The OTU profile was obtained after 16S rDNA data analysis
To study the urine microorganisms in obstructive UR, we performed 16S sequencing on urine samples from 84 Chinese participants, including 34 patients with UR caused by lithiasis (stone UR), 25 patients with UR due to the urinary tract tumors (tumor UR), and 25 healthy controls. These negative controls were not sequenced, as the agarose gel electrophoresis showed that no electrophoretic band of microbial DNA, which indicates that there was no artifact and contaminant in the environment. A total of 727 MB of 300 bp paired-end reads were generated after sequenced on the MiSeq platform, and the average number of reads per sample was 36,072 ± 3,376 reads (Table S1). Two samples from the stone UR group, S18 and S32, were removed for their low assigned microbial reads, which were less than 8000. Finally, we obtained 310 OTUs from 25 controls, 32 stone UR and 25 tumor UR individuals (Table S2). The clinical and demographic characteristics of all remaining UR patients and controls are shown in Table 1. In terms of age, there was no significant difference between the stone UR group and control group, while the tumor UR group subjects were older than the other two groups. There was no statistically significant difference in gender among these three groups
Table 1
Clinical characteristics of the enrolled participants
Clinical indexes | Control (n = 25) | Stone UR (n = 32) | Tumor UR (n = 25) | P-value Control vs Stone UR | P-value Control vs Tumor UR | P-value Stone UR vs Tumor UR |
Age(year) | 42.92 ± 20.15 | 52.94 ± 14.40 | 61.60 ± 13.24 | 0.054 | 0.0013 | 0.034 |
Gender | | | | | | |
Female | 17(68%) | 21(65.62%) | 16(64%) | 1 | 1 | 1 |
Male | 8(32%) | 11(34.38%) | 9(36%) |
DBIL(µmol/L) | 2.50 ± 0.70 | 4.14 ± 2.74 | 4.56 ± 5.61 | 0.030 | 0.25 | 0.49 |
IBIL(µmol/L) | 4.36 ± 1.85 | 6.53 ± 2.67 | 5.46 ± 2.76 | 0.043 | 0.58 | 0.12 |
eGFR | 97.56 ± 23.70 | 70.63 ± 32.19 | 40.30 ± 33.32 | 0.013 | 0.00010 | 0.0032 |
ALP(µ/L) | 91.08 ± 30.99 | 91.70 ± 33.98 | 144.92 ± 234.24 | 0.96 | 0.79 | 0.81 |
UA(µmol/L) | 317.20 ± 54.00 | 367.96 ± 103.13 | 454.01 ± 160.95 | 0.24 | 0.039 | 0.063 |
RBC(/µL) | 9.45 ± 10.60 | 691.23 ± 1348.03 | 3489.10 ± 6187.54 | 0.0015 | 0.068 | 0.56 |
WBC(/µL) | 55.45 ± 123.05 | 2110.97 ± 5085.15 | 1472.82 ± 3973.84 | 0.0096 | 0.0026 | 0.59 |
pH | | | | | | |
> 6.5 | 2(8%) | 12(37.5%) | 7(28%) | 0.27 | 0.68 | 0.56 |
≤ 6.5 | 9(36%) | 17(53.1%) | 16(64%) |
Urobilinogen | | | | | | |
Positive | 1(4%) | 3(9.4%) | 2(8%) | 1 | 1 | 1 |
Negative | 10(40%) | 27(84.4%) | 20(80%) |
Proteinuria | | | | | | |
Positive | 0(0%) | 20(62.5%) | 16(64%) | 0.00016 | 0.000086 | 0.76 |
Negative | 11(44%) | 10(31.2%) | 6(24%) |
DBIL: Direct bilirubin IBIL: Indirect bilirubin eGFR: estimated glomerular filtration rate |
ALP: Alkaline phosphatase UA: uric acid RBC: red blood cell WBC: white blood cell. |
Increased urine microbial diversity in stone UR and tumor UR individuals
To depict the bacterial richness of each group, we randomly sampled the same amount of reads from each sample and performed rarefaction analysis to estimate the observed OTUs that could be identified from these sequences. As shown in Fig. 1A, all three curves had reached plateaus, which indicated that the amount of sequenced data were sufficient to detect the microbial feature. The acquisition rate of OTUs in control samples was strikingly lower than that in stone UR and tumor UR groups. Measured by the Shannon index, the urine microbial diversity of the stone UR and tumor UR was significantly greater than that of healthy controls, however, there was no significant difference between the two UR groups (Fig. 1B, Shannon: controls vs stone UR, P = 1.10 × 10− 2; controls vs tumor UR, P = 4.57 × 10− 5; stone UR vs tumor UR, P = 0.29). Likewise, the Simpson index of stone UR and tumor UR was significantly higher than that of controls (Fig. 1C, Simpson: controls vs stone UR, P = 3.70 × 10− 2; controls vs tumor UR, P = 1.61 × 10− 3; stone UR vs tumor UR, P = 0.28). The Chao1 index and Ace index, which measure the richness of the community, showed that the richness of microbiome in two UR groups was significantly higher than that in controls (Fig. 1D, Chao1: controls vs tumor UR, P = 1.25 × 10− 3; controls vs tumor UR, P = 6.68 × 10− 5; stone UR vs tumor UR, P = 0.15; Ace: controls vs tumor UR, P = 1.18 × 10− 3; controls vs tumor UR, P = 7.41 × 10− 5; stone UR vs tumor UR, P = 0.19). In addition, a Venn diagram (Fig. 1E) was used to characterize the overlapped OTUs among the three groups. A total of 212 OTUs were shared among the 3 groups, and there were 20, 3, and 2 OTUs were unique to the controls, stone UR and tumor UR groups, respectively. It is worth noting that there were as many as 42 OTUs shared between the stone UR and tumor UR groups.
To further investigate whether there were differences among the three groups in the urine microbiota spectrum, PCoA was performed based on the unweighted UniFrac distances of the 16S rRNA sequence at the OTU level. There were differences in β-diversity among the three groups, as shown in Fig. 1F (PERMANOVA, pseudo-F statistic: 11.88, P = 1.00 × 10− 3). In addition, differences between each of the two groups were further evaluated based on unweighted and weighted UniFrac distances and Bray-Curtis dissimilarity (Table S3) These results showed that both UR groups were significantly different from the control group, while the microbial compositions of two types of obstructive UR could not be separated.
Altered urine microbial communities in stone UR and tumor UR
To explore the specific microbial signature of stone UR and tumor UR, we evaluated the relative abundance of taxa in three groups and found that the two UR groups were significantly different from the control group at the phylum level. However, the stone UR patients and the tumor UR patients had similar urine microbial compositions. Proteobacteria was the most abundant phylum in three groups, followed by Bacteroidetes (Figure S1A). Compared with controls, stone UR and tumor UR individuals exhibited a significant increase in the phylum Bacteroidetes in urine (Figure S1B). In accordance with the phylum level, the urine microbial compositions of the two types of obstructive UR patients were similar at the genus level, but both were different from that of the control group (Figure S1C). It is noteworthy that there was a total of 44 bacteria with significant differences between any two of the three groups at the genus level (q < 0.01, Wilcoxon rank sum test, Fig. 2A). Twelve out of the 44 are displayed in Fig. 2B. Elizabethkingia, Proteus, Sphingomonas, Pseudomonas, Acinetobacter, Sphingobacterium and Myroides were overrepresented in the stone UR and tumor UR groups. In contrast, Lactobacillus, Streptococcus, Gardnerella, Prevotella and Atopobium, which were decreased in stone UR and tumor UR patients, were enriched in controls.
To further confirm the specific bacteria associated with obstructive UR, LEfSe was used, which identified 14 discriminative features, and their relative abundances significantly varied between stone UR individuals and controls, which was completely consistent with the result in Fig. 2A, as evaluated by the Wilcoxon rank sum test (Figure S2A). Curvibacter was a newly found bacterium enriched in the tumor UR group, while Escherichia was found enriched in control group (Figure S2B).
Evaluation of the connections among these different genera was performed by a Spearman correlation test. Significant positive correlations were found in genera enriched obstructive UR, such as for Gluconacetobacter and Myroides (R = 0.98, P = 1.24 × 10− 60); Gluconacetobacter and Sphingobacterium (R = 0.94, P = 2.97 × 10− 40); Pseudomonas and Comamonas (R = 0.85, P = 6.08 × 10− 24); and so on. Likewise, positive correlations were also found in genera that were enriched in control subjects. More interestingly, the bacteria enriched in urine of the two obstructive UR patients were negatively correlated with those enriched in controls (Fig. 2C), such as Myroides and Lactobacillus (R = -0.71, P = 7.93 × 10− 14); Sphingobacterium and Lactobacillus (R = -0.68, P = 3.45 × 10− 12); Elizabethkingia and Atopobium (R = -0.46, P = 1.44 × 10− 5); and so on. We further explored the association of the urine microbiome with clinical manifestations (Fig. 2D) and found that there were some significantly negative correlations between the estimated glomerular filtration rate (eGFR) level and the microorganisms that were enriched in obstructive UR patients, such as Pseudomonas, Methylobacterium, Elizabethkingia, and so on. On the contrary, positive correlations were observed between the uric acid (UA) level and the genera that were enriched in obstructive UR patients, such as Pseudomonas, Stenotrophomonas, Sphingomonas, and so on.
Classification of disease status using bacterial genus-level biomarkers
To explore the potential diagnostic value of the urine microbiome in stone UR and tumor UR, we constructed a random forest classifier to discriminate urinary retention samples from control samples. We only selected the microorganisms that were significantly enriched in stone UR and tumor UR to construct the classification models. Finally, 30 and 34 genera signatures were selected for further analysis for stone UR and tumor UR, respectively. The cross-validation error curve distribution was obtained from five trials of five-fold cross-validation. Both 8 biomarkers were selected as the optimal marker set to distinguish stone UR or tumor UR from the control group (Figure S3A, B). The performance of these optimal marker models was assessed by 100 random ROC analyses, and the average AUC value achieved 92.29% between the stone UR and control group and 97.96% between the tumor UR and control group (Fig. 3A, D). The average MDA for the random 100 times of these optimal markers are shown in Fig. 3B and Fig. 3E. These results showed that 7 out of 8 bacteria were identical, which were used to distinguish stone UR or tumor UR from the control group, including Mycobacterium, Agrobacterium, Ralstonia, Delftia, Acinetobacter, Methylobacterium and Sphingomonas. Here, we further validated our previous finding that stone UR and tumor UR have approximate urine microbial background although they result from different obstruction causes. The POD value was significantly increased in stone UR group versus control group, and a similar trend was also found in tumor UR group (Fig. 3C, F; P < 2.2 × 10− 16 for both comparisons).
To select the optimal marker set of stone UR and tumor UR based on the urine microbiome, we used the genera with P < 0.05 which were measured by the Wilcoxon rank sum test. Only Sediminibacterium was selected from two different bacteria as the optimal marker after five repeated five-fold cross-validation trials Figure S3C). The average AUC value was 58.44% after running 100 random ROC analyses (Figure S3D). The MDA of Sediminibacterium is shown in Figure S3E. The probability of tumor UR was significantly higher than stone UR in the tumor UR group (stone UR vs tumor UR, P = 1.00 × 10− 15), which was consistent with these clinical data (Figure S3F).
Microbial functional altered in stone UR and tumor UR
To explore whether the function of the urine microbiome of stone UR and tumor UR has changed, we used PICRUSt to predict the functional components of the 16S rRNA gene sequencing data of all samples. The OTU profile of these three groups was aligned to level 3 of the KEGG database, and COG abundance was calculated (Table S4, Table S5). Principal component analysis (PCA) based on the KEGG pathways showed that the control group was strikingly separated from stone UR and tumor UR groups (Fig. 4A; PERMANOVA, pseudo-F statistic: 13.05, P = 1.00 × 10− 3), which was consistent with the result based on COG categories (Figure S4A; PERMANOVA, pseudo-F statistic: 11.97, P = 1.00 × 10− 3). We found that both obstructive UR groups were significantly different from control group, while the functional structure between stone UR and tumor UR was similar (Fig. 4B). Twenty-five KEGG pathways were differentially enriched between each two of the three groups (adjusted p-value < 0.01, Wilcoxon rank sum test). There were 14 pathways involved in membrane transport, signal transduction, genetic information processing, carbohydrate metabolism and nucleotide metabolism, such as that for purine and methane, which were significantly reduced in the stone UR group and tumor UR group. We observed 11 pathways that were increased in the stone UR and tumor UR groups, including amino acid metabolism and energy metabolism. Intriguingly, the abundance of pathways associated with membrane transport functions, such as ABC transporters and the phosphotransferase system (PTS), was negatively correlated with these genera enriched in stone UR and tumor UR groups. In accordance with the KEGG function result, most of the energy metabolism, carbohydrate metabolism and amino acid metabolism pathways had changed in the COG annotation (Figure S4B). There was a strong positive correlation between the significantly enriched functions in both obstructive UR groups and their enriched genera (Fig. 4C). For instance, Pseudomonas, Acinetobacter and Sphingomonas were significantly related to valine, leucine and isoleucine degradation and glycine, serine and threonine metabolism. These amino acids may be the product of bacteria as Corynebacterium glutamicum can produce amino acids on a large scale reported previously [30]. ABC transporters, the PTS, transporters and other ion-coupled transporters that were significantly enriched in controls were mainly negatively correlated with patients with UR microbial features. The same trend was found in the COG function analysis with these significantly different bacteria (Figure S4C).