February 21, 2023

Spatial Transcriptomics and Molecular Stratification of Cancer Patients Using Pathology Images

5th Call, iDoc

Short Summary

Digital pathology is an established technology for cancer diagnosis in hospitals. The patterns and structures (morphology) in tumor tissue samples are central for categorizing cancer patients in broad prognostic groups, but the utility for personalized treatment is limited. With the emergence of next-generation sequencing (NGS) and single-cell genomic profiling technologies, we can now obtain a molecular understanding of cancer at the level of the individual patient with strong predictive value. In this iDoc study, we design two novel workstreams (WS) in close collaboration between the professorship for Computer-aided Image Analysis in Pathology at USZ / UZH (Koelzer) and the Biomedical Informatics Group at ETHZ (Rätsch). We investigate a) the changes in molecular signaling pathways and b) mutations in cancer genes in the context of tumor structure and form to aid the personalized diagnosis and prognosis of cancer patients. We develop methods for integrative bioinformatic analysis of molecular and morphological data. We will test the developed methods in the real-world setting of malignant melanoma and endometrial cancer (EC) in clinical patient samples.

Goals

The goal of the iDoc is to develop AI-based methods for the application to digital and molecular pathology datasets to improve the prognostication of cancer patients. The main focus will be on the following two work streams (WS), both balancing method development and translational application: WS1 (DigPathWays) aims to generate novel approaches to investigate the gene expression patterns in cancer tissue at single-cell resolution and in a spatially resolved manner using clinically established histopathology images.
WS2 (Morpho-molecular pathology) will focus on developing and applying image-based classifiers for molecular subtypes based on DNA mutations in oncogenic driver genes using clinical trial samples of the PORTEC EC cohort.

Significance

The methods to be developed in WS1 have the potential a) to strongly reduce the cost of obtaining spatially resolved gene expression profiles, b) to be faster (anticipated turn-around in minutes, vs. days-weeks with established transcriptomics platforms), and c) for easy deployment in established digital pathology workflows. WS2 will a) provide the scientific community with a tool to classify research cohorts molecularly without the need for molecular tests, b) allow to generate novel biological insights into specific molecular subtypes of EC and their intra- and inter-patient heterogeneity, and will c) promote the translation of image-based precision diagnostics to histopathology workflows.

Background

Recent studies indicate that the spatially-resolved gene expression profiles in cancer tissue hold the potential to predict response to treatment and patient survival. Access to a cohort of 126 malignant melanoma patients from the multi-institutional Swiss Tumor Profiler Study allows us to perform our analysis cost-effectively. The cohort provides matched digital pathology, immunohistochemistry, bulk- and single-cell RNA-seq datasets, complete clinical outcome data to develop novel spatially resolved transcriptomics methods using only the image data as an input. Targeted treatment of cancer patients via molecular stratification is obtained by sequencing tumor DNA. Still, it has been challenging to translate into clinical practice due to the high sequencing costs. Here, we will aim to develop novel morpho-molecular stratification methods collaborating with the TransPORTEC consortium using inexpensive, clinically established pathology images as an input. The PORTEC trials represent the world’s most extensive collection of molecularly classified EC samples with complete outcome information and digital images.

Publications

Fremond, S., Andani, S., Wolf, J. B., Dijkstra, J., Melsbach, S., Jobsen, J. J., … & Bosse, T. (2022). Interpretable deep learning model to predict the molecular classification of endometrial cancer from haematoxylin and eosin-stained whole-slide images: a combined analysis of the PORTEC randomised trials and clinical cohorts. The Lancet Digital Health. (DOI: https://doi.org/10.1016/S2589-7500(22)00210-2).