Related Experiment Video
Updated: Mar 29, 2026
11:02
Induction of Invasive Transitional Cell Bladder Carcinoma in Immune Intact Human MUC1 Transgenic Mice: A Model for Immunotherapy Development
Published on: October 30, 2013
21.9K
Spatial Immune Profiling and AI-Based Classifiers Identify Predictors of BCG Therapy Outcomes in High-Risk
Melinda Lillesand1,2, Marie Austdal3, Jakub Mroz4
1Department of Pathology, Stavanger University Hospital, 4011 Stavanger, Norway.
Cancers
|March 28, 2026
Summary
High recurrence rates in bladder cancer treatment underscore the need for better biomarkers. Spatial profiling reveals immune cell patterns linked to Bacillus Calmette-Guérin (BCG) immunotherapy response, offering potential for improved patient stratification.
Area of Science:
- Oncology
- Immunotherapy
- Spatial Biology
Background:
- Bladder cancer treatment is costly due to high recurrence and intensive surveillance.
- Bacillus Calmette-Guérin (BCG) immunotherapy is standard for high-risk non-muscle-invasive bladder cancer (NMIBC), but 50% of patients do not respond.
- Predictive biomarkers for BCG response are lacking, hindering personalized treatment strategies.
Purpose of the Study:
- To validate BCG response subtypes at single-cell spatial resolution.
- To identify spatial immune and stromal features associated with BCG immunotherapy response and failure in NMIBC.
- To develop a predictive model for BCG response using imaging mass cytometry (IMC) data.
Main Methods:
- Applied single-cell IMC to 82 BCG-treated high-risk NMIBC samples.
- Utilized unsupervised clustering to identify cell phenotypes and quantified cluster abundances.
- Developed and trained a convolutional neural network-based gated attention multiple instance learning model (IMC-GA-MIL) to predict BCG response.
Main Results:
- Identified 18 distinct cell clusters; enriched immune cells within the tumor compartment were protective in BCG responders.
- BCG nonresponse correlated with increased fibroblast-dominant clusters relative to tumor cells.
- Plasma cell-dominant clusters predicted progression; IMC-GA-MIL model achieved 90% accuracy in predicting BCG response.
Conclusions:
- Spatial single-cell profiling and IMC-GA-MIL can identify spatial immune and stromal features linked to BCG failure.
- Specific cell cluster abundances and spatial arrangements are associated with treatment outcomes.
- Findings are exploratory and require validation in larger, independent cohorts for clinical application.
