Periodic Reporting for period 1 - SpatialOrganoids (Profiling the emergence of phenotypic heterogeneity in breast cancer organoids)
Período documentado: 2021-01-01 hasta 2022-12-31
Breast cancer shows the highest cancer incidence in women and extensive research focused on elucidating the mechanisms for the onset and progression of the disease. On a cellular level, breast cancer is characterized by extensive intra- and inter-patient heterogeneity in cellular and molecular phenotypes. Intra-tumor heterogeneity often limits the effect of cancer therapies due to the outgrowths of therapy-resistant clones of cancer cells. Inter-patient heterogeneity on the other hand limits the broad applicability of therapies across patient groups and personalized cancer therapies need to be developed. To date, the molecular events on a cell-to-cell level that lead to the emergence of tumor heterogeneity are largely unknown. Understanding the mechanisms of how intra- and inter-tumor heterogeneity forms will open avenues to improving cancer therapies against breast cancer which forms 13% of all new cancer cases worldwide.
The overall objectives of the project include the generation of breast cancer model systems, specifically breast cancer organoids, to profile the emergence of phenotypic heterogeneity. I use imaging mass cytometry (IMC) as a type of multi-parametric microscopy technology to study phenotypic changes over time (the growths of organoids) and space (location differences in cellular phenotypes). Finally, I perturb the growth of organoids using common cancer drugs to study their effect on the emergence of phenotypic heterogeneity. The final dataset allows me to predict possible clonal outgrowths of cancer cells within breast cancer organoids and derive targeted therapies.
The project resulted in datasets of three organoid lines treated with seven anti-cancer drugs over time periods of two to four weeks. I observed heterogeneous expression patterns between organoid lines and between time-points while treatment often had an “all or nothing” effect on organoid growth either inducing growth arrest or no phenotypic changes at all. In parallel, the project resulted in an extensive computational framework to support in-depth analysis of multiplexed imaging data.
The overall objectives of the project include the generation of breast cancer model systems, specifically breast cancer organoids, to profile the emergence of phenotypic heterogeneity. I use imaging mass cytometry (IMC) as a type of multi-parametric microscopy technology to study phenotypic changes over time (the growths of organoids) and space (location differences in cellular phenotypes). Finally, I perturb the growth of organoids using common cancer drugs to study their effect on the emergence of phenotypic heterogeneity. The final dataset allows me to predict possible clonal outgrowths of cancer cells within breast cancer organoids and derive targeted therapies.
The project resulted in datasets of three organoid lines treated with seven anti-cancer drugs over time periods of two to four weeks. I observed heterogeneous expression patterns between organoid lines and between time-points while treatment often had an “all or nothing” effect on organoid growth either inducing growth arrest or no phenotypic changes at all. In parallel, the project resulted in an extensive computational framework to support in-depth analysis of multiplexed imaging data.
The work performed during the project can be split into an experimental and computational part.
On the experimental side of the project, I have grown eight breast cancer organoid lines and developed a collection, embedding, cutting and staining workflow for generating imaging mass cytometry data from breast cancer organoids. To estimate organoid growth, I used a large particle FACS employing reference beads to accurately measure organoid size. In parallel, I optimized staining conditions to acquire 3D imaging data using a confocal microscope. To further characterize clinically relevant features of the organoids I optimized embedding and cutting strategies to acquire IHC images of HER2, ER and PR of selected organoid lines.
The main focus of the experimental part of the project was put on developing a drug screening approach to perturb organoid growth using seven commonly used drugs targeting signaling pathways in breast cancer. As part of the drug screening approach, I developed a heavy metal barcoding strategy allowing me to pool 20 conditions per embedding.
I observed that organoids grow in a heterogeneous fashion in which few organoids form large structures while the majority remain in a small potentially senescent state. Depending on the organoid line, different morphologies can be detected. These range from small and dense to large and hollow structures.
Studying the expression of 40 breast cancer relevant proteins in organoid sections revealed clear distinctions in terms of overall and spatial expression patterns between organoid lines. Major differences arise from the expression of cytokeratins and other epithelial markers including EpCAM and E-Cadherin. Studying the expression changes over organoid growth showed an initial lack of ER and PR expression, both markers used for diagnosing the type of breast cancer.
When growing breast cancer organoids under drug treatment, a number of effects can be observed. First, tamoxifen which targets ER signaling showed no influence on growth and protein expression. Second, inhibitors of the EGF signaling pathway showed strong growth inhibition and down-regulation of phosphorylation of S6. Third, MEK and MYC inhibitors show efficient growth reduction without blocking EGF signaling, however, the exact mechanism is not resolved yet.
The obtained results have been presented at seminars at the University of Zurich and ETH Zurich. Currently, I’m preparing a manuscript describing the findings of the time-course treatment of four representative breast cancer organoid lines.
On the computational side, I have developed a deep learning-enabled strategy for segmenting organoids after sectioning as well as an extensive workflow for analyzing IMC data.
Detecting organoid slices in images is challenging due to differences in size and shape, clumping of objects and the presence of biological debris. To tackle this segmentation issue, I have implemented a deep learning-based instance segmentation approach leveraging existing Mask RCNN frameworks. This approach results in higher accuracy than shallow-learning alternatives including watershed segmentation based in Ilastik-generated pixel probability maps.
In addition, I developed and documented an extensive computational framework for the analysis of IMC data as existing analysis approaches were poorly documented or difficult to apply in a unified framework. The developed framework is openly accessible at https://bodenmillergroup.github.io/IMCDataAnalysis/ and is already highly used by the larger community.
I have presented the developed framework for data analysis at multiple conferences and taught its use at a number of workshops and seminars. Parts of these frameworks were published in a Science Immunology publication (https://www.science.org/doi/10.1126/sciimmunol.abk1692) and is prepared for publication as a protocol paper (https://www.biorxiv.org/content/10.1101/2021.11.12.468357v1).
On the experimental side of the project, I have grown eight breast cancer organoid lines and developed a collection, embedding, cutting and staining workflow for generating imaging mass cytometry data from breast cancer organoids. To estimate organoid growth, I used a large particle FACS employing reference beads to accurately measure organoid size. In parallel, I optimized staining conditions to acquire 3D imaging data using a confocal microscope. To further characterize clinically relevant features of the organoids I optimized embedding and cutting strategies to acquire IHC images of HER2, ER and PR of selected organoid lines.
The main focus of the experimental part of the project was put on developing a drug screening approach to perturb organoid growth using seven commonly used drugs targeting signaling pathways in breast cancer. As part of the drug screening approach, I developed a heavy metal barcoding strategy allowing me to pool 20 conditions per embedding.
I observed that organoids grow in a heterogeneous fashion in which few organoids form large structures while the majority remain in a small potentially senescent state. Depending on the organoid line, different morphologies can be detected. These range from small and dense to large and hollow structures.
Studying the expression of 40 breast cancer relevant proteins in organoid sections revealed clear distinctions in terms of overall and spatial expression patterns between organoid lines. Major differences arise from the expression of cytokeratins and other epithelial markers including EpCAM and E-Cadherin. Studying the expression changes over organoid growth showed an initial lack of ER and PR expression, both markers used for diagnosing the type of breast cancer.
When growing breast cancer organoids under drug treatment, a number of effects can be observed. First, tamoxifen which targets ER signaling showed no influence on growth and protein expression. Second, inhibitors of the EGF signaling pathway showed strong growth inhibition and down-regulation of phosphorylation of S6. Third, MEK and MYC inhibitors show efficient growth reduction without blocking EGF signaling, however, the exact mechanism is not resolved yet.
The obtained results have been presented at seminars at the University of Zurich and ETH Zurich. Currently, I’m preparing a manuscript describing the findings of the time-course treatment of four representative breast cancer organoid lines.
On the computational side, I have developed a deep learning-enabled strategy for segmenting organoids after sectioning as well as an extensive workflow for analyzing IMC data.
Detecting organoid slices in images is challenging due to differences in size and shape, clumping of objects and the presence of biological debris. To tackle this segmentation issue, I have implemented a deep learning-based instance segmentation approach leveraging existing Mask RCNN frameworks. This approach results in higher accuracy than shallow-learning alternatives including watershed segmentation based in Ilastik-generated pixel probability maps.
In addition, I developed and documented an extensive computational framework for the analysis of IMC data as existing analysis approaches were poorly documented or difficult to apply in a unified framework. The developed framework is openly accessible at https://bodenmillergroup.github.io/IMCDataAnalysis/ and is already highly used by the larger community.
I have presented the developed framework for data analysis at multiple conferences and taught its use at a number of workshops and seminars. Parts of these frameworks were published in a Science Immunology publication (https://www.science.org/doi/10.1126/sciimmunol.abk1692) and is prepared for publication as a protocol paper (https://www.biorxiv.org/content/10.1101/2021.11.12.468357v1).
The project has profiled the growth of a number of breast cancer organoid lines, the effect of drug perturbations and extended the framework for computational analysis of spatially-resolved single-cell data. The results of the breast cancer organoid experiments highlight the effect of drug therapies on cancer growth. To better model patient-specific responses, a larger number of organoid lines and an adjusted medium need to be worked on. The developments on the computational side so far facilitate the training of researchers with less bioinformatic experience to perform their own data analysis therefore increasing the impact of imaging mass cytometry as technology on understanding and developing therapies against common diseases.