Gut Dysbiosis in Experimental Kidney Disease: A Meta-Analysis of Rodent Repository Data

Abstract

Significance Statement

Alterations in gut microbiota contribute to the pathophysiology of a diverse range of diseases, leading to suggestions that chronic uremia may cause intestinal dysbiosis that contributes to the pathophysiology of CKD. Various small, single-cohort rodent studies have supported this hypothesis. In this meta-analysis of publicly available repository data from studies of models of kidney disease in rodents, cohort variation far outweighed any effect of experimental kidney disease on the gut microbiota. No reproducible changes in animals with kidney disease were seen across all cohorts, although a few trends observed in most experiments may be attributable to kidney disease. The findings suggest that rodent studies do not provide evidence for the existence of “uremic dysbiosis” and that single-cohort studies are unsuitable for producing generalizable results in microbiome research.

Background

Rodent studies have popularized the notion that uremia may induce pathological changes in the gut microbiota that contribute to kidney disease progression. Although single-cohort rodent studies have yielded insights into host-microbiota relationships in various disease processes, their relevance is limited by cohort and other effects. We previously reported finding metabolomic evidence that batch-to-batch variations in the microbiome of experimental animals are significant confounders in an experimental study.

Methods

To attempt to identify common microbial signatures that transcend batch variability and that may be attributed to the effect of kidney disease, we downloaded all data describing the molecular characterization of the gut microbiota in rodents with and without experimental kidney disease from two online repositories comprising 127 rodents across ten experimental cohorts. We reanalyzed these data using the DADA2 and Phyloseq packages in R, a statistical computing and graphics system, and analyzed data both in a combined dataset of all samples and at the level of individual experimental cohorts.

Results

Cohort effects accounted for 69% of total sample variance (P<0.001), substantially outweighing the effect of kidney disease (1.9% of variance, P=0.026). We found no universal trends in microbial population dynamics in animals with kidney disease, but observed some differences (increased alpha diversity, a measure of within-sample bacterial diversity; relative decreases in Lachnospiraceae and Lactobacillus; and increases in some Clostridia and opportunistic taxa) in many cohorts that might represent effects of kidney disease on the gut microbiota.

Conclusions

These findings suggest that current evidence that kidney disease causes reproducible patterns of dysbiosis is inadequate. We advocate meta-analysis of repository data as a way of identifying broad themes that transcend experimental variation.

Introduction

It has been widely suggested that alterations in the composition and function of the gut microbiota may play an important role in the development of CKD.¹ One potential mechanism involves kidney disease altering the gut microenvironment and exerting selective pressures on bacterial populations, causing structural or functional alterations to the resident microbiota which have been termed “uremic dysbiosis.”^2–4 It has been suggested that abnormal gut microbiota may drive progression of CKD and its complications through increased generation of uremic toxins, reduced production of beneficial metabolites such as short-chain fatty acids, or disruption of the mucosal barrier of the gut resulting in chronic inflammation.

There is little consensus, however, as to exactly which microbial changes are caused by kidney disease. In humans, case-control studies have identified changes in host microbiota associated with CKD,^5–10 but confounding factors including dietary restrictions and use of gut-active or antimicrobial medications make the effects of kidney disease itself hard to define.

Several animal studies have described relative changes in the abundances of various microbial taxa in experimental kidney disease. Most notably, Vaziri et al.¹¹ demonstrated that 175 operational taxonomic units were differentially abundant in the gut microbiota of rats after either subtotal nephrectomy or sham surgery, concluding that kidney disease profoundly affects the gut microbiota. However, other animal studies have yielded contrasting results,^12–14 and because each of these animal studies used a single, small cohort of animals, the generalizability of the findings they describe is limited. We have previously demonstrated the extent of variability in the intestinal microbiota between two batches of animals from the same supplier and the ability of such batch differences to influence the metabolomic phenotype of host rats.¹⁵

In this meta-analysis, we reanalyze publicly available datasets describing the gut microbiota of animals with or without experimental kidney disease from two online repositories, comprising a total of 127 rodents across ten experimental cohorts, to attempt to find common microbial signatures that transcend batch variability and may be confidently attributed to the effect of kidney disease.

Methods

Selection of Studies

We searched the Sequencing Reads Archive (SRA) operated by the National Center for Biotechnology Information (NCBI) for relevant studies, using the search term (uremia OR uraemia OR kidney OR renal) AND (microbiome OR microbiota) AND (rodent OR rat OR mouse OR mice), on May 24, 2021. This search returned gut microbiota data from 412 experimental subjects across 14 studies, which were assessed for suitability for inclusion using the Run Selector facility. Eligibility criteria were use of rodent subjects, use of experimental techniques to induce chronic (>2 weeks) uraemia, and use of non–culture-dependent, DNA-based tools to assess the gut microbiota. Exclusion criteria included use of experimental interventions other than the induction of kidney disease; however, in some studies using a four-group design (e.g., control, control plus intervention, kidney disease, and kidney disease plus intervention), data from animals in the nonintervention control and kidney disease groups were included.

Eight studies were excluded: three because there was no induction of kidney disease (NCBI BioProject IDs PRJNA576633, PRJNA596575, and PRJNA325943), two which used RNA rather than DNA sequencing (PRJNA631843 and PRJNA492322), one which used an acute kidney injury rather than CKD model (PRJDB6225), one which studied kidney tissue rather than gut microbiota samples (PRJEB27588), and one which included only human samples (PRJEB11419). Seven of these excluded studies have subsequently been published.^16–22

The remaining six studies were included,^{12–15,23,24} including our own study (Randall 2019) which comprised data from two cohorts of animals. Each of these studies has been published in a peer-reviewed journal.

One cohort (Al-Asmakh 2020) included sequencing samples from the ileum, cecum, and colon for each animal; we elected to include only cecal samples in this analysis to match most samples from other rat cohorts.

We also included data from two further cohorts of our own (Randall 2021a and Randall 2021b) which are also publicly available through the SRA with the publication of this article. These cohorts have not previously been published in journals; the experimental conditions of these animals and subsequent sample analysis and DNA sequencing are described below.

Finally, phylochip microbiota data from the older, Greengenes repository were obtained for a final study, Vaziri 2013,¹¹ which was the first major study to claim an effect of kidney disease on the gut microbiota. All other phylochip datasets also in the Greengenes repository were manually screened for eligibility using the criteria above, but none were suitable.

Data Processing

Datasets downloaded from the SRA were converted into fastq format using the fastq-dump software from the SRA toolkit. Raw sequences were analyzed in R version 3.6.1, using the DADA2 pipeline (version 1.4),²⁵ with each dataset preprocessed separately because of differences in primer pairs and sequencing quality, with filtering and trim parameters being optimized for each dataset. One dataset (Al-Asmakh 2020) used widely separated primer pairs (337F/805R) which meant that after adjusting for quality, only a very small proportion of reads could be successfully merged, and so for this dataset the decision was made to include only forward reads to avoid bias. Two datasets (Mishima 2015 and Kikuchi 2017) used 454 pyrosequencing instead of Illumina paired-end sequencing, and so only longer, forward reads were available for these datasets.

Amplicon sequencing variants (ASVs) were aligned against Silva v138²⁶ to assign taxonomy. Raw abundance data of ASVs were used with taxonomic assignments and sample metadata to create phyloseq objects for each cohort.²⁷ These phyloseq objects were retained for analysis within each dataset at the level of individual ASVs but then agglomerated at family level and merged to allow analysis of the whole dataset as described below.

Phylochip data for the Vaziri 2013 dataset were substantially different in nature from the sequencing data of all the other datasets, partly because of the nature of the data acquisition (consisting of fluoroscopic intensity scores for each of several thousand probes on the chip, rather than simply those sequences present in the sample), and partly because the taxonomic identities attributed to the different 25-mer probes on the phylochip are incommensurable with the modern Silva taxonomy. Thus, otu table, taxonomy, and meta-data were combined for this dataset to allow it to be individually analyzed in phyloseq in parallel with the other datasets, but this dataset was not agglomerated and merged into the whole dataset object for combined analysis.

Quantification and Statistical Analysis

Combined Analysis of Whole Dataset

A combined dataset was constructed to permit comparison between microbial communities from all samples (excluding the Vaziri 2013 dataset), irrespective of the sequencing methodologies and primer pairs used.

To allow this, taxa from the individual cohort phyloseq objects were agglomerated to the family level (the lowest taxonomic level at which all sequencing variants received a confident taxonomic identity), using the tax_glom function in phyloseq (version 1.36.0). Taxa were manually renamed across datasets to allow comparison of like with like between cohorts; then, a combined taxonomy, meta-data, and ASV table were used to construct a phyloseq object incorporating all samples.

These data were rendered compositional using centered log-ratio transformation through the transform function in the R package microbiome (version 1.14.0),²⁸ and redundancy analysis (RDA) was performed using the ordinate function in phyloseq which was plotted using the plot_ordination function. Scores and loadings were extracted from the RDA model and used to calculate spatial means and the vector between control and uremic samples within each cohort on the combined RDA axes. The ADONIS function in R package vegan (version 2.5.7)²⁹ was used for permutational analysis of variance (PerMANOVA) calculations.

Separate Analysis of Individual Datasets

Each cohort was then analyzed independently at the level of individual ASVs, without agglomeration at higher taxonomic levels. RDA and PerMANOVA were performed using the same methods as for the combined dataset. In addition, alpha diversity analyses were performed on log-ratio transformed datasets using the estimate_richness function in phyloseq, and beta dispersion was calculated for control and uremic groups using the betadisper function in vegan.⁷² Abundance data from the combined phyloseq object were aggregated to the phylum level and rendered compositional before being used to generate the bar charts demonstrating compositional community abundance.

To reflect the composition nature of microbiota datasets and to allow for multiple hypothesis correction, testing for ASVs displaying differential abundances according to kidney disease was performed for all cohorts using the analysis of the composition of microbiomes (ANCOM2) statistical framework.³⁰ Code for ANCOM2 (version 1) was obtained from GitHub (https://github.com/FrederickHuangLin/ANCOM, accessed August 26, 2019) and used according to default parameters. ANCOM analysis was performed using data agglomerated at family, order, class, and phylum levels to pick out differences between control and uremic samples at each of these levels. For the data presented in Supplemental Table 4, only taxa detected at a cutoff of 0.7 were treated as significant, and at each level, the differentially abundant taxa were listed in descending order according to their W score. On this table, to allow a crude comparison of the significance of association, is the two-sample t test; in some cases, this is higher than the set α of 0.05, but these all actually had an adjusted significance of <0.05 after multiple hypothesis correction. A simple ratio between mean abundance in kidney disease animals and mean abundance in controls is presented to show whether kidney disease animals had increased or decreased abundance relative to controls.

Experimental Method for the Two Previously Unpublished Datasets

This article includes data from two experimental cohorts (Randall 2021a and Randall 2021b) that were not previously published in a peer-reviewed format. Details of these animal experiments are provided here.

All animal experiments were conducted in accordance with the United Kingdom Home Office Animals (Scientific Procedures) Act 1986, with local ethical committee approval. All animal work was performed in the Biological Services Units of Queen Mary University of London at Charterhouse Square and complied fully with all relevant animal welfare guidance and legislation (United Kingdom Home Office Project License Number PPL 70/8350 and P73DE7999).

Randall 2021a Cohort

This cohort consisted of two experimental groups from a larger study which had been designed to investigate the effects of lactulose on the gut microbiota, using a 2×2 experimental design (control with lactulose, control without lactulose, uremic with lactulose, and uremic without lactulose). Only samples from the nonlactulose groups were included in this analysis. In the whole experiment, 27 male wild-type outbred Wister IGS rats were obtained at age 7 weeks from Charles Rivers (Kent, United Kingdom). During a week of acclimatization, rats were swapped between cages each day for a week to homogenize resident microbiota. Seventeen underwent subtotal nephrectomy (SNx), and ten underwent sham procedures. Subtotal nephrectomy involved exteriorization of the left kidney with decapsulation and removal of the upper and lower poles and subsequent replacement of the middle pole only, followed by total right nephrectomy 2 weeks later. Sham procedures involved exteriorization, decapsulation, and replacement of the left kidney, followed by the same procedure on the right kidney 2 weeks later.

Four weeks after the completion of surgery, lactulose was administered mixed into drinking water to eight SNx animals and six controls, with the remaining animals in each group (nine SNx, six sham) continuing to receive tap water. Only samples from these latter groups were included in the meta-analysis. All animals received free access to RM1 standard rodent diet (SDS dietary services, Essex, United Kingdom) and water and were housed under standard 12-hour light-dark cycles in individually ventilated cages. There were up to four rats per cage, and the animals housed according to surgical procedure, with no cohousing between batches.

In weeks 5–8 postsurgery, rats underwent individual housing in metabolism cages weekly to allow the collection of a 24-hour urinary specimen which was frozen at −80°C until the time of analysis. Rats were killed by using lethal injection of sodium thiopentone (LINK Pharmaceuticals, Horsham, United Kingdom), and cecal fluid was expressed, stored in foil, and snap-frozen in liquid nitrogen and then at −80°C until the time of analysis. Blood samples were taken by cardiac puncture, and after centrifugation, the serum was frozen at −80°C until the time of analysis. Data describing the housing, weights, and serum biochemistry are presented in Supplemental Table 5.

Randall 2021b Cohort

Twenty male wild-type C57/BL6 mice were obtained from Charles Rivers at age 7 weeks. After a week-long period of acclimatization, ten animals were placed on an intervention diet (RM1 with 0.15% adenine as published previously by our group³¹), whereas ten remained on standard RM1 diet (both diets from SDS dietary services, Essex, United Kingdom). All animals received free access to food and water and were housed under standard 12-hour light-dark cycles, with five animals in each individually ventilated cage. Mice were weighed weekly and housed individually in metabolism cages every 4 weeks to allow the collection of a 24-hour urinary and fecal specimen. All mice were sacrificed 18 weeks after the start of the experimental protocol (at age 26 weeks), by using lethal injection of sodium thiopentone (LINK Pharmaceuticals, Horsham, United Kingdom). Cecal fluid was stored at −80°C pending DNA extraction. Data describing the housing, weights, and urine volumes are presented in Supplemental Table 6.

DNA Extraction and Next-Generation Sequencing

DNA from cecal fluid samples from both cohorts was extracted using the PowerSoil kit from Qiagen, according to manufacturer's instructions. PCR was performed in house using barcoded 27F/338R primer pairs, targeting the V1/V2 hypervariable region of the 16S rRNA gene. PCR was performed in a sterile 96-well plate using Phusion Green Hot Start II High Fidelity PCR Master Mix (ThermoFisher Scientific), using an initial denaturation step for 5 minutes at 98°C followed by 25 cycles of 98°C for 10 seconds, 53°C for 30 seconds, 72°C for 45 seconds, and a final extension of 72°C for 10 mintues. Normalization of DNA concentrations was performed using SequalPrep Normalization Plates (ThermoFisher), and DNA was quantified using a Qubit 4 Fluorometer (also ThermoFisher). Pooled samples were then sent for next-generation sequencing at the DNA Sequencing Facility, Department of Biochemistry, University of Cambridge. Samples with <4500 reads were excluded from further analysis, and all remaining fastq files were uploaded to the NCBI SRA database. BioProject identifiers for the Randall 2021a and Randall 2021b cohorts are presented in Table 1.

Table 1.

Protocols for animal cohorts and techniques used for molecular characterization of gut microbiota in the datasets included in this study

Cohort	NCBI BioProject ID	Host Organism	Number (Control/Uremic)	Method for Induction of Uremia	Proportional Increase in Serum Creatinine in Renal Failure Versus Control	Proportional Increase in Serum Urea in Renal Failure Versus Control	Age at Time of Sacrifice (Wk)	Sample Type	Molecular Study Method	16S Region Studied	Sequencing Depth (Mean Reads/Sample)	Assigned ASVs in Dataset
Al-Asmakh 202024	PRJNA662560	Male Wistar rat	12 (6/6)	0.75% adenine feed	4.97	9.16	14	Cecal fluida	Illumina sequencing	V3/V4	182,389	2453
Kikuchi 201714	PRJDB3917	Male Sprague-Dawley rat	13 (6/7)	5/6 nephrectomy	2.72	Not reported	42	Feces	454 pyrosequencing	V1/V2	3000	550
Kikuchi 201923	PRJDB6437	C57/BL6 mouse, sex not stated	10 (5/5)	0.2% adenine feed	4.33	Not reported	16	Feces	Illumina sequencing	V1/V2	34,366	369
Mishima 201512	PRJDB2888	Male C57/BL6 mouse	12 (6/6)	0.2% adenine feed	Not reported	4.1	15	Feces	454 pyrosequencing	V1/V2	8297	1022
Nanto-Hara 202013	PRJDB6792	Male C57/BL6 mouse	15 (8/7)	0.2% adenine feed	5.56	3.2	16	Feces	Illumina sequencing	V1/V2	30,890	631
Randall 2019a15	PRJNA525754	Male Wistar rat	14 (6/8)	5/6 nephrectomy	2.73	2.36	18	Cecal fluid	Illumina sequencing	V3	175,607	1331
Randall 2019b15	PRJNA525754	Male Wistar rat	10 (4/6)	5/6 nephrectomy	2.47	2.15	18	Cecal fluid	Illumina sequencing	V3	193,520	1228
Randall 2021a	PRJNA692608	Male Wistar rat	12 (6/6)	5/6 nephrectomy	2.96	3.01	18	Cecal fluid	Illumina sequencing	V1/V2	12,481	712
Randall 2021b	PRJNA692608	Male C57/BL6 mouse	17 (9/8)	0.15% adenine feed	Not measured	Not measured	26	Cecal fluid	Illumina sequencing	V1/V2	19,393	983
Vaziri 201311	NA	Male Sprague-Dawley rat	11 (5/6)	5/6 nephrectomy	1.87	Not reported	16	Feces	Phylochipb	All	NA	NA

ASV, amplicon sequencing variant; NCBI, National Center for Biotechnology Information.

^aThe publicly available phylochip data from the Vaziri 2013 dataset consist of mean fluouroscopic intensity data from 4,522 probes each consisting of a 25 bp DNA strand against a portion of the 16S rRNA gene unique to one bacterial taxon. Each of these probes was treated as a separate ASV for the purposes of phyloseq analysis.

^bThe Al-Asmakh 2020 dataset included paired samples from the ileum, cecum, and distal colon of each animal; we included only the cecal fluid samples to match the majority of other rat datasets.

Results

Published Datasets Exhibit Significant Variability in Experimental Technique

A search of the NCBI Sequence Read Archive³² and the Greengenes repository³³ was conducted to discover publicly available data describing the bacterial composition of the gut microbiota in experimental kidney disease (Figure 1). Data were included from seven published studies, including one specifically designed to investigate the effect of kidney disease on the gut microbiota,¹¹ three designed to investigate bacterial production of uremic toxins,^14,15,23 and three investigating gut-acting medications for the improvement of the uremic syndrome (only samples from nonintervention animals were included).^12,13,24 In addition, two unpublished but publicly available datasets were included from our own group; a description of the experimental treatment of cohorts is included in the methods section.

Figure 1.

Consort diagram of the meta-analysis. Searches were conducted on May 24, 2021. The NCBI BioProject IDs of excluded studies were PRJDB6225, PRJEB27588, PRJEB11419, PRJNA576633, PRJNA596575, PRJNA631843, PRJNA325943, and PRJNA492322; seven of these studies are published.^16–22 BioProject IDs of included studies are presented in Table 1. All datasets in the Greengenes repository were manually screened, but none met inclusion criteria. NCBI, National Center for Biotechnology Information; SRA, Sequencing Reads Archive.

In total, we included data from 127 animals, 73 rats across six cohorts and 54 mice across four cohorts. There were significant differences between these datasets in the animal strains used, the methods used to induce kidney disease, the age of animals at the time of sacrifice, and the sample types used as well as in the methods of DNA amplification and sequencing employed (Table 1).

Raw sequencing data were reanalyzed using the DADA2 pipeline and the Silva (v138) taxonomy database.²⁶ There was a broad but nonsignificant positive correlation between sequencing depth (mean reads per sample) and observed species richness (amplicon sequencing variants, ASVs, per sample), Spearman rank coefficient 0.55, P=0.133.

Cohort is the Key Driver of Variation across All Datasets

Sequencing data from different cohorts were agglomerated at the family level (the lowest taxonomic level at which all ASVs were assigned a clear taxonomic identity) and combined to allow broad trends in variation to be visualized across all datasets (Figure 2). RDA revealed clustering to be most significantly influenced by cohort, with the Al-Asmakh 2020 and Kikuchi 2017 cohorts completely separate from other samples and only the Randall 2019a and Randall 2019b cohorts, which comprised animals obtained a few weeks apart from the same supplier, displaying broadly overlapping ordination. We have previously shown, however, significantly differential ordination between these two cohorts when analyzed in a reduced dataset comprising only these samples.¹⁵

Figure 2.

Ordination plot of redundancy analysis of combined, log-ratio transformed data from all sequencing samples, agglomerated at the family level. Each point represents an individual sample, circles represent samples from control animals, triangles represent samples from uremic animals, and colors represent samples from different cohorts.

PerMANOVA of the log-ratio transformed datasets was used to establish how much variation could be attributed to different experimental variables; this revealed that cohort accounted for the largest amount of variation (69% of variance, P<0.001), with host species (rat versus mouse) accounting for 13.3% (P<0.001).

Other significant associations were found between clustering and primer type (V1/V2 versus V3 versus V3/V4, 23.9% of variation, P<0.001), method of inducing kidney disease (surgery versus adenine feed, 13.2% of variation, P<0.001), sequencing methodology (454 pyrosequencing versus Illumina, 9.7% of variance, P<0.001), and sample type (feces versus cecal fluid, 6.7% of variance, P<0.001), although these variables were closely associated with cohort.

Treatment effect (control versus uremic) did influence sample clustering, but to a much lesser extent (1.9% of variance, P=0.026). Scores and loadings from the RDA model were interrogated to understand the basis for this small observed effect of kidney disease, and it became clear that while control and uremic samples were not significantly separated in axes 1 and 2, when plotted on axes 3 and 4 of the RDA model, there was a significant shift between control and uremic samples in both dimensions (Supplemental Figure 1). Loadings for these axes revealed that a “uremic” deflection along both was notably associated with increased abundances of certain families from class Clostridia (including Peptostreptococcaceae, Peptococcaceae, Clostridiaceae, and Christenellaceae), Supplemental Table 1.

Significant Compositional Differences Exist between Experimental Cohorts

Each experimental dataset was agglomerated at the phylum level, and community composition was plotted for each sample to allow comparison at a high taxonomic level between cohorts (Figure 3).

Figure 3.

Proportional abundances of bacterial phyla in all experimental cohorts. Rat cohorts are on the top row and mouse cohorts below. Each vertical bar represents a sample from a single animal, grouped within cohorts with control samples on the left and uremic samples on the right. Because of the nature of phylochip analysis, the Vaziri 2013 cohort included data for 43 phyla, including many making negligible contributions to the overall population; thus, for the Vaziri 2013 cohort, only the 12 most abundant phyla are shown and a different legend is provided to reflect the different taxonomy used in phylochip analysis compared with the other datasets. The relatively high abundances of minor phyla in the Vaziri 2013 samples may reflect increased prominence of these probes in the design of the phylochip.

Dominant phyla in all samples were Firmicutes, accounting for 76% of reads in rat samples versus 40% in mouse samples (P<0.001), and Bacteroidetes (58.7% of sequencing reads in mouse samples but only 9.9% in rat samples; P<0.001). Mouse samples were simpler than those from rats, with the contribution of these major taxa accounting for an average of 98.4% of reads in mice, but only 84% in rats (P<0.001).

The Al-Asmakh 2020 and Kikuchi 2017 datasets seemed to be outliers compared with other rat cohorts. The Al-Asmakh 2020 cohort demonstrated clear differences between control and uremic samples at the phylum level, with substantial increases in Proteobacteria and Actinobacteria in uremic animals, accounting for an average of 39.9% and 9.4% of reads, respectively, in this group. Conversely, samples in the Kikuchi 2017 cohort were very simple, with reads from the phylum Firmicutes accounting for 98.5% of reads across all samples and kidney disease having no discernible effect. In other datasets, there were no obvious high-level community differences between control and uremic samples.

Kidney Disease Increases Alpha Diversity in Rats but Not in Mice

There were no significant differences in alpha diversity between control and uremic animals across the whole dataset (Supplemental Figure 2, Supplemental Table 2). However, samples from rats were found to have higher alpha diversity than samples from mice (significantly so for observed ASVs per sample, 338 in rats versus 232 in mice, P=0.006, and the related Chao1 and ACE indices).

Among rat samples, those from uremic animals showed higher alpha diversity than those from controls across all measures, significantly so for the Shannon (4.135 in control versus 4.656 in uremic, P=0.011), Simpson (9.952 versus 9.975, P=0.01), and Inverse Simpson (40.74 versus 62.59, P=0.012) indices. Although this was chiefly driven by the highly diverse uremic samples in the Al-Asmakh 2020 cohort, a trend toward increased alpha diversity was seen universally across all measures of diversity in every rat cohort. No measures of alpha diversity showed significant differences between control and uremic samples in mice.

Beta dispersion was assessed to test the hypothesis that kidney disease increases the heterogeneity of gut communities, but no reproducible differences were seen between groups, Supplemental Table 3.

Samples from Control and Uremic Animals Cluster Apart in Most Cohorts

Plots of RDA ordination for log-ratio transformed datasets at the level of individual ASVs were constructed for all cohorts (Figure 4). Samples from uremic animals clustered separately from those from control animals in most cohorts, and this was confirmed using PerMANOVA which quantified significant between-group differences associated with kidney disease in seven of the ten cohorts. Nevertheless, significantly divergent clustering between similarly treated animals (seemingly attributable to cage effects) was seen in several cohorts, especially the Mishima 2015, Nanto-Hara 2020, Randall 2019a, and Randall 2021b cohorts.

Figure 4.

Ordination plots of compositionally transformed data for all cohorts at the level of individual amplicon sequencing variants. Each point represents a sample from an individual animal, colored according to treatment (control versus uremic). R² and P values from permutational analysis of variance analysis of the same data are superimposed on each plot. Divergent clusters between similarly treated animals seen in the Mishima 2015, Nanto-Hara 2020, Randall 2019a, and Randall 2021b are attributed to caging effects.

Kidney Disease is Associated with Reduced Abundances of Health-Associated Taxa and an Increased Abundance of Opportunistic Species in Some Cohorts

The ANCOM2 methodology was used to assess differential abundances of bacterial taxa between control and uremic samples in all cohorts, at each taxonomic level between individual ASVs and phyla (Supplemental Table 4).

In two cohorts (Al-Asmakh 2020 and Vaziri 2013), a classically dysbiotic picture emerged with reductions in health-associated taxa (prominently genus Lactobacillus, also Bacteroides and Akkermansia), an increase in Gram-negative bacteria (including families from the class Gammaproteobacteria, such as Enterobacteriaceae and Pseudomonadaceae) and an increase in families from the high GC content, Gram-positive phylum Actinobacteria (including Corynebacteriaceae and Bifidobacteriaceae).

However, these changes were not seen universally, and in some cohorts—including the Randall 2019a and Kikuchi 2017 cohorts which did not show significantly differential clustering in ordination plots and PerMANOVA—there were no differentially abundant taxa between control and uremic groups at any taxonomic level. In other cohorts, conflicting results were seen, such as in the two mouse cohorts Mishima 2015 and Nanto-Hara 2020, where some Lactobacillus species were seen to increase in abundance in samples from uremic animals, an opposite trend from that seen in the Al-Asmakh 2020 and Vaziri 2013 cohorts.

To assess whether similar trends were seen across multiple groups, the mean relative abundance of all families was compared between control and uremic animals within each sequencing cohort (Figure 5). There were no families or organisms for which kidney disease caused uniform changes in relative abundance across all cohorts. However, two highly prevalent taxa showed a trend to reduced abundances in uremic animals (Lachnospiraceae, the most proportionally abundant family overall, present at lower relative abundances in uremic animals in 7/9 cohorts, and Lactobacillaceae, the third most proportionally abundant family overall, present at lower relative abundances in 7/8 cohorts). Other highly abundant families (including Oscillospiraceae and Ruminococcaceae) did not show anything approaching a uniform association with kidney disease. Several lower-abundance taxa showed relatively uniform increases in uremic animals, including Clostridiaceae, increased in 8/9 cohorts; Erysipeltrichaceae, increased in 7/9 cohorts; Peptostreptococcaceae, increased in 6/7 cohorts; Tannerellaceae, increased in 6/7 cohorts; and Eggerthellaceae, increased in 7/9 cohorts. In most cohorts, the difference in mean relative abundances between control and uremic samples was small, and there were no families in which the mean relative abundances were significantly different between control and uremic animals.

Figure 5.

Relative abundances of the 15 most abundant families of bacteria in the overall dataset in uremic versus control animals within different experimental cohorts. Graphs are presented in order of decreasing overall abundance, with Lachnospiraceae having the highest overall relative abundance, Muribaculaceae the second highest, Lactobacillaceae the third highest, and so on. Each point represents the mean proportional contribution of a particular family of bacteria within control or uremic animals in a given cohort; when a family is represented by fewer dots than the number of studies, this is because that family was not detected in all studies; bars link control and uremic animals within a given cohort so that upward slopes suggest that the family forms a higher proportion of the microbiota in uremic animals within that cohort and downward slopes indicate that the family forms a higher proportion of the microbiota in control animals in that cohort. None of the differences between the average proportional abundances between control and uremic animals were significant by two-sample unequal variances t test, and the average of all slopes on a given graph was never significantly different from zero by the one-sample t test.

Discussion

This meta-analysis demonstrates that between-cohort differences eclipse the effect of experimental kidney disease in explaining compositional variation between samples of rodent gut microbiota. The limitations of animal models of kidney disease means that it is impossible to say whether the greater degree of “uremia” seen in human subjects on dialysis may exert more selective pressure than is seen in the animal models described in the animal studies reanalyzed here. Similarly, limitations of 16S rRNA gene amplicon sequencing (which describes only the composition and not the activity of the microbiota) mean it is impossible to determine on the basis of the data described here whether a similar population of gut microbes may be exhibiting a different metabolic phenotype in the gut environment of a host organism with kidney disease than they would in the gut environment of a metabolically healthy host. Nevertheless, these findings do fundamentally question the idea that kidney disease itself, in the absence of drug, dietary, or behavioral interventions commonly used in humans with CKD, causes distinct and reproducible changes in the composition of the gut microbiota.

Although it is possible that differences in experimental technique and sequencing methodologies may account for some of the cohort differences described in this meta-analysis, we suggest that the majority is likely to be accounted for by baseline differences in the gut microbiota of animals used in different studies. These results are consistent with previous reports showing batch variation to be a major confounder in microbiome research³⁴; microbial variation being demonstrated on the basis of animal vendor,^15,35 differences in husbandry,^36–41 animal strain, sex, and even diurnal rhythm.^42–44

Significant heterogeneity between experimental cohorts makes it difficult to describe with confidence any reproducible pattern of “uremic dysbiosis.” For example, whereas in the Al-Asmakh 2020 and Vaziri 2013 cohorts large and classically “dysbiotic” effects were associated with induction of kidney disease, in the Kikuchi 2017 and Randall 2019a datasets (which used the same intervention—five sixths nephrectomy, in the same host species, rats), there was no discernable difference at all between the microbiota of experimental groups. These data suggest that microbial community changes may vary widely between groups of experimental animals in response to a similar biological insult and pose significant questions about the generalizability of any single-cohort rodent experiments where the gut microbiota might play a significant physiological or pathological role. Publication bias risks obscuring genuine heterogeneity of response by highlighting experiments reporting more striking results.

A drawback of most of the studies included in this analysis is that caging was defined by treatment class (i.e. control animals were housed with other controls and uremic animals with other uremics), presumably because of the practicalities of administering a modified feed, or allowing for different recovery times after sham surgery or subtotal nephrectomy. The consequence of separate housing is that it becomes impossible to distinguish the effects of kidney disease from the diverging effects that would be seen in microbiota between any two groups of animals housed in separate cages. Profound cage effects were seen between different groups of similarly treated animals in a number of the cohorts analyzed (Mishima 2015, Nanto-Hara 2020, Randall 2019a, and Randall 2021b), and in several of the cohorts, there were ASVs present in high abundance in one experimental group but totally absent in the other group; we feel this situation is highly likely to reflect cage effects rather than the biological effect of kidney disease. Interestingly, the fewest differences between control and uremic groups were seen in cohorts where attempts had been made to reduce caging effects, by moving animals between cages before the initiation of surgery to homogenize microbial populations (Randall 2021a), by housing control and uremic animals together after postoperative recovery (Randall 2019a and Randall 2019b), or by housing all animals in individual cages (Kikuchi 2017).

We conclude that single-cohort studies comparing control and intervention animals are an unsuitable tool for investigating the role of the gut microbiota in health and disease. Future microbiome experiments should assess baseline gut microbiota in all animals before experimental interventions are undertaken, allowing comparison of longitudinal changes in bacterial taxa and allowing every animal to act as its own control. Studies should ideally use several batches of animals, take measures to reduce caging effects, and—if describing “dysbiosis”—should demonstrate trends that transcend batch and cage variability, as has been advocated elsewhere.^45,46 We furthermore suggest that meta-analysis of published data from different studies, such as this one, can be used to discriminate batch effects from genuine biological trends.

In the specific context of experimental kidney disease, no trends in microbial populations could be demonstrated that were present universally in uremic animals across all animal cohorts. However, the following features were each seen in many cohorts, implying that the common factor of kidney disease may be causative: increased alpha diversity (in samples from rats); an increase in lower-abundance taxa including non–Lachnospiraceae Clostridia, Gammaproteobacteria and Actinobacteria; and a decrease in core, health-associated taxa, particularly Lactobacillus and Lachnospiraceae. It is possible that the effect of kidney disease is to disrupt dominant members of the gut microbiota and create an environment where less prevalent, opportunist organisms, varying at species level between cohorts, can expand in number. However, it must be noted that some animal cohorts (e.g. Kikuchi 2017 and Randall 2019a) did not display even these broad trends. Reassuringly, the effects we describe in our reanalysis of others' data are broadly the same as those reported by the authors in the original descriptions of their research.

These data also present a broadly similar picture to results in published human studies. We are aware of 20 studies describing the molecular characterization of the gut microbiota in kidney disease; findings from these studies are summarized in Table 2.^{5–11,47–59} There are many differences between these studies in the patient populations included sequencing techniques used and statistical analysis used, and many of these studies do not meet quite basic requirements for modern microbiological work, as discussed in a recent review article. Similarly, humans living in different parts of the world, consuming radically different diets and receiving different enteral medications, are likely to exhibit far greater heterogeneity in gut microbiota than experimental animals consuming standardized animal feeds.⁶⁰ While most of these studies compare patients with end-stage kidney disease on dialysis with healthy controls, several consider those with much milder renal impairment.^5,10,52,54 Similar to the rodent data reviewed in this article, there is a significant degree of heterogeneity in the results described by the authors, alongside some broader themes that may reflect the effect of kidney disease, and in some cases, match trends in the animal data presented here. Measures of alpha diversity were reported in 11/20 studies: in six diversity, it was reduced in kidney disease^{5,8,10,47,55,57}; in three, it was not significantly different^6,9,11; and in two, it was increased.^50,52 Eleven studies reported the results of ordination between samples: in five, there was clear separation between control and kidney disease samples^6,8–10,47; in two samples, there was partial separation^49,57; and in five, there was no separate clustering,^{5,11,52,54,61} although two of these studies noted that samples from kidney disease subjects showed a greater degree of beta dispersion than those from controls.^11,54 At the level of individual microbial taxa, changes in abundance between subjects with kidney disease and controls were especially notable for Enterobacteriaceae (where the whole family, or subtaxa within it, were increased in subjects with kidney disease in 13 studies,^{5,7,8,11,47,49,50,52,53,56–58} but decreased in five^8,9,47,50,56), Clostridia (increased in subjects with kidney disease in four studies^7,8,10,49 but decreased in one⁵¹), Bifidobacteria (decreased in subjects with kidney disease in three studies^5,7,56 but increased in one⁹), and a group of well-recognized, health-associated producers of short-chain fatty acids (genera Roseburia, Faecalibacterium, Romboustia, Blautia, and Eubacterium decreased in subjects with kidney disease in nine studies,^{6,8–10,47–49,51,57} but increased in two^5,47).

Table 2.

Summary of studies describing genetic analysis of human microbiota in subjects with and without CKD

Study	Subjects	Degree of Uremia	Alpha Diversity in Uremic Subjects	Ordination	Taxa Increased in Uremic Subjects	Taxa Decreased in Uremic Subjects	Authors' Additional Comments, Uremic Subjects
De Angelis5 2014, Italy	Nonprogressive IgA nephropathy, n=16; progressive IgA nephropathy, n=16; healthy controls, n=16	Starting eGFR (MDRD formula): nonprogressors, 76 ml/min; progressors, 30 ml/min; health controls, 96 ml/min	Decreased	No clear separation	f.Streptococcaceae, f.Eubacteriaceae, f.Alcaligenaceae, f.Enterobacteraeceae, f.Coriobacteraceae	f.Lactobacillaceae, f.Bacteroidaceae, f.Prevotellaceae, f.Bifidobacteraceae	Lower concentrations of culturable bacteria; fewer active Bacteroidetes and more active Firmicutes comparing RNA with DNA sequencing. Altered metabolomic phenotype in patients with progressive disease.
Al-Obaide7 2017, USA	Advanced diabetic CKD, n=20; healthy controls, n=20	Mean eGFR (CKD-EPI formula): patients with CKD, 16.5 ml/min; healthy controls, not stated	Not stated	Not stated	g.Clostridium g.Escherichia g.Enterobacter g.Acinetobacter g.Proteus g.Lactobacillus	g.Bifidobacteria	Changes in gut microbiome linked to increases in serum concentration of uremic toxins and indicators of increase gut permeability.
Xu8 2017, China	Advanced CKD, n=32; healthy controls, n=32	Classified using measured GFR: In the advanced CKD group, all but one had GFR <15 ml/min; health controls all had GFR >90 ml/min	Decreased	Striking separation of healthy and uremic samples on PCoA ordination plots	f.Enterobacteriaceae, f.Corynebacteriaceae, g.Clostridium, g.Adlercreuzia	g.Prevotella, g.Coprococcus, g.Megamonas, g.Sutterella, g.Enterobacter, g.Acidaminococcus, g.Dorea, g.Roseburia, g.Streptococcus, g.Streptococcaceae, g.Lactobacillales	Microbiome alterations reported to explain increased serum trimethylamine oxide (TMAO) concentrations in serum and to increase mouse serum TMAO concentrations after microbial transfer experiments.
Li9 2019a, China	Predialysis CKD5, n=24; hemodialysis, n=29; healthy control, n=69	Mean eGFR (CKD-EPI formula): predialysis CKD, 5.3 3 ml/min; healthy controls, 126.1 ml/min	Decreased	Striking separation of healthy and uremic samples on PCoA ordination plots	p.Firmicutes, g.Lachnoclostridium, g.Neisseria, g.Streptococcus, g.Subdoligranulum, g.Bifidobacteria	p.Bacteroidetes, g.Bacteroides, g.Faecalibacterium, g.Prevotella, g.Eshchericia_Shigella, g.Enterobacter	Alterations in gut bacteria reported to correlate with serum concentrations of uremic toxins.
Li10 2019b, China	CKD, n=50; healthy control, n=22	Mean eGFR (method not stated): CKD, 22.4 ml/min; healthy controls, 97.1 ml/min	Nonsignificantly reduced	Striking separation of healthy and uremic samples on PCoA ordination	p.Verrucomicrobiota, g.Lactobacillus, g.Clostridium IV, g.Alloprevotella, g.Paraprevotella, g.Clostridium sensu stricto, g.Desulfovibrio	p.Actinobacteria, g.Akkermansia, g.Parasutterella	Suggestion that alterations in gut bacterial may influence host inflammatory phenotype.
Stadlbauer47 2017, Austria	End-stage CKD, n=30; healthy control, n=21	Of the patients with CKD, all were receiving dialysis (15 hemodialysis and 15 peritoneal dialysis). Healthy controls had a mean GFR (method not stated) of 77.6 ml/min	Reduced	Striking separation of all three groups (hemodialysis, peritoneal dialysis, healthy controls)—e.g. groups completely separate from the other	c.Bacilli, c.Epsilonproteobacteria c.Gammaproteobacteria, c.Alphaproteobacteria, g.Thalassospira, g.Eisenbergiella, g.Ruminococcaceae, g.Coprococcus g.Escherichia-Shigella, g.Streptococcus, g.Enterobacter, g.Blautia	c. Erysipelotrichia, g.Coprococcus, g.Holdemanella, g.Asteroleplasma, g.Paraprevotella, g.Prevotella, g.Romboutsia, g.Lachnospiracea UCG008, g.Pelomonas	Suggestion that alterations in gut bacterial may influence host inflammatory phenotype.
Jiang6 2017, China	End-stage predialysis CKD, n=52; healthy control, n=60	Mean eGFR (formula not stated): predialysis CKD, 6.86 ml/min; healthy controls, 98.03 ml/min	No different	Significant separation between groups	f.Bacteroidaceae, g.Bacteroides, g.Parabacteroides	g.Roseburia, g.Faecalibacterium, g.Prevotellaceae, g.Prevotella, g.Coprococcus	Reduction in total bacterial DNA in fecal samples in ESRD.
Wang50 2012, China	End-stage CKD, n=30; healthy control, n=10	Not stated except that the kidney disease patients were “end stage.”	Species count increased	Not stated	g.Klebsiella,  g.Proteus, g.Escherichia, g.Enterobacter, g.Pseudomonas		20% of ESRD patients had bacterial DNA present in blood, from species found to be overgrown in the gut, suggesting increase bacterial translocation in CKD.
Lun49 2019, China	CKD, n=49; healthy control, n=24	Not stated for either group except that 13 of the patients with CKD were receiving hemodialysis, while the others were not.	Not stated	Partial separation between groups	g.Bacteroides, g.Escherichia_Shigella, g.Parabacteroides, g.Ruminococcus_ gnavus, g.Ruminococcus torques, g.Weissella, g.Flavonifractor, g.Ruminiclostridium5, g.Sellimonas, g.Erysipelatoclostridium, g.Eggerthella, g.Clostridium_innocuum.	g.Dialister, g.Eubacterium rectale, g.Carnobacterium, g.Lachnospira, g.Subdoligranulum, g.Eubacterium_coprostanoligen, g.Coprococcus, g.Roseburia, g.RuminococcaceaeUCG_009, g.RuminococcaceaeNK4A214, g.LachnospiraceaeFCS020, g.Ruminococcus1, g.Romboutsia, g.Butyricicoccus, g.Collinsella, g.RuminococcaceaeUCG_003, g.Eubacterium_halliigroup, g.Tyzzerella3, g.LachnospiraceaeUCG_001.
Wang51 2020, China	Hemodialysis, n=233; healthy control, n=69	Of the patients with CKD, all were receiving hemodialysis. Healthy controls had a mean GFR (CKD-EPI formula) of 104 ml/min	Not stated	Ordination not described, although statistical measures suggested that ESRD strongly affected the microbiome.	s.Eggerthella lenta,  g.Flavonifractor, g.Ruminococcus g.Fusobacterium	g.Prevotella, g.Clostridium, g.Roseburia s.Faecalibacterium prausnitzii s.Eubacterium rectale	Fecal metabolomics also showed significantly different concentrations of gut-derived products including uremic toxins and short-chain fatty acids between patients and controls.
Joossen59 2018, Belgium	Longitudinal sampling of 17 patients with end-stage kidney disease; comparison with a wider population cohort of 1106 controls	Not stated	Not stated	No separate clustering of samples from patients with CKD compared with a wider cohort of population samples	Not reported. Authors report no common signature of kidney disease.	Not reported. Authors report no common signature of kidney disease.	Six taxa associated with differential production of uremic toxins.
Vaziri11 2013, USA	End-stage kidney disease, n=24; healthy control, n=10	Of the patients with CKD, all were receiving hemodialysis. Healthy controls had a mean serum creatinine of of 70.7 µmol/L	Relative richness no different between groups	Not separately clustered, but more widerly dispersed than control samples	f.Brachybacterium f.Nesterenkonia f.Catabacter f.Peptostreptococcaceae f.Catenibacterium f.Polyangiaceae f.Alteromonas f.Enterobacteriales_Enterobacteriaceae f.Halomonadaceae f.Methylococcaceae f.Moraxellaceae f.Pseudomonadaceae f.Thiothrix	None significant
Barros55 2015, United Kingdom, Spain, USA	Early kidney disease; patients from the TwinUK registry dataset, n=855, of which 7.4% had eGFR <60 ml/min	The overall cohort had a mean eGFR (CKD-EPI formula) or 83.1 ml/min; not reported for the individual groups	Not reported	Not separately clustered, but more widely dispersed than control samples	None	g.Ruminococcus g.Lachnospira g.Oscillospira	Also looked at metabolomic signatures in this cohort with mild renal impairment and bacterial associated with higher concentrations of specific toxins.
Gradisteaunu58 2019, Romania	Diabetic kidney disease, n=9; healthy controls, n=5	Not stated for either group	Not reported	Not reported	g.Turicibacter f.Enterobacteriaeceae		The authors note this was a pilot study limited by small sample size
Miao55 2018, China	Hemodialysis patients, n=21, of whom 14 were subsequently started on lanthanum carbonate; Healthy control, n=20	Of the ESRD patients, all were receiving hemodialysis. Healthy controls are reported to have had “normal renal function”	Possibly lower in ESRD; no reporting of significance tests	Not reported	No significance tests	No significance tests
Wang56 2012, Taiwan	Peritoneal dialysis patients, n=29; healthy controls, n=41	Of the ESRD patients, all were receiving peritoneal dialysis. No reporting of renal function in healthy controls.	Not reported	Not reported	s.Pseudomonas aeruginosa	g.Bifidobacteria s.Lactobacillus plantarum s.Lactobacillus paracasei s. Klebsiella pneumoniae
Guirong57 2018, China	CKD, n=100; healthy controls, n=53	Of the patients with CKD, 16 had received a kidney transplant while 84 were receiving hemodialysis of the healthy controls	Reduced	Partial separation on ordination plots	g.Bacteroides f.Enterobacteriaceae	g.Lachnospira, f.Ruminococcaceae, g.Faecalibacterium
Jiang48 2016, China	CKD, n=65; healthy controls, n=20	The patients with CKD were selected from all stages (1–5) or predialysis CKD and had a mean eGFR (formula not stated) of 55.6 ml/min; healthy controls had a mean eGFR of 105 ml/min	Not assessed	Not assessed		g.Roseburia s.Faecalibacter prausnitzii	This study did not assess the full microbiome, but rather used quantitative PCR with primers against the two species listed left and showed both to be reduced in patients with CKD.
Tao52 2019, China	Biopsy-proven diabetic nephropathy with albuminuria but preserved excretory function, n=14; type 2 diabetes without nephropathy, n=14; healthy controls, n=14; household contacts of diabetic nephropathy patients, n=14	Mean eGFR (method not stated) for each group was diabetic nephropathy, 93.3 ml/min; diabetes without nephropathy, 93.5 ml/min; healthy control, 96ml/min; household contact, 106.9 ml/min.	Increased in diabetic nephropathy compared with non-nephropathy diabetics	Relatively poorly separated	f.Coriobacteriaceae g. Escherichia-Shigella	g. Prevotella_9
Shi53 2014, China	End-stage kidney disease, n=52; healthy control, n=10	Of the ESRD group, 22 were receiving hemodialysis and the other 30 had significant renal dysfunction (serum creatinine >707 µmol/L).	Not stated	Not stated	p. Proteobacteria	p.Firmicutes p.Bacteroidetes

Many questions remain in seeking to define the relationship between kidney disease and the gut microbiota. First, what microbiological factors underlie the heterogeneity of changes seen in response to kidney disease in the host organisms? Are there features of organisms, or consortia of organisms, that make them more or less able to survive in a uremic environment? Longitudinal studies showing how different taxa fare over time as a host organism becomes uremic may be helpful in this regard (and single-cohort experiments, properly conducted, may still be useful here), as may in vitro testing of microorganisms for their urea tolerance by batch culture.

Second, might the functions of the microbiota change in kidney disease, even if structural shifts vary? For example, in the setting of nonalcoholic fatty liver disease, it has been shown that functional changes in the microbiota precede compositional changes, but nevertheless affect host phenotype.⁶² The family Lachnospiraceae (which tended to reduce in relative abundance in uremic animals) includes numerous species which are highly metabolically active and associated with production of short-chain fatty acids, which have a variety of beneficial health outcomes.^63,64 Gammaproteobacteria,^65,66 which we have shown to be variably increased in the gut microbiota of uremic animals, possess the metabolic potential for production of toxic molecules from dietary protein^67,68 and production of forms of lipopolysaccharide which have been associated with an exaggerated inflammatory response to systemic endotoxemia.^69,70 Studies using different metaomics techniques (metagenomics, metatranscriptomics, metabolomics, and metaproteomics) may answer these key mechanistic questions. For example, recent metabolomic data have suggested that fears of increased toxin generation by bacteria in the context of CKD may be wide of the mark.⁷¹

Finally, is the microbiome of uremic animals amenable to the kinds of modifications that have been demonstrated in other contexts, for instance using prebiotic or probiotic interventions? Studies looking at manipulation of the microbiota for therapeutic purposes may lead us to view the microbiome in CKD not primarily as a factor in pathology, but rather as a potential therapeutic resource.

We conclude that single-cohort, intervention/control rodent studies are not fit for purpose in describing the effect of experimental conditions on gut microbiota or on the wider host phenotype where the host:microbiota interactions might be a key pathophysiological factor in the disease process studied. We conclude further that there is no definite and reproducible effect of experimental kidney disease on the rodent gut microbiota but that trends seen in several different experiments may be caused by the effects of kidney disease. Finally, we advocate meta-analysis of repository data as a way of addressing experimental variation and identifying trends that transcend batch effect.

Supplementary Material

jasn-34-533-s001.pdf ^{(506.3KB, pdf)}

Disclosures

K. Mccafferty reports Consultancy: Oncacare; Research Funding: Astra Zeneca; Honoraria: Vifor Fresenius, Bayer, Pharmacosmos, Napp, AstraZeneca; and Speakers Bureau: AstraZeneca, Bayer. M. Curtis reports Consultancy: GSK. All remaining authors have nothing to disclose.

Funding

The direct experimental work presented here was supported by the Barts Health Diabetic Kidney Disease Center (DKC, Barts Health Grant Reference Number 577/2348). L. Hoyles was funded by UK Med-Bio (Medical Research Council Grant Number MR/L01632X/1).

Author Contributions

M. Curtis, L. Hoyles, K. McCafferty, D.W. Randall, and M.M. Yaqoob conceptualized the study; L. Hoyles and D.W. Randall were responsible for data curation; L. Hoyles, J. Kieswic, and D.W. Randall were responsible for formal analysis; D.W. Randall was responsible for investigation, project administration, and visualization; J. Kieswic and D.W. Randall were responsible for methodology; J. Kieswic was responsible for resources; L. Hoyles was responsible for funding acquisition; M. Curtis, L. Hoyles, K. McCafferty, and M.M. Yaqoob provided supervision; D.W. Randall wrote the original draft; and M. Curtis, L. Hoyles, J. Kieswic, K. McCafferty, D.W. Randall, and M.M. Yaqoob reviewed and edited the manuscript.

Data Sharing Statement

All data are available via the online SRA and Greengenes repositories using accession numbers as described in the text.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/JSN/D637.

Supplemental Figure 1. Ordination plot showing axes 3 and 4 from the redundancy analysis model of all samples.

Supplemental Figure 2. Alpha diversity between control and kidney disease animals.

Supplemental Table 1. Loadings for axes 3 and 4 in the RDA model, the basis for the scores plot in.

Supplemental Table 2. Alpha diversity.

Supplemental Table 3. Beta dispersion.

Supplemental Table 4. Bacterial taxa showing significant differences in abundance between control and uremic samples at each taxonomic level within each cohort.

Supplemental Table 5. Animal data for the Randall2021a cohort.

Supplemental Table 6. Animal data for the Randall2021b cohort.