 It's a pleasure to be here. Many, many are really interested young scientists eager to learn more about different techniques in biophysics. Simulations is just one of them, but let me try to convince you that it can generate lots of added value to experimental sciences. In our team we have one of these must-be things and one of them is that every PhD student has to pass a course in techniques in biophysics. It's basically a book about 600 to 800 pages where we don't discuss any simulation techniques. Instead, we just go through different experimental ones. What I'm saying here is that for a simulation person it's absolutely crucial to understand how experiments are done, how to interpret experiments, what really is measured in experiments. Because if you are doing simulations with what is understanding, then it's very difficult or impossible to couple your simulation predictions to experiments. Say in science you have to prove your predictions and it's very easy to make simulation models. It's much more difficult to make simulation models, which are valid, consistent with biology. And to that end we absolutely need to collaborate experimental people such that these predictions we are doing by simulations are able to match reality. We started this work about 20, 25 years ago. My personal history in computer simulations or computers is a bit longer. I think I was 10 plus when I got my first computer, that was week 20. You might think that one of these ancient persons in this context, that's true. Since week 20 it was the first ever computer that was sold to the big markets, not just for hardcore scientists in the labs, but really to the big markets. And I was brainwashed about its excellence when I was working on Commodore 64, which is known as one of the classics. I learned different computing languages, how to go down into the hardware, so to do machine language assembler and so forth. And I have been in this business ever since. So this figure at the bottom left is the first supercomputer we had in Finland, that's Cray XMP. It looks funny, but during those times it was really massively powerful. These days your mobile phones, I think, are able to generate computing capacity that is beyond Cray XMP. But during those years one was able to do quite a bit of things already during those years to complement experiments. So what we have done in our team in Helsinki or in Finland in general is to try to understand biology, living systems. So the motto here is that we do computer simulations and theory to improve health. This also implies that most of the collaborations we have are directed into biomedical teams, working cell biology, pharmacology, structural biology and so forth. And we have lots of these collaborations. The context where we are working is to understand signaling. And that is highlighted by this figure on the left. We have memory receptors, receptors which are embedded in cell membranes. And these are like local post offices in our daily life. You have letters coming from somewhere. They are delivered to the local post office, which tries to put them into order. After doing that, they are sent to the homes somewhere nearby. And now if something goes wrong in this process, but these letters are delivered to a wrong address, then obviously something bad happens. However, if letters are not delivered at all, at all for whatever reason, that's even worse news. Or if a same letter is published 100 times and delivered to 100 addresses, and 99 of them are wrong, then that's also bad news. In every one of these cases what usually happens is that you get some sort of disease. That's not fun. But key to understand signaling is really to understand memory receptors, which are the hubs, they are the sort of centers of communication. And the difficulty arises from the point that not many membrane receptor structures are known. There are hundreds of thousands of structures in a PDB databases collecting protein structures for proteins which are water soluble. This is quite easy to do. I think that it's easy, but it's easier to do compared to membrane receptors. And meanwhile for membrane receptors, a number of structures is quite modest. Just a few hundred basically. And that of course makes studies of membrane receptors quite difficult, both experimentally because interpretation data is difficult if you don't know the structure and also in terms of the simulations because we can't do simulations unless we know the structure to start with. We can do dynamics, but to start with dynamic studies through simulations we need to be static image of what the protein exactly looks like, that the chemistry is exactly the right one. The topics we explore by these membrane receptor investigations is to understand how their function can be modulated. It could be lipids in the membrane. It's common knowledge that if you have a bad diet, then the lipid composition in the membrane is somehow altered. If you don't eat properly, then the lipid composition in the brain will be altered and I would guess as an educated guess that it doesn't have very nice consequences. So please eat properly, please. Then we have sugars, glygans, which are attached to the proteins and they have some really, really extraordinary functions which we don't understand and it's one of the least understood topics at the moment in membrane biophysics. And then you have other modifications including mutations. Quite often the disease arises because of the mutations because it's changing the activation mechanism. Proteins are activating themselves without any particular reason simply because of the mutations. So all these ideas are what we are doing in our team to give experimental teams. Of course the cornerstone, the key cornerstone of membrane biophysics is experimental science because when you are really measuring reality what takes place in a living system, but there are limits and the most crucial limit in my opinion is survival resolution. If you go into the state of the art techniques, the super resolution microscopy which allows you to see what happens in a cell, the resolution is usually about 30 to 50 nanometers. Some people state that it's just a few nanometers, maybe five. But these are cases which are published under quite extreme conditions for example for a crystal or under semi-crystalline conditions. And that's not really a typical situation where super resolution microscopies are used. I think this is a quite fair statement even for state and other techniques that 30 nanometers is close to the minimal resolution you can probe. Now what do you see in 30 nanometers? Lipids are usually of a size of maybe two nanometers. Proteins, their size is quite diffuse, but it's of the order of five to ten nanometers, typically each. And that implies that if you have a protein complex comprised of two or more proteins bound together using super resolution microscopies, you can't identify what proteins are bound to one another. You are not able to see the structure. You just can't because the resolution is not good enough. For that purpose you have other techniques, say X-ray crystallography, NMR, cryoEN, but then again you have static images. And quite often in these techniques you have conditions which can be used to debate the quality of your data. For example crystallization, everybody knows, but the structures you get after crystallization might be quite different from the native ones. CryoEN is better, but in cryoEN the technology is just improving. It's very difficult to get high resolution structures. We are able to identify the positions of every atom in detail. So these techniques are emerging, but the point is that we have a gap in terms of resolution and also time resolution that experiments are not able to cover. And that's the reason why we are doing atomistic simulations. Atomistic and coarse-grains simulations. Atomistic ones are those where we identify these molecules such that we consider them atom by atom for proteins, lipids, carbohydrates, nucleic acids, water. In every one of them every atom is included. The state of the art here is roughly a few million atoms in a simulation system which means that the box size in 3D for the simulation box is about 10 to 20 nanometers which is already quite good. And it can be increased if you have access to really major supercomputers. These simulation times are now up to, say, a millisecond and maybe a bit more. And that of course depends on the system size you are exploring because we score a hand in hand regarding the computational capacity to carry other simulations. Then there are so-called coarse-grained molecular simulation models where one instead of describing the molecules in atomistic detail you now cluster atoms together. You create so-called beads represented by these spheres. Every one of these spheres tries to, in an effective way, describe what happens inside this sphere even a cluster of five to ten atoms. If you are clever enough, you can identify molecular shapes which are representing the right topology. You still maintain the right physics. For example, electrostatic interactions, phase behavior, and so forth. But of course you are losing most of the chemistry, for example, hydrotembonic capacity because you are no longer describing these molecules atom by atom. But the point is that by reducing computational complexity you are able to increase system sizes and time scales quite dramatically. I said that in atomistic simulations one can do simulations up to a millisecond. In coarse-grained simulations you can do simulations up to seconds at least depending on the amount of coarse-graining you carry out. This is the only slide where I am using any equations. This is high school physics. We have Newton's equation motion, mass, acceleration, force. For every atom, I, one by one. Then we have a force arising from a gradient of the potential at a given time. This potential arises from different terms under vast interactions which are basically dispersion interactions then electrostatics for two charged particles one at a time then bonded interactions for two particles bonded together three particles bonded together, four particles bonded together for example in steroid-backbones, in steroid molecules and so forth. This is a sort of minimal standard what people are using that if you have these terms then you are able to reach quantitative agreement between simulations and experiments. Quantitative agreement. That's quite amazing since if you look at these equations describing these interactions we have quite a few parameters. Sigma describing the hardcore size of a given atom. Epsilon describing the depth of the potential well between two particles under these one revolve interactions. Then we had two parameters. The typical distance, a sort of equilibrium distance between two particles when they are bonded together for example in a horogravan chain and then their force constant. Another two particle parameters. Just a few more and then we have these parameters. Three, five, seven, eight, nine, so roughly ten parameters for every atom. Type. Then how many atom types you have? Well, carbon, nitrogen, oxygen, hydrogen at least. And then you essentially have a system where you have forty to one hundred free parameters. Forty to one hundred free parameters in simulation model. How many parameters do you need in order to fit an elephant? I had done that. If you are using adaptive fitting then it's about thirty. If you have thirty parameters you can make an excellent elephant. If you want to get the shape without all these fine details ten parameters is enough. Ten parameters to fit an elephant. And here in the simulation model we have forty to one hundred free parameters. So does this make any sense? Well, pretty much it does because most of these parameters can be extracted from experiments. For example equilibrium distance between two particles. That is based on crystallization, electron microscopy, spectroscopy. Then other parameters as well. These force constants can be found from spectroscopic measurements too. Then it's basically just one parameter which is difficult to estimate. And that is with sigma. Van der Waas radios describing the size of an individual atom because this depends on the context of other atom types what other molecules are nearby with atom you are considering. So the sigma parameter is difficult. And that is typically determined from quantum mechanical simulations, quantum chemistry. But all in all it's not an impossible exercise to develop simulation models which are really valid where the simulation predictions are consistent experimental findings. Here's one example. One of the really big questions in the field is to understand protein folding and whoever is able to design a theory that based on a sequence of residues, amino acids in a row you can predict the final three-dimensional structure for this amino acid sequence. Whoever does that will certainly be granted the Nobel Prize. I believe so. I'm not in the comic scene but I'm pretty confident that if you are able to resolve to be a protein folding problem then you are one of the people on the short list for the Nobel Prize. We currently don't have this theoretical understanding but we have techniques able to predict protein folds. This is a paper published already ten years ago by one of the really big simulation teams in biophysics in blue where is the simulated structure for one of the sequences starting from a completely random fold. And this red one is the one that has been found from experiments. So just look at this comparison. Red versus blue. Blue being the simulation prediction. Red being the experimental observation. And now here for twelve different proteins. There's an excellent match. The results were not put in. We just used one of these interaction descriptions known as a force field to start a simulation for these amino acid sequences from a random fold and they ended up in a confirmation that matches experimental reality. So these models we are using simulations, they do make sense. Of course we need also quite a bit of computing power. The first computers were designed in 1970s after the Second World War. ENIAC was the first one. MANIAC 1 was the second. The third one was called MANIAC 2. Don't ask me where these names are coming from but they are really nice. These computers were based on electron tubes and obviously the computing capacity was not very large. These computers were able to do roughly ten to four operations per second. These days your mobile phones are much, much, much, much more efficient compared to these computers. But nonetheless these also stayed of the art in the 1950s. These days we have better computers. With supercomputers we are using about ten to fourteen times more effective compared to ENIAC able to do ten to fourteen operations per second and this is what you will see even. This is just one of the examples we are doing in our simulations in Helsinki. We have a lipid membrane comprised of many different lipids. You have cholesterol, swingomyelin, phosphatidylkalines, other lipids, the different bearing amounts of unsaturation and then we have a protein. Protein embedded in a membrane. We are looking at the influence of the lipids. So if you change the lipid composition, what happens for the protein? We are adding ligands through glycosylation to be protein. Then again what happens? All of these different factors are affecting the probability of a protein to bind to its racks and other ligands and so forth. And these simulations can cover times up to a millisecond. So why this millisecond timescale is important? Since for quite many proteins, or for most of the receptors in membranes, the activation takes place about a millisecond. The whole process of activation, starting from ligand binding, going through the conformational changes of the protein, then on the cytosolic side doing the chemistry such that some processes, for example, the phosphorylation takes place. So you receive a signal, you transfer a signal through the membrane and then you submit the signal from the cytosolic side to the next destination. This whole process is about a millisecond. And now we have computing capacities and simulation models able to simulate these systems and add any other mystic detail over the millisecond timescale. We are right now going through this paradigm change where simulations are able to study the whole process of a signaling event, starting from activation to the delivery of a signal to the final destination inside the cell. Here's another example based on coarse-grained molecular simulations. Here we don't describe these particles atomistically. We have lots of different lipids in the membrane. Then we have different proteins whose sizes are different. In this particular case there are about ten different proteins. What is important is to look at the dynamics. Usually a protein diffusion is about ten to five times slower compared to lipid diffusion. In principle this implies that proteins don't move. If you are looking at a microsecond timescale, you don't observe proteins to diffuse at all in the plane of the membrane. But because of these coarse-grained simulations where we are able to probe the timescales of seconds, you are able to see how proteins are assembling the environment. They are forming complexes. So proteins are getting together to form protein-protein complexes where many proteins together from some sort of oligomers. You can follow all these dynamic events in semi-molecular, in a semi-realistic manner. Semi-reality meaning that the physics is the right one but some of the chemistry is lost because of the coarse-graining procedure. And then after doing these coarse-graining simulations you are able to find grain with coarse-graining models back to atomistic representation and continue atomistic simulations to look at defined details. And this is a nice way to understand how the biology really happens. All right, having said that, let me move on and take some time to describe some examples about how computer simulations can be used to understand biology. I like this image because this is highlighting that computer simulations can do something important. I think that none of us debates the importance of your heart. If our heart is not working properly then... Yeah, yeah, you know bad things happen. But what keeps our heart beating? It's a pretty complex procedure but there is one particular protein which is important known as beta-2-adenetic receptor. It's one of the two protein copper receptors which essentially does the following. It receives a signal, a ligand binding to the receptor. It's activated through activation. It sends a signal to the cardiac tissue which releases calcium and calcium in turn is regulating the beating of the heart. And this protein therefore is the key molecule which is involved in this regulation process of keeping your heart beating. This protein is somehow impaired. If its function is impaired, then news are not nice. How this protein is modulated? That's another question because it's embedded in a memory and quite recent experiments have revealed that cholesterol is playing some sort of important role. The experiments done by Daniel Müller and his group in Basel they figured out that if you have cholesterol bound to the environment around this protein then the femostability of the protein is increased. Femostability meaning that the fluctuations are weaker somehow weakened because of cholesterol. If you don't have cholesterol interacting with the receptor then these femofluxations are increasing. So the receptor is somehow just a kind of wild animal doing something but not doing what it's expected to do. We started doing simulations to better understand what is going on. Here is an example. The protein beta2aR is here in the middle. We are looking at the system from top. These particles are round. They are liquid molecules. We don't show every molecule, every lipid molecule. We only show here the cholesterol molecules and every one of these moving molecules is cholesterol despite the fact that we are using different colors. We just identify some of the lipids by color since they interact the protein surface longer compared to the rest of the cholesterol molecules. And now to understand the main message look at this blue one. It comes from bottom right. It's coming closer. It kicks other cholesterol molecules away because it wants to go into the sort of one to say that this protein is mine. Get away, this is mine. Now after binding to the protein surface it remains there for microseconds. So there's some high affinity site possibly an allosteric binding site where this carcin molecule likes to be. After doing the simulations many times on different conditions we figured that cholesterol has an effect on the protein behavior. There's quite a bit of data here but let me just illustrate what is going on. We have a plot describing the correlation between the width on the excess error binding site where the light is binding and then on the cheap protein binding site on the cytosolic side of the membrane where the cheap protein is binding. So LL is a width of a ligand binding site. LG is a width of a cheap protein binding site in this protein. And here we show a graph for a histogram as a function of ligand binding site width and cheap protein binding site width. BVART cholesterol. We observe that there are quite strong fluctuations between two confirmations. One here and another here. It basically says that when we don't have cholesterol in a membrane then this protein is like fluctuating back and forth over time quite strongly. However, if you have cholesterol included then the protein confirmation is fixed to a unique one where the width of the ligand binding site and cheap protein binding site are the same. They don't fluctuate much and that implies that if you add cholesterol about 10 mole percent then instead of having these strong fluctuations you have a confirmation where the protein is just very mildly fluctuating but these fluctuations are very, very modest. And one knows that cholesterol is involved in the activation process. And one story short, it turns out that cholesterol and lipids, generally speaking, are able to modulate membrane receptor structure and also dynamics, for example, through allosteric binding. Then as another example, one can consider glycans. Experimentally it's known that glycosylation where glycans are added to the protein structure are important for protein activation, protein transport, and so forth. So there are very good reasons to better understand what with glycans are doing. Also based on experimental literature this chemical content of a glycans is important. Quite often one has galactose, mannose and fucose which are the abandoned. Glycans are attached. However, if you have diseases such as cancer when the content of C-alic acids is very, very pronounced compared to healthy tissue. So simply by identifying the glycans attached to the proteins one can see whether the tissue is somehow damaged. And it has, it's going along a trajectory that ends up in a major disease. How much do we understand about glycosylation and its effects on protein functionality? Very little. Absolutely very little. And there are many reasons why we don't understand. Quite many proteins that have many glycosylation sites but it doesn't mean that every one of them are glycosylated. Then if you're adding mass spec then you are able to see what glycans are there in the proteins but mass spec doesn't tell where are these glycans. Or to which glycosylation sites they have been attached to. Then if you want to understand the effects of glycans one by one it's a tedious process where you first have to mutate every other glycosylation site except for one then add glycans to the specific glycosylation site you would like to explore and then look at its effects and then you do that one by one for every one of the glycosylation sites. Then you consider two glycans at the same time in different spots, three glycans in different places having all the others sort of mutated that they do not form glycans. And it's a very tedious exercise absolutely. I'm not saying that in simulations it's much easier since there is a universal number of different combinations that one has to explore. But at least the basic principles can be studied quite easily because you always have a primary part in end glycans for example which can be used and then you can tune for the rest of the glycan such that you understand the basic information what the glycans are really doing for example for visibility of the protein for the glycans trying to attack these proton receptors. So in simulations we can do very controlled experiments to have at least a basic understanding about what glycans are doing in the first place. Here's my example we did a few years ago this is EGFR Epidemagrofactor receptor which has been studied in experiments a lot. I did a survey in a web of science many years ago where there are 40,000, 45,000 papers published in EGFR and this survey was indeed done many, many years ago so nowadays I think it's 50,000 plus. So I would say that this protein is important for health. And yes there are glycosolation sites about 11 in this particular protein. We decided to glycosolate every one of them all together to look at what is the extremal effect if you add as many glycans as possible. And these glycans are shown by these transparent Legos bound to the structure whereas some when we sort of surface of a protein that is facing water some glycans are next to the membrane what interface interacting with the membrane so to understand what we are doing here are two simulations. On the left we have a non-glycosolated case and the right hand side we have a glycosolated situation having these 11 different glycans attached to the protein. Please focus on the membrane water interface. Peace for each and down here. What we observe is that for the non-glycosolated EGFR the membrane is very strongly bound to the membrane. However for the glycosolated situation this ectodomain is lifted upwards closer to the water phase. So somehow for some reason it's not bound to the membrane as strongly as in the non-glycosolated situation. And this turns out to be true. These glycans they are acting as entropy barriers they are lifting this ectodomain further to the water phase simply by creating an entropic barrier such that this ectodomain is not able to bind to the membrane as strongly as in the case without the glycans. And it turns out basic experiments reported in the same paper that if you have the glycans in place they are somehow affecting the visibility of the ligand binding site. So the ligand binding to the receptor is dependent on the visibility of the glycans around the ligand binding site. Another story is described here for CD2. It's a very nice protein because it also has a biological function in adhesion and then it's pretty small. So it's easy to simulate. To make a long story short, it turns out that this receptor is very strongly bound to the membrane water interface as depicted here if the lipid composition is matching non-graft conditions that is evad cholesterol and evad swinger lipids and when this protein is not glycosolated. However, if you do glycosolate this protein you add these glycans shown in red and then you also change the lipid composition to graph-like conditions where you have lots of cholesterol swinger lipids. Then this ectodomain just lifts up and starts standing upright as depicted here. The question is why does it happen? What is causing this? Well, there are two reasons. First, just like in the EGFR case, these glycans next to the membrane water interface are acting as entropy barrier. They are pushing this ectodomain to stand upright. They are trying to avoid a situation where this protein ectodomain would fall down to the membrane. But it's not just a full story. At the same time, electrostatics is important. Quite many of residues in this protein are charged and these charges are interacting with the polar head groups of the lipids and they track this receptor down to the membrane water interface. And that depends on the lipid composition. To prove that, what we did is that we took a native CD2 protein and then we deprotonated all the charged residues one by one in the ectodomain. So here are the residues which are typically charged. So what we did was that we essentially neutralized every one of them by mutation. And after these charged residues were neutralized to not have any charge anymore, it turned out that the CD2 ectodomain which was lying along the membrane just started lifting up as I described and started standing upright. So there are two effects which are affecting the ectodomain confirmation and also the visibility of its ectodomain and the binding site of the ligands binding to this protein and these two effects are entropic interactions because of the ligands and then electrostatic interactions. So these two effects are working in unison in a concerted action. Lipids and ligands working together. Then I think I have some time to... So let me discuss another idea based on a process which also many of you would consider important that is how to stay alive, breathing, oxygen. Already in high school we were told that when we breathe oxygen gets into our lungs when it gets into blood circulation where hemoglobin takes oxygen and carries that to the cells in our body. So hemoglobin is important. But what high school physics and chemistry on biology did not tell is that how on earth the oxygen in our lungs is able to get through the lung sort of actant into the blood circulation to see where hemoglobin is. There's a barrier, a biological barrier with lung sort of actant which is a mixture of lipids and proteins and nobody understands the structure. Yes, nobody understands what the lung sort of actant really is. It's some sort of a very complex network, mess or something but it has a very important role since it regulates the transport of oxygen from our lungs to blood circulation and also the transport of carbon dioxide as a toxic compound from our body back to the lungs that we can breathe it away. So how does it do it? How does the lung really carry out this function? It would be important to understand this because there are quite many lung-based diseases which are very difficult to create and if you don't understand how lung act really functions or how oxygen is permeating through the lungs or factant then it's quite difficult to figure out any ways to treat these diseases. Many of these diseases are particularly important for newborn babies because the lungs are so sensitive, so subtle and any damage might compromise their health. Another very current disease is patient ARDS Accurate Dispiratory Dispressed Syndrome ARDS which is a disease of its own a pretty severe one but this is also reflected in COVID-19 since one of the side effects of COVID-19 is that the lungs are not working properly because of excess moisture which is impairing the lung function. So we still have a basic question how does oxygen permeate through lungs or factant? It turns out that there's one or maybe two proteins which are particularly important in this process. There are four proteins which are important in lungs known as SPA, SPB, SPC, SPD. The guys who invented these proteins were not very innovative in figuring out the names of these proteins but these are easy to remember. SPA and SPD are not so important. They are playing a role in certain aspects of lung function but they are not crucial for our health. However, SPB, the surfactant protein B that is crucially important because if you don't have it, then you die. Amen. That's it. So SPB, its function should be understood. SPC is also important because basic experiments done by Hespereskill and his collaborators, SPB is somehow fostering the activation of SPB. Most likely SPB and SPC they somehow work together on a molecular level. Details are not understood, but they do something together. But please keep in mind SPB, surfactant protein B that is the key protein because that is involved in oxygen transfer through the lung surfactant. To do some simulations to better understand this process we started from experiments done by Hesuses lab in Madrid. They identify low-resolution structures for SPB complexes so it's not just a single SPB molecule but a mixture of many SPB proteins found together. And the experiments still predict that they are forming some sort of complex which has a total shape hole in the middle and then stuff around it. We used experimental data to which we were fitting our simulation models first to create a diameter of SPB proteins using supposin B as a template because they are belonging to the same family. Most likely they have quite a bit of similarity. So that was a template. So that we furthermore created oligomass of these SPB dimers such that the match experimental structures was as good as possible. And I'm not saying that this hexamer we are using here in the simulations is the only one, the only oligomer having a role in biology but it was the smallest one which was able to fit all the experimental data and constructing the simulation models. So this seems to make sense. And Henna Hezos was also involved in this work so his input was highly invaluable in creating a simulation model. It turns out that when we simulated this SPB complexes next to a membrane or interface these protein complexes bind to a membrane. This is pretty strong binding. And then lipids are migrating through the hole in this protein complex. We also found that some lipids are favored more than the others. In particular cholesterol and PG they have lots of contacts with the protein in particular for residues from 30 to 60. So these graphs are showing the probability of having a contact with a protein for different lipid types. DPPC, which is a most abundant lipid non-chlorifactant POPC, which is slightly unsaturated force for the glucerol, which is charged and based on experiments with our PG no binding takes place. And cholesterol about 10 to 15 mole percent of lipids in non-chlorifactant are cholesterol and that's also important for lung function. And here is on the x-axis just a residue number for the amino acids in the sequence. So we see that cholesterol and PG are favored more than PCs regarding the membrane protein contacts. Then in different simulation studies we identified that if you don't have PG then this lipid profile inside this protein complex does not take place. So if you don't have PG then somehow some of the functions of this SBB complex are compromised. Without cholesterol you also see the same thing that this lipid profile inside this SBB complex disappears, but only under conditions where you have a physiological lipid profile having all of these lipids, TPPC, POPC, cholesterol, POPG found in reality when you observe that these lipid profiles are inside the SBB complex. And that tells but quite likely oxygen could be able to permeate through these SBB complexes too. Here's some additional data describing the contact frequency of different lipids when this SBB complex is bound to a membrane interface. PC doesn't show much, but if you have PG or cholesterol in a membrane then you see pretty strong binding in particular for cholesterol whereas some specific cholesterol binding points binding sites sort of on one of the principle interfaces since there are two interfaces in the SBB complex one on both sides. This is what we are now proposing based on this data together experimental data published earlier but SBB is forming complexes which are creating routes for oxygen to permeate from the water phase into the membrane structures one by one get through the lunchbox and membranes of a complex shape finally ending up in blood circulation and find hemoglobin. There are quite many predictions here but at least we have a hypothetical picture that can be used as a basis for testing this scenario using both simulations additional simulations and experiments. All right, I think I still have a bit of time so let me go through the last example here for Sabin. This is one of the hot topics in the field right now since there's some new strap for our data coming for the Sabin protein in the context of lipidroplet formation. Lipidroplets are those which we are using quite often if you go and buy some fast food say Big Macs if you take one Big Mac or two or three at the same time when there's a quite a bit of energy ending up in your body and your body is not able to use all that energy right away it has to be stored and this energy is stored in terms of the lipids inside lipidroplets. Lipidroplets are basically storage vehicles for energy. After eating we are formed and then if you're on diet don't eat then bit by bit the lipidroplets the content I use for energy production inside of self and they also have other functions but this is just one of the key ones. The big question is that how do these lipidroplets actually form? Experiments tell that there are some key proteins which are involved about which droplet formation is very difficult to take place or it doesn't happen at all and the most important one is Sabin which is a protein in the ER where it's doing something and I'm quite vague here because I don't know and I think nobody knows and it's based on public data nobody understands what exactly Sabin is doing so we started doing simulations together the Eleanor Aikkonen team here in Helsinki who were doing experiments in the same project to understand how lipids interact with the Sabin structure. We used the most recent crystal structures for the lumen or Sabin bound to the membrane water interface when we are using other techniques to create the trans membrane domain structure such that this lumen is embedded in the membrane and this realistic atomistic simulation model together the coarse-grains simulations were used to understand how different lipids are interacting with a Sabin protein. It turns out that the key molecule is TAC cryasal glycerols. In the absence of Sabin, TACs are clustering they are forming clusters of many TACs bound together but they don't do anything else they are not able to form lipid droplets effectively. Experiments on the other hand say that if you have these TAC clusters and if you also have Sabin then you just wait a bit and then these lipid droplets they emerge you can identify them from the inside of a cell in a living cell. So what happens here? Simulations predict that these TAC clusters after forming they find Sabin. They are binding to the Sabin. Sabin is like a nucleation site for creating larger and larger TAC clusters and it's collecting all possible TACs together to be embedded inside the slave Sabin structure. What happens after this point is not really understood yet but our simulations identified that there are specific residues in the Sabin structure which are making Sabin to be nucleation sites. There's one particular serine S166 identified here which is interacting with the glycerol group of three glycerides TACs. I think this image on the left depicts this process quite nicely. We have one particular TAC diffusing inside this lumen inside the Sabin protein. This blue trajectory describes a diffusion of a TAC molecule when the TAC is not interacting with serine 166. So it's really diffusing quite rapidly. However, when this residue is interacting with this particular residue, it binds there. So it's not interacting. So it's not interacting. So it's not interacting with the residue. It binds there and remains there. And diffusion is blocked. It no longer moves. So there's one particular residue in the Sabin structure which is catching TACs. And if it's able to catch one of them, then the TAC molecule which has high affinity catches another TAC, which in turn catches another. So bit by bit the TAC cluster inside the Sabin structure increases in size, increases further. And then what we observe is the formation of the TAC cluster inside the Sabin lumen. Quite intriguing. The nucleation site arises because of specific protein vapid interaction. And that is highlighted by this tiny image right here. The details are given in the paper, but here again the point was that as simulations were able to predict new phenomena, they were able to explain what happens in experiments because simulation findings were completely consistent experiments done in Elina Konens lab. And all in all there is quite a bit of added value given by simulations to understand experimental data. All right, let me conclude here. There are many springs in biomolecular simulations. First of all, we are able to provide nanoscale inside of a spatial scales and time scales which can't be reached by experimental techniques. That's one thing. In this particular context simulations are able to both predict and also confirm experimental findings help to understand experimental findings how lipids and other modulation mechanisms are able to alter protein behavior. And obviously you can use simulation techniques for all possible biophysical processes. The key point is that none of the simulations are dependent on one particular aspect and that is that you have accessed protein data, protein structure data. If you don't have any data for the protein structure it's very difficult to get started. Of course you can create a protein model for your own body. It might be a toy model having no coupling to reality. It must be safer that you have collaborators to inquiry OEM, X-ray studies, NMR, you name it to find protein structures. And after you have a very good protein structure you can use that as a basis in the biomolecular simulations to understand where it happens. You can use proteins doing dynamically. And finally one of the key limitations also is that of course your simulation model is at most as good as its brains. The brains being the force field is interaction description. If there's something wrong in your force field then yes, then you can't trust your simulation predictions very well. But the force field is used in the field right now. For example, Gen36, OBLS, Amber they are pretty good. Quite often the generic behavior is given correctly and also the quantitative numbers are better and better over time compared to experiments. Okay, hopefully I was able to convince some of you that collaborations between simulation groups and experimental teams makes sense and doing simulations would also be fun. Thank you. Thank you very much, Ilpo. So I'm sorry I missed a few because I had to listen into another meeting. Farah-Aleldas, Life of Zoom. But I especially liked your, this with them, shaping this triglystride stuff. I think that is very cool. I think it's very important. And often forget. And we look so much into biomembrane that we forget that lipids are everywhere in every form and shape that we don't understand completely how they behave. So is there any questions for Ilpo? Anybody? There is a raised hand. Who is that? That's Yan Yan, good on you. Go for it. So I had a question about, in the example you showed with the, what was the title of the slide, large scale behavior of kind of proteins and in a membrane with coarse grained models. You said that you could then find grain the results back to I'm guessing kind of atomistic level. How exactly does that process work? That makes sense. It's a partly an engineering problem and partly a biological problem. So when you create coarse grained models, of course the amount of detail you can include is limited. This is obvious. And then you do long time scale simulations on coarse grained model. You end up with something. For example, we find a snapshot of your simulation which you want to find grain back to the atomistic representation. It's quite easy in principle. You just take these blobs which contain many atoms. You use some educated guess how they might be there inside these blobs and different blobs can be connected to each other that the carbon-carbon bonding and so forth, all of these conditions are respected. You can easily do that. But what is the most reliable final structure after doing this fine graining? Since there is a universal number of different options, how to do that? Yes, you are clever. You understand this point. So in practice, one has to test different scenarios and end up with some educated guess which makes the most of sense. And what we do after that is that after the fine graining process has been done, instead of a coarse grained model, you have an atomistic model matching the last snapshot of your coarse grained simulation, then you slowly try to equilibrate your atomistic system. So you don't take the atomistic system after fine graining as a proof, but you let the system equilibrate. So you continue this simulation atomistically for a few nanoseconds, maybe 100 nanoseconds. It depends on how long it takes until you see that the system has truly equilibrated and there are no major driving forces to change the protein behavior anymore. So it's a very careful process. The engineering part is that you can always find some sort of mapping from the coarse grained to the atomistic representation. But the biology is that you have to very kindly, very smoothly find conditions through equilibration that your model seems right. And this is completely fine. You can always do atomistic simulations over quite long times matching or starting from the coarse grained model and then you observe better how atomistic behavior affects your system properties. Thank you. A very good question. Okay. She's my boss, you know, Will. So is there any other question? For Ilpo. I think it is very good your emphasis that experiment and simulations always have to go hand in hand. It's always tempting when you see this simulation with beautiful pictures and it looks always convincing while a scatting curve looked horribly boring and so on. But if you can link the two together and make some sense of it, it will be very important. And also we are struggling when we do reflectometry for instance on model membrane to do the data fitting in a realistic way. And I mean it for a flat by layer it's IDL, DPPC and so on. It's not a major thing but when you start to curve and buckle and stuff like that even on a supported layer then you are lost. And then it will be very helpful. It's complicated but it will be helpful for the evaluation of the data and also predict what you can expect. It's an important aspect. So when I started I highlighted this idea that it's very easy to design simulation models but it's much more difficult to design simulation models which are matching reality. Okay. Thank you so much. Jing has a question. You're very welcome. Yes, so I think I can just shout out. Yes, I want to ask a question that you have a I remember you have a model, a membrane and then you have protein and then you say if you have the some weakened glycolisation and then it is like more leaving the membrane. And you say it is because of the entropic and the electrostatic some kind of effects. I wonder how to understand this entropic effect. Why if you have some binding and then the entropic is like changed. It's like a beach effect. If you don't have any these balloons when you are taking sun at a beach then you are just lying along the ground. However, if you have a balloon you can put it under your belly and you try to lie along the beach again. You can't do that because the balloon is between your belly and the beach. So the balloon is lifting your body upwards towards the air. So these glycans are doing basically the same thing. The primary effect is entropic interaction that they are taking some space which the protein is not able to occupy and by taking space they are enforcing the protein to lift themselves from the membrane water interface to the water phase. That is what I mean by the entropic effect. They are taking space which other proteins are not able to occupy at the same time. Okay. What about the hydrophobic effect? Will that be important in this case? Hydrophobic effects are including dissimulations indirectly. Hydrophobic effect is not really an interaction. It's a mixture of different interactions which end up in an effect which we know as the hydrophobic effect. It's basically a difference between water loving and water hating interactions. We describe that in the force fields in terms of hydrogen bonding which is an electrostatic interaction and then through one of our interactions which also in part is a kind of electrostatic interaction. Okay. For example, if we don't have hydrogen bonding per C in the simulation model, hydrogen bonding appears because of electrostatic interactions. Thank you. Thank you for clarifying. It's important to know that it's not a hydrophobic force with a hydrophobic effect. Let's thank Ilpo again. We will have five minutes break before we have the next lecture. I see that Motoma is already there. Did you ever meet Motoma when you were with Ola? Can you hear me now? Yeah, we can hear you. Opa, thanks a lot for a great lecture. I joined from the middle, but I enjoyed a lot and I love your papers. Great to see you. You both have Ola and Paavo in the comment, so to speak. Very nice. Ola made me one of a really good friends of Sushi and Seaweed. I invited him two years ago to Kyoto and he held a lecture in front of Japanese talking about Sushi. It was very interesting. I enjoyed it. He has written many nice books about Sushi and Seaweed. Beautiful books.