Offering a touch of computer magic to stem cell biologists

The scientists used EpiMogrify to predict molecules needed to control the cell state and fate of cardiac muscle cells (left) and astrocytes (right)

When the internet was first introduced to the world in the 1970s, it allowed people to send messages from one terminal to another. But many did not care much for it. After all, they could already send messages to others—by writing a letter or even sending a telegram.

Fast forward to 2018. When Enrico Petretto, Deputy Director of the Centre for Computational Biology at Duke-NUS, together with Owen Rackham, an assistant professor with the Cardiovascular and Metabolic Disorders Programme at Duke-NUS and the lead researcher of the project, introduced the idea of replacing the slow trial-and-error method used by stem cell biologists to culture and grow cells with the help of a computer programme, they, too, received a similar response.

“The basic need [for EpiMOGRIFY] was to try to improve the culturing of cells in the lab as well as the transdifferentiation of cells,” Petretto said, referring to the process where a stem cell transforms into a different cell type.

He added, “Some of my colleagues who work with stem cells, they believed that computational methods are ‘magic’.”

When told about this idea [of EpiMogrify], Ray Dunn, an associate professor specialising in regenerative medicine at Nanyang Technological University Singapore's Lee Kong Chian School of Medicine and a friend of Petretto, said, “I thought he’d graduated from Hogwarts because he was working on something that I thought would never work.”


Creating a stepping stone for stem cell scientists

For Rackham, having a tool like EpiMOGRIFY was very important for scientists who wanted to try to systemise making cell therapies. It was necessary that the algorithm could coax the cells to follow the fate mapped by the transcription factors after being converted from different adult cells with the right nourishment and environment.

“It is not enough to have a cell express the genes that are needed to turn it into, say, a heart cell. This cell then has to sit in an environment that invites it to be a heart cell,” said Rackham.

Normally, the process for culturing cells is a slow and steady one, explained Dunn. “I was using the toolkit provided by Mother Nature—the whole field of stem cell biology is following that path. That means following a cell from its embryonic stage at a slow and steady pace until it has differentiated into the cell type of interest.”

For Dunn that could mean dedicating as many as 45 days to coaxing his cells to turn into insulin-producing pancreatic beta cells.

Petretto wanted to speed up the process while also creating a starting point that allows both scientists new to the field and stem cell experts focusing on a new cell type to cut out the laborious trial-and-error method by narrowing the possible combinations for them.

“It [EpiMOGRIFY] can be used by people who already know how to make their preferred cell type and people who are just starting to work on making new cells,” Petretto said.

But EpiMOGRIFY doesn’t just offer one set of factors. It provides a large repertoire of molecules that scientists can choose from. One advantage of that is that it enables researchers to pick the cheapest combination of factors needed to complete their experiment.

“The traditional way to culture may sometimes mean you need to buy expensive reagents, expensive molecules. With our approach, in several cases, you can exchange molecule A for molecule B if they provide the same result as predicted by EpiMOGRIFY, with the difference that maybe molecule B is tenfold cheaper,” said Petretto.


Coining an algorithm for healthy cell growth

But before such ‘magic’ could be done, the algorithm had to first be written.

Turning to publicly available data libraries that catalogued how different cell types organise their genome so that the right genes can be switched on and off at the right time, Petretto, Rackham and then PhD student Dr Uma Sangumathi Kamaraj had to go mining for patterns.

“What was really challenging in those early days of the project was that labs processed their data differently,” recalled Kamaraj. 

After about a year of pulling together the data, Kamaraj and the team spotted that genes that were important in determining a cell’s fate all seemed to be next to a long string of a specific epigenetic activation mark, called H3K4me3 (short for a tri-methylation (me3) of the fourth lysine (K4) on the third histone (H3)). Like compactus shelves in a vault, histones organise the DNA so that it can fit into the small space of a cell’s nucleus with only the genes necessary for the host cell’s function readily accessible.

“Genetic code that is preceded by a lot of H3K4me3 tends to signal that this gene is important for cell type,” said Rackham. “We found this to be the case in more than 100 cell and tissue types.”

Kamaraj set to work, developing an algorithm that could identify the necessary signalling molecules and factors that could either maintain cell identity or enhance directed differentiation. While using the available data, she also had to check the latest literature to ensure that her source data was as up to date as possible.

While Kamaraj and the team worked on the computational element, Rackham’s long-time collaborator Jose Polo, a professor at Monash University, and his team tested the programme’s predictions in the lab, starting with astrocytes, a type of nerve cell.

“We were measuring it against Matrigel,” said Rackham, referring to the gelatinous protein mixture derived from mouse tumour cells that is used by cell biologists to recreate the complex environment in which cells live.

“While we have shown that our predicted cultures were as good or even better than those grown in Matrigel,” Rackham said, “what is most important is that we have using a computational approach, for the first time, grown these cells in a clearly defined environment that is free from any other animal product eliminating environmental reasons for batch-to-batch variations.”

Click on the images above to enlarge

Growing to be the first in the commercial world

Knowing that computational biology is a strongly competitive field, Petretto and his team took a calculated gamble in 2019 by filing for a provisional patent.

“With so many potential competitors, we decided to file the provisional application first even though that could mean we’d lose priority [against anyone who may have filed since the filing of our first application] if we didn’t generate the additional data within 12 months [and refiled the application],” Rishu Srivastava from the Centre for Technology and Development (CTeD) explained.

After filing the application, the team had to focus on their biggest task: generating enough data to demonstrate that the programme, which covers 111 tissue and cell types, works in more than just astrocytes.

“That was a very anxious period for me. Not only did we have only 12 months to complete the work for the patent, but I also needed a publication to graduate from my IBM PhD programme,” said Kamaraj.

While Kamaraj worked on the manuscript, Polo and his team moved on to test the predictions in a second cell type—heart muscle cells or cardiomyocytes.

To validate the algorithm in further cell types, Rackham’s team applied for a Technology Acceleration Programme grant from NUS. With the additional funding, they could test the algorithm in other cells types such as endothelial and smooth muscle cells, among others. Just as the funding and resources were in place, the team was banned from the lab because of the COVID-19 pandemic.

When the Circuit Breaker period hit, the team was forced to stop their lab work for an entire month. But they managed to receive permission for one team member to go back to lab, allowing them a sprint to the finish line.

“In that last month, we were doing everything that we could to finish,” said Kamaraj.

And they did. All the effort that they had put in amounted in not only the completion of EpiMOGRIFY but also Kamaraj earning her well-deserved PhD after months and months of developing the algorithm.

“This project,” said Petretto, “reflects that in science, a lot of prep work is needed. Even with the excitement that comes alongside new innovations, there are also a lot of tedious tasks before these innovations can work.”

While it may be a far cry from the enchanting Hogwarts, computational methods, like EpiMOGRIFY, are certainly giving developmental and stem cell biologists their own kind of magic.

Uma Sangumathi Kamaraj (centre) marks the moment she completes her PhD with her team and mentors, Owen Rackham (top, left) and Enrico Petretto (top, right) // Credit: Uma Sangumathi Kamaraj