How we actually measure your pet's gut microbiome.

The clinical problem with how the gut microbiome has been measured.

The veterinary microbiome conversation has a recurring failure mode. A vet runs a faecal culture on a dog with diarrhoea, the laboratory isolates Escherichia coli or Clostridium perfringens, and an antibiotic course is prescribed against the organism named on the report. The intention is correct. The method is not. And the consequences, in a gut community that is already depleted in the bacteria that resist enteropathogen proliferation, can be the opposite of what the antibiotic course was meant to achieve (Suchodolski 2022; Pilla et al. 2020).

Suchodolski's 2022 review in Veterinary Clinical Pathology (Suchodolski 2022) lays out, in careful detail, the methods available for analysing the canine and feline gut microbiome, what each one is able to resolve, and where each one fails. It is, in effect, the methodological starting point for a clinician trying to interpret a microbiome result, or a researcher trying to choose an assay for a new study. The post that follows works through the methods in the order a clinician is likely to encounter them, with the published evidence for and against each, and ends with the assay BAARK actually runs and why.

Why bacterial culture is not a measurement of the gut microbiome.

The dominant feature of the canine and feline gut microbial community, by orders of magnitude, is strict anaerobiosis. The bulk of intestinal bacteria require oxygen-free conditions and specialised growth media that standard veterinary diagnostic laboratories are not equipped to provide. Traditional aerobic culture therefore captures a small, biased subset of the community: the facultative aerobes that grow on standard agar in ambient air. This subset is dominated by organisms such as E. coli, C. perfringens, and various streptococci (Suchodolski 2022).

Because these organisms are isolated from sick animals, they are routinely reported as pathogens. The same organisms are routinely present in healthy animals at comparable abundance. A study cited in the Suchodolski review put the point directly: three aliquots of each of several faecal samples, taken from both healthy dogs and dogs with chronic diarrhoea, were sent to three independent veterinary reference laboratories for culture-based assessment. The three laboratories disagreed with each other on which organisms were present. They reported "dysbiosis" more often in the healthy dogs than in the diarrhoeic dogs. And they isolated haemolytic E. coli more often from the healthy animals (Suchodolski 2022). The molecular Dysbiosis Index, run on the same samples, correctly separated healthy from diarrhoeic dogs.

The clinical implication is straightforward. A faecal culture result that names E. coli or C. perfringens in a diarrhoeic patient does not establish those organisms as the cause of the diarrhoea, and a treatment decision built on that culture report has weak foundations. Aerobic faecal culture has a legitimate role in isolating a small number of frank enteric pathogens for antimicrobial susceptibility testing, where the clinical history justifies it. As an assessment of the gut microbiome, it is unsuited to the question.

16S rRNA gene sequencing: useful for research, limited for reporting.

The technical platform that built the published veterinary gut microbiome literature is amplicon sequencing of the 16S ribosomal RNA gene. The gene is present in all bacteria and archaea, contains nine variable regions interspersed with conserved regions usable for primer binding, and at full length (approximately 1,500 base pairs) carries enough informative positions to distinguish many bacterial species (Johnson et al. 2019). PCR amplification of one or two variable regions, followed by short-read sequencing, allows hundreds to thousands of samples to be processed in parallel at low per-sample cost.

The contribution of 16S sequencing to veterinary microbiome science is not in dispute. It is how we know that culture misses most of the community. It is how we know that chronic enteropathies in dogs are associated with a depletion of Faecalibacterium, Fusobacterium, Blautia, Turicibacter, and Clostridium hiranonis, and an enrichment of Streptococcus and E. coli (Suchodolski 2022; AlShawaqfeh et al. 2017). It is the platform on which the Dysbiosis Index was built. The case against using it as the reporting platform of a clinical microbiome service, however, is a series of well-documented technical limitations.

Primer bias.

The choice of which variable region to amplify (V1-V2, V3-V4, V4 are the common targets) and the exact primer sequences used both bias which bacteria are captured and at what apparent abundance. Two laboratories analysing the same sample with different primer sets produce different bacterial proportions (Suchodolski 2022). This makes cross-study comparison difficult and is one of the reasons the field has struggled to establish reference intervals for 16S relative abundances.

Genus-level resolution as a practical ceiling.

The short-read 16S protocols that dominate published microbiome studies sequence approximately 250 to 450 base pairs of the gene. Johnson and colleagues (2019), in a critical re-evaluation published in Nature Communications, used both in silico and sequence-based experiments to test what 16S variable-region targeting can resolve. Their conclusion: short-read sequencing of 16S sub-regions, including V4, cannot match the taxonomic accuracy of full-length 16S sequencing, and is generally limited to genus-level resolution. Full-length 16S sequencing, available on long-read platforms, can provide species-level resolution where the intragenomic copy variation of the gene is appropriately handled. For a report that names organisms to species (let alone strain), short-read 16S is structurally insufficient.

Relative abundance, not absolute.

A 16S dataset reports the proportion of reads assigned to each taxon, not the absolute count of organisms per gram of faeces. The data are compositional: an increase in one taxon's read share necessarily reduces every other taxon's read share, even where the absolute abundances of the other taxa have not changed. This is the formal reason why apparent "decreases" in 16S abundance can be artefactual, and it is the reason qPCR-based assays, which return absolute concentrations against a standard curve, are more directly comparable across patients and time points (AlShawaqfeh et al. 2017).

Bacteria only, and a partial view of bacteria at that.

The 16S gene is a bacterial and archaeal marker. Fungi are profiled with an entirely different marker, the internal transcribed spacer (ITS) region of the ribosomal operon, in a separate assay. DNA viruses have no universal marker gene. Together, archaea, fungi, and viruses make up approximately 2% of the faecal microbiome in dogs and cats (Suchodolski 2022). For a clinical question that involves Malassezia overgrowth, Candida co-occurrence, or virus carriage, a 16S assay returns no signal at all.

No functional gene content.

16S is a marker gene. It identifies the organism that carries it, but it carries no other genomic content. Antimicrobial resistance genes, virulence-associated genes, bile acid 7α-dehydroxylation genes, and the bacterial pathway genes that produce indoles, p-cresol, trimethylamine, and short-chain fatty acids sit elsewhere in the genome. A 16S dataset can predict functional content statistically using tools such as PICRUSt2, but the prediction is an inference from taxonomic composition against reference genomes, not a direct measurement.

Each of these limitations is acceptable in a research context where the question is well-defined, the analysis pipeline is standardised across samples, and the relative-abundance frame matches the question. None of them is acceptable as the foundation of a clinical-style microbiome report that needs to name organisms, quantify functional capacity, and survive comparison across laboratories.

The Dysbiosis Index: a validated qPCR assay, used inside its scope.

The most useful clinical product to come out of the canine microbiome literature is the quantitative-PCR-based Dysbiosis Index (DI), developed and validated at the Texas A&M Gastrointestinal Laboratory (AlShawaqfeh et al. 2017). The DI does not attempt to characterise the whole community. It uses seven qPCR assays, run against standard curves, to quantify the absolute abundance of seven canine gut bacterial taxa that prior 16S and qPCR work had shown to be consistently altered in chronic enteropathies: Faecalibacterium, Turicibacter, Streptococcus, E. coli, Blautia, Fusobacterium, and Clostridium hiranonis. A trained mathematical algorithm combines the seven absolute abundances into a single index, with reference intervals derived from a 95-dog healthy training set and 106-dog chronic-enteropathy validation set.

The performance numbers from the original paper (AlShawaqfeh et al. 2017) are 74% sensitivity and 95% specificity for separating healthy dogs from dogs with histologically confirmed chronic enteropathy. Subsequent independent studies have shown that the DI is reproducible across samples, sensitive to interventions such as faecal microbiota transplantation and broad-spectrum antibiotic exposure, and analytically validated to a standard that none of the published 16S protocols can match (Suchodolski 2022). A feline version has since been developed and reported (Sung et al. 2022, cited in Suchodolski 2022).

The constraint of the DI is its scope. The assay sees the seven targets it was designed to see. Where the relevant biology is in those seven taxa, the DI catches it. Where the relevant biology is in Akkermansia muciniphila, a specific Prevotella species, a Bifidobacterium, a fungal commensal, a virulence gene, or an antimicrobial resistance cassette, a targeted qPCR panel is silent. This is a feature, not a defect, of a targeted assay: precision and reproducibility come from focusing on a narrow target set. It does mean that a microbiome service whose entire output is a Dysbiosis Index has structurally limited the questions a vet or owner can ask of it.

Shotgun metagenomic NGS: every DNA fragment in the sample.

Shotgun metagenomic next-generation sequencing approaches the problem from the opposite direction. Rather than amplifying a single marker gene before sequencing, shotgun metagenomics extracts total DNA from the sample and sequences every fragment. There is no primer pair to bias the result. There is no fixed marker region to limit resolution. The reads are assigned across all kingdoms simultaneously, so bacterial, archaeal, fungal, and DNA-virus hits all emerge from the same run (Suchodolski 2022).

Several things become possible that 16S cannot deliver.

Species and, with adequate coverage, strain resolution. Reads sample the whole genome rather than a single variable region, so taxonomic classification has many more informative positions to work from. With long-read sequencing chemistries and curated reference databases, species-level assignment is routine and strain-level resolution becomes feasible where coverage supports it. This matters in any context where strain is the clinically informative variable: enterohaemorrhagic versus commensal E. coli, toxigenic versus non-toxigenic C. perfringens.

Cross-kingdom coverage from the same reads. Fungal hits including Malassezia, Candida, and Aspergillus, archaeal hits including methanogenic Methanobrevibacter, and DNA virus hits are all picked up in proportion to their DNA abundance in the sample. No separate ITS workflow is required, and no virus-specific protocol is needed for DNA viruses.

Direct measurement of functional gene content. The reads carry the gene sequences themselves, not just the marker. Antimicrobial resistance gene carriage can be quantified by mapping against curated AMR databases. Virulence-factor gene presence can be screened against pathogen-associated gene catalogues. The metabolic pathway capacity of the community (indole production from tryptophan, p-cresol production from tyrosine, trimethylamine production from carnitine and choline, butyrate production from carbohydrate fermentation, bile-acid 7α-dehydroxylation) can be measured as gene-level read coverage rather than inferred from taxonomy. This is the layer of the report that distinguishes a functional microbiome view from a taxonomic one.

Read provenance that supports re-analysis. A shotgun dataset is, in effect, a permanent biochemical record of the sample. As reference databases improve, as new pathogens or resistance determinants are characterised, and as new functional gene catalogues are released, an existing shotgun dataset can be reclassified without resampling the animal. A 16S amplicon dataset is fixed at the resolution of the primer and the region.

The historical objections to shotgun metagenomics were cost and bioinformatic complexity. Deep short-read shotgun, sequenced to several gigabases per sample on Illumina platforms with full assembly and binning, is genuinely expensive and computationally intensive. Two technical developments have changed the calculation.

The first is shallow shotgun metagenomics, which uses substantially lower per-sample sequencing depth on short-read platforms while still delivering species-level taxonomy and a useable functional gene readout, at a cost only modestly above 16S (Suchodolski 2022). The second is the maturation of Oxford Nanopore sequencing. The R10.4.1 flow cell chemistry, paired with the Q20+ basecalling reagents, raises mean single-read accuracy to approximately 99%, which is sufficient for accurate species-level assignment on full-length 16S amplicons and on metagenomic reads (Zhang et al. 2023). The long-read format also avoids the need for de novo assembly of short reads to recover gene-cassette-length information, and pairs well with classifiers designed for full-length data such as Emu, which uses an expectation-maximisation algorithm to refine species-level abundance estimates against a reference database (Curry et al. 2022).

Methods at a glance.

Methods for analysis of the canine and feline gut microbiome. Adapted and extended from Suchodolski (2022), incorporating long-read 16S evidence from Johnson et al. (2019), Curry et al. (2022), and Zhang et al. (2023).

The star organism: Clostridium hiranonis and the bile-acid story.

The bacterium that has emerged as the most consistently clinically informative marker in canine microbiome science is Clostridium hiranonis. In the healthy canine gut, C. hiranonis is the principal organism responsible for the 7α-dehydroxylation of primary bile acids (cholic and chenodeoxycholic acid) into secondary bile acids (deoxycholic and lithocholic acid) (Suchodolski 2022).

Secondary bile acids are not chemically passive. They are FXR and TGR5 receptor ligands with anti-inflammatory effects on intestinal epithelium. They contribute to maintenance of intestinal barrier integrity. And they inhibit the growth of Clostridioides difficile, C. perfringens, and several pathogenic E. coli strains, both directly and through competitive metabolism of the substrate (Suchodolski 2022; Pilla et al. 2020). When C. hiranonis is depleted, primary bile acids accumulate in the colon, and the conditions that ordinarily suppress enteropathogens are removed. The dysbiotic gut then becomes more permissive for the organisms that the original symptom pattern was attributed to.

The clinical implication is direct, and uncomfortable for routine practice. A broad-spectrum antibiotic course prescribed against an organism named on a faecal culture report can, in a dog whose chronic enteropathy is partly driven by C. hiranonis depletion, accelerate the depletion. The microbiome literature has repeatedly shown long-lasting effects of antibiotics, particularly metronidazole, on the canine faecal microbiome and bile-acid profile (Pilla et al. 2020; Suchodolski 2022). An assay that quantifies C. hiranonis (whether by Dysbiosis Index qPCR or by shotgun metagenomic species-level assignment) and reports the functional bile-acid metabolic capacity of the community can support better treatment decisions than one that does not.

From "who is there" to "what are they doing".

Suchodolski's 2022 review makes the case explicitly: the next phase of clinical microbiome diagnostics is the measurement of function, not just composition. Two complementary platforms support that move.

Metabolomics measures microbial-derived small molecules directly. The relevant pathways in companion animal medicine are now well-characterised. Dietary tryptophan is converted by intestinal bacterial tryptophanases (widespread across E. coli, several Bacteroides, and other taxa) into indole, which is hepatically conjugated to indoxyl sulfate. Indoxyl sulfate accumulates in chronic kidney disease and predicts disease progression in cats and dogs (Suchodolski 2022; and see our gut-kidney axis post for the full evidence). Dietary tyrosine and phenylalanine are converted by gut bacteria into p-cresol, sulfate-conjugated to p-cresyl sulfate, with a comparable nephrotoxic profile. Dietary carnitine and choline are converted via bacterial CutC/CutD into trimethylamine, hepatically oxidised to trimethylamine N-oxide (TMAO), associated with cardiovascular disease in humans and increasingly in companion animals. Carbohydrate fermentation by Faecalibacterium, Roseburia, and several Lachnospiraceae produces the short-chain fatty acids butyrate, propionate, and acetate, which feed colonocytes and modulate systemic inflammation.

Functional metagenomics measures the same biology one step upstream: rather than measuring the metabolite, it measures the gene-encoded capacity of the community to produce it. A shotgun dataset can be queried for tryptophanase gene abundance, CutC/CutD gene abundance, p-cresol biosynthesis pathway gene abundance, bile-acid 7α-dehydroxylation gene abundance (the bai operon), and butyrate kinase / butyryl-CoA:acetate CoA-transferase gene abundance, directly. The community can look unremarkable at the level of bulk taxonomic diversity and still carry a functional gene-content profile that is informative for clinical risk, and the reverse is equally true (Suchodolski 2022). The combination of taxonomic and functional data, both readable from the same shotgun reads, is where the field has been heading for several years.

Where BAARK fits.

What this means for vets ordering microbiome tests.

The practical question for a vet whose patient might benefit from a microbiome assessment is which assay the chosen laboratory actually runs. The questions worth asking, in roughly the order they should be asked:

• Culture or molecular? A culture-based report has the limitations Suchodolski's 2022 review lays out and is not a measurement of the gut community.
• If molecular: 16S, qPCR, or shotgun? 16S is research-grade. qPCR-based Dysbiosis Index is analytically validated within its scope. Shotgun metagenomic NGS reaches further.
• At what taxonomic level does the report name organisms? Short-read 16S generally caps at genus. Long-read 16S and shotgun reach species; shotgun reaches strain with adequate coverage.
• Bacteria only, or cross-kingdom? If the clinical question involves Malassezia, Candida, or DNA virus carriage, 16S will return nothing.
• Does the report include functional gene content? Bile-acid metabolism, indole and p-cresol production, AMR carriage, virulence factor presence: all are gene-level measurements, not inferable from 16S without statistical extrapolation.
• Is the platform validated against a species-matched reference cohort? Reference intervals derived from human data do not transfer cleanly to dogs and cats, and intervals derived from one breed size or age class do not transfer cleanly to others.
• Is the result reproducible if the sample is split? qPCR-based assays generally are. Sequencing-based assays vary by laboratory.

The answers determine whether the test will inform the clinical conversation or merely supplement it with noise. For research-context use, where the question is well-defined and the analysis pipeline is standardised across the study, 16S remains a reasonable choice. For an individual-patient clinical microbiome report, the field has moved on, and shotgun metagenomic NGS with appropriate long-read chemistry and reference databases is the technical platform that supports the questions a clinician is likely to want to ask.

The bottom line.

The gut microbiome is large, mostly anaerobic, and dominated by organisms that standard veterinary diagnostic methods were not built to see. Aerobic bacterial culture captures a small biased subset and is not a measurement of the community. Short-read 16S amplicon sequencing is the platform on which most of the published evidence has been built, and it has well-documented limitations as a clinical reporting platform: primer bias, genus-level resolution, relative-abundance compositionality, bacterial-only coverage, and no functional gene content. The canine Dysbiosis Index is the analytically validated qPCR assay in the space, reproducible and clinically informative within the seven taxa it targets. Shotgun metagenomic NGS reads every microbial DNA fragment in the sample, covers bacteria, fungi, archaea, and DNA viruses, resolves to species and strain, and reads the functional gene content of the community directly. Long-read sequencing on Oxford Nanopore R10.4.1 chemistry, paired with classifiers designed for the longer reads, has brought species-level accuracy and per-sample cost within reach of routine clinical use.

This is the methodology layer behind every BAARK report. Decisions about treatment, diet, antimicrobials, and clinical management belong with the treating veterinarian, alongside the history, examination, and conventional diagnostics that a clinical decision requires. The microbiome view is one input into that conversation, and the value of that input depends, more than anything else, on the method that produced it.