Portal de Periódicos da CAPES

Alex Mogilner

2019; Elsevier BV; Volume: 29; Issue: 19 Linguagem: Inglês

10.1016/j.cub.2019.07.077

1879-0445

Alex Mogilner,

Genetics, Bioinformatics, and Biomedical Research

Alex Mogilner grew up in Yekaterinburg (formerly Sverdlovsk), in the Ural Mountains separating the European part of Russia from Siberia. He went to the Ural Polytechnic Institute and finished in 1985 with a MSc degree in engineering. He then did research in theoretical physics for a few years in one of the academic institutes in Yekaterinburg and received a Candidate of Science degree (vaguely similar to a PhD) in 1990. Alex continued his theoretical physics research at the University of Manitoba for a year, after which time he attended grad school at the University of British Columbia (UBC) and earned a PhD in applied mathematics. After being a postdoc within the Math/Molecular Biology program at the University of California, Berkeley (UCB) for a year, he became a faculty member at the University of California, Davis, and after 17 years there he moved to New York University, in 2014. Over the last 25 years, Alex’s focus has been on the mathematical and computational modeling of the cell cytoskeleton, cell motility, and mitosis. His research team, together with experimental collaborators, have proposed models for actin force generation — using an elastic polymerization ratchet — lamellipodial and mitotic spindle mechanics, and the search-and-capture assembly of mitotic spindles. How did you get into cell biology? Ever since I was a teenager, I have thought that there is no higher calling than to solve scientific puzzles. I chose to study theoretical physics — which was all glamor in those pre-Chernobyl times — over biology, and years after I realized that this was a big mistake; I am not Einstein (dah!) capable of going after cool big problems, and ‘small’ incremental questions of physics bore the hell out of me. Then it hit me: biology was full of great ‘big-picture’ puzzles fit for normal people to crack, and solving some of these required quantitative thinking. Long story short, I first started doing mathematical biology — analyzing the mathematical modeling of biological phenomena — using biology as inspiration for producing elegant mathematics, and then I realized that it was much more fun to help experimentalists answer biological questions by using modeling as a tool. Luckily, my advisors — first Leah Keshet at UBC and then George Oster at UCB — showed me the beauty and mysteries of cytoskeletal molecular machines, and with time I learned enough cell biology not only to research it but even to teach it. How does modeling work in cell biology? There are many possible ways, the most straightforward and least painful of which is just reading current literature, seeing some gaps in our understanding, and realizing that these gaps could be bridged if one translates available quantitative data into a set of equations or rules of a computer game involving key molecular structures. Basically, one hypothesizes what the mechanism is based on some published data, then runs computer experiments and tests the predictions against some other published data. An example of this came during my early life in research: a powerful concept of the polymerization ratchet seemed to explain how Listeria is propelled by growing actin filaments inside the host cell. The idea was that the pathogen exhibits Brownian movement, and any time it randomly moves forward actin filaments grow a little, prop it up, and block its way backward. With Oster, we realized that Listeria is too large and slow to be the cause of the observed fast rates of propulsion. We hypothesized that the key was that the filaments are not rigid but elastic, and they are also much more nimble than the pathogen, so the filaments rapidly and repeatedly bend backward, grow, and push Listeria with the resulting elastic force. This elastic ratchet model had to be modified a few times over the years to keep up with increasingly detailed and complex data, of course. I should emphasize that this form of mathematical, or computational, modeling is a small fraction of modern computational biology. A much larger fraction is ‘omics’: using data and computer algorithms to find complex correlations, motifs, modules, and clusters in bionetworks. This is what I like to call — without any sarcasm but with admiration — ‘discovery without thinking’. During the past few decades, modeling and data analysis were largely separate, but lately they have started to merge. A more painful but also more exciting and impactful way to model is to do it in collaboration with experimentalists. The big difference is that, if such collaborations start early in the project, you have to come up with an initial model based on very limited data, then hope that simulations generate interesting predictions that help to design new experiments, data from which leads to changes in the model, and so on. The pain is that such an iterative feedback loop between theory and experiments repeatedly ruins beautiful models by introducing inconvenient data and messes with pre-conceived qualitative concepts. If both sides are serious — if modelers don’t ignore the messy data and if experimentalists actually listen to modelers and think with them — the results can be spectacular. One of my favorite examples of this is our joint studies of keratocyte’s lamellipodial motility with the labs of Julie Theriot, Kinneret Keren, and Alexander Verkhovsky. The collaboration started from puzzling over strange correlations between cell shape, cell speed, and the inhomogeneity of actin network density across the lamellipodium. After a few months of thinking and tinkering with the data, we realized that a force balance model roughly explains the organization of the lamellipodium: at its sides the low-density actin filaments are stalled by membrane tension, at the leading edge the high-density actin network protrudes against the membrane tension, and at the rear the membrane tension and myosin contraction pull forward debris of the disassembled actin network. This simple model not only explained the mysterious correlations but also made many predictions, which we have tested over the years. I should emphasize that, in such cases, the model is created by all team members, including experimentalists — after all, thinking hard about qualitative molecular mechanisms is at the heart of modeling. I am very lucky to have had a few other partnerships, with the labs of Jonathan Scholey, Daniela Cimini, and Alexey Khodjakov to study mitotic spindle dynamics, Min Zhao to investigate galvanotaxis, Laurent Blanchoin and Manuel Thery to work on in vitro and in silico reconstitution of actin dynamics, and Mary Baylies to solve some puzzles of muscle cell development. Here, I only mention the long-term research programs, but I have also enjoyed participating in very cool stand-alone joint projects with other colleagues. What are the thrills in this line of work? I’ll tell you a dirty little secret. When modelers or experimental/modeling teams write papers, often the order in which the results are presented is arranged for dramatic effect: initial data are described first, then a model based on these data is presented, then the model makes a surprising, counterintuitive prediction, and finally this prediction is verified when the rest of the data are presented. In reality, more often than not, the model is built and fine-tuned with more data available than it seems from the paper, and one can foresee modeling predictions before the mathematical analysis and computer simulations are performed. Once in a while you get lucky, and a sense of wonder emerges: simulations tell you something completely unexpected and cool that sheds light on the mechanism behind the data. This produces an unrivaled high. Here are two examples. When we were working on the elastic ratchet model more than 20 years ago, we were initially only thinking about how the filaments produce enough force and grow fast enough. Then, just by accident, we looked at predictions for the filaments growing at various angles and saw that the most effective filaments were pushing obliquely — at an angle — because filaments growing straight toward the boundary were too stiff to bend enough. We did not make a big deal out of this at the time, but soon after that Thomas Pollard and R. Dyche Mullins, in the process of assembling the dendritic nucleation model, realized that there was a connection between that prediction and the fact that the Arp2/3 complex creates branching angles of 70 degrees between mother and daughter filaments, making the majority of the filaments at the leading edge of the cell very effective. Another, more recent example comes from our collaboration with Alexey Khodjakov on mitotic spindle assembly: we were discussing his lab’s measurements of dynamic kinetochore shapes and puzzling over the fact that kinetochores are large in the early prometaphase. The issue here is that, while the large kinetochores make big targets for centrosomal microtubules, potentially accelerating spindle assembly, their size also means that it is easy for them to be connected to a wrong spindle pole, increasing the number of errors. My former postdoc Raja Paul ran fast simulations just mimicking the observed geometry, and we were stunned to see that, due to some peculiarities of this geometry, the in silico spindle with large dynamic kinetochores assembles both faster and more accurately. I have got to say that there is an even greater kind of thrill than this: when it turns out that your students or postdocs are smarter, faster, and more successful than you are. It just feels great when you can influence and guide a brain for a short while that comes up with better stuff than yours… It sounds as though there is excitement and thrills all the time. Do you not have any worries? I wanted to say that the greatest worry is funding, but though it is true that funding for basic science is becoming harder to get — and this forces us into wasting a lot of time and doing unnecessary crap — it is a pretty boring and trivial complaint, and also a universal one. A somewhat more serious and less boring worry is that modern science is becoming too fast — I’d even say frenetic — not allowing us time to acquire an increasingly detailed and quantitative understanding of very basic systems. For example, there are many gaps in our knowledge of the lamellipodia of single motile cells on flat surfaces, but interest in closing these gaps is waning… Collective cell migration has become a red-hot topic, yet how are we to understand the cohesive movements of multiple cells if we do not yet fully understand how single cells turn? This is not a very original complaint either, plus my wife tells me that I sound like an old person with this kind of talk. Here is my main worry: together with a significant fraction of other computational cell biologists, I have been caught unprepared by the current ‘big data’ revolution. When the data were scarce, it was perfectly adequate in any given study to come up with a single model, which was more often than not reductionist and conceptual rather than detailed. Now that the data are becoming plentiful, it is not only sufficient but also necessary not to limit ourselves to a single simple model, which is often biased by our scientific tastes and preferences. It is much more scientifically honest and useful for the process of biological discovery to start with multiple possible models. In fact, ideally, a computer program has to create a ‘library’ of possible models. Then, the experimental quantitative data have to be used — again by a computer — to screen these models until we are left with a hierarchy of a few good predictive models. This is easier said than done. We and others tried something like this recently, with some success, and we saw that we are flying in the dark so far — the methods and computer tricks that we use are very ad hoc and not scalable yet. Also, we will have to embrace difficult new methods — such as machine learning — to make sense of the sea of data. I was skeptical of this at first, but we are now trying to use them, and I am amazed at the results. There is probably no way that my group can learn these new methods enough to become experts though. So, maybe even modelers will soon not be able to survive without collaboration with other modeling labs, such that each lab will specialize in one set of modeling methods. (Experimental colleagues tell me that it is already impossible to keep up with state-of-the-art science without extensive collaborations.) For example, we recently started to collaborate with Gaudenz Danuser, whose lab focuses on acquiring and analyzing ‘big data’ and systems-level models while we help with reductionist modeling. Or maybe we are at the end of PI-driven science, and it is time for large centralized academic institutes… What is the best advice that you have been given? I don’t remember who told me this anymore, but many years ago when I was riddled with career anxieties and uncertainties somebody told me to snap out of my useless fretting and focus on a few simple things: make sure that you enjoy thinking about your area of science, concentrate on what you do well and collaborate on the rest, learn to enjoy writing and giving talks, and remember that life and science are unpredictable, so most likely you are worrying about the wrong things… What do you do for fun? Before moving to New York, I was an avid hiker. From New York, it takes too long to get to really cool trails, so I did what I wanted to do for decades: I started taking tango lessons. The best thing about it was that I met my wife on the dance floor. I have been obsessing about tango ever since…