The microbiome and cancer immunotherapy.

The human observation that started this.

Checkpoint inhibitors, the drugs that release the brakes on a patient's own T cells, produce meaningful survival benefits in some cancers including melanoma and non-small cell lung cancer, but only in a subset of patients. A great deal of work has gone into explaining that subset, and the gut microbiome turned out to be one of the variables that tracked with it.

Three papers in Science set the field going, all in melanoma or epithelial tumours. Gopalakrishnan and colleagues profiled the oral and gut microbiome of 112 melanoma patients receiving anti-PD-1. Among the 43 patients with faecal samples, 30 responders and 13 non-responders, responders showed significantly higher alpha diversity (P < 0.01) and a higher relative abundance of the family Ruminococcaceae (P < 0.01). Germ-free mice given faecal transplants from responding patients showed enhanced antitumour immunity.

Matson and colleagues, in the same volume, analysed baseline stool collected before treatment from metastatic melanoma patients, combining 16S ribosomal RNA sequencing, shotgun metagenomic sequencing and quantitative PCR for selected bacteria, and reported a comparable association. Routy and colleagues added the drug-interaction result: antibiotics inhibited the clinical benefit of checkpoint inhibitors in patients with advanced cancer, and faecal transplant from responding patients into germ-free or antibiotic-treated mice improved the effect of PD-1 blockade, while transplant from non-responders did not.

The strongest version of that evidence is a transplant experiment. Faecal transplant from a drug-responsive patient into a non-responsive one converted a subset of prior non-responders in a human clinical trial, allowing them to benefit from treatment. That is a real result and it is the reason the field exists.

The follow-up analysis is the part that rarely gets quoted. Extensive pre- and post-treatment analysis of microbiome composition, local metabolism and expression of immune-related genes failed to show significant differences between the patients who became responsive and those who did not. As the reviewers put it, while successful in proving the concept of faecal transplant to be safe and potentially beneficial, the effects remain correlative.

So the intervention worked and nobody can yet say what changed. Relatedly, although several bacterial species of interest have been identified, no universal responder signature has been identified. There is no list of organisms you could test for.

A second trial, run by Davar and colleagues in patients whose melanoma was refractory to anti-PD-1, gave responder-derived faecal transplant alongside the drug. It was well tolerated and provided clinical benefit in 6 of 15 patients. Fifteen patients is a small trial, and the authors present it as a proof of concept rather than a treatment.

Then came the counterweight, and it is the paper to read if you are being sold a microbiome cancer test. Lee and colleagues ran shotgun metagenomic sequencing on stool collected before treatment from 165 checkpoint-inhibitor-naive patients with advanced cutaneous melanoma across five observational cohorts, then pooled that with 147 metagenomes from earlier published studies, 312 samples in total. The association between the gut microbiome and response held, but it was cohort-dependent, and a machine learning analysis found limited reproducibility of microbiome-based signatures across cohorts. Several hundred patients after the first three papers, there is still no organism list you could score a human patient against, let alone a dog.

Why dogs were proposed as the bridge.

The argument for a canine model is that mice are too far from people and people are too hard to experiment on. Mouse and human gut microbiome composition may be too dissimilar for discovery of all the relevant microbial biomarkers, and even within mouse work the results have been unstable: it is not clear why effects were seen in some but not all mice, and studies identifying similar mechanisms named different organisms as responsible.

Dogs sit between the two. They develop naturally occurring cancers rather than induced ones, and genetic studies show notable overlap in the tumour genetics of human and canine osteosarcoma, melanoma, mammary tumours, gliomas and lymphoma. They have intact immune systems and a lifetime of established gut communities, they live in our houses and breathe our air, and canine housemates show similar microbiome profiles to one another in the way that human partners do.

Their gut communities are also closer to ours than the usual laboratory species. The five most prominent phyla in dogs are Firmicutes, Fusobacteria, Bacteroides, Proteobacteria and Actinobacteria, a composition more similar to humans than other commonly studied mammals. One comparison of 129 stool samples from 64 dogs found substantially greater overlap with the human gut gene catalogue than the mouse does.

Caninised checkpoint antibodies against PD-1, CTLA-4 and PD-L1 are now in development, which is what makes the proposal practical rather than theoretical.

The organism that goes the other way.

There is one detail in this literature that should stop anyone from reading human microbiome findings straight across into dogs.

Fusobacterium nucleatum in humans is associated with colorectal cancer and is thought to trigger inhibitory T cell receptors that suppress the anti-cancer immune response. In dogs, fusobacteria are associated with maintaining gut health. Fusobacteria are one of the five dominant phyla in the healthy dog, and their loss is one of the changes seen in canine chronic enteropathy.

The same genus, opposite meanings, in two species that share a sofa. Any test that scored a dog's sample against human cancer associations would get this exactly backwards.

Not everything diverges. Increased E. coli is associated with canine intestinal lymphoma, and both canine and human inflammatory bowel disorders show reduced community diversity with overgrowth of organisms with genotoxic potential such as Bacteroides fragilis and E. coli. But the Fusobacterium reversal is the standing warning against assuming the rest transfers.

What has actually been shown in dogs.

This is the shortest section in the post, and that is the point.

There is no published study of the gut microbiome and immunotherapy response in dogs. Not one. Every statement connecting the two in the current literature is written in the conditional: that microbiome correlative studies in dogs with intact immune systems may yield useful observations, that canine clinical trials should consider collecting stool samples for microbiome biomarkers, that large animal models will be useful for trialling interventions.

What exists in dogs is descriptive: the phyla present, breed and household effects, the instability of the puppy microbiome, and disease associations in enteropathy and lymphoma. None of it is a treatment-response finding, because no such study has been done.

The canine cancer data that does exist is worth reading, because it points the other way to the human findings. Gavazza and colleagues compared faecal microbiota in 12 dogs with multicentric B-cell stage III to IV lymphoma against 21 healthy dogs, using Illumina sequencing of 16S ribosomal RNA genes together with quantitative PCR for selected bacterial groups. Alpha diversity was significantly lower in the lymphoma dogs, community structure separated on principal coordinates analysis (P = 0.001), and linear discriminant analysis effect size identified 28 differentially abundant bacterial groups. On qPCR, dogs with lymphoma carried significantly less Faecalibacterium spp. (q < 0.001), Fusobacterium spp. (q = 0.032) and Turicibacter spp. (q = 0.043), and significantly more Streptococcus spp. (q = 0.041).

Read the Fusobacterium line against the previous section. Dogs with lymphoma had less of the genus that is enriched in human colorectal cancer. That is the species divergence showing up in real canine cancer data, not as a theoretical caution.

A second and smaller study by Mahiddine and colleagues profiled dogs with stage IV multicentric lymphoma against healthy dogs on MiSeq sequencing, and reported the phylum Actinobacteria and two species, Corynebacterium amycolatum and Streptococcus lutetiensis, in higher proportions in the affected dogs. The authors describe these as potential biomarkers, call the work a pilot study, and state that further investigation is needed to understand any mechanism.

Both studies compare dogs that have cancer with dogs that do not. Neither tells you whether changing a dog's microbiome would change how that dog responds to any treatment, and neither was designed to.

The reviews are also silent on whether microbiome profiling has any clinical use in veterinary oncology today. They make no claim either way, which in a field this promotional is itself informative.

What this means for an owner.

A faecal microbiome test cannot tell you whether your dog will get cancer, cannot tell you whether a cancer treatment will work, and cannot be used to select or modify cancer therapy. There is no canine evidence base for any of those uses, and in humans, where the evidence is far stronger, there is still no responder signature to test for.

What the field does support is the opposite direction of travel: that dogs with spontaneous tumours and intact immune systems are a good population in which to answer the question, and that stool collected during canine oncology trials would be worth banking. That is a research argument, not a consumer one.

Where BAARK fits.

The bottom line.

Faecal transplant converted some non-responders into responders in a human melanoma trial, and the analysis afterwards could not identify what had changed. No responder signature exists. No canine study of the microbiome and immunotherapy response has been published at all. And the one genus where dogs and humans have been directly compared points in opposite directions in the two species. This is an interesting research programme with a genuine case for using dogs, and it is not a product.