Modeling and inference of spatial intercellular communications and multilayer signaling regulations using stMLnet

Jinyu Cheng, Lulu Yan, Qing Nie, Xiaoqiang SunbioRxiv 2022 · DOI

TL;DR

This paper introduces stMLnet, a novel computational framework that leverages spatial transcriptomics data to model and infer complex, multi-layered cell-to-cell communication (CCC) networks. By integrating spatial information with molecular interaction data, stMLnet provides a more accurate and mechanistic understanding of how cells communicate and regulate gene expression within their tissue environment, outperforming existing methods.

What Problem Does This Paper Address?

Understanding how cells communicate within a tissue is fundamental to comprehending complex biological processes like development, immunity, and disease. While many methods exist to infer cell-cell communication (CCC) based on gene expression data, they often fall short in two critical areas: 1) they typically ignore the spatial context of cells, and 2) they fail to fully capture the intricate, multi-step signaling cascades that occur both between cells (intercellular) and within cells (intracellular). The advent of spatial transcriptomics (ST) technologies, which provide gene expression data alongside precise spatial coordinates of cells, offers a unique opportunity to address these limitations. However, effectively utilizing this rich spatial information to build comprehensive CCC models remains a significant challenge.

Methods

stMLnet tackles this challenge by employing a four-component framework:

  1. Prior Network Construction: It integrates existing knowledge about molecular interactions from various databases, creating comprehensive knowledge graphs for ligand-receptor (LigRecDB), receptor-transcription factor (RecTFDB), and transcription factor-target gene (TFTGDB) relationships.
  2. Multilayer Network Inference: It constructs a multilayer signaling network by combining prior knowledge with gene expression data. This involves identifying potential ligand-receptor pairs based on co-expression and spatial proximity, and then inferring downstream signaling pathways involving transcription factors and target genes.
  3. Spatially Dependent LR Signaling Activity Quantification: This is a key innovation. stMLnet models the signaling strength of ligand-receptor interactions as a function of the physical distance between cells. It uses a diffusion model for ligands and the law of mass action to quantify how the spatial arrangement of cells influences communication strength. Crucially, it employs a reciprocal relationship between distance and signaling activity, which is shown to be more effective than constant or radial basis functions.
  4. LR-Target Regulation Quantification: To understand how intercellular signals translate into intracellular gene expression changes, stMLnet uses an explainable tree-based machine learning model (random forest regression). This model links the quantified LR signaling activity to target gene expression, providing importance scores that reveal the contribution of specific ligands, receptors, or LR pairs to regulating target genes. This allows for prioritization of key regulators and identification of complex regulatory circuits.

Datasets

The study utilized a variety of datasets for benchmarking and application:

  • Simulation Data: 100 sets of synthetic data were generated to rigorously test the performance of different distance-weighting functions within stMLnet.
  • Perturbation Datasets: 8 real datasets of cell type-specific perturbations, primarily from breast cancer cell lines (e.g., TGFB1 or IL6 stimulation, NRP1 knockout, CXCR4 mutation), were used to benchmark LR-target prediction accuracy against known responses.
  • Spatial Transcriptomics Datasets: Six ST datasets from four different technologies were used to demonstrate stMLnet's applicability and robustness across varying spatial resolutions and gene coverages:
  • seqFISH+ (mouse cortex)
    • Slide-seq v2 (mouse hippocampus)
    • MERFISH (mouse hypothalamus)
    • Visium (breast cancer, glioma, and COVID-19 lung tissue)
  • RNA-seq Data: RNA-seq data of macrophages isolated from mice (GSE69104) was used to verify CSF1R-target regulation predictions in the glioma dataset.

The paper also mentions the use of publicly available datasets from sources like 10x Genomics, SeuratData, and NCBI GEO.

Key Insights and Ideas

  • Spatial context is crucial for CCC: The paper strongly emphasizes that ignoring cell location leads to incomplete and potentially inaccurate CCC models. stMLnet's distance-dependent signaling model is a significant step towards integrating this spatial information.
  • Multilayer networks capture complexity: Moving beyond simple ligand-receptor pairs to include transcription factors and target genes provides a more mechanistic view of how intercellular signals are transduced intracellularly.
  • Reciprocal distance weighting is superior: The study demonstrates that a reciprocal relationship between cell-cell distance and signaling strength is more effective than constant or radial basis functions for modeling spatial CCC. This suggests that closer cells don't just have a higher chance of communicating, but the strength of that communication is inversely proportional to distance.
  • Positive feedback loops in disease: stMLnet identified significant positive feedback circuits in the COVID-19 lung microenvironment involving alveolar epithelial cells, macrophages, and monocytes, which may contribute to hyperinflammation.
  • Explainable AI for biological insights: The use of random forest regression allows not only for prediction but also for quantifying the importance of individual regulators, providing actionable insights into key signaling pathways.

Key Results

  • Benchmarking Performance: stMLnet significantly outperformed existing methods (NicheNet, CytoTalk, MISTy) in predicting LR-target regulations on multiple perturbation datasets, achieving AUROC scores greater than 0.72 and AUPRC greater than 0.55 on most datasets.
  • Spatial Model Validation: The reciprocal distance-weighting function in stMLnet showed superior performance in both synthetic and real-world perturbation data compared to constant or radial basis functions.
  • Application to Diverse ST Data: stMLnet demonstrated effectiveness and reliability across various ST technologies (seqFISH+, Slide-seq v2, MERFIS, Visium) and spatial resolutions.
  • COVID-19 Insights: The analysis of COVID-19 lung tissue revealed positive feedback circuits between alveolar epithelial cells, macrophages, and monocytes, suggesting a role in hyperinflammation.
  • Glioma Immunotherapy: In gliomas, stMLnet identified key interactions between glioma cells and macrophages, including CSF1R and IGF1 signaling, and highlighted the role of NFKB1 in downstream pathways, aligning with experimental findings.

Limitations and Future Work

  • Simplified Assumptions: The paper acknowledges several simplifications in its diffusion model, such as approximating ligand concentration with ligand gene expression and using receptor gene expression to represent protein levels.
  • TF-Target Information: The TF-target information is currently derived from general databases and is not cell-type specific, which could be a limitation.
  • Future Directions: The authors propose integrating ST data with emerging single-cell multi-omics data (e.g., proteomics, ATAC-seq) to develop more comprehensive multiscale models for better CCC and gene-gene regulation inference. They also plan to develop novel multiscale models that integrate multi-omics data across multiple layers.

Summary updated

Abstract

Abstract Multicellular organisms require intercellular and intracellular signaling to coordinately regulate different cell functions. Although many methods of cell-cell communication (CCC) inference have been developed, they seldom account for both the intracellular signaling responses and global spatial information. The recent advancement of spatial transcriptomics (ST) provides unprecedented opportunities to better decipher CCC signaling and functioning. In this paper, we propose an ST -based m ultilayer net work method, stMLnet, for inferring spatial intercellular communication and multilayer signaling regulations by quantifying distance-weighted ligand–receptor signaling activity based on diffusion and mass action models and mapping it to intracellular targets. We benchmark stMLnet with existing methods using simulation data and 8 real datasets of cell type-specific perturbations. Furthermore, we demonstrate the applicability of stMLnet on six ST datasets acquired with four different technologies (e.g., seqFISH+, Slide-seq v2, MERFIS and Visium), showing its effectiveness and reliability on ST data with varying spatial resolutions and gene coverages. Finally, stMLnet identifies positive feedback circuits between alveolar epithelial cells, macrophages, and monocytes via multilayer signaling pathways within a COVID-19 microenvironment. Our proposed method provides an effective tool for predicting multilayer signaling regulations between interacting cells, which can advance the mechanistic and functional understanding of spatial CCCs.
BibTeX
@misc{cheng2022modeling,
  author = {Jinyu Cheng and Lulu Yan and Qing Nie and Xiaoqiang Sun},
  title = {{Modeling and inference of spatial intercellular communications and multilayer signaling regulations using stMLnet}},
  year = {2022},
  doi = {10.1101/2022.06.27.497696},
  howpublished = {bioRxiv},
}
Figure