Nutrients 2023 , 15 , 2688
6of 20
provided cooler bag with a cooler block. Samples received at the research center were immediately placed in a freezer at − 80 ◦ C. Genomic DNA was extracted using the ZymoBIOMICS™ 96 MagBead DNA kit (Zymo Research Corp., Irvine, CA, USA) integrating a double lysis (mechanical and chemical) on the Precellys Evolution homogenizer (Bertin Instruments, Montigny-le-Bretonneux, France). DNA extraction was performed on the KingFisher Flex automaton (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s instructions. Once obtained, the DNA solutions were assayed by fluorimetry with the Qubit device (Thermo Fisher Scientific, Waltham, MA, USA). 2.7.2. Libraries Preparation and Shotgun Metagenomic Sequencing Fragmentation of the extracted total DNA was performed using the FS DNA Library Prep Set kit (MGI Tech, Shenzhen, China). After ligation of adapters to each sample, the libraries generated were purified on magnetic beads. Library size was verified by capillary electrophoresis on at least 10% of samples. After quantification by fluorimetry, the libraries were normalized and pooled before denaturation, single-strand circularization, and sequencing using the DNBSEQ-G400 platform (MGI Tech). 2.7.3. Analysis of Overall Association and Taxonomic Profile The MiRKAT family of tests was used to assess overall association between taxonomic compositional profiles and treatment group [29–31]. These are regression-based association tests based on kernels that have been proposed specifically for microbiome data and allow covariate adjustment and repeated measurements. Jaccard and Bray–Curtis beta diversity scores were used to quantify dissimilarity between compositional profiles at several taxonomic ranks. Participant sex, age, and time since menopause were added as covariates. Assessment of microbiota components showing differential abundance between treat- ment groups was evaluated using CoDA-lasso enriched with stability analysis. CoDA-lasso is a multivariate approach that fits a regularized logistic regression model with an addi- tional constraint on the regression coefficient due to the compositional nature of relative abundance data [32]. The set of relevant taxa are those corresponding to non-zero coeffi- cients in the solution of CoDA-lasso. Since the approach can lead to some false positives, a stability analysis was also applied to solutions from CoDA-lasso [33]. It involves refitting the model several times on independent bootstrap samples of the data and picking only those components that are selected almost always, so to delete false positives detected by chance in just a few of the re-samplings. Such stability analysis was performed using 100 bootstrap re-samplings, each comprising 80 percent of the available samples. Only components selected in at least 90% of the replicates were chosen in the final result. The relevance of each selected taxon to discriminate between the treatment groups was investigated assessing variable importance in prediction with random forests. Each random forest comprised 500 non-pruned classification trees. Reported results comprised the selected taxa sorted according to their relative relevance in prediction, a sample estimate of the log-fold difference between the mean abundance of each group, and a heatmap of the taxa prevalence in each compared group. 2.7.4. Quantification of SCFA For the quantification of short-chain fatty acids (SCFA), fecal samples were divided in two aliquots, one for the lyophilization, and the second for a direct measure of the molecules of interest in order to obtain the dry weight-normalized absolute concentration. SCFA (acetic acid, propionic acid, butyric acid, valeric acid, caproic acid, isobutyric acid, isovaleric acid, and isocaproic acid) were measured on the second aliquot of test material via a gas chromatography–flame ionization detector (GC-FID) method as described by De Weirdt et al. [34].
Powered by FlippingBook