCLC samples for PD-L1 expression by IHC at the 1% and 50% cutoff for tumor cell positivity, the two cut offs reported in the label for second and first line NSCLC patient selection for treatment with pembrolizumab. PD-L1 staining ≥ 1% tumor cells was not significantly associated with survival (hazard ratio = 0.87, CI [0.4, 2], p = 0.74), and PD-L1 staining on 50% tumor cells had a trend with overall survival but did not reach statistical significance (hazard ratio = 0.40, CI [0.1, 1.3], p = 0.13) (Fig. 2c, d). Additionally, tumor mutational burden by whole exome sequencing was available for 19 of the patients (see Additional file 4: Table S4 for baseline characteristics). The tumor mutational burden ranged from 13.7 to 26.8 mutations/MB and was positively correlated with the number of smoking pack years (spearman coefficient 0.43). In this limited patient cohort, TMB was lower in adenocarcinoma (n = 11, median = 20.3 mutation/Mb) than in squamous cell carcinoma (n = 6, median = 23.3 mutation/Mb) (p value = 0.01, Fisher test), and was not significantly associated with survival (hazard ratio = 1.91, CI [0.6, 6.2], p = 0.25). In this small cohort, TIS was still significantly associated with overall survival (p = 0.02, data not shown). Finally, we assessed whether any of the biomarkers were associated with one another, and observed that PD-L1 staining on tumor cells and TMB were positively correlated with tobacco exposure, but the other biomarkers were not strongly associated with each other (Fig. 2e). Specifically, PD-L1 IHC staining was not significantly with TMB (spearman coefficient − 0.16, p value 0.53), and the TIS was not significantly correlated with either PD-L1 immunohistochemical staining (spearman coefficient 0.20, p value 0.25), or TMB (spearman coefficient − 0.22, p value 0.38).

We next evaluated a number of other predefined immune gene signatures contained within the IO360 panel for their association with response in the NSCLC cohort and identified several signatures that were statistically significant, including IFNγ signaling, lymphoid cells, T cell, NK cells, cytotoxic cells, exhausted CD8 T cells, macrophages, stroma, inflammatory chemokines, and immunoproteasome (Fig. 3a). The majority had strong positive correlation with TIS and with each other (Fig. 3b). Multivariate analysis showed that none of these gene signatures significantly predicted for response after correcting for TIS (data not shown). The signatures were then evaluated to see which were also predictive of overall survival in the NSCLC cohort, and only TIS, stroma, and lymphoid signatures were significantly associated (Additional file 5: Figure S1). In the multi-tumor cohort, several signatures were associated with response, but only IFNγ signature remained significant after adjusting for TIS (p-value = 0.03) (Additional file 6: Figure S2A); whereas TIS, immunoproteasome, lymphoid and hypoxia signatures were predictive of survival (Additional file 6: Figure S2B). These findings indicate that TIS is a robust biomarker of both response and survival following PD-1 checkpoint blockade in both NSCLC and multi-tumor cohorts from real world patients.