Osteoarthritis and the gut microbiome in dogs.

The claim, and why it travelled so far.

Osteoarthritis is the joint condition every owner of an older dog will recognise. Cartilage thins, bone remodels at the joint margins, the synovium becomes inflamed, and the dog slows down. Dunn and Jeffries (2022) describe it as an unrecognised pandemic in the human literature, the leading cause of chronic disability in many populations, and the picture in dogs is similar. By the time most owners notice the stiffness, the joint changes are already well established.

For most of the twentieth century, osteoarthritis was framed as a wear-and-tear problem. Mechanical load and age were the explanations, and the cellular biology that the joint did in response was treated as downstream of that loading. The framing shifted in the past fifteen years, as it became clear that low-grade systemic inflammation has a causal role rather than a reactive one (Guido et al. 2021). Cytokines, adipokines, and microbial products were all proposed as the upstream drivers. Among those, the gut microbiome attracted the loudest attention.

The hypothesis is intuitive. The gut contains the largest density of bacteria in the body. Those bacteria produce metabolites, and some of those metabolites are inflammatory. The intestinal barrier is one tight-junction protein layer thick. If the barrier becomes permeable, microbial products leak into the portal circulation, reach the joints, and trigger the chondrocyte and synoviocyte responses that drive cartilage breakdown. Guido and colleagues (2021), in a systematic review of the mechanistic literature, set out this gut-joint axis in detail: zonulin upregulation, tight-junction disassembly, raised plasma lipopolysaccharide, Toll-like receptor 4 activation on chondrocytes, and matrix metalloproteinase secretion that degrades cartilage matrix. The mechanism is plausible, and there are good experimental data behind each step in isolation.

The strong version of the hypothesis is that the gut bacterial community itself is altered in osteoarthritis, and that the altered community is part of what drives the disease. This is the claim that has been tested most often, and it is the claim that has had the most trouble replicating.

What Stevens 2024 found, in 93 dogs.

The largest published canine osteoarthritis microbiome study to date is by Stevens and colleagues (2024), published in Arthritis & Rheumatology. They recruited 93 client-owned pet dogs from a translational pain research program at North Carolina State University. Seventy-four had clinical osteoarthritis pain, validated by the Liverpool Osteoarthritis in Dogs questionnaire, Canine Brief Pain Inventory, orthopaedic examination, and radiographic confirmation. Nineteen were clinically healthy controls with no history of impairment and no orthopaedic or neurological abnormalities.

Faecal samples were sequenced with whole-genome shotgun metagenomics on Illumina NovaSeq, profiled with MetaPhlAn2 and HUMAnN2, and analysed in R. The osteoarthritis-pain group was significantly older than the healthy group (8.6 ± 2.9 versus 5.2 ± 3.3 years; P < 0.0001) and had higher body condition scores (P = 0.01). Pain scores and Liverpool Osteoarthritis in Dogs scores were, as expected, much higher in the osteoarthritis group.

The microbiome result was clean and negative. Alpha diversity did not differ between groups: observed species P = 0.87, Shannon P = 0.67. Principal coordinate analysis on Bray-Curtis dissimilarity revealed no compositional separation between healthy and arthritic dogs. The Firmicutes-to-Bacteroidetes ratio, frequently invoked as a coarse dysbiosis index, was not different between groups. Thirteen taxa were nominally significant at P < 0.05, but none of those associations remained significant after Benjamini-Hochberg correction for multiple comparisons. Random forest classification identified Bacteroides vulgatus as the top species-level predictor of osteoarthritis status, but the relative abundance distributions overlapped so heavily that the unadjusted comparison itself only reached P = 0.001 (false discovery rate-adjusted P = 0.27).

There were two findings in Stevens 2024 worth holding onto. The first is that Akkermansia was more abundant in dogs with higher physical activity, regardless of whether those dogs had osteoarthritis. This is consistent with the human literature linking Akkermansia to lean body composition and metabolic health, but it is an activity finding, not an osteoarthritis finding. The second is that Clostridium hiranonis, the most abundant species across all samples, was either absent or at very low abundance in the dogs with the most severe pain-related interference with function. That observation did not survive the multiple-comparisons gauntlet either, but the magnitude of the shift is interesting enough to flag for replication.

The authors are explicit about the implication. The dogs in their study, like the human osteoarthritis patients in the parallel Loeser 2022 cohort, did not show a dysbiotic gut community when measured rigorously. The investigators conclude that intestinal permeability and circulating microbial products, rather than bulk community composition, are the more likely contributors to the inflammatory environment that supports osteoarthritis progression.

What Balouei 2025 found, in 175 Beagles.

Balouei and colleagues (2025), publishing in Animals, looked at a different and in some ways more controlled population. They sequenced faecal samples from 175 Beagles housed in a single colony, providing the kind of diet and environmental homogeneity that pet-dog cohorts cannot offer. The dogs were stratified into three age classes: Junior (20-46 months, n = 43), Adult (47-92 months, n = 58), and Senior (93-168 months, n = 74). A subset of 136 dogs was then split into 69 healthy and 81 with osteoarthritis for the disease-status comparison.

Age had a measurable effect on the microbiome. Senior dogs had significantly lower Shannon and Chao1 alpha diversity than younger animals. Linear discriminant analysis identified taxa that differed by age class: Blautia, Erysipelotrichaceae, and Clostridium enriched in Junior dogs; Prevotella, Streptococcus, and Ruminococcaceae in Adults; Prevotella and Ruminococcus in Seniors. These age-related signals are real and replicate the directional findings reported in other canine ageing studies.

The osteoarthritis comparison delivered a different verdict. There were no significant differences in alpha or beta diversity between healthy and arthritic dogs, controlling for age. Inter-individual variability was high. The authors state, plainly, that aging and osteoarthritis do not induce significant shifts in microbial beta diversity in this colony. The taxon-level analysis did identify some signal. Peptococcus and Peptostreptococcus were enriched in osteoarthritic Adult dogs; Clostridiaceae and Coprobacillus in osteoarthritic Senior dogs. Both genera in the Adult comparison are proteolytic, consistent with the kind of fermentation shift Guido and colleagues (2021) implicated in barrier dysfunction. But the magnitude was small, and it was not visible above the noise of overall community variation.

Two studies, two methods, two populations, and the same headline result: when you control for age and correct for multiple comparisons, there is no clean compositional signal that distinguishes osteoarthritic dogs from healthy ones at the level of bulk diversity. The taxon-level signals that do appear are small, biologically plausible, and not yet replicated across cohorts.

The Cintio 2020 outlier, and why size matters.

The exception in the canine literature is the often-cited 2020 paper by Cintio and colleagues, also published in Veterinary Sciences. They compared 14 dogs with hip and elbow osteoarthritis to 13 healthy controls. Both groups were fed a semi-moist diet supplemented with omega-3 fatty acids. Faecal samples were taken at baseline and after 45 days; blood biochemistry, faecal short-chain fatty acids, and 16S faecal sequencing were performed.

The Cintio findings are taxon-level and directionally informative. Plasma C-reactive protein was elevated in arthritic dogs (0.41 versus 0.15 mg/dL at baseline; 0.33 versus 0.12 at day 45, p < 0.05), consistent with low-grade systemic inflammation. Linear discriminant effect-size analysis identified Megamonas as approximately 2-fold higher in arthritic dogs, while families Paraprevotellaceae, Porphyromonadaceae, and Mogibacteriaceae, and several genera within them, were higher in healthy dogs. The molar proportion of butyrate among faecal short-chain fatty acids was slightly but significantly higher in arthritic dogs (7.7% versus 7.0%, p < 0.05), an unexpected direction given butyrate's usual association with gut health.

Cintio 2020 is the paper that the gut-joint axis literature in dogs leans on most heavily. Stevens 2024 cites it directly as the work their negative result contrasts with. The issue is sample size. Cintio's n = 27 is small enough that several taxon-level associations would be expected to reach P < 0.05 by chance under any reasonable comparison structure, and the paper does not report a false discovery rate correction. In the larger Stevens cohort, where 13 taxa cleared the same unadjusted threshold and zero survived correction, the Cintio results would also have been reduced or eliminated by the same statistical procedure.

This is not a criticism of the Cintio team. The 2020 paper was an early-stage exploratory study and was framed as such by its authors, who note that their results require further investigation in larger cohorts. The point is that the gut-joint axis hypothesis in dogs was built on a small early signal that two larger and more recent studies, using different populations and different sequencing strategies, did not reproduce at the level of bulk community composition.

The signal that does replicate: serum LPS.

If the dysbiosis story is fragile, the gut-joint axis hypothesis is not dead. It is more interesting now than it was, because the level of analysis has shifted from community composition to microbial product. The most informative single piece of evidence in the recent literature comes from a parallel human cohort, by Loeser and colleagues (2022) in Arthritis & Rheumatology, the same group that ran Stevens 2024.

Loeser and colleagues recruited 92 obese adults from the Johnston County Osteoarthritis Project. Fifty had both hand and knee osteoarthritis (Kellgren-Lawrence grade 2 or higher); 42 had no hand osteoarthritis and Kellgren-Lawrence grade 0 or 1 knees. Stool was sequenced by 16S rRNA, and blood was analysed for 80 cytokines, lipopolysaccharide, and lipopolysaccharide-binding protein. To test causality, they also performed faecal transplants from pooled case and pooled control samples into germ-free mice, fed those mice a Western diet for 40 weeks, and scored the histological osteoarthritis that developed.

The microbiome result was negative: no significant difference in the faecal microbial community between obese hand-and-knee osteoarthritis cases and obese controls. Of 80 cytokines tested, only osteopontin, a pro-inflammatory cytokine, was elevated in cases (24,381 versus 16,620 relative fluorescence units, p = 0.01). The result that mattered was the lipopolysaccharide measurement. Serum lipopolysaccharide was 104.9 ± 45.8 endotoxin units per millilitre in cases and 61.3 ± 33.9 in controls, p < 0.0001. That is roughly a 1.7-fold increase, with a tight enough distribution that the difference is unmistakable.

And the faecal-transplant arm was the cleanest causal test. Germ-free mice colonised with case microbiota and fed the Western diet showed beta-diversity separation from controls and 15 differentially abundant taxa at baseline. But there were no differences in any histological osteoarthritis measure between the case-transplant and control-transplant mice after 40 weeks. The case microbiota did not transfer the disease. Whatever was elevating serum lipopolysaccharide in the osteoarthritis patients was not encoded in the bulk transferable community.

The implication is structural. A faecal microbiome that looks compositionally similar to a healthy one can still be leaking more lipopolysaccharide into the systemic circulation. The variable that matters is not the bacteria; it is the barrier between the bacteria and the bloodstream.

The mechanism, in the form that survives the data.

Guido and colleagues (2021) reviewed the mechanistic literature on gut permeability and osteoarthritis across 21 included studies, most of them in rodent models. The pieces of the mechanism are well characterised in isolation, and the chain runs as follows.

Diet, age, obesity, and antibiotic exposure can each upregulate zonulin, a host-derived protein that reversibly disassembles intestinal tight junctions. Tight junction disassembly increases paracellular permeability. Microbial products from the gut lumen, lipopolysaccharide foremost among them, translocate into the lamina propria and from there into the portal circulation. Hepatic clearance handles some of the load; the rest reaches the systemic circulation. Once in the joint, lipopolysaccharide engages Toll-like receptor 4 on articular chondrocytes, which mount an innate immune stress response and secrete matrix metalloproteinases. Matrix metalloproteinases degrade the collagen and proteoglycan scaffolding of articular cartilage. Synoviocytes from osteoarthritis joints similarly upregulate the inflammasome receptors NLRP1 and NLRP3 when exposed to lipopolysaccharide in vitro, completing the inflammatory loop. Guido and colleagues report that one validated study found a positive association between synovial lipopolysaccharide concentration, joint inflammation, and osteoarthritis severity.

That is a mechanism. What it does not require is a wholesale rearrangement of the gut community. A subtle shift in proteolytic species, or no taxonomic shift at all with an increase in barrier permeability, would deliver the same systemic lipopolysaccharide signal. This is consistent with the Stevens, Balouei, and Loeser findings: the community looks compositionally similar to a healthy one, but the downstream product reads differently.

One implication of placing the action at the barrier rather than the community is that the same composition can produce different systemic outputs depending on host state. Diet, obesity, age, and antibiotic exposure all act on barrier integrity through partly overlapping pathways. A dog on a low-fibre, high-fat diet can have a faecal community indistinguishable from a dog on a balanced diet on standard 16S metrics, and still have a measurably higher rate of bacterial product translocation. The host variables that move barrier function are also the ones that move osteoarthritis risk independently. The microbiome variable is not the bacterial census; it is the interface between the bacterial census and the rest of the body.

This is also why the Akkermansia signal from Stevens 2024 is worth a second look. Akkermansia muciniphila is a mucin-degrading specialist that lives in the colonic mucus layer and, in mouse and human studies, supports mucus turnover and tight junction integrity. Its abundance in the Stevens cohort tracked physical activity rather than osteoarthritis status, but the two variables are themselves coupled: less active dogs gain weight, become more sedentary, and the conditions that allow Akkermansia to thrive degrade. Whether Akkermansia abundance is a useful marker for the barrier side of the gut-joint axis in dogs has not been formally tested. It is a reasonable hypothesis to chase with the data that is now being generated at scale.

The human signal that does cross over.

The largest piece of human evidence for a directional bacterial-pain association comes from Boer and colleagues (2019), described in detail by Dunn and Jeffries (2022). Boer evaluated the faecal microbiome of 1,427 knee osteoarthritis patients and controls from the Rotterdam III cohort, with replication in 867 participants from the Lifelines-DEEP study. The headline was that four bacterial clades, all within the lineage class Bacilli, order Lactobacillales, family Streptococcaceae, genus Streptococcus, were associated with knee pain and with magnetic resonance imaging evidence of knee effusion. Guido and colleagues (2021) report that in the same cohort, Streptococcus species accounted for around 20% of the total Firmicutes population in osteoarthritis patients and correlated positively with pain scores and negatively with joint functionality.

The Boer 2019 finding is the closest thing in the field to a replicated, large-cohort, taxon-level association. It is also a pain-and-effusion association, not a structural osteoarthritis-versus-no-osteoarthritis association: an alpha-diversity reduction in the same cohort disappeared after correction for body mass index. The Streptococcus-pain signal in humans has not yet been chased rigorously in dogs.

What the cat data show.

Almost nothing. The published literature on feline osteoarthritis microbiomes is essentially empty. None of the three direct canine studies covered here had a feline arm. The reviews by Dunn and Jeffries (2022) and Guido and colleagues (2021) discuss rodent and human evidence and do not address cats. The Loeser 2022 work is human.

This is a real gap, and not a small one. Feline osteoarthritis is at least as prevalent as the canine form, possibly more so, and it is systematically under-diagnosed because cats present with subtle behavioural changes rather than the obvious lameness that prompts a vet visit in dogs. The pathophysiology is comparable. The intervention options are narrower, given feline drug-metabolism sensitivities. There is a clear case for parallel work in cats, and it has not yet been done at any scale.

For now, the honest framing is that the gut-joint axis literature in cats is theoretical. Anything in a feline-specific microbiome report on this subject is extrapolation from canine and human data. That is a limitation BAARK takes seriously and is positioned to address; the company's combined faecal and oral profiling across the cat reference cohort is one of the few datasets that will be able to investigate the question prospectively as more cats are sequenced.

Diet, weight, and intervention, without the overclaim.

The data on intervention are mixed, and the honest reading is that the strongest evidence is for the interventions that were already well supported on other grounds.

Weight management is the most reliably supported. Body condition score correlates with osteoarthritis severity in the Stevens 2024 cohort, intestinal permeability is increased in obesity in both animal models and humans, and weight loss reduces both mechanical joint loading and the systemic inflammatory signal that obesity drives. Guido and colleagues (2021) note that exercise and weight loss are among the strongest evidence-supported interventions for osteoarthritis pain and progression independently of the microbiome story, and both are also associated with reduced bacterial translocation. The microbiome science does not change the recommendation; it adds to the list of reasons it works.

Exercise on its own, separate from weight loss, also has measurable microbiome effects. Stevens 2024 observed that Akkermansia was higher in dogs walking longer daily distances, and the Guido 2021 review notes that physical activity is associated with reduced gut permeability in rodent models and human cohorts. The mechanism is not fully resolved; bile-acid flux, mucus turnover, and gut motility have all been proposed. The practical implication for an osteoarthritic dog is that the same activity programme that loads the joints in a controlled, range-of-motion-favourable way also reduces the gut-derived inflammatory signal that contributes to joint inflammation. Two separate interventions, one behaviour.

Dietary fibre is biologically reasonable but under-tested in canine osteoarthritis specifically. Fibre fermentation produces short-chain fatty acids, which support colonocyte energetics, intestinal barrier integrity, and a reduction in lipopolysaccharide translocation. The animal-model literature reviewed by Dunn and Jeffries (2022) supports fibre and prebiotic supplementation as protective in diet-induced osteoarthritis rodent models. There are no comparable clinical trials in dogs.

Probiotics are at an earlier stage. Dunn and Jeffries (2022) summarise rodent studies in which Lactobacillus acidophilus, L. rhamnosus, and multi-strain probiotic complexes reduce osteoarthritis pain and histological severity in monosodium iodoacetate models. The single human trial they cite, a 2017 study, reported reductions in pain scores and serum C-reactive protein in human knee osteoarthritis patients after a 60-day course of Lactobacillus casei-supplemented milk. The dose, the strain, and the formulation matter, and the canine probiotic literature in osteoarthritis is essentially absent.

Faecal microbiota transplantation is a much harder argument to make in this context. The Loeser 2022 result was that transplanting osteoarthritis-case microbiota into germ-free mice did not transfer worse disease. That does not rule out a therapeutic application of donor faecal transplant, but it sharply weakens the case for community composition as the operative variable.

Where BAARK fits.

The bottom line.

The gut-joint axis is a real concept with a real mechanism. The version of it that gets repeated most often, that osteoarthritic dogs and cats have a dysbiotic gut community that distinguishes them from healthy ones, does not survive the larger and more carefully controlled studies. Stevens 2024 (n = 93), Balouei 2025 (n = 175), and Loeser 2022 (n = 92, human) all return negative results on diversity. Taxon-level signals exist but are small, inconsistent across cohorts, and not yet replicated.

What does replicate is the lipopolysaccharide signal. Circulating endotoxin is elevated in osteoarthritis patients without a corresponding shift in the bacterial community that produces it. The variable that does the work is intestinal barrier integrity, not bulk community composition. The mechanism downstream, Toll-like receptor 4 on chondrocytes driving matrix metalloproteinase secretion, is well characterised.

The feline side is, for now, a research gap rather than a settled science. BAARK's reports are framed accordingly. Wellness profiling, not a diagnostic test. The microbiome view of osteoarthritis is one input into the conversation with your veterinarian, alongside the orthopaedic examination, the pain scoring, and the radiographs that the vet uses to make the actual clinical decisions.