Lucie Delemotte is an Associate Professor in Computational Biophysics at the KTH Royal Institute of Technology in Stockholm, Sweden. She is also the head of a research group focusing on understanding the molecular basis for the function and modulation of membrane proteins.

Other than that, Lucie loves all kinds of mathematical models, practices knitting and kickboxing, and believes in the next generation of scientists.

I spoke with Lucie Delemotte about the place of computational biophysics in ion channel research, molecular dynamics simulations, coronavirus and even kickboxing. I learned a lot from Lucie and I think many experimental biophysicists will learn something new from this interview.

Have a good read.

---

---

So, Lucie, let’s start with the way by which you came to the position you are at now. Could you tell me a bit about this?

Sure. I’m actually trained in theoretical chemistry in Nancy, France, at the University of Lorraine, and that’s where I got my PhD. I was initially interested in biology but didn’t like how biology was taught so I kind of slipped into chemistry and physics because it was less classification-based and more mechanism-based. During my master studies, I realized that you could study biological systems with physical methods and that’s where I was hooked, basically. So, I decided to do my PhD in that field with Mounir Tarek. Mounir had a background in membrane biophysics using computational methods and he had projects on ion channels and membrane proteins among others. And I kind of begged to have a project on ion channels because they seemed really cool machines and so he was nice enough to get me that. So, that’s how I got started with ion channels. But then, this means I didn’t have a background in traditional biophysics and it was also at the time where these methods from computational chemistry or computational physics they were not so well accepted by biophysicists. And I think I worked quite hard to get to the point now where we have lots of collaborations with people that are interested. And so, when I finished my PhD I went to the US for a postdoc at Temple University in a group led by Michael L Klein. There, I basically continued the project I was working on during my PhD. And then, I did a short bit in Switzerland before I got the position at the KTH Royal Institute of Technology in Stockholm, Sweden. So, now, I’m an Associate Professor in Computational Biophysics at KTH and ion channels have stayed kind of at the center of what I still like. Now we have so many high-resolution structures of different ion channels that we are really in a position where we can greatly contribute to the field.

And what about SciLifeLab? In your profile I saw that you have double affiliation at KTH and SciLifeLab.

SciLifeLab is a Swedish-wide institute that hosts people from various universities having common research interests. So I’m from KTH but in the same building I have colleagues from Karolinska Institute, Stockholm University, Uppsala University and others. SciLifeLab also provides a national infrastructure, so the instrumentation that is too big or too expensive for host universities to have, such as cryo-electron microscopes or next-generation sequencers. I think it’s quite unique and innovative institution, and I’m not sure that we have something similar in other countries.

So, you are a computational biophysicist. Could you tell me a little bit about computational biophysics and why it is important for the ion channel research?

OK. I’ll start with the 3D structure. I think people are generally convinced that it’s important to have a 3D picture of an object that you’re studying and the main reason for this is that physiological phenomena start with atoms. And then chemistry and physics happens at the molecular level. Nowadays structural biology makes the covers of Nature and Science all the time. However, what structural biology actually gives you is a static structure of a well-behaved experimental sample. But in fact, any protein, including an ion channel, is actually a moving object, it’s dynamic, so in order to fully understand how it works, we need to have a dynamic insight. Even if you get several structures of different states, let’s say an open and a closed channel, you still haven’t worked out how precisely it opens and closes. So if you want later to not only understand, but also modify, how channels open and close, it’s important to have the entire picture. And here comes computational biophysics, which uses computational algorithms to understand complex biological phenomena and processes at an atomistic level. Out of different computational techniques, in our lab we mostly use a technique called molecular dynamics (MD) simulations which allows us to simulate protein behavior over time. Basically we build movies of how proteins behave dynamically. So getting a structure is a static snapshot and molecular dynamics is creating the movie of the response of the structure to its environment and to perturbations. So let’s say we put a drug in – what does that change in terms of the response of the molecule? Apply a voltage – what does that do to the ions flowing through the channel? We can answer these kinds of questions. Another really useful technique that we use is molecular docking. Docking is finding places where molecules, typically drugs, bind. This technique doesn’t involve reconstructing the time evolution for the movie but it can be a starting point for MD simulations later. So, you find where the drug binds and then you want to see if it stays there or how it moves within the cavity, for example. These sort of things could be really useful for drug development.

But how reliable are these molecular simulations?

Molecular simulations allow us to see things before the experiments.

LUCIE DELEMOTTE

I think it depends on many factors. Typically we need to consult with the group that wants to do something about how feasible and how reliable the answer will be. It starts with the quality of the ion channel structure and it’s also important to know if you have structures of several states or not. Let’s say it’s a drug discovery project: do you already know where the drug binds and with what confidence? Do you have a structure where the drug is already bound? That would change a lot the answer that you’ll get. But, broadly speaking we are at a stage where we have predictive power, which probably was not the case 5 years ago. I have an example of a story about HCN channels that we published last year. We got a structure of HCN channel in 2017 and that was in one specific state – in the closed resting state. We ran long simulations to activate the channel and see how it would activate and open. We came up with the very different model from what was known for other voltage-gated channels. And then the structure was published by the MacKinnon’s group, and we realized that our predictions were correct, so we could predict how the activation would occur. The structure revealed the same features, but our simulations got there first. Of course, it was much more convincing once there was also the structure that showed it, but I think that’s a good example that we can see things before the experiments at this point. Nevertheless, I have to say that we also see a lot of wrong things, and that’s why we often need the experimental counterpart. We predict something and then we design an experiment based on this and we validate the simulation. In my opinion, it doesn’t work to do simulations isolated from the rest of the community. Everything is done always in collaboration with people who do structural biology and functional recordings.

So, it seems that you are heavily dependent on the availability of protein structures. Does it mean that if you don’t have a structure – you don’t have a job?

That’s about right. But we can also do some molecular modeling. So, if you have a protein for which you don’t have a structure, then you can use homology modeling, which is using another structure as a template, finding sequence homology and constructing a model of your system. But homology models are generally of less good quality. I remember there was a time where we were interested in mammalian sodium channels together with a group of Mohamed Chahine in Quebec but we didn’t have structures at that time. And so we had to build homology models. But what we now know is that it’s dangerous to assume homology from just the sequence. In the field of voltage-gated ion channels, until recently, everybody modelled everything based on the Shaker architecture, assuming that all voltage-gated ion channels are domain-swapped. And nobody thought that there could be channels with a non-domain swapped architecture. Then, non-domain swapped channel structures were revealed! The pieces are conserved, which is why you cannot know from just the sequence that there could be kind of dramatic changes in the structure. So, we need to be extremely careful when modeling proteins, even though in some cases that works very well.

And from your experience, how do experimental biophysicists treat computational approaches in ion channel research? Are they open to this or reluctant?

More and more people are becoming convinced that computational approaches are actually very useful in ion channel research.

LUCIE DELEMOTTE

I think more and more people are becoming convinced that it’s actually very useful. Now if I go to conferences, I have to turn down requests from people, which was really not the case when I started. I remember the first ion channel Gordon Conference that I went to. That was 2010 I believe. Then, nobody showed up at my poster. Nobody. I was very disappointed. And nowadays it’s like the opposite. But the methods have also matured, such that it’s actually possible now to teach people to use the methods themselves. I’ve had cases of colleagues who have sent their PhD students-electrophysiologists to my lab so they could learn structural modeling and MD simulations. For example, for an electrophysiologist a very useful tool is mutagenesis. And maybe you don’t want to do a whole scan of an entire helix. Maybe you’d rather just focus on a specific region and maybe the structure is not quite enough to tell you what to mutate. So, simulations can really help you in that case to predict what the effect of the mutation might be and then if it turns out to be interesting you can test it in the lab.

So the interest to our work is there, but I have to tell you that I generally get a lot more positive feedback from the young guard than from the previous generation who are not always interested in the fact that we can solve problems they have been working on for a long time. Although, I must say I have many wonderful seasoned collaborators so this cannot be completely generalized. The next generation is anyways generally much more open to new approaches and much more interdisciplinary. So I think we’ll be fine.

Could you describe a typical day of computational biophysicist?

Well, you sit at the computer, you stare at models and you prepare systems for molecular dynamics simulations. Then you send them to supercomputers and then there is a lot of analysis to be done. Simulations a very noisy and there is a lot of data, so you need to kind of clean and extract data you need. We use a lot of Python and we mostly use libraries that other people have developed, so it is a community effort. For example, we use MDTraj and MDAnalysis libraries that specialize in analysis of molecular dynamics simulations. For visualizations we mostly use VMD, but also PyMOL. And for molecular dynamics simulations we also use GROMACS, which has been developed by my colleagues in our institute. My lab has focused quite a bit recently on automating the data analysis by using machine learning to not bias the analysis by prior knowledge. We are automating more and more, but it still difficult to abstract the human intuition completely from that. Interestingly, because it’s an interdisciplinary type of area where we join computer science, math, physics, biology, chemistry, I realized that my group members all do things a little bit differently. I come from chemistry, so I stare a lot at models and like turning them around, trying to understand what’s going on. But I also have group members that come from a background in maths or computer science and they like to do a lot of data analysis prior to visualization. So, everybody kind of adapts a little bit depending on where they come from and I think that’s beautiful because we learn a lot from other people’s ways of doing things.

And how does your lab look like – desks with computers?

Yeah. It’s very boring. Very interesting people but at the same time very boring open office with a bunch of desks and computers. No rigs, no chemicals, no danger zones.

Do you call it an office or a lab?

We call it a lab, but for any person coming in, they’ll see an office.

And the research that you are doing, is it more fundamental or applied?

I think it’s both, but it’s more fundamental than applied. It’s fundamental, because it’s really about extracting physical principles. But also I have to say that recently, in collaboration with the group of Fredrik Elinder, we submitted a paper that is trying to work out what the mode of action of a specific class of molecules on the channel is. And this work borders on applied because we hope to understand principles that then will guide future drug design. Obviously, computational biophysics is a powerful tool in drug design. In traditional drug discovery you get a structure and you try to plug a hole or active site. What we contribute on top of this is an understanding that structure is dynamic and cycles between different states. Ion channels are very complex things and they have different states that they cycle thought, and it might be completely useless to target a given state when you have to target another one instead. So you need to understand what the cycle looks like to do good drug discovery. And then, once you find a drug that binds specifically in a given state, you can propose modifications to that scaffold to make it have a higher affinity to get it to stick longer in a specific place.

What is your most satisfying discovery?

I think it’s probably the HCN story I mentioned earlier. I am very proud of many of the projects we do, especially the recent ones. But that one had a specific favor because we saw something very unexpected: the voltage sensor helix, instead of moving up and down rigidly, was bending and breaking into two parts! We were even initially worried that what we were seeing was an artifact of the simulations since it was so unexpected. But then we came across some experimental results from 2004 that were compatible with what we saw, so that reassured us a bit. Eventually, our collaborators did some experiments to verify our predictions and it was very satisfactory to see them actually be correct.

OK, just to have a superficial understanding of your work, imagine that you have an ion channel structure in two states, closed and open. How long does it take to simulate a channel opening?

Let me try to explain it that way. The typical timeframe for channel opening would be milliseconds or more. And in our simulations each atom is treated as an individual particle. To give an idea of the order of magnitude, the number of particles in a system comprising an ion channel in a membrane with a solution of water and ions is somewhere between one hundred thousand and five hundred thousand. And we have to follow the movement of each atom. The atoms are very light and move very fast, so you need to have a very short time step of femtoseconds. So, you see that your simulation time step is femtoseconds, whereas the timescale you want to reach is millisecond – and it means that you should repeat your simulation about 10¹² times. That’s quite a lot of time steps to do. With the computer resources that we have right now, we can reach nanoseconds routinely and microseconds rarely. So, with a powerful personal computer we can get tens of nanoseconds per day, and it means that depending on a project, we could need 10-100 days to get to a microsecond range. And you should also remember about statistics. If you see something happening once it’s not statistically significant, so you need to repeat your simulation several times. How many times? It depends on the system, but in any case it’s a lot of computer time. That’s why we need supercomputers to do our simulations for such complex projects. So, in order to get to micro- or millisecond we modify our algorithms and use some techniques like “enhanced sampling” which allow you to see larger timescales, but you’re losing some information in the process. So you are paying some type of price, either losing spatial resolution or the dynamical information. But you can still learn something, you just have to be mindful of the price you are paying.

Just by looking from the outside, there is an impression that it doesn’t really matter to you which channel to study – if you have a structure you can work with it even without knowing what the channel is. Is it so, and how you decide which channel to concentrate on?

We choose our channels based on available structures and also, just by talking to people, attending events, reading papers, seeing gaps and feeling like you can address them. But, even though it’s somewhat easier to jump from one channel to the other in our field than it is when you do wet-lab experiments, it still takes expertise. You have to know your molecule. And you make mistakes in the beginning.

For example, 3-4 years ago we got into the business of G-protein coupled receptors through collaboration and I found it very difficult to catch up with the literature and get to know people and understand what the challenges are. So we made a bunch of mistakes at the beginning even though we had a very good collaborator.

Another example is our very recent coronavirus project designed to simulate the activity of virus envelope protein, which, in fact, is an ion channel. And I want to tell you that it’s not that easy actually to get started with a new protein that you know nothing about. So, again, there we had to learn things about this specific protein and the structures of the previous coronavirus. So, that’s a recent example of trying to jump to a new system, and it has not been as easy as it may seem.

I’m curious to know more about your coronavirus project. How come that you started working on it?

Well, it was really spontaneous. I think it was a combination of things. In this period when many labs where shutdown all the people that do computational work kept working from home, so there was almost no downtime for us. And of course, looking at the situation, you ask yourself, how can I help? Actually, many people jumped on this; they started looking at available structures and analyze good potential drug targets based on previous experience. And most of them kind of settled on the spike protein or viral proteases. But, to be honest I resisted this initially and I thought: “Let people who know coronavirus work on coronavirus.” And then, eventually, I learned about this envelope protein, which got much less attention from the scientific community and I thought that’s a way to combine our expertise together with being potentially useful for the world in that specific context. I figured that even if we don’t manage to design a drug for COVID-19 hopefully we’ll learn something about coronavirus in general that can be useful in the future. Or maybe something even broader, like how the membrane influences this ion channel. So, that’s how I kind of designed the project. And I want to mention that one of the big reasons why this project becomes possible is Folding@home, a distributed computing platform allowing us to use multiple personal computers to perform our simulations. Basically, the idea of Folding@home is that everybody who has a computer or any computing device can download a piece of software called the Folding@home client, run this on their computer and execute the algorithm of the MD simulation to help us accumulate the data that we need to figure out the scientific problem. And I find it fantastic that many people want to dedicate their resources to COVID-19 studies and allow us running simulations on their computers. It means a great deal for us. And I really like the outreach part of Folding@home so you get to talk directly to users and it’s a good way to educate people, to tell them about the importance of molecular biophysics.

And also, for the group members I think it felt quite nice to be able to work on this, because the COVID-19 crisis was so overwhelming and affecting everyone. So just the fact that you are working on something that is interesting for everyone is empowering and you fill like you regain a little bit of control over the whole situation.

Do you collaborate with companies on any projects?

No, unfortunately, and I need to work at that better, because, you know I understood it only very recently, this could have a direct impact on your funding. I had a grant that was rejected recently where they said that it looked interesting but lacked a partner company. So the short answer is no, I don’t, and yes, I should. But, it’s not so easy actually to find contacts, and I thought maybe your IonChannelLibrary could become a platform where scientists and companies could meet. I think that to some extent it’s really a question of getting people in the same room and like getting them to talk and understanding better what their needs are. But it’s probably not so easy to do because everybody is a little bit overworked and maybe a little bit conservative and even scared about this type of partnership.

Have you ever thought about starting your own company in the computational biophysics field?

The job that I have now is very similar to having a startup. I do fundraising, managing a small group of people and getting stuff off the ground.

LUCIE DELEMOTTE

If I had several lives, I might start a company. You know it’s a full-time job. I’m already having a lot to do with all the projects I have in my lab. But, you know what, in my opinion, the job that I have now is very similar to having a startup. I started my lab 4.5 years ago and basically, I do fundraising, managing a small group of people, getting stuff off the ground, so it’s very similar. Except it’s maybe not your personal money and then you also have teaching responsibilities which could take a considerable amount of your time. Actually, one of my friends and colleagues, Chris Ing, who did a PhD in computational biophysics applied to ion channels, co-founded a company ProteinQure in Toronto, Canada, which is specializing in combining molecular simulations, machine learning and high-performance computing algorithms to perform structure-based drug design. So, if I were to start a company, this would be the type of company that I would have liked to create.

And at the end of our interview I’d like to ask you about what you do when you don’t do science. What do you do after a busy working day full of math and programming?

Well, my hobbies are somewhat opposite. I do kickboxing and I also do knitting. I’m not a kickboxing champion, I just do it for myself and I’m still a little bit intimidated by the whole process of fighting on the ring. I’ve been a couple of times scared by unexpected sparring with unknowns. You know, some people are just really strong by their nature. One time I got paired with a big Swedish guy who was also very broad, I mean just really big in every direction. We were supposed to do some exercises together, so it was fine in principle, but I was like: “Don’t even touch me”. And he was like: “No, no, don’t worry. I’m not putting force. I’m just using my weight.” And in a second I was like flying to the other side of the training room. So much force. I couldn’t believe it. So, that taught me that people who have technique, even if they don’t put force they can, you know, kick your … Yeah, I remember, that evening I was knitting till late night in order to calm down before I could sleep. 🙂

---

---

I’m thankful to Dr. Lucie Delemotte for taking the time to talk with me and sharing her story and insights.

---

Images by Lucie Delemotte.