Cell-Cell Communication Analysis in Healthy and Malignant Bone Marrow Using Comunet Algorithm on Human Scrnaseq Data

Intercellular communication (IC) is essential in healthy bone marrow (BM) and seems to be altered in malignant disease, such as acute myeloid leukemia (AML). Availability of large single-cell (sc) RNAseq data sets facilitates a better understanding of how changes in IC contribute to development and maintenance of malignant neoplasms. Although some attempts have been made to investigate IC between different bone marrow cell types, a large-scale computational approach covering the whole complexity of the communication processes is still missing. We aimed to use our algorithm COMUNET (Cell cOMmunication exploration with MUltiplex NETworks) (Solovey and Scialdone, 2020) to gain a deeper understanding of IC in healthy and AML BM and to generate new data driven hypothesis about changes in BM IC during malignant transformation.

We developed COMUNET to assess cell-cell communication between distinct cell types using single-cell or sorted bulk RNAseq data. In contrast to other algorithms published to date, COMUNET is capable of assessing differential communication between multiple conditions, which allows us to perform multi-sample comparison of communication patterns in healthy and malignant bone marrow. For our analysis, we used two publicly available scRNAseq datasets of healthy (Oetjen et al., 2018) and AML (van Galen et al., 2019) human bone marrow. We analyzed the intercellular activity of each cell type and compared IC changes between several healthy individuals, as well as AML patients at diagnosis, during therapy, and in remission.

We first analysed the similarity of communication among healthy bone marrow samples from several individuals by performing pairwise comparisons. Here two parameters were the most important: i) whether a ligand-receptor pair is present in both samples, and ii) if the ligand-receptor pair is used by the same populations of cells in both samples. We found that in healthy bone marrow samples, there is a cluster of ligand-receptor pairs, which is present in all analysed samples and is used by similar cell populations. These ligand-receptor pairs most probably represent the base line communication pattern. We could also identify ligand-receptor pairs that were used specifically in some of the samples, which might be explained by differences in haematopoietic status between individuals. We then proceeded to analyse the communication activity of individual cell types. Of the 8 analysed hematopoietic populations, monocytes showed the highest variety of ligands and receptors used, compared to B-cells, cytotoxic T-lymphocytes (CTL), granulocyte-monocyte progenitor (GMP), natural killer (NK), plasma cells, early and late erythrocytes. In younger bone marrow, there was a tendency for a higher number of ligands and receptors used by each population, as well as a slight increase in the average number of partners for immune cell populations. Immune cell populations showed a higher cumulative activity compared to GMP and erythroid populations.

In the AML samples, we identified a dramatic change of communication patterns in non-tumor cells at the diagnosis. Under treatment, we observed a shift of the communication patterns towards the levels obtained from healthy bone marrow samples, and finally in remission, no statistically significant difference was observed between the remission samples and healthy bone marrow samples. We also observed an increase in the variety of ligands and receptors used and in cumulative activity of immune cell populations at diagnosis, as well as normalization of these parameters in remission.

In conclusion, COMUNET allows us to characterize e IC patterns in scRNAseq BM data and identified disease-driven changes in communication patterns, as well as a normalization of the IC when a complete remission was achieved. While the activity state of individual cell type was not affected by the size of this cell population, we noticed that overall communication measurement is sensitive to population loss due to structural changes in the BM, such that the results largely depended on a harmonized population size of all cell types. This is of great importance when interaction with rare cell populations (e.g. T-cell subsets) is studied and has to be kept in mind during data acquisition. We will continue to analyse more data sets and further develop our algorithm to generate new data driven hypotheses for a deeper understanding of haematopoietic neoplasms.