Spaceflight-associated changes in the gut microbial diversity and community structure
To evaluate the effect of spaceflight on the gut microbial alpha diversity (i.e., within-sample diversity), we computed the number of species observed in each sample (i.e., richness) and Shannon index (i.e., a diversity index accounting both evenness and richness) at the species level. We found that both the number of observed species and Shannon index significantly varied across RR-1 experimental groups (P = 0.0057 and P = 3.28 × 10−4, respectively, Kruskal-Wallis test) and were higher in Flight and Ground groups relative to Basal and Vivarium groups (Fig. 1b). Since the primary difference between Flight/Ground and Basal/Vivarium groups is the conditions of animal housing (i.e., habitat hardware, temperature, humidity, and CO2 levels; Fig. 1a), this observation indicates that the ISS rodent housing environment alters the richness and evenness of the murine gut microbial community. Interestingly, when comparing Flight and Ground animals, we found a slight increase in Shannon index (P = 0.022, Mann-Whitney test) but not in the number of observed species (P = 0.721, Mann-Whitney test) in Flight samples. Since the Ground animals were housed using the same ISS habitat equipment as the Flight animals under matched conditions of temperature, humidity, and CO2 levels in an ISS Environmental Simulator (ISSES), our observations suggest that factors specific to spaceflight induce an elevation in the evenness but not richness of the gut microbial community in mice.
We also observed significant differences in the gut microbial community structure among RR-1 groups (i.e., beta diversity analysis). Using principal component analysis (PCA) on isometric-log-ratio (ILR)-transformed species-level data, we found clear segregation of samples by the experimental group (P < 1 × 10−4, PERMANOVA; Fig. 1c). While the largest difference was between Flight/Ground samples and Basal/Vivarium samples, Flight samples were also significantly segregated from Ground samples (P = 7 × 10−4, PERMANOVA; Fig. 1c). Since the RR-1 groups were each associated with a distinct set of experimental conditions (Fig. 1a), we formulated the PERMANOVA test to replace animal groups with these associated factors in an additive model, in order to obtain an approximated estimation of the contributions of each condition to the overall variance of the gut microbial composition. Our analysis found that ISS housing conditions (i.e., habitat, temperature, humidity, and CO2 levels) accounted for 36.3% and spaceflight-specific factors accounted for 6.6% of the overall variance at the species level (for higher taxonomic levels, see Additional file 1). Thus, in addition to ISS housing conditions, we demonstrate that spaceflight-specific factors strongly modulate the composition of the gut microbiome.
Reproducible effects of spaceflight on the murine gut microbiome composition
Spaceflight-associated changes in the gut microbiome composition have been reported in a recent study of fecal samples collected in mice onboard a space shuttle during the STS-135 mission in 2011 [5]. It is thus of interest to compare RR-1 data to STS-135 data, in order to test the reproducibility of spaceflight-associated changes in the gut microbiome. Comparing two different microbiome datasets in a formal statistical setting remains a challenge due to dataset-specific biases associated with biological and technical factors, such as animal or population cohorts, experimental conditions, sequencing strategies, data analysis methods, and many others. Despite these challenges, if changes in microbial compositions are reproducible, the microbial differential abundance patterns in two datasets are expected to involve a similar set of microbes with comparable amplitudes and directions of changes. We developed a statistical method, named STARMAPs, to capture this similarity by projecting samples from a second microbiome dataset onto the PCA axes that separate the groups of the first dataset. This method assumes that, when the group differences in two datasets are similar, the samples of each dataset in the microbial taxon space segregate by their respective groups in a similar fashion and that the PCA axes capturing the group segregation in the first dataset can also capture the similar group segregation in the second dataset.
To evaluate the performance of STARMAPs, we simulated pairs of datasets, each with 10% of the species that were differentially abundant with a given fold change (FC). The differential abundance patterns in a given pair of datasets were set to be either similar (i.e., involving the same set of differentially abundant species) or distinct (i.e., involving totally non-overlapping sets of differentially abundant species). We applied STARMAPs to each of the simulated pairs of datasets and compared the results to this “ground truth” for an evaluation of STARMAPs performance (Additional file 2: Figure S1). At the typical cutoff of omnibus P < 0.05, the specificity of STARMAPs was very high under all of the simulated conditions, while the sensitivity of the test varied in each of the scenarios. In the first simulation (Simulation 1; Additional file 2: Figure S1, left), we considered the influence of sample size in each of the dataset. Expectedly, when the differential abundance amplitude was small [i.e., log2(FC) = 1], the sensitivity of STARMAPs decreased as the sample size decreased. However, STARMAPs performed very well regardless of the sample size when the simulated amplitude of differential abundance was moderate or high [i.e., log2(FC) ≥ 2]. Since dataset-specific biological and technical biases are expected to cause differences in the amplitudes of differential abundance between datasets, in Simulation 2 (Additional file 2: Figure S1, middle), we introduced random variations into the differential abundance amplitude in the second dataset of the dataset pair, and tested whether STARMAPs can still capture the similarity between the pair of datasets. As the introduced variance increased, the sensitivity of STARMAPs decreased, particularly when the mean differential abundance was small [i.e., log2(FC) = 1]. However, when the mean differential abundance increased, the negative impact of this variation on test sensitivity reduced, indicating STARMAPs is well suited to identify differential abundance patterns that are similar but not necessarily identical in two datasets. Another expected consequence of dataset-specific biological and technical biases is the differences in the set of microbial species uncovered in each of the dataset, which was simulated in Simulation 3 (Additional file 2: Figure S1, right). As expected, the sensitivity of STARMAPs worsened as the proportion of commonly observed taxa in the pair of datasets decreased, due to loss of information. However, the decrease in sensitivity caused by low proportions of commonly observed taxa was in part compensated by the increase in the effect size. It is of interest to note that, when considering a similar effect in two microbiome datasets, it is likely that the proportion of taxa that are differentially abundant in both datasets is higher than the proportion of taxa that are commonly present in both datasets. In our simulation, all species in the second dataset have the same chance of not being found in the first dataset, and therefore, our simulation represented a harsher condition. Nevertheless, our simulations suggest that the performance of STARMAPs was satisfactory over a range of scenarios, particularly when the differential abundance amplitudes were relatively large.
We next applied SATRMAPs to test whether spaceflight-associated changes in the gut microbiome during the RR-1 mission were similar to the STS-135 mission. Like RR-1, the mouse research onboard STS-135 involved a flight and a ground group with matched diet, habitat equipment, and environment (i.e., an environmental simulator was used), and a significant difference in the microbial community structure between the two groups was reported [5]. Using STARMAPs, we found that the differences in fecal microbial composition between the flight and the ground animals in the STS-135 mission were similar to those between RR-1 Flight and Ground animals (Fig. 2a; omnibus P = 0.032, STARMAPs). It can be noted that the directions of differences between Flight and Ground samples in the two missions were similar but not parallel with each other (cosθ = 0.33; θ is the angle between the directions of group differences in the two datasets). Aside from the technical differences in the microbiome profiling methods, this may be due to the differences in the mission duration (i.e., 13 days for STS-135 vs. 37 days for RR-1) or the sample collection strategies. STS-135 samples were collected from animals euthanized after return to Earth, while the RR-1 samples were collected from frozen carcasses of mice euthanized in orbit. Nonetheless, our findings indicate that space environmental factors produce robust and reproducible effects on the murine gut microbiome composition.
Lack of similarity between spaceflight- and radiation-induced microbiome changes
We next sought to understand the contributions of specific space-associated factors to microbiome changes during spaceflight. It has been hypothesized that cosmic radiation is a unique environmental factor that can modulate the gut microbiome in space [3]. Previous Earth-based studies have indeed found changes in the gut microbiome in animals exposed to radiation that was similar in type to cosmic radiation. One study exposed mice to high-linear energy transfer (LET) radiation (600 MeV/n 16O) at doses of 0, 0.1, 0.25, or 1.0 Gy and reported changes in the gut microbiome composition and functional potential 10 and 30 days after the exposure [7]. Another study fed rats on either a high-iron diet or a diet with adequate iron for 14 days and then exposed the animals to low-LET radiation (137Cs fractionated radiation at 0.375 Gy/day) every other day for 16 days with a total dose of 3 Gy while continuing the assigned diets [5]. This study reported an altered relative abundance pattern of microbial orders that were associated with the diet and radiation exposure [5]. In order to test whether exposure to radiation significantly contributed to the microbiome changes during spaceflight, we used STARMAPs to compare the microbiome differences between RR-1 Flight and Ground groups to the space-type radiation-induced microbiome changes in these two Earth-based rodent studies (Fig. 2b, c). In both datasets, changes in the gut microbial community structure in response to radiation exposures were observed in our re-analysis at the species level (Additional file 3: Figure S2), confirming an effect of space-type radiation on the gut microbiome. However, the radiation-induced changes were not found to share significant similarity to those during RR-1 spaceflight (Fig. 2b, c). Although the exact nature of radiation exposure during RR-1 is unknown, radiation dosimetry data [8] recorded inside the space shuttle cabins during previous STS missions suggest that the total radiation dose and the dose rate (dose per day) during each mission were at least two and three magnitudes lower, respectively, than those used in the two Earth-based studies, which considered space environment beyond ISS and space shuttle orbits. It can be expected that the radiation exposure during RR-1 was likely to be similar to the STS missions, since ISS and space shuttles operate in similar obits. Therefore, our observation, together with the expected dose of RR-1 radiation exposure, suggests that space radiation alone during RR-1 is unlikely to be the predominant contributor to the observed microbiome changes and implies significant contributions from other space environmental factors.
Spaceflight-associated changes in taxon abundance and inferred functional gene content
The altered microbial community structure among RR-1 groups was associated with altered relative abundance patterns that can be clearly seen at the family level (Fig. 3a). To identify specific microbial taxa affected by spaceflight, we used the ALDEx2 analysis package, which operates on centered-log-ratio (CLR) transformed sequencing data for a compositionally coherent inference of differential abundance [9]. Within the RR-1 dataset, at false discovery rate (FDR) < 0.05, 5 phyla, 6 classes, 10 orders, 15 families, 20 genera, and 18 species of bacteria were differentially abundant among the four experimental groups (Fig. 3b). Consistent with the PCA results, the predominant differences were observed between Flight/Ground samples and Basal/Vivarium samples, highlighting the strong impact of ISS rodent housing conditions on the gut microbiome composition. A number of taxa (1 order, 1 family, 5 genera, and 6 species), however, were significantly (FDR < 0.05, ALDEx2) differentially abundant between Flight and Ground groups, while an additional set of taxa (1 phylum, 1 class, 2 families, 6 genera, and 6 species) were suggestively (P < 0.05 but FDR > 0.05, ALDEx2) differentially abundant between the two groups (Fig. 3b). For example, the abundance of the bacteria in the phylum of Bacteroidetes, while was lower in the Ground/Flight animals compared to Basal/Vivarium animals, was also suggestively decreased (P < 0.05 but FDR > 0.05, ALDEx2) in the Flight animals compared to Ground animals. This change, together with a trend of elevated abundance of the Firmicutes phylum, led to a significantly increased Firmicutes-to-Bacteroidetes (F/B) ratio (Fig. 3c; P = 0.014, Mann-Whitney test, Flight vs. Ground), consistent with our previous findings in a twin astronaut during his 1-year spaceflight mission [4]. Firmicutes and Bacteroidetes are the two most common and abundant bacteria phyla found in the mammalian gastrointestinal tract. A change in the F/B ratio can be a sensitive marker, or serve as a proxy, of overall microbiome changes associated with many conditions. Examples include changes in the F/B ratio in patients with obesity [10], during aging in humans [11], and in response to dietary fiber particle size [12]. In addition, the relative abundance of Tyzzerella (a genus in the Lachnospiraceae family, Clostridiales order) was significantly decreased (FDR < 0.05, ALDEx2) in Flight animals compared to Ground animals, while the abundance of a few other genera of the Lachnospiraceae family were significantly (FDR < 0.05, ALDEx2) or suggestively (P < 0.05 but FDR > 0.05, ALDEx2) increased in Flight animals (Fig. 3b), revealing opposite effects of spaceflight on relatively close-related taxa. Similar patterns were observed in the Ruminococcaceae family, in which the Ruminococcaceae UCG-010 genus showed a significantly increased (FDR < 0.05, ALDEx2) while the Hydrogenoanaerobacterium genus showed a suggestively decreased (P < 0.05 but FDR > 0.05, ALDEx2) abundance in the Flight animals compared to Ground animals. Finally, the relative abundance of the Staphylococcus genus of the Bacillales order was similar among Flight, Vivarium, and Basal samples, while the Ground samples appeared to be distinctively high (Fig. 3b), suggesting ISS rodent housing conditions and space-specific factors may induce opposite changes in the abundance of these microbes.
We next investigated the functional implication of these spaceflight-induced changes in gut microbial compositions. We used the software package PICRUSt2 to infer microbial gene content from 16S rRNA gene data and aggregated relative abundance of functional genes into metabolic pathways [13]. We then used ALDEx2 to identify differentially abundant pathways among RR-1 experimental groups. To capture the dominant functional features of spaceflight and ISS housing environment effects, we used a permissive threshold of FDR < 0.1. At this threshold, we found 619 pathways differentially abundant among groups (Additional file 1), 174 of which were differentially abundant between Flight and Ground animals (Fig. 3d). Hierarchical clustering of these 174 pathways based on the CLR-transformed relative abundance revealed three clusters, each with a unique differential abundance pattern and highlighting a specific mode of energy metabolism (Fig. 3d, e). Cluster I consists of a set of pathways that involve compounds used or produced by pyruvate fermentation, including carbohydrate degradation, aromatic compound degradation, carboxylate degradation, amino acid biosynthesis, lipid biosynthesis, and synthesis of polysaccharides. The relative abundance of genes in Cluster I pathways was low in Ground animals and higher in Flight animals. However, except for several pathways, Flight samples were not significantly different from the combined Basal and Vivarium samples (Fig. 3d and Additional file 1). This differential abundance pattern contrasted that of Cluster II, which contains a number of pathways related to utilizing amines as sources of nutrients and energy. The relative abundance of Cluster II pathway genes was high in Ground animals and was lower in Flight animals. In a few pathways (e.g., 4-aminobutanoate degradation I and III, urea degradation II, and putrescine degradation I; Fig. 3d and Additional file 1), gene abundance in Flight animals was also lower than Basal/Vivarium animals. Finally, Cluster III pathways are involved in electron transfer and biosynthesis of cofactors needed for aerobic and anaerobic respiration. Flight animals showed the lowest relative abundance of genes in this cluster, and Ground animals appeared to be intermediate between Flight and Basal/Vivarium animals. Taken together, our analysis of inferred microbial gene content revealed an increased abundance of fermentation genes and a decreased abundance of genes for respiration and amine degradation in Flight animals compared to the housing condition-matched Ground mice. This finding is consistent with a shift in energy metabolic capability in the gut microbiome during spaceflight.
It is worth noting that the choice of reference genome catalog influences the accuracy of microbiome gene content predictions. A recently developed integrated mouse gut metagenome catalog (iMGMC) has been shown to enhance the accuracy of PICRUSt predictions in mice [14], providing a useful resource for inferring the functional capability of the murine gut microbiome. We thus performed PICRUSt2 functional prediction with the iMGMC reference and compared the results with those obtained with the default reference, in order to ensure that the inference described above was robust. Using the iMGMC reference, PICRUSt2 analysis uncovered 592 of the 868 pathways that were uncovered with the default reference and 3 additional pathways (Additional file 4: Figure S3A; Additional file 1). This discrepancy is likely due to the fact that iMGMC reference, at its current stage, contains a small set of 16S rRNA-linked functional genomes (i.e., 484 genomes) that are specific to the murine gut microbiome, as opposed to the PICRUSt2 default reference, which contains a set of > 20,000 genomes of various origins. Despite this major difference, the predicted abundance of the commonly uncovered pathways and their differential abundance patterns between Flight and Ground animals derived using these two references were largely similar (Additional file 4: Figure S3B–D; Additional file 1). Given these observations, we continued our analysis with the functional predictions made using the PICRUSt2 default reference for a more inclusive analysis, in order to sufficiently capture the functional capability of the gut microbiome under the unique environment of spaceflight.
Associations between the expression of host genes in the liver and inferred gene abundance of microbial metabolic pathways in the gut during spaceflight
To further understand the functional implications of spaceflight-associated changes in the gut microbiome, we utilized RNA-seq data in the liver of RR-1 mice stored in NASA’s GeneLab data repository [15, 16] and tested the correlations between the liver transcriptome of the host animal and the inferred relative gene abundance of microbial metabolic pathways in the gut, with the hypothesis that microbial metabolic potential and host metabolism are altered in coordination during spaceflight. We focused on the subset of microbial pathways that have been identified with differential inferred gene abundance between Flight and Ground animals (i.e., the 174 pathways in Fig. 3d), and performed the correlation analysis with multiple testing adjustment on a per-pathway basis in order to capture the dominant patterns of transcriptomic variations relevant to each microbial pathway of interest. The number of host genes significantly correlated (FDR < 0.1) with each microbial pathway were highly variable, ranging from a few thousand to only a few or even none (Fig. 4a and Additional file 1). For each microbial pathway with significantly correlated host genes, we identified biological processes and pathways that were enriched with those genes. This analysis revealed a number of host functions that covaried with gut microbial metabolism under the spaceflight and control conditions (Fig. 4b). Microbial 1,2-dichloroethane degradation (a Cluster I pathway in Fig. 3d) was positively correlated with genes encoding rhodopsin-like G-protein-coupled receptors (GPCRs) and was negatively correlated with genes encoding glycoproteins. In addition, microbial pathways of putrescine degradation, 4-aminobutanoate degradation, and glutathione-glutaredoxin redox reactions (Cluster II pathways in Fig. 3d) were positively correlated with host genes that were enriched in a number of pathways, most notably ribosome, proteasome, mitochondria, redox processes, lipid metabolism, and cell-cell adhesion. Lastly, microbial conversion of acetate to acetyl-CoA (a Cluster III pathway in Fig. 3d) was positively correlated with the expression of host genes involved in lipid metabolism, for which acetyl-CoA is a key intermediate.
We note that these correlations could be due to independent responses of the liver transcriptome and the gut microbiome to the ISS housing and spaceflight conditions, and are not necessarily indicative of interactions between liver functions and the gut microbial metabolic potential. Indeed, the majority of the correlations between microbial pathways and hepatic gene expression were no longer significant (FDR > 0.1) when partial correlations controlling for experimental groups were computed (Fig. 4a), suggesting these relationships reflected only a concurrence under the spaceflight and control conditions. Nonetheless, several potential microbial-host interactions were observed. The microbial pathway converting acetate to acetyl-CoA was associated with 121 genes, 48% of which remained significantly correlated (FDR < 0.1) when partial correlations were computed. In addition, about 26% of the genes correlated with the microbial putrescine degradation pathway remained significantly correlated (FDR < 0.1) after controlling for experimental groups. Enriched biological functions of these partially correlated genes confirmed a positive association between host protein metabolic genes (e.g., ribosome and proteasome; Fig. 4c and Additional file 1) in the liver and the capability of putrescine degradation by microbes in the gut. Putrescine is one of the most common polyamines that can be synthesized or uptaken by mammalian cells [17]. While polyamines are essential for many physiological functions, inhibited protein synthesis by excessive exogenous polyamines has been observed in a murine mammary carcinoma cell line [18]. Therefore, our observations raise an intriguing possibility that the decreased abundance of gut microbial putrescine degradation genes during spaceflight leads to a putrescine surplus, and in turn, to inhibition of host protein synthesis and metabolism.