Serum metabolic profiling identified a distinct metabolic signature in patients with idiopathic pulmonary fibrosis – a potential biomarker role for LysoPC

For the pilot and validation study, serum samples of patients with stable IPF (n = 10, n = 11 respectively) and control subjects (n = 10, n = 11 respectively) were obtained from the Department for Pulmonary Medicine, University Hospital, Inselspital, Bern, Switzerland and stored by −80 °C until analysis. IPF was diagnosed according to the current international guidelines of 2011 [1]. High resolution chest CT scans of IPF patients were evaluated by a radiologist experienced in ILD diagnosis. If necessary, a histological evaluation including face-to-face multidisciplinary discussion with the treating pulmonologists, pathologists and radiologists was done. Clinical baseline characteristics, including smoking habits and lung function, were assessed and documented. Part of the serum samples from IPF patients have been obtained within the clinical trial NCT02173145 at baseline and from the idiopathic interstitial pneumonia (IIP) cohort study. The study protocol was approved by the local Ethics Committee (Swiss Ethics Committee, Bern (KEK 002/14 and 246/15 or PB_2016–01524)) and all participants signed written informed consent to participate.

Serum metabolites were extracted with organic solvent for protein precipitation. Frozen serum was gently thawed at 4 °C. 250 μL of −20 °C cold methanol:acetonitrile (1:1, v/v) containing chlorpropamide as internal standard (1 μg/mL) was added to 50 μL serum in a 1.5 mL CapLock tube. Samples were shortly vortexed, stored at −80 °C for 30 min and then centrifuged twice at 14′000 g for 20 min at 4 °C. The supernatants were transferred into total recovery liquid chromatography – mass spectrometry (LC-MS) vials. In addition to the serum samples, a blank sample was similarly extracted (water collected in a similar CapLock tube used for sample preparation). Pooled group samples were generated by combining 50 μL of each samples from the same group. A pooled quality control (QC) sample was generated by mixing 50 μL of the pooled group samples. The final serum extracts were stored at 6 °C until analysis.

The serum extracts were analyzed in a randomized block design order on a high resolution mass spectrometer (Synapt G2-S HDMS, Waters, Milford, MA, USA) coupled to a 2D UPLC Acquity I-Class system (Waters, Milford, MA, USA). Two μL of the extracted metabolites were separated on a ACQUITY UPLC HSS T3 column (1.0 mm × 100 mm, 1.8 μm, Waters) at a flow rate of 0.17 mL/min. The mobile phase was composed of 0.1% (v/v) formic acid in (A) 1% (v/v) methanol in H₂O and (B) methanol. The following LC gradient was applied: 0 min: 100% mobile phase A, 1 min: 100% A, 11 min: 1% A, 13.0 min: 1% A, 13.1 min: 100% A, 15 min: 100% A. The column temperature was set to 50 °C and the autosampler to 6 °C. Standard mass spectrometric parameters were 0.5 kV and 20 V for capillary voltage and cone voltage, respectively. Desolvation and source temperature were kept at 450 °C and 120 °C, respectively. Cone and desolvation flows were set to 150 L/h and 800 L/h, respectively. Leucine-Enkephalin ([M + H]⁺m/z = 556.2766, [M-H]⁻m/z = 554.2620) was acquired every 20 s for lock mass correction. Mass spectra were acquired at a scan time of 0.2 s in the MS^E resolution mode between m/z 50 and 1200. In addition, data-dependent acquisition (DDA) MS/MS experiments were performed on the pooled group and QC samples. Before sample analysis, the system was equilibrated by injecting 10 times the QC sample and the QC sample was then analyzed after every fifth sample injection. A system suitability test (SST) containing standards was measured at the beginning and the end of the analytical run to ensure retention time stability, intensity stability and mass error ≤ 8 ppm. The instrument was controlled via MassLynx (version 4.1, Waters). The pilot study set was analyzed with one replicate per IPF patient and healthy control sample, whereas for the validation set, duplicate analysis was applied.

To improve structural elucidation of the regulated metabolites, traveling wave ion mobility mass spectrometry (TWIM-MS) of extracted serum was additionally performed. Extracted serum was separated under the same chromatographic conditions described above. Mass spectra were acquired in the positive and negative ionization mode at a scan time of 0.5 s in the HDMS^E resolution mode between m/z 50 and 1200. The TWIM wave height was set at 40 V and the wave velocity was ramped from 1000 to 300 m/s. Precursors were fragmented after the TWIM separation in the transfer T-cell set and transferred to the TOF at a velocity of 220 m/s. The ion mobility was calibrated with poly-DL-alanine, which allowed the determination of the collision cross section (CCS) in nitrogen [16]. A lipid standard mixture of phosphatidylcholines (PC(18:0/0:0), PC(18:2/0:0), PC(12:0/12:0), PC(14:0/14:0), PC(18:1/18:1), PC(16:0/0:0) obtained from Echelon Biosciences) was prepared in acetonitrile:isopropanol:water (2:1:1, v/v) and used as quality control for the CCS measurements. Two μL of this lipid standard mixture at 0.1 and 1.0 μg/mL were analyzed by UHPLC-TWIM-MS in the positive and negative ESI mode, respectively. Data processing was performed using UNIFI (version 1.8.2.169, Waters, Millford, USA).

After lock-mass correction and chromatographic alignment with Progenesis QI (version 2.2, Nonlinear Dynamics, Newcastle, UK), ion patterns were deconvoluted between 0.45–11.5 min. As possible ions, [M + H]⁺, [M + Na]⁺, [M + K]⁺, [M + H-H₂O]⁺, [M + H-2H₂O]⁺, [M + 2H]²⁺, [M + 3H]³⁺, [2M + H]⁺, [2 M + Na]⁺ and [2M + K]⁺ were defined in the positive ion mode, [M-H]⁻, [M-H-H₂O]⁻, [M + HCOOH-H]⁻, [M-2H]²⁻, [M-3H]³⁻ and [2M-H]⁻ in the negative ion mode. The abundances of the features were normalized to all compounds. Features eluting between 0.5–10.5 min with a chromatographic peak width ≥ 0.05 min, having a coefficient of variation of the peak area ≤ 30% in the QC samples, singly charged ions and ions having a maximum abundance across all samples ≥200 (positive mode, arbitrary threshold) or ≥150 (negative mode, arbitrary threshold), were selected for further evaluation. In addition, the metabolic features having an abundance in blank higher than in QC and which were at least twice more abundant in blank were excluded. Known drug/drug metabolites taken from patient were also excluded from the dataset.

The normalized abundances of each filtered metabolic features were subjected to multivariate and univariate statistical analysis. For the validation set, the features abundances of the duplicate analysis were averaged. For multivariate statistical analysis, data were subjected to Pareto scaling prior to orthogonal projections of latent structures discriminant analysis (OPLS-DA) with SIMCA (version 14, Umetrics, Umeå, Sweden). Results were further integrated if the n-fold (n = total number of sample, leave-one-out cross validation approach) cross-validated correlation Q²(cum) was ≥0.5, if the cross validatory ANOVA p-value was ≤0.05 and if the permutation test was passed (20 permutations, R² > R²_permutation and Q²_permutation < 0). Discriminative features were selected based on a VIP_PRED score ≥ 1.5 and absolute p (corr) ≥ 0.5. The confidence interval of the selected features was also not including 0. With this approach, only metabolic features with strong model contribution and high reliability were selected. Univariate statistical analysis was performed to consistently evaluate the filtered metabolic features and to screen for additional metabolites. For the pilot study, univariate analysis of variance (ANOVA) was based on arcsinh transformed data from Progenesis QI. For the validation study, ANOVA was performed with SIMCA and was based on log2 transformed and Pareto scaled normalized abundances. The false discovery rate (FDR) due to multiple testing was estimated according to Benjamini Hochberg [17]. FDR corrected p-values (i.e. q-value) threshold was set at q ≤ 0.05.

Structure and formula of the isolated discriminative metabolic features were searched against the Human Metabolome Database (HMDB, version 3.6) and LIPID MAPS Structure Database (LMSD, version from December 6, 2016) with a mass accuracy of 8 ppm. Metabolic features with only one adduct ion were assumed to be [M + H]⁺ and [M-H]⁻ ions in the positive and negative ionization mode, respectively.

In the pilot study, 10 IPF patients and 10 age- and sex-matched healthy controls were included. The mean age of IPF patients and of healthy controls was 67.8 respectively 68.7 years. Ninety percent of all participants were male in both groups. Lung functional measurements showed a reduced median forced vital capacity (FVC) of 2.4 L (mean 65% predicted) in IPF patients and a median FVC of 4.1 L (mean 115% predicted) in controls. Average 3-year mortality prediction in IPF patients was in 42.1% according to the currently used GAP staging method (stage II) [2]. Eighty percent of IPF patients and controls had a history of cigarette smoking.

For the validation study, blood sampling from 7 IPF patients was performed 2 weeks after the initial pilot study. Four recently diagnosed IPF patients and all healthy controls were newly collected for the validation study. One sample of the control group was excluded due to analytical problems during sample preparation. Eleven IPF patients and 10 healthy controls remained for further analysis. The baseline characteristics of patients and controls of both pilot and validation study are summarized in Table 1.

The differentially regulated serum metabolites in healthy participants and IPF patients were searched for by non-targeted metabolic profiling using UHPLC-HRMS. The flowchart for sample preparation and analysis is illustrated in Fig. 1. After data processing and filtration, 2426 and 1740 metabolic features were detected in the pilot study by UHPLC-HRMS in the positive and negative electrospray ionization (ESI) mode, respectively. Principal component analysis (PCA) of the non-targeted metabolic profiles measured by UHPLC-HRMS suggested differences between healthy controls and IPF patients (Fig. 2). In the negative ESI mode, one IPF sample was defined as outlier and excluded from further statistical analysis, otherwise multivariate and univariate statistical analysis of the data did not isolate differentially regulated metabolic features (data not shown). Acceptable OPLS-DA model characteristics were obtained (Table 2) and based on the selection criteria, 67 and 103 discriminative features were selected by univariate and/or multivariate analysis in the positive and negative ESI mode, respectively (data not shown).

To validate our results, a new set of samples was profiled by UHPLC-HRMS. After data processing and filtration, 1229 and 1059 metabolic features were detected by UHPLC-HRMS in the positive and negative ESI mode, respectively. Based on the PCA score plots (data not shown), one healthy participant was considered as outlier and was further excluded from statistical analysis. The PCA score plots obtained after data filtration and exclusion of this participant are shown in Fig. 3. Similar to the pilot study, satisfactory multivariate model parameters were obtained in both ionization modes (Table 2). Based on our selection criteria, 103 and 364 discriminative features were selected by univariate and/or multivariate analysis in the positive and negative ESI mode, respectively (data not shown).

To identify common metabolic features regulated between healthy participants and IPF patients in the pilot and validation study, the retention time and m/z ratio of the discriminative metabolic features isolated in each study by multivariate and/or univariate analysis were compared. Among them, 3 and 27 metabolic features were commonly found in the positive and negative ionization mode, respectively. Based on their abundance profile (Additional file 1: Figure S1), we hypothesized that the metabolic features mainly detected in the IPF samples were related to drugs and/or drug metabolites. After exclusion of these features, 10 metabolic features remained differentially regulated in the pilot and were confirmed in the validation study. Their corresponding fold changes in abundance and potential annotations found against the metabolite databases are summarized in Table 3.

The identity of inosine was confirmed by comparing the retention times and fragmentation patterns obtained in serum to an authentic inosine standard. This purine metabolite was upregulated in the pilot study and downregulated in the validation study. Only three metabolic features eluting at 5.94 min (m/z = 332.242) and 9.59 min (m/z = 512.334 in the positive ESI mode and m/z = 556.325 in the negative ESI mode) were significantly upregulated in IPF patients in both studies.

The metabolic features eluting at 5.94 min was potentially assigned against the HMDB database to 3-hydroxydecanoyl carnitine (chemical formula C₁₇H₃₃NO_5, mass error − 2.7 ppm, isotope similarity 96%). The detection of the characteristic acylcarnitine fragment [C₄H₅O₂]⁺ at m/z 85.0284 was confirmed by UHPLC-TWIM-MS measurement (mass error − 4.8 ppm, Additional file 2: Figure S2). No significant differences were found in the global acylcarnitine profile (data not shown, profile extracted as in Bally et al. [18]).

At 9.59 min, three discriminative metabolic features were two times more abundant in the serum of IPF patients compared to healthy participants (Table 3, m/z = 512.334 in the positive ESI mode, m/z = 556.325 and 496.303 in the negative ESI mode, with the latter being significant only in one study). Presumably, these co-eluting ions belonged to the same metabolic features. In the negative ESI mode, the detected ion pattern agreed with a phosphatidylcholine (PC) or sphingomyelin [M-CH₃]⁻ and [M + CHOO]⁻ ions. These ion patterns were confirmed by the analysis of PC lipid standards (data not shown) and agreed with the literature [19, 20]. The MS/MS spectra acquired in the positive and negative ESI mode (DDA experiments, Additional file 3: Figure S3) further suggested the detection of a LysoPC with the chemical formula C₂₄H₅₀NO₈P (mass error − 1.0 ppm, isotope similarity 98%). No LysoPC structure matching the MS data was found against the LIPID MAPS database and the literature. Therefore, we assumed a LysoPC structure with an ether or hydroxylated acyl chain in the sn-1 or sn-2 position (Scheme 1). We further characterized this unknown LysoPC with TWIM-MS and measured a CCS of 227.5 Å² for the protonated ion (Table 4). Excellent CCS agreements with the literature (Table 4, CCS error < 1.1%) were obtained for the PC standard mixture confirming the excellent inter-laboratory reproducibility of CCS values [16]. A single TWIM peak was detected for the unknown LysoPC, which suggested the detection of a single LysoPC having a compacter conformation than PC(16:0/0:0). To our knowledge, the influence of ether or hydroxylated acyl chain on the LysoPC CCS has not been reported. It is therefore difficult to hypothesize the most probable structure for this unknown LysoPC based on the TWIM-MS measurement. The abundance profiles of all significantly upregulated metabolic features detected in the pilot and validation study (3-hydroxydecanoyl carnitine, and 2 metabolic features of LysoPC) are shown in Fig. 4.