This feature on cell-cell communication is adapted from part of Chapter 4 of the 2023 Spatial and Single-cell Playbook.
Organisms are complex entities, made up of many different tissues. Tissues themselves are also complex entities, made up of many different types of cell. Single-cell transcriptomics has proved incredibly valuable for decomposing tissues into their constituent cell types and for finding rare, novel, cell types. However, cells do not act in isolation and identifying constituent cell types only makes up half the story of what a tissue is.
Tissues are, in fact, dynamic and cells are in constant communication with each other. This communication allows the tissue to function properly by allowing cells to coordinate their activities. Within disease, it is not the activity of a single cell, but the coordination of the whole tissue that often goes awry. If we want to understand how a tissue functions in both health and disease, then measuring cellular interactions is the next layer of information that is necessary to acquire.
This feature will introduce the concept of cell-cell communication and will showcase the different methods that are available to quantify cellular interactions from single-cell RNA sequencing data. It will also introduce novel methods to take advantage of spatial transcriptomic data, to provide a jumping off point for anyone interested in performing this analysis.
How can we measure cell-cell communication?
Cells have various ways of communicating (see Figure 1). This communication typically takes the form of ligand secretion. Here, ‘sender’ cells secrete ligands that can bind to the corresponding receptor proteins on the plasma membrane of ‘receiver’ cells.
Figure 1. Forms of chemical signalling between cells. Image Credit: Biology 2e. Provided by: OpenStax. Located at: http://email@example.com. License: CC BY: Attribution. License Terms: Access for free at https://openstax.org/books/biology-2e/pages/1-introduction
Ligands can bind to receptors on the same cell that secreted them, which is known as autocrine signalling. However, ligands mostly bind to receptors on other cells in the immediate environment of the sender, known as paracrine signalling. Ligands can also pass directly between cells if there is a physical connection between cells, known as gap junctions. Autocrine, paracrine and gap signalling all have a distance limit, as ligands can only travel so far. Endocrine signalling, on the other hand, allows cells to communicate from any location in the body, secreting hormones into the bloodstream to reach distant targets.
Paracrine and endocrine signalling are the most informative forms of signalling for gauging how cells are communicating with each other. These types of communication happen over physical distances, and, unfortunately, we currently do not have the technological capacity to track ligands and hormones in real-time to truly measure cell-cell communication. Hence, the analysis of cell-cell communication with modern technology is limited.
Current methods, instead, measure cell-cell communication as an inferred analysis. It is performed downstream from single-cell sequencing and spatial data, and uses the expression of genes and proteins for specific cell messengers (ligands) and the complementary receptors. This can isolate which cell population could be in paracrine communication with each other.
Single-cell methods for cell-cell communication
Single-cell transcriptomic or proteomic data provides expression levels of known ligands and known ligand-receptors within cell clusters. Using databases of known ligand-receptor interactions, paracrine cell-cell communication can then be inferred based on expression of ligands and their corresponding receptors (see Figure 2).
Figure 2. Principles of cell–cell communication inference. (A) Cells can secrete ligands that diffuse and can bind to receptors. This is likelier to occur for receiver cells that are closest to the sender cell and when there is sufficient receptor expression. The blue and orange cells represent different cell types. For the blue cells, darker shades represent stronger ligand expression. (B) Cell–cell communication can be inferred from scRNA-seq at either the individual cell or cell cluster level, but spatial distances between cells are lost. (C) Using spatial transcriptomics to infer cell–cell communication preserves spatial distances between cells but potentially at the loss of single-cell or gene resolution. Figure and Legend taken from Almet, et al. 1
Many tools exist to estimate communication from single-cell data. They can be classified into two classes of methods. There are (1) methods that simply measure the levels of ligands and of receptors in cell clusters to ascertain the level of communication between these cells; and (2) methods that estimate downstream intracellular activities (e.g., gene expression changes) alongside ligand-receptor expression rates to try to estimate whether a ligand has bound and affected genomic processes in the cell. Cell-cell communication should, in principle, alter the transcriptomic state of the receiver cell, and the second class of methods tries to validate the inferred communication by measuring the downstream effects too.
All methods operate on a basic assumption; that transcriptomic data is a good proxy for cell-cell communication events. This is a proxy because paracrine cell-cell communication happens via proteins (not RNA) and is spatially constrained (something single-cell RNA-seq data cannot represent). Hence, a level of caution is necessary when interpreting the output of cell-cell communication methods.
The first class of tools use ligand/receptor expression estimates. They are the most common. Several popular tools in this category include CellPhoneDB 2 and CellChat 3, which have well curated ligand-receptor databases to infer cell communications. These tools assess co-expression of ligands and receptors across cell populations and use permutation tests to determine whether there is significant co-expression between specific cell types. As stated, these tools assume that RNA expression of ligand protein in cell A and RNA expression of receptor protein in cell B is a proxy for functional communication between cell types. The curated ligand-receptor databases inform the tool as to which ligands and receptors are known to bind.
These methods estimate cellular communication between cell populations rather than between cells of a population, which is realistically where a lot of communication will occur. A new tool, NICHES 4, tackles this problem and takes advantage of the statistical power provided by single-cell data. While most methods average ligand expression across a cell cluster as a measure of the cell type’s ligand expression, NICHES, by comparison, operates at the single-cell level, measuring the interactions of every cell with another cell, comparing them in an iterative pairwise manner, measuring true cell-cell communication. Finding a small number of cells expressing the complementary ligand-receptor pairs implies a highly specific communication network.
An overview and benchmarking of the major ligand/receptor tools in this category was carried out in 2022, which the reader should refer to5. The outcome of the review process saw the creation of a new tool – LIANA – that takes single-cell RNA data and establishes a common interface with all methods and resources that exist for cell-cell communication, providing a consensus ranking for the method’s prediction (Figure 3). This review found that the ligand/receptor methods that estimate cell-cell communication between populations make cell-cell communication inferences that are robust and concordant with both proteomic and spatial ligand/receptor data5, so should be taken seriously.
Figure 3. Overview of LIANA. LIANA takes any annotated scRNA dataset and establishes a common interface to all the resources and methods in any combination. LIANA also provides a consensus ranking for the method’s predictions. Image Credit: https://saezlab.github.io/liana/
Downstream Signalling Tools
For the second class of tool, those which estimate the downstream effects of signalling, NicheNet 6 is the most well-known. NicheNet, like the first class of tools, assumes RNA expression of ligand and receptor is proxy for the proteins and functional communication. It goes beyond the other methods by also assuming that the receiver cells (with the receptors) will experience a signal propagation affecting master gene regulators and transcription factors, which can be measured. Hence, enrichment of downstream target genes is also incorporated into the measure of cell-cell communication, and NicheNet makes use of prior knowledge databases of downstream gene effects to do this.
This approach effectively prioritises cell-cell communication inferences based on downstream biology; hence it helps reduce the large number of ligand-receptor interactions generated by the first class of tools. However, by prioritising certain interactions, it inevitably ignores potentially true cell-cell interactions that don’t have the expected downstream effects according to the database.
Recently, NicheNetv2 and MultiNicheNet have also been released7. NicheNetv2 adds experimentally determined target genes for over 100 ligands, grounding the predicted downstream effects in biology. MultiNicheNet tackles a problem in the cell-cell communication field – appropriately analysing multiple samples and conditions. In brief, MultiNicheNet infers the differentially expressed ligand-receptor pairs between conditions and the downstream target genes, while accounting for inter-sample heterogeneity.
Spatial methods for cell-cell communication
A biological reality for secreted ligands is that they can only travel a certain distance once secreted 1. With the influx of spatial transcriptomic data, the opportunity is arising for a more accurate assessment of cell-cell communication. With x and y coordinates for transcripts and cells, we can calculate communication measures within a defined niche of cells around the sender cell, to which the ligands can disperse. This measures communication in the way it actually operates. Figure 4 displays the tasks and outputs of spatial analysis of cell-cell communication.
Figure 4. Spatial Analysis of cell-cell communication (A) Integrating spatial and scRNA-seq data provides combined information and can allow cell type assignment and spatial distance estimate data (B) Current spatial cell–cell communication inference methods output: a cell–cell or cluster–cluster network due to ligand–receptor binding and more general intercellular gene regulatory networks in space. Figure taken from Almet, et al. 1
One tool mentioned earlier, NICHES, was also built to be used for spatial data. Microenvironment signalling is assessed by limiting the analysis of ligands and receptors to spatial neighbours. Other tools such as CellPhoneDBv3 have also expanded to include spatial information by restricting ligand-receptor analysis to defined cellular neighbourhoods, and NicheNet will likely go the same way.
Specific methods have been developed to take advantage of spatial transcriptomic data. Fundamental spatial packages such as Squidpy 8, MISTy9 and Giotto10 have ligand and receptor analysis methods using the spatial context. There are also graph-based methods, such as SpaTalk 11, which evaluate intercellular and intracellular communication. Newer mathematical methods, such as COMMOT 12, look promising. Rather than assessing ligand and receptor levels, these methods work out optimal transport systems for ligands within spatial constraints, and use them to predict cell communication patterns across a tissue.
Combining single-cell and spatial data from the same tissue is another promising avenue, using the cell type annotations of the former and the spatial context of the latter (See Figure 4A). A tool, Renoir 13, has recently been released for this purpose, and can infer communication niches, as well as identify the major ligands in each niche, uncovering key players in cell communications.
There are challenges with current spatial methodologies to assess cell-cell communications. Ideally, spatial methods require transcriptomic data that is recorded at subcellular resolution and is transcriptome-wide. There are several technologies that are transcriptome-wide, but these often lack spatial resolution, meaning multiple cells are represented by one ‘spot’, limiting the ability to measure cell-cell communication. Several technologies are also quantifying transcripts at subcellular resolution, but are limited in the number of targets they can assess, and transcriptome-wide measurements are needed for these enrichment-based approaches.
Ultimately, to get the most out of spatial data for cell-cell communication, it needs to be hi-plex, allowing the visualisation of many ligand and receptor genes and/or proteins in the same section. As single-cell and spatial proteomics develop to include more targets, we can hope that direct ligand and receptor protein assessment will be commonplace for assessing cell-cell communication.
If you would like to know more, there are valuable reviews on cell-cell communication that you can read1,5,14. Furthermore, the Single-Cell Best Practices resource15 presents a step-by-step walkthrough of popular cell-cell communication methods such as CellPhoneDB and NicheNet.
1. Almet, A.A., Cang, Z., Jin, S. & Nie, Q. The landscape of cell–cell communication through single-cell transcriptomics. Current opinion in systems biology 26, 12-23 (2021).
2. Efremova, M., Vento-Tormo, M., Teichmann, S.A. & Vento-Tormo, R. CellPhoneDB: inferring cell–cell communication from combined expression of multi-subunit ligand–receptor complexes. Nature protocols 15, 1484-1506 (2020).
3. Jin, S. et al. Inference and analysis of cell-cell communication using CellChat. Nature communications 12, 1088 (2021).
4. Raredon, M.S.B. et al. Comprehensive visualization of cell–cell interactions in single-cell and spatial transcriptomics with NICHES. Bioinformatics 39, btac775 (2023).
5. Dimitrov, D. et al. Comparison of methods and resources for cell-cell communication inference from single-cell RNA-Seq data. Nature communications 13, 3224 (2022).
6. Browaeys, R., Saelens, W. & Saeys, Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nature methods 17, 159-162 (2020).
7. Browaeys, R. et al. MultiNicheNet: a flexible framework for differential cell-cell communication analysis from multi-sample multi-condition single-cell transcriptomics data. bioRxiv, 2023.06. 13.544751 (2023).
8. Palla, G. et al. Squidpy: a scalable framework for spatial omics analysis. Nature methods 19, 171-178 (2022).
9. Tanevski, J., Flores, R.O.R., Gabor, A., Schapiro, D. & Saez-Rodriguez, J. Explainable multiview framework for dissecting spatial relationships from highly multiplexed data. Genome biology 23, 1-31 (2022).
10. Dries, R. et al. Giotto: a toolbox for integrative analysis and visualization of spatial expression data. Genome biology 22, 1-31 (2021).
11. Shao, X. et al. Knowledge-graph-based cell-cell communication inference for spatially resolved transcriptomic data with SpaTalk. Nature Communications 13, 4429 (2022).
12. Cang, Z. et al. Screening cell–cell communication in spatial transcriptomics via collective optimal transport. Nature Methods 20, 218-228 (2023).
13. Rao, N. et al. Charting spatial ligand-target activity using Renoir. bioRxiv, 2023.04. 14.536833 (2023).
14. Liu, Z., Sun, D. & Wang, C. Evaluation of cell-cell interaction methods by integrating single-cell RNA sequencing data with spatial information. Genome Biology 23, 1-38 (2022).
15. Heumos, L. et al. Best practices for single-cell analysis across modalities. Nature Reviews Genetics 24, 550-572 (2023).