We included all consecutive patients with advanced non-small cell lung cancer treated with nivolumab (3 mg/kg d1d15, given intravenously) as second or later line therapy in the Department of Respiratory Diseases and Thoracic Oncology of the Hospital Ambroise Pare, Boulogne-Billancourt, France.

Demographic and pathological data were collected. Patient records were reviewed to determinate if they had been treated with oral or intravenous antibiotherapy two months before until one month after the beginning of nivolumab, defining EUA, given for any reason. The class of antibiotic, indication, route of administration and duration were collected. Tumor response was determined by Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1, based on CT scans every 4 injections, by an expert thoracic radiologist and validated in multi-disciplinary meeting. Overall response rate (ORR) was defined as the proportion of patients with complete or partial response. Clinical benefit was defined as nivolumab given more than 6 months.

Plasma citrulline concentration was prospectively measured at months (M) M0, M2, M4 and M6 by ion exchange chromatography. Low citrulline rate was defined as concentration less than 20 μM, and high citrulline rate as concentration equal or more than 20 μM.

Blood microbiome was assessed by sequencing of variable regions (V3-V4) of the 16S rRNA bacterial gene at M0 (for correlation with citrulline, tumor response and clinical benefit) and M2 (for correlation with EUA), (Vaiomer, Toulouse, France). Briefly, PCR amplification was performed using 16S universal primers targeting the V3-V4 region of the bacterial 16S ribosomal gene. The joint pair length was set to encompass 467 base pairs amplicon thanks to 2 × 300 paired-end MiSeq kit (V3). For each sample, a sequencing library was generated by addition of sequencing adapters. The detection of the sequencing fragments was performed using MiSeq Illumina® technology. The targeted metagenomic sequences from microbiota were analyzed using the bioinformatics pipeline established by Vaiomer from the FROGS guidelines. Operational taxonomic units (OTUs) were produced with single-linkage clustering and taxonomic assignment was performed in order to determine community profiles [21].

PDL1 immunohistochemistry (IHC) was performed using an automated method (Leica) and the E1L3N anti-PD-L1 antibody (Cell Signalling 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. The IHC was considered as being positive if at least one tumour cell out of 100 analyzed tumor cells was positively stained.

We compared continuous variables using Mann-Whitney test or Fisher test and categorical variables using Chi² test, according to EUA and citrulline rate (low versus high). ORR and clinical benefit rates comparisons were also performed by Chi² test, according to EUA and citrulline rate (low versus high). Concerning blood microbiome analyses, reads obtained from the MiSeq sequencing system have been processed using Vaiomer bioinformatics pipeline. The steps included quality-filtering, clustering into OTUs with the Swarm algorithm and taxonomic affiliation. Linear discriminant analysis (LDA) Effect Size (LEfSe) algorithm was used to identity statistically significant differences in microbiome composition according to clinical condition (tumor response, clinical benefit, citrulline rates, EUA) [22]. OS and progression-free survival (PFS) analyses were performed using Kaplan–Meier method (p-value calculated by log rank test). Multivariate analyses on OS and PFS were performed using Cox proportional hazards model. Statistical analyses (except blood microbiome analyses) were performed using Xlstat 2018 (Addinsoft, France).

Seventy-two patients were treated with nivolumab between July 2014 and September 2017. Median duration of nivolumab treatment was 87 days (interquartile range IQR 36–138). Median number of nivolumab injections was 6 (IQR IQR 3–9). At the end of the follow-up, 11 patients were still receiving nivolumab. Mean follow-up time was 500 days (IQR 401–599).

Patients were mostly male (62%), median age 68.8 year-old (IQR 62.7–73.7), former or current smokers (87%), PS 0–1 at the beginning of nivolumab (63%), with stage IV disease (86%) (Table 1). 99% of the patients were Caucasian (only one Asian patient was noted in the group who did not receive EUA). Forty-five patients (63%) had an adenocarcinoma and 62 (86%) received 1 line of treatment before nivolumab.Tab1

Characteristics of the patients, according to early use of antibiotics (EUA) and citrulline rates at baseline

PD-L1 status was available for 23 patients (32%). From these patients, 15 patients (65%) had positive (≥1%) PD-L1 staining, and 10 patients (43%) had high PD-L1 expression (≥50%).

ORR with nivolumab was 29% (n = 21) and clinical benefit rate was 28% (n = 20). Median PFS was 2.9 months (IQR 1.8–4.8) and median OS was 11.1 months (IQR 5.6-NR).

Thirty patients (42%) received EUA. Median duration of antibiotic treatment was 9.5 days (IQR 7–14). No significant clinical or histomolecular difference was observed between patients with or without EUA, except lower hemoglobin rate in case of EUA (Table 1). EUA was prescribed 51 times on the 30 patients (one patient could receive EUA several times). The most frequently prescribed antibiotics were beta-lactamines (n = 36/51) and vancomycine (n = 4/51) (Additional file 1: Figure S1). Oral administration was predominant (65%) and most common indications were respiratory infections (n = 22 on the 45 available indications), intestinal infections (n = 5/45), skin infections (n = 4/45) and catheter infections (n = 4/45) (Additional file 1: Figure S2). Thirty-three percent of the patients who received EUA were hospitalized (n = 10/30). ORR was 37% (n = 11) with EUA vs 24% (n = 10) without EUA (p = 0.276). Clinical benefit rate was 27% (n = 8) with EUA vs 29% (n = 12) without EUA (p = 0.859). Median PFS with and without EUA were respectively 2.8 months (IQR 1.4–5.1) and 3.3 months (IQR 1.8–7.3) (p = 0.249) (Fig. 1a). Median OS were respectively 5.1 months (IQR 3.4-not reached NR) and 13.4 months (IQR 10.6-NR) (p = 0.027) (Fig. 1b). Considering the use of antibiotics only during the month before nivolumab, median PFS was 1.8 months with antibiotics vs 3.0 months without (p = 0.341) and OS were 5.1 vs 13.3 months (p = 0.039), respectively.Fig1Fig. 1

Progression-free survival (a) and overall survival (b) according to early use of antibiotics (EUA)

Thirty-five patients accepted plasma collection for citrulline and blood microbiome analyses. Ten patients (29%) had low citrulline rates at baseline (< 20 μM). These patients had more often PS 2 (70%) than patients with high citrulline rates (12%) (p = 0.001), and lower hemoglobin rate (p = 0.006) (Table 1). Baseline citrulline tended be associated with ORR (52% with high citrulline vs 20% with low citrulline, p = 0.084) and was associated with clinical benefit (56% with high citrulline vs 0% with low citrulline, p = 0.002). Patients with clinical benefit of nivolumab had significantly higher plasmatic citrulline rates than other patients, at M0, M2 and M4, keeping high rates at M6 (Fig. 2).Fig2Fig. 2

Evolution of plasma citrulline concentrations during nivolumab treatment, according to clinical benefit

Blood microbiome analyses were feasible for all 35 patients at M0. Most bacteria were from the Proteobacteria phylum (39%), followed by Bacteroidetes (30%), Actinobacteria (20%) and Firmicutes (7%) (Additional file 1: Figure S3). This repartition was as expected in the literature [16].

Patients experiencing response and/or clinical benefit had different blood microbiome profile at M0 (Fig. 4, Additional file 1: Figure S4, Additional file 1: Figure S5). The detection of Peptostreptococcaceae (n = 6), Lewinella (n = 4), Paludibaculum (n = 5) and Holophagae (n = 4) at M0 was significantly associated with both clinical benefit and response. Blood microbiome of patients who did not experience clinical benefit and/or tumor response was significantly enriched in Gemmatimonadaceae at M0 (n = 7). The presence of Gemmatimonadaceae DNA in blood at M0 was associated with both low response rate (14% vs 50% for other patients, p = 0.09) and high rate of progression at M2 (86% vs 29% for other patients, p = 0.006).Fig4Fig. 4

Incidence of DNA detection in blood at baseline of bacteria associated with tumor response (a) and clinical benefit (b)

Multivariate analysis on PFS including EUA (yes vs no), PS (0–1 vs 2), hemoglobin rate (continuous variable), albumin rate (< 30 g/l vs ≥30 g/l), Kras mutation (yes vs no), PDL1 status (< 1 vs ≥1%), citrulline concentration at baseline (< 20 μm vs ≥20 μM) and presence of Gemmatimonadaceae on blood microbiome analysis at baseline was performed. Low citrulline rate (hazard ratio HR = 3.8; IC95% 1.4–99.9; p = 0.008), PS 2 (HR = 2.1; IC95% 1.1–4.1; p = 0.023), and hemoglobin rate (HR = 0.8; IC95% 0.0–0.9; p = 0.005) were independently associated with PFS (Table 2). The presence of Gemmatimonadaceae on blood microbiome analysis at baseline tended to be associated with worse PFS (HR = 2.9, IC95% 0.9–9.2; p = 0.073).Tab2

Multivariate analyses (Cox model) on PFS and OS

Multivariate analysis on OS including the same variables showed that EUA (HR = 2.2; IC95% 1.1–4.8; p = 0.038), Kras mutation (HR = 0.4; IC95% 0.0–0.9; p = 0.020) and presence of Gemmatimonadaceae on blood microbiome analysis at baseline (HR = 16.4; IC95% 3.9–68.5; p < 0.001) were independently associated with worse OS (Table 2).