Cefazolin shifts the kidney microbiota to promote a lithogenic environment

Mice kidneys harbor a unique microbial signature

Given pragmatic issues in establishing and assessing a kidney microbiota in humans, comprehensive assessment of the kidney microbiota was initially examined in mice. To quantify viable bacterial loads in mouse kidneys, aseptically retrieved punch biopsies of three, 10-week-old SWR/J mice housed in a clean barrier facility were plated undiluted in seven different media, incubated, and colony-forming units (CFUs) quantified. Data revealed a significant increase in bacterial density from the renal cortex to medulla, then ureter (Fig. 1a,b; Paired chi-square x² = 5.024, p = 0.023 (a) or t-test cortex:medulla- t = 3.96; p = 0.0036; cortex:ureter- t = 6.47; p = 0.00002; ureter:medulla- t = 2.17; p = 0.09 (b)). Negative controls did not produce CFUs. An RNA-targeted, fluorescence in situ hybridization (FISH) analysis of kidneys stained with a universal probe for bacteria shows the presence bacteria in the medulla and cortex (Fig. 1c) and is consistent with the distribution of bacteria assessed by culture-based means.

To evaluate the presence of a kidney microbiota using criteria established for low biomass microbial communities¹⁹ and quantify the impact of antibiotics on kidney bacteria, one-hundred sixty SWR/J mice were exposed to antibiotics, with or without recovery, in parallel with non-exposed controls (Fig. 1). Animals exhibited an antibiotic-dependent difference in water intake and urine output similar to past studies²⁰, but not food intake or body mass (Two-way ANOVA and post-hoc, Holm’s corrected, paired t-tests; t = 16.436, p < 0.001 for water, t = 10.23 p = 0.001 for urine; Fig. 2). A total of 239 urine samples and 40 kidneys were subjected to high-throughput sequencing of the V4 region of the 16S rRNA gene. There was an average of 8336 + /- 598 reads per sample, with 2821 + /- 62 mapping to host reads. Kidney and urine samples were significantly different than positive and negative controls in terms of microbial community composition (Permanova f = 3.98, p = 0.001; Fig. 3a). All kidney and urine samples surpassed the sequencing threshold to provide a statistically representative snapshot of microbial communities based on rarefaction analysis, defined as the sequencing depth that >90% of samples had a slope of <0.01 (Fig. 3b). While 5.34% of high-quality reads were removed from urine samples as Eukaryotic/host, 52% of reads were removed from kidney specimens as Eukaryotic/host (ANOVA, f = 5.73, p < 0.001; Fig. 3c).

PERMANOVA analysis, based on weighted UniFrac dissimilarity, revealed that the murine kidneys and urine microbiomes were unique (f = 97, p = 0.001; Fig. 1d). Taxa in kidneys and urine were dominated by the Pseudomonadota and Bacilliota, with relatively high numbers of Actinobacteriota and low levels of Bacteroidota (DESeq2; FDR < 0.05 for significantly different phyla; Fig. 1e), consistent with voided human urine^7,9,21,22 and in contrast to predominant taxa in murine stool²³. We detected significantly higher species richness/evenness in urine compared to kidneys (paired t-test; PD_whole_tree t = 3.24, p = 0.19; Margalef t = 914, p = 0.0009; Shannon t = 2.7 × 10¹¹, p = 6.9 × 10^-12; equitability t = 4.5 × 10¹⁷, p = 5.6 × 10^-18; Fig. 1f). Acinetobacter, a common urobiome taxon²², was most enriched in urine compared to kidneys (DESeq2; FDR < 0.05 for significantly different taxa; Fig. 1g, Supplementary data S1). Pseudomonas was most enriched in kidneys, compared to urine (Fig. 1g), a taxon that is resident in other organs, such as lung tissue²⁴.

Importantly, vesicoureteral reflux (VUR) assays²⁵ were conducted in 12-weeks male SWR/J to assess this renal defect phenotype in this breed used for the experiment. (Fig. 1h). Five animals were tested under the protocol described in the methods, and not any reflux was recorded from 30 to 150 cm², observing exit of dye through the urethra at the highest pressure.

Long-term use of cefazolin impacts the LUT of mice, with recovery

Long-term use of cefazolin had a significant impact on bacterial composition and richness of the urine, even with a recovery period (2-way Anova f = 12.6, p = 0.0007; Fig. 2a; 2-way Anova f = 8.5, p = 0.021; Fig. 2b; Permanova f = 3.94, p = 0.018; Fig. 2c). Short-term antibiotic exposure did not impact the composition or richness (2-way Anova f = 0.014, p = 0.909; Fig. 4a; 2-way Anova f = 4.064, p = 0.05; Fig. 4b; Permanova f = 0.98, p = 0.32; Fig. 4c).

To identify taxa positively or negatively impacted by antibiotic treatment in urine samples, we conducted a differential abundance analysis of amplicon sequence variants (ASVs) in animals that received a long-term course of antibiotics from baseline to the last day of exposure, compared to non-exposed controls (DESeq2, FDR < 0.05 Fig. 2d, Supplementary data S2). Genera of known uropathogens Pseudomonas (8 ASV’s), Proteus (6 ASV’s), and Enterococcus (10 ASV’s) significantly increased in abundance with antibiotic use (p < 0.01), indicative of antibiotic resistance, whereas Planococcaceae (7 ASV’s), Bacilliales (32 ASV’s), Staphylococcus (6 ASV’s), and Lactococcus (9 ASV’s) were most negatively impacted by intervention (FDR < 0.05). With the exception of Staphylococcus, infections by these taxa are rare and instead may represent the commensal urinary flora. Acinetobacter (19 ASV’s) most benefitted from time in the trial without antibiotic use (FDR < 0.05).

Previous studies suggest that bacteria from the Enterobacteriaceae and Lactobacillus taxa, present in kidney stones and urine, are associated with pro- or anti-lithogenic influences, respectively, on kidney stone development^7,9,15. As such, we sought to evaluate the susceptibility of these specific taxa to antibiotic disturbance, as the use of antimicrobials have been associated with the development of stones^7,26. Proteus, a causative agent of UTIs and struvite stones²⁷, is from the Enterobacteriaceae family and was one of the taxa that increased in abundance most after antibiotic use (6 ASV’s, FDR < 0.05), while Lactobacillus was significantly inhibited by antibiotics (Fig. 2d). Molecular and culture-based analysis of all Enterobacteriaceae and Lactobacillus at the endpoint for all groups, revealed that there was no impact of antibiotics on the Enterobacteriaceae, but a significant reduction of Lactobacillus (Two-way ANOVA f = 2.097, p = 0.04; Fig. 2e, f = 10.665, p = 0.013; Fig. 4d). Antibiotic-associated dysbiosis is consistent with published clinical urobiome data^7,28,29 as well as antibiotic sensitivity testing across multiple Lactobacillus³⁰ and Enterobacteriaceae³¹ species.

Proteomic techniques have revealed an upregulation of proinflammatory cytokines in the kidneys and urine of stone formers³². In our study, we observed a significant, antibiotic-associated increase in the urinary cytokines IL-1β and IL-6, but not IL-18 (Two-way ANOVA, f = 16.49, p = 0.002; Fig. 2f), which could reflect urinary dysbiosis-driven inflammation due to higher levels of Enterobacteriaceae³³ or an oxalate-driven inflammatory response due to reduced oxalate degradation^34,35.

Murine kidney microbiota is stable, but shifts towards uropathogenic bacteria with cefazolin exposure

To evaluate the stability of the murine renal microbiota, we quantified alpha diversity of the kidney microbiome longitudinally for animals not exposed to antibiotics, similar to previous analyses in low microbiome echosystems³⁶. While antibiotic exposure did not significantly change the number of microbial species detected in the kidneys (paired t-test, t = 0.5, p = 0.6’ Fig. 3a), there was a shift in overall microbial composition based on weighted UniFrac dissimilarity (Permanova, f = 3.94, p = 0.03; Fig. 3b). In contrast, there was no impact of a short-term course of antibiotics for the kidney microbiome (Fig. 4e; one-way ANOVA, N = 4-6/group, df=9, f = 0.03 for Abx; p = 0.867; Fig. 4f; Permanova, N = 4-5/group; df=8, f = 1.153,p = 0.654). There were no significant changes in renal microbial alpha diversity over 30 days (Pearson’s correlation, r = 0.14, 0.45, 0.5, 0,42 and p = 0.6, 0.2, 0.16, 0.2 for equitability, Margalef, PD_whole_tree, and Shannon; Fig. 3c), indicative of a stable community³⁷. Similar to the LUT, Acinetobacter, Lactobacillus and Lactococcus (1 ASV each) were lost in the kidneys with antibiotic treatment (DESeq2; FDR < 0.05). Tepidimonas and Lysinibacillus (1 ASV each) proliferated with antibiotic administration (Fig. 3d, Supplementary data S3; FDR < 0.05). There was no effect of a short-term course of antibiotics on species number or composition (paired t-test, t = 0.03, p = 0.867; Fig. 4e).

To validate molecular data, an RNA-FISH protocol was utilized to localize and quantify microbial DNA through hybridization in mouse kidney tissue. Three targeting probes were used for all bacteria, Enterobacteriaceae, or Lactobacillus. The protocol was validated with pure cultures from either the Enterobacteriaceae or Lactobacillus taxa, isolated from mouse urine (Fig. 5a–c). The EUB (universal), PB (Enterobacteriaceae) and GC (Lactobacillus) probes hybridized to the intended targets of all bacteria, Enterobacteriaceae, and Lactobacillus. There was co-localization with DAPI signals with all probes, and no cross hybridization observed in triplicate samples. Kidney slides stained with DAPI and any one of the three probes exhibited bacterial signals primarily located in the medulla, in or around the renal tubules and collecting ducts (Fig. 5d–f). Negative controls, which included stained slides without tissue or bacteria (Fig. 5g) did not exhibit any bacterial signals. Additionally, serial sections of mice kidney samples stained with randomly scramble probe sequences did not exhibit fluorescent signals, compared to their counterpart sections from the same samples stained with bacterial targeting probes (Fig. 5h–j). Quantitative analysis of bacteria was independently performed by two analysts blinded to sample identifiers. Analysis of surroundings and borders of kidney tissue did not show the presence of randomly distributed bacteria, suggesting no contamination during slide processing³⁸. Histologic analysis revealed no indications of calcium oxalate deposition or kidney injury. Quantitative results show a significant loss of total microbial density in the kidneys with antibiotics (paired t-tests, t = 12.69 p = 0.04), with a clear impact on Lactobacillus (t = 9.81, p = 0.04), but not on Enterobacteriaceae (t = 0.04, p = 0.316; Fig. 3e), consistent with molecular (Figs.  2e, 3d) and culture-based (Fig. 4d) analyses in urine. These findings suggest a shift in kidney microbial communities caused by antibiotics towards antibiotic-resistant, uropathogenic Enterobacteriaceae and away from uroprotective bacteria such as Lactobacillus. Validation of results were obtained through antibiotic resistance assays on a kidney stone-associated strain of E. coli (ATCC 43886)^7,13 and a strain of L. crispatus that is negatively associated with urinary stone disease^7,8,9,39, such that L. crispatus was significantly more susceptible to three out of four antibiotics tested compared to E. coli (Fig. 3f; p = 0.02). Genomic analysis revealed that the genome of E. coli (ATCC 43886) harbors more antibiotic resistance genes than L. crispatus (Fig. 6a).

Occasional outbreaks of Enterobacteriaceae hybridization in the kidneys were apparent and clusters were often observed (Fig. 3g). These observations were consistent with molecular data which saw three of the 10 mice exhibiting very high Enterobacteriaceae counts compared to the other animals (Fig. 6b), which may result from virulence factors, as suggested by the higher number of genes for biofilm formation, stress response, and competition than L. crispatus (Fig. 6a). Molecular comparison of Enterobacteriaceae and Lactobacillus in the kidneys, reveals that either one or the other taxa was dominant, indicative of potential direct competition (Fig. 6c).

Specific strains of Lactobacillus and E. coli compete and differentially impact calcium oxalate crystal metrics through known non-canonical pathways

To determine the direct impact of candidate bacteria on the growth of calcified crystals, we utilized a CDC bioreactor system, which is a specialized chemostat that mimics the constant flow, shear forces, temperature, and biochemical environment of the kidneys, while providing a removable surface for biofilm growth⁴⁰. Uroprotective L. crispatus and stone-associated strain of E. coli (ATCC 43886) were incubated in the bioreactor individually and in co-culture to evaluate their influence on CaOx crystallization.

Artificial urine media (AUM; Supplementary data S4) was run at a constant flow into the chemostat system, which was placed on a stir plate at 37 °C and at a constant stir rate. The system was either run sterile or inoculated with candidate bacteria at equal starting densities. After 72hrs, removable polycarbonate coupons were analyzed with scanning electron microscopy (SEM) and energy dispersive x-ray spectroscopy (EDS). Aliquots of media from each experiment were used to quantify bacterial density and viability. No culturable bacteria were recovered from sterile controls. For SEM and EDS, a pure, clinically extracted CaOx stone was included for crystal morphology analysis, along with in vitro CaOx crystals for validation of bioreactor studies.

In qRT-PCR analysis, when grown alone, 100% of sequences were attributed to E. coli or L. crispatus, relative to a universal bacterial primer (paired t-test against L. crispatus, t = 7.173, 8.459 and p = 0.0003, 0.0002 for co-cultures and E. coli; Fig. 7a; paired t-test against L. crispatus, t = 8.54, 9.76 and p = 0.0009, 0.0005 for co-cultures and E. coli; Fig. 7b). When co-cultured, E. coli (ATCC 43886) comprised ~90% of the culture after three days, despite equal inoculation densities (Fig.  S7a, b). No detectable DNA was recovered with qRT-PCR from sterile controls. The E. coli strain (ATCC 43886) exhibited significantly higher viability compared to L. crispatus, in pure culture. However, the proportion of viable bacteria was significantly reduced in co-culture (ANOVA, f = 52.81, p = 0.0002; Fig. 7c). The more effective competitiveness by E. coli (ATCC 43886) was also reflected in the number of genes for competition in genome analyses (Fig. 6a).

The CaOx crystals were styloid laminated sheets in a clinical pure CaOx stone (Fig. 4a), consistent with CaOx monohydrate⁴¹, which is generally found on the surface of CaOx stones. In sterile media, crystals exhibited an octahedral structure, consistent with CaOx dihydrate crystals (Fig. 4b)⁴². Additionally, spherical structures less than 1μm in diameter were common, which appeared to aggregate into amorphous, then octahedral structures (Fig. 4b), consistent with a recently reported, non-canonical pathway for CaOx crystallization and growth⁴³. To confirm abiotic origins of these nanospheres, we filter sterilized fresh AUM and observed media under light microscopy after one hour. Imaging showed a few shapes compatible with CaOx dihydrate crystals, but also showed many of the styloid crystal shapes and nano-sized spheres that were the same size and shape as those observed in SEM analysis (Fig. 7d). No culturable bacteria were recovered from this media, similar to sterile bioreactor runs. When stone-associated strain of E. coli (ATCC 43886) was grown by itself, octahedral structures were abundant along with nanospheres that appeared to aggregate to octahedral crystals (Fig. 4c). Bacteria were often found attached to octahedral structures. In contrast, when L. crispatus was grown alone, only small, amorphous crystal structures formed, with no bacterial attachment (Fig. 4d). When E. coli (ATCC 43886) and L. crispatus were co-cultured, styloid, CaOx monohydrate crystal structures were observed (Fig. 4e), indicative of an entirely different microbe-crystal interaction. Quantification of crystal size in bioreactor studies showed E. coli (ATCC 43886) grown alone produced the largest crystals, followed by sterile media, then L. crispatus grown alone (ANOVA, f = 52.54, p < 0.001; Fig. 4f). However, the influence on crystallization, based on plate-based crystal size assays, of other ATCC purchased species of lactic-acid bacteria and different strains of E. coli isolated from calcium-based stones was different from the findings observed with the bioreactor experiments. (ANOVA, f = 78.97, p < 0.001; Fig. 4g). Collectively, data suggest that this strain of E. coli (ATCC 43886) facilitates the growth and aggregation of CaOx crystals into larger and more complex octahedral structures through a non-canonical pathway, whereby CaOx nanospheres aggregate into amorphous structures and continue aggregating until the octahedral structures develop and further aggregate⁴⁴. In contrast, L. crispatus appears to inhibit this process at the amorphous phase.

The differential interaction between bacteria and CaOx crystallization could either be mediated through oxalate production/degradation, by the production of pro-/anti-lithogenic molecules, or by the interaction of proteins on the bacteria surface on crystal faces¹⁵. Genome analysis revealed that both species have an equal number of oxalate metabolism genes (Fig. 6a). A previous report showed that healthy individuals had a higher abundance of L. crispatus oxalate-degrading genes in their urine compared to CaOx stone formers⁹. In vitro oxalate-degrading assays revealed that L. crispatus was able to degrade significant amounts of oxalate, confirming genome and metagenomic analyses, but the stone-associate E. coli (ATCC 43886) produced significant amounts of oxalate (Two-way ANOVA, f = 23.47, p < 0.001; Fig. 7e).

Genome analysis revealed that E. coli (ATCC 43886) harbors more genes that overlap with known lithogenic factors, compared to L. crispatus, which includes genes for urea and metal interactions (Fig. 6a). In an energy dispersive x-ray spectroscopy (EDS)-based assay, peaks of calcium, carbon, and oxygen were observed in crystals formed by E. coli (ATCC 43886) and L. crispatus, consistent with CaOx crystals (Fig. 5a). The largest CaOx peaks were observed in the clinically extracted stone, along with the crystals from the sterile and E. coli (ATCC 43886) experiment. Other elements were present in lower abundance, such as sodium, chloride and nitrogen. Binomial dissimilarity matrix analysis of the whole elementome revealed treatment-specific effects with significant differences based on the presence or absence of bacteria (Permanova, f = 4.85, p = 0.001; Fig. 5b, ANOVA, f = 1.47-3.83, p = 0.004-0.206; Fig. 8a–h). Crystals that developed when L. crispatus was present were enriched in sulfur (Fig. 8f) and phosphorus (Fig. 8h), known calcium competitors linked to lower CaOx urolith formation^{44,45,46,47,48}. In contrast, crystals formed when E. coli (ATCC 43886) was present were enriched in chloride (Fig. 8g) and iron (Fig. 8c), known promoters of CaOx crystals^49,50,51.

Metagenomic and FISH data from human renal specimens show microbial signature differentiated by tissue proximity, age, and ethnicity

To evaluate the presence of microbial signatures in human renal specimens, we conducted an RNA-FISH analysis of kidney biopsies and autopsies, using the universal bacterial probe (EUB). Image analysis revealed bacteria in all samples, typically reserved to lumen spaces (Fig. 6a-d). Bacterial density, estimated as the relative area of bacterial stain compared to total kidney tissue, was not significantly different between biopsies and autopsies, which suggests that bacterial proliferation post-mortem did not contribute to bacterial signals (paired t-test, t = 2.195, p = 0.177; Fig. 6e).

To further evaluate the presence of bacteria in human kidneys, bacterial signatures in an independent population of kidney specimens were re-analyzed from a previously published transcriptomic dataset of micro-dissected glomeruli and tubuli that underwent polyA enrichment prior to sequencing (BioProject PRJNA725213)⁵². There were 101 sequenced samples from healthy participants available, which included 71 glomeruli and 30 tubuli. There was an average of 38,646,509 + /− 1,301,280 total reads/sample, with 19,086,347 + /− 639,898 mapping to host reads, and 56,191 + /− 2934 mapped to prokaryotes after removal of potential contaminants and low-quality reads. We observed significant differences in the ratio of high-quality sequences mapping to prokaryotes vs. human reads between glomeruli and tubuli (paired t-test, t = 26.99, p < 0.001; Fig. 9a). Importantly, nearly all reads exhibited 100% sequence homology with their microbial references, which minimizes the possibility of false microbial annotations (Fig. 9b). Wih this analysis, we observed significantly greater species richness in the glomeruli compared to tubuli (paired t-test, t = 25.37, p = 0.0007; Fig. 9c), with a posititve correlation with increasing age (Pearson’s correlation, r = 0.43, -0.14 and p = 0.0003, 0.45 for glomeruli and tubuli; Fig. 9d). The composition of the microbial signatures was significantly different between the glomeruli and tubuli, assessed with a Bray-Curtis dissimilarity index (Permanova, f = 28.58, p = 0.001; Fig. 9e). Bacterial phyla indentified in glomeruli and tubuli were dominated by Pseudomonadota and Bacilliota (Fig. 9f), consistent with previous urobiome studies^7,21. Taxa that differentiated the glomeruli and tubuli, based on a DESeq2 analysis, were primarily Streptomyces (38 ASV’s), Pseudomonas (22 ASV’s), Burkholderia (16 ASV’s), and Streptococcus (15 ASV’s) in the glomeruli and Bacillus (17 ASV’s) in the tubuli (DESeq2, FDR < 0.05; Fig. 9g; Supplementary data S5). Similar results were obtained with an ANCOM analysis (ANCOMBC, FDR < 0.05; Fig. 9h). Ethnicity was a significant driver of composition and species richness in two-way analyses (Permanova, f = 2.85, p = 0.007; Fig.  S9i; ANOVA, f = 6.191, p = 0.0007; Fig. 10a). We did not see differences in the kidney microbial signatures by gender identity (Two-way ANOVA f = 0.061, p = 0.805; Fig. 10b; Permanova f = 1.37, p = 0.253; Fig. 10c). Negative controls used in this analysis for decontamination were variable, but significantly different from positives, independent of source (Fig. 10d).

Scrutinization of microbiome data

To scrutinize and support the validity of the microbiome data generated from our analysis, we used a hybrid approach. First, we conducted statistical decontamination with Decontam, which removes 90% of contamination with minimal false negatives⁵³ and we also conducted a knowledge-based removal of common taxa found in laboratory reagents⁵⁴. Statistical decontamination led to a 23% reduction in the number of unique taxa in the human data compared to a 19% when bacteria was censored through a knowledge-based approach. In mouse data, statistical decontamination led to a 2.3% reduction in unique taxa compared to a 35% reduction with the knowledge-based approach. It must be noted, however, that in mouse studies, multiple sources of negatives that were directly tied to the experimental procedures were used as negatives for statistical decontamination. Additionally, many of the taxa that have been identified as reagent contaminants⁵⁴ that we used for the knowledge-based scrutinization, such as Enterobacter, Escherichia, Corynebacterium, Kocuria, Bacillus, Streptococcus, and others, have been recurrently identified in the urinary tract microbiome⁵⁵. As such, removal of these taxa would necessarily lead to a number of false negatives.

To assess the impact of decontamination methods on the statistical differences shown here, we compared the statistical significance in the beta diversity of human glomeruli vs. tubuli and mouse kidneys vs. urine. While the knowledge-based decontamination method led to a reduction in the statistical significance for both human and mouse data, beta diversity analyses were highly significant regardless of the decontamination method chosen (Fig. 11a, b). Additionally, to assess the impact of contamination on the statistical differences between groups, we simulated contamination in both the human and mouse data by using a random number generator to simulate universal, stochastically distributed taxa, similar to what would be seen with reagent-based contamination. In this analysis, we found that the number of contaminants exhibited a significant negative correlation with the beta diversity test statistic in both datasets (Pearson’s correlation, r = -1, p < 0.001; Fig. 11c, Pearson’s correlation, r = -96, p < 0.001; Fig. 11d).