Chromatof

GC × GC-TOF-MS and UPLC-IMS-QTOF-MS/MS data were subjected to a series of processing procedures, including baseline correction, denoising, smoothing, time-window splitting, deconvolution, and peak alignment using LECO Chroma TOF (LECO Corp.) and Progenesis QI (Waters Corp.) software, respectively.

GC-MS data were processed using the LECO ChromaTOF® software (LECO Corporation, MI, United States) as described earlier (Doerfler et al., 2013 (link)) with slight modifications. Peak areas of analytes were divided by extract weight and peak areas of internal standards. For analytes relatively quantified in runs injected with split rate 1:50, areas of phenyl-β-d-glucopyranoside were used and analytes quantified in runs with split rate 1:5 were normalized to pentaerythritol in respective runs. Peak annotation were done according to Metabolomics Standard Initiative (Goodacre et al., 2007 (link); Sumner et al., 2007 (link)) and stated in Supplementary Table 2.

Analysis of GC-MS raw data, including peak extraction, baseline adjustment, deconvolution, alignment, and integration, was carried out on Chroma TOF (V 4.3x, LECO) software. Metabolites identification was conducted by matching the mass spectrum and retention index of the samples with those of the standards available in the LECO-Fiehn Rtx5 database and NIST05 libraries.
All experiments were run in sextuplicate (n = 6). The ANOVA (one-way ANOVA) was conducted using SPSS 22.0 (IBM, Armonk, NY, USA). Principal component analysis (PCA) was carried out using SIMCA-13.0 software (Umetrics, Aubagne, France) to analyze the data of sourdough VOMs. The heatmap was created using “heatmap2” package in R. Bacterial composition analysis was performed using Majorbio Cloud Platform (www.majorbio.com, accessed on 14 June 2021), a free online platform.

Data processing was performed using ChromaTOF^® (version 4.51.6.0; LECO). A signal-to-noise ratio of 150 was used with a baseline offset of 0.8. The peak widths for the first and second dimensions were 30 s and 0.15 s, respectively. Analyte identification was carried out using the National Institute of Standards and Technology (NIST) Mass Spectral Library, where a match threshold of 80% was required. Analytes were aligned using the statistical compare tool within the software, where a spectral match threshold of 60% was required to match analytes across samples. After alignment, the analytes were normalised against the internal standard. Known artefacts and environmental contaminants were excluded, and a final compound table was created. Blank subtraction was subsequently done using the blank water samples (n = 5), whereby the analyte needed to be present in an abundance of more than 50% of the blank water samples to be retained. In addition, analytes that were not present in more than three of the biological replicates per sample type were removed. The resulting analytes were then normalised against the cell count (cell/mL; Supplementary Table S2). Compounds were grouped into their respective compound classes to investigate volatilome variability across the treatment and control groups.

Data were pre-processed and integrated according to a SOP as described previously (Dunn et al. 2011 (link)). UPLC–MS data were converted from the raw instrument datafile to NetCDF files and subsequently XCMS was applied for peak deconvolution and alignment separately for each analytical batch. Due to the untargeted nature of the UPLC–MS analysis, the number and identity of common peaks detected in each batch differed considerably. Thus, each of the 20 batched XCMS chromatographic peak-area data matrices consisted of N_b metabolite features (where b = 1…20; with N_b associated m/z and retention times) × 85 samples (60 subjects plus 25 integrated QC samples). GC–MS data were deconvolved and matched to a reference database of 259 metabolites applying ChromaTof (Leco) separately for each analytical batch. This produced 20 chromatographic peak-area data matrices of 259 metabolite features (with associated EI-MS spectrum and retention index) × 80 samples (60 subjects plus 20 integrated QC samples). If a given metabolite was not detected in a given batch then the associated matrix element was replaced with a missing value (NaN; not-a-number).