Feces from axSpA patients and HCs was collected from two university based tertiary hospitals, Seoul Saint Mary’s Hospital and Konkuk University Medical Center (Seoul, South Korea). The inclusion criteria for axSpA patients were as follows: 1) Fulfillment of the 2009 Assessment of SpondyloArthritis International Society classification criteria for axSpA (15), and 2) aged > 19 years old. AxSpA patients with another autoimmune diseases, inflammatory arthritis, active infections, malignancy, or pregnancy were excluded. Baseline demographics, laboratory results, medications, and disease history were collected. Disease-associated parameters such as Ankylosing Spondylitis Disease Activity Index Score with C-reactive protein (ASDAS-CRP) (16), Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) (17), Bath Ankylosing Spondylitis Functional Index (18), and a patient global assessment were recorded when the fecal samples were collected. Peripheral blood samples were also obtained. The present study was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. The experimental protocol was approved by the Institutional Review Boards of Seoul Saint Mary’s Hospital (KC18TESI0627) and Konkuk University Medical Center (KUH 2019–08–015). Written informed consent was obtained from each participant prior to enrollment.

Human peripheral blood mononuclear cells (PBMCs) were collected from heparinized blood by standard Ficoll-Paque density gradient centrifugation (GE Healthcare Biosciences, Uppsala, Sweden). Human PBMCs were incubated with CD4-coated magnetic beads (Miltenyi Biotec, Bergisch Gladbach, Germany) and isolated on MACS separation columns (Miltenyi Biotec). CD4⁺ T cells (5 x 10⁵ cells) were seeded into 48-well plates (Nunc, Rosklide, Denmark). Th17 cell-polarizing conditions, the sorted CD4⁺ T cells were stimulated with plate-bound anti-CD3 (0.5 μg/mL), anti-CD28 (1 μg/mL), anti-IFN-γ (2 μg/mL), anti-IL-4 (2 μg/mL), TGF-β (2 ng/mL), and IL-6 (20 ng/mL) with F. prausnitzii (1 or 10 μg/mL) or butyrate (0.5 or 1 mM) for 72 h to investigate the effect of F. prausnitzii or butyrate on CD4+ T cell differentiation.

Cytokine expression was analyzed via intracellular staining with the following antibodies: Alexa-647-conjugated anti-IL-17 (catalog number 560437) and Alexa-700-conjugated anti-CD4 (catalog number 557922) (BD Biosciences, San Diego, CA, USA). Before staining, cells were stimulated for 4 h with phorbol myristate and ionomycin in the presence of GolgiStop (BD Biosciences, San Diego, CA, USA). To analyze regulatory T cell (Treg) populations, PBMCs were stained with Alexa-700-conjugated anti-CD4, PerCP-Cy5.5-conjugated CD25 (catalog number 560503), and Alexa-647-conjugated FoxP3 (catalog number 561184) (BD Biosciences, San Diego, CA, USA). Th17 and Treg populations were determined using specific antibodies. After surface staining for 30 min, the cells were permeabilized with Cytofix/Cytoperm solution (BD Biosciences, San Diego, CA, USA). Thereafter, the cells were intracellularly stained with fluorescent antibodies. Flow cytometry was performed with the aid of a cytoFLEX Flow Cytometer (Beckman Coulter, Brea, CA, USA), and flow cytometry data were analyzed using FlowJo (Tree Star, Ashland, OR, USA).

Briefly, a 96-well plate (Eppendorf, Hamburg, Germany) was coated with monoclonal antibodies against IL-17A (DY317; R&D systems, Minneapolis, MN, USA), and IL-10 (DY217B; R&D systems) at 4°C overnight. After blocking with a phosphate-buffered saline (PBS)/1% bovine serum albumin/0.05% Tween 20 solution for 2 h at room temperature (22–25°C), the test samples and standards (recombinant IL-10 and IL-17A) were added to the 96-well plate and incubated at room temperature for another 2 h. The plates were washed four times with PBS and Tween 20 and then incubated with biotinylated human monoclonal antibodies against IL-10 and IL-17A for 2 h at room temperature. After washing, streptavidin-alkaline phosphatase-horseradish peroxidase conjugate (Sigma, St Louis, MA, USA) was added to the wells for 30 min, followed by incubation with 1 mg/mL p-nitrophenyl phosphate (Sigma, St Louis, MA, USA) dissolved in diethanolamine (Sigma, St Louis, MA, USA) to develop the color reaction. The reaction was stopped by the addition of 1 M NaOH, and the optical density of each well was measured at 405 nm.

PBMCs obtained from axSpA patients (N = 4) were separated from buffy coats using Ficoll-Hypaque. Red blood cell-free cells were seeded onto 24-well plates at 5 x 10⁵ cells/well and incubated at 37°C for 2 h to separate the floating and adherent cells. The adherent cells were washed with sterile PBS. The preosteoclasts were further cultured in the presences of 10 ng/mL macrophage colony-stimulating factor (M-CSF), 100 ng/mL receptor activator of RANKL (PepproTech, London, UK), and butyrate (0.5 mM) for 4 days to generate osteoclasts. The medium was changed every 2 days and osteoclasts were generated after 8–10 days.

Twelve-week-old male SKG mice weighing 20–25 g at the start of the experiment were purchased from CLEA Japan Inc. (Tokyo, Japan). A maximum of three animals per cage were housed in a room with temperature (20–26°C) and light (12-h light-dark cycle) controlled conditions. The mice had free access to a gamma-ray-sterilized diet (TD 2018S; Harlan Laboratories, Indianapolis, IN, USA) and autoclaved R/O water. All animal research was conducted in accordance with the Laboratory Animals Welfare Act, the Guide for the Care and Use of Laboratory Animals, and the Guidelines and Policies for Rodent Experimentation provided by the Institutional Animal Care and Use Committee (IACUC) at the school of medicine of The Catholic University of Korea (Approval number: CUMS–2020–0312–03). The IACUC and Department of Laboratory Animals at the Catholic University of Korea, Songeui Campus, was accredited by the Korean Excellence Animal Laboratory Facility in accordance with the Korean Food and Drug Administration in 2017 and full accreditation by the Association for Assessment and Accreditation of Laboratory Animal Care International was acquired in 2018.

SKG mice were injected with 3 mg of curdlan (Wako Chemicals, Richmond, USA) dissolved in a 200-μl volume using a 26.5-G needle via intraperitoneal injection and oral administration. After 1 week, animals were equally divided into groups based on the mean values ​​of their arthritis score. Then, 0.9% saline and sodium butyrate (100 mg/kg) were administered orally once a day.

The arthritis score index for disease severity was defined as follows: 0.1, a single site of interphalangeal swelling; 0.5, moderate swelling of the wrist or ankle; and 1.0, severe swelling of the wrist or ankle. The maximum possible score per mouse was 5.8. The scoring was carried out by two independent observers without knowledge of the experimental and control groups.

Tissues were collected from each group 15 weeks after the induction of spondyloarthritis. Tissues were fixed in 10% formalin solution, decalcified using Calci-Clear (National Diagnostics, Atlanta, USA), and embedded in paraffin. Sections of 4- to 5-μm thickness were cut, dewaxed using xylene, dehydrated through an alcohol gradient, and then stained with hematoxylin and eosin (H&E) and safranin O.

Paraffin-embedded sections were incubated at 4°C with the following primary monoclonal antibodies: anti-IL-1β (catalog number NB600-633) (Novus Biologicals, Littleton, CO, USA), anti-IL-17 (catalog number ab91649), and anti-TNF-α (catalog number ab6671) (Abcam, Cambridge, UK). The samples were then incubated with their respective secondary biotinylated antibodies, followed by a 30-min incubation with a streptavidin-peroxidase complex. The reaction product was developed using 3, 3-diaminobenzidine chromogen (Dako, Santa Clara, CA, USA).

The H&E-stained joint samples were scored for inflammation, bone erosion, and cartilage damage. Inflammation was scored according to the following criteria: 0, no inflammation; 1, slight thickening of the lining layer or some infiltrating cells in the underlying layer; 2, slight thickening of the lining layer plus some infiltrating cells in the underlying layer; 3, thickening of the lining layer, an influx of cells in the underlying layer, and the presence of cells in the synovial space; and 4, the synovium highly infiltrated with many inflammatory cells (19). Cartilage damage was evaluated by staining with safranin O and the extent of damage was scored as follows: 0, no destruction; 1, minimal erosion (limited to single spots); 2, slight-to-moderate erosion in a limited area; 3, more extensive erosion; and 4, general destruction. In addition, spinal samples were scored for the condition of the nucleus pulposus (NP) and annulus fibrosus (AF). The NP was scored as follows: 0, normal; 1, slightly condensed; 2, moderately condensed (fibrosis); 3, severely condensed (scattering cells); 4, mild replacement by fibrous cartilaginous tissue; and 5, moderately or severely replaced by fibrous cartilaginous tissue. The AF was scored as follows: 0, normal; 1, serpentine; 2, mildly reversed contour (internally convex); 3, thickened with central ingrowth or rupture; 4, vague borderline with NP; and 5, indistinct.

Cell pellets were prepared from the spleens of spondyloarthritic SKG mice. To examine the population of T helper cells, the cells were stained with PerCP-conjugated anti-CD4 (catalog number 45-0042-82) and APC-conjugated anti-CD25 (catalog number 102012) antibodies (eBioscience; Biolegend, San Diego, CA, USA), and then permeabilized and fixed with CytoFix/CytoPerm (BD Biosciences, San Diego, CA, USA) according to the manufacturer’s instructions. The cells were then further stained with PE-conjugated anti-FoxP3 (catalog number 12-5773-82), APC-conjugated anti-interferon (IFN) (catalog number 505810), and FITC-conjugated anti–IL-17 (catalog number 11-7177-81) (eBioscience; Biolegend, San diego, CA, USA).

Spleen samples were collected from normal SKG mice. Splenocytes were incubated with CD4-coated magnetic beads and isolated using magnet-activated cell sorting separation columns (Miltenyi Biotech, Auburn, CA, USA). To establish Th17 cell-polarizing conditions, the sorted CD4⁺ T cells were stimulated with plate-bound anti-CD3 (0.5 μg/mL), anti-CD28 (1 μg/mL), anti-IFN-γ (2 μg/mL), anti-IL-4 (2 μg/mL), TGF-β (2 ng/mL), and IL-6 (20 ng/mL) for 72 h.

The amounts of IL-17 and TNF-α in culture supernatants from mouse cells were measured by a sandwich ELISA (R&D Systems, Minneapolis, MN, USA). Horseradish peroxidase-avidin was used for color development. The absorbance was determined at a wavelength of 450 nm on an Multiskan SkyHigh Microplate Spectrophotometer (Thermo Fisher, Waltham, MA, USA).

Using feces samples from 33 axSpA patients and 20 HCs, the V3-V4 amplicon sequencing of 16S rRNA data was performed with an Illumina MiSeq reagent kit v3 (2 x 300 bp, Illumina, USA). The PCR primers, i.e., forward (CCTACGGGNGGCWGCAG) and reverse (GACTACHVGGGTATCTAATCC), were designed from the hypervariable regions (V3-V4) of 16S rRNA. PCR was performed using the following conditions: 95°C for 3min, followed by 25 cycles of denaturation at 95°C for 30 sec, primer annealing at 55°C for 30 sec, and extension at 72°C for 30 sec, with a final elongation at 72°C for 5 min. Sequencing libraries were then constructed using a TruSeq^® DNA PCR-Free Sample Preparation Kit (Illumina, USA), TruSeq^® Nextera XT index primer (Illumina), and 2x KAPA HiFi HotStart ReadyMix (Roche, Basal, Switzerland) using the purified PCR products. Subsequently, paired-end reads were generated by sequencing on the MiSeq platform after determining the quality of the library with the Tapestation 4200 platform (Agilent Technologies, Santa Clara, CA, USA) and a Qubit Fluorometer (Thermo Fisher, Waltham, MA, USA). Metagenome analysis data was submitted to National Center for Biotechnology Information BioProject (PRJNA863406).

Microbiomes were identified by the bioinformatics procedures described in Choi et al. (20) After trimming the low-quality sequences and adapter sequences using Cutadapt (21) and Trimmomatic (22), the paired-end reads were assembled using the PEAR (23) tool. An average of 46,560 high-quality reads per sample (median: 29,385 bp; range: 11073–179248 bp) were obtained, in which the average length and quality score were 413.3 bp and 37.24 bp, respectively. The 16S rRNA sequencing data were processed using QIIME2 (ver. 2021.8) (24). Single-end reads for 53 microbiomes from axSpA and HC samples were imported into QIIME2. Sequence quality control and feature table construction were completed using the Divisive Amplicon Denoising Algorithm 2 (DADA2) QIIME2 plugin. The cleaned sequences were then used for the identification of operational taxonomic units (OTUs) with singleton OTUs excluded. Information on the taxonomic hierarchy from phylum to species of each sample was defined by aligning the sequence to the SILVA reference database (release 138) (25) by the pretrained naive Bayes classifier (26) at a 99% minimum similarity level. Only taxa with at least 0.1% relative abundance in each group were used to generate a taxonomic profile graph based on the final OTU table. Linear discriminant analysis effect size (LEfSe) (27) was applied for investigating the compositional information of the microbiomes differentially enriched between axSpA and HC samples.

Using QIIME2 software, the number of observed OTUs and the Chao 1 index were measured to estimate α-diversity, and principal coordinate analysis (PCoA) was used for estimating β-diversity (28). All graphs and diagrams were constructed with the ggplot2 package in R (ver. 4.1.2). The Wilcoxon rank-sum test was used for comparisons between groups. The PICRUSt2 tool was used for predicting the metabolic functions of bacteria (29), where the metabolic functions were estimated by mapping the composition of the identified bacteria into the KEGG database. To identify different metabolic functional abundances between groups, Statistical Analysis for Metagenomic Profiles (STAMP) was used (30) with a Bonferroni correction-adjusted P value < 0.01 as a significance cutoff.

Continuous variables were tested for normality and then analyzed using Student’s t test or the Mann-Whitney U test. The data are expressed as means ± standard deviations (SD) or median with interquartile range (IR). The relative abundance of each taxon was calculated through total sum normalization methods. Nonparametric statistical analyses such as the Mann-Whitney test and Kruskal-Wallis test were performed to compare the relative abundance of taxa and α-diversity of each group using Prism 8 (GraphPad Software, Inc., San Diego, CA, USA). A permutational multivariate analysis of variance (PERMANOVA) with 999 permutations for pairwise tests was carried out to analyze the β-diversity of each group using qiime2-2021.8 and silva 13899 similarity. In all analyses, P < 0.05 indicates statistical significance.

A total of 33 axSpA patients and 20 HCs were included. Thirty-two of 33 axSpA patients were positive for human leukocyte antigen (HLA)-B27 and 51.5% used a TNF inhibitor. Mean age and percentage of males were higher in patients with axSpA than in HCs (P = 0.02 and P = 0.01, respectively). Other information is summarized in Table 1 .

Patients with axSpA had significantly different microbial diversity and composition compared to the HC group. As shown Figure 1 , the number of observed OTUs and Chao1 index measuring α-diversity was smaller in the axSpA group ( Figure 1A ), while PCoA plots for estimating β-diversity indicated that the bacterial communities in axSpA samples were significantly different from those of HCs (P < 0.01) ( Figure 1B , left panel). In addition, the estimation of Bray-Curtis dissimilarity showed that microbial communities showed significantly more diversity between groups than within groups ( Figure 1B , right panel).

We then examined differences in microbial compositions according to taxonomical rank from phylum to species. For instance, at the phylum level, the Bacteroidota and Proteobacteria phyla were abundant in axSpA patients, whereas the Firmicutes phylum was abundant in HCs ( Supplementary Figures 1A, B ). At the family level, Streptococcaceae, Bacteroidaceae, Enterobacteriaceae, and Veillonellaceae were abundant in axSpA patients, whereas Ruminococcaceae, Lachnospiraceae, Coriobacteriaceae, and Erysipelotrichaceae were abundant in HCs ( Supplementary Figures 1C, D ). At the genus level, Bacteroides and Faecalibacterium emerged to be the second most enriched bacteria in axSpA and HCs, respectively ( Figures 1C, D ). At the species level, Bacteroides (g) and Streptococcus (g) were significantly more abundant in axSpA patients than in HCs ( Figures 1E, F ), whereas F. prausnitzii, Blautia (g) was significantly more abundant in HCs than in axSpA patients ( Figures 1E, F ). Among the microbial species that differed significantly between the two groups, only Veillonella (g) showed a positive correlation with disease activity score (ASDAS-CRP) (r = 0.37, P = 0.03) ( Supplementary Figure 2 ).

We then predicted microbiome-driven metabolic functions using the PICRUSt2 tool. We observed that enrichments of several microbial pathways differed significantly between the axSpA and HC groups ( Figure 2 ), suggesting that abnormal changes in microbial composition may be associated with the pathogenesis of axSpA. For instance, several inflammation-associated pathways, including cell motility, secretion, and lipopolysaccharide biosynthesis proteins; glycosyltransferases (31); glycosphingolipid biosynthesis (32); and the citrate cycle (33), were more abundant in axSpA patients than in HCs. Surprisingly, some anti-inflammatory pathways, such as ubiquinone biosynthesis (34) and phosphatidylinositol signaling (35), were also more abundant in the axSpA group. The subgroup analysis between axSpA with or without TNFi showed that pentose/glucuronate interconversions, galactose metabolism, pentose phosphate pathway, and phenylpropanoid biosynthesis were relatively more abundant in axSpA patients who did not use TNFi than axSpA patients who used TNFi ( Supplementary Figure 3 ).

PBMCs and CD4⁺ T cells derived from axSpA patients (N = 4) were cultured under the administration of F. prausnitzii or butyrate with anti-CD3 (0.5 μg/mL). When F. prausnitzii was added to PBMCs of axSpA, decrease of CD4⁺ IL-17A⁺ T cell population (F. prausnitzii 10 μg/mL) and increase of CD4⁺ CD25^high Foxp3⁺ T cell population (F. prausnitzii 1, 10 μg/mL) were observed ( Figure 3A ). Also, the level of IL-17 decreased and levels of IL-10 increased in culture media of PBMC culture ( Figure 3B ). In CD4⁺ T cell purified experiment, the CD4⁺ IL-17A⁺ T cell population was lower in the F. prausnitzii 5 μg/mL treatment group compared to the control, whereas CD4⁺ CD25^high Foxp3⁺ T cells population did not change by F. prausnitzii ( Supplementary Figure 4A ). The levels of IL-17A in culture media were downregulated in the F. prausnitzii 10 μg/mL treatment groups, and IL-10 was increased in the F. prausnitzii 1 μg/mL and 10 μg/mL groups ( Supplementary Figure 4B ). CD4⁺ IL-17A⁺ T cell population decreased by butyrate in PBMC culture ( Figure 4A ). The addition of butyrate 0.5 mM downregulated the CD4⁺ IL-17A⁺ T cell population and CD4⁺ CD25⁺ Foxp3⁺ T cell population in CD4⁺ T cell purified experiment ( Supplementary Figure 5 ). Similarly, TRAP⁺ multinucleated cell counts were significantly lower in the butyrate 0.5 mM + M-CSF + RANKL treatment group than in the M-CSF + RANKL group ( Figure 4B ).

Curdlan-immunized SKG mice were treated with or without butyrate 100 mg/kg/day for 15 weeks (N = 5 for each group). The arthritis score was significantly lower in the butyrate-treated group, but the incidence rate of arthritis was similar between the two groups ( Figure 5A ). On H&E and Safranin- O staining of the hind feet, the degree of inflammation, erosion, and cartilage destruction were diminished in butyrate-treated mice ( Figure 5B ). in addition, NP and AF scores were downregulated in butyrate-treated group ( Figure 5C ). Proinflammatory cytokines (IL-1β, IL-17, TNF-α)-expressing cells were decreased in the butyrate-treated group ( Figure 5D ).

Splenocytes were harvested from the butyrate-treated and non-treated groups at 15 weeks after curdlan injection, then flow cytometry were performed to examine the IL-17A⁺ CD4⁺ and CD25⁺ Foxp3⁺ CD4⁺ T cell populations. The results showed significant decrement of the IL-17A⁺ CD4⁺ T cell population in the butyrate-treated group ( Figure 6A ). The CD25⁺ Foxp3⁺ CD4⁺ T cell population was increased in the butyrate-treated group ( Figure 6B ).

Splenocytes were next obtained from SKG mice not immunized with curdlan and cultured under Th17 polarizing conditions. The in vitro results demonstrated a decrease in the IL-17A⁺ CD4⁺ T cell population and in the levels of IL-17A and TNF-α in culture media in the butyrate-treated group ( Supplementary Figure 6 ).