We analyzed treatment related adverse events and clinical variables of 46 patients with potentially resectable stage IIIA NSCLC in the neoadjuvant setting with chemoimmunotherapy included in the NADIM clinical trial (NCT03081689), an open-label, multicenter, single-arm phase II trial done at 18 hospitals in Spain.

Patients included in the study were treated with nivolumab (360 mg), paclitaxel (200 mg/m²) and carboplatin (area under the curve (AUC) 6; 6 mg/mL per min), on day 1 of each 21-day cycle, for three cycles before surgical resection.13 Main exclusion criteria were the presence of active autoimmune or infectious disease, ongoing systemic corticosteroid or other immunosuppressive therapy, history of symptomatic grade 3 or 4 interstitial lung disease, clinically significant concurrent malignancies and previous treatment with checkpoint inhibitors.

Adverse events and abnormal laboratory findings were graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events V.4.0. Investigators determined whether adverse events were treatment related according to the study protocol and standard regulatory requirements.

A central committee confirmed all Pn cases and considered adverse events related with neoadjuvant treatment those developed within 180 days after the administration of first dose of neoadjuvant nivolumab. Patients with Pn were considered to be those who had clinical and/or radiological findings of Pn of any grade (such as dyspnea and pleural pain), once other clinical entities had been ruled out (namely infectious diseases, pulmonary thromboembolism, or tumor progression, among others).

Clinical variables as well as molecular parameters (at diagnosis and after neoadjuvant treatment, along with their treatment variation) were compared between patients who developed Pn as an adverse effect to treatment and patients who did not, in order to search for markers associated with this irAE.

Basal PD-L1 tumor proportion score (TPS), tumor mutational burden (TMB) and specific mutations were retrieved from.13 Briefly, PD-L1 immunohistochemistry assay (22C3 pharmaDx, Code SK006; Dako, Glostrup, Denmark) was used to assess diagnostic PD-L1 TPS following manufacturer’s instructions.

TMB Library generation and sequencing of samples was performed on an Ion Chef System and S5 Sequencer. Specifically, 20 ng of extracted DNA was treated with heat labile uracil-DNA glycosylase and used for library preparation using the Oncomine Tumor Mutation Load Assay. Eight samples were loaded at 50pM and sequenced onto an Ion 540 chip. Reads were aligned to hg19 using Torrent Suite 5.12 and BAM files were transferred to Ion Reporter 5.12 for variant calling. TMB was computed using the TMB filter chain and the TMB algorithm 3.0. Germline variants were filtered out using a germline filter-chain based on population databases. Additionally, specific mutations using the Oncomine Variants 5.12 filter whose variant allele frequency was greater than 5%, reached ≥60 of coverage and a p≤0.05 were reported.

Patients had laboratory blood tests before each 21-day treatment cycle to monitor complete blood cell counts and biochemical parameters. Hemoglobin, leucocytes, neutrophils, eosinophils, lymphocytes, monocytes, platelets and LDH levels were retrieved from laboratory blood tests at two timepoints: prior treatment initiation and after neoadjuvant treatment but before surgery. Additionally, Lung Immune Prognostic Index (LIPI), neutrophil to lymphocyte ratio (NLR), monocyte to lymphocyte ratio (MLR), platelet to lymphocyte ratio (PLR), derived NLR (dNLR), and platelet to monocyte ratio (PMR) were calculated. dNLR was defined as neutrophils/(leucocytes-neutrophils). LIPI was estimated as the sum of two factors, dNLR greater than three and lactate dehydrogenase (LDH) greater than 333 IU/L, defining three groups (good, 0 factors; intermediate, 1 factor; poor, 2 factors).22

Cell and plasma isolation, immunophenotyping of peripheral blood mononuclear cells (PBMCs) and immunoassays for detecting cytokines in plasma were obtained and processed as previously described 23 . Briefly, 29 samples of peripheral blood mononuclear cells (PBMCs) at baseline or after neoadjuvant treatment, but presurgery, were isolated by lymphoprep gradient centrifugation and cryopreserved with freezing medium until use for flow cytometry analysis. Immunophenotyping of general and specific subpopulations of circulating natural killer (NK) cells (CD3-CD56+), T cells (CD3+), cytotoxic T cells (CD3+CD8+), helper T cells (CD3+CD4+), T-NK like cells (CD3+CD56+), B cells (CD3-CD19+) and monocytes (CD14+) with different activation and checkpoint markers was determined by multicolor panels on a MACS Quant 10 cytometer (Miltenyi Biotec) and analyzed with the FLOWJO software, as previously described.23 When data from different antibody panels were available for the same cell population median value among them was used. Information derived from anti-PD-1 clone PD1.3.1.3 labeling in posttreatment samples was not included in the analysis due to nivolumab binding competition to PD-1 receptor.

Plasma fraction of 30 patients at baseline and 34 patients after neoadjuvant treatment was collected after gradient centrifugation and stored diluted 1:2 with RPMI at −80°C. Levels of 200 cytokines related to cancer biomarkers were measured using G-Series Human Cytokine Antibody Array 4000 (RayBiotech) following manufacturer protocols.

Tissue and blood samples were used for the analysis of the T cell receptor (TCR) repertoire, both at diagnosis and after neoadjuvant treatment. TCR Library preparation and sequencing was carried out as previously described.24 Briefly, RNA input for PBMCs-derived libraries using the Oncomine TCR Beta–LR Assay was 25 ng, whereas for formalin-fixed, paraffin-embedded (FFPE) derived libraries using Oncomine TCR Beta–SR Assay, was 100 ng. For PBMCs-derived libraries, equal volumes from eight samples at 25 pM were pooled for sequencing on an Ion 530 chip. For FFPE-derived libraries, equal volumes from up to 32 samples at 25 pM were combined for sequencing on an Ion 540 chip.

Once the libraries were templated on an Ion Chef System, they were sequenced in the Ion GeneStudio S5 Series and analysis was done via Ion Reporter V.5.12. Convergence, Shannon’s diversity index and evenness were provided by Ion Reporter Software as a standard output. Dynamics of expanded and contracted clones in blood and tissue were calculated as percentages and clonal space of clones that increased or decreased their frequency after treatment, relative to pretreatment repertoire.

To test for clinical and molecular data categorical associations with Pn development, we performed a Fisher’s exact test, in variables such as sex, histology, smoking status, pathological and clinical response, number of cycles, specific mutations and LIPI score. Non-parametric Mann-Whitney U test was performed for clinical variable analysis (age and packs-year) and quantitative molecular parameters, such as blood counts, cytokines, TCR repertoire and immune populations. To analyze changes related to neoadjuvant treatment specific of Pn development, differences between paired postsamples and presamples were calculated for each patient group and Mann-Whitney U test was performed. Kaplan-Meier survival analysis was performed using the log-rank test in order to identify difference between patients who developed pneumonitis and those who did not. The receiver operating characteristic (ROC) curve analysis was calculated to predict the potential value of the parameters as predictors of pneumonitis development, and AUC values over 0.75 were considered good enough predictors to be included in this study. Since the research was designed as a discovery study, p values were not adjusted in order to maximize the finding of new biomarkers and the generation of new hypothesis. P<0.05 was considered statistically significant for all tests. Finally, the post hoc nature of the analyses, together with the large number of variables analyzed and the small number of patients, requires confirmation of the biomarkers found in independent cohorts.

The cohort of patients included 46 patients with NSCLC stage IIIA, according to American Joint Committee on Cancer seventh edition criteria, recruited across 18 centers in Spain between April 26, 2017 and Aug 25, 2018. All of them were treated with neoadjuvant chemoimmunotherapy according to protocol described in Material and Methods.

At the time of the data cut-off, with a median follow-up of 24.0 months (IQR 21.4–28.1), 12/46 patients (26.1%) developed a clinical and/or radiological treatment related Pn within 180 days after receiving the first dose of nivolumab. The median to Pn development was 109.5 day (IQR 58–130).

We found no difference between patients who developed Pn and those who did not regarding sex (p=1.000), age (p=0.910), smoking status (p=0.749), packs-year (p=0.196), histological subtype (p=0.316), or clinical response (p=0.836). No significant differences were found between patients who had complete pathological response (CPR) and those who had not (p=0.480) or those who had CPR or major pathological response and those with incomplete pathological response (p=0.398). We found no differences in progression-free survival (p=0.761) or overall survival (p=0.726) between patients who developed Pn and those who did not. No association between number of nivolumab cycles received and Pn development was found (p=0.409) (table 1).

We also wanted to investigate any relationship between Pn development and molecular characteristics of the tumors, specifically, PD-L1 TPS, TMB or specific baseline mutations. 28 of 46 patients had valid data for PD-L1 TPS (six developed Pn), and 29 had valid data for baseline mutation analysis (five developed Pn).

No association between PD-L1 TPS (Pn 45%, IQR 0%–100%; Non-Pn 32.5%, IQR 0%–75%; p=0.491) or TMB (Pn 7.6, IQR 4.2–55.8; Non-Pn 5.9, IQR 3.6–9.0; p=0.298) and Pn development were observed (figure 1A). Mutations in KEAP1, ARID1A, RB1, HNF1A, TP53 or KRAS genes did not show significant association to Pn development. However, despite the low number of cases, the presence of ARID1A mutations showed a trend to be associated to Pn development (p=0.068). Out of three patients who presented the mutation, two developed Pn (66.7%), while only three cases of Pn were described in the 26 non-mutated patients (11.5%) (figure 1B and table 2).

Other less frequent mutations found in our cohort of patients (n<3 cases) were STK11 (one mutated patient who did not develop Pn), EGFR (one mutated patient who did not develop Pn) and BRAF (one mutated patient who developed Pn). No mutations were found in RET, ROS, MET, or ALK.

We further sought to identify Pn associated biomarkers in patient cell blood counts at baseline (n=46) or postneoadjuvant treatment (n=45). Different standard parameters in the hemograms from all patients were analyzed, as well as the ratios and scores derived from this data.

No significant differences were observed for hemoglobin, platelets, eosinophils, basophils, lymphocytes and LDH. (online supplemental figure S1A and online supplemental table 1). At diagnosis, only a trend to higher monocyte counts in patients who develop Pn, was observed (Pn 0.75, IQR 0.63–1.15; non-Pn 0.70, IQR 0.50–0.84; p=0.079) (figure 2A). However, in postneoadjuvant treatment samples, patients who developed Pn showed increased levels of neutrophils (Pn 4.31, IQR 3.38–4.84; non-Pn 3.1, IQR 2.56–4.15; p=0.041), as well as, a trend to higher leukocyte counts (Pn 7.61, IQR 6.16–8.27; non-Pn 5.95, IQR 4.8–7.11; p=0.063) (figure 2A).

Related to blood counts ratios, there were no significant differences in NLR, dNLR, MLR, and LIPI between patients with and without Pn (online supplemental figure S1B). However, patients who developed Pn showed a trend to lower platelet-to-lymphocyte ratio (PLR; Pn 105.3, IQR 70.8–161.8; non-Pn 146.9, IQR 103.8–214.2; p=0.054) and statistically significant lower platelet-to-monocyte ratio, at baseline(PMR; Pn 317.3, IQR 204.6–443.8; non-Pn 436.0, IQR 353.9–597.0; p=0.012). These differences were reduced after neo-adjuvant treatment (figure 2B).

To identify candidate immune populations to serve as peripheral blood biomarkers, we performed flow cytometry analysis in preneoadjuvant and postneoadjuvant samples from 29 patients, of which 8 developed Pn. We characterized general monocytes, NK, T and B cells as well as additional immune subpopulations (online supplemental table 2).

Similarly, as described above for blood counts, we saw higher percentage of monocytes in pretreatment samples, characterized by CD14 expression detected by flow cytometry, in patients who had developed Pn (Pn 36.6, IQR 26.6–43.4; non-Pn 25.7, IQR 17.4–35.7; p=0.017) (figure 3A). The percentage of CD14+ population at diagnosis could serve as a good predictive biomarker since its area under the ROC curve value was 0.791 (95% CI 0.613 to 0.970) (figure 3A). Nevertheless, deepening in the different monocyte subsets, we did not see significant differences between classical (CD14+CD16−; Pn 84.0, IQR 60.7–88.6; non-Pn 78.9, IQR 57.7–88.1; p=0.591), intermediate (CD14+CD16+; Pn 2.8, IQR 1.7–5.7; non-Pn 2.0, IQR 1.2–4.6; p=0.283) and non-classical (CD14-CD16+; Pn 2.1, IQR 1.0–6.2; non-Pn 1.5, IQR 0.6–2.9; p=0.317) monocytes (online supplemental figure S2A). In postneoadjuvant treatment samples, there was no statistical significance, neither in CD14+ general monocytes (figure 3A), nor in their different subpopulations (online supplemental figure S2A).

Regarding B cells, Pn patients had a trend to lower percentage of CD3-CD19+B cell population in pretreatment samples (Pn 12.6, IQR 7.9–23.7; non-Pn 22.5, IQR 15.1–38.8; p=0.064), that reached statistical significance after neoadjuvant treatment (Pn 5.44, IQR 3.43–11.94; non-Pn 24, IQR 14.15–35.15; p=0.002) (figure 3B). However, we did not see any differences regarding to CD25+ and CTLA4+ B cells subpopulations in both pretreatment and posttreatment samples (online supplemental figure S2B).

Patients who had developed Pn showed a significant higher percentage of CD3-CD56+ NK cells at diagnosis (Pn 11.2, IQR 2.8–15.0; non-Pn 3.3, 1.4–6.9; p=0.019), and after neoadjuvant treatment (Pn 6.57, IQR 4.66–15.3; non-Pn 2.74, IQR 1.33–4.79; p=0.005) (figure 3C). Among NK cells, we detected that PD1− percentages were elevated in patients who develop Pn (Pn 76.8, IQR 75.5–80.9; non-Pn 73.6, IQR 66.3–76.7; p=0.032), but no differences in PD1+ cells were observed in these patients (Pn 2.5, IQR 0.8–5.2; non-Pn 1.4, IQR 0.4–3.1; p=0.407) (figure 3C). Their AUC values for ROC curves were 0.786 (95% CI 0.581 to 0.990) for general NK cells and 0.762 (95% CI 0.577 to 0.947) for PD1− subset (figure 3D).

Related to T cells, we found a trend to lower percentage of CD3+ general T cells at diagnosis in patients with Pn (Pn 54.6, IQR 51.0–66.3; non-Pn 67.9, IQR 57.6–74.0; p=0.064), that reached statistical significance after neoadjuvant treatment (Pn 50.3, IQR 38.21–65.10; non-Pn 70.2, IQR 63.45–76.03; p=0.001) (figure 4A). In addition, patients who developed Pn suffered a greater reduction in the percentage of CD3+ T cells with treatment (Pn −8.18, IQR −13.68–5.73; non-Pn 3.9, IQR −4.9–16.63; p=0.045) (figure 4A). Looking deeper in T cells subpopulations, we found no significant differences neither in CD3+CD4+ helper T cells nor CD3+CD8+ pretreatment or posttreatment samples. However, in pretreatment samples Pn patients had a trend to lower percentage of CD3+CD4+PD1+ T helper subpopulation (Pn 1.2, IQR 0.4–4.9; non-Pn 5.2, IQR 1.3–16.6; p=0.088) that was completely absent for CD3+CD8+PD1+ T cytotoxic subpopulation (Pn 1.8, IQR 0.0–8.0; non-Pn 2.9, IQR 0.3–10.4; p=0.329) (online supplemental figure S3A).

In addition to immunophenotyping, we characterized the T cell repertoire by massive TCR sequencing, in both, tumor (22 samples at diagnosis, 4 Pn; and 38 samples at surgery, 12 Pn) and blood (30 pretreatment samples, 8 Pn; and 35 postneoadjuvant samples, 10 Pn). We found no differences in TCR diversity, convergence, and expanded or contracted clones, between patients who developed Pn and those who did not, for any of the timepoints or compartments (online supplemental figure S3B and online supplemental table 3). However, we found significant lower TCR evenness levels in postneoadjuvant tissue samples (Pn 0.89: IQR 0.84–0.90; non-Pn 0.90, IQR 0.88–0.94; p=0.028) (figure 4B).

In order to describe additional blood biomarkers, we performed further protein analysis in baseline (30 patients, 8 developed Pn) and postneoadjuvant (34 patients, 10 developed Pn) plasma samples. Among protein measured (online supplemental table 4), significantly lower baseline levels of teratocarcinoma-derived growth factor 1 (TDGF1) (Pn 582.5, IQR 272.0–1150.4; non-Pn 2174.6, IQR 775.8–3289.8; p=0.013) were found in patients who develop Pn (figure 5A). On the contrary, patients who develop Pn showed significantly higher levels of plasmatic macrophage stimulating protein (MSP) (Pn 1564.1, IQR 1477.2–1599.9; non-Pn 1358.6, IQR 1144.4–1521.2; p=0.006), Poly(A)-specific ribonuclease (PARN) (Pn 774.5, IQR 482.0–1031.3; non-Pn 480.6, IQR 251.0–653.0; p=0.017) and E-cadherin (Pn 1321.5, IQR 843.9–2076.7; non-Pn 683.2, IQR 555.5–947.3; p=0.022) at diagnosis. Also, a trend for higher CCL22 levels, also known as macrophage-derived chemokine (MDC), was found in patients who developed pneumonitis (Pn 2281.8, IQR 1342.3–3168.0; non-Pn 941.7, IQR 612.5–1970.2; p=0.055) (figure 5A). However, these differences were lost after neoadjuvant treatment, only a trend to a lower TDGF1 (Pn 339.4, IQR 81.1–1821.6; non-Pn 1969.2; IQR 691.8–6402.1; p=0.076) and a trend to higher MDC (Pn 1784.3, IQR 1243.9–3051.0; non-Pn 1132.8, IQR 606.1–1778.7; p=0.089) were observed (figure 5A). Referring to postneoadjuvant treatment samples, Pn patients showed significant higher relative levels of proinflammatory cytokines such as angiopoietin-1 (ANG-1), epithelial growth factor (EGF), Chemokine (C-C motif) ligand 16 (CCL16), neutrophil gelatinase-associated lipocalin (NGAL), macrophage migration inhibitory factor (MIF), platelet-derived growth factor (PDGF-AA), bone morphogenetic protein 4 (BMP-4), beta nerve growth factor (b-NGF), hepatocyte growth factor (HGF), neurotrophin-4 (NT-4), urokinase-type plasminogen activator receptor (uPAR) and vascular endothelial growth factor (VEGF) (online supplemental figure S4A). Likewise, a differential impact of treatment on the plasma levels of latency-associated peptide (LAP), chemokine (C-X-C motif) ligand 1, 5, 6 (CXCL1, CXCL5, CXCL6), Cathepsin S, CCL4 (MIP-1b), EGF, and metalloproteinase inhibitor 1 (TIMP-1) was observed between patients with and without Pn (online supplemental figure S5A).

AUC values for ROC curves to predict Pn in baseline samples were 0.801 (95% CI 0.645 to 0.958) for TDGF1; 0.838 (95% CI 0.695 to 0.982) for MSP; 0.790 (95% CI 0.604 to 0.976) for PARN; and 0.788 (95% CI 0.566 to 0.991) for E-Cadherin (figure 5B).