Back on Their Feet: Neuroscientist highlights how CSCS resources support new spinal rehabilitation treatment

February 28, 2023 - by Simone Ulmer

In the journal Nature, the scientists led by EPFL neuroscience professor Grégoire Courtine and Jocelyne Bloch, neurosurgeon at Lausanne University Hospital (CHUV), reported in November last year how they were able to regain some motor function through targeted epidural electrical stimulation (EES) in nine patients who had been paralysed by a spinal cord injury. As part of their research, the scientist created a spatially resolved molecular atlas of the neurons of the spinal cord. With the help of the machine learning method "Augur", which was applied to the database of this neural atlas, the CSCS supercomputer "Piz Daint" was able to, among other things, identify the neurons that support the recovery of walking after paralysis. The study showed that the neurons that are important for restoring control of leg movement in a mouse model are surprisingly not those that are important for walking movement in healthy mice. Conversely, neurons that are not required for walking before spinal cord injury are essential for restoring walking with EES.

What are the backgrounds of the many researchers who conducted the study?

JS: We have everything, starting from neurosurgeons, obviously, who work with the patients and engineers that are involved in all of the human work. Then, once we start getting into the more basic science side of it, there are experimental neuroscientists doing all the experiments. And then, as we start moving through the paper, we start getting into the molecular biology of it all. So, we have people who are experts in the wet lab doing single cell biology and all that side of things. And then, once we started collecting this data set, which was at the time really complex in terms of single cell biology, myself and one of the first authors on the paper, Michael Skinnider, started working on analysing this. We worked to develop a few of these new machine learning methods such as Augur and Magellan that would help us answer some of these questions that we were having trouble with. Finally, towards the end of the paper when they started doing a lot of optogenetics and things like this, it required, again, those experimental people to do complicated experiments in mice.

What exactly was your role?

JS: We collected single cell sequencing data from a lot of different experimental groups — from mice that were walking and mice that weren't walking, mice that had a spinal cord injury, mice that had rehabilitation, etc. And the question was, within each of these conditions, what are the neurons that are the most important? For instance, if the mouse has a spinal cord injury but is walking, how can we, from a computational point of view, answer this question? That was kind of the big challenge. At that time there were a few different ways in the field that people thought they might be able to do something like that, such as looking at genes that are known to be important for the activation of neurons. But what we found was that really none of these methods yielded consistent results. Therefore we tried a lot of different approaches, and in the end we had to come up with a new method to be able to do that. That became a huge part of the project, and the method itself ended up being its own paper that we published a year or so ago. That method was the only way we were able to answer this question and to find these neurons. 

By "the method", you mean all the RNA sequencing work with which you succeeded in creating the single cell molecular atlas of the neurons of the spinal cord…

JS: Exactly. So, it was to take all of this data, all these different conditions, tens of thousands of cells, 50 different kinds of neurons, and try to figure out across all these different conditions what's the common thread? That was where the method's development came in to try and come up with a way to do that robustly. 

This was the moment where supercomputing and "Piz Daint" come into play?

JS: Yes, absolutely. When you're doing these method development projections, especially in the context of single cell biology, the data sets are getting really big. We're sequencing millions of cells in a single experiment, and with more than 20,000 genes in any given cell, the data gets big quick.  

How important was the support from the supercomputer for your research?

JS: Trying to validate methods becomes a huge computational task that you can only do with resources like CSCS has. You need to run everything through a lot of different parameters to optimize all the features of all the methods and to test it on many other data sets. For that, we pull a lot of people's public data downloaded onto the servers, reanalyze it, try our method on their data sets and things like that; and it becomes extremely costly from a computational point of view. So, it's not something that would even be reasonably done on a laptop. In fact, it would be impossible. We would probably estimate that, to develop a method like that, it's in the order of thousands of years of compute time. 

Did you use "Piz Daint" exclusively to identify the key neurons in the recovery of walking ability, or was the high-performance computer used for other purposes?

JS: The main project that we use it for — ’to try and understand how spinal cord injury is different in different kinds of animals’ — is still ongoing. Some animals are able to spontaneously repair their spinal cord and some aren't; for example, us humans cannot but a fish or a salamander can spontaneously regrow the spinal cord. We have a grant from CSCS to have continued time to specifically work on that. 

So this means you use "Piz Daint" beyond machine learning methods.

JS: One of the things that's very interesting about this kind of work is you never necessarily know when you're going to need a new method. Sometimes when we start these projects, we might think we're just going to analyze the data and it's all going to be okay. But sometimes we need to kind of reinvent the wheel and come up with something new. That’s when having access to that kind of resource is super important. 

How does CSCS support you in this?

JS: I think one of the things that we're certainly finding is that this field (single cell biology) is a really cool field to be part of, because it's advancing so quickly in terms of what we can measure and learn. But one of the problems with that is the compute requirements are scaling with it. Year by year, we're just requiring more and more compute resources, so we definitely will be having to submit more applications and get more compute time because it looks like it's just going to keep going up. Fortunately, we found that whether it's environment problems or software installation issues that are hindering optimizing the runtime of things, the CSCS team is really great in helping us whenever we need it. 

Looking ahead, the results of the EES study are based on a clinical trial and give hope to patients with spinal cord injuries. When will it be available for affected patients?

JS: A few years ago, there was a company that spun off from the lab — ONWARD Medical, and that's now its main focus — to take these treatments and start putting them into larger clinical trials and eventually into the market. I've had the fortunate experience of being involved in some of this in the context of a very similar treatment where we stimulate the spinal cord to control people's blood pressure instead of helping them walk. It's going to happen; it just takes some time. The requirements to do all the clinical trials and to scale those trials, to get the FDA's approval and put this thing on the market, this is all obviously extremely expensive and time consuming. I think the general estimate for something like spinal cord stimulation to help people walk would probably be somewhere in the realm of five to 10 years before we would see that start to go on the market. The bar for the clinical trials is very high, and to execute those takes some time. 

Is the treatment suitable for all people with spinal cord injuries?

JS: There are a few groups in the world that have been working on this, and they've collectively done this kind of treatment on people, including males and females, people with severe injury, less severe injury, older people and younger people. And I would say the general consensus is that everybody seems to improve and they do see some value in the treatment. I would suspect that as we start to do this in more and more people, it will become clear for whom it works best. 

Are there things that still need to be improved? What are the next steps?

JS: I think on the clinical side, one of the biggest challenges is to make all of the technology usable, not only by the patients but by the people who are delivering it to them. The surgeons, physiotherapists, and engineers that are going to be working at various hospitals around the world need this to be straightforward to use so that patients can get the most out of it. That's a big challenge. And then on the scientific side, I think exposing that one kind of neuron can have such a profound effect across all these different kinds of behavioral conditions within the mice in the context of spinal cord injury was really striking to us; and that's led us down a whole new path of trying to understand how not only those neurons, but other kinds of neurons become so important after spinal cord injury. And how does that change in the context of different kinds of spinal cord injury, or different species? Does that hold in larger mammals? Is it the same? Is it different? And how can we potentially really tap into those neurons with a kind of clinically relevant method? That's where we're now headed. 

Will you for sure use high-performance computing in the future? 

JS: Absolutely! This kind of work cannot be done without these kinds of resources. Having a centralized system in Switzerland that is accessible to us and is beyond the scale of what's typically found in any given institution is extremely valuable and very useful. We're lucky to have it. It gives us a big advantage. So, I hope that CSCS keeps expanding and keeps working with us. I would suspect we'll be users of this system for years to come.

(Image above: EPFL)

This article may be used on other media and online portals provided the copyright conditions are observed.