Profiling the gut and oral microbiota of hormone receptor-positive, HER2-negative metastatic breast cancer patients receiving pembrolizumab and eribulin

Abstract

Aim: Changes in host-associated microbial communities (i.e., the microbiota) may modulate responses to checkpoint blockade immunotherapy. In the KELLY phase II study (NCT03222856), we previously demonstrated that pembrolizumab [anti-programmed cell death protein 1 (PD-1)] combined with eribulin (plus microtubule-targeting chemotherapy) showed encouraging antitumor activity in patients with hormone receptor (HR)-positive/human epidermal growth factor receptor 2 (HER2)-negative metastatic breast cancer (mBC) who had received prior treatments. Methods: A total of 58 fecal and 67 saliva samples were prospectively collected from a subset of 28 patients at baseline (BL), after three treatment cycles, and end of treatment. Shotgun metagenomics, 16S rRNA gene amplicon sequencing, and bioinformatics and statistical approaches were used to characterize fecal and oral microbiota profiles. Results: Treatment caused no substantial perturbations in gut or oral microbiota, suggesting minimal drug-related microbial toxicity. Bacteroides and Faecalibacterium were the dominant gut microbiota genera, while Prevotella and Streptococcus were present in both oral and gut samples, highlighting potential gut-oral microbial interactions. Additionally, clinical benefit (CB) appeared to be associated with gut-associated Bacteroides fragilis (B. fragilis) and a BL oral abundance of Streptococcus ≥ 30%. Notably, B. fragilis NCTC 9343 supernatant induced dose-dependent lactate dehydrogenase (LDH) release from the MCF-7 (HR-positive/HER2-negative) BC cell line. Conclusion: These findings suggest that specific gut and oral microbiota may modulate the effectiveness of combinatory anti-BC therapies, potentially through the action of microbial metabolites.

Keywords: 16S rRNA gene amplicon sequencing; Microbiota; breast cancer; eribulin; immunotherapy; metastatic breast cancer; pembrolizumab; shotgun metagenomic sequencing.

Figure 1

Relative abundances of (A) gut and (B) oral microbiota profiles showed little changes throughout the study treatment, thereby indicating that pembrolizumab and eribulin do not confer microbiota toxicity; (C) There were 57 common genera between the oral and gut microbiota profiles.

Figure 2

(A) Heatmap of metabolic pathways identified to have a LDA > 2.0 using LEfSe analysis for the clinical feature: time to PFS. Unstratified abundances of all pathways produced by Humann3 were run through LEfSe. Pathways identified as a discriminative feature as a LDA > 2.0 was then selected to create the heatmap. We categorized PFS into two groups: > 6 months and ≤ 6 months; (B) Antibiotic resistance genes identified in CALADRIO samples using Resfinder associated with CB. LDA: Linear discriminant analysis; LEfSe: linear discriminant analysis effect size; PFS: progression-free survival; CB: clinical benefit.

Figure 3

Samples with a relative abundance of > 30% Streptococcus came from patients who experienced a CB. (A) Heatmap by Bray-Curtis dissimilarity matrix on saliva samples by the top ten genera; (B) further visualization showed this tended to be BL samples. This difference between oral Streptococcus in CB and no-CB group was lost at consecutive time points (C), most likely due to a combination of antibiotic usage and a lower number of EOT samples provided. Mann-Whitney U test was performed on the CB vs. no-CB group where no significance was observed. Antibiotics, categorized by class, taken by patients based on clinical metadata. Not all patients enrolled in the CALADRIO study had complete metadata about antibiotics, which is why the number of patients presented differs from the number of patients enrolled (D). No clustering of gut microbiota data was observed according to CB status, suggesting no overall taxonomic changes (E). Mann-Whitney U was performed to compare the relative abundances of Streptococcus by time point. CB: Clinical benefit; BL: baseline; EOT: end-of-treatment.

Figure 4

LEfSe analysis of the salivary microbiota data showed five genera associated with no-CB and one for CB (A). B. fragilis in the gut is associated with CB. LEfSe analysis (B) showed that B. fragilis is associated with CB. Box plots (C) further showed that patients with CB consistently had a greater relative abundance of B. fragilis than no-CB. This observation was most significant at EoT [Mann Whitney U: P = 0.009, U = 0, median (yes, no) = 1.95, 0.38, range (0.00-11.73)]. LEfSe: Linear discriminant analysis effect size; CB: clinical benefit.

Figure 5

Antibiotic resistance presence in B. fragilis MAGs based on ResFinderFG. MAGs: Metagenome assembled genome.

Figure 6

B. fragilis NCTC 9343 supernatant harvested during growth (0-36 h) could significantly (^***P < 0.001) stimulate more LDH release, measured by OD₄₉₀, from MCF-7 (HR[+]/HER2[-]) breast cancer cells in vitro compared to BHI control (A). However, this was not reflected as cell death since viability measured by OD₄₉₀ (B) was not significantly different than BHI (^*P < 0.05). Linear regression analysis showed that this was a dose dependent relationship (slope = 0.0186, P-value = 4.52 × 10^-8%) (C). All conditions had n = 16, OD490 readings are corrected for background noise i.e., BHI media. Welch’s two sample t-test was used to compare the means of BHI and B. fragilis for both (A) and (B). LDH: Lactate dehydrogenase; BHI: brain-heart infusion.