Effect of bacterial glycosyltransferases on resistance to Lewis Y-targeting antibody-drug conjugates.

Abstract

e15013 Background: Bacteria in the tumour microenvironment (TME) influence anticancer drug pharmacodynamics, potentially contributing to variability in clinical trial outcomes. Previously, we demonstrated that microbial-derived proteins modulate topoisomerase inhibitors, explaining differences between Japanese and US populations. We now extend this to antibody-drug conjugates (ADCs) showing bacterial glycosyltransferases disrupt ADC targets like the Lewis Y antigen (LeY). Despite preclinical promise, the LeY-directed ADC, SGN-15, failed to demonstrate sufficient efficacy in non-small cell lung cancer (NSCLC) trials, leading to discontinuation. We hypothesised bacterial proteins in the TME modulate LeY targeting, contributing to tumour-specific therapeutic failure and geographic variability in outcomes. Methods: Analysis was performed using BioCorteX’s knowledge graph and proprietary engines v20250124_015926. Tumour microbiome profiles were obtained from 4436 primary tumours of 5 cancer types, including NSCLC, across geographically diverse cohorts. Bacteria-derived proteins were identified, and ranked by abundance of microbial genera producing a given protein. Sequence homology, complemented by 3D structural analysis, was used to evaluate mimicry of human enzymes involved in biosynthesis, modulation, and metabolism of LeY. Functional impacts on ADC efficacy were inferred using a "Numbers Needed to Interact" (NNI) metric. Results: Bacteria expressing LeY-interacting proteins were significantly enriched in NSCLC compared to other cancer types (68% vs 18%, p<0.01). Functional analyses highlighted enzymes involved in Lewis antigen biosynthesis, potentially generating competing glycans that impair LeY-targeted ADCs. Structural analysis revealed similarities between bacterial and human fucosyltransferases, suggesting mimicry that may impair ADC binding. NSCLC showed lower NNI scores for SGN-15 compared to other cancer types (1.5 vs 10.5, p<0.01). Regional bacterial enzyme prevalence emphasises contributions to geographic disparities in ADC efficacy. Conclusions: This study provides the first evidence that bacterial glycosyltransferases in the TME may interfere with ADC efficacy, potentially contributing to failure of LeY-targeting ADCs such as SGN-15 in NSCLC. Mimicry of human fucosyltransferases by bacterial enzymes reveals a novel resistance mechanism that builds on prior literature implicating H pylori in Lewis antigen modulation. These findings underscore the global relevance of tumour-associated bacterial ecosystems in shaping cancer treatment outcomes and suggest geographic variability in tumour microbiome composition may influence ADC efficacy. Incorporating an understanding of drug-bacteria interactions in ADC development pipelines could improve patient stratification and optimise the probability of success for clinical trials.