Differences in Gut Microbiota Proles and Functions Between End-stage Renal Disease and Healthy Populations

Background: Patients with end-stage renal disease (ESRD) have extremely high risks of mortality and morbidity, as well as altered gut microbiota and impaired intestinal barrier function. The translocation of gut-derived molecules in ESRD contributes to systemic complications. In this study, we evaluated the gut microbiome difference in ESRD patients compared to age- and gender-matched subjects without kidney disease in discovery and validation cohorts. Results: Compared to controls with normal renal function, an increased α-diversity and distinct β-diversity were found in ESRD subjects. The increase in α-diversity was correlated with protein-bound uremic toxins, particularly hippuric acid. A higher microbial dysbiosis index (MDI) was found in ESRD patients with the following enriched genera: Facealibacterium, Ruminococcus, Fusobacterium, Dorea, Anaerovorax, Sarcina, Akkemansia, Streptococcus, and Dysgonomonas. MDI at the genus level demonstrated highly differentiated accuracies between ESRD and control subjects in the discovery cohort (area under the curve [AUC] of 81.9%) and between ESRD and control subjects in the validation cohort (AUC of 83.2%). On functional enrichment analysis with gut metabolic modules, ESRD subjects presented with increased saccharide and amino acid metabolism when compared with matched controls. Conclusions: An enriched but dysbiotic gut microbiota was presented in ESRD patients, in which the bacteria that were present increase amino acid metabolism linked to the production of protein-bound uremic toxins.


Biodiversity of the gut microbiota
Results of α-diversity analysis (observed operational taxonomic unit [OTU], Chao1, Shannon, Simpson, and inverse Simpson) differed signi cantly among ESRD participants and matched controls in the discovery and validation cohorts (observed OTU, P < 0.001; Chao1, P < 0.001; Shannon, P < 0.001; Simpson, P < 0.001; inverse Simpson P < 0.001; Fig. 2A). The differences in microbial composition (β-diversity) between ESRD patients and normal subjects are illustrated in the PCoA plot (P < 0.001; Fig. 2B). Distinct microbial compositions were found at different taxonomic levels in the discovery and validation cohorts (All P < 0.001; Figure S1).

Overall gut microbiota composition comparisons between ESRD patients and healthy controls
Gut microbiota composition and structure differed between ESRD patients and control subjects in the discovery cohort. At the phylum level, ESRD patients had lower levels of Bacteroidetes than controls. However, no differences were found in the levels of Firmicutes. Lower proportions of classes Bacteroidia and Betaproteobacteria and orders Bacteroidales and Burkholderiales were observed in ESRD patients when compared with controls. At family and genus levels, ESRD patients exhibited increased relative abundances of Ruminococcaceae and Ruminococcus but decreased Prevotellaceae and Pravotella ( Figure S2). Other differences are presented in Figure S2.

Speci c microbial taxa associated with ESRD patients and healthy controls
To identify the most relevant taxa responsible for the differences between ESRD patients and controls, discriminant analysis by linear discriminant analysis (LDA) effect size (LEfSe) method was performed and a cladogram was produced to represent the connections among taxa of differential abundance at different taxonomic levels (Fig. 3A). LDA scores suggested that differentially abundant taxa were potential biomarkers. Discriminating features between the two groups across different taxonomic levels were observed ( Figure S3). In ESRD patients, predominant genera were Facealibacterium, Ruminococcus, Fusobacterium, Dorea, Anaerovorax, Sarcina, Akkemansia, Streptococcus, and Dysgonomonas compared to control subjects. In subjects with normal kidney function, predominant genera were Prevotella, Lachnospira, Megamonas, Sutterella, Dialister, Acidaminococcus, Rothia, Megasphaera, 5-7N15 (belonging to the family of Bacteroidaceae), and Catenibacterium compared to ESRD patients (Fig. 3B). We found increases in a total of 67 taxa in ESRD patients, including 20 species, 23 genera, 10 families, 6 orders, 5 classes, and 3 phyla (Table S1).

Microbial dysbiosis as ESRD marker
To illustrate the discriminative value of gut dysbiosis for ESRD, we constructed the microbial dysbiosis index (MDI) to differentiate ESRD subjects and controls. A higher MDI was found in ESRD patients (Fig. 4A). Genus-level MDI was associated with an AUC value of 81.9% in the discovery cohort and AUC value of 83.2% in the validation cohort (Fig. 4B). To evaluate the MDI at different taxonomic levels, prior discriminative values were found at the genus, species, and family levels ( Figure S4). To consider the confounders that may in uence MDI, multivariate-adjusted logistic regression was performed. After adjusting for covariates (including diabetes mellitus, hypertension, and hyperlipidemia) in the logistic regression models, ESRD patients continued to demonstrate an increased risk of microbial dysbiosis compared to matched controls (Table 2). Functional characterizations of the microbiome in ESRD patients and controls Analysis of gut metabolic modules (GMMs) revealed the alteration of pathways in ESRD patients' gut microbiota in the discovery and validation cohorts. Compared to controls, ESRD patients had higher numbers of mapped genes with GMM-predicted metabolism of saccharides (lactose, arabinose, fucose, xylose, and pentose), amino acids (glutamate, alanine, cysteine, threonine, arginine, and histidine), and other metabolites (glycerol, pyruvate, lactate, mucin) (Fig. 5). The full results of GMM enrichment analysis are presented in Table S2. From the generated Clusters of Orthologous Groups (COGs), ESRD patients demonstrated increased gut microbial functions related to signal transduction histidine kinase, glycosyltransferase, Fe 2+ -dicitrate sensor, and betagalactosidase ( Figure S5). As for the enrichment of predicted Traditional Kyoto Encyclopedia of Gene and Genomes (KEGG) Orthology (KO) groups, ESRD patients showed increased gut microbial functions associated with multiple sugar transport system, antibiotic transport system, and beta-galactosidase, as well as decreased functions of the lipoprotein-releasing system and major facilitator superfamily (MFS) transporters ( Figure S6).

Exploration of the relationships between α-diversity and host parameters in ESRD patients
We found positive correlations between microbial-derived metabolites (hippuric acid [HA], IS, and PCS) and α-diversity in the discovery cohort. Circulating HA level was positively associated with microbial richness (r = 0.45, P < 0.001), Shannon index (r = 0.34, P < 0.01), Simpson index (r = 0.23, P < 0.05), and inverse Simpson index (r = 0.37, P < 0.001). No signi cant association was found between α-diversity and other clinical parameters, such as demographic data, comorbidities, and clinical laboratory data (Fig. 6). Considering that protein intake predisposes such patients to the development of protein-bound uremic toxins, we evaluated the α-diversity differences between high and low normalized protein catabolic rate (nPCR), which re ects daily dietary protein intake in stable dialysis patients. ESRD patients with nPCR value of more than1.2 g/kg per day had higher α-diversity (Shannon index, Simpson index) than ESRD patients with nPCR value of less than 1.2 g/kg per day ( Figure S7). These ndings implied that higher protein intake predisposes ESRD patients to the development of protein-bound uremic toxins that further predict microbial diversity.

Discussion
In this study, we compared differences in gut microbiota between ESRD patients and age and sex-matched controls in discovery and validation cohorts. ESRD patients had higher α-diversity (within-sample diversity), distinct β-diversity (between-sample diversity), and increased gut microbial dysbiosis compared to the healthy population. In general, the gut microbiota of a healthy individual is more diverse than that of a diseased individual. However, this is not the case in ESRD patients. Diversity is fundamental to ecology as an indicator of the state of an ecosystem due to its relationships with stability, productivity, and functioning [25]. However, a higher diversity is not always better [25][26][27]. For example, increased α-diversity has been found in Parkinsonism [28].
No consistent reduction in diversity among patients relative to healthy individuals was found on the meta-analysis of case-control studies [29]. Host genetics, geographical region, environmental exposures (including dietary habits and drugs), and lifestyle factors are all essential to gut microbial diversity [30,31]. In our study, microbiota diversity indices positively correlated with protein-bound uremic toxins (IS, PCS, HA) suggesting that diet drives biodiversity in the ESRD patients. The relationship between protein and microbiota diversity is further supported by higher microbial diversity in the high nPCR group. Similar nding reporting a positive correlation between urea levels (represent the protein intake) and microbiota diversity supports this claim [32]. Furthermore, a high-protein intake increases the plasma levels of protein-bound uremic toxins and urea in healthy individuals [33] and ESRD patients [34]. Thus, high-protein intake may increase gut microbiota diversity.
Recent metabolomics reports have indicated that certain protein-bound uremic toxins are strongly and positively associated with α-diversity. [35,36] Our results demonstrated that the production of amino acid metabolites affected the composition of gut microbiota, previously identi ed by metabolites, such as hippurate and PCS, in non-CKD subjects [35,36]. Our discovery of the positive relationship between IS and Shannon index agrees with previous studies, reporting a positive association between indole-containing compounds (Indolepropionate) consumption and Shannon diversity among non-CKD subjects [35].
In allogeneic stem cell transplantation patients, a positive association between the urinary level of microbiota-derived indole and microbial diversity has been found [37]. In contrast, no signi cant correlation has been demonstrated between IS/PCS and α-diversity in pediatric patients with ESRD [23]. Therefore, more studies are needed to explore this issue.
Individuals with CKD, especially ESRD, are often advised to follow restrictive diets based on individual nutrients such as sodium, potassium, and phosphorus to minimize circulating electrolyte imbalance and uid retention. In the early stages of CKD, low protein intake is recommended to preserve kidney function and limit circulating nitrogenous waste. Patients that develop ESRD (complete kidney failure) and undergo dialysis are advised to increase protein consumption to more than the recommended amounts for healthy individuals and CKD patients. This is based on the need to preserve lean body mass [38,39]. Current international recommendations for daily dietary protein intake are 0.8 g/kg for the general population [40], 0.6 to 0.8 g/kg for non-dialysis CKD patients [18,[41][42][43], and 1.1-1.4 g/kg for ESRD patients on dialysis [18,[43][44][45]. A high protein diet is recommended for ESRD patients on dialysis as reduced protein intake is associated with increased all-cause mortality [46,47]. Insu cient protein intake in ESRD patients may lead to di culty counteracting protein loss and catabolism during dialysis [48]. A high protein diet, similar to the Western diet (which is rich in animal proteins and fats), stimulates the overgrowth of proteolytic bacteria, resulting in dysbiosis and accumulation of proteolytic-derived uremic toxins [49]. Also, impaired small intestine protein digestion and amino acid absorption in ESRD patients results in more proteins reaching the large intestine [50]. Prolonged colonic transit time is not only associated with high richness and diversity of the microbiota but also higher bacterial protein catabolism, facilitating increased protein fermentation by proteolytic (putrefactive) bacteria for energy metabolism [51][52][53][54][55][56]. In our study, increased microbial function related to amino acid metabolism was demonstrated in ESRD patients on GMMs enrichment analysis. However, increased α-diversity can be a double-edged sword in ESRD. Our ndings of an association between proteinbound uremic toxins and microbial diversity may partly explain this paradox in ESRD patients.
In addition to the impact of protein intake and uremic toxins on α-diversity in ESRD patients, reduced consumption of fruits, vegetables, and dietary ber to avoid potassium overload causes gut dysbiosis [56]. Other contributing factors include the rise in gastrointestinal luminal pH rise due to uremic milieu (ammonia and ammonium hydroxide) [7,13,57,58] and complex drug exposure (e.g., antibiotics, phosphate binders, and iron) [59][60][61]. As expected, we found higher MDI in ESRD patients than in controls, which may help to differentiate ESRD patients from subjects without kidney disease. The association between high MDI and ESRD persisted even after adjusting for diabetes mellitus, hypertension, and hyperlipidemia. Dysbiosis of the gut microbiota in patients with kidney disease is characterized by a decrease in bacterial species with saccharolytic fermentation activity (e.g., Lactobacillus and Prevotella) and enrichment of bacterial strains with proteolytic fermentation activity (e.g., Bacteroides and Clostridium), which leads to the increased levels of circulating uremic toxins followed by chronic in ammation [62]. The previously reported results just correspond to the increased abundance of Clostridium and several Bacteroides strains in our ESRD patients compared to healthy controls.
In the present study, there were distinct gut microbial compositions in ESRD patients and healthy subjects. Increased abundance of several taxa has been reported [7, 22-24, 63, 64], such as the phylum Actinobacteria, class Erysipelotrichi, orders Enterobacteriales and Erysipelotrichales, families Enterobacteriaceae, Verrucomicrobiaceae, Clostridiaceae, and Coriobacteriaceae, and genera Faecalibacterium, Desulfovibrio, and Cloacibacillus (Table 3). In line with the ndings of a previous report, microbial families that possess urease (Enterobacteriaceae) [7] and indole/p-cresyl-forming enzymes (Clostridiaceae, Verrucomicrobiaceae, and Enterobacteriaceae) [7,65] were enriched in ESRD patients in our study. These microbial families harbor genes that encode tryptophanase-tyrosine indol-lyase, suggesting their essential roles in uremic toxins production. In an experimental study, it was con rmed that bacterial tryptophanase from Bacteroides species processes tryptophan to indole [66]. We also found an abundance of species Bacteroides ovatus, Bacteroides uniformis, Bacteroides fragilis, and Bacteroides acidifaciens in ESRD patients that might contribute to the elevation of uremic toxins. In addition, the Ruminococcaceae family, which can ferment tyrosine to p-cresol [67], was enriched in ESRD patients in our study, as well as, other bacterial genera reportedly increased with higher levels of protein-bound uremic toxins, such as Akkermansia and Blautia [68]. In theory, the in ux of uremic toxins and urea into the gastrointestinal lumen provoke the overgrowth of bacteria that produce urease, uricase, indole, and p-cresol forming enzymes, generating a vicious cycle of in ammation and oxidative stress in ESRD patients [22]. Increased abundance of phylum Verrucomicrobia [63] and family Enterobacteriaceae [13,64] in CKD patients has also been found. However, Actinobacteria phylum and Akkermansia genera are reduced in CKD patients [63] compared to ESRD patients. A high protein diet in experimental animal models leads to the enrichment of the Akkermansia genus [69]. Thus, the discrepancy regarding Akkermansia abundance between CKD and ESRD can be partly explained by differences in protein intake. There were several limitations to this study. For example, the cross-sectional design led to the inability to prove causality, and the sequencing of the 16S rRNA gene limited the analysis to the strain level. In addition, subjects were Asian ESRD patients. Their genetics, epigenetics, environmental factors, and diet may differ from those of other populations. For instance, the phylum Bacteroidetes has been reported to be most dominant in healthy Chinese adults [70,71]. The switch from Prevotella enterotype to Bacteroides enterotype in ESRD patients [24] could only be observed in participants from the validation cohort but not from the discovery cohort. Lastly, our targeted protein-bound uremic toxin approach allowed for precise metabolite measurements but was limited to prede ned target metabolites. In terms of its strengths, a large number of ESRD participants were enrolled in this discovery and validation study to analyze 16S rRNA amplicon pro ling, enabling us to adequately investigate microbial diversity, gut dysbiosis, and microbial function.

Conclusions
ESRD patients have a distinct microbial composition, increased microbial diversity, predicted gut dysbiosis, and increased microbial function related to amino acid metabolism compared to healthy controls. Circulating protein-bound uremic toxins, especially HA, explained the increased microbial diversity in ESRD patients. Our ndings demonstrated an association between gut microbiota and kidney failure treatment strategy, especially in terms of dietary recommendations.

Study participants
From August 2017 to February 2018, 194 ESRD patients from the dialysis units of KMUH and TTCH, Taiwan, were enrolled. Eligible participants received regular hemodialysis three times per week, 3.5-4 hours each time with high-ux dialyzers. Control participants without kidney disease were enrolled from the health management center at Taichung Veterans General Hospital, Taiwan, from January 2015 to December 2015. Participants with previous malignancies, prior gastric surgery, or who had received antibiotics within three months before enrollment were not included. Experiments were carried out following the
Comorbidities, laboratory data, and clinical variables Sociodemographic data, age, sex, medical history, and prescribed medications were obtained for all participants from electronic health care records. Diabetes was de ned as HbA1C of 6.5% or higher or the use of oral antidiabetic agents or insulin. Hypertension was de ned as blood pressure of 140/90 mmHg or higher or the use of blood pressure-lowering drugs. Hyperlipidemia was de ned according to physician diagnosis or the use of lipid-lowering medications.
Blood samples were obtained from patients after overnight fast through arteriovenous access before a scheduled midweek hemodialysis session.
Fecal sample collection and DNA extraction All participants provided a stool sample that was immediately frozen after home collection and delivered by using cooler bags within 24 hours via commercial transport to the laboratory (Germark Biotechnology, Taichung Bioinformatics analysis of the 16S rRNA amplicon was conducted as previously described [73]. Brie y, on a per-sample basis, pair-end reads were merged using USEARCH (v8.0.1623), with 8 bp minimum overlap of reading pairs [74]. Merged sequences were processed using Mothur (v1.35.1) to remove reads shorter than 450 bp or longer than 550 bp, as well as any read not meeting the minimum quality score of 27 [75]. Reads containing ambiguous base or homopolymer exceeding 8 bp were discarded. Chimeric sequences were identi ed and removed by USEARCH (reference mode and 3% minimum divergence).
Quality-ltered and non-chimeric reads were analyzed (UPARSE pipeline [76]) to generate OTUs per sample at a 97% similarity cutoff. The identi ed OTU representative sequences were then aligned using USEARCH to determine the corresponding taxonomy from the Greengenes reference database (version 13.5) [76,77]. OTUs without a hit or with only a weak hit were excluded from the following analysis.

Analysis of richness and biodiversity
The α-diversity measures the richness and evenness of taxa within each sample. The β-diversity compares the taxa between pairs of individual samples. The α-diversity indices were estimated with the R package "phyloseq". The species richness indices (Observed, Chao 1) and species diversity (Shannon, Simpson, Inv Simpson) were tested by Student's t-test [78]. The β-diversity was assessed by computing the Bray-Curtis distance and displayed via Principal Coordinates Analysis (PCoA) to evaluate microbial compositional similarities between ESRD patients and controls [79]. Between-class analysis of inertia percentage was performed by the R package "ade4" using Monte-Carlo test with 10,000 permutations to assess the statistical signi cance between groups [80].

Bacterial community comparisons between ESRD and control subjects
A pseudo-count of 0.0001 was added to the relative abundance (in percentage) before the logarithmic transformation [81]. Differential abundance analysis was performed using the Wilcoxon rank-sum test. Only taxa with average abundance > 0.2% and sample coverage > 10% were included in the differential analysis. LEfSe was applied to determine the taxa of signi cantly differential abundance between the ESRD patients and controls, evaluated with α of 0.05 (Kruskal-Wallis and Wilcoxon tests) and effect size threshold of 2 using the stand-alone implementation [82]. The results were plotted on a cladogram based on phylogenetic relationships among taxa.

Microbial dysbiosis index analysis
MDI is an index that contrasts the abundance of enriched taxa in case and control cohorts [83]. In the present study, MDI was determined by the logarithm of accumulated relative abundance of enriched taxa (LEfSe [82]) in ESRD and control subjects. The 10% of globally non-zero minimal relative abundance was used in the logarithm if the sum was zero. MDI was analyzed at six taxonomic levels from phylum to species, and its distribution was visualized. The empirical probability of ESRD was predicted by the proportion of ESRD over all subjects (i.e., control and ESRD) in the corresponding MDI. Logistic regression was applied to model MDI versus ESRD probability. The receiver operating characteristic curve was generated from all samples, and the most signi cant area determined the best taxonomic level for classi cation under curve. MDI of best sensitivity and speci city was used for validation of predicted performance.

Gut metabolic module analysis
Functional analysis is a critical advantage of shotgun sequencing data. KEGG annotation methods hold redundant information and are not suitable for the interaction between host and microorganism. Thus, we investigated GMMs [84] to show the functional changes in ESRD patients. GMMs are bacterial and archaeal metabolic pathways associated with the human gut, mainly anaerobic fermentation processes. Each module is composed of prokaryotic and archaeal KEGG Orthology (KO) groups to describe an enzymatic process that converts the input compound to output metabolite. GMMs were rst grouped by their position in the gut metabolic map (i.e. input, central, and output; module numbers 75, 11, and 17, respectively), then by 10 metabolic categories and 30 subcategories. There were 103 GMMs in total. To infer the GMM pro le, the KO pro le was rst predicted by PICRUSt [85], then subjected to Omixer-RPM [86] with default settings except for output format, which was set to 2. To identify GMMs with differential abundance, enrichment analysis was performed by twotailed Wilcoxon test with Benjamini-Hochberg false discovery rate correction for multiple testing.
Bacterially derived protein-bound uremic toxin pro ling Circulating free form IS, PCS, and HA were measured by mass spectrometry as previously described. Brie y, each ESRD patient's serum sample (300µL) was packed into a centrifugal lter device (Amico Ultra 3K, MerckMillipore). After centrifugation, the supernatant was evaporated and re-dissolved with 100 uL 30% acetonitrile (MeCN) and 0.1% formic acid. Subsequently, 10 µL IS-d4 (internal standard purchased from Sigma-Aldrich, 1000 ng/mL) were added to the sample and ltrated with 0.22 µm polytetra uoroethylene (PTFE) lters for mass spectrometer analysis. The tandem mass spectrometry system was equipped with Micro electrospray ionization (ESI) ion source, coupled with Acella 1250 Ultra-high performance liquid chromatography (UHPLC

Competing interests
The authors declare that they have no competing interests.

Funding
The funding sources did not play a role in the design or conduct of the study, collection, management, analysis, or interpretation of the data, or preparation, review, or approval of the manuscript. This study was funded by grants from the Ministry of Science and Technology, Taiwan (MOST 107-  Enrichment analysis was performed to identify gene functions of differential abundances in gut microbiota between ESRD patients and control subjects. Functional classi cation of the predicted metagenome content of the microbiota of ESRD using the gut metabolic module (GMM) in the discovery cohort and validation cohort. Signi cance was considered for an adjusted P<0.05 Figure 6 Explore the demographic data, comorbidities, clinical laboratory data, protein-bound uremic toxins, and alpha diversity relationship in the discovery cohort. The correlation measures between two continuous variables were tested by Spearman correlation. The correlation measures between categorical and continuous variables were tested by the Point Biserial correlation

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.