AI helps cells pull themselves together

US scientists have overcome a major stumbling block in the creation of mini-organs, programming cells to take on the desired shape rather than relying on 3D printing or external “scaffolds”.

This “inside out” approach, described in a paper in the journal Cell Systems, could signal a paradigm shift in how mini-hearts, kidneys and brains are grown on the lab bench – a technique used to study disease that may one day lead to personalised organ transplants.

The team, led by bioengineer Todd McDevitt at Gladstone Institutes in the US, was driven by an enduring issue with state-of-the-art ways of producing mini-organs such as 3D printing. The cells just won’t stay put.

Making a mini-organ or “organoid” starts when scientists take a person’s skin cell and, using the right mix of agents, turn it into an “induced pluripotent stem cell”. This IPS cell is the blank cheque of biology, capable of becoming almost any cell type. 

Grow it into a mini-kidney, say, and you can reproduce kidney diseases and test treatments in a dish sitting on your lab bench. But how faithful that model is depends on the physical organisation of the cells; to mimic a real deal kidney, 3D printing is often used.

But cells, much like unruly teenagers, have a mind of their own and will often wander away from their printed position.

McDevitt’s team wanted to own those cellular minds and so took control of two genes that together make up something of a joystick that directs how the cells organise.

CDH1 and ROCK1 figure heavily in the complex moves that lead to the final configuration of a group of cells. The pair influences stickiness and repulsion between cells, the surface tension that makes them spherical and their overall speed of migration. {%recommended 9677%}

The researchers used the editing tool CRISPR to knock out the two genes at various stages in the evolution of a clump of cells. Their aim was to make a bull’s eye pattern, a shape that’s common in human development, including in early embryo formation.

To detect that aspirational pattern, they engineered another tweak – making the cells fluoresce when CDH1 and ROCK1 were neutralised.

But there was a problem.

Factor in all the potential time points where the genes could be knocked out, the proportion of cells to be targeted, and a host of other variables, and the researchers calculated they’d need to do nearly 9000 trial-and-error experiments.

So they called on AI. They trained a machine learning model to compute the precise pattern of gene knockouts needed to realise their dream shape.

“Machine learning can predict what movie you might like based on your viewing history, but it can also generate new insights into biological systems by mimicking them,” says co-author Demarcus Briers, from the Boston University Bioinformatics Program.

“Our machine-learning model allows us to predict new ways that stem cells can organise themselves, and produces instructions for how to recreate these predictions in the lab.”

That model hit a bull’s eye, quite literally, allowing the team to produce the concentric pattern of cells they were aiming at.

“We’ve shown how we can leverage the intrinsic ability of stem cells to organise,” says McDevitt. “This gives us a new way of engineering tissues, rather than a printing approach where you try to physically force cells into a specific configuration.”

Ultimately, that concrete target shape will give way to a target in the abstract, one with potential to shift the life course.

“We’re now on the path to truly engineering multicellular organization, which is the precursor to engineering organs,” McDevitt says. “When we can create human organs in the lab, we can use them to study aspects of biology and disease that we wouldn’t otherwise be able to.”