Defining the relationship between the gut microbiome and response to combination immune checkpoint blockade across cancer types

Immune checkpoint inhibitors (ICIs) targeting the proteins Programmed cell Death protein 1 (PD-1) and Cytotoxic T Lymphocyte associated protein 4 (CTLA-4) have revolutionised the treatment of many cancers, and can be used in combination for synergistic anti-tumour effect. Nevertheless, response to ICIs remain unpredictable, and we lack reliable ‘biomarkers’ to predict this a priori. Studies have demonstrated the importance of the gut microbiome (referring to the sum of microorganisms, predominantly bacteria, and their genomic content in our digestive tracts) in mediating ICI response, creating enthusiasm for a gut microbial biomarker to guide ICI therapy. However, though associations can be found at the study level, previous meta-analyses have failed to define a microbial ‘signature’ (that is, a pattern of quantifiable traits, such as taxon abundances) for response that is generalisable across cohorts. At the outset of this work, I hypothesised that one reason for this may be that previous work primarily profiled gut microbiota to the species- or genus- level, which may lack the resolution to capture the precise, strain-specific markers of response. As such, in this thesis I attempt to define a generalisable microbiome signature of response by utilising deep shotgun metagenomic sequencing from a large, multi-centre Australian cohort of patients with advanced-stage, diverse, rare cancers (n= 106 discovery cohort). All patients were treated with combination immune checkpoint blockade (CICB), namely ipilimumab (anti-CTLA-4) and nivolumab (anti-PD-1) therapy, with a uniform clinical schedule. Using a genome-resolved metagenomics and supervised machine learning approach, I demonstrate that sub-species (strain) level gut microbial quantifications can reliably predict response to CICB across cancer types, and outperforms models built using clinical or species-level microbial quantifications. To externally validate this gut microbial signature, I subsequently analysed another original cohort of patients with advanced biliary tract cancer treated with ICIs, as well re-analysed publicly available data from 6 comparable studies (n= 394 validation cohorts). Notably, performance was better in CICB versus anti-PD-1 treated cohorts. Vice versa, predictive models trained using cohorts treated with anti-PD-1 monotherapy performed better in other such cohort, providing evidence that gut microbial response signatures may be tumour agnostic, but regimen-specific. Finally, by interrogating feature importance within the gut microbial signature, we could identify 22 strains disproportionately predictive (7 positive, 15 negative) of CICB response. Strikingly, 4 of the 7 strains belonged to a specific clade within the highly diverse genus Faecalibacterium. Overall, this body of work serves to highlight the potential for deep shotgun metagenomic predictors of immunotherapy response, but suggests their development should be regimen, rather than cancer, specific. In the future, strain-level analysis of bacterial isolates is warranted.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2025

Author dc:creator

• Gunjur, Ashray

Advisors dc:contributor.advisor

• Adams, David
• Lawley, Trevor

• microbiome
• cancer immunotherapy
• immune checkpoint inhibitors
• rare cancers

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/62ef6f81-0d81-4654-b480-2e0921260ffc/download
• http://purl.org/NET/rdflicense/allrightsreserved

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.119109

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/385532