Spatially-resolved multiscale models shed light into personalized drug treatments

Spatially-resolved multiscale models shed light into personalized drug treatments Alejandro Madrid#1, Alfonso Valencia#*2, Arnau Montagud#1 #Barcelona Supercomputing Center (BSC), Plaça Eusebi Güell, 1-3, Barcelona, Spain 1[email protected], 3[email protected] *ICREA, Pg. Lluís Companys 23, Barcelona, Spain 2[email protected] Keywords— agent-based simulations, spatial transcriptomics, cancer Extended ABSTRACT Multiscale models have been very helpful in tissue biology by providing novel hypotheses to uncover mechanisms and novel treatments that tackle diseases of interest. PhysiBoSS is an open-source software which combines intracellular signalling using Boolean modelling (MaBoSS) and multicellular behaviour using agent-based modelling (PhysiCell) [1]. Since 2018, it has been successfully used in tissue simulation of diseases such as cancer and COVID. PhysiBoSS can use Boolean models personalized using bulk omics data and its simulation can be set-up using manuallydesigned 3D architectures defining an extracellular matrix (ECM). In spite of these advances, current simulations use rather simplistic 3D set-ups and much remains to be done to have a simulation that accurately produces results that can seem like the real tissue. To have spatially-resolved personalized multiscale models, we have taken advantage of a novel technique: single cell spatial transcriptomics. This new technique allows to have the spatial information of unique cells combined with its transcriptomic state [2]. We hereby present a user-friendly workflow called “PhysiBoSS-spatial” for setting up PhysiBoSS simulations using spatial transcriptomics data. This workflow enables the translation of a slice of spatial transcriptomics in the initial disposition of the multiscale model. Users can modulate the number of cell types to be captured (and which), the resolution of the clustering and the cell-type annotation. To showcase the use of this workflow, we hereby present the set-up of the simulations using breast cancer spatial omics datasets and their simulation with different drugs treatments. A. Clustering and annotation of the spatial omics datasets First, we use the raw spatial omics dataset and re-analise them to know which subtype of cells are present in the data. For this, our pipeline processes and analyses the datasets using Scanpy [3]. For the clustering, the pipeline allows the user to choose from Scanpy internal metrics or the ESTIMATE score. ESTIMATE is a tool which provides scores for tumor purity, stromal cells presence and infiltration level of immune cells in tumor tissues [4]. Additionally, immune and endothelial cells are annotated using the Python package CellTypist to increase the accuracy identifying these cell types due to their low number in some samples [5]. B. Boolean Models personalization PhysiBoSS embeds Boolean Models of signaling pathways in each agent to integrate genetic and environmental cues to their phenotype simulation. To have personalized Boolean models, we modified the tool PROFILE [6] with our breast Fig. 1. Initial stage of the simulation from breast cancer spatial transcriptomics data (a) Clustering and annotation analysis of a breast cancer sample of spatial transcriptomics. Three cancer subtypes are shown with the endothelial cells. (b) Initial setup in PhysiBoSS of the tissue using our workflow, where the endothelial cells are displayed as a source of oxygen. (c) Same figure than b, but with the ECM depicted in yellow. cancer spatial omics. As a first trial, we decided to personalize 3 different cancer subtypes and the endothelial cells. Thanks to the Boolean Model personalization our workflow recovers another layer of heterogeneity from the spatial transcriptomic data. C. Creating the set up for PhysiBoSS The pipeline then combines the spatial positions of each subtype of cells with the personalized Boolean Models to create all the files used for the initialization of the simulation. We have to highlight that in this first version we identify the endothelial cells as microvasculature where the oxygen, nutrients and drugs will appear in the simulation, each one with its own biophysical properties (Figure 1). Our pipeline is also able to represent the extracellular matrix in our simulations as an obstacle that the cells cannot move through unless they produce metalloproteases to degrade it using an addon created by Ruscone et al. [7]. In this first version, we consider ECM to be present in the positions with the lowest (or no) level of RNA counts (see an unit test in Figure 2). Fig. 2. Unit test for the interaction between ECM and cells. (a) Final stage of a simulation without cell-ECM interaction (b) Final stage of a simulation with cell-ECM interaction, where the cells cannot move through the ECM. D. Discussion and future steps PhysiBoSS-spatial is able to faithfully reconstruct a spatial omics 2D slice as an initial point for a multiscale simulation by integrating different information from the spatial transcriptomics data set: spatial position, cell type, and the transcriptomics to obtain personalized Boolean models. With this workflow users can create all the files needed to perform PhysiBoSS simulations in a straightforward way without having prior experience in spatial transcriptomic analysis or ESTIMATE or PROFILE tools. For the advanced users, our workflow allows the modification of the most important parameters in the scripts of the pipeline. We will improve the PhysiBoSS-spatial workflow in the future and we will focus on the annotation of spatial omics to create more complex and heterogeneous simulations by including other cell types, like immune and stromal cells, and a better characterization of the cancer subpopulations. ACKNOWLEDGEMENTS This work has been partially supported by the European Commission under the PerMedCoE (H2020-ICT-951773). References [1] Letort, G., Montagud, A., Stoll, G., Heiland, R., Barillot, E., Macklin, P., ... & Calzone, L. (2019). PhysiBoSS: a multi-scale agent-based modelling framework integrating physical dimension and cell signalling. Bioinformatics, 35(7), 1188-1196. [2] Rao, A., Barkley, D., França, G. S., & Yanai, I. (2021). Exploring tissue architecture using spatial transcriptomics. Nature, 596(7871), 211-220. [3] Wolf, F. A., Angerer, P., & Theis, F. J. (2018). SCANPY: large-scale single-cell gene expression data analysis. Genome biology, 19, 1-5. [4] Yoshihara, K., Shahmoradgoli, M., Martínez, E., Vegesna, R., Kim, H., Torres-Garcia, W., ... & Verhaak, R. G. (2013). Inferring tumour purity and stromal and immune cell admixture from expression data. Nature communications, 4(1), 2612. [5] Domínguez Conde, C., Xu, C., Jarvis, L. B., Rainbow, D. B., Wells, S. B., Gomes, T., ... & Teichmann, S. A. (2022). Cross-tissue immune cell analysis reveals tissuespecific features in humans. Science, 376(6594), eabl5197. [6] Montagud, A., Béal, J., Tobalina, L., Traynard, P., Subramanian, V., Szalai, B., ... & Calzone, L. (2022). Patient-specific Boolean models of signalling networks guide personalised treatments. Elife, 11, e72626. [7] Ruscone, M., Montagud, A., Chavrier, P., Destaing, O., Bonnet, I., Zinovyev, A., ... & Calzone, L. (2022). Multiscale model of the different modes of cancer cell invasion. bioRxiv, 2022-10. Alejandro Madrid Valiente received his Bsc in Biochemistry and Biomedical Sciences from the Universitat de València, Spain in 2021. Now he is doing his MSc in Bioinformatics for Helath Sciences at the Universitat Pompeu Fabra, Spain. He is working in his MSc Thesis at the Computational Biology group of the Barcelona Supercomputing Center (BSC).