Periodic Reporting for period 2 - TWIGA (Tissue-wide identification of genome alterations in cancer)
Okres sprawozdawczy: 2023-04-01 do 2024-09-30
Our focus on cancer is easy to motivate as it impacts the individual, health care, and society. We provide innovative technology for early detection, tumor progression, relapse, and resistance. The ongoing developmental work takes advantage of novel spatial knowledge to identify and understand genomic mutations in cancer.
The main objectives of the proposal are to (i) develop a robust spatial barcoding technology to investigate genomes from a tissue section, mapping the spatial distribution of gains and losses of chromosomal information (copy number variations,CNVs ) and, eventually, point mutations within the same tissue section (ii) develop a computational framework to merge high-resolution microscopy imaging data with molecular data using machine learning and deep learning principles to visualize histology, gene expression profiles, cell types, and genomic alterations (iii) investigate, describe, and compare the genomic landscape in prostate and breast cancer to elucidate the impact of genomic alterations on disease.
The main objectives of the proposal are to (i) develop a robust spatial barcoding technology to investigate genomes from a tissue section, mapping the spatial distribution of gains and losses of chromosomal information (copy number variations,CNVs ) and, eventually, point mutations within the same tissue section (ii) develop a computational framework to merge high-resolution microscopy imaging data with molecular data using machine learning and deep learning principles to visualize histology, gene expression profiles, cell types, and genomic alterations (iii) investigate, describe, and compare the genomic landscape in prostate and breast cancer to elucidate the impact of genomic alterations on disease.
As outlined in the application, we proposed two strategies for surveying genomic alteration in cancer samples using the genome or the transcriptome as a starting point. We provided preliminary data on both routes and continued the research in parallel. A highlight from the project is that we have demonstrated that our spatial transcriptome-wide data can successfully predict chromosomal gains and losses on cancer samples (Erickson et al., Nature 2022). An attractive feature of our spatial approach is the cost-saving; we generate spatial genomic maps without additional tissue sections and reagents for genome analysis.
Furthermore, we provided preliminary data that we could use deep learning to fuse microscopy images and gene activity (transcriptome) to achieve a super-resolution view of tissues. This work was published (Bergenstråhle et al., Nature Biotech 2022) and demonstrated that deep learning could be used for these purposes; this study also provided insight that 3D models can be constructed using only microscopy images if the training set includes molecular spatial data. Establishing 3D models will be essential to understanding heterogeneity in the tumor ecosystem. We have also used deep learning to automatically reconstruct a 3D view from hundreds of tumor sections (Ekvall et al., Nature Methods 2024).
Several novel biological insights into prostate cancer (see below, Erickson et al., Nature 2022) and breast biology have been obtained. Today, we are using this TWIGA technology to study metastasis in prostate cancer (in collaboration with Dr Lamb). We also have data on the precursor lesions in prostate cancer that confirms the wide occurrence of precursors in prostate cancer. We are also using TWIGA technology to provide the spatial components of breast cancer to be part of an unprecedented new breast cancer atlas with rich metadata.
Furthermore, we provided preliminary data that we could use deep learning to fuse microscopy images and gene activity (transcriptome) to achieve a super-resolution view of tissues. This work was published (Bergenstråhle et al., Nature Biotech 2022) and demonstrated that deep learning could be used for these purposes; this study also provided insight that 3D models can be constructed using only microscopy images if the training set includes molecular spatial data. Establishing 3D models will be essential to understanding heterogeneity in the tumor ecosystem. We have also used deep learning to automatically reconstruct a 3D view from hundreds of tumor sections (Ekvall et al., Nature Methods 2024).
Several novel biological insights into prostate cancer (see below, Erickson et al., Nature 2022) and breast biology have been obtained. Today, we are using this TWIGA technology to study metastasis in prostate cancer (in collaboration with Dr Lamb). We also have data on the precursor lesions in prostate cancer that confirms the wide occurrence of precursors in prostate cancer. We are also using TWIGA technology to provide the spatial components of breast cancer to be part of an unprecedented new breast cancer atlas with rich metadata.
Our publication in Nature 2022, Erickson et al., represents a significant technical and scientific achievement. This paper used spatial gene expression data in cancer samples to describe the successful mapping of chromosomal gains and losses over a cancer tissue section. Our paper uses barcoded spatial glass slides, with the significant improvement being that we can place chromosomal gains and losses in a precise tissue context. In particular, we use a whole cross-section of prostate cancer, providing an unbiased data-driven approach to identifying multiple tumors without prior knowledge. This approach allowed us to establish evolutionary trees over the history of tumor development and place them into a tissue context. This was an important advancement in the field that previously mainly used laser cell microdissection (LCM) to select areas (biased) for genetic investigation, a selection usually based on tumor morphology.
One of our breakthroughs in this paper was that we identified tissue with normal morphology but contained genetic alterations similar to the tumor. This finding would not have been possible with other LCM methods focusing on cancer alone. We denoted this altered benign epithelia as representing a new spatial entity with somatic mutations that suggest an early event in tumor development. This opens up exciting possibilities for developing early diagnostic biomarkers and represents an unexpected discovery.
One of our breakthroughs in this paper was that we identified tissue with normal morphology but contained genetic alterations similar to the tumor. This finding would not have been possible with other LCM methods focusing on cancer alone. We denoted this altered benign epithelia as representing a new spatial entity with somatic mutations that suggest an early event in tumor development. This opens up exciting possibilities for developing early diagnostic biomarkers and represents an unexpected discovery.