Stability of the individual gut microbiota profile in confined environment

The intestinal microbial communities of the six crewmembers of the MARS500 project (subject 5001 to 5006) were tracked over time during the 520 days of ground-based space simulation. Longitudinal faecal sampling also included collection of stools before entering the isolation facility, and a number of samples after the exit from the modules, up to 6 months later. A mean of 27 faecal samples per crewmember were collected. Each sample was characterized via next-generation sequencing of V3–V4 hypervariable region of the 16S rRNA gene, allowing for a total of 5,377,450 high-quality sequence reads (mean per subject, 33,820; range, 7759–91,366). Reads were clustered into 54,836 chimera- and singleton-filtered OTUs at 97% sequence similarity.

The gut microbiota dynamics were reconstructed across time and subjects, based on the variation of highly abundant OTUs, according to the normalization strategy described by David et al. [5]. As shown in Fig. 1, these trajectories revealed a strong individuality in the ecosystem structure and its dynamics during the prolonged isolated confinement in the MARS500 infrastructure, with a unique succession of individual microbial profiles. Even at phylum level, there were apparent microbial signatures that characterized each individual in his free-living conditions (i.e., before entering the isolation module) and accompanied him during and after the simulated interplanetary flight. The observed individual microbiota profiles at phylum level were consistent with the range of the phylum-level microbiota variation in the human population (Additional file 1: Figure S1). Specifically, Proteobacteria members were distinctive of subject 5002 (mean relative abundance across time series, 12.3%), Bacteroidetes were almost completely missing in subject 5004 (mean relative abundance across time series, 0.5%), and Verrucomicrobia were undetectable in the microbiota of subjects 5003 and 5006.

It should be pointed out that no detailed, individual-specific information on the micro- and macronutrient composition of the different variants of diet designed for the MARS500 astronauts throughout the whole project is available to the authors, making impossible any correlation with the microbial changes observed. However, based on available data, it is reasonable to assume that the diets were similar among astronauts across the mission and, thus, unlikely to make a major contribution to the inter-individual differences in the gut microbial dynamics.

Notwithstanding the strong individuality, the intestinal microbial communities of the six crewmembers shared 14 OTUs, all belonging to the Firmicutes phylum, and mainly to the Lachnospiraceae family (9 OTUs) (Additional file 2: Table S1; Additional file 3: Figure S2). All these OTUs were already present in the individual microbiota configurations prior to entry into the MARS500 module.

When exploring the variation of the beta (i.e., inter-astronaut) diversity of the gut microbiota over time, it is worthy to note that, unlike the weighted UniFrac distances that showed apparently random fluctuations, without a significant trend over time, the unweighted UniFrac values followed a downward trajectory, with a significant inverse association with the time spent in the MARS500 isolation facility (quantile median regression test: RC range, regression coefficients scaled to the full variation of UniFrac distances, −9362.98; RC sd, regression coefficients scaled to one standard deviation, 1900.01; P value generated by boot-strap analysis, 4E−5) (Additional file 4: Figure S3). In particular, immediately upon entry into the module, the median unweighted UniFrac distance was about 4% lower than the initial value (i.e., in free-living conditions) and reached a reduction of up to 9% after about 7 months spent in spacecraft.

Temporal dynamics of gut microbiota components

The highly abundant OTUs used to reconstruct the individual microbiota trajectories over the MARS500 study were further analyzed by evaluating the OTU propensity to variation over time, as the magnitude of change above or below the median relative abundance (Fig. 1 and Additional file 2: Table S1).

Interestingly, some behavioral patterns of OTUs were shared among subjects. Specifically, OTUs attributable to Roseburia faecis (OTU_ID 84029, family Lachnospiraceae), Faecalibacterium prausnitzii (OTU_ID 543524, family Ruminococcaeae), and when present, Akkermansia muciniphila (OTU_ID 35867, family Verrucomicrobiaceae) were identified as highly variable during the whole period of confinement, regardless of the initial microbial configuration. In particular, the F. prausnitzii OTU was especially affected in the second half of the MARS500 mission, showing for all subjects the lowest relative abundance values around about 1 year of confinement. On the other hand, where detected, the A. muciniphila OTU decreased dramatically in the gut microbiota of all crewmembers when returning to their own free-living conditions after the end of the mission. The OTUs associated with Bacteroides (including B. caccae, B. eggerthii, B. fragilis, B. ovatus, B. uniformis, and unclassified species) were instead among the most perturbed in the initial phase of the mission, typically reaching for each subject the highest relative abundance values within the first 30 days of confinement. Conversely, the OTUs classified as Dorea (when available, D. formicigenerans, OTU_ID 181619) and Coprococcus catus (OTU_ID 616283) were highly stable in the intestinal microbiota of all crewmembers.

Identification of microbiota steady states and ecological succession in time

Temporal dynamics of the gut microbiota were further investigated for each crewmember, by identifying microbial steady states and characterizing their ecological succession over time (Fig. 2 and Additional file 5: Figure S4). Steady states were interpreted in the context of recent theories of microbiome ecology, predicting that ecosystems may exist under multiple states (to be intended as “stable” configurations/equilibria of the microbial community) [5]. For each subject, steady states were identified through the same method described by David et al. [5], i.e., using a cluster-defining height threshold in a Jensen-Shannon Distance (JSD)-based tree. A total of 5 different steady states were identified for individuals 5002 and 5004; 6 for 5001, 5003, and 5005; and 7 for 5006. As already described for the individual microbial profiles, the phylogenetic structures of steady states were largely dominated by Firmicutes but retained a strong individual fingerprint in terms of taxonomic composition of the microbial communities, with variable percentages of Bacteroidetes, Proteobacteria, Actinobacteria, and Verrucomicrobia members (Additional file 5: Figure S4). According to our data, the temporal succession of steady states followed a highly personalized non-linear trajectory, with different temporal persistence and times of recurrence for each of them. Except for 5006, whose microbiota was deeply unstable, with steady states continuously alternating with one another, the intestinal microbial communities of the other crewmembers remained stable for periods lasting on average 2 months, with steady states generally recurring one up to three times over the entire sampling period. Interestingly, only for subjects 5004 and 5006, we observed a return to the initial microbiota configuration, after exiting the module, while new steady states were consolidated for the other crewmembers.

Steady state-based temporal dynamics of the crewmembers differed from each other also for the microbial ecology of the establishment of a steady state compared to the previous one. Specifically, we used a SourceTracker/z-score combining approach to measure the relative contribution of microbial sources other than indigenous ones in defining the steady state structures. According to our data, within each time series, the transition from one steady state to the next generally retained a strong individual signature, indicative of a rearrangement of the autochthonous microbial asset (contribution of the individual microbial communities up to 96%). Nevertheless, in each steady state-based trajectory, we could identify a major change event (three for subject 5006) in the microbial configuration, associated with a higher relative contribution of allochthonous microbial sources than indigenous ones (median contribution of the individual microbial communities, 10%). These major events were variously distributed along the individual timelines but occurred on average after 340 days (range, 330–360 days) of isolated confinement for subjects 5002, 5004, and 5006.

According to a Random Forests analysis [12], the vast majority of steady state discriminatory OTUs belonged to the Lachnospiraceae and Ruminococcaceae families (Fig. 2). The main exceptions were represented by Bifidobacterium OTUs that discriminated two steady states in the first half of the MARS500 simulation, for subjects 5003 and 5006, and one steady state of subject 5004 in the second half of the study and after the return to real life. On the other hand, Proteobacteria OTUs, specifically belonging to Sutterella and Enterobacteriaceae, discriminated steady states between 2 and 7 months of confinement for subjects 5001 and 5002, respectively.

To further explore the patterns of microbiota variation across the individual steady states over the MARS500 study period, for each crewmember, we established co-abundance associations of the most abundant OTUs and then clustered correlated OTUs into three or four co-abundance groups (CAO; Fig. 2 and Additional file 5: Figure S4) [13]. The differences in the succession of individual CAO profiles, with intra- and inter-CAO rearrangements, confirmed the personalized alternating of multiple reversible configurations of the gut microbiota in a single time series, mainly characterized by different proportions of Ruminococcaceae and Lachnospiraceae SCFA producers.

MARS500 mission

The MARS500 programme was financed by the European Programme for Life and Physical Sciences in Space (ELIPS) and involves scientists from across Europe. It was performed by the State Scientific Center of the Russian Federation – Institute of Biomedical Problems (IBMP) of the Russian Academy of Sciences, and consisted of three isolation studies: a 14-day pilot study (completed in November 2007), a 105-day pilot study (completed in July 2009), and the main 520-day study, simulating a complete space flight to Mars, which is the focus of the present manuscript. A multinational crew, composed of six adult male volunteers (three selected by the Russian Federation, two by the European Space Agency, and one by the China National Space Administration; mean age, 31.8 years; range, 27–38 years), entered an isolation facility in the IBMP, Moscow, on June 3_, 2010, where they remained in continuous temporal and spatial confinement till November 4, 2011. During the stay in the spacecraft-like habitat, consisting of four hermetically sealed interconnected modules and one external module to simulate the Martian surface, they performed realistic activities of a roundtrip mission to Mars following a weekly work schedule, including, among others, operative work and meetings, exercise, scientific experiments covering the areas of physiology, psychology, biochemistry, immunology, biology, and microbiology, and even simulated emergency events. The parameters of the module habitat, determining physiologically comfortable proportion of the main components of gas media and their total pressure, complied with the standard “Cosmonauts’ habitat in a manned space ship”. Regularly (not less than once a month), the microbial contamination of the gas media was assessed, and selected surfaces from the habitable, medical, and utility module were swabbed over time, in collaboration with the crewmembers. The overall microbial load in the air and on different surfaces during the entire mission was found to be moderate compared to non-confined occupied rooms, sampled at the German Aerospace Center, private households and suburbs in Cologne (scientific experiment MICHAm, Microbial Ecology of Confined Habitats and Human Health, conducted in the context of MARS500) [15], and in any case, the CFU counts did not exceed the maximum allowed aboard the ISS [25]. Resources of water and food, whose composition reflected the diet used in the International Space Station (ISS) [26], were limited as in a real space flight. Specifically, two water supply systems were used: the system of drinking water supply (drinking water and for food cooking) and the system of domestic water for hygienic demands. The water quality in the first system was checked every 2 weeks. In the other system, water was from the centralized system of water supply of Moscow. No episodes of increased microbial contamination were reported. Regarding diet, the composition of food rations fulfilled the recommendations of World Health Organization (WHO) and also Russian-American norms on the food composition of rations for the crew of ISS. Specifically, three food rations were designed: (1) first variant, from the 1st till the 250th day of the mission (time of flight from Earth to Mars); (2) second variant, from the 251st till the 270th day for three crewmembers (simulation of the egress to the Martian surface); (3) third variant, from the 271st till the 520th day for the three crewmembers participating in the egress to the planet surface (returning to Earth), and from the 251st till the 520th day for the other three crewmembers. The food rations included different types of products, ready or semi-ready for consumption, by Russian, European, Korean, and Chinese firms, with up to four menu variants, providing on average 15.1% protein, 33.4% fat, and 51.2% carbohydrate. Detailed information about the entire project is available at http://www.esa.int/Our_Activities/Human_Spaceflight/Mars500 [11]. All scientific investigations carried out in the context of the MARS500 project were approved by the Ethics Committee of IBMP, and all crewmembers gave their written informed consent.

Sampling

Crewmembers were asked to collect faecal samples 10 days before entering the isolation facility, at various time intervals throughout the entire 520-day simulation experiment, and after exiting the module up to 6 months later. Samples were collected in sterile vials, after cleaning the table top of the toilet with a hydrogen peroxide solution, washing hands with disinfectant gel, and wearing disposable gloves. In particular, for each time point inside the MARS500 modules, the vials were stored at 4 °C and, when samples from all six astronauts had been collected, they were moved outside the ground-based experimental facility (NEK facility) to the IBMP lab, where they were transferred to a −80 °C freezer into a Ziploc bag. Every 6 months, the samples collected were shipped to Italy in dry ice. For an overview of sampling times for each crewmember, please see Fig. 1.

Microbial DNA extraction and Illumina MiSeq sequencing

Total bacterial DNA was extracted from faeces using the repeated bead-beating plus column method [27] with only minor modifications [28]. Briefly, cell lysis was achieved by introducing three 1-min steps in a FastPrep instrument (MP Biomedicals, Irvine, CA) at 5.5 movements per second, in the presence of 500 mM NaCl, 50 mM Tris-HCl pH 8, 50 mM EDTA, 4% (w/v) SDS, four 3-mm glass beads, and 0.5 g of 0.1-mm zirconia beads (BioSpec Products, Bartlesville, OK). After incubation at 95 °C for 15 min and centrifugation at full speed for 5 min to pellet stool particles, nucleic acids were precipitated by adding 10 M ammonium acetate and one volume of isopropanol. Seventy percent ethanol-washed pellets were resuspended in TE buffer, treated with 10 mg/ml DNase-free RNase at 37 °C for 15 min, and then subjected to protein removal and column-based DNA purification following the manufacturer’s instructions (QIAamp DNA Stool Mini Kit; QIAGEN, Hilden, Germany). The V3–V4 hypervariable region of the 16S rRNA gene was amplified using the 341F and 805R primers with added Illumina adapter overhang sequences as previously reported [29]. Amplicons were purified with a magnetic bead-based cleanup system (Agencourt AMPure XP; Beckman Coulter, Brea, CA). Indexed libraries were prepared by limited-cycle PCR using Nextera technology and further cleaned up as described above. The final library, prepared by pooling samples at equimolar concentrations, was denatured with 0.2 N NaOH and diluted to 6 pM with a 20% PhiX control. Sequencing was performed on Illumina MiSeq platform using a 2 × 300 bp paired end protocol, according to the manufacturer’s instructions. Sequencing reads were deposited as entire raw data in the National Center for Biotechnology Information Sequence Read Archive (NCBI SRA; BioProject ID PRJNA358005) and separately for each sample, along with available metadata, in the MG-RAST database (http://metagenomics.anl.gov/mgmain.html?mgpage=project&project=mgp79314).

Computational and statistical analyses

Raw sequences were processed using a pipeline combining PANDAseq [30] and QIIME [31]. High-quality reads were clustered into OTUs at 97% sequence similarity using UCLUST [32]. Taxonomy was assigned using the RDP classifier against the Greengenes database (May 2013 release). The filtering of chimeric OTUs was performed by using ChimeraSlayer [33]. All singleton OTUs were discarded.

For the analysis of the gut microbiota dynamics over the entire Mars mission simulation, the normalization technique developed by David et al. [5] was used. Briefly, for each crewmember: (i) time points were normalized in the standard manner so that the sum of all fractional OTU abundances at a given time point was 1; (ii) highly abundant OTUs, accounting for 90% of median time point reads, were selected; (iii) each time point was normalized to a reference community that was computed for each sample based on other time points with a similar community structure. Specifically, reference OTU values were computed using a weighted median across time series, with time point weights set to be (1 − j)² and j being the pairwise Jensen-Shannon Distance (JSD) score to the sample being normalized.

For each time series, steady states (i.e., “stable” configurations/equilibria of the gut microbial community) were identified clustering samples in a hierarchical-Ward linkage tree based on JSD and choosing a common tree height of 4 as the lowest across time series, that still allowed identifying sample groups (i.e., steady states) that (i) showed significant correlations between samples within the group (multiple testing using the q-value method) and (ii) were statistically significantly different from each other (permutational MANOVA using the JSD matrix as input, function Adonis of the vegan package in R). We assumed that such clusters approximated steady states of each individual microbiota, as microbial configurations with precise values of taxon relative abundance and defined functional roles. All steady states displayed significantly different inter-relationships from each other (P < 0.001). The taxonomic profiles for each steady state were generated taking the median value of relative abundance for each OTU in the samples being part of that steady state.

Within each subject, the dynamics of formation of the microbiota steady states was assessed by estimating, for each steady state (except for the first one), the percentage of microbiota not deriving from the previous steady state using SourceTracker [34], and then normalizing values by a z-score approach. In brief, for each crewmember, the relative abundance profiles of each steady state were alternately treated as sinks with the microbiota configurations of the immediately preceding steady state in the timeline of that subject as well as of the other individuals as sources. We then considered exogenous (i.e., other subjects and unknown) sources, obtaining a list of scores ranging from 0 to 1, and applied a z-score approach to these values, for each subject separately. In this way, when the z-score was 0, the change in the microbiota structure of that steady state corresponded to that observed on average; when the z-score was positive, the change was greater than observed on average; when the z-score was negative, the change was less than observed on average.

Co-abundance groups of OTUs (CAO) were determined as described in Claesson et al. [13], using OTUs with a mean relative abundance among steady state profiles more than 0.1%. Wiggum plots were created using Cytoscape 3.2.1. Steady state discriminatory OTUs were identified using the Random Forest machine learning algorithm [12]. Briefly, Random Forests is a powerful classifier that identifies the best subset of features (here, relative OTU abundance) at discriminating between categories (steady states within each subject).