NEW YORK – Researchers in the UK and Australia have homed in on a 22-strain gut microbiome signature they believe could predict response to immune checkpoint inhibitor combinations across diverse cancer types.
In a new study published in Nature Medicine last week, researchers including Ashray Gunjur, a medical oncologist and researcher at the Olivia Newton-John Cancer Research Institute in Australia and the Wellcome Sanger Institute in the UK, identified unique strains of bacteria from patients' gut microbiomes that could predict whether they'd respond to treatment with combined anti-PD-1 and anti-CTLA-4 therapy.
"What was reassuring with our study was that the patient cohort was so incredibly diverse," Gunjur said.
The clinical trial that the researchers used as a discovery cohort for their microbiome signature, the Phase II CA209 clinical trial, enrolled patients across multiple rare solid tumor types, including upper gastrointestinal cancers, gynecological cancers, and neuroblastomas. Of the 120 patients on the trial, 108 had received prior treatment. All patients on the study received a combination of Bristol Myers Squibb's PD-1 checkpoint inhibitor Opdivo (nivolumab) and CTLA-4 checkpoint inhibitor Yervoy (ipilimumab) for up to four doses, then continued on with maintenance Opdivo for up to two years or until their cancers progressed or they experienced unacceptable toxicity.
To identify their microbiome signatures, the researchers retrospectively analyzed these patients' fecal samples alongside their clinical outcomes. They performed deep shotgun metagenomic sequencing of the baseline fecal samples for 106 patients on the trial. Then, using a bespoke, genome-resolved metagenomics approach, Gunjur and his colleagues discovered baseline strain-level subspecies associated with treatment response.
That the researchers zeroed in on specific strains of various bacterial species, rather than just identifying the species themselves, was a key strength of the study, Gunjur said.
"There can be a lot of variation within a species in terms of the function of the bug," he said, noting that certain strains of Escherichia coli, the "prototypical pathogenic bacteria," can be completely benign, while others can cause terrible illness.
"The difference might be just a few genes," he said. "The only way we've discovered that is by testing lots of different strains of the same species."
For the present Nature study, Gunjur and colleagues identified 22 bacterial strains that were deemed most informative in predicting patient outcomes.
"The number 22 was honestly a bit arbitrary," he said, explaining that the bugs they chose were all within half as impactful of the most impactful one, which was a strain of Faecalibacterium.
That 22-strain signature "essentially recapitulated the predictive power of the full 1,397 quantifications" that they started with. Gunjur pointed out that there could be an opportunity to narrow that list further or even to increase it by including more strains of gut microbiome bacteria that fell just outside the threshold. "We didn't really define the optimal number of strains, but 22 was a realistic number of strains for us to scrutinize further," he said.
External validation
After homing in on those strains, Gunjur and colleagues set out to validate them in larger patient cohorts as well as confirm they were tumor-agnostic, as opposed to driven by only the certain types of cancers in the Opdivo-Yervoy study.
To confirm the pan-tumor nature of the signature's predictive ability, they tested the associations with clinical outcomes in various groupings of patients in the discovery cohort. For instance, they'd leave out the gynecological rare cancers and validate the signature in just the cancers that fell into the upper gastrointestinal or neuroblastoma groups and then would leave one of the other groups out in the next validation round.
The researchers also used data from various cohorts of patients treated on external immune checkpoint inhibitor clinical trials to validate their signature across cancer types and different regimens. The meta-cohort included 364 patients total.
Interestingly, they found that their signature held up well, but only for patients treated with the combined Opdivo-Yervoy regimen, not with Opdivo alone.
Gunjur said he and his colleagues were initially surprised by this finding.
"Historically, we've always sort of lumped them together — anti-PD-1 and the combo with anti-CTLA4 — sort of feeling that the anti-PD-1 therapy was the more active drug carrying the combo," he said. "But these findings show that we really need to recognize that these are distinct regimens working in distinct ways."
Beyond validating the signature, Gunjur and his coauthors wanted to learn more about the function of individual strains and whether some were playing an outsized role in driving patients' treatment responses — or, conversely, their resistance — relative to others. To do this, they ran various models and machine learning algorithms and found that, ultimately, a few strains were "disproportionately important." One was a specific strain of Faecalibacterium.
The researchers interrogated the genomes of the top 22 strains to understand their functional potentials and why some were linked to positive immunotherapy responses while others were the oppositive.
"Getting down to the strain level helps us start to associate specific functions to response and nonresponse phenotypes to really understand the direction of that relationship," Gunjur said.
Looking ahead, he said it will be important to isolate the strains and investigate whether they have individual sway on patient outcomes or whether the predictive power is rooted in the interplay between the microbiome strains.
"Do they synergize or are they kind of individuals?" he asked hypothetically. "That's the type of experiment that would really need to happen to be able to justify if these 22 strains are the ideal number for a predictive biomarker."
According to Gunjur, this is exactly the type of research that's "next on the horizon" for his team. In his lab at the Wellcome Sanger Institute, he said researchers are building a culture collection and using state-of-the-art genomic technology to study the biology of each individual species and strain.
For now, the signature isn't quite ready to be used in a routine clinical capacity to guide immunotherapy treatment decisions, but Gunjur said that is part of the goal, big picture. This is especially true given the historically inconsistent predictive power of routinely used immunotherapy biomarkers like PD-L1 expression or tumor mutational burden. These biomarkers might work well in certain cancer types but not others, and even have significant heterogeneity within cancer types, he pointed out.
"We could have a predictive biomarker that uses a stool test, and based on the test, say 'Ok, well this is the likelihood of response to immunotherapy,'" he said, adding the caveat that there would be quite a lot of work required to develop a test that would give oncologists and their patients clear answers as quickly as possible.
Beyond the predictive biomarker vision, Gunjur is also enthusiastic about the prospect of microbiome modulators.
"Unlike a lot of other biomarkers, the gut microbiome is actually modifiable," he said. "There are avenues for altering people's diets and even replacing a whole microbial ecosystem. The potential to enhance immunotherapy responses is a really exciting potential application."
Elsewhere, researchers and early-stage biotechs have also jumped on this potential. For instance, academic and commercial scientists have shown that fecal microbiome transplants can convert certain patients from immunotherapy nonresponders into responders. Others have been trying to develop oral capsules with specific bacteria strains meant to improve immunotherapy responses.
Gunjur said these strides are all encouraging, but that the field needs to come together and harmonize their methods of collecting and analyzing stool samples, among other variations in how this research is conducted.
"Microbiome research is still pretty young," he said. "It's only within the last decade that we've appreciated how important to health the gut microbiome is. Understandably, it will take time for these efforts to harmonize. But we're already seeing progress."