During early development, tissues and organs begin to flourish through the displacement, division, and growth of many 1000’s of cells.
A team of MIT engineers has now developed a solution to predict, minute by minute, how individual cells fold, divide and rearrange within the earliest stages of a fruit fly's growth. The latest method could in the future be used to predict the event of more complex tissues, organs and organisms. It could also help scientists discover cell patterns that correspond to early-onset diseases reminiscent of asthma and cancer.
In a study published today within the journal, the team presents a brand new deep learning model that learns after which predicts how certain geometric properties of individual cells will change as a fruit fly develops. The model captures and tracks properties reminiscent of a cell's position and whether it touches a neighboring cell at a given time.
The team applied the model to videos of developing fruit fly embryos, each of which begins as a group of about 5,000 cells. They found that the model could predict with 90 percent accuracy how each of the 5,000 cells would fold, shift and rearrange, minute by minute, in the course of the first hour of development because the embryo transforms from a smooth, uniform shape into more defined structures and features.
“This very first phase known as gastrulation and lasts about an hour as individual cells rearrange themselves on a matter of minutes,” says study creator Ming Guo, an associate professor of mechanical engineering at MIT. “By accurately modeling this early phase, we are able to begin to uncover how local cell interactions result in the emergence of worldwide tissues and organisms.”
The researchers hope to use the model to predict cell-by-cell development in other species reminiscent of zebrafish and mice. Then they’ll begin to discover patterns common to all species. The team also imagines that the strategy might be used to detect early patterns of disease, reminiscent of asthma. Lung tissue from asthmatics differs significantly from healthy lung tissue. How asthma-prone tissue initially develops is an unknown process that the team's latest method could potentially uncover.
“Asthmatic tissue shows different cell dynamics when recorded live,” says co-author and MIT doctoral student Haiqian Yang. “We envision that our model could capture these subtle dynamic differences and supply a more comprehensive representation of tissue behavior, potentially improving diagnostics or drug screening assays.”
The study's co-authors are Markus Buehler, McAfee Professor of Engineering in MIT's Department of Civil and Environmental Engineering; George Roy and Tomer Stern of the University of Michigan; and Anh Nguyen and Dapeng Bi of Northeastern University.
Dots and foams
Scientists typically model the event of an embryo in one in every of two ways: as a degree cloud, where each point represents a single cell as a dot moving over time; or as “foam,” which represents individual cells as bubbles that shift and slide against one another, much like the bubbles in shaving cream.
Rather than choose from the 2 approaches, Guo and Yang adopted each.
“There is a debate about whether to model the model as a degree cloud or as a foam,” says Yang. “However, each are essentially different methods of modeling the identical underlying diagram, which is a sublime solution to represent living tissues. By combining these as one diagram, we are able to highlight more structural information, reminiscent of how cells are connected to one another as they rearrange themselves over time.”
At the center of the brand new model is a “dual-graph” structure that represents a developing embryo as each moving dots and bubbles. Through this dual representation, researchers hoped to capture more detailed geometric properties of individual cells, reminiscent of the position of the cell nucleus, whether a cell touches a neighboring cell, and whether it folds or divides at any given time.
As a proof of principle, the team trained the brand new model to “learn” how individual cells change over time during fruit fly gastrulation.
“The overall shape of the fruit fly at this stage is roughly an ellipsoid, but during gastrulation, gigantic dynamics happen on the surface,” says Guo. “It ranges from completely smooth to forming a series of folds at different angles. And we would like to predict all of those dynamics, moment by moment and cell by cell.”
Where and when
For their latest study, the researchers applied the brand new model to high-quality videos of fruit fly gastrulation recorded by their collaborators on the University of Michigan. The videos are hour-long recordings of developing fruit flies, recorded at single-cell resolution. In addition, the videos contain labels of the perimeters and nuclei of individual cells – data that’s incredibly detailed and difficult to acquire.
“These videos are extremely prime quality,” says Yang. “This data may be very rare since you get submicron resolution of all the 3D volume at a reasonably fast frame rate.”
The team trained the brand new model using data from three of 4 fruit fly embryo videos, allowing the model to “learn” how individual cells interact and alter as an embryo develops. They then tested the model on a wholly latest fruit fly video and located that it could predict with high accuracy how many of the embryo's 5,000 cells modified from minute to minute.
Specifically, the model could predict properties of individual cells with about 90 percent accuracy, reminiscent of whether they may fold, divide, or proceed to share an edge with a neighboring cell.
“In the tip, we predict not provided that this stuff will occur, but in addition when,” says Guo. “For example, will this cell detach from this cell in seven minutes or eight minutes? We can tell when that may occur.”
The team believes that the brand new model and dual-graph approach should, in principle, give you the option to predict the cell-by-cell evolution of other multicellular systems, reminiscent of more complex species and even some human tissues and organs. The limiting factor is the provision of high-quality video data.
“From a modeling perspective, I feel it’s finished,” Guo says. “The real bottleneck is the info. If we now have high-quality data on specific tissues, the model might be directly applied to predict the event of many more structures.”
This work is supported partly by the US National Institutes of Health.
