Study population and sample collection
We analyzed the samples collected during the CEREMI trial (ClinicalTrials.gov identifier NCT02659033), a prospective, randomized open-label clinical trial conducted at the Clinical Investigation Center of the Bichat-Claude Bernard Hospital (Paris, France) from March 2016 to August 2017. The trial was sponsored by Assistance Publique - Hȏpitaux de Paris and approved by French Health Authorities and by the Independent Ethics Committee Ȋle-de-France-1. All procedures were conducted in compliance with good clinical practice and the Declaration of Helsinki. Full details of the trial have been reported elsewhere [13].
Briefly, healthy volunteers of both genders aged between 18 and 65-year-old without exposure to antibiotics in the preceding 3 months nor the history of hospitalization in the last 6 months were prospectively included after obtention of their informed consent. A total of 22 healthy volunteers were randomly assigned (1:1 ratio) to receive for three days either 1 gram of ceftriaxone once a day (n = 11) or 1 gram of cefotaxime three times a day (n = 11). Antibiotic treatment was administered as 30-minute intravenous infusions. For each volunteer, 12 faecal samples were collected (Supplementary Figure S1): before the beginning of treatment at days − 15, -7 and − 1; every day during treatment at days 1, 2, and 3, and after the end of treatment at days 4, 7, 10, 15, 30 and 90.
Bacterial counts
Sample collected at days 15, -1, 4, 10, and 30 were analyzed to determine the total bacterial counts (Supplementary Figure S1). Aliquots containing 200 mg of faeces were diluted 200,000 times in a physiological solution (8.5 g/L NaCl). Samples were filtered for debris removal from faecal solutions using a sterile syringe filter (pore size 5 µm; Sartorius Stedim Biotech GmbH, Göttingen, Germany). Then, 1 mL of the microbial cell suspension obtained was stained with 1 µL SYBR Green I (1:100 dilution in dimethylsulfoxide; shaded 15 min incubation at 37°C; 10,000 concentrate, Thermo Fisher Scientific, Waltham, MA, USA). The flow cytometry analysis of the microbial cells present in the suspension was performed using a C6 Fortessa flow cytometer (BD Biosciences, Franklin Lakes, NJ, USA). Fluorescence events were monitored using the FITC filter 505LP 530/30 nm and perCP filter 635LP 695/40 nm optical detectors. Forward and sideways-scattered light was also collected. The BD Accuri CFlow software was used to gate and separate the microbial fluorescence events on the FL1–FL3 density plot from the faecal sample background. A threshold value of 200 was applied to FSC/SSC light. The gated fluorescence events were evaluated on the forward–sideways density plot, to exclude remaining background events and to obtain an accurate microbial cell count.
Metagenomic analysis of the bacterial microbiome
All samples were analysed through shotgun sequencing for bacterial microbiome analysis.
DNA extraction of stool samples and shotgun sequencing
DNA extraction from aliquots of all faecal samples was performed following IHMS SOP P7 V2 (Supplementary Figure S1) [35]. DNA was quantitated using Qubit Fluorometric Quantitation (ThermoFisher Scientific, Waltham, MA, USA) and qualified using DNA size profiling on a Fragment Analyzer (Agilent Technologies, Santa Clara, CA, USA). Three µg of high molecular weight DNA (> 10 kbp) was used to build the library. Shearing of DNA into fragments of approximately 150 bp was performed using an ultrasonicator (Covaris, Woburn, MA, USA) and DNA fragment library construction was performed using the Ion Plus Fragment Library and Ion Xpress Barcode Adapters Kits (ThermoFisher Scientific, Waltham, MA, USA). Purified and amplified DNA fragment libraries were sequenced using the Ion Proton Sequencer (ThermoFisher Scientific, Waltham, MA, USA), generating 22.2 ± 1.8 million reads of 150 bp (on average) per sample.
Microbial gene count table
To create the gene count table, the METEOR software was used [36]: first, reads were filtered for low-quality by AlienTrimmer [37]. Reads that aligned to the human genome (identity > 95%) were also discarded. The remaining reads were trimmed to 80 bases and mapped to the Integrated Gut Catalogue 2 (IGC2) [38], comprising 10.4 millions genes, using Bowtie2 [39]. The unique mapped reads (reads mapped to a unique gene in the catalogue) were attributed to their corresponding genes. The shared reads (reads that mapped with the same alignment score to multiple genes in the catalogue) were attributed according to the ratio of their unique mapping counts of the captured genes. The resulting count table was further processed using the R package MetaOMineR v1.31 [40]. To decrease technical bias due to different sequencing depths and avoid any artifacts of sample size on low-abundance genes, read counts were ‘rarefied’ using 20M high-quality reads (a threshold chosen to include all samples) using a random sampling procedure without replacement. The downsized matrix was finally normalized dividing gene read counts per gene length x100, as a proxy of gene coverage. Since gut microbiota has been found to be enriched in species from the oral cavity upon antibiotic treatment [41], the same process was repeated on an oral microbiota catalogue of 8.4 million genes [42].
Metagenomic Species (MGS) profiles
The IGC2 and the oral catalogues were organized into 1990 and 853 Metagenomic Species (MGS, cluster of co-abundant genes), respectively, using MSPminer [42–44]. After removing duplicated MGS (i.e., MGS present in both catalogues), we were left with 2741 MGS. The relative abundance of an MGS was computed as the mean abundance of its 100 ‘marker’ genes (that is, the genes that correlate the most altogether). If less than 10% of ‘marker’ genes were seen in a sample, the abundance of the MGS was set to 0. MGS abundance profiles were finally normalized to estimate the proportion of each species in the microbiota (sum of all species abundance = 1).
Bacterial microbiome richness of each sample was evaluated as the number of unique species (MGS) identified. Bacterial microbiome structure is evaluated according to species abundance.
Determination of the Enterobacterales counts
During the CEREMI trial, faecal samples from all volunteers (Supplementary Figure S1) were stored at 4°C after emission and transmitted to the bacteriology laboratory after blinding. One hundred mg of faeces were suspended in 1 mL of brain–heart infusion broth containing 30% glycerol and stored at -80°C. Enterobacterales were counted by plating serial dilutions of broth on Drigalski agar (bioMérieux, Marcy-l’Etoile, France).
Determination of the resistome and β-lactamasome
The IGC2 and the oral catalogues were annotated for the Antibiotic Resistant Determinants (ARD) using a two steps approach. First, potential ARD homologs were selected among catalogue genes using BLASTP against Mustard antibiotic resistance determinant database (http://www.mgps.eu/Mustard) [15]. Genes with ≥ 50% identity for ≥ 90% alignment coverage were selected and tested using pairwise comparative modelling (PCM), a 3-dimensional modelling-based approach [15]. This allowed the identification of a non-redundant list of 19061 ARD from 21 families of which 5 beta-lactamase families: 627 blaA genes, 463 blaB1, 463 blaB3, 181 blaC and 89 blaD.
The richness of the resistome was evaluated as the number of copies of genes mapping to one of the identified ARD. The relative abundance of the β-lactamasome was computed as the proportion of copies of genes mapping to any beta-lactamase family among all copies of genes mapping to one of the identified ARD.
Determination of the β-lactamase activity
β-lactamase activity of the faecal content was analysed in all samples (Supplementary Figure S1). For extraction of faeces, samples (stored at -65°C) were thawed on ice for 30 min, where after 140–380 mg of faeces material was transferred to a 2-ml Eppendorf tube by means of a spatula. Ice-cold HZn buffer (50 mM (2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) buffer, pH 7.5, supplemented with 50 µM ZnSO4) was then added to the faeces material at a concentration of 5 ml/g faeces. Samples were briefly mixed by means of vortexing and incubated horizontally for 1 hour under mild agitation. Sample were clarified by two centrifugation steps of 15 minutes and 30 minutes (4°C), respectively, in which the supernatant was transferred to a new 2-ml Eppendorf or finally 1.5-ml screw-cap tube.
Assays for determination of β-lactamase activity were performed in HZn buffer using 3–20 µL of freshly clarified faeces sample kept at 4°C. Reactions were carried out in a final volume of 200 µL with 100 µM nitrocefin (Cayman Chemical Company, Ann Arbor, MI, USA). In the first assay, 10 µL of sample was tested for the hydrolysis activity of nitrocefin. This assay was, subsequently, repeated with an adjusted sample volume if necessary. Assays were performed in 96-well microplates (SpectraPlate-96, PerkinElmer, Waltham, MA, USA) using an automated liquid handling Janus Integrator system (PerkinElmer, Waltham, MA, USA) and nitrocefin hydrolysis was monitored spectrophotometrically at a wavelength of 482 nm (EnVision microplate reader, PerkinElmer, Waltham, MA, USA). All assays always included a buffer control to assess substrate stability.
Metagenomic analysis of the phage microbiome
The phage microbiome was analyzed in samples collected at days 15, -1, 4, 10, and 30 (Supplementary Figure S1). Phage isolation was performed using a polyethylene glycol (PEG) concentration step, as previously recommended [45]. One gram of faecal samples was weighed and homogenized in 40 mL of phosphate buffered saline (PBS) (Sigma-Aldrich, Saint-Louis, MO, USA). The sample was then agitated with a mechanic laboratory agitator for 1 hour at 4°C, centrifuged at 17,000 g for 5 min and the supernatant was filtered at 2 µm and 0.45 µm. Phages were then concentrated using PEG. One molar solid NaCl and 10% (v/v) PEG 8000 (Sigma-Aldrich, Saint-Louis, MO, USA) were dissolved into the filtered culture fluid and incubated overnight at 4°C as recommended for a constant and stable precipitation. The solutions with the phages were pelleted by centrifugation at 5,250 g for 1 hour at 4°C and re-suspended in 500 µL of SM buffer (NaCl 100mM, MgSO4.7H2O 8mM, Tris-Cl 50mM). Samples were treated with 10 U/ml of DNAse (Sigma-Aldrich, Saint-Louis, MO, USA) for 30 min at 37°C followed by 10 min at 65°C to stop the reaction. DNA was then extracted using the commercial kit "Phage DNA extraction" (Norgen biotek Corp, Thorold, ON, Canada). DNA was purified on a sephadex column (Sigma-Aldrich, Saint-Louis, MO, USA), measured with Qubit dsDNA HS Assay kit (ThermoFisher Scientific, Waltham, MA, USA), and sequenced with the Illumina HiSeq2500 PE_250 bases method using the Kit Nextera XT with an input of 1 ng DNA. The sequence reads of the six samples of the same volunteers were pooled. They were trimmed to remove the Illumina adapters and remove low-quality reads using Atropos (v1.1.18) [46] with parameters: atropos trim -m 100 –q 20,20 –trim-n. The resulting reads were assembled using SPAdes 3.15.2 [47] with the metaviralSPAdes mode. At this step we obtained 22 pools of contigs (1 pool per volunteer). Gene prediction was made using Prodigal (v2.6.3) [48] with –p meta option. We excluded genes lacking start and stop codons. In order to focus our analysis on contigs sufficiently large to study genetic contexts, we excluded contigs with less than 3 open reading frames (ORFs).
In order to create a non-redundant catalogue of contigs, the 22 pools of contigs were concatenated and clustered with cd-hit-est (v4.8.1) [49]. The sequence identity threshold was 0.95, the alignment must cover 90% of the shorter sequence and a sequence was clustered into the most similar cluster that meets the threshold. We used viralVerify 1.1 [47] to classify the non-redundant contigs as viral or non-viral, and only viral contig were selected for further analysis. Then, we mapped each sample read on the “viral non-redundant contigs catalog” using bowtie2 (v2.4.2 local –very-fastlocal options) [39] and exploited SAM files with samtools (v1.3.1 with the following commands: views, sort, index, idxstat) [50]. As a result, we obtained a matrix (matrix count) representing the number of reads of a sample (columns) mapping each contig reference catalog (rows) in the dataset. All the matrix counts were rarefied at 3 003 762 reads with the ”rarefy” function of the vegan package in R [51].
The phage microbiome richness was computed as the number of phage contigs identified in each sample.
Determination of fungal load and Candida albicans DNA concentration
The fungal loads and Candida albicans DNA concentration were analysed in all available samples. For each faecal sample, 250 mg were processed using the repeated bead beating plus column protocol described elsewhere [52] (Supplementary Figure S1). A FastPrep-24™ device (MP Biomedicals, Santa Ana, CA, USA) was used instead of a Mini-BeadbeaterTM. Faecal DNA levels were quantified with the Qubit dsDNA Broad Range Kit (Invitrogen, Waltham, MA, USA) and only samples with a concentration above 50 ng/µL were considered in the analysis.
A TaqMan qPCR protocol, using a double dye MGB 5' 6-FAM-labelled probe (Eurogentec, Seraing, Belgium), with the following conditions: 2 min at 50°C, 10 min at 95°C, 15 sec at 95°C and 1 min at 65°C, the last two steps repeated for 45 cycles, was used to measure fungal DNA levels [53]. Samples were processed in two sets of duplicates, in two independent rounds. The fungal load was estimated for each sample as the ratio of the fungal DNA levels to the faecal DNA levels [54].
A TaqMan qPCR protocol in the following conditions: 2 min at 50°C, 10 min at 95°C, 15 sec at 95°C and 1 min at 62°C, the last two steps repeated for 45 cycles, was used to quantify C. albicans DNA levels. 7.5 µL of the extracted faecal DNA, at a 1:10 dilution, were used as a template, using probes and primers described by Guiver et al., 2001, at 0.1µM and 0.2 µM, respectively [55]. Samples were processed in two sets of duplicates, in two independent rounds.
The presence of qPCR inhibitors in the samples were verified in all samples, diluted at 1:10, using the Universal Exogenous qPCR Positive Control for TaqMan® Assay (Eurogentec, Seraing, Belgium), with a Cy®5-QXL®670 Probe system (Eurogentec, Seraing, Belgium). Manufacturer’s recommendations were followed, using the target Ct > 30 option.
Targeted-metagenomic analysis of the fungal microbiome
All samples were processed to study the fungal microbiota (Supplementary Figure S1). The Internal Transcriber Spacers (ITS) 1 region was targeted for the preparation of the amplicon libraries. The amplicons were produced by PCR using the ITS1F and ITS2 primers in the following conditions [56, 57]: 95°C for 3 min, 25 cycles of 95°C for 30 secs, 55°C for 30 sec and 72°C for 30 sec, 72°C for 5 min and cooling at 4°C, and their size were verified with a Bioanalyzer DNA 1000 chip (Agilent Technologies, Santa Clara, CA, USA). The purification of the amplicon was performed with AMPure XP (Beckman Coulter, Brea, CA, USA) as described in the 16S Metagenomic Sequencing Library Preparation guide [58]. The adapters were attached with the Nextera XT Index Kit (Illumina, San Diego, CA, USA) and index PCRs were done in the following conditions: 95°C for 3 min, 8 cycles of 95°C for 30 secs, 55°C for 30 secs and 72°C for 30 secs, 72°C for 5 min and cooling at 4°C. AMPure XP (Beckman Coulter, Brea, CA, USA) was used to purify the PCR products and a Bioanalyzer DNA 1000 chip allowed their verification and their quantification. Samples concentrations were normalized at 4 nM and 5 µL of each diluted sample were pooled into a library and a PhiX sequencing control was prepared by the manufacturer’s guidelines. Libraries were sequenced on Illumina MiSeq platform (Illumina, San Diego, CA, USA) with the MiSeq Reagent Kit V3 in 300bp paired-end.
The sequencing allowed the recovery of 8’819’635 amplicons from ITS1 region. The SHAMAN pipeline was used to remove the singletons and chimera amplicons, resulting in a total of 56’634 amplicons [59]. The remaining amplicons were clustered in 4648 OTUs using a cut-off value of 97% identity. 551 OTUs could be associated to fungal sequences using the Unite database and on these OTUs, 340 were present in at least two faecal samples and were kept for further analysis. A first round of annotation was performed on SHAMAN against the UNITE database (rev. 8.0) and then a second round was performed against a more recent release of UNITE (rev. 8.2). The OTUs that could not be annotated after these two rounds submitted to a classic BLASTN and only hits matched with a similarity above 97% to reference genomes were conserved. The abundances and weighted non-null normalized counts tables were generated with SHAMAN [59].
The richness of the fungal microbiome was computed as the number of unique fungal OTUs identified in each sample.
Non-targeted analysis of the metabolome
The metabolome was analyzed in all collected samples (Supplementary Figure S1). Experimental methods and parameters for the non-targeted approach were carried out by liquid chromatography and high-resolution mass spectrometry (LC-HRMS) as detailed in [60, 61]. Briefly, eight volumes of frozen acetonitrile (-20°C) containing internal standards (labelled IS mix of amino acids at 10 µg/mL) were added to 100 µL serum samples and vortexed. The resulting samples were then sonicated for 10 min and centrifuged for 2 min at 10 000 g at 4°C. Supernatants were incubated at 4°C for 1 hour for a slow protein precipitation process. Samples were centrifuged for 20 min at 20 000 g at 4°C. Supernatants were transferred to another series of tubes and then dried and stored at -80°C before LC-HRMS analyses. Pellets were diluted 3-fold and reconstituted with H2O/ACN (20/80).
Non-targeted approach experiments were performed using a HILIC phase chromatographic column, ZIC-pHILIC 5µm, 2.1 × 150 mm at 15◦C (Merck, Darmstadt, Germany), and on a UPLC Waters Acquity (Waters, Milford, MA, USA) coupled to Q-Exactive mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA). Processing steps were carried out using the R software [62]. Peak detection, correction, alignment and integration were processed using XCMS R package with CentWave algorithm [63, 64] and workflow4metabolomics platforms [65]. The resulted datasets were log10 normalized, filtered and cleaned based on quality control (QC) samples [66]. The features were then putatively annotated based on their mass over charge ratio (m/z) as well as retention time using a local database as described previously [67] and then validated based on MS/MS experiments. The remaining features were either characterized using public repositories [68, 69] or discarded when feature status are still unknown to remove noise and artifact signals. The relative abundance of all annotated chemical features was then summed and computed as a total signal, named ‘total useful signal’, for each sample. The richness of the metabolome was computed as the number of unique chemical species identified in each sample.
Analysis of the cholesterol conversion rate
The microbiota-dependent catabolism of the cholesterol in faeces was analyzed in all collected samples (Supplementary Figure S1). Sterols and stanols were extracted from faeces as follow. Faeces were weighted (~ 50 mg) and resuspended in 1% formic acid to a final concentration of 167 µg/µL. The mixture was homogenized using a Precellys Evolution instrument (Bertin Instruments, Montigny-le-Bretonneux, France) using the ‘soft program’. Volume equivalent to 1 mg of dried faeces was supplemented with deuterated internal standards (cholesterol d7 and coprostanol d5) and sterols and stanols extracted with 1.2 mL of methanol/chloroform (2:1 v/v) and 320 µL deionized water. Phase separation was triggered with 400 µL chloroform and 400 µL water. The mixture was centrifuged for 10 minutes at 3600 g and the lower phase was collected and dried. Sterols and stanols were derivatized for compatibility with GC-MS analysis using 60 µL of BSTFA (with 1% TMCS). The solution was heated at 80°C for 1h, dried and resuspended in 0.1% BSTFA in cyclohexane before injection in the GC-MS. Samples were injected at 250°C in split mode and sterols/stanols separated on a 50mx0.25mm, 0.25µm DB-5MS column. Sterols/stanols were ionized using electronic impact (EI) and analyzed in SIM mode using m/z 329 as quantitative ion and m/z 368 as qualitative ion for cholesterol and m/z 370 and m/z 215 as quantitative and qualitative ions for coprostanol respectively. The rate of cholesterol conversion to coprostanol was computed as the ratio of faecal coprostanol concentration and the sum of faecal coprostanol and cholesterol concentrations.
Analysis of the biliary acid transformation
The metabolism of the biliary acids in faeces was analyzed in all collected samples (Supplementary Figure S1). All chemicals and solvents were of the highest purity available. Cholic acid (CA), deoxycholic acid (DCA), chenodeoxycholic acid (CDCA), ursodeoxycholic acid (UDCA), lithocholic acid (LCA), hyocholic acid HCA, hyodeoxicholic acid (HDCA), glyco and tauro derivatives were obtained from Sigma-Aldrich (Saint Quentin Fallavier, France). 3α-sulfate derivatives were a generous gift of J. Goto (Niigita University of Pharmacy and Applied Life Science, Niigata, Japan) and 7α-cholic acid (CA-7S) was from Cayman Chemical (Ann Arbor, MI, USA). 23-NOR-5β-cholanoic acid-3α,12α diol, all muricholic acids, glyco, tauro derivatives and iso, keto bile acids were purchased from Steraloids Inc (Newport, RI, USA). Acetic acid, ammonium carbonate, ammonium acetate and methanol were of HPLC grade and purchased from Sigma-Aldrich (Saint Quentin Fallavier, France).
Bile acid molecular species concentrations were measured by HPLC coupled to tandem mass spectrometry (HPLC-MS/MS) as previously described with slight modification [70]. Two microlitres of an internal standard solution (23-nor-5β-cholanoic acid-3α, 12α-diol at 1 mg/ml) was added to 10–50 mg of faeces lyophilized samples using a Lyovapor L200 (Buchi, Villebon-sur-Yvette, France). For 15–20 mg lyophilized faeces samples, 2 ml of NaOH (0.1 M) was added and incubated for 1 h at 60°C before adding 4 ml of water. The solution was homogenized by two 10 s runs in an Ultra-Turrax disperser (IMLAB, Lille, France). The preanalysis cleanup procedure was achieved by centrifugation (12 000 g for 20 min) followed by solid-phase extraction using reversed-phase silica cartridges (HLB Oasis, Waters, Milford, MA, USA) and we used a 5500Q-trap (AB Sciex, Framingham, MA, USA) as mass spectrometer.
The hydrophobicity index reflects BA hydrophobicity, taking into account the concentration and the retention time of different BAs on a C18 column with a methanol gradient; lithocholic acid has the highest retention time, tauroursodeoxycholic acid-3S has the lowest.
The ability of the gut microbiota to metabolize the biliary acids was computed as the ratio of the secondary biliary acids (LCA and DCA) to the total concentration of the faecal content in biliary acids.
Data analysis
Data were classified between ‘high-dimensionality’ data relative to the taxonomic composition of the bacterial, phage and fungal microbiomes, and metabolome, and ‘low-dimensionality’ data (richness of the bacterial, phage and fungal microbiomes, richness of the resistome and metabolome, relative abundance of the β-lactamasome, total bacterial counts, β-lactamase activity, fungal load and Candida albicans DNA levels, cholesterol conversion rate, and biliary acid transformation).
Baseline was defined at day 0, and the baseline sample was defined as the sample obtained at day − 1. If this sample at day − 1 was not available, the sample obtained at day − 7 was considered as the baseline, or the one obtained at day − 15 if this latter was also missing.
For ‘high-dimensionality’ data, we computed for each subject the Spearman’s correlation coefficient (s) of the taxonomic composition of the studied system between baseline and each sampling day. These correlation coefficients were used to evaluate the change from baseline of the taxonomic composition of the system. Among ‘low-dimensionality’ data, all variables, except those relative to the richness of the systems, were log10 transformed before analysis, and we computed the change from baseline at each sampling time as the difference of the values at each time.
In order to study the variability between subjects and within subjects for each variable before the administration of antibiotics, we analysed the ‘low-dimensionality’ data using a linear mixed effect model (lmer function of R package lme4), treating subjects as random effects.
In order to study the perturbation of systems, we computed a raw distance from the baseline, that increases with the extent of the perturbation of each system, regardless of the direction of the perturbation. It was calculated at each sampling time as 1 - s² for ‘high-dimensionality’ data (with s being the Spearman’s correlation coefficient as described above) and as the absolute change from baseline for ‘low-dimensionality’ data.
Raw distances from baseline were normalized to address the effect of intra-individual variations of the systems before the start of antibiotic treatment. Normalization was made for each subject by dividing distances from baseline by the individual average of the distances from baseline computed before the beginning of antibiotic treatment (at days − 7 and − 15). In the case of missing samples at days − 7 and − 15, the normalization was used as the median of the average raw distances computed for all other subjects.
We studied the effect of antibiotics on the gut content using both fixed endpoints (days 4, 7, 10, 30 and 90) and areas under the curve between baseline and days 10 (AUCD0−D10) and 30 (AUCD0−D30). Metrics used were the changes from baseline for ‘low-dimensionality’ data and the normalized distances for ‘high-dimensionality’ data. AUCs were computed using the trapezoidal rule, using the actual date and time of stool emission. AUCs were standardized using the observed delay between baseline and the actual time of collection of the day 10 (for AUCD0−D10) or day 30 (for AUCD0−D30) sample. We used the non-parametric Wilcoxon test to compare these metrics at fixed sampling times, the AUCD0−D10, and the AUCD0−D30 to 0 for ‘low-dimensionality’ data, and to 1, 10 or 30 for ‘high-dimensionality’ data and their AUCD0−D10 AUCD0−D30, respectively. We also compared the effect of the two antibiotics on the microbiota using the non-parametric Wilcoxon test. All statistical tests were bilateral, with a type-I error fixed to 0.05.
Next, we defined for each subject and system the maximal perturbation as the maximal normalized distance from baseline observed between the baseline and day 10, and maximal resilience as the minimal normalized distance from baseline observed on days 15, 30 or 90. Pairwise relations between the level of maximal perturbation for each system was investigated using Spearman’s correlation coefficients and comparing them to 0. A similar analysis was performed to study the relationship between the maximal resilience of systems.
Finally, the relationship between the composition of the microbiome at baseline (studied using the ‘low-dimensionality’ data) and the maximal perturbation and resilience of studied systems was assessed using the Spearman’s correlation coefficient and its test to 0.