A loss of microbial diversity in DKD
Through the rigorous clinical inclusion and exclusion criteria, a total of 359 fecal samples were obtained for our study, including 179 healthy controls and 180 DKD patients. In agreement with Figure S1A, B, an average of 28349 reads per sample were collected in DKD group, lower than that 29348 of Con group. With sufficient sample size, number of detected OTUs showed a marked decrease in DKD compared with healthy controls (Figure S1C). For instance, there were total 2331 OTUs in DKD group, while 2997 in Con group (Figure 1A). Rank-abundance curve indicated a relatively uniform and diverse OTU profile in Con group (Figure S1D). By calculating index of alpha diversity, we observed a decrease of OTU richness (Ace for DKD vs. Con: 468.60 vs. 545.21; P=1.34e-05; Chao for DKD vs. Con: 542.97 vs. 453.49; P= 1.08e-06) and diversity (Shannon for DKD vs. Con: 3.65 vs. 3.62; P=5.19e-06; Simpson for DKD vs. Con: 0.09 vs. 0.14; P=0.0048) in DKD compared with Con (Figure 1B-E). To visualize spatial distribution of gut microbiome among all samples, (un)weighted UniFrac algorithm was used to compute beta diversity. PCoA analysis revealed significant difference between DKDs’ and healthy controls’ microbial community (PERMANOVA test for unweighted UniFrac distance: R2=0.0074, P=0.001; Figure 2F; PERMANOVA test for weighted UniFrac distance: R2=0.0036, P=0.197, Figure S2A). Same results were also found in NMDS analysis based on Bray-Curtis distance (ANOSIM test: R=0.01342, P=0.004; Figure S2B) or unweighted UniFrac dissimilarity (ANOSIM test: R=0.02529;P=0.001; Figure S2C).
Dysbacteriosis in gut of DKD
At the phylum level, proportion of the four dominant bacteria (i.e., Firmicutes, Bacteroidetes, Proteobacteria and Actinobacteria) could achieve 95% of all sequences (Figure S3A). Compared with Con group, relative decrease of phyla Bacteroidetes and relative increase of Proteobacteria were observed in DKD, accompanying with increased Firmicutes/Bacteroidetes ratio (Figure 2A). At the class level, bacteria including Gammaproteobacteria, Verrucomicrobiae and Erysipelotrichia were abundant in DKD, while Betaproteobacteria and Bacteroidia were relatively decreased (Figure 2B). Correspondingly, a total of 16 bacteria at the order level and 31 bacteria at the family level were identified as significantly different between two groups (Figure 2C, 2D). In more detail at the genus level, we observed relative increase of potential pathogens consisting of Peptostreptococcus, Streptococcus, Peptostreptococcaceae_incertae_sedis, Erysipelotrichaceae_incertae_sedis, Enterococcus and Escherichia-Shigella, enrichment of sulphate-degrading bacteria Desulfovibrio, as well as decrease of “beneficial bacteria” including Roseburia, Blautia, Bacteroides and Faecalibacterium in DKD patients. Notably, relative abundance of Akkermansia, a mucin-degrading bacteria, was also reversely increased in DKD (Figure 2E, Figure S3B).
To evaluate microbial differences at the taxonomic level, LEfSe analysis was performed on microbial content of the 180 DKD patients and 179 healthy controls (LDA score >2.0, P value <0.05). Besides those mentioned above, decreased abundance in genera including Lachnospira, Pseudobutyrivibrio and Anaerostipes and increased abundance in Bifidobacterium, Lactobacillus were observed in DKDs. We also demonstrated that genera Bacteroides (LDA=4.90, P=1.95e-19) and Escherichia_Shigella (LDA=4.65,P=7.92e-25) were strongly correlated with the disease status. All results above showed gut microbiome of DKD remarkably deviated from that of healthy status. (Figure S4).
DKD identification based on OTUs
Considering microbial striking difference in DKDs compared with healthy controls, a strategy based on potential microbiota-targeted markers for distinguishing DKD from healthy status was proposed. As mentioned in ‘Statistical analysis’ section, we performed random forest analysis on more than 1000 OTUs (number of sequences > 0.0001) selected by Wilcoxon test (P<0.05 and Q<0.05). As a result, 328 OTUs were identified as key markers and distribution of the first 73 OTUs (importance value >0.001) in 180 DKD patients and 179 healthy controls were shown in heatmap. Differential relative abundance between DKD and Con was observed (Figure 3). We also delineated the profile of top abundant 50 OTUs in each sample and found a relatively insignificant difference between DKD and Con group (Figure S5).
With five-fold cross validation performed on random forest model, an OTU combination consisted of 5 OTUs (model 1) was identified with the highest diagnostic value. As shown in Figure 4A, this optimal set had the minimum cv error-rate. These OTUs’ contribution and stability in construction of disease classifier were respectively evaluated by decreasing accuracy and Gini (Figure 4B-C). Same identification method was further implemented on a range of OTUs whose abundance larger than 0.0005 and 22 OTUs were identified as another microbial combination with diagnostic value (Model 2; Figure 4D). Their function in model construction was shown in Figure 4E-F.
We applied ROC curve in the training set containing 118 DKD patients and 132 healthy controls and found combination of 5 OTUs could separate DKDs from healthy controls with 83.71% accuracy, while 81.56% in model 2 (Figure 5A). Although average POD values in DKD group were higher than Con group in both OTU sets, there was a more significant discrepancy in model 1 (Figure 5B, 5C). Remaining 62 DKDs and 47 healthy controls were incorporated into testing set to assess diagnostic efficiency of the two OTU sets. As expected, area under ROC curve (AUC) could achieve 80.89% in model 1 and 76.75% in model 2. Higher POD value in DKD than Con group was also observed in Figure 5D, 5E. These data suggested that increasing OTU markers couldn’t improve predictive performance.
Microbiome-associated functional changes in DKD
PICRUSt was used to analyze microbiome-associated metabolism in disease or healthy status. A total of 145 KEGG pathways showed distinct relative abundance between DKD and Con group (LDA score >2.0 and P value <0.05). Functional categories including membrane transportation (ABC transporters and phosphotransferase system), cell motility (bacterial motility proteins), lipid metabolism (fatty acid, glycerolipid, glycerophospholipid metabolism and unsaturated fatty acids biosynthesis), amino acid metabolism (aromatic amino acids, glutathione and others metabolism) and carbohydrate metabolism (pyruvate metabolism) showed higher levels in DKDs, while metabolism of cofactors, vitamins and specific amino acids (i.e., alanine, aspartate, glutamate, arginine and proline), oxidative phosphorylation were decreased in DKDs compared with healthy controls (Figure 6).