A total of 825 fecal samples were finally included for 16S rRNA analysis. The study design and process were shown in Figure 1.
3.1 Gut microbiome profiles changed dramatically in untreated MN
We compared the composition and alteration of gut microbiome in 115 UMNs and 115 age-and-gender matched HCs. The baseline characteristics were shown in Table S1. Results showed that both α-diversity and β-diversity of microbiome in UMNs had a significant decrease compared with HCs (Figure 2A, B, Figure S1A-B, Table S2). The composition also changed obviously in UMNs (Figure S1 C). The components of PC1 and PC2 account for 79.56% proportion of total variance in two dimensions by using PCoA, suggesting enormous difference of two groups (Figure 2B and Figure S1D, E). Observed OTUs in UMNs was markedly decreased as shown by Venn diagram, rarefaction and rank abundance distribution curves (Figure S1 F-H).
At phylum level, there was a certain increase of the relative abundance of Proteobacteria and Actinobacteria, while the average abundance of Bacteroidetes decreased a lot in UMNs (Figure 2C, Figure S2 A-C and Table S3, S4). In terms of genus level, Bacteroides, Prevotella and Lachnospiraceae_unclassified decreased significantly in UMNs. However, the relative abundance of Escherichia−Shigella, Subdoligranulum, Bifidobacterium and Enterobacteriaceae_unclassified increased the most evidently in UMNs versus matched HCs (Figure 2D, Figure S2 D and Table S5, S6). The relative abundance and distribution of the key OTUs in each sample was shown in the way of heatmap (Figure 2E, Table S7). The spearman correlation analysis showed the correlation between laboratory examination and key OTUs in Figure 2F, G. The alteration of metabolic pathway of gut microbiome was also found changed in UMNs (Figure S2 E, F), including increase of osmoprotectant transport system. These results indicated that the gut microbiome composition and function had specific alteration in UMN patients.
3.2 Establishment and validation of the diagnostic model.
To explore whether the gut flora could be used to identify MN patients, 115 UMN patients and 115 matched healthy controls were randomly divided into training group (UMN=71, HC=71) and test group (UMN=44, HC=44). By means of five-fold cross-validation on random forest model, we obtained 7 crucial OTUs including OTU27 (Lachnospira), OTU113 (Lachnospira),OTU232 (Lachnospiraceae_unclassified),OTU559 (Agathobacter),OTU619 (Roseburia),OTU667 (Faecalibacterium) and OTU722 (Lachnospiraceae_unclassified) to establish the diagnostic model (Figure 3A, B). The POD and the ROC showed the model had a high efficiency with an AUC of 98.36% (cut-off value: 0.4988, sensitivity: 0.9306, specificity: 0.9306) in training group (Figure 3D). The test group also showed high AUC of 92.02% (cut-off value: 0.468, sensitivity: 0.9535, specificity: 0.814) (Figure 3E, F). In cross regional validation phase, composed by 44 untreated MN from South China and 44 age and gender matched HCs, the model showed good discrimination efficiency as well, with an AUC of 95.31% (cut-off value: 0.523, sensitivity: 0.9091, specificity:0.8958, Figure 3G H).
Previous studies reported medication could influence gut microbiome. So, we tested whether the model could be used in TMNs. Results showed POD values were markedly higher in TMNs than HCs, with an AUC of 96.44% (cut-off value: 0.6315, sensitivity: 0.9373, specificity: 0.9122, Figure 3I, J). Regional cohort of 65 MN patients (treated or untreated) from East China and matched 57 HCs showed AUC was as high as 90.24% (cut-off value: 0.4465, sensitivity: 0.9231, specificity: 0.7544, Figure 3K L). These results suggested medication did not influence the model efficiency. It could be widely used in MN patients, whether applied medication or not.
To further test its efficiency in different disease states, we divided TMN into high protein group (HPRO, n=84) and low protein group (LPRO, n=171). With equal number of age-and-gender matched HCs respectively, these two cohorts were recruited into the diagnostic model. Results showed AUCs were 96.02% (cut-off value: 0.6195, sensitivity: 0.9643, specificity: 0.881) in HPRO group (Figure 3M, N) and 96.5% (cut-off value: 0.615, sensitivity: 0.9357, specificity: 0.9181) in LPRO group (Figure 3O, P), indicating the model was also suitable across MN states.
3.3 Medication of MN did not influence gut microbiome
To further explore the influence of medication on gut microbiome, we compared them among 78 UMNs, 108 TMNs and 100 HCs based on the matched age and gender. Urine protein, serum creatinine, albumin, triglyceride and total cholesterol were also matched in these UMNs and TMNs. Clinical data of three groups was shown in Table S8. Observed OTUs were decreased in either UMNs or TMNs compared with HCs (Figure 4A). PCoA showed the gut microbiome composition was similar between UMNs and TMNs, while quite different from HCs (Figure 4B, Figure S3A, B). Similar results were obtained by Plsda and ANOSIM analyses (Figure 4C, D, Table S9).
The analysis of relative abundance also suggested similar changes of gut microbiome in UMNs and TMNs compared with HCs. Microbial composition at phylum level showed Proteobacteria increased mostly while Bacteroidota decreased significantly either in UMNs or TMNs compared with HCs (Figure 4E, F, G, Figure S3B, Table S10). Similar pattern was obtained at genus level (Figure S3C and Table S11), suggesting synchronized microbial changes in UMNs and TMNs. The Venn diagram and heatmap indicated UMNs and TMNs were more alike while quite different from HCs (Figure S3D, Figure4H and Table S12). KEGG pathways also showed great differences of UMNs or TMNs when compared with HCs (Figure S3E, F). The specific of functional pathway changed a lot, including increase of human diseases in both UMNs and TMNs. And it might be related to the pathogenesis of MN. These findings were quite different from our previous expectation, since the microbiome seems not to be affected much by medication.
3.4 The changes of gut microbiome remained stable across different disease states of MN.
Composition of gut microbiome between HPROs and LPROs were compared (clinical data was shown in Table S13). Interestingly, observed OTUs and α-diversity were similar in HPROs and LPROs and it was not affected by increasing of sample size as shown in rarefaction curves (Figure 5A). ANOSIM analysis suggested there was no statistical difference between HPROs and LPROs (Figure 5B and Table S14). PCoA analysis also suggested the gut microbiome of HPROs resembled with that in LPROs (Figure 5C, Figure S4A, B, C). Venn diagram showed 797 out of 897 OTUs were shared between both groups (Figure S4 D). The similarity of OTU composition was further shown in way of heatmap (Figure 5E and Table S15). It was hard to found the differences in the relative abundances of microbial composition in neither phylum nor genus level (Figure S4 E, F and Table S16-S19).
We further divided these individuals into training group (HPRO=56, LPRO=114) and test group (HPRO=28 LPRO=57). With five-fold cross-validation on random forest model, 8 key OTUs were obtained for the new model establishment including OTU48 ([Ruminococcus]_torques_group), OTU90 (Ruminococcaceae_Incertae_Sedis), OTU144 (Butyricicoccus), OTU178 (Saccharimonadaceae), OTU251 (Parabacteroides), OTU292 (Lachnospiraceae_UCG-010), OTU301 (Peptococcus) and OTU843 (Oscillospiraceae_uncultured) (Table S20). Nevertheless, the model did not suggest good discriminative ability (Figure S4G-J). These results implied that the changes of gut microbiome might not be associated with the disease states of MN.
3.5 Successful construction of MN rat model depends largely on gut microbiome
Patient samples indicated relationship between gut microbiome and MN onset. To further explore the role of gut microbiome in the pathogenesis of MN, we established MN rat model. The grouping and process of model construction was shown in figure 6A. The dynamic changes of average urine protein in each group were shown by means of point-fold line chart in Figure 6B. Scatter plots showed that at baseline, each group had no statistical differences in urinary protein (Figure 6C). What interested us was that the use of antibiotics significantly prohibited fully presentation of MN phenotype (Figure 6D, Group Mod+Con, Mod+MN verse Group Model). What’s more, even in control rats, the use of antibiotics also significantly lowered the urinary protein levels (Figure 6D, Group Nor+Con, Nor+MN verse Group Normal). This effect quickly disappeared when antibiotics discontinued (FMT for 1 week, Figure 6B E, Group d-f). Compared with Model group, FMT groups (group e, f), both from healthy people and MN patients, showed delayed and reduced presentation of phenotype (Figure 6B). Contrast to expectations, FMT using MN feces hardly had any effect in aggravating urinary protein excretion until sacrifice. On the contrary, healthy feces could partially replicate the MN model (Figure 6B, E-G). This phenomenon indicates that naturally existed microbiome is essential for the pathogenesis of MN. Lack of natural microbiome prevents MN development, while restoring it, but not MN microbiome, could at least partially reconstruct MN phenotype.
In terms of histological changes, HE, PAS and Masson staining showed significant glomerular damages in Model rats at the end of the study. Consistent with urinary protein, FMT using MN feces showed more mild glomerular damages than that using healthy feces (Figure 6H). Electronic microscope showed more details at the basement membrane which further verified what we got using light microscope (Figure 6I). Histological results confirm that the naturally existed gut microbiome plays as a prerequisite condition in MN pathogenesis, while changed gut microbiome in MN patients is more likely to play a protective role.
3.6 Analysis of rat fecal microbiome confirmed effective elimination and transplantation of microbe
The fecal samples of rats at different time points were collected and underwent 16S rRNA sequencing. Since the excretion of urinary protein after model establishment was largest at 1 week, we compared gut microbiome among these three points, which were defined as Nor (fecal collection at 0w in Normal group), ME (fecal collection at 0w when Model was Established in Model group) and AM (fecal sample collection at 1w After Model was established in Model group). Gut microbiome of MN model altered from control rats at the point of 0w. These changes got more significant at 1w, even when the urinary protein excretion began to relieve at this point (Figure 7A, Figure S5A and Table S21-S24), indicating a delay in microbiome adaptation. The composition of gut microbiome differed from each other analyzed by Plsda among three groups of fecal samples. And heatmap showed the key OTU composition also changed obviously when compared with Nor in both ME or AM group (Figure 7B, Figure S5B, Table S25, S26). Observed OTUs of microbiome were markedly decreased after using antibiotics (AB, fecal collection at 0w in group e), as shown by cloudplot, PCoA and ANOSIM analyses (Figure 7C, D E, Figure S5C-E and Table S27), indicating success in intestine cleaning. Composition of FMT groups (group Mod+Con and group Mod+MN) differed from each other (Figure 7F, G). Each of them differed greatly from control or Model group as shown by Plsda (Figure 7H and Table S28), suggesting successful transplantation of different feces. Differences in composition of gut microbiome between these four groups were also shown in heatmap and PCoA diagrams (Figure S5E, F Table S29). Analysis of rat feces confirms gut microbiome was successfully eliminated and transplanted in different groups. Together with physiological and pathological results, it demonstrates natural gut microbiome is the crucial factor responsible for MN development.