Modelling the flow

As a PhD student at the Harvard-MIT programme in health sciences and technology, Connor Verheyen was part of a team working on granular hydrogels. These gellike materials can be injected into the body to heal injured tissues or even to manufacture new tissues. The team was led by Jennifer Lewis, the Hansjörg Wyss professor of biologically inspired engineering at Harvard University, and also included Ellen Roche, an associate professor of mechanical engineering and a member of the Institute for Medical Engineering and Science at MIT. Granular hydrogels are formed by densely packing microgels – tiny building blocks or ‘bioblocks’ that can fit together like Lego.

“It’s an inherently modular system where you can imagine plugging and playing different building blocks to get these macroscale ensemble materials that have different properties in terms of mechanics, chemical ingredients and so on,” Verheyen, now a postdoctoral researcher, explains. “The fact that you’ve got all these squishy Lego materials to play with makes it really interesting to think about how you can mix and match these different populations to arrive at a formulation that will do exactly what you want it to do in the body.”

As well as modularity, granular hydrogels have another very useful feature. “Compared to these bulk continuous hydrogels they’ve used previously, the fact that they are made up of all these little building blocks means that when you compact them, depending on how tightly you compact them, you can have this really nice, intricately connected pore structure,” says Verheyen. This porosity enables cells and blood vessels a way to access and fill out the scaffolding.

Self-healing properties

Granular hydrogels can act as either a solid or a liquid, meaning it’s possible to create formulations where the hydrogel remains stable in a syringe, but when it’s time to use them, they can be turned into fluid and travel through a nozzle. “Once they come out the other side, they immediately start self-healing and regaining their elastic cell-like properties, so they’ll maintain themselves in whatever shape you’ve patterned them in,” Verheyen explains. “That makes them really useful, because it allows you to deliver them non-invasively, so you don’t have to cut patients open to get them there. Once delivered, they self-heal and hold their shape again, and so that kind of combination of properties makes them really attractive from both a tissue engineering standpoint and from an in vitro or organ fabrication standpoint.”

The project to create the granular hydrogels proved more challenging than expected, however, requiring him to experiment with changing features of the gels so that they were optimised for delivery by injection. “The initial work was a lot of very frustrating trial and error,” Verheyen says. “It doesn’t sound so hard. You just make these little squishy building blocks; you compact them and then you should get them to inject – we found that was very much not the case.”

With the arrival of the pandemic in 2020, Verheyen worked on Covid research, during which he acquired computational modelling skills. He realised that some of the data science and machine learning tools he’d been using could be applied to the work on hydrogels. Computational fluid dynamics (CFD) involves using computers to model the flow of liquids in a particular environment, making it possible to predict how a fluid will behave. In medical devices, the modelling could help determine the optimal size and shape of a device for delivering certain drugs, or predict how blood will flow in a catheter. The advantages of using computer models particularly at the start of a project are they are a cheaper and less time-intensive way of testing a hypothesis than using physical materials.

Rigorous and reproducible

Initially, his colleagues were a little sceptical; and rightly so, according to Verheyen. After a few preliminary test building models, he decided that the approach was worth pursuing. He made sure it was as rigorous as possible, he says: “One of the things that is frightening about the machine learning approach is that you can throw something in, and it will spit something out, but you’re not always sure if it’s useful.” The work of building a computational model was time-consuming. Verheyen’s experience with Covid-19 research had primed him to make sure it was a “rigorous, reproducible, computational pipeline”. It took several months before he reached a finalised workflow that he was confident would lead to a good model that could be robustly validated.

The work involved breaking down the process of assembling the hydrogels into separate stages, each of which was then modelled separately using data from the earlier experiments. In the first stage, the model analysed how bioblock properties are affected by the starting material of the blocks and how they are assembled. In the second stage, the bioblocks were packed together to form the granular hydrogels. The modelling enabled the team to identify the different factors influencing the injectability of the final gel. These included the size and stiffness of the bioblocks, the viscosity of the interstitial fluid between the blocks, and the dimensions of the needle and syringe used to inject the gel. Having modelled this process from beginning to end, the researchers are now able to use the model to predict the best way to create a material with the qualities needed for a particular application, instead of going through a laborious process of trialand- error for each new material.

One of the things Verheyen finds attractive about the approach is that having used the data to build a model, it can then be shared through the opensource software repository GitHub, making it fully reproducible by anyone anywhere in the world. Rather than simply picking and sharing the three most successful trials, he says, researchers share everything: “You’re saying, ‘Here’s everything that we did, some of which was failed, some of which was OK, some of which went well. I don’t fully understand what’s going on behind the scenes there, but I can leverage these tools that are really effective at learning from these high-dimensional spaces.’ You essentially make that entire thing completely transparent and explicit for everyone else.” Following an open science approach means more and more training data becomes available, allowing other scientists to build on it to create better models.

Using hydrogels deep inside the body

Verheyen says the long-term vision for the research is to translate the initial work into clinical deployment. When it comes to translating the research into practical medical use, the focus will initially be on superficial applications, such as soft tissue repair. The team are interested, however, in looking at how to use the granular hydrogels deeper inside the body to repair 3D-tissue defects, for example in the intracardiac space, in the gastrointestinal tract or in the brain. “We’re trying to think about it with our engineering hat, so we’re thinking about what is hard to get to, because in that case you want a minimally invasive solution,” he says. Verheyen and his colleagues are also interested in defects that vary between patients and require a customised solution. “The intersection of those types of problems are where this type of material can really shine, and so a lot of those deep tissue defects, you don’t want to have to cut the patient open to get to them, and they are really different from patient to patient. It makes it really hard to manufacture something off the shelf and so potentially we could use this type of approach instead and use these building blocks to manufacture something in there.”

Verheyen is also interested in tackling the “inverse problem”, working from the clinical information to create a solution, “for example, you know you need a needle that’s several centimetres long, you know what the diameter should be, you know the modulus of the tissue that you’re trying to match.” Those parameters can then be used to restrict the design space to guide the researcher about where to search to find an “optimal formulation” to address the problem.

While computational modelling techniques have been adopted widely in other engineering fields, Verheyen says take-up has been slower in medical engineering, which has more traditionally been “an empirical trial and error field”. He believes the situation is changing, however, though more work needs to be done to make the technique both more approachable and more rigorous. Computational modelling is complex, and the techniques need to be adopted cautiously to reduce the risk of error. “It’s sometimes harder if you don’t have a background in it to know that it’s not being used properly,” he says. Nonetheless, he believes that adoption will gradually spread: “I think at some point, it will just be another tool like anything else we have in our toolbox.” ­