Poster Presentation 35th Lorne Cancer Conference 2023

TrajectoryNet reveals temporal transcriptional relationships driving the mesenchymal-to-epithelial transition in tumour formation (#155)

Shabarni Gupta 1 , Alexander Tong 2 , Manik Kuchroo 3 , Beatriz San Juan 1 , Laura Rangel 1 , Brandon Zhu 4 , John G. Lock 5 , Christine L. Chaffer 1 6 , Smita Krishnaswamy 2 7
  1. The Garvan Institute of Medical Research, Darlinghurst, NSW, Australia
  2. Department of Computer Science, Yale University, New Haven, CT, USA
  3. Department of Neuroscience, Yale University, New Haven, CT, USA
  4. Yale College, New Haven, CT, USA
  5. Systems Microscopy, University of New South Wales, Sydney, NSW, Australia
  6. St Vincent’s Clinical School, University of New South Wales, Sydney, NSW, Australia
  7. Department of Genetics, Yale University, New Haven, CT, USA

Rationale: Metastasis is broadly considered incurable and primarily associated with cancer-related mortality1. Metastasis is mediated through cancer stem cells (CSCs) endowed with tumour-initiating capacity which are chemotherapy-resistant. CSCs have the ability to create heterogeneity critical to robust tumour growth. This is achieved through their ability to self-renew or to differentiate into non-CSC progeny. These cell-state transitions are regulated through intrinsic programs such as the epithelial-to-mesenchymal transition (EMT), which drives non-CSC to CSC de-differentiation, and the reverse mesenchymal-to-epithelial transition (MET) that drives CSC to non-CSC differentiation2. This study aims to define the transcriptomic transitions that drive CSC into a chemotherapy-sensitive non-CSC state, through MET. 

Design: We developed a neural ODE-based framework called TrajectoryNet3 that learns continuous dynamics from static transcriptomic data and applied it to time lapsed single cell expression measurements from an in-vitro triple negative breast cancer (TNBC) MET differentiation system. This led to the identification of 23 core transcriptional factors that were temporally associated with the emergence of the epithelial trajectory (non-CSC state). We constructed a gene regulatory network to sketch the transcriptional circuitry underlying the MET. The regulatory effect of one such gene, the estrogen-related receptor alpha (ESRRA) was validated using orthogonal approaches including RNAi, western blotting and immunofluorescence.

Results: TrajectoryNet identified a subpopulation of cancer cells enriched for tumour initiating properties, enabling the refinement of cell-surface markers to purify CSC populations.  We mapped the temporal gene expression program driving MET using TrajectoryNet and identified a sub-network initiating MET through early elevated expression of ESRRA. Knockdown of ESRRA increased CDH1 (E-cadherin) expression. Downregulation of ESRRA within tumourspheres was validated temporally, together with the concomitant downregulation of ZEB1 and upregulation of CDH1.

Conclusions: TrajectoryNet presents an innovative approach to characterizing the dynamic molecular programs derived from time point based -omic datasets. Using this approach, we define the key transcription factors driving the MET enabling translational research aimed at driving CSCs out of their aggressive state as a means to inhibit metastasis and chemotherapy-resistant disease4.

  1. Chaffer CL., and RA. Weinberg. "A perspective on cancer cell metastasis." Science 331.6024 (2011): 1559-1564.
  2. Chaffer, CL., et al. "Normal and neoplastic nonstem cells can spontaneously convert to a stem-like state." Proceedings of the National Academy of Sciences 108.19 (2011): 7950-7955
  3. Tong, A, et al. "TrajectoryNet: A dynamic optimal transport network for modeling cellular dynamics." International conference on machine learning. PMLR, 2020.
  4. San Juan, BP., et al. "Targeting phenotypic plasticity prevents metastasis and the development of chemotherapy-resistant disease." medRxiv (2022).