Clinical signs and pulmonary histopathological lesions are different in mice that succumbed to or survived the influenza infection
To assess the heterogeneity in the host response to influenza, we infected specific pathogen-free (SPF) C57BL/6 mice with virulent (GX) and attenuated (HB) H7N9 virus using the LD50 approach (see the “Materials and methods” section for the two virus details). Consequently, C57BL/6 mice were infected with LD50 dose of the virulent H7N9 strain (the GX group), the same load of attenuated virus (the HB group), or administered phosphate-buffered saline (PBS) (the negative control NC group). Animal body weights and survival were monitored, and fecal samples were collected daily after the infection. All mice in the GX group exhibited obvious clinical signs of infection, including weight loss, anorexia, shivering, and a rough hair coat. The survival rate in this group was 53.3% (16/30) (Fig. 1a). By contrast, all mice in the NC and HB groups survived and exhibited no obvious symptoms of infection. Further, based on whether they succumbed to or survived from the infection, the mice in the GX group were divided into the death group (n = 14, GX.DG) and survival group (n = 16, GX.SG), respectively. Of the GX.SG group, 10 mice were randomly selected from the 16 mice as the representative of this group. The percentile curves for the mouse body weights revealed that the mice in the GX.DG group lost more body weight than those in the GX.SG group (Fig. 1b). To further evaluate the differences between the GX.DG and GX.SG groups, a repeated experiment was performed. At day 10 post-infection, the mice in the GX group that died formed the GX.DG group, whereas those mice that survived and started to show signs of improvement comprised the GX.SG group. Indeed, both autopsy and histological examination revealed that the lung lesions in the GX.DG mice were notably more severe than those in the GX.SG mice (Fig. 1c and Additional file 1: Fig S1). Taken together, these findings demonstrated that mice that succumbed to or survived infection with the same dose of influenza virus have clearly different clinical signs and obvious histopathological differences, representing a heterogeneous response to infection and confirming that the LD50 approach can be used to differentiate between individual-specific responses to infection.
The transfer of fecal microbiota from mice that survived virulent influenza infection increases the resistance of the recipient mice to influenza
To investigate whether the differences in the responses of mice to influenza infection are associated with the gut microbiota, we next performed fecal microbiota transplantation (FMT) (Fig. 1d). The effect of fecal microbiota transferred from mice that succumbed to or survived the virulent infection on infection with the GX virus in recipient mice was then assessed. FMT of the fecal microbiota collected from GX.SG mice on either day 3 or 9 post-infection significantly increased the survival of the recipient mice to influenza infection compared with the resistance of mice transplanted with fecal microbiota from uninfected mice (NC donor) or that of PBS-treated mice (Fig. 1e, f). By contrast, FMT of the fecal microbiota collected on day 9 post-infection from GX.DG mice reduced the survival rate of the recipient mice after influenza infection (Fig. 1f). These findings indicated that the fecal microbiota of mice that survived the infection likely contains specific intestinal microbes that provide protection against influenza.
Gut microbiota exhibits obvious differential characteristics depending on the severity of influenza infection
To identify the anti-influenza intestinal microbes of mice that survived the infection, we next used 16S rRNA gene sequencing to analyze the fecal microbiota of mice daily, during the 16 days that the experiment lasted. To more intuitively analyze the data, samples from the 16 time points were grouped into four infection stages: stage 1 (day 0), control; stage 2 (days 1–4), the stable weight stage; stage 3 (days 5–8), the weight loss stage; and stage 4 (days 9–15), the death stage. The Shannon diversity index analysis indicated that the abundance and diversity of the bacterial community followed a decreasing trend at stage 4 in the GX.DG mice but not in the GX.SG, HB, or NC mice (Additional file 2: Fig S2). Further, the principal coordinate analyses (PCoAs) of weighted UniFrac distances indicated a notable shift in the microbiome community structure in the GX.DG samples with disease progression, particularly at stage 4, whereas only a slight shift was observed in the GX.SG and HB groups. By contrast, almost no shift was observed in the NC group (Fig. 2a). An intergroup analysis showed similar results: the sample clustering at the last infection stage revealed differences between the GX.DG group and the other three groups (Additional file 3: Fig S3).
Furthermore, a detailed phylum-level taxonomic analysis of the intestinal microbiota revealed obvious changes in the bacterial taxonomic composition in the gut microbiota of mice infected with the GX strain, particularly in the GX.DG mice (Fig. 2b). Several phyla exhibited similar variation trends in the GX.DG and GX.SG groups; for example, in both groups of mice, Elusimicrobia and Proteobacteria became more abundant than in the NC group, while Bacteroidetes became less abundant than in the NC group as infection proceeded, although the changes in the Elusimicrobia abundance were more notable (Fig. 2c and Additional file 4: Fig S4). Aside from these similarities, several differences were apparent between the GX.DG and GX.SG groups: for example, during the infection, Verrucomicrobia abundance significantly increased only in the GX.DG group, while Actinobacteria abundance clearly increased only in the GX.SG group, compared with the NC group (Fig. 2c). In addition, the variation trends of Firmicutes abundance visibly fluctuated, with higher abundance in GX.DG than in GX.SG at the third stage post-infection, but the other way around at the fourth stage (Additional file 4: Fig S4). These results indicated that infection with the virulent H7N9 virus strain significantly changed the composition of the gut microbiota, with clear differences between infected vs. non-infected as well as dead vs. survived mice groups.
To further detail the effect of influenza infection on the gut microbiota, we performed taxonomic analyses of the intestinal microbiota at the genus level. Based on the variation trends of gut microbes in the four mouse groups, the differentially abundant microbes were divided into six groups (Fig. 2d and Additional file 5: Fig S5): (1) bacteria with increased abundance only in the GX.DG group (Akkermansia, Bacteroides, Parabacteroides, unidentified Gastranaerophilales, Butyricimonas, and unidentified Ruminococcaceae), (2) bacteria whose abundance first increased and then decreased only in the GX.DG group (Enterorhabdus, Adlercreutzia, and Candidatus Saccharimonas), (3) bacteria with increased abundance in both the GX.DG and GX.SG groups (Elusimicrobium, [Eubacterium]_coprostanoligenes_group, and Ruminococcaceae_UCG-005), (4) bacteria whose abundance decreased only in the GX.DG group (Lachnospiraceae_NK4A136_group, Lactobacillus, and Parasutterella), (5) bacteria whose abundance increased in the GX.DG and GX.SG groups and particularly in the HB group (Turicibacter and Allobaculum), and (6) bacteria whose abundance increased in the GX.SG group and decreased in the GX.DG group (Bifidobacterium). The analysis indicated different genera of the gut microbiota exhibit differential responses to influenza infection, with some common features shared by individual genus depending on the outcome of the infection. Based on the correlations between the abundances of gut microbes and the health status of the infected mice, we speculated that Lachnospiraceae_NK4A136_group, Parasutterella, Lactobacillus, and Bifidobacterium might play important roles in the host’s defense against influenza infection.
Bifidobacterium pseudolongum and Bifidobacterium animalis are strongly associated with the survivability of influenza-infected mice
Since the above taxonomic analysis indicated that specific bacterial genera could impact the host’s response to infection, we moved on to analyze the bacteria at the species level. Accordingly, 177 representative samples from all groups collected on days 0, 2, 5, 8, 9, 10, 11, and 15 post-infection were selected for metagenomic sequencing analyses. For gene richness characterization, a rarefaction analysis was performed. The estimated gene richness values almost approached saturation in all groups, indicating that the sequencing data had sufficient coverage and that only very few genes may be undetected (Additional file 6: Fig S6). The results of PCoA analysis showed a notable shift in the gut microbiota composition in the GX.DG samples at the later infection stage (Fig. 3a), consistent with those of the 16S rRNA gene sequencing analysis. This indicated that the starting dataset could be used for meaningful analyses of changes in bacterial abundance at the species level.
We then analyzed the co-abundance gene groups (CAGs) of all genes to uncover any correlations between species abundance and the host’s health status. Overall, 1157 CAGs were identified, and 98 of these groups were annotated (Additional file 7: Table S1). The relative abundance of the annotated CAGs in the four mouse groups is shown in Fig. 3b. Indeed, CAGs enriched in the GX.DG group differed from those in the NC group: the former exhibited greater abundance of Escherichia coli, Bacteroides, Parabacteroides, Streptococcus agalactiae, and Helicobacter and lower abundance of Lactobacillus (CAGs of these bacteria were listed in Additional file 7: Table S1). Although the CAG components in the GX.SG and HB groups were very similar to those in the NC group, several specific components were identified. For example, the GX.SG group exhibited greater abundances of B. pseudolongum and B. animalis, and the HB group was enriched in Lachnospiraceae. For the comparative analysis of the GX.SG and GX.DG groups, or the GX.SG and NC groups, the GX.SG-associated CAGs were classified as enriched in the GX.SG group (odds ratio (OR) score > 2). To further identify the dominant specific species in the GX.SG group and their correlations with disease severity, Spearman’s correlation network of the CAGs enriched in the GX.SG group was generated (data for days 9–15 are shown in Fig. 3c, d). Compared with the GX.DG group, the GX.SG group exhibited a greater abundance of 18 CAGs (Fig. 3c), most of which were previously found to confer health benefits [16, 17]. In addition, compared with the NC group, the GX.SG group was enriched in 12 CAGs (Fig. 3d), of which B. animalis (2 CAGs) and Clostridium sp. (2 CAGs) were shared with the above correlation network. This indicated these two species might play an important role (e.g., protective role) during influenza infection. Next, a random forest classifier was used to classify the GX.SG and GX.DG groups using the 18 CAGs enriched in the GX.SG group compared with the GX.DG group. Tenfold cross-validation was repeated five times, and the area under the curve (AUC) of receiver operating characteristic (ROC) was used as the scoring method to evaluate the accuracy of the classifier on a testing dataset (Fig. 3e). Based on the test set, an average AUC of 85.39% was obtained with a 95% confidence interval (CI) of 76.77 to 94%, indicating that this model has a powerful diagnostic potential for predicting infected mice’s prognosis/severity. Moreover, the contribution of each CAG was evaluated based on the mean decrease in accuracy (Fig. 3f). Among the 18 CAGs analyzed, the CAGs annotated as B. pseudolongum (CAG-982 and CAG-979), Lactobacillus sp. (CAG947, CAG944, CAG950, and CAG952), and B. animalis (CAG980 and CAG981) contributed most to the identification of the GX.SG and GX.DG groups. These findings further confirmed the close correlation between the survivability of infected mice and the presence of specific bacteria, especially B. pseudolongum, Lactobacillus, and B. animalis.
Further, to get the detailed differences of the gut microbiota at each time point, the effect size analysis (LEfSe) of linear discriminant analysis (LDA) was performed (Fig. 3g, h and Additional file 8: Fig S7). At day 0 post-infection, the abundances of gut microbes have no apparent differences between the groups GX.SG and GX.DG or NC. However, the abundances of B. pseudolongum and B. animalis were obviously increased in the GX.SG mice at 2 days post-infection compared with their abundances in the GX.DG or NC mice (Fig. 3g, h). The elevated distribution of B. pseudolongum in GX.SG mice persisted throughout the infection process, with B. animalis also found at elevated levels at days 10 and 11 post-infection (Additional file 8: Fig S7). In addition, to validate the metagenomic analysis, qRT-PCR was performed to detect B. pseudolongum and B. animalis of fecal samples collected from another repeated experiment. As shown in Additional file 9: Fig S8, the populations of B. pseudolongum and B. animalis were actually enriched in GX.SG vs. in GX.DG at day 2 and day 10 post-infection, but not at day 0 post-infection. These suggested that the variation in the abundances of B. pseudolongum and B. animalis between the GX.DG and GX.SG groups was associated with the differential responses of these mice to influenza infection and not due to the differences in the initial abundances of these gut microbes.
Based on the objective cause-precedes-effect law in a causal relationship, and on the above-described FMT experiments and distribution characteristics of gut microbes in the GX.SG group, we hypothesized that the gut microbiota increases the resistance of the host mouse to virulent H7N9 virus infection by increasing the abundance of endogenous B. pseudolongum and/or B. animalis.
Gavage of survival-associated microbes protects the host against influenza
To verify the role of B. pseudolongum and B. animalis on host resistance to influenza infection, SPF mice were treated with an antibiotic solution (ATB) [18] and then administered the two bacterial strains (in single or in combination) by oral gavage before infection. Oral administration of B. animalis alone or two-bacterium combination significantly increased the survival rate of recipient mice, but gavage with B. pseudolongum alone had no obvious effect (Fig. 4a). Further, the weight loss was most clearly improved on days 6–10 after the infection by gavage with the two-bacterium combination, and somewhat improved by gavage with B. animalis alone (Fig. 4b). To eliminate any possible artifacts (e.g., residual microflora) associated with the ATB-treated mice, germ-free mice were orally administered with B. animalis alone, two-bacterium combination, or PBS. Similarly, gavage with B. animalis alone and two-bacterium combination both significantly increased the survival of germ-free mice. The only difference is that in germ-free mice, the administration with B. animalis alone provided almost equal protection as the combination gavage, but in ATB-pretreated mice, there was only half as much protection by gavage with B. animalis compared to combination gavage (Fig. 4c). These observations demonstrated that the transfer of B. animalis alone provided obvious protection against influenza and that this protection could be enhanced by the additional administration of B. pseudolongum in non-sterile mice.
A parallel experiment involving ATB-pretreated mice as described above was performed to determine the lung virus titer and cytokine levels and to examine histological changes. On days 3 and 5 post-infection, the lung virus titers in mice administered with B. animalis alone or the two-bacterium combination were significantly lower than those in PBS-treated mice (Fig. 4d). However, the cytokine levels in the lungs of mice administered with B. animalis alone or the two-bacterium combination showed different trends between the two time points compared to those in mice administered with PBS (Fig. 4e): the levels of the majority of tested cytokines were significantly elevated in the lungs of mice administered with B. animalis alone or the two-bacterium combination on day 3 post-infection; by contrast, on day 5 post-infection, the levels of several cytokines in these two groups, especially the two-bacterium combination group, became lower than those in PBS-treated mice. Although only IL-6 levels significantly decreased in the blood of mice administered with B. animalis alone and the two-bacterium combination on day 5 post-infection, most cytokines displayed an obvious increase in these two groups on day 3 post-infection (Additional file 10: Fig S9). The data indicated that the lower lung virus titers in mice administered with B. animalis alone or the two-bacterium combination compared with those in mice administered with PBS were accompanied by a greater induction of cytokines at the early phase of H7N9 infection. This suggested that B. animalis likely protects hosts against influenza infection by modulating the host immune response.
In addition, histological examination revealed that the lungs of mice administered with PBS or B. pseudolongum alone displayed severe pathologies on day 7 post-infection, including congestion, inflammatory cell infiltration, and deciduous cells in the bronchial lumen, even the alveoli almost disappeared. By contrast, the lung lesions of mice administered with B. animalis alone or the two-bacterium combination were obviously improved (Fig. 4f). This further confirmed the anti-influenza effect of B. animalis.
B. animalis contributes specific metabolic functions to the gut microbiome to protect the host against influenza infection
To further understand the mechanism through which B. animalis mediates the anti-influenza effect, we performed a functional metagenome comparison of the gut microbiome between the GX.SG and GX.DG or NC groups. Overall, 2278 KEGG Orthology (KO) terms were enriched in the GX.SG group compared with the GX.DG or NC groups (Additional file 11: Table S2). The distribution characteristics of these KO terms at each time point are shown in Fig. 5a. The comparison of the GX.SG and GX.DG groups revealed an enrichment of KO terms in the GX.SG group mainly at the early and late stages of infection. However, in the comparison of the GX.SG and NC groups, the KO terms enriched in the GX.SG group were mainly observed at the early and middle stages of infection. Further, the number of enriched KO terms in the first comparison (1606) was clearly higher than that in the second comparison (1061). This suggested that the functional structure of the gut microbiota in the GX.SG group was more similar to that in the NC group.
Further, the enriched metabolic pathways in the GX.SG group revealed by the comparisons between the GX.SG and GX.DG groups as well as between the GX.SG and NC groups displayed similar distribution characteristics as those found in the analysis of KO terms (Additional file 12: Fig S10). More importantly, on days 2 and 5 post-infection, many KO terms were enriched in the GX.SG group compared with both the GX.DG and NC groups (double-positive enriched KO terms) (Fig. 5a), implying these KO terms very likely play important roles in defense against influenza. Our above data demonstrated that B. animalis was significantly enriched in the GX.SG mice at 2 days post-infection compared with the GX.DG or NC mice, but not at 5 days post-infection (Fig. 3g, h and Additional file 8: Fig S7), suggesting that B. animalis provides more anti-influenza effect at 2 days post-infection than at 5 days. Thus, we subsequently analyzed the functional composition of the double-positive enriched KO terms at 2 days post-infection and the contribution of B. animalis to these terms/functions. As shown in Fig. 5b, 39 functions were identified among the double-positive enriched KO terms (here, only the KO terms labeled with a module number were shown; other KO terms without module number were listed in Additional file 11: Table S2), which involved five pathways, including carbohydrate and lipid metabolism, energy metabolism, environmental information processing, genetic information processing, and nucleotide and amino acid metabolism. B. animalis or B. pseudolongum participated in most of these 39 functions and all of the five pathways. Of these pathways, valine/isoleucine biosynthesis (M00019), lysine biosynthesis and the DAP dehydrogenase pathway (M00526), and coenzyme A (CoA) biosynthesis (M00120) are uniquely enriched in B. animalis. These results indicated that B. animalis may mediate a protective effect against influenza through promoting the biosynthesis of valine, isoleucine, lysine, and CoA.
Oral administration of valine or intraperitoneal injection of CoA protects the host against influenza
To verify the above functional metagenomic analysis, ATB-pretreated mice were gavaged with valine, isoleucine, or lysine, or were intraperitoneally injected with CoA, and then intranasally challenged with the H7N9 virus. As shown in Fig. 6a, b, f, g, oral administration of valine or intraperitoneal injection of CoA significantly increased the survival rate of recipient mice and improved the weight loss. By contrast, oral administration of lysine only improved the weight loss, and oral administration of isoleucine had no effect. These results demonstrated that oral administration of valine or intraperitoneal injection of CoA can provide protection in vivo against influenza. However, in vitro, both valine and CoA showed no obvious anti-influenza effect in A549 cells (Additional file 13: Fig S11), suggesting that the anti-influenza effects of valine and CoA might be indirect. To further understand how valine and CoA mediate the anti-influenza effect in vivo, the parallel animal experiments were performed. Then, the lungs were collected to determine virus titer, cytokine levels, and histological changes (Fig. 6c–e, h–j). On day 5 post-infection, both oral administration of valine and intraperitoneal injection of CoA can significantly reduce virus titers and levels of IL-1β, IL-6, and IL-10 of the lungs and at the same time upregulate IFN-β and IFN-γ of the lungs. In addition, on day 7 post-infection, the treatments of both valine and CoA also alleviated the lung lesion compared with that of PBS. The difference is that, on day 0 post-infection, intraperitoneal injection of CoA can obviously induce the expressions of TNF-α, IL-1β, IL-6, and IFN-γ in the lungs, but oral administration of valine only stimulates IFN-γ. These data suggested that the anti-influenza effect of valine could be due to its immunoregulatory properties after influenza infection, by contrast, that of CoA could be because it stimulates the innate immune response in advance.