Gut Microbiome and Osteoarthritis: Insights From the Naturally Occurring Canine Model of Osteoarthritis

Abstract

Objective

The purpose of this study was to enhance the current knowledge of the relationship between the gut microbiome and osteoarthritis (OA) and associated pain using pet dogs as a clinically relevant translational model.

Methods

Fecal samples were collected from 93 owned pet dogs. Dogs were designated as either clinically healthy or OA pain using validated methods. Metagenomic profiling was performed through shotgun sequencing using the Illumina NovaSeq platform. MetaPhlAn2 and HUMAnN2 were used to evaluate bacterial taxonomic and pathway relative abundance. Comparisons between healthy and OA‐pain groups were performed individually for each taxa using nonparametric tests following Benjamini and Hochberg adjustment for multiple comparisons. Permutation analysis of variance was performed using Bray‐Curtis distance matrices. All downstream analyses were completed in R.

Results

No significant differences between healthy and OA‐pain dogs were observed for alpha and beta diversity. We found 13 taxa with nominally significant (P < 0.05) associations with OA case status, but none of the associations remained significant after adjustment for multiple comparisons. No differences in alpha or beta diversities or the Firmicutes to Bacteroidetes ratio were found regarding pain severity, mobility or activity level, age, or body composition score.

Conclusion

Similar to recent studies in humans, the present study did not demonstrate a significant difference in the fecal microbial communities between dogs with OA pain and healthy control dogs. Future research in this naturally occurring model should expand on these data and relate the gut microbiome to gut permeability and circulating proinflammatory and anti‐inflammatory molecules to better understand the influence of the gut microbiome on OA and OA pain.

INTRODUCTION

Our understanding of the mechanisms driving osteoarthritis (OA) remains incomplete. The role of the gut microbiome in OA is a developing area of interest that holds promise for addressing some of these knowledge gaps. ¹ Ankylosing spondylitis and rheumatoid arthritis have been associated with some degree of alteration in the composition of the gut microbiota. ² Conversely, recent evidence suggests that for obesity‐related OA, microbial dysbiosis may not be a primary driving factor, but rather, intestinal permeability and low levels of endotoxemia could be playing a significant role. ³ Both lines of evidence suggest that the intestinal environment plays a key role in OA pathogenesis.

Pet dogs have emerged as a potential high‐fidelity model for human OA because of their shared environment with people, high prevalence of naturally occurring OA, and availability of validated outcome measures for measuring pain and mobility. ⁴ Given these features, the present study aimed to enhance the current knowledge of the relationship between the gut microbiome and OA using pet dogs as a clinically relevant translational model for naturally occurring OA. Our primary focus was on the association among gut microbial composition, OA, and, more specifically, OA‐associated pain.

MATERIALS AND METHODS

Animals

Client‐owned dogs of any sex or breed, greater than one year of age, and weighing ≥10 kg that had been enrolled in clinical research studies at North Carolina State University College of Veterinary Medicine's Translational Research in Pain Program were included. These dogs had been thoroughly evaluated (as described below) and designated as either clinically healthy with no evidence of OA‐associated pain or as having OA‐associated pain. Dogs were required not to have received any analgesics (including nonsteroidal anti‐inflammatories, gabapentin, amantadine, tramadol, polysulfated glycosaminoglycans, or opioids) for four weeks before screening and sample collection. Dogs were excluded if they had any concomitant diseases other than OA. All original studies and sample collection were conducted with informed and written owner consent and Institutional Animal Care and Use Committee approval (approvals 17‐110‐O and 16‐184‐O).

Subject phenotyping

Through a standardized screening process, dogs were phenotyped and designated either clinically healthy (n = 19) or as suffering from OA pain (n = 74) based on general and orthopedic examination, validated Client Reported Outcome Measures (Liverpool Osteoarthritis in Dogs [LOAD] index and the Canine Brief Pain Inventory [CBPI]), ⁴ objectively measured activity, and radiographic evaluation. The CBPI consists of a pain severity score (PSS) and a score relating to how pain interferes with function (pain interference score; PIS). The LOAD questionnaire also incorporates questions that pertain to the amount of exercise the dog receives. If pain was elicited from a joint during orthopedic examination, the severity was graded as previously described. ⁵ The presence of OA was confirmed radiographically. Healthy, non‐OA‐pain dogs were required to have no history of impairment recognized by the owner (LOAD score ≤9) and no abnormalities detected on orthopedic or neurologic examination. OA‐pain dogs were required to have a six‐month history of impaired mobility (LOAD score ≥10) as reported by the owner and at least one appendicular joint with both pain on manipulation (score ≥1) and radiographic evidence of OA.

Fecal sample collection

Fecal samples were obtained from dogs at the time of screening. Samples weighing approximately 1.0 to 5.0 g were collected using gentle digital palpation, immediately frozen, and stored at −80°C until shipment for analysis. Once ready for shipping, each sample was thawed, divided into 0.5‐ to 1.0‐g aliquots in cryovials, and refrozen. Samples were shipped frozen, on dry ice, to Diversigen for microbiome analysis.

Microbiome and statistical analysis

Whole genome shotgun sequencing (WGS) was used for DNA sequencing of the samples. Illumina NovaSeq WGS with a 2×150 bp pair‐end read protocol was used as a platform for sequencing. Standard trimming was completed using trimmomatic. Filtering of the sequences was performed by sliding window technique (sliding window = 4:20, min = 75) and Bowtie2 to filter host reads. Taxonomic profiling of the microbial community was determined by using the MetaPhlAn2 computational tool (https://github.com/biobakery/MetaPhlAn2). Functional profiling of the microbial community was completed by HUMAnN2 software (https://github.com/biobakery/humann) with mapping to customized per‐sample pangenomes and the UniRef90 database (https://www.uniprot.org/help/uniref). Metabolic pathways were determined by mapping to the MetaCyc database (https://metacyc.org). Alpha (within‐sample) diversity measures, including Shannon and observed species number metrics, were estimated using default parameters. Associations of alpha diversities with OA status were examined by using Mann‐Whitney U test. Bray‐Curtis beta diversity was used to calculate distances between microbiome samples at the genus level. Principal coordinate analysis (PCoA) based on the Bray‐Curtis dissimilarity index was used to summarize these results. Random forest classification was used to identify bacterial species most associated with predicting OA versus healthy dogs. Comparisons between healthy and OA‐pain groups were performed individually for each taxon. Secondary analyses were conducted to evaluate whether the gut microbiota was associated with pain severity and function, sex status (ie intact, neutered and spayed dogs), age, and body condition. All comparisons were performed using nonparametric tests (Mann‐Whitney or Kruskall‐Wallis as applicable) following Benjamini and Hochberg adjustment for multiple comparisons. All downstream analyses were done in R.

RESULTS

A total of 93 dogs (74 OA‐pain dogs and 19 healthy dogs) were included. The demographics and breeds for the dogs are detailed in Supplementary Table 1. Groups were significantly different for age (P < 0.0001), with a mean ± SD age of 5.2 ± 3.3 years in the healthy control group and 8.6 ± 2.9 years in the OA‐pain dog group. The groups did not differ based on sex distribution (P = 0.17) or body weight (P = 0.31). The OA‐pain group had slightly, but significantly, higher body condition scores (BCS) (P = 0.01). As expected, the OA‐pain group had significantly higher total joint pain scores (P < 0.0001) and LOAD scores (P < 0.0001) compared to the healthy control group. Mean ± SD joint pain scores were 5.9 ± 3.5 and 0 ± 0, and LOAD scores were 21.9 ± 7.7 and 4.1 ± 2.7 in the OA‐pain and healthy groups, respectively.

OA‐pain versus healthy dogs

No significant differences between healthy and OA‐pain dogs were observed for alpha diversity (P = 0.87) and the Shannon diversity index (P = 0.67). There were no significant differences between groups in the overall distribution of taxa or relative abundances of the 25 taxa with the lowest false discovery rate (FDR)–adjusted P values. PCoA identified no compositional differences between healthy and OA‐pain dogs (Figure 1A). There were also no significant differences in relative abundance of the 25 most abundant taxa or most abundant pathways based on alpha and Shannon diversity between the two groups by sex status or between sexes. Age and BCS had no association with alpha diversity, beta diversity, or Firmicutes to Bacteroidetes ratio (FBR). However, several species were considered significantly different before adjusting P values. We found 13 taxa with nominally significant (P < 0.05) associations with OA case status, but none of the associations remained significant after adjustment for multiple comparisons (Table 1). The taxa that were more represented in dogs with OA were Bacteroides vulgatus, Eubacterium dolichum, Collinsella stercoris, Clostridium ramosum, and Ruminococcus torques. Figure 1C shows the taxa with the highest contribution to the differences between the healthy and OA‐pain dogs, with a rightward shift indicating greater abundance in healthy dogs and a leftward shift indicating greater abundance in OA‐pain dogs. No significant differences between healthy and OA‐pain dogs were observed for FBR. Lachnospiraceae bacterium 6 1 63FAA was present in 100% of the healthy dogs and was not present in any of the dogs with OA pain. Random forest analysis showed that B vulgatus was used as the most important species variable for prediction of the OA‐pain status; however, there was no difference between the groups for B vulgatus abundance (Figure 1B).

Figure 1.

(A) Principal coordinate analysis showing the Bray‐Curtis dissimilarity between the healthy (normal) and OA‐pain dogs. (B) Group comparison of Bacteroides vulgatus abundance; (C) Graph showing the taxa with the highest contribution to the differences between the healthy and OA‐pain dogs, with a rightward shift indicating greater abundance in healthy dogs and a leftward shift greater abundance in OA‐pain dogs. OA, osteoarthritis; PCoA, principal coordinate analysis.

Table 1.

Relative abundance of taxa between healthy control dogs, and dogs with OA‐pain^*

Taxa	P value	FDR‐Adj. P value	Mean of healthy	Mean of OA
Bacteroides vulgatus	0.001	0.27	0.000505263	0.010367568
Eubacterium dolichum	0.002	0.27	0.000884211	0.006756757
Lactobacillus gasseri	0.005	0.27	0.000142105	0
Mycoplasma canis	0.005	0.27	1.00 × 10−4	0
Shigella sonnei	0.006	0.27	1.00 × 10−4	9.46 × 10−6
Catenibacterium mitsuokai	0.006	0.27	0.064152632	0.033074324
Collinsella stercoris	0.007	0.27	5.26 × 10−6	0.007032432
Eubacterium biforme	0.017	0.61	0.029873684	0.026244595
Clostridium ramosum	0.029	0.70	0.000421053	0.002668919
Ruminococcus torques	0.036	0.70	0.005047368	0.008032432
Streptococcus lutetiensis	0.047	0.70	0.021357895	0.018975676
Enterococcus avium	0.050	0.70	3.16 × 10−5	1.76 × 10−5
Lactococcus lactis	0.050	0.70	3.16 × 10−5	6.76 × 10−6

^*Numbers represent relative abundance of taxa. Taxa that are in bold were more represented in dogs with OA. Constructed using MetaPhlAn2, db_v20 (https://bitbucket.org/biobakery/metaphlan2). FDR‐Adj., adjusted false discovery rate; OA, osteoarthritis.

Activity, pain severity, and client‐reported outcome measures

Shannon diversity was significantly lower in healthy dogs compared to OA‐pain dogs that walked approximately 0.6 to 1.2 miles daily. There was a significant difference in beta diversity noted between the two groups when they walked 0 to 0.6 miles (P = 7.95 × 10⁵) and 0.6 to 1.2 miles daily (P = 4.87 × 10¹⁵); however, no significant differences in beta diversity were present between the groups for dogs walking longer distances (1.2‐1.9 and 1.9‐2.5 miles). Looking at the overall distribution of taxa, although no significant differences were noted between abundance of species in healthy and dogs with OA after FDR adjustment based on daily walking distance, several species were significantly different prior adjusting the P values based on the daily walking distance. Erysipelotrichaceae bacterium 21 3 and Bacillus megaterium were more represented in healthy dogs, that walk 0 to 0.6 miles and B vulgatus and R torques were more represented in OA dogs that walk 0 to 0.6 miles compared to dogs without OA. Streptococcus lutetiensis and Clostridium spiroforme were more represented in healthy dogs and B vulgatus and E dolichum more represented in OA dogs that walk 0.6 to 1.2 miles. There were no functional pathway differences in relation to distance walked. Akkermansia was more abundant in dogs with higher physical activity (dogs with better mobility, based on question 1, LOAD Lifestyle) compared to those with lower physical activity, regardless of OA status.

There were no significant differences in alpha or beta diversity between different levels of the PSS or PIS calculated from the CBPI, nor were there any significant associations between the relative abundance of the 26 most abundant taxa or most abundant pathways and PSS or PIS. However, B vulgatus and Lactobacillus acidophilus appeared to be more enriched in dogs with the highest levels of owner‐reported pain. When looking at PIS scores (degree to which pain interferes with function), the only consistent finding was that Clostridium hiranonis, which is the most abundant species among all samples, was either absent or of very low abundance in those samples from dogs with the highest levels of pain interference with function (based on ability to walk, run and climb stairs).

Based on unadjusted P values, there were some pathways that were significantly different between healthy dogs and those with OA pain (Table 2). Most of these pathways were more represented in healthy dogs compared to dogs with OA pain, with some exceptions (PWY‐7221: guanosine ribonucleotides de novo biosynthesis; COMPLETE‐ARO‐PWY: superpathway of aromatic amino acid biosynthesis; PWY‐7219: adenosine ribonucleotides de novo biosynthesis; PWY‐5103: L‐isoleucine biosynthesis III; and PWY66‐399: gluconeogenesis III). No significant differences between healthy and OA‐pain dogs were observed for complete blood cell count, chemistry panel, or urinalysis values when controlling for FDR.

Table 2.

Functional pathway abundances in the healthy control dogs and dogs with OA pain^*

Pathway	P value	FDR‐Adj. P value	Mean of healthy	Mean of OA
PWY‐5367: petroselinate biosynthesis	0.001	0.34	1.26 × 10−3	4.81 × 10−4
PWY‐7234: inosine‐5'‐phosphate biosynthesis III	0.003	0.34	3.51 × 10−3	1.92 × 10−3
PWY‐6859: all‐trans‐farnesol biosynthesis	0.009	0.34	1.58 × 10−3	9.69 × 10−4
PWY‐6123: inosine‐5'‐phosphate biosynthesis I	0.010	0.34	4.50 × 10−3	2.89 × 10−3
PWY‐6606: guanosine nucleotides degradation II	0.014	0.34	9.26 × 10−4	4.74 × 10−4
PWY‐6124: inosine‐5'‐phosphate biosynthesis II	0.015	0.34	3.97 × 10−3	2.55 × 10−3
PWY‐5030: L‐histidine degradation III	0.015	0.34	4.21 × 10−5	1.32 × 10−4
PWY‐6270: isoprene biosynthesis I	0.016	0.34	3.75 × 10−3	2.12 × 10−3
ANAEROFRUCAT‐PWY: homolactic fermentation	0.018	0.34	3.57 × 10−3	2.07 × 10−3
PRPP‐PWY: superpathway of histidine, purine, and pyrimidine biosynthesis	0.021	0.34	3.20 × 10−3	2.11 × 10−3
PWY‐7221: guanosine ribonucleotides de novo biosynthesis	0.022	0.34	1.42 × 10−2	1.72 × 10−2
PWY‐7039: phosphatidate metabolism, as a signaling molecule	0.024	0.34	3.53 × 10−4	5.27 × 10−5
PWY‐7560: methylerythritol phosphate pathway II	0.032	0.34	3.54 × 10−3	2.09 × 10−3
COMPLETE‐ARO‐PWY: superpathway of aromatic amino acid biosynthesis	0.033	0.34	1.06 × 10−2	1.29 × 10−2
PWY‐7219: adenosine ribonucleotides de novo biosynthesis	0.035	0.34	2.38 × 10−2	2.69 × 10−2
PWY‐5103: L‐isoleucine biosynthesis III	0.035	0.34	8.48 × 10−3	1.04 × 10−2
FERMENTATION‐PWY: mixed acid fermentation	0.038	0.34	1.86 × 10−3	1.23 × 10−3
PWY‐6113: superpathway of mycolate biosynthesis	0.038	0.34	6.26 × 10−4	3.65 × 10−4
PWY‐6612: superpathway of tetrahydrofolate biosynthesis	0.039	0.34	1.49 × 10−3	1.09 × 10−3
PWY‐5173: superpathway of acetyl‐CoA biosynthesis	0.041	0.34	8.16 × 10−4	7.72 × 10−4
PWY‐2941: L‐lysine biosynthesis II	0.041	0.34	9.83 × 10−3	7.92 × 10−3
PWY66‐399: gluconeogenesis III	0.043	0.34	1.05 × 10−5	6.62 × 10−5
PWY‐841: superpathway of purine nucleotides de novo biosynthesis I	0.045	0.34	5.09 × 10−3	3.74 × 10−3
SER‐GLYSYN‐PWY: superpathway of L‐serine and glycine biosynthesis I	0.047	0.34	3.16 × 10−3	2.05 × 10−3

^*Constructed using HUMAnN2 (https://bitbucket.org/biobakery/humann2/wiki/Home) and MetaCyc (https://metacyc.org). FDR‐Adj., adjusted false discovery rate; OA, osteoarthritis.

DISCUSSION

Overall, the present study did not demonstrate a significant difference in the fecal microbial communities between dogs with OA pain and healthy control dogs. This is in contrast to work by Cintio et al, ⁶ which compared the fecal microbiomes of 14 dogs with hip and elbow OA to those of 13 healthy control dogs. In that study, arthritic dogs exhibited elevated relative abundances of the Megamonas genus and reduced relative abundances of the Paraprevotellaceae, Porphyromonadaceae, and Mogibacteriaceae families when compared to healthy controls. The present study did find an increase in the genus Akkermansia in more active dogs, but this was true regardless of OA status. Given the reported association of Akkermansia abundance with obesity‐related OA, ³ , ⁷ this may be worth exploring further in future studies. OA dogs did initially show a higher abundance of B vulgatus, which is a propionate‐producing bacterial species that has been shown to modulate intestinal inflammation, the direction of which depends on the specific strain. ⁸ , ⁹ This difference did not remain significant after adjusting for multiple comparisons; however, it is worth further investigation because it suggests that intestinal inflammation may be linked to OA.

There is evidence in the human literature of an association between gut microbial composition and OA pain. Boer et al ¹⁰ reported a significant association between Streptococcus species abundance in stool, OA‐related knee pain, and knee inflammation. Yu et al ¹¹ generated data suggesting that the Methanobacteriaceae family, Desulfovibrionales order, and Ruminiclostridium genus played a causal role in the development of OA and that several bacterial taxa had a protective role. However, such findings have not always been repeatable, and the results of different studies in the field vary widely (see the review by Liu et al ¹ ). Conversely, a recent study in humans by members of our group found no significant differences between the fecal microbiome of healthy patients and those with OA. ³ That study had a similar number of participants as the present study, although their ratio of OA to control patients to those with OA was closer to 1. Similar to our study, patients with OA were significantly older and had a significantly higher body mass index (comparable to BCS in our study). As such, our results in a naturally occurring model of OA support previous findings in humans that it may not be a dysbiotic microbiota that contributes to the development of OA, but this does not negate the potential role of intestinal permeability in the development of OA. ³

Although specific intestinal microbial profiles may not always distinguish between OA and non‐OA individuals, subtle changes in the microbiome may alter gut microbe or microbial product translocation. Elevated levels of lipopolysaccharide (LPS) and LPS‐binding protein (LBP), two markers that are frequently used as a reflection of decreased intestinal barrier function, are associated with increased OA severity in humans. ¹² In ongoing unpublished work, we found LBP, but not LPS, was positively (and significantly) associated with the number of joints with OA pain in dogs (n = 20). Furthermore, it has also been demonstrated that mice that received a fecal transplant from a metabolically compromised human had a greater acceleration of histologic OA severity, increased proinflammatory plasma biomarkers and LPS, decreased messenger RNA expression of tight junction proteins in the colon, and evidence of bacterial translocation through the gut endothelium visualized using fluorescent in situ hybridization. ¹²

Although not significant after adjusting for multiple comparisons, there are notable patterns within the top functional pathways identified through HUMAnN2 analysis. For instance, all three inosine‐5′‐phosphate biosynthesis pathways are represented in the top 10 pathways, and all are enriched in healthy dogs compared to OA‐pain dogs. Previous research has shown that adenosine and its metabolite inosine exert both (mainly) anti‐inflammatory effects on human synovial cells in vitro. ¹³ The adenosine ribonucleotides de novo biosynthesis pathway was enriched in OA dogs, but the difference between the groups was quite small compared to the inosine pathways. Further investigation is warranted to test whether inosine‐producing bacteria are in fact protective against OA pain.

The present study had several limitations. As expected, dogs with OA pain were significantly older than dogs without OA pain. In future studies, age‐matched controls should be sought, as difficult as that is. Dogs were not stratified by breed across groups, leaving breed as a potential confounding variable. Previous research has suggested potential breed‐related differences in gut microbial diversity and relative abundance ¹⁴ ; however, contradicting evidence has also been reported. ¹⁵ Additionally, the sample size was relatively small, with an imbalance between the healthy group and OA‐pain group. Lastly, this study did not measure markers of intestinal permeability such as proinflammatory cytokines or LPS. This is an area that should be investigated further in pet dogs, especially considering their use as a translational model for human OA.

Although overall the results were “negative,” the data provide valuable further insight on the question of whether the microbiome differs between subjects with painful OA and healthy controls and, therefore, the role of dysbiosis in the etiology of OA. The results highlight the need for further exploration in this area.

AUTHOR CONTRIBUTION

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr Lascelles had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design

Carter, Enomoto, Lascelles.

Acquisition of data

Enomoto, Lascelles.

Analysis and interpretation of data

Stevens, Norris, Arbeeva, Enomoto, Nelson, Lascelles.

Supporting information

Disclosure form

Supplementary Table 1: Demographic characteristics of study subjects.