Tumoroscope: a probabilistic model for mapping cancer clones in tumor tissues

Shadi Darvish Shafighi, Agnieszka Geras, Barbara Jurzysta, Alireza Sahaf Naeini, Igor Filipiuk, Łukasz Rączkowski, Hosein Toosi, Łukasz Koperski, Kim Thrane, Camilla Engblom, Jeff E. Mold, Xinsong Chen, Johan Hartman, Dominika Nowis, Alessandra Carbone, Jens Lagergren, Ewa SzczurekbioRxiv 2022 · DOI

TL;DR

This paper introduces Tumoroscope, a novel probabilistic model that integrates pathological images, whole exome sequencing (WES), and spatial transcriptomics (ST) to map cancer clones within tumor tissues at high resolution. It addresses the critical challenge of understanding tumor heterogeneity by not only identifying clone locations but also deconvoluting their gene expression profiles, offering a comprehensive view of tumor organization.

What Problem Does This Paper Address?

Cancer is driven by the accumulation of mutations, leading to distinct subpopulations of cells called "clones." The spatial distribution and genetic makeup of these clones are crucial for understanding tumor progression, treatment response, and patient survival. However, existing technologies struggle to map these clones directly within the tumor tissue at a high resolution. While WES can identify mutations, it loses spatial information. Spatial transcriptomics can map gene expression across tissue but provides only aggregated signals from small groups of cells (spots), making it difficult to infer the underlying clonal composition and genotypes. This paper tackles the challenge of precisely mapping cancer clones and their associated gene expression profiles within the complex tumor microenvironment.

Methods

Tumoroscope is a probabilistic graphical model that integrates three key data types:

  • Pathological images (H&E): Used to identify cancerous regions and estimate the number of cells within each ST spot.
  • Whole Exome Sequencing (WES): Provides the genetic blueprint of the tumor, allowing for the reconstruction of cancer clones and their genotypes.
  • Spatial Transcriptomics (ST): Offers spatially resolved gene expression data, where each spot represents a small group of cells.

The core of Tumoroscope involves:

  1. Clone Reconstruction: Using WES data and methods like Canopy, the evolutionary tree of somatic mutations is inferred, defining the genotypes and frequencies of different cancer clones.
  2. Spatial Deconvolution: Tumoroscope models each ST spot as a mixture of these identified clones. It uses a probabilistic framework to estimate the proportion of each clone within every ST spot, leveraging cell counts from H&E images and mutation data from ST reads.
  3. Gene Expression Deconvolution: A regression model is then employed to infer the gene expression profiles of individual clones. This model uses the inferred clone proportions in each spot and the measured gene expression in that spot to deconvolve clone-specific expression levels.

The model is designed to be robust to noise in cell count estimations and can handle the inherent limitations of ST data, such as aggregated signals.

Datasets

The authors evaluated Tumoroscope on two main datasets:

  • Prostate Cancer Dataset: A publicly available dataset from Berglund et al. [24], providing H&E images, bulk WES, and ST data.
  • Breast Cancer Dataset: A newly generated dataset from this study, also including H&E images, bulk WES, and ST data from three breast tumor sections.

Simulated datasets were also extensively used to evaluate the model's performance under various conditions, including different numbers of mutations, clones, coverage levels, and noise in cell counts.

Key Insights and Ideas

  • Integrated Data Approach: The power of combining H&E, WES, and ST data is demonstrated, overcoming the limitations of each individual technology for a comprehensive view of tumor heterogeneity.
  • Deconvoluting Mixtures: Tumoroscope's ability to explicitly model and deconvolve the mixture of clones within ST spots is crucial for accurate spatial mapping, outperforming methods that treat spots as single-cell equivalents.
  • Phenotypic vs. Genotypic Similarity: The study suggests that phenotypic similarity (gene expression profiles) might drive the spatial co-occurrence of clones, rather than just genotypic similarity, opening new avenues for understanding clone interactions.
  • High-Resolution Spatial Mapping: Tumoroscope provides unprecedented spatial resolution for clone mapping, down to the scale of ST spots (roughly 1-40 cells), enabling the identification of localized clonal patterns.

Key Results

  • Accurate Clone Proportion Estimation: Tumoroscope accurately estimates clone proportions in ST spots, demonstrating robustness to noise in cell count estimates and performing well across various simulation scenarios.
  • Spatial Patterns Revealed: Applied to real breast and prostate cancer data, Tumoroscope reveals distinct spatial patterns of clone colocalization and mutual exclusion within tumor sub-areas.
  • Clone-Specific Gene Expression: The model successfully infers gene expression profiles for individual clones, highlighting potential functional differences between them and suggesting clone-specific roles in tumor progression.
  • Superior Performance: Tumoroscope significantly outperforms an alternative method (cardelino) in accurately assigning clonal composition to ST spots, especially when dealing with clone mixtures.

Limitations and Future Work

  • Technological Limitations: The resolution of current ST technology limits the ability to resolve individual cells, and the sequencing depth of ST data can impact mutation detection accuracy. Future improvements in ST technology (e.g., smaller spots, whole gene body sequencing) will further enhance Tumoroscope's capabilities.
  • Data Integration: While Tumoroscope integrates multiple data types, the accuracy of clone inference is dependent on the quality and availability of WES data for robust genotype reconstruction.
  • Validation: The finding that phenotypic similarity drives clone co-occurrence needs further validation with additional patient samples and independent datasets.
  • 3D Mapping: The authors suggest that their framework could be extended to profile adjacent tumor samples to create 3D maps of clones, enabling predictions of proliferating areas and tumor expansion sites.
  • Clinical Applications: Tumoroscope opens up new avenues for cancer research, potentially aiding in understanding disease progression and informing clinical applications.

Summary updated

Abstract

Abstract Spatial and genomic heterogeneity of tumors is the key for cancer progression, treatment, and survival. However, a technology for direct mapping the clones in the tumor tissue based on point mutations is lacking. Here, we propose Tumoroscope, the first probabilistic model that accurately infers cancer clones and their high-resolution localization by integrating pathological images, whole exome sequencing, and spatial transcriptomics data. In contrast to previous methods, Tumoroscope explicitly addresses the problem of deconvoluting the proportions of clones in spatial transcriptomics spots. Applied to a reference prostate cancer dataset and a newly generated breast cancer dataset, Tumoroscope reveals spatial patterns of clone colocalization and mutual exclusion in sub-areas of the tumor tissue. We further infer clone-specific gene expression levels and the most highly expressed genes for each clone. In summary, Tumoroscope enables an integrated study of the spatial, genomic, and phenotypic organization of tumors.
BibTeX
@misc{shafighi2022tumoroscope,
  author = {Shadi Darvish Shafighi and Agnieszka Geras and Barbara Jurzysta and Alireza Sahaf Naeini and Igor Filipiuk and Łukasz Rączkowski and Hosein Toosi and Łukasz Koperski and Kim Thrane and Camilla Engblom and Jeff E. Mold and Xinsong Chen and Johan Hartman and Dominika Nowis and Alessandra Carbone and Jens Lagergren and Ewa Szczurek},
  title = {{Tumoroscope: a probabilistic model for mapping cancer clones in tumor tissues}},
  year = {2022},
  doi = {10.1101/2022.09.22.508914},
  howpublished = {bioRxiv},
}

Loading related content...

Figure