In a recent study published in Nature Medicine, researchers investigated the composition of the gut microbiota as a predictive marker for immune checkpoint blockade (ICB) responsiveness and toxicity.
ICB drugs that target programmed cell death protein 1 (PD-1), programmed death-ligand 1 (PD-L1), or cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) have transformed cancer therapy. Combination ICB has synergistic anticancer efficacy but is associated with variable responses and adverse immune-related side effects. Current tumor-agnostic biomarkers for PD-1 inhibition have limitations and depend on accessible tumor tissue. Gut microbiota composition may impact clinical response to ICB; however, identifying microbiome-based biomarkers for ICB response is difficult due to strain-specific measurement.
Study: A gut microbial signature for combination immune checkpoint blockade across cancer types. Image Credit: CI Photos / Shutterstock
About the study
In the present study, researchers analyzed deep shotgun metagenomic sequences from stool samples of the phase 2 CA209-538 trial cohort of rare cancer patients treated with ICB drugs such as nivolumab and ipilimumab. They investigated whether a patient's gut microbiota composition may predict ICB responsiveness and toxicity.
The trial included 120 individuals with histologically advanced stage solid-organ tumors from five hospital networks in Australia. The tumor histologies were diverse and categorized into three predetermined groups: upper gastrointestinal and biliary tumors (UGB), rare gynecological cancers (GYN), and neuroendocrine neoplasms (NEN). The majority of patients had previously undergone systemic anticancer treatments, and all participants in the trial were treated with a combination of ipilimumab and nivolumab for ≤4.0 doses (induction) and nivolumab maintenance therapy for ≤2.0 years or till progressive disease (PD) development.
The researchers predicted that distinct strains may predict combination ICB effectiveness in the sample. They used objective response versus progression (RvsP), defined as a RECIST best overall response (BOR) of partial response (PR) or complete response (CR) vs. cPD or PD, as the primary endpoint. They eliminated individuals with a BOR of stable disease (SD) since it is ambiguous in a pan-cancer sample. The secondary endpoint was to create 'tumor-agnostic' indicators for combination ICB responses by exploiting CA209-538's trial design. Most participants provided pretreatment fecal samples, and there were no significant clinical differences between microbiome-evaluable individuals and those unsampled.
Researchers sequenced 106 baseline stool samples using deep shotgun metagenomics to better understand patients’ gut microbial compositions. They adopted the genome-resolved strategy, using metagenome-assembled genomes (MAGs) and applicable Genome Taxonomy Database (GTDB) species reference genomes (SRGs) to create a study-specific microbial strain database for reference. Based on the patients' BOR and intersample beta diversity, the researchers used the Aitchison distance to determine if there were significant compositional variations.
The researchers used PFS12 and supervised machine learning algorithms to measure the performance of 15 clinical and microbiological parameters. They concentrated on strain-RvsP classifiers due to their superior performance and incremental advantage over ordinary clinical variables. They compared forecasts to patient BOR results and found strain abundances to be significant factors. They also analyzed the genomes of the top 22 strains to improve their understanding of their functional potential and performance in combination with ICB and anti-PD-1 monotherapy cohorts. Strain-RvsP RF classifiers were trained and evaluated across all strain abundances and cohort combinations.
Results
The findings indicated that strain-specific microbe abundances can improve machine-learning estimations of immune checkpoint blockade response and one-year progression-free survival compared to models based on species-rank quantifications or pretreatment clinical variables. Through the meta-analytical research of intestinal metagenomes obtained from six studies (364 individuals, validation cohort), the team found cross-country and cross-cancer validity of microbial strain-response characteristics using concordant ICB training and testing regimens (drugs targeting PD-1 or those combined with drugs targeting CTLA-4).
The results show that strain-resolved microbial abundances enhance machine-learning predictions of ICB response and 12-month progression-free survival compared to models based on species-rank quantifications or pretreatment clinical variables. The overall (OS) and progression-free survival (PFS) rates were more stable among histological cohorts. The team observed a significant positive relationship between albumin and BOR, especially in individuals with fast clinical development.
The BOR metadata characteristic explained the highest microbiological variation among the 23 pretreatment clinical and technical sample information examined. However, patient progression-free survival or overall survival at one year showed minimal microbial variation. Strain-responsiveness signatures were consistent across cancers, although clinical variables alone were insufficient predictors of RvsP. The research location and deoxyribonucleic acid (DNA) extraction kit were the most significant causes of microbial variation throughout the meta-cohort.
The study suggests that future gut microbiome diagnosis must depend on treatment regimens rather than cancer type. Assessing a patient's baseline gut microbiota composition may help predict ICB response and toxicity. In a phase 2 study of Australian patients with uncommon malignancies, researchers utilized strain-resolved metagenomic classification to identify 22 gut microbial strains linked to ipilimumab and nivolumab responses. The data support strain resolution in gut microbial ICB indicator development, generalizability across cancer types and geographic regions, and regimen disaggregation in future studies.
How AI is advancing mammographic density-based breast cancer risk prediction