Extracting Intercellular Signaling Network of Cancer Tissues using Ligand-Receptor Expression Patterns from Whole-tumor and Singlecell Transcriptomes

Abstract

The burgeoning notion of cellular heterogeneity within a tumor has attracted much attention in cancer research 1,2 . In addition to genetic heterogeneity embodied by the diversity of genomes in each cancer cell due to genome instability 3-5 , it is increasingly recognized that the nongenetic variability of cell phenotypes within an isogenic (clonal) cell population contributes to functional heterogeneity of cancer cells 6,7 . Even not considering the variety from stromal cells (endothelium, stromal fibroblasts and infiltrating immune cells) but just within the tumor parenchyma, the neoplastic cells raise the possibility of functional specialization and cooperation among the cancer cells themselves [8][9][10][11] . Slow proliferating cancer stem cells vs. fast proliferating vs. migratory cells are examples of (overlapping and dynamic) subsets of cells among the neoplastic cells 12,13 . Cooperation would require cell-cell communication which is mainly mediated by physical cell-cell contact, shared extracellular-matrix and secreted diffusible signaling molecules [14][15][16] . While interaction between distinct genetic clones of cancer cells in a tumor, has received attention in past years 17, 18 the diversification of cell phenotypes within the same clone (non-genetic heterogeneity 7,19,20 ) must also be considered as a source of interaction between distinct types of cells. Here we study crosstalk between cells via secreted (paracrine) soluble or matrix proteins as well as cell surface signaling (juxtacrine) proteins that bind specifically to their cognate receptors. Herein the principle of autocrine stimulation in which a cell secretes a growth factor and also expresses its cognate receptor has long been used to explain the observed co-expression of both ligand and receptor in the tumor cell population which was originally considered to be homogeneous and consisting of only one cell type 21 . But the notion of the diversity of cell subtypes among the tumor parenchyma cells which can differ in their capacity to secrete different sets of mediators and to express distinct receptors for these signaling molecules by necessity entails that networks of

Paired-cell sequencing enables spatial gene expression mapping of liver endothelial cells

“…We used a database of known ligand-receptor pairs28,29 to identify potential interactions between these liver sub-populations (Supplementary Fig. 2, Supplementary Data 3).…”

DiSiR: a software framework to identify ligand-receptor interactions at subunit level from single-cell RNA-sequencing data

“…This database is a part of FANTOM5 and includes a table reporting the list of ∼2500 LR pairs over 144 primary cell types, which represents a large-scale map of cell-to-cell communication in human. Zhou et al [26] investigated these interactions in the transcriptomes of ∼4,000 single cells isolated from melanoma patients, including both malignant tumor cells and stromal and immune cells. By systematically comparing the LR pair expression patterns among seven major cell types present in the melanoma-derived single-cell data (melanoma cells, T cells, B cells, macrophages, NK cells, cancer-associated fibroblasts (CAF) and endothelial cells), the authors were able to build a cell-cell communication map indicating which cell type interacts with each other, based on specific LR links.…”

“… (a) Recall and precision, across different threshold values, are defined in relation to positive and negative groups for DiSiR vs. CellPhoneDB and ICELLNET (AUC, area under the curve). Best precision in DiSiR is obtained for Th = 0 (optimal threshold value) (b) Cell-cell interaction network between cancer and the normal cells in the melanoma (our gold-standard reference) extracted in [26]. …”

“…Here, we evaluate the performance of DiSiR using a “gold-standard” LR interaction database [17] from the literature. Zhou et al [26] analyzed these interactions in scRNA-seq data from melanoma patients and generated a cell-cell communication network for melanoma. We use this network as the ground truth and benchmark the performance of DiSiR against well-known available permutation-based methods, e.g., CellPhoneDB [22] and ICELLNET [10].…”

Spatial compartmentalization of signalling imparts source-specific functions on secreted factors

“…In RNAseq datasets, manual curation has been applied to define lists of relevant factors and receptors expressed by specific cell types. From these lists, interactions have been predicted computationally, generating models that represent ongoing interactions (Choi et al, 2015;Efremova et al, 2020;Kumar et al, 2018;Rezza et al, 2016;Verma et al, 2018;Zhou et al, 2017). In all cases, these models were created using data gathered at a single time point, and fail to capture the temporal dynamics of "cellular interactomes" as they exist in vivo over the course of a physiological response.…”