Antibiotics ameliorate lupus-like symptoms in mice

Antibiotics given post disease onset attenuated lupus

The onset of autoimmune responses in female MRL/lpr mice is as early as 6 weeks of age²⁶. To determine the effects of antibiotics on active disease in lupus-prone MRL/lpr mice, we treated female mice with a combination of antibiotics (ampicillin, neomycin, metronidazole and vancomycin⁵) started at 9 weeks of age and post disease onset. The treatment led to enlarged ceca as expected for antibiotic treatments, whereas the overall body weight did not change (data not shown). Spleen and mesenteric lymph node (MLN) weights were significantly decreased with antibiotic treatment compared to controls (Fig. 1A). The serum level of IgG autoantibodies against double-stranded (ds) DNA was also significantly reduced by the mixed antibiotic treatment (Fig. 1B). As kidney inflammation (or lupus nephritis) affects up to 60% of lupus patients^27,28, we determined the renal function by measuring proteinuria and kidney histopathology. Both glomerular and tubulointerstitial scores were significantly decreased with antibiotic treatment compared to controls (Fig. 1C). The antibiotics also significantly reduced proteinuria (Fig. 1D), suggesting improvement of renal function. Together, these results indicate amelioration of lupus-like disease in female MRL/lpr mice with post-disease-onset antibiotic treatment.

We next sought to understand the underlying mechanism of how antibiotic treatment ameliorated lupus in female MRL/lpr mice by examining immune cell differentiation and inflammatory mediator production (Fig. 2). IL-6, known to promote lupus disease in both human and mouse due to its ability to increase T-helper (Th)17 differentiation and IL-17 production^29,30, was significantly decreased in the serum with antibiotic treatment (Fig. 2A). We then quantified IL-17-producing cells in the spleen and kidney by using flow cytometry. As anticipated, antibiotic treatment significantly reduced the percentages of IL-17⁺ cells (Fig. 2B), CD3⁺CD4⁺RORγ⁺ Th17 cells and Lin⁻CD4⁻RORγ⁺ group 3 innate lymphoid cells (ILC3s) (Fig. 2C). Both Th17 cells and ILC3s are known producers of IL-17 in the spleen³¹, and their decrease suggests systemic downregulation of IL-17 with antibiotic treatment. In the kidney, the double negative (DN) T cells and Th17 cells are major sources of IL-17³² and they both were significantly reduced with post-disease-onset antibiotic treatment (Fig. 2D and Fig. 2E, respectively). As IL-17 was too low to be detected in the circulation, we measured IL-17 protein in the splenic extract that reflected the systemic level of IL-17. Post-disease-onset antibiotic treatment significantly reduced the level of splenic IL-17 (Fig. 2F). These results suggest that antibiotics may attenuate lupus-like disease in female MRL/lpr mice by suppressing the production of IL-17 from Th17 cells and ILC3s in the spleen and from DN-T and Th17 cells in the kidney. Interestingly, we also found a significant increase in serum IL-10 with antibiotic treatment (Fig. 2G) that correlated with an increase in the number of IL-10-producing cells in the MLN (Fig. S1A). IL-10 is known as a protective cytokine in MRL/lpr mice³³.

Antibiotics given post disease onset reshaped gut microbiota

It has been recently reported that Th17 cells can migrate from the gut to the kidney to facilitate the development of lupus nephritis³⁴, whereas the generation of Th17 cells in the gut is dependent on the gut microbiota³⁵. We thus characterized the gut microbiota in antibiotics-treated mice. While antibiotic treatment initiated post disease onset did not decrease the bacterial diversity (Fisher index, Fig. S1B), it did reduce the bacterial load of fecal microbiota by 2 magnitudes (Fig. S1C). Based on 16 S rRNA sequencing analysis, the overall structure of the remaining gut bacteria was distinct from untreated animals (Fig. 3A; p < 0.01, PERMANOVA; red vs. blue symbols). In addition, the overall structure of the gut microbiota was different between the time points before (3 and 8 weeks of age) and after (11 and 15 weeks of age) antibiotic treatment initiation at 9 weeks of age (also in Fig. 3A; p < 0.01, PERMANOVA; comparison within the red symbols). Moreover, the antibiotic-treated group before given the treatment (3 and 8 weeks of age) shared similar gut microbiota composition with the Control group, but the antibiotic treatment initiated at 9 weeks of age appears to have altered the bacterial composition at 11 and 15 weeks of age (Fig. 3B). Further analysis on specific bacteria groups revealed that antibiotic treatment significantly increased the abundance of Lactobacillus spp. and significantly decreased the abundance of Lachnospiraceae (Fig. 3C), two groups of bacteria previously shown to be associated with improved or deteriorated symptoms in MRL/lpr mice, respectively³. L. agilis, L. brevis, L. mucosae and L. reuteri, in particular, were below detection limit before antibiotic treatment, whereas their abundance increased to about 5% after antibiotics-mediated enrichment. In addition to these changes, treatment with antibiotics removed significant amounts of Bacteroidales and Clostridiales while increasing the relative abundance of Bacillales (Fig. 3D). These results suggest that antibiotics given post disease onset reshaped the gut microbiota, removing potentially harmful bacteria (e.g., Lachnospiraceae) and enriching those that are associated with better disease outcomes (e.g., Lactobacilli).

Vancomycin recapitulated the disease-attenuating effects of mixed antibiotics

A cocktail of 4 different antibiotics is impractical to implement as a treatment for lupus and more likely to induce resistance. Vancomycin alone, however, can remove Gram-positive bacteria such as Clostridial species (Lachnospiraceae) but spares Lactobacilli^17–20, making it a favorable choice as a potential intervention against lupus progression in MRL/lpr mice. Importantly, vancomycin is not absorbed in the intestine^36,37 and its effects are limited to targeting commensal bacteria in the gut lumen. We thus examined whether oral treatment of vancomycin initiated at 9 weeks of age could attenuate lupus. As anticipated, the relative abundance of Lactobacillus spp. was significantly elevated with vancomycin treatment (Fig. 4A). While the overall body weight did not change (Fig. S2A), vancomycin treatment significantly decreased the weight of spleen, MLN and major lymph nodes (Fig. 4B). Furthermore, vancomycin significantly reduced the level of circulating anti-dsDNA IgG (Fig. 4C), proteinuria (Fig. 4D), and renal histopathological scores (Fig. 4E). In contrast, neomycin, an antibiotic with a broad spectrum of activity against both Gram-positive and Gram-negative bacteria³⁸, did not affect the severity of lupus disease when given starting from 9 weeks of age. This indicates that the decrease in bacterial load, achieved by both vancomycin and neomycin treatments, was not the reason for disease attenuation. Together, these results suggest that vancomycin given post disease onset recapitulated the attenuated disease phenotype seen with mixed antibiotic treatment.

Vancomycin reshaped the gut microbiota and differentially affected KEGG pathways

We collected weekly fecal samples from vancomycin-treated mice and determined longitudinal changes of gut microbiota composition by using 16 S rRNA sequencing. The diversity of gut microbiota was largely reduced upon vancomycin administration (Fisher index, Fig. S2B). Similar to the mixed antibiotic treatment, vancomycin removed Clostridiales right after treatment initiation and Bacteroidales at most of the observed time points (Fig. 5A). Many other groups of bacteria were also removed by vancomycin, including Desulfovibrionales and Turicibacterales, whereas Enterobacteriales were enriched. Anaeroplasmatales, on the other hand, was increased by vancomycin treatment at the later time points. These results suggest that vancomycin given during active disease reshaped the gut microbiota in MRL/lpr mice.

We next established mathematical networks to model the gut microbiota changes at the phylum level (Fig. 5B). While some relationships remain the same with and without vancomycin treatment, such as the positive influence of Bacteriodetes on Verrucomicrobia (nodes 2 and 7), overall, treatment with vancomycin significantly affected the interaction among different bacterial groups. Firmicutes (node 4), for example, were shown to positively influence Verrucomicrobia (node 7) only in the vancomycin-treated group. Both Lactobacillaceae (“good” bacteria) and Lachnospiraceae (possibly “harmful” bacteria in this model) belong to the phylum Firmicutes. Another example is Proteobacteria (node 5), which are commonly used to represent Gram-negative bacteria. In untreated mice, Proteobacteria were involved in a complex interaction network with Bacteroidetes (node 2), Actinobacteria (node 1) and Verrucomicrobia (node 7), suggesting that Gram-negative bacteria may contribute to lupus pathogenesis in MRL/lpr mice. However, these interactions were absent in vancomycin-treated mice, suggesting that Gram-negative bacteria no longer play an important role to promote lupus in the presence of vancomycin.

To get insights into functional categories and pathways affected by antibiotics treatment, we performed PICRUSt analyses, and analyzed KEGG level 3 pathways with DESeq. 2. The analysis showed that the presentation of functional pathways was relatively stable over time for the control group, whereas treatment with vancomycin produced the most significant changes of the functional pathways at 9 weeks of age and 4 days after the initiation of antibiotic treatment (Fig. 6A). Some of the changes sustained beyond 9 weeks of age in the vancomycin group, while others returned to the baseline levels. Among the pathways with significant changes (Table S1), many exhibited similar trends as in our previous publication³. These include the vancomycin-mediated upregulation of Peptidoglycan biosynthesis and Transcriptional factors pathways that were associated with improved lupus-like symptoms in MRL/lpr mice, as well as vancomycin-mediated downregulation of Glyoxylate and dicarboxylate metabolism, Histidine metabolism, and Phenylalanine, tyrosine and tryptophan biosynthesis pathways that were associated with deteriorated lupus-like symptoms in MRL/lpr mice.

In addition to analysis of the time course, we also determined differential presentation of functional pathways between control and vancomycin groups when data from 9–15 weeks were averaged (Fig. 6B). This analysis showed 75 functional pathways that were significantly altered, including the Phenylalanine, tyrosine and tryptophan biosynthesis pathway that was associated with more severe lupus disease³ and significantly downregulated by vancomycin treatment. Another important pathway, Lipopolysaccharide biosynthesis, was also significantly downregulated by vancomycin regardless of sampling time. Detailed analysis of PICRUSt data on the ortholog level (Fig. S3 and Table S2) revealed vancomycin-mediated downregulation of 12 functional genes within the Lipopolysaccharide biosynthesis pathway. When the relative levels of these genes were plotted over time, they exhibited the same pattern of a sharp decrease at 9 weeks of age, followed by gradual recovery from 10–11 weeks of age (Fig. 6C). Importantly, a majority of these genes were Lpx genes involved in lipid A biosynthesis³⁹. Lipid A is the endotoxic component of LPS. These analyses suggest that vancomycin may attenuate lupus-like disease in MRL/lpr mice by downregulating the relative abundance of Gram-negative bacteria and the LPS endotoxin.

Vancomycin decreased intestinal permeability and the plasma level of LPS

LPS accelerates lupus progression in several lupus-prone mouse models^21–25. A leaky gut may allow for the translocation of Gram-negative bacteria across the intestinal epithelium, leading to an increase of LPS—a cell wall component of Gram-negative bacteria—in the circulation. We thus determined the intestinal permeability of vancomycin-treated mice by measuring the diffusion of orally gavaged (for small intestine) or rectally administered (for colon) FITC-conjugated dextran. The result showed that vancomycin treatment significantly decreased the permeability of small intestine (Fig. 7A) and had a trend to decrease colonic permeability (Fig. S4A). Neomycin, on the other hand, did not change intestinal permeability (Fig. S4B). In addition, vancomycin significantly increased the epithelial expression of barrier-forming tight junction transcripts Occludin, ZO-1 (Fig. 7B), Cldn1 and Cldn3 (Fig. S4C), whereas the transcript level of pore-forming tight junction protein Cldn2 did not change with vancomycin treatment (data now shown). While the intestinal epithelium appeared intact regardless of treatment, significantly less inflammation was observed with vancomycin treatment (Fig. 7C). Further studies that directly measured LPS in the circulation indicated that vancomycin indeed significantly decreased the serum level of LPS (Fig. 7D). This is consistent with significantly reduced bacterial translocation to MLN with vancomycin treatment (Fig. 7E). Together, these results suggest that vancomycin may attenuate lupus-like disease in MRL/lpr mice by reducing the “leakiness” of the gut epithelium and preventing the translocation of LPS and/or LPS-containing bacteria from the gut lumen to the circulation.

Mice and antibiotic treatment

MRL/lpr mice (stock #000485) were purchased from The Jackson Laboratory (Bar Harbor, ME) and maintained in a specific pathogen-free facility. All mice used were female as lupus has a strong female bias. Antibiotic mixture (1 g/l ampicillin, 1 g/l neomycin, 1 g/l metronidazole and 0.5 g/l vancomycin) was given in the drinking water starting from 9 weeks of age till euthanasia at 16 weeks of age. For single antibiotic treatment, 2 g/l vancomycin or 2 g/l neomycin was given in the drinking water from 9 weeks of age till euthanasia at 15 weeks of age. The drinking water with antibiotics was refreshed every 5 days. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Institutional Animal Care and Use Committee (IACUC) of Virginia Tech College of Veterinary Medicine (Animal Welfare Assurance Number: A3208-01). For anesthesia and euthanasia, isoflurane and CO₂ were used, respectively, according to the IACUC protocol. All experiments were performed in accordance with relevant guidelines and regulations.

Microbiota 16S and PICRUSt inferred metagenomics analyses

Fecal microbiota samples were obtained by taking an individual mouse out of the cage and collecting a fecal pellet. The 9-week fecal pellet was taken 4 days after the initiation of antibiotic treatment. To avoid cross-contamination, each microbiota sample was collected by using a new pair of sterile tweezers. All samples were stored at −80 °C till being processed at the same time. Sample homogenization, cell lysis and DNA extraction were performed as previously described^3,57. The V4 region of purified 16 S rRNA gene amplicons were sequenced bi-directionally (paired-end 150 bp) on an Illumina MiSeq. Microbiome data analysis was performed as previously described^3,57. Briefly, OTUs were picked by usearch, taxonomy was assigned against Greengenes reference database. Microbial diversity measures including Fisher and observed species were calculated by using QIIME. The datasets generated and analyzed during the current study are available in the NCBI number SRP102626.

Bacterial metagenomes were predicted using PICRUSt⁵⁸ by comparing 16 S to database gg13.5 then to Integrated Microbial Genomes (IMG). The OTUs were mapped to gg13.5 database at 97% similarity by QIIME’s “pick_closed_otus” script. The OTUs abundance was normalized using 16 S rRNA gene copy numbers from known bacterial genomes. The normalized OTUs were used for metagenomes prediction in PICRUSt. The predicted Kyoto Encyclopedia of Genes and Genomes (KEGG) orthologs⁵⁹ was summarized to level-3 functional categories and compared among groups by using the Statistical Analysis of Metagenomic Profile package⁶⁰. Differentially represented gene families were identified by two-sided Welch’s t-test with Storey’s false-discovery-rate correction. DESeq. 2 analysis⁶¹ in R environment was performed with read count data for statistical analysis [R 2016, version 3.3.2 https://www.r-project.org/]. We compared between treatments and control for each week (vancomycin vs. control at 8 week, 9 week, etc.). We also compared consecutive weeks for each genotype respectively (e.g. 8 week vs. 9 week for vancomcyin), and a time course analysis which tests interaction between time and treatment. Functional categories were considered significantly differentially abundant between conditions if their adjusted P value was ≤ 0.001 and their absolute value of log₂ fold change was ≥ 2. Categories with low average count were filtered before the analysis. Read counts for each KEGG category were normalized by DESeq. 2 and averaged between replicates. The average read counts were centered and scaled before used for clustering analysis to produce a heatmap.

Mathematical modeling

A standard way in mathematically describing the temporal dynamics of interacting species is the generalized Lotka-Volterra or predator-prey system, which reads as follow

where (y(t)) is the diagonal matrix with the state variable y(t) = [y ₁(t), …, y _n(t)]^T (describing the temporal abundances of the different organisms), r is the vector of intrinsic growth rate parameters, and A is the interaction matrix that characterizes the influences within the network of each species, e.g., the entry a _ij of the matrix A describes the influence of species j on the growth of species i [M. Chung, J. Krueger, and Mihai Pop. Robust Parameter Estimation for Biological Systems: A Study on the Dynamics of Microbial Communities. In revision at Mathematical Bioscience, 2017. [arXiv Preprint]]. Hence, the network of the species is identified by this interaction matrix A.

To determine the interaction A we assumed that the dynamics were at an asymptotically stable equilibrium $\overline{\mathit{y}}$, where $\overline{\mathit{y}}$ was determined by the data (Fig. 5A). Mathematically such equilibria are given by solution of the nonlinear equation

where ℜℯ and is the real part of λ _j and λ _j is the j ^th eigenvalue of the matrix diag(r+A $\overline{\mathit{y}}$) + diag($\overline{\mathit{y}}$)A.

Hence, we need to find A such that the above equations are fulfilled. Note that this is a nonconvex problem and multiple solution may exist. More precise, multiple network configuration may describe the same asymptotically stable equilibrium $\overline{\mathit{y}}$. We used a repeated Monte Carlo sampled direct search method [Wright, Stephen and Nocedal, Jorge, Numerical Optimization, Springer, 2006] to identify solution A of above problem, see Fig. 5B.

Renal function

Urine samples were collected biweekly and all samples were stored at −20 °C till being processed at the same time. We used Pierce Coomassie Protein Assay Kit (Thermo Scientific) to test the total protein level in the mouse urine. When mice were euthanized at 14 weeks of age, kidneys were fixed in formalin for 24 h, paraffin embedded, sectioned, and stained with Periodic acid–Schiff (PAS) at the Histopathology Laboratory at Virginia Maryland Regional College of Veterinary Medicine. Slides were read with an Olympus BX43 microscope. All slides were scored in a blinded fashion by a certified veterinary pathologist (Cecere). Glomerular lesions were graded on a scale of 0 to 3 for each of the following five categories: increased cellularity, increased mesangial matrix, necrosis, the percentage of sclerotic glomeruli, and the presence of crescents. Similarly, tubulointerstitial lesions were graded on a scale of 0 to 3 for interstitial mononuclear infiltration, tubular damage, interstitial fibrosis, and vasculitis.

Endotoxin quantification and enzyme-linked immunosorbent assays (ELISA)

Separated serum after blood clotting was saved at −20 °C until use. We used Pierce LAL Chromogenic Endotoxin Quantitation Kit (Thermo Scientific) to measure serum endotoxin level by following the kit’s instructions. For detection of anti-double-stranded DNA (dsDNA) IgG, we used previously described methods²⁶. Serum IgG, IL-6 and IL-10 concentrations were determined with mouse IgG (Bethyl Laboratories), IL-6 (Biolegend) and IL-10 (Biolegend) ELISA kits according to the manufacturer’s instructions.

Intestinal permeability and colonic histopathology

In vivo intestinal permeability assay to assess barrier function was performed using Fluorescein isothiocyanate conjugated dextran (FITC-dextran, Sigma-Aldrich) method. For permeability of the small intestine, mice were water starved overnight then orally gavaged with FITC-dextran (40 mg/100 g body weight). For colonic permeability, FITC-dextran (20 mg/100 g body weight) were rectally administered twice with a 2-h break. After 4 h (for colonic permeability, 4 h after initial administration), mice were anesthetized and blood was collected and saved in the dark. Serum was then prepared from the blood, diluted 1:1 with PBS, and measured in a 96-well microplate in duplicates for the concentration of FITC in the serum by Glomax (Promega) with an excitation wavelength of 485 nm and an emission wavelength of 528 nm, using serially diluted FITC-dextran as the standards. Colonic histopathology was graded on a scale of 0 to 4 for each of the following six categories: inflammation, edema, epithelial defects, crypt atrophy, hyperplasia, and dysplasia⁶². Only inflammatory histopathology was observed in our samples. All slides were scored in a blinded fashion (Eden).

Cell isolation and flow cytometry

Spleen, MLN and kidney were collected and mashed in 70-μm cell strainers with C10 media (RPMI 1640, 10% fetal bovine serum, 1 mM sodium pyruvate, 1% 100 MEM non-essential amino acids, 10 mM HEPES, 55 μM 2-mercaptoethanol, 2 mM L-glutamine, 100 U/ml penicillin-streptomycin, all from Life Technologies, Grand Island, NY). For splenocytes, red blood cells were lysed with RBC lysis buffer (eBioscience, San Diego, CA). To isolate intestinal epithelial cells (IECs), the entire intestine including small intestine and colon was opened longitudinally and cut into pieces. The pieces were incubated twice in EDTA-DTT solution and intensively vortexed to harvest IEC-enriched fractions. For surface marker staining, cells were blocked with anti-mouse CD16/32 (eBioscience), stained with fluorochrome-conjugated antibodies, and analyzed with Attune NxT flow cytometer (Thermo Scientific). For intracellular staining, Foxp3 Fixation/Permeabilization kit (eBioscience) was used. Anti-mouse antibodies used in this study include: CD3-APC-eFluor 780, CD8-PE-Cy7, CD4-PerCP-Cy5.5, RORγT-PE, CD3e-biotin (eBioscience); CD45-APC-Cy7, IL-10-BV421, IL-17A-APC, CD49b-biotin, CD19-biotin (Biolegend, San Diego, CA); Biotin-FITC (MACS). Flow cytometry data were analyzed with FlowJo.

Reverse transcription-quantitative polymerase chain reaction (RT-qPCR)

Isolated IECs were homogenized with Bullet Blender homogenizer (Next Advance, Averill Park, NY) and total RNA was extracted with RNeasy Plus Mini Kit (Qiagen, Valencia, CA) according to the manufacturers’ instructions. Genomic DNA was removed by digestion with RNase-free DNase I (Qiagen). Reverse transcription was performed by using iScript cDNA Synthesis Kit (Bio-Rad, Hercules, CA). Quantitative PCR was performed with iTaq Universal SYBR Green Supermix (Bio-Rad) and ABI 7500 Fast Real-Time PCR System (Applied Biosystems, Grand Island, NY). Relative quantities were calculated using Villin as the housekeeping gene. Primer sequences for mouse Villin, ZO-1, Occludin, Cldn1 and Cldn3 are available upon request.

Tissue extraction for ELISA

Total protein was extracted from spleen according to a published method⁶³ for the measurement of IL-17 with ELISA. Briefly, frozen splenic tissue was homogenized in ice-cold extraction buffer containing protease inhibitor mixture (Sigma-Aldrich). After homogenization, samples were maintained at 4 °C for 2 h, followed by centrifugation at 12,000×g for 20 min. Supernatants were collected and analyzed for IL-17A with ELISA (Biolegend).

Measurement of bacterial translocation

MLN was homogenized in PBS at 0.1 g tissue/ml PBS. After short spin-down, 100 μl supernatant was added to a blood agar plate (Hardy Diagnostics) and cultured at 37 °C for 36 h. The number of colonies was counted and calculated as colony-forming units (CFU) per g tissue.

Statistical analysis

Statistical analyses were performed by using R version 3.0.2 (sequencing data) or Prism GraphPad (non-sequencing data). Principal coordinate analysis was tested for significance by a permutational multivariate method PERMANOVA⁶⁴. Analysis of non-sequencing data was performed with nonparametric Mann-Whitney test for the comparison of 2 groups, and nonparametric Kruskal-Wallis test for the comparison of 3 groups. The results were considered statistically significant when p < 0.05.

Competing Interests

The authors declare that they have no competing interests.