A multi-tiered map of EMT defines major transition points and identifies vulnerabilities

Indranil Paul, Dante Bolzan, Ahmed M. Youssef, Keith A. Gagnon, Heather Hook, Gopal Karemore, Michael Oliphant, Weiwei Lin, Qian Liu, Sadhna Phanse, Carl A. White, Dzmitry Padhorny, Sergei Kotelnikov, Guillaume P. Andrieu, Amanda X. Chen, Pingzhao Hu, Gerald V. Denis, Dima Kozakov, Brian Raught, Trevor Siggers, Stefan Wuchty, Senthil K. Muthuswamy, Andrew EmilibioRxiv 2021 · DOI

TL;DR

This study presents a comprehensive, multi-layered map of the epithelial-to-mesenchymal transition (EMT) in human mammary epithelial cells, revealing key molecular players and transition points. By integrating data from twelve different "omic" layers across ten time points, the researchers provide an unprecedented resource for understanding this critical cellular process in cancer.

What Problem Does This Paper Address?

Epithelial-to-mesenchymal transition (EMT) is a fundamental cellular process crucial for embryonic development, wound healing, and unfortunately, cancer progression. In cancer, EMT is linked to stemness, metastasis, and resistance to therapy. While many genes associated with EMT are known, a deep, dynamic, and multi-faceted understanding of the molecular events that orchestrate the transition from an epithelial to a mesenchymal state (and back) has been lacking. Specifically, how different molecular layers (like proteins, metabolites, and RNA) interact and contribute to the transition, and how these interactions change over time, remains unclear. This gap hinders the development of effective therapeutic strategies targeting EMT.

Methods

The authors employed a highly integrative, multi-omic approach. Key techniques include:

  • Mass Spectrometry (MS)-based Proteomics: Quantifying proteins across various cellular compartments (whole cell, nucleus, membrane, extracellular vesicles) and secreted proteins. They also performed phosphoproteomics, N-glycoproteomics, acetylomics, and peptidomics.
  • Metabolomics: Untargeted analysis to quantify small molecules and track metabolic reprogramming.
  • Transcriptomics: Measuring mRNA and microRNA levels using microarrays.
  • Single-Cell RNA Sequencing (scRNAseq): To capture cellular heterogeneity and identify transcription factors driving EMT at a single-cell level.
  • Bioinformatics and Systems Biology:
    • Phylogenetic clustering and SOMs (Self-Organizing Maps): To analyze temporal dynamics and identify patterns of co-regulation across different molecular layers.
  • Covariance analysis: To assess the concordance (or discordance) between different molecular layers.
  • Integrative causal modeling (using CausalPath and PCSF): To build hierarchical mechanistic models of EMT and identify key regulatory pathways and vulnerabilities.
  • Pathway enrichment analysis: To understand the functional implications of observed molecular changes.

Datasets

The study generated and analyzed a novel, large-scale dataset from human mammary epithelial cells (MCF10A) undergoing TGFβ-induced EMT. This dataset comprises:

  • Quantifications of over 60,000 molecules in total, spanning multiple time points (0, 4 hrs, 1, 2, 3, 4, 5, 6, 8, and 12 days) and twelve distinct molecular layers (including whole cell, nuclear, membrane, extracellular vesicle, secreted, and phosphoproteome, as well as metabolomics, mRNA, and miRNA).
  • Single-cell RNA sequencing data from 1,913 individual cells.

This dataset is described as an "unprecedented resource" and is made accessible through an interactive website.

Key Insights and Ideas

  • Discordance Between Layers: A major surprise was the surprisingly low concordance between different molecular layers, particularly between mRNA and protein levels, and even among different proteomic layers. This highlights that mRNA abundance is a poor proxy for protein function and that systems behavior cannot be understood by looking at a single layer alone.
  • Metabolic Reprogramming Precedes Transcriptional Changes: Metabolomics data revealed that metabolic reprogramming occurred as early as 4 hours after TGFβ stimulation, preceding significant changes in other layers, suggesting a crucial early role for metabolism in initiating EMT.
  • Post-Transcriptional Regulation is Key: While transcription plays a role, especially in early EMT, the study suggests that post-transcriptional and post-translational mechanisms become increasingly dominant in driving EMT, particularly at later stages.
  • EMT is a Continuum with Distinct States: The data supports the idea of EMT as a continuum but also identifies distinct intermediate states (E/M-1 and E/M-2) with unique molecular signatures.
  • Novel Regulators Identified: The multi-omic approach uncovered potential novel transcription factors and signaling molecules that were not previously linked to EMT, likely because their changes were specific to certain cellular compartments or secreted fractions.

Key Results

  • Comprehensive Molecular Atlas: The study quantified over 60,000 molecules across 12 omic layers and 10 time points, creating a detailed temporal map of EMT.
  • Identification of Transition Points: The analysis identified key transition points in EMT, notably an E→E/M transition around day 2 and an E/M→M transition around day 5.
  • Differential Regulation: Over 10,000 molecules were found to be significantly regulated during EMT. Phosphoproteome and secretome showed the highest fractional changes, indicating extensive intra- and extra-cellular signaling.
  • Metabolic Pathways Activated Early: Pathways like arachidonic acid metabolism, glycerophospholipid metabolism, and lysine degradation were activated within 4 hours of TGFβ stimulation.
  • scRNAseq Reveals Heterogeneity: Single-cell analysis revealed heterogeneous responses to TGFβ and identified potential novel transcriptional regulators of EMT.
  • Causal Modeling Predicts Vulnerabilities: Integrative causal modeling identified key factors driving EMT and predicted vulnerabilities that were experimentally validated using combinatorial inhibition, demonstrating the predictive power of the model.
  • Ligand-Receptor Interactions Mapped: The study identified 67 upregulated and 12 downregulated ligand-receptor pairs, suggesting novel inter-cellular communication pathways involved in EMT.

Limitations and Future Work

  • In Vitro Model: The study was conducted using an in vitro cell line (MCF10A). While a widely used model, future work could validate these findings in more complex in vivo systems and primary human tumors.
  • Temporal Resolution: While ten time points were analyzed, finer temporal resolution might reveal even more dynamic regulatory events.
  • Experimental Validation: While some predictions were experimentally validated (e.g., combinatorial inhibition), further validation of novel identified regulators and pathways is warranted.
  • Interactive Website: The authors provide an interactive website for accessing the data, which is a valuable resource for future hypothesis generation and research. This suggests a future direction of enabling researchers to explore the multi-omic landscape of EMT and identify specific molecular players relevant to their own research questions.

Summary updated

Abstract

Summary Epithelial to mesenchymal transition (EMT) is a complex cellular program proceeding through a hybrid E/M state linked to cancer-associated stemness, migration and chemoresistance. Deeper molecular understanding of this dynamic physiological landscape is needed to define events which regulate the transition and entry into and exit from the E/M state. Here, we quantified >60,000 molecules across ten time points and twelve omic layers in human mammary epithelial cells undergoing TGFβ-induced EMT. Deep proteomic profiles of whole cells, nuclei, extracellular vesicles, secretome, membrane and phosphoproteome defined state-specific signatures and major transition points. Parallel metabolomics showed metabolic reprogramming preceded changes in other layers, while single-cell RNA sequencing identified transcription factors controlling entry into E/M. Covariance analysis exposed unexpected discordance between the molecular layers. Integrative causal modeling revealed co-dependencies governing entry into E/M that were verified experimentally using combinatorial inhibition. Overall, this dataset provides an unprecedented resource on TGFβ signaling, EMT and cancer.
BibTeX
@misc{paul2021multi,
  author = {Indranil Paul and Dante Bolzan and Ahmed M. Youssef and Keith A. Gagnon and Heather Hook and Gopal Karemore and Michael Oliphant and Weiwei Lin and Qian Liu and Sadhna Phanse and Carl A. White and Dzmitry Padhorny and Sergei Kotelnikov and Guillaume P. Andrieu and Amanda X. Chen and Pingzhao Hu and Gerald V. Denis and Dima Kozakov and Brian Raught and Trevor Siggers and Stefan Wuchty and Senthil K. Muthuswamy and Andrew Emili},
  title = {{A multi-tiered map of EMT defines major transition points and identifies vulnerabilities}},
  year = {2021},
  doi = {10.1101/2021.06.01.446492},
  howpublished = {bioRxiv},
}
Figure