Learning a conserved mechanism for early neuroectoderm morphogenesis

Background

This preprint addresses a fundamental aspect of developmental biology: morphogenesis, the process by which cells organize into tissues and organs during early embryonic development. Central to this study is the Bone Morphogenetic Protein (BMP); and, more specifically, the morphogenetic mechanism by which BMP specifies cell fate and tissue architecture of the neuroectoderm. The study delves into the dynamics of the cytoskeleton, particularly focused on myosin II, and cell adhesion molecules like E-cadherin. The authors’ unique approach allows for the quantitative analysis of developmental processes, bridging empirical observations with theoretical modeling to uncover the rules governing tissue formation. This study opens up exciting new avenues for exploring the molecular and biophysical underpinnings of development and their conservation across the animal kingdom, promising further insights into fundamental biology.

 

Key Findings

In Drosophila, BMP signaling is shown to orchestrate a gradient of cell adhesion molecules, notably E-cadherin, which influences the distribution of myosin. This distribution is critical for tissue flow during germ-band extension (GBE), a key developmental process. The research leverages neural network models to predict tissue dynamics based on myosin distribution, with the predictions validated by examining mutants that disrupt the adhesion pattern, highlighting the necessity of an embryo-wide adhesion pattern (a “patterned control field”) for proper tissue flow.

 

To further elucidate the mechanisms of how BMP controls neurectoderm morphogenesis, the study utilized a sophisticated combination of tissue cartography, principal component analysis (PCA), and sparse identification of nonlinear dynamical systems (SINDy).

 

The researchers began by isolating the smallest set of proteins that allow for forecasting mechanical behavior from initial conditions, leveraging a morphodynamic atlas of Drosophila gastrulation. This atlas contains in toto protein expression movies for various patterning genes and components of the cytoskeleton. By training deep neural networks (NNs) with these movies, the team was able to identify predictive biochemical quantities without specifying a physical or biological model beforehand. This approach revealed that knowledge of the initial myosin distribution is sufficient for a NN to forecast tissue flow in wild type embryos for 20 minutes with high accuracy.

 

To gain insights beyond the black-box predictions of NNs, the team performed PCA on coarse-grained fields to identify the most crucial features of myosin patterns and tissue flow. PCA helped decompose the data into principal components (PCs), revealing that a small number of PCs can account for the dynamics of proteins and tissue flow during germ-band extension (GBE). This analysis showed the development of a dorso-ventral-patterned myosin field and characteristic flow patterns with four vortices, suggesting that GBE can be described by simple and reproducible rules across embryos.

 

For a more detailed understanding of the dynamics and to derive interpretable rules for neuroectoderm morphogenesis, the study employed the SINDy algorithm. This method constructs dynamic equations from a large library of possible terms, fitting the data with as few terms as possible. By generating a library of biologically motivated terms, the researchers learned interpretable biophysical equations for the coupled dynamics of myosin and E-cadherin. These equations encapsulate rules and feedback mechanisms, such as myosin detachment, tension recruitment, and strain-rate feedback, modulated by the local E-cadherin concentration.

 

In human models, the study extends these findings to demonstrate the conservation of BMP signaling pathways in regulating neuroectoderm morphogenesis. Using stem cell-based models of the neural tube, it shows that similar mechanisms of adhesion patterning and BMP signaling are operational, suggesting a fundamental role for these processes in neuroectoderm development across bilaterians.

 

What I Like About the Preprint

This preprint exemplifies that the recent advances in interpretable machine learning techniques can indeed be used to study complex biological systems. The integration of data-driven methods with developmental biology in order to elucidate the conserved mechanisms of morphogenesis establishes a new paradigm of investigation; this preprint not only advances our understanding of developmental biology, but also showcases the power of interdisciplinary approaches in uncovering complex biological processes.

 

Future Directions and Questions for the Authors

• Given the predictive model’s success in capturing tissue dynamics, could this approach be extended to model the impact of direct mechanical perturbations (e.g., by tissue ablation) on morphogenesis?
• How fast is the deep learning model at inference – could you potentially make forecasts in close to real-time, as the time lapse movies are being acquired?
• With the role of E-cadherin in guiding myosin gradient formation and tissue flow, could you speculate on the potential mechanisms that ensure the precise patterning of E-cadherin during development?

Tags: bmp, data-driven methods, drosophila, stem-cell model

doi: https://doi.org/10.1242/prelights.36579

Read preprint (No Ratings Yet)