Gut metabolomics profiling of non-small cell lung cancer (NSCLC) patients under immunotherapy treatment

From April 2016 to March 2017 a cohort of 11 NSCLC patients aged 44 to 82 years (median age 68 years; 8 males and 3 females) were recruited at the Department of Clinical and Molecular Medicine, Sant’ Andrea Hospital, Sapienza University of Rome.

As inclusion criteria the following were considered: adult subjects, age > 18 years; NSCLC diagnosed by histology; Eastern Cooperative Oncology Group (ECOG) performance status ≤ 2; anti-PD-1 nivolumab employed as second-line treatment; acceptable pulmonary, cardiac, liver renal, and bone marrow functions.

As exclusion criteria the following were considered: autoimmune diseases; symptomatic interstitial lung disease and other noteworthy comorbidities; systemic immunosuppression; previous treatment with immune-stimulatory antitumor agents, including checkpoint-targeted molecules.

Nivolumab was proven at ordinary dosage of 3 mg/kg each 2 weeks till disease progression or undesirable toxicity. Radiological response was assessed with Response Evaluation Criteria in Solid Tumors (i-RECIST) Criteria and classified according to disease control (complete response, partial response and stable disease) and progressive disease. Toxicity was reported consistently with National Cancer Institute Common Terminology Criteria for Adverse Events (version 4.0) and toxicity valuation was performed at day 1 of every cycle until end of cure. The PFS was defined as the time from patient registration on clinical trial until the first documented tumor progression or death from any cause. The OS was defined as the time from patient registration to death from any cause. We defined as early progressors (EPs), patients experiencing disease progression within 3 months from the beginning of nivolumab, and long responders (LRs) patients presenting PFS longer than 12 months. The study was conducted according to good clinical practice guidelines and Helsinki declaration. The final version of the protocol was approved by the Institutional Ethics Committee (Ethical Committee n 4421, “Sapienza University”). Patients gave informed consent according to the guidance of the hospital ethics committee and with the approval of regulatory agencies. The fecal samples collected from NSCLC patients for gut microbiota metabolome profiling, were handled and processed for biobanking and integration processing at the OPBG Human Microbiome Unit and NMR-based Metabolomics Laboratory, Sapienza University of Rome.

Genomic DNA from stool samples was manually extracted using QIAmp Fast DNA Stool mini kit (Qiagen, Germany), according to the manufacturer’s instructions.

Amplification of the variable region V3–V4 from the 16S rRNA gene (~ 460 bp) was carried out using the primer pairs described in the MiSeq rRNA Amplicon Sequencing protocol (Illumina, San Diego, CA).

The so obtained DNA amplicons were then cleaned-up by AMPure XP beads (Beckman Coulter Inc., Beverly, MA, USA). After second amplification step using a unique combination of bar-coded Illumina Nextera forward and reverse adaptor-primers, the final library was cleaned-up using 50 μL of AMPure XP beads and quantified using Quant-iT™ PicoGreen^® dsDNA Assay Kit (Thermo Fisher Scientific, Waltham, MA). Finally, the library was diluted in equimolar concentrations (4 nM), pooled together and sequenced on an Illumina MiSeqTM platform according to the manufacturer’s specifications.

Paired-ends reads were trimmed for their quality, read length and chimera presence using Qiime v1.8. pipeline [24]. Sequences were organized into Operational Taxonomic Units (OTUs) with a 97% of clustering threshold of pairwise identity and representative sequences were aligned using PyNAST v.0.1. software [25] against Greengenes 13_08 database [26] with a 97% threshold of similarity. Ecological and statistical analyses were performed using phyloseq and DESeq2 packages from R software [27, 28].

Fecal water was obtained as previously described [31]. After sample collection, 2 out of 11 NSCLC samples were excluded for insufficient sample collection. Briefly, ¹H-NMR spectroscopy analysis was performed on 500 mg of feces suspended with 1 mL of D₂O–PBS–NaN₃ buffered solution. Each sample was vortexed for 2 min and then centrifuged for 25 min at 10,000 rpm and 4 °C to obtain fecal water. In total, 600 μL of supernatant was collected and analyzed according to Brasili et al. [31] NMR analysis was carried out at 298 K by using a Bruker Advance 400 spectroscope (Bruker BioSpin GmbH, Rheinstetten, Germany), equipped with a magnet operating at 9.4 Tesla (400.13 MHz for ¹H frequency). 1D ¹H NMR experiment was performed employing the standard presaturation presat pulse sequence. Spectral width was set to 6009 Hz (15 ppm) and 64 scans were collected for each spectrum with a presaturation pulse length of 2.00 s and a relaxation delay of 6.55 s. The spectra were collected with 64 K points for an acquisition time of 5.5 s [32]. The assignment of resonances was done by 2D homonuclear NMR Total Correlated Spectroscopy (TOCSY) and heteronuclear Single Quantum Coherence (HSQC) experiments. TOCSY experiments were recorded at 298 K with a spectral width of 15 ppm in both dimensions, using 8 K × 256 data points matrix, repetition time of 2 s and 80 scans with a mixing time of 110 ms. HSQC experiments were acquired with a spectral width of 12 ppm in proton dimension and 200 ppm in the carbon dimension, using 8 K × 256 data points matrix for the proton and the carbon dimensions, respectively, with a repetition delay of 2 s and 96 scans. The assignment was confirmed according to Human Metabolome Data Base [33] and own laboratory database. 1D ¹H NMR spectra were processed and quantified by using the ACD Lab 1D-NMR Manager 12.0 software (Advanced Chemistry Development, Inc., Toronto, ON, Canada), whereas the MestReC software (Mestrelab Research SL, Santiago de Compostela, Spain) was used to assess 2D-NMR spectra. The quantification of metabolites was obtained by comparison of the integrals (normalized for number of protons) of specific signals with the IS trimethyl silyl propanoic acid (TSP) integral and then normalized for feces weight (expressed as µmol/g).

To characterize the differences between NSCLC subjects from a single- and a multi-omic standpoint (i.e., ¹H-NMR- and GC–MS-based metabolomics), a data analytical strategy based on the use of multivariate chemometric methods were carried out on the integrated dataset of 9 patients considering both GC–MS and ¹H-NMR data. However, in order to have reliable results, prior to data processing, the raw data matrix of GC–MS/SPME metabolites was cleaned by retaining only those molecules which were detected in at least 80% of the investigated subjects. Such a screening was not needed in the case of ¹H-NMR data. Since the two blocks of data came from different experimental techniques and had different variances and because of high inherent variability, data were block scaled after individual autoscaling. Metabolomics multivariate data analysis was carried out using in house written functions running under MATLAB (R2015b; The Mathworks, Natick, MA) environment. Principal components analysis (PCA) was used to analyze inherent clustering, to identify outliers and significant metabolites. Mann–Whitney U test was then applied to assign significant differences at the level of single metabolite in the PCA model in particular, a p-value of 0.05 was considered as threshold for statistical significance.

Eleven patients with stage IV NSCLC treated with second-line nivolumab were enrolled in this study. Baseline clinical characteristics of patients are summarized in Table 1. Among them, 10 patients had squamous-cell carcinoma and one adenocarcinoma. Median PFS and OS were 7.5 and 7.7 months, respectively. Four patients were EPs, while 7 LRs (Table 1).

The metagenomics data have been analyzed in a preliminary way and no cluster formation has been highlighted between EPs and LRs in the structure of microbiota (Additional file 1: Figure S1).

In particular, the analysis performed by Bray–Curtis dissimilarity (Additional file 1: Figure S1, panel A) does not highlight the existence of a specific cluster related to the clinical condition of EP or LR. Furthermore, this is confirmed in the distribution of the main bacterial genera for each sample (Additional file 1: Figure S1, panel B).

Each fecal sample was analysed to determine both VOCs and non-volatile metabolites, in an untargeted fashion, thus capturing large numbers of known and uncharacterized metabolites, including those of potential microbial origin.

Two-hundred and twenty-three VOCs for all 11 NSCLC patients were identified and quantified by GC–MS/SPME, and grouped into the following 18 chemical classes: alcohols (n 44); esters (n 31); aldehydes (n 21); ketones (n 35); alkenes (n 12); alkanes (n 17); acids (n 8); phenols (n 4); terpenes (n 18); sulfur compounds (n 2); hydrazine (n 1); azetidine (n 1); indoles (n 7); pyridine (n 1); amines (n 13); furans (n 2); pyrazine (n 2) and aromatic hydrocarbons (n 4) (data not shown). After data reduction, a matrix with 24 metabolites detected in the 80% of samples was considered for computation (Additional file 2: Table S1).

Fourty-nine non volatile metabolites were detected (i.e., acids, amino acids, amines, and sugars) and quantified (µmol/g) by ¹H-NMR for 9 NSCLC patients (Additional file 3: Table S2).

In order to have a multi-omics overview, the integration of GC–MS/SPME and ¹H-NMR data was carried out through a low-level data fusion approach. Therefore, PCA was applied to the matrix obtained by concatenation of GC–MS/SPME and ¹H-NMR data after block-scaling. The PCA scores plot (PC1 vs. PC2; PC1: 28.82%; PC2: 17.80%), displayed in Fig. 1, panel a, showed a clear separation between EP and LR patients (Fig. 1, panel a). In particular, the separation between the two groups occurred mainly along the axis of PC1; therefore, in order to assess which metabolites mainly contributes to the observed separation, only the variables having the highest absolute value of the loadings on such component were considered (Fig. 1, panel b).

Interestingly, the levels of tridecane could be measured only in fecal metabolome of EPs patients, being absent in that of LR ones. Particularly, the differences between levels of propionic acid (p value = 0.016), lysine (p value = 0.032), nicotinic acid (p value = 0.016), tridecane (p value = 0.032) 2-pentanone (p value = 0.016) and p-cresol (p value = 0.016) were statistically significant, as inferred by U-Mann–Whitney test. PCA model took into account the covariance of propionic acid with the other SCFAs (Fig. 2).