All consecutive patients with advanced NSCLC treated with ICIs in our Department between June 2015 and December 2017 were screened. Patients who underwent tumor response (ie, complete or partial response) with ICI, lasting at least 6 months and with secondary tumor progression, were included in the study. All patients had CT-scan of thorax and abdomen and brain imaging every 3 (with pembrolizumab) or 4 (with nivolumab) cycles. Tumor response was assessed locally in multidisciplinary board, including a radiologist specialized in thoracic oncology, and reviewed at the time of WES analyzes according to RECIST 1.1. Best overall response rate (ORR), progression-free survival (PFS) and overall survival (OS) were collected, with a cut-off point on 11 June 2019. Demographic and histomolecular data were retrospectively collected.

PDL1 IHC was performed using an automated method (Leica) and the E1L3N anti-PD-L1 antibody (Cell Signaling Technology) diluted to the 1/80th on 4 µm slides from the treatment-naïve diagnostic samples. The assay was performed using human amygdala as positive control, and IgG as isotype negative control. Proportion (percentage) of positive tumor cells (membranous staining) was recorded. A minimum of 100 tumor cells was needed for interpretation.

After signature of a consent form, blood samples were collected at diagnosis and every 2 months during ICI treatment. One 10 mL-EDTA (Ethylenediaminetetraacetic acid) tube was collected and whole blood samples were conserved at −80°C for constitutive DNA analyzes. Two additional 10 mL-EDTA tubes of peripheral blood were taken, and plasma was isolated within 1 hour after centrifugation and immediately conserved at −80°C.

ctDNA from plasma samples was extracted using AS1480 Maxwell RSC ccf DNA Plasma Kit (Promega, USA), according to manufacturer’s instructions. DNA from whole blood samples (used for constitutive DNA) was extracted using AS1010 Maxwell 16 Blood DNA Purification Kit (Promega, USA), according to manufacturer’s instructions. DNA concentrations were calculated using Multiscan GO reader, V.1.01.10 (ThermoFisher Scientific, France).

WES was performed on samples taken at the beginning of ICI treatment and at the time of late progression. Library preparation, exome capture, sequencing and data analysis have been done by IntegraGen SA (France). Genomic DNA was captured using Twist Human Core Exome Enrichment System (Twist Bioscience)+IntegraGen Custom.15 Sequence capture, enrichment and elution were performed according to manufacturer’s instruction and protocols (Twist Bioscience) without modification except for library preparation performed with NEBNext Ultra II kit (New England Biolabs). For library preparation 150 ng of each genomic DNA were fragmented by sonication and purified to yield fragments of 150–200 bp. Paired-end adaptor oligonucleotides from the NEB kit were ligated on repaired, a-tailed fragments then purified and enriched by 7 PCR cycles. Five hundred ng of these purified Libraries were then hybridized to the Twist oligo probe capture library for 16 hours in a singleplex reaction. After hybridization, washing and elution, the eluted fraction was PCR amplified with eight cycles, purified and quantified by QPCR to obtain sufficient DNA template for downstream applications. Each eluted-enriched DNA sample was then sequenced on an Illumina NovaSeq as Paired-end 100 reads. Image analysis and base calling was performed using Illumina Real Time Analysis with default parameters. We aimed a tumorous exome depth of x270 and constitutive exome depth of x65.

Between June 2015 and December 2017, 79 consecutive patients with advanced NSCLC were treated by ICIs in our department (nivolumab in second line or more, n=73; pembrolizumab in first line, n=6). We identified 10 patients (12.7%) who experienced initial and durable (more than 6 months) tumor response (all with partial response), and secondary tumor progression after 6 months of treatment. Eight patients had paired blood samples (diagnosis and progression) available for analyzes. Main characteristics of the patients are reported in table 1. Median age was 70.2 years (range 43.5–84.8), five patients (63.0%) were male, and all had lung adenocarcinoma. Seven patients received nivolumab in second-line or more, and one received pembrolizumab as first-line treatment because of a high PD-L1 (50%) expression. Of the seven patients treated with nivolumab, best tumor response with previous chemotherapy line was partial response (n=1; 14.3%), stability (n=4; 57.1%) and tumor progression (n=2; 28.6%). Routine molecular screening at diagnosis showed a KRAS mutation within the tumors of five patients (62.5%). One patient (12.5%) had a tumor with high-level Met amplification (confirmed by fluorescence in situ hybridization). Three patients presented a BRAF mutation (G464V, n=2; G469V, n=1), alone (n=1) or associated with a KRAS mutation (n=2). Two patients (25.0%) did not have alteration of KRAS, BRAF, EGFR, HER2 nor MET at diagnosis. Best tumor response with ICIs was partial response for all patients, all of which occurred at the first tumor evaluation. Clinical outcomes with ICI are presented in table 2. Median number of ICI cycles administered was 22 (range 12–42). Median PFS with ICIs was 12.1 months (IQR 9.0–14.6) (online supplementary figure 1A). Median OS with ICIs was not reached (NR) (IQR 23.5 – NR) (online supplementary figure 1B). Seven patients (87.5%) received a new systemic treatment after progression under ICIs. Median post-ICIs OS was NR (IQR 11.5–NR) (online supplementary figure 1C). At the time of progression with ICI, the majority of the patients experienced an extrathoracic progression (n=5; 62.5%). One patient had tumor progression restricted to the central nervous system (CNS) with appearance of brain metastases and carcinomatous meningitis.

At ICI initiation, ctDNA was detected in all eight plasma samples, with a median plasma DNA concentration of 3.3 ng/µL (range 2.0–15.7). Mean tumorous exome depth was x233 (range 195–315). WES analyzes did not show any KRAS mutation in any samples. Absence of KRAS mutation for patients #1, #2, #4 and #6 was confirmed using an NGS panel on these plasma, whose analytical sensitivity is around 0.5%.16 Two samples (patients #3 and #8) harbored a TP53 mutation (VAF of 8.3% and 6.8%), and one (patient #8) had a KEAP1 mutation (VAF of 9.0%). Summary of molecular profile is presented in table 3. No gene amplification was observed.

At ICI initiation, the eight samples presented a low bTMB, with a median of 0.04 mutations per Mb (range 0.0–8.2). Median number of CNAs was 170.0 (range 116.0–347.0), almost exclusively loss of heterozygosity (LOH). Concerning cancer-related genes, number of CNAs was 28.5 (range 15.0–79.0).

At the time of late progression, ctDNA was detected in all eight plasma samples at the time of late progression, with a median plasma DNA concentration of 3.1 ng/µL (range 2.6–4.4). Mean tumorous exome depth was x233 (range 147–321). WES analyzes did not show any Kras mutation in any samples. However, we found both atypical Braf mutations (G464V with VAF of 10.6%, and G469V with VAF of 3.0%, respectively) that were present at diagnosis for patients #2 and #5. Interestingly, we found in patient #2 the emergence of 2 somatic mutations in Wnt pathway related-genes: a SNV LRP1B mutation (VAF 11.5%, splice-acceptor sequence) and a S213F DVL1 mutation (VAF 4.1%, inducing a missense on the protein sequence). Another sample (patient #5) presented a P192H RET mutation (VAF 4.2%) inducing a missense on the protein sequence. A summary of the mutational profile of the eight samples is presented in table 3. No gene amplification was observed.

For patient #6 who presented a progression located exclusively in the CNS with brain metastases and carcinomatous meningitis (resulting in paraplegia and sphincter disorders), we performed a NGS analysis (Colon-Lung panel) on spinal fluid. We found a different tumor clone within the spinal fluid, with absence of KRAS mutation (detected in diagnostic sample), the presence of an exon 18 G719A activating EGFR mutation (VAF 4.0%), a PTEN mutation (VAF 2.5%) and a TP53 mutation (VAF 2.0%). Comparison of polymorphisms between this sample and the sample at the time of diagnosis was concordant and confirmed they corresponded to the same patient. Patient #6 received therefore subsequent treatment after ICI with osimertinib, considering that thoracic disease was controlled at this time, resulting in a rapid clinical efficacy (spectacular improvement of neurologic symptoms) and stabilization of lesions on brain and spinal MRI at 2 months. Unfortunately, the patient experienced further CNS tumor progression 4 months after the beginning of osimertinib. New NGS analysis in spinal fluid at the time of progression with osimertinib showed the persistence of the exon 18 EGFR, PTEN and TP53 mutations, the absence of KRAS mutation but the apparition of gene amplifications (EGFR, DDR2 and AKT1) that may explain the tumor progression.

The eight samples at the time of progression with ICI still had a low bTMB, with a median of 0.05 mutation per Mb (range 0.0–1.81). MSI status was negative for all tumors (median 0.2%; range 0.0–0.4).

Median number of CNAs was 183.0 (range 127.0–228.0), almost exclusively LOH, without numerical statistical difference compared with the beginning of ICI treatment. However, looking at cancer-related genes, a majority of samples (n=5, 62.5%) at ICI progression showed a dramatic increase of CNAs (mean relative increase of +86% of cancer-related CNAs), with two patients (#7 and #8) with more than +130% relative increase of CNAs, whereas other patients (n=3, 37.5%) had moderate decrease of CNAs (mean relative decrease of −45% of cancer-related CNAs), compared with the beginning of ICI. All of the cancer-related CNAs concerned LOH. Interestingly, we found new loss of gene expressions concerning Wnt-related genes (LRP5, patient #6; SETD6, patient #8), gene associated with antigen presentation (B2M, patient #4), gene associated with recognition of tumor damage associated molecular patterns (TLR8, patient #4), tumor necrosis factor (TNF)-induced inflammation (GADD45B, patients #1, #3 and #8; TNFRSF17, patient #8), and LOH of immune checkpoint PD-L2 (PDCD1LG2, patient #8).

A summary of comparative molecular profiles is presented in table 3 and figure 1.