MultiNicheNet: modeling cell-cell communication from multi-sample multi-condition single-cell transcriptomics data Problem setting Studying intercellular communication is essential for an improved understanding of tissue biology and disease pathophysiology. Advances in single-cell and spatial transcriptomics help address this need through their ability to generate molecular profiles of cells within a tissue1. Several computational approaches have been developed to investigate cell-cell communication from these profiles2. Most tools infer cell-cell communication by predicting interactions between ligands (membrane-bound or secreted extracellular proteins) expressed by sender cell types and receptors expressed by receiver cell types3. These tools provide a comprehensive list of potential expressed ligand-receptor pairs, but they don’t provide a functional understanding of cell-cell communication because they don’t infer the response of the receiver cell type in reaction to the ligand-receptor binding. Other tools approach the cell-cell communication inference problem differently by incorporating downstream signaling of ligand-receptor interactions. We previously developed NicheNet (https://github.com/saeyslab/nichenetr) which predicts downstream affected target genes of expressed ligand-receptor pairs by combining the expression data of interacting cells with a model of ligand-target regulatory potential4. NicheNet then prioritizes expressed ligand-receptor pairs according to how strongly their predicted targets are enriched in the receiver cell type (their so-called ligand activity). Both types of approaches have been applied successfully to study both communication in steady state and differences in communication between conditions. In the context of differential cell-cell communication analysis, the first type of tools will prioritize differential cell-cell communication patterns based on the differential expression of the ligand-receptor pairs. In contrast, NicheNet predicts “differentially active” ligand-receptor interactions for which prior knowledge supports that they could function upstream of the DE genes in a receiver cell type of interest. However, both prioritization approaches might be useful. Moreover, both types of tools suffer from additional limitations when applied to infer differential cell-cell communication from multi-sample scRNA-seq data (e.g., from a cohort of several patients and healthy controls). Running the current cell-cell communication tools in their default mode on multi-sample data will generate results after pooling all cells across samples. This approach is statistically inadequate because it ignores sample-to-sample variation. These issues have already been discussed extensively in the context of classic differential expression (DE) analyses5–7. Noteworthy, this pooling procedure is also suboptimal from a biological perspective because it ignores that cell-cell communication occurs within one sample. This is an important issue because of the expected rise in multi-sample datasets due to technological advances, for example, in sample multiplexing8. In parallel to this evolution, more and more datasets are added to existing atlases in projects like the Human Cell Atlas9. These atlases consist of several healthy and diseased samples of several tissues from multiple individuals. Deciphering the role of cell-cell communication in the pathogenesis of these diseases requires tools that can correct for the source of origin of the data and relevant clinical covariates. Ideally, these tools should also be able to exploit the wealth of these multi-sample multi-condition datasets and tackle more complex questions than just pairwise comparisons (such as comparing therapy response or disease progression over time between several diseases). In summary, there is a need for dedicated differential cell-cell communication tools that consider both the expression and activity of ligand-receptor pairs and that can handle the challenges and exploit the opportunities of multi-sample scRNA-seq datasets. Aim and rationale for the approach To address this need, we propose MultiNicheNet (https://github.com/saeyslab/multinichenetr), a novel tool for differential cell-cell communication analysis from multi-sample multi-condition scRNA-seq data. The rationale behind MultiNicheNet is to build upon the principles of state-of-the-art approaches for DE analysis of multi-sample scRNA-seq data6. As a result, the algorithm considers inter-sample heterogeneity, can correct for batch effects and covariates, and can cope with complex experimental designs to address more challenging questions than pairwise comparisons. Details of the suggested approach The main idea behind MultiNicheNet’s prioritization strategy is to uncover essential interactions by considering several complementary aspects informative for cell-cell communication inference. As ideal ligand-receptor pairs, we consider those that are more strongly expressed in the condition of interest, for which predicted target genes are enriched in the receiver cell type, that are also cell-type specific, and present in most samples of the condition of interest. These criteria are calculated by applying state-of-the-art DE approaches like muscat6. The MultiNicheNet software package does not only provide this prioritization framework, but it also provides possibilities for further downstream analyses and the generation of several intuitive visualizations. These visualizations let users explore the data behind the predictions, which is essential to inform them before proceeding to experimental validation. We applied MultiNicheNet to scRNA-seq data of several tissues and diseases (breast cancer, squamous cell carcinoma, MIS-C, and lung fibrosis)10–13. These applications demonstrate that MultiNicheNet both retrieves known biology and generates novel hypotheses, including the possible identification of previously undescribed subgroups of patients. Additional data modalities, such as spatial co-localization from spatial transcriptomics data and proteomics, were used to further validate some of the top predictions. How it will affect the broader field We anticipate that MultiNicheNet will be a useful tool for studying dysregulated cell-cell communication patterns from patient cohort scRNA-seq data. This might lead to improved insights into disease pathogenesis, indicate potential treatment strategies, and identify potential biomarkers for patient stratification. References 1. Tanay, A. & Regev, A. Scaling single-cell genomics from phenomenology to mechanism. Nature 541, 331–338 (2017). 2. Armingol, E., Officer, A., Harismendy, O. & Lewis, N. E. Deciphering cell–cell interactions and communication from gene expression. Nat. Rev. Genet. 1–18 (2020) doi:10.1038/s41576-020-00292-x. 3. Dimitrov, D. et al. Comparison of methods and resources for cell-cell communication inference from single-cell RNA-Seq data. Nat. Commun. 13, 3224 (2022). 4. Browaeys, R., Saelens, W. & Saeys, Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat. Methods 17, 159–162 (2020). 5. Squair, J. W. et al. Confronting false discoveries in single-cell differential expression. Nat. Commun. 12, 5692 (2021). 6. Crowell, H. L. et al. muscat detects subpopulation-specific state transitions from multi-sample multi-condition single-cell transcriptomics data. Nat. Commun. 11, 6077 (2020). 7. Zimmerman, K. D., Espeland, M. A. & Langefeld, C. D. A practical solution to pseudoreplication bias in single-cell studies. Nat. Commun. 12, 738 (2021). 8. Mylka, V. et al. Comparative analysis of antibody- and lipid-based multiplexing methods for single-cell RNA-seq. Genome Biol. 23, 55 (2022). 9. Regev, A. et al. The Human Cell Atlas. eLife 6, e27041. 10. Bassez, A. et al. A single-cell map of intratumoral changes during anti-PD1 treatment of patients with breast cancer. Nat. Med. 27, 820–832 (2021). 11. Ji, A. L. et al. Multimodal Analysis of Composition and Spatial Architecture in Human Squamous Cell Carcinoma. Cell 182, 497-514.e22 (2020). 12. Hoste, L. et al. TIM3+TRBV11-2 T cells and IFNγ signature in patrolling monocytes and CD16+ NK cells delineate MIS-C. J. Exp. Med. 219, e20211381 (2021). 13.Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573-3587.e29 (2021).

4D-NicheNet: modeling of spatiotemporal cell-cell communication dynamics by using prior knowledge on ligand-to-target signaling paths Problem setting Studying intercellular communication is essential for an improved understanding of tissue biology and disease pathophysiology. Advances in single-cell and spatial transcriptomics help address this need through their ability to generate molecular profiles of cells within a tissue1. Several computational approaches have been developed to investigate cell-cell communication from these profiles2,3. Most tools infer cell-cell communication by predicting interactions between ligands (membrane-bound or secreted extracellular proteins) expressed by sender cell types and receptors expressed by receiver cell types. Whereas these methods provide a comprehensive overview of potential expressed ligand-receptor pairs, they don’t provide a functional understanding of cell-cell communication because they don’t infer the response of the receiver cell type in reaction to the ligand-receptor binding. To address this problem, we previously developed the computational method NicheNet (https://github.com/saeyslab/nichenetr)5. NicheNet starts from human or mouse gene expression data of interacting cells and combines this with a prior model calculated through integrating prior knowledge about ligand-to-target signaling paths. Specifically, we applied graph algorithms like Personalized PageRank6 on an integrated network of known ligand-receptor, signal transduction, and gene regulatory interactions to link ligand-receptor pairs to the target genes they might influence. NicheNet then uses this general prior model to interpret the data of putatively interacting cells: it prioritizes which ligands are most likely to influence the expression in the receiver cell , which target genes are affected by each prioritized ligand, and which signaling mediators may be involved. Our benchmark5 demonstrated that NicheNet can accurately predict active ligands given an observed gene expression signature after in vitro ligand stimulation for the majority of datasets. The accuracy of ligand activity predictions is harder to systematically assess in vivo, but several studies reported experimental validation of some of the top predictions7–12. Although NicheNet is a stepping stone in modeling downstream responses of ligand-receptor interactions, its current version still has several limitations. A first limitation is that the quality of the predictions depends on how representative the prior knowledge of ligand-to-target signaling paths is for the context/cell types under study. We can thus expect that predictions will be less accurate for less-studied ligands and less-studied receiver cell types. Another limitation concerns the types of expression data that NicheNet can be applied to. The basic version of NicheNet was developed to analyze cell-cell communication in steady-state or to study pairwise differences between conditions. This means that it cannot handle other exciting data types for studying cell-cell communication, such as longitudinal datasets. Moreover, the typical datasets analyzed now are scRNA-seq data from dissociated cells, meaning that information about the spatial structure in the tissue is lost. Because of the relevance of tissue architecture in cell-cell signaling, it is another limitation that the current NicheNet algorithm cannot handle spatially resolved transcriptomics data properly. Aim and rationale for the approach The goal of this research project is thus to further develop the NicheNet framework to address the limitations mentioned above. We hope that the developed approach will improve the current ability to model cell-cell communication in space and time, while using accurate and context-appropriate prior knowledge to better link ligand-receptor pairs to their downstream response. Details of the suggested approach The first part of the project involves improving the prior knowledge of ligand-to-target signaling paths. The first idea for this part is to continuously include additional relevant data sources to the currently integrated networks (e.g., cytokine signatures from CytoSig13). The second idea is to provide integration with context-specific data. We are currently exploring how to integrate context-specific gene regulatory networks inferred from complementary data such as (sc)ATAC-seq (e.g., by using tools like SCENIC+). In the near future, we will investigate how to include CRISPR perturbation data. We anticipate that downstream signatures obtained from (context-specific) perturbation of ligands/receptors/transcription factors might greatly improve how we link ligand-receptor interactions to downstream responses in the prior knowledge model. The second part of the project involves extending NicheNet to take advantage of the wealth of information provided in spatial transcriptomics and longitudinal data. The main idea here is to exploit the expected covariance between ligand-receptor expression and target gene expression. For spatial data, we expect that real target genes of ligand-receptor pairs will show higher expression in cells that are physically close to cells that show high expression of the ligand-receptor pair. For temporal data, we expect to observe a delay in time between an increase in ligand-receptor expression in sender cells and an increase in target gene expression in receiver cells. Because some target genes that are induced at an early timepoint will encode for ligands or receptors that might be involved in cell-cell communication at a later timepoint, we might start inferring intercellular signaling feedback and cascade mechanisms over time. In the hope of better discriminating between merely expression correlation between ligand-receptor pairs and target genes, we will use the improved NicheNet model to find links that are supported by plausible prior knowledge. How it will affect the broader field We anticipate a substantial interplay between this computational framework and wet-lab experimental work. First, NicheNet generates hypotheses from omics data to guide wet-lab validation of potentially crucial cell-cell communication patterns in the system under study. In the other direction, NicheNet will greatly benefit from ongoing efforts in experimentally determining downstream effects of important mediators in cell-cell communication (e.g., through CRISPR screens). The power of the approach is that the more and better prior knowledge becomes available, the better cell-cell communication patterns will be deciphered from omics data of interacting cells. In the end, we hereby hope to contribute to an improved understanding of the role of cell-cell signaling in health and disease. References 1. Tanay, A. & Regev, A. Scaling single-cell genomics from phenomenology to mechanism. Nature 541, 331–338 (2017). 2. Almet, A. A., Cang, Z., Jin, S. & Nie, Q. The landscape of cell-cell communication through single-cell transcriptomics. Curr. Opin. Syst. Biol. (2021) doi:10.1016/j.coisb.2021.03.007. 3. Armingol, E., Officer, A., Harismendy, O. & Lewis, N. E. Deciphering cell–cell interactions and communication from gene expression. Nat. Rev. Genet. 1–18 (2020) doi:10.1038/s41576-020-00292-x. 4. Dimitrov, D. et al. Comparison of methods and resources for cell-cell communication inference from single-cell RNA-Seq data. Nat. Commun. 13, 3224 (2022). 5. Browaeys, R., Saelens, W. & Saeys, Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat. Methods 17, 159–162 (2020). 6. Jeh, G. & Widom, J. Scaling Personalized Web Search. in Proceedings of the 12th International Conference on World Wide Web 271–279 (ACM, 2003). doi:10.1145/775152.775191. 7. Bonnardel, J. et al. Stellate Cells, Hepatocytes, and Endothelial Cells Imprint the Kupffer Cell Identity on Monocytes Colonizing the Liver Macrophage Niche. Immunity (2019) doi:10.1016/j.immuni.2019.08.017. 8. Guilliams, M. et al. Spatial proteogenomics reveals distinct and evolutionarily conserved hepatic macrophage niches. Cell 185, 379-396.e38 (2022). 9. Mair, F. et al. Extricating human tumour immune alterations from tissue inflammation. Nature 605, 728–735 (2022). 10. Hoste, E. et al. OTULIN maintains skin homeostasis by controlling keratinocyte death and stem cell identity. Nat. Commun. 12, 5913 (2021). 11. Stakenborg, M. et al. Enteric glial cells favor accumulation of anti-inflammatory macrophages during the resolution of muscularis inflammation. Mucosal Immunol. 1–13 (2022) doi:10.1038/s41385-022-00563-2. 12. Karras, P. et al. A cellular hierarchy in melanoma uncouples growth and metastasis. Nature 1–9 (2022) doi:10.1038/s41586-022-05242-7. 13. Jiang, P. et al. Systematic investigation of cytokine signaling activity at the tissue and single-cell levels. Nat. Methods 18, 1181–1191 (2021).