Salivary microbiota composition during the study
To inspect how the salivary microbiota reacts in a confined environment, we characterized samples collected during the entire duration of the Mars500 mission (720 days in total) by 16S rRNA gene amplicon sequencing of the variable region V3-V4 (Table 1).
Subject
|
Earth to Mars (First Variant)
|
Mars to Earth (Third Variant)
|
Follow-up (Normal Diet)
|
Total
|
5001
|
7
|
5
|
3
|
15
|
5002
|
7
|
5
|
3
|
15
|
5003
|
7
|
5
|
3
|
15
|
5004
|
7
|
5
|
3
|
15
|
5005
|
7
|
5
|
2
|
14
|
5006
|
7
|
5
|
2
|
14
|
Table 1: Number of salivary samples collected during the study. The number of samples collected during each step of the study was reported for each crewmember. Marginal totals were added for subjects and simulated journeys together with the grand total that was reported in the bottom right corner of the table.
Table S1 summarizes all phases of the mission whereas Figure 1a reports the sampling scheme used in this work. Amplified sequences formed 1890 amplicon sequence variants (ASVs) with a median number of 172.00 ASVs per sample (ranging from 81 to 317). A total of 4,337,540 sequences specifically aligned to an ASV resulting in a sequencing depth ranging from 20,084 to 116,809 and a median value of 47,044 (for additional information about sequence analysis pipeline and the number of sequence obtained in each pre-processing step see Supplementary material, Supplementary Table S2, and Supplementary Figure S1 and S2). All replicates reported an accuracy higher than 0.96 with a Spearman’s rank correlation () that ranged between 0.94 and 0.98 (Supplementary Table S3, S4 and Supplementary Figure S3). Rarefaction curves reached a plateau above 15k reads suggesting an adequate sequencing depth for all samples (Supplementary Figure S4). Good’s coverage estimator ranged between 99.99% and 100.00% across all samples indicating that roughly 0.01% of the reads in a sample came from ASVs that appear only once in that sample (Supplementary Table S5).
Roughly, 99% of sequences aligned to variants that came from known bacterial taxa (Table S2). Supplementary Table S6 shows the overall taxonomic composition of samples whereas Figure S5 reports the phylogenetic tree reconstructed from ASVs. At phylum level Firmicutes, Bacteroidetes, Actinobacteria, Proteobacteria, and Fusobacteria accounted for more than 97% of the total number of reads assigned to taxonomically annotated ASVs (Figure 1b and Supplementary Table S6). The total bacterial diversity (namely the alpha diversity) remained constant during the mission with no significant differences detected between the isolation period and the follow-up, across different diets, and across subjects (Table S7 and Figure 1 panels c, d, and e). Also, time did not impact bacterial diversity as showed in Figure 1f (random mixed model fitted using crewmembers as random intercept: slope lower than 0.002; Supplementary Table S8).
Effect of food and time on salivary microbiota
We inspected differences across samples (namely beta diversity) using non-metric multidimensional scaling (nMDS) on quantitative and qualitative indexes. Samples showed a similar distribution with all index tested (Figure 2a): Sorensen index and unweighted unifrac distance (qualitative analysis), and Bray-Curtis and weighted unifrac distance (quantitative analysis). As opposed to alpha diversity, subjects, diets, and time significantly contributed to shape the salivary microbiota with different percentage of variance explained depending on the index but never exceeding 10% of the total variance (Figure 2b). For all diversity indexes (except for the Sorensen index which reported a significant effect of subjects) the dispersion of tested factors was homogeneous, meaning that only the composition of samples varied among groups as highlighted by the permutational analysis of variance reported above. Diet impacted on bacterial genera usually present in the salivary microbiota of healthy subjects—such as Actinomyces, Veilonella, and Fusobacterium [18]—but also on Peptostreptococcus, Haemophilus, Megasphaera, and Prevotella, which have been correlated to different disorders of the oral cavity (such as periodontitis, dental caries, and oral lichen planus) [19–21]. Bacterial species classified as Alloprevotella, Fusobacterium, Dialister, Veilonella, and Megasphaera followed the same pattern: decreasing abundance passing from the the first to the third diet and then back to starting values during the follow up (even if the shift was not significant). Other species like Haemophilus and Prevotella reported a significantly higher abundance during a single diet—namely Haemophilus was more abundant during the follow up and the abundance of Prevotella was higher during the first diet. At phylum level, diets impacted more on Bacteroidetes and Firmicutes. Within them, 4 out of 28 (14%), and 3 out of 7 (43%) genera reported (at least) a significant difference during diet changes (Figure S6).
To explore the effect of time on bacterial diversity we used change-point analysis on both within-subject (Figure S7 panel a) and between-subjects diversity (Figure S7 panel b). Within-subject diversity measures changes in the salivary microbiota of each crewmember through time, whereas between-subjects diversity compares the salivary microbiota of different crewmembers at each time point (Figure 3a and b). Three segments significantly divided within-subject diversity with two change-points at 123 days and 480 days. Between-subjects diversity was not segmented since the overall model gave better results than the segmented one according to the genetic algorithm used during optimization (Figure 3b). The overall between-subjects model had an effect size of -0.00004 which means that after 520 days of isolation the overall diversity decreased by 0.02179. The effect of time on within-subject diversity was indeed higher than the one observed for between-subjects diversity. During the first 123 days the effect modeled was 0.00103 reflecting an average increase of 0.12636 for all crewmembers. After the first change-point, within-subject diversity started to decrease with a regression parameter of -0.00064 (average decrease during the second segment of -0.22835). After the second change point, which roughly matched the end of the isolation period (Figure 3c and d), the within-samples diversity started to increase again. At the end of the follow-up period diversity increased again of 0.29474 exceeding the average value detected in the first day of isolation (Table 2).
Days
|
|
SE
|
|
|
|
Within-subject
|
|
|
|
|
|
1 - 123
|
0.00103
|
0.00027
|
3.86
|
12.00
|
.0023
|
124 - 480
|
-0.00064
|
0.00012
|
-5.19
|
36.00
|
< .0001
|
481 - 720
|
0.00123
|
0.00020
|
6.16
|
14.23
|
< .0001
|
Between-subjects
|
|
|
|
|
|
1 - 720
|
-0.00004
|
0.00001
|
-3.89
|
424.19
|
.0001
|
Table 2: Temporal changes of salivary microbiota. Within-sample diversity was divided into three segments following change-point analysis whereas between-samples diversity was modeled on the full time period since no change-points were detected. Results of mixed effect models fitted for each segment were reported in the table. , regression parameter (slope of the model); SE, standard error; , t-value (also known as “standardized” regression parameter); , degrees of freedom; , p-value.
Resilience of salivary microbiota
The average abundance of ASVs correlates with their persistence, the number of subjects in which a given ASV was detected at each time point. Figure 4a shows the increasing trend of log-transformed abundance with an R-squared value of 0.72 (, 95% CI , ). Time-resolved clustering produced two groups of ASVs: one, called inconsistent micriobiome (Cluster 1), included variants detected in a small number of subjects at each time point, whereas the other (Cluster 2), called stable microbiota, included variants detected in the vast majority of subjects during the whole mission (Figure 4a and Figure S8, panel a and b). The inconsistent microbiota showed low average persistence in respect with the stable microbiota but it contained the largest amount of variants (1746 ASVs against 144 of stable microbiota). Unlike stable ASVs, subjects lost and acquired inconsistent ASVs both during and after the isolation period (Figure S8 panel b and c). Stable ASVs were detected in roughly 30% of all subjects at each time point (26 samples on 88) with sporadic losses and acquisitions (Figure 4a and Figure S9).
We represented the acquisition and loss of bacterial species during the whole mission using networks. At each time point we linked subjects to ASVs detected in their salivary microbiota forming a bipartite network structure which reflected the underlying bacterial community structure. The loss and acquisition of bacterial ASVs was shown in supplementary video S1 where green squares represent subjects, red circles represent inconsistent microbiota, and light blue circles represent stable microbiota As shown in the video, the topology of the networks did not change in time, but at each time point subjects acquire/release bacterial species from/into the environment, except for stable ASVs which are shared by most crewmembers and thus (almost) always present in central part of the network. The number of new edges formed and destroyed passing from one time point to another slightly decreased in time (mixed effect model 95% CI for formed edges [-0.08, -0.02] and destroyed edges [-0.08, -0.01] Table S9). The end of the isolation period significantly increased the average number of formed edges (namely acquired ASVs) of 28 but the trend was still negative (Figure 4b). The number of formed edges was independent from the number of lost edges (Spearman’s = -0.04; p-value = 0.750). The salivary microbiota structure did not change at the beginning, during, and after the isolation period reporting a similar network topology (Figure 4c and supplementary video S1). Amplicon sequence variants clustered into cluster 1 had a marginal position in all networks, linking only one (or a few) subjects—in many cases no link was reported for ASVs of cluster 1 since they are intermittently present in most subjects as reported in supplementary video S1. In contrast, ASVs clustered into cluster 2 had a more central position in all community networks reporting connections with at least four subjects in the majority of the cases. The topology of the networks was confirmed by the centrality analysis performed on ASVs of both cluster 1 and 2. The centrality of ASVs assigned to cluster 2 was higher than those assigned to cluster 1 at every time point, highlighting the central role of these variants in respect with the whole community structure (Figure 4d, Wilcoxon rank sum test: all p-values < 0.01). Also, the abundance reported a similar effect with ASVs of cluster 2 showing higher values during the whole experiment (Figure 4e, Wilcoxon rank sum test: all p-values < 0.01).
Drivers of diversity
To inspect drivers of beta diversity along and after the isolation period, we fitted a linear model for each ASV detected in the salivary microbiota of crewmembers. The Mars500 mission time-scale was divided into three stages according to the changepoints detected for within-subject diversity. Crewmembers showed a different number of ASVs reporting a trend of diversity similar to the one reported in Figure 3c. The number of ASVs showing a significant effect of time and changepoints ranged from 3 to 28 depending on the subject (Figure 5a and Table S10). Stable ASVs—namely those grouped into Cluster 2 according to their prevalence in time—that showed a significant trend of diversity enriched the saliva of four out of six crewmembers, if compared with the overall occurrence (Figure 5b). The fraction of stable ASVs in subjects 5004 and 5005 was roughly six times higher than the average fraction of stable ASVs, whereas subjects 5002 and 5006 reported a fraction three times higher than the average.