Rapamycin treatment ameliorates HLA-B27-mediated gut inflammation and alters the microbiome in experimental spondyloarthritis

Hemizygous HLA-B*27:05 and human β₂m-transgenic (HLA-B27-Tg) Lewis rats carrying 55 copies of HLA-B27 and 66 copies of human β₂m in the 33–3 locus were used for this study. Aged-matched wild type (WT) Lewis rats were used in parallel. A total of 19 WT rats and 24 B27-Tg rats were used for the in vivo experiments, and an additional 4 WT animals and 9 B27-Tg animals for ex vivo studies. Rats were bred and housed in Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) approved facilities on the NIH Bethesda campus. Carbon dioxide (CO₂) was used for animal euthanasia in accordance with the guidelines for euthanasia of rodents using CO₂. Briefly, animals were placed in a 10-liter chamber and exposed to CO₂ using a flow rate of 3–7 liters per minute, achieving a CO₂ displacement rate of 30-70% per minute, until respiration ceases. Death was confirmed using accepted methods. All animal experiments were approved by the Animal Care and Use Committee at the National Institute of Arthritis and Musculoskeletal and Skin Disease (NIAMS).

Throughout the treatment course, animals were evaluated weekly for diarrhea via a standard stool scoring system. Stools from each animal were evaluated at treatment initiation and once weekly thereafter. Stools were assigned a score of 0 for solid stools, 1 for soft stools, 2 for semiliquid stools, and 3 for watery stools by veterinary facility staff.

For histologic evaluation, tissue was obtained from the distal colon or from the apex of the cecum of each animal. Tissue was obtained from vehicle-treated WT rats (n=9) or B27-Tg rats (n=11), as well as rapamycin-treated WT rats (n=10) and B27-Tg rats (n=13). Tissue was paraffin embedded, sectioned into 2–4 unique sections and stained with hematoxylin and eosin (H&E). Each tissue section was given a histological score in a blinded fashion by two independent observers according to a previously published scoring system (18). These scores were then averaged together to obtain the final histological score for the given animal. The tissue was evaluated for four separate categories: the presence or absence of gut-associated lymphoid tissue (GALT), degree of goblet cell loss, immune cell infiltrate, and area of tissue affected. Goblet cell loss, inflammatory infiltrate, and percent inflammation was scored from 0-3. The GALT category was given a score of 0 or 1 to indicate the presence or absence of GALT within a tissue slice. The scores for each subcategory were then summed together obtain a final composite histological score (histoscore).

Distal colon tissue samples obtained from rats (female and male) were homogenized in TRIzol reagent (Thermo Fisher Scientific). RNA was isolated using a standard phenol-chloroform protocol. RNA quantity and quality was assessed (Agilent) and RNA with an RNA integrity number (RIN) >7.0 was used for all RNA-seq analyses. Libraries were prepared according to the manufacturers guide (Illumina). The Illumina Novaseq 6000 system was used to perform 50-base, paired-end sequencing. Raw data were mapped to a customized genome based on rat rna6 with human transgenes HLA-B27 and hβ₂m added to account for their transgenic expression in the experimental model. Partek Genomics Suite 7.0 was used to calculate Reads per Kilobase Million (RPKM) and to analyze differential gene expression via ANOVA. For further analysis, only genes where at least one sample had a max expression of RPKM > 1, coefficient of variance (C.V.) > 0.3, and a minimum of |2-fold| change, and p<0.05 were used. Partek Genomics 7.0 was also used for principal component analysis (PCA), and for generating hierarchical clustering heat maps and Volcano plots.

Immune cells were isolated from colon tissue as described previously with some modifications (19). For these experiments 4 WT animals and 9 B27-Tg animals were used. Briefly, after removing connective tissue, colon tissue was washed in PBS several times and then cut into 0.5 cm sections in Hank’s balanced salt solution (HBSS, without Ca²⁺/Mg²⁺, Thermo Fisher Scientific) supplemented with 5% FBS and 1% Antibiotic-Antimycotic (Anti-Anti) (Thermo Fisher Scientific). To detach mucosa from the epithelial layer, sections were incubated in 1 mM DTT in HBSS supplemented with 5% FBS for 20 min at 37°C with shaking (120 RPM). Supernatants were removed and tissues were vortexed briefly and then placed in 5% FBS/1% Anti-Anti HBSS containing 300 U/mL collagenase-II (Worthington Biochemical) for 40 min at 37°C with shaking. Cells were pelleted, resuspended in supplemented HBSS, followed by gradient centrifugation with Lympholyte (Cedarlane) to collect the mononuclear fraction. Cells were then either plated in complete RPMI (10% FBS) for ex vivo experiments and treated as described or lysed in TRIzol reagent (Thermo Fisher Scientific) for RNA isolation. RNA was then isolated with the Zymo Direct-zol RNA miniprep kit, according to manufacturer’s protocol.

Total RNA was reverse transcribed with the iScript cDNA Synthesis kit (Bio Rad), according to manufacturer’s guidelines. To determine relative gene expression, quantitative PCR (qPCR) was performed via commercially available Taqman Assay (Thermo Fisher Scientific) primers for rat Il23a (Rn00590334_g1), rat Il17a (Rn01757168_m1), Il17f (Rn01757244_m1), Il1a (Rn00566700_m1), Il1b (Rn00580432_m1), Il22(Rn01766097_m1), with rat Actb (Rn00667869_m1) as the reference gene. The qPCR was performed using an QuantStudio 6 Flex.

Immune signature analysis as previously described (20) with some modifications. Homologous gene signature IDs were translated from rat to human using the UPPER function in excel. Duplicate gene IDs were then removed, with the duplicate with the greatest average expression (RPKM) kept. The ImSig program was then performed on the RNA-seq transcriptome for each animal. A conservative correlation threshold (r = 0.7) was used for feature selection and immune signature analysis (ImSig R package, version 1.0.0).

Luminal DNA was isolated from cecum and colon using DNeasy Kit (Qiagen, Valencia, CA) according to manufacturer’s instructions. 16S rRNA genes were amplified using primers as specified by the EMP sequenced on Illumina MiSeq. The greengenes database was used for alignment. Data were quality-filtered using quantitative insights into microbial ecology 2 (QIIME2) and taxa were summarized to kingdom, phylum, class, order, family, genus, and the species level (Phylogenetic Level 7). QIIME2 was used to visualize PCoA plots for beta diversity. Differential abundance was assessed for magnitude and significance using linear discriminant analysis (LDA) effect size (LefSe) analysis. We compared relative abundance of the gut microbes (|LDA|>2, p <0.05) in rapamycin-treated with their wild-type and vehicle-treated controls in the cecal and colon lumen.

Host transcriptome cecum and colon gene expression data was correlated with 16S rRNA sequencing of the respective colon and cecum lumen, as previously reported (21) with slight modifications. Pearson’s correlation coefficient (r) was calculated and significant values (p <0.001) were subjected to hierarchical clustering. Microbes correlating with at least 5% of the total transcripts and transcripts that correlated with least of 4 microbes (2.5% of total) are shown.

One-way parametric ANOVA in GraphPad Prism Version 10.3.1. (LaJolla, CA) was used to assess differences between means, with p < 0.05 considered statistically significant (*). Data was assessed for normality (Kolmogorov-Smirnov’s test) prior to using a one-way parametric ANOVA (Tukey’s multiple comparisons). For non-normally distributed data, a Kruskal Wallis test was used (Dunn’s multiple comparisons). For transcriptome analysis, differentially gene expression was evaluated using ANOVA on log-2 transformed RPKM data in Partek Genomics Suite as described in the method section above. Unless otherwise stated, the figures show three independent experiments. Simple linear regressions were performed to assess linear correlation in GraphPad prism, with the 95% confidence interval shown, or in Partek Genomics Suite (Pearson’s correlation) for the inter-omics analysis as described in the methods section above.

Gut inflammation is a major component of experimental SpA in high copy number B27-Tg rats. In our colony, inflammation generally begins at 8-weeks of age and worsens over time (8). To assess the effect of rapamycin on gut inflammation, 11 to 22-week-old B27-Tg rats and age matched WT animals were treated for 5 weeks with rapamycin or vehicle as described (5) (Figure 1A). Before (Day 0) and during the rapamycin treatment stool samples were collected and scored. B27-Tg rats exhibited elevated stool scores before the start of treatment, consistent with the presence of gut inflammation (Figure 1B). Stool scores were significantly reduced compared to vehicle-treated rats after two weeks of rapamycin treatment, and remained normal throughout the treatment period (Figure 1B). Rapamycin did not alter the stool scores of WT rats.

After 5 weeks of treatment, tissue samples were collected, and gut inflammation was assessed by histological scoring of the colon (Figures 1C, D) and cecum (Supplementary Figures 1A, B). Increased histological scores in vehicle-treated B27-Tg rats reflect loss of goblet cells, the presence of inflammatory cell infiltrates, and the relative amount (percent) of inflamed tissue (Supplementary Figures 1C, D). Rapamycin treatment reduced inflammation in B27-Tg rats significantly, consistent with the decreased stool scores (Figures 1C, D; Supplementary Figures 1C, D).

Together these data confirm that rapamycin treatment reduces gut inflammation in HLA-B27 Tg animals.

Next, we analyzed the effect of rapamycin on the gut transcriptome. Distal colon tissue collected at the experimental endpoint was analyzed for differences in gene expression via RNA-seq. Principal Component Analysis (PCA) of the transcriptome showed that samples from the B27-Tg vehicle-treated group (red) clustered distinctly from the vehicle-treated WT group (dark blue), with greater intra-group variability observed for B27-Tg animals (Figure 2A). Rapamycin treatment of B27-Tg rats (green) restored much of the transcriptome, driving the cluster to overlap substantially with the WT groups (dark and light blue), including a reduction in intra-group variation (Figure 2A; top inset). The colon transcriptome of WT animals was marginally affected by rapamycin treatment, indicating that in the absence of inflammation the effect of rapamycin on colon gene expression is minimal (Figure 2A; bottom inset).

Volcano plots were generated to visualize differentially expressed genes. Samples from vehicle-treated B27-Tg rat colon showed the highest level of differentially expressed genes when compared to the corresponding WT controls (Figure 2B, left). Rapamycin treatment of B27-Tg rats dramatically reduced the number of differentially expressed genes compared to vehicle-treated WT rats (Figure 2B, second panel) or compared to rapamycin-treated WT animals (Supplementary Figure 2A, left panel). For comparison, differentially expressed genes in rapamycin-treated B27-Tg rats compared to vehicle-treated B27-Tg rats are shown in Supplementary Figure 2A (right panel). It should be noted that rapamycin treatment of WT rats resulted in minimal gene expression differences compared to vehicle-treated controls (Figure 2B, third panel). The number of differentially expressed genes for each comparison and the overlap between groups is shown in the Euler diagram (Figure 2C), demonstrating that rapamycin is almost exclusively affecting gene expression in the inflamed B27-Tg colon tissue.

Further analysis of individual genes whose expression is reduced by rapamycin treatment shows near normalization of pro-inflammatory cytokines including Il17a, Il17f, Il22, Tnf, Il1a, Il1b, and Il23a, as well as other markers of gut inflammation, and Nfkbiz, an IkappaB family member that regulates the NF-κB pathway (Figure 2D). Analysis of the downregulated gene set (Supplementary Figure 2A, right panel, blue dots) revealed pro-inflammatory pathways, such as ‘inflammatory response’, ‘Th17 cell differentiation’, ‘T cell receptor signaling’, and ‘NF-kB and TNF-signaling’ (Supplementary Figure 2B). Immune signature analysis of the colon transcriptome revealed correlations between the relative abundance of T cells, B cells, NK cells, and macrophages, and histology scores in vehicle-treated B27-Tg rats and showed that rapamycin treatment reduced the relative abundance of each cell population except macrophages (Supplementary Figure 2C). These results correspond with the decrease in inflammatory infiltrate observed in rapamycin-treated B27-Tg animals compared to their corresponding vehicle-treated B27-Tg controls. Together, these data suggest that rapamycin treatment led to a major reduction in HLA-B27-mediated inflammation with reduced pro-inflammatory gene expression, without affecting the transcriptome of healthy WT animals.

To look for direct effects of rapamycin on cytokine expression, bulk immune cells were isolated from the colons of 5–6-month-old untreated B27-Tg and WT rats. Cytokine expression was measured using qPCR, and significantly increased expression of Il17a, Il17f, Ifng, Il22, Il1a, Il1b, Il23a and Tnf was found (Figure 3A), similar to whole tissue transcriptome results. Next, we treated isolated bulk immune cells from B27-Tg rats with 50 nM rapamycin or vehicle for 22 hours and measured expression of these cytokines. Rapamycin significantly reduced the expression of Il17a, Il17f, and Ifng (Figure 3B). Differences in expression of Il23a, Il1a, Il1b, and Tnf transcripts were not observed, and while Il22 expression was lower in all treated samples, the difference was not statistically significant. These results confirm the strong effects of rapamycin on Th17 T cell cytokine expression (22) and support important roles for these cytokines in vivo in HLA-B27-mediated gut inflammation.

We previously reported dysbiosis that correlated with inflammation in the colon and cecum of B27-Tg rats (8). To determine whether rapamycin affects HLA-B27-associated gut dysbiosis we analyzed the microbial composition of the cecum lumen using 16S rRNA sequencing. Principal Coordinate Analysis (PCoA) revealed that after rapamycin treatment, B27-Tg cecal lumen microbiota were more similar to rapamycin-treated WT rats (Figure 4A; Supplementary Figure 3A; dark green vs. light blue symbols) than vehicle-treated WT animals (Figure 4A dark blue symbols), suggesting independent effects of rapamycin on gut microbiota (i.e. effects not only secondary to the resolution of inflammation). This was confirmed by a direct comparison of the rapamycin-treated WT rats with WT vehicle-treated animals (Figure 4A; Supplementary Figure 3A; light blue vs. dark blue symbols). This indicates that not only does rapamycin treatment restore many of the changes associated with dysbiosis in B27-Tg rats, but it also alters the gut microbiome in the absence of inflammation (Supplementary Figure 4).

We next examined the effect of rapamycin on cecal microbes at the genus or family level (Figures 4B, C). The 14 microbes with the highest relative frequency in each group of rats are shown in Figure 4B. There is a notable increase in Blautia in vehicle-treated B27-Tg rats, which correlates positively with cecum histology scores (Figure 4C), confirming our previous findings (8). Other microbes showing elevated abundance and correlating with histology scores include Ruminococcaceae, Marvinbryantia and Prevotellaceae (Figure 4C). Rapamycin reduced the relative frequencies of Blautia, Ruminococcaceae, Marvinbryantia, and Prevotellaceae, although the latter was not statistically significant and was quite variable in non-inflamed cecum lumen (Figure 4C). Host transcripts significantly correlated with microbial abundance were subjected to hierarchical clustering, revealing several microbial taxa such as Blautia, Prevotelleaceae, Marvinbryantia, and Ruminococceae, correlating with pro-inflammatory genes such as Il17, Il23, and type II IFN signaling (Supplementary Figure 3B). This inter-omic analysis confirms and extends our previous findings (8).

Rapamycin treatment also significantly reduced the relative frequency of Lachnospiraceae which was modestly increased in inflamed B27-Tg gut (although not statistically significant) compared to the WT controls and correlated weakly with histological scores. Interestingly, Helicobacter was strongly reduced in B27-Tg cecal tissue and correlated negatively with cecal histological scores (Figure 4C). Rapamycin treatment restored Helicobacter in B27-Tg rats to the level seen in WT controls. Lactobacillus was slightly decreased in B27-Tg rats and correlated negatively with histological scores but was more abundant after rapamycin treatment of both WT and B27-Tg rats (Figure 4C). Together these data show rapamycin affects the abundance of microbes in B27-Tg cecum lumen which associate with inflammation, as well as having independent effects on microbial composition.

The RNASeq data are deposited in the NCBI GEO repository, accession number GSE308302↗. Microbiome data are deposited in the NCBI BioProject repository, accession number PRJNA1333912.