 Okay, so I would like to introduce Professor Lisbeth Garis. So she's a professor in the department of biomechanics and computational tissue engineering at the University of Liège in Belgium. And she's also working at the Prometheus Division. That is part of the Catholic University of Leven, so where they have a strong interaction with clinicians. So Lisbeth has very strong expertise in building models in order to assess precise clinical questions and then to calibrate those models from experiments that are directly derived from the clinical experience. Thank you, Jerome. That's a lot to cram into one hour of talk, but I'll try to live up to the expectations. Good afternoon, everybody, or good morning, everybody. This does not seem to be working. So I'm staying in the musculoskeletal system as Jerome was doing, but I'm slightly shifting clinical perspectives. So whereas Jerome was looking at the intervertebral disc and all the problems related to back pain, we're working in the field of tissue engineering. And this slide is basically all the justification that we need to tell you why we are doing tissue engineering. It shows you, just for the U.S. alone, the clinical problem that is arising in the need for donor organs or for new organs to treat patients. You see in green the number of donors, in brown the number of transplants because every donor can have multiple organs that can be transplanted, but in gray you have the number of people on the waiting list. And this is not just for, you know, the organs that everybody thinks about, like heart and lungs and liver, but also for bone. So in bone tissue engineering, this is also part of our problem that we try to save to work on. Now, just a few words on bone. For those who do not know it yet, so for those who do, it's just one slide. Bone is an intelligent material. It has sensors. It can adapt to its surroundings, to its loading that it experiences. And it is capable of healing, even scarless healing in the right conditions. So it looks like nature has it all solved, but in 5-10% of the cases, there is still a problem. There is a non-union, which probably has a lot of underlying causes. And at some point, you will not die from bone that is not breaking, but you will become very much incapacitated. You might lose a limb, which has a very big economic impact on society and on your own life. The problem specifically that we're working with, the patient population that we're developing our treatments for, is the patients are children with neurofibromatosis. In Dutch, the title reads, she has been walking around for 16 months with a broken leg that was taken when she was about two years old and finally she had surgery when she was four and a half years old and the leg was still broken. So this is the pathology that we're not specifically working on because it's a big pathology. It has a lot of symptoms and one of them is musculoskeletal. But this is a specific patient population that is termed in orthopedic terms, the orthopedic nightmare because there is no treatment and a lot of times these children end up losing their limb. So what we want to do is actually come up with a tissue engineering strategy so that in the end when we treat those children that we will be able to give them when they have finally surgery, when they're big enough to undergo the surgery, that we can implant a living implant, a piece of bone of about three centimeters that will help close the gap and that we will be healthy tissue and not the diseased bone tissue that they had at the side of the fracture. So tissue engineering, I think everybody is more or less familiar with the definition. It basically comes down to wanting to create a living implant, one that can really integrate into the body and become functional once it's integrated. Now the problem with tissue engineering at this point is that there is, well, the wording is all chosen wrong. We are not really engineering things at the moment or many groups are really doing just trial and error so there is very little engineering apart from the biomaterial side and the bioprocessing side. And also the tissue definition is slightly fuzzy in tissue engineering itself. So we believe in our group that engineers really have more to do in tissue engineering, a bigger role to play in tissue engineering and that can be in terms of quantifying and optimizing products, processes, and in vivo responses. And one of the ways to do that, of course, in our opinion, is using computational modeling. And ideally we have, in our group, we have the experimental people, the clinical people. So we would be able to do all of the questions and all of the models in this nice integrative scheme where you have a question, you have a model, you analyze a model, you do the experiment and so on and you get more insight. Now before I start, I want to make a few preliminary remarks. One of that is that the research that you will be seeing is not just from one of my students or just from one of, just my team alone. This is really a larger consortium that was needed to do that work. So these are the people that in some way or another are involved, not necessarily all of them directly in the research that you will be showing, but by simply being together you get some influences from the experimental researchers and you learn to look at problems in a different way. So the way we are structured in our group is we have, we start with the basics, the ground substances on the left-hand side, so the cells, the growth factors and the materials. We have people working on those subjects specifically and then we have people looking into mechanisms of action. So when I put all of that together, I put it in the body what happens. And then a bunch of people are responsible for trying to translate this to the clinic. So translating means both technology transfer which is the bioprocessing part and then the clinical transfer which is the clinical, that preclinical work. And at the bottom going over all of these other boxes you have the enabling technologies and those are imaging, modeling, nanotechnologies and data management. Now even if it looks all nice and dandy well integrated, developing a common language is still difficult. Blanca already indicated and I think James yesterday also mentioned it. The experiments that you need as a modeler are not always so very interesting for the biologist. So it is still and after 10 years it still remains tricky and a challenge to get those data that you need and to come up with experiments where the modeler can get interesting things and that are at the same time also useful for biologists. Now if you have that problem in your group I can recommend this paper, Can a Biologist Fix a Radio? Very importantly this has been written by a biologist. This actually in a very funny way indicates why it is useful to have a model or a common language to discuss biological problems. And the third preliminary remark is that in this presentation I will just be talking about examples and I will not go into the details and the nitty-gritty of every one of the models that I will be showing. But we did do go through the whole modeling loop as establishment and then the selection parameter optimization, sensitivity analysis and then trying to learn more about the model. Blanca already mentioned a number of techniques like she talked about Bayesian for the uncertainty. There is way more indeed to do. There is a sloppy parameter analysis as she indicated. Jerome already talked about the design of experiments so all these kinds of things have their place in the model development cycle and we just edited the book and this is a bit publicity for myself on all of these different techniques that you could use in order to complete such a model cycle. So back to tissue engineering. What I want to show today are actually four different examples for categories of problems not necessarily clinical problems but also biological problems where we try to give or answer some questions by using or by developing a model. And those are the four C's that we work on. So the cells, the carriers, the culture and then finally the clinics. So starting with the cells the first thing you need to know is when you look at tissue engineering or developmental engineering as we now call it to have a more biomimetic approach is that we use biology as an inspiration as the example of what we want to do and most of the bone in biology in our body is developed through something called endocontrolosfication where you have the formation of cells coming together then they form this, this is the top half of a bone where they form this condensation and then in this condensation you will get differentiation to chondrocytes to hypertrophic chondrocytes you get the growth plate or sorry you get the primary ossification center you see your bone forming and then you get the same thing happening again at the ends of the bone which is creating then ultimately the growth plate here in between. And so we believe that if we can recapitulate this endocontrolosfication process so going through a cartilage intermediate to ultimately form bone that we will have more robust and better outcome in biology and there have been some proof of concept experimental papers to kind of underline or give power to that assumption. So now what are the questions? If we want to go, if we want to take our stem cells and we want to drive them into the chondrogenic lineage now you have cells here that are stem cells that are at the top of this kind of energy landscape and when you give them a push be it a chemical or a mechanical or an electrical push they will start rolling down into the valleys of this landscape so they will start differentiating. Now the question is where are my cells in this landscape? Is it still a stem cell? Is it really a chondrocyte? Is it something in between? Is it already on its way out? And how stable is that state? So am I really sitting here nicely in a valley like this? Or am I actually in between two valleys? That is a situation that you don't want because if you're in between two valleys and you implant the cell like that in the body well then it might decide to go back to that site where you expected it to be or it might roll to the other side so instead of producing bone you might actually end up with stable cartilage which is not what you want so we need to understand how we can push cells into a certain state and once they're there how we can keep them there because those are two separate questions. Now the way we do that is by looking at the regulatory networks of these cells and there are very different ways many ways to have a model of such a regulatory network and if you just in most papers you will find something like this graph that is connecting with arrows, the different players that the people were interested in but you can go all the way from this graph you can add interactions so you can actually model over time and quantitatively the interactions between the different players and you can go from a very phenomenological model all the way to a mechanistic model where you put values on all the different steps. In our case with the endocontrol ossification or with the growth plate what you see here is a simplified version of the network that is representing the controgenic differentiation within the growth plate. So you have growth factors that are here in blue, you have a number of transcription factors and then you have other proteins that are involved like the other ones which are the matrix proteins that are involved in this whole process. Two important ones are SOX9 which is representative of a controsite and runix2 which is representative of a hypertrophic controsite which then will ultimately induce bone formation. So how does it work if we use one of these models? So let's start with a simple Boolean model in a Boolean model what you do is you say all my genes are either on or off, one or zero and they can interact by having an additive or a negative effect so you can have an end function an or function or a non-function. So basically what happens is with this kind of networks you induce your network by starting with putting one certain genes on and other genes off and then you look at the interaction so if A and B are necessary to have to upregulate C as long as A and B are zero C will always be zero but once A and B are one C will also become one. So when you start by initializing this this is very well possible that you have changed the network so you have put things on that will induce the upregulation for instance the next step of other genes in the network. So you initialize a Boolean model and then you go through step by step until you actually reach a state where no matter how many steps you have the model take you can return the same result that is a stable state. So we initialized the network with just some of the markers for chondrocytes like SOX9 and then what we ended up with is a number of genes and gray is one and white is zero a combination of genes that is reminiscent of a resting zone chondrocytes so before it actually participates when you then change something and in this case that was the Indian Hedgehog that is a signal that normally comes from outside from the Piri chondrum so from the boundaries of the growth plate when you change that when you put that on then the network starts evolving again until it reaches a second stable state which is that of the proliferative chondrocyte you see that because you have SOX9 which is upregulated black is actually two so it's kind of a multi-level Boolean black is two so SOX9 is expressed and Ronyx2 is not expressed then we move further and physically in the growth plate the cells are migrating down so at some point you lose the PTHRP which also comes from the site and then you see that the network evolves to a final stable state and that is actually reminiscent of the hypertrophic layer of the chondrocytes you pass in between a stable state where Ronyx2 and SOX9 are both present and that is actually the pre-hypertrophic state but that is not a stable state that's simply a transition zone because you have a physical distance between the proliferative cell layer here and the hypertrophic cell layer here this is just another representation of that in black you have the results from our model and in pinkish you have the results so what we cannot have is gradients of course because well we have just one cell that represents a certain layer but you see that it's kind of over matching so what we can conclude from that or what we concluded from that is that with the set of genes we had in the network we were able to recreate or capture some of the essential behaviors of a growth plate chondrocytes now the problem was that with this Boolean model you're either on or off and there is no time so we have genes we have proteins and these actions there are different time levels and in a Boolean model there is no concept of time so everything happens just as fast so the you have the phosphorylation phosphorylation and the transcription all happen at the same time level which is not the case so we made a new model framework which is an additive model which still sits between 0 and 1 but it is in continuous way so your gene can be anywhere between 0 or 1 or your protein and you have fast processes and slow processes so for every box that you saw in the network we actually have two values one represents the gene the other one represents the protein and then there can be interactions between the gene level and the protein level and the proteins will be much faster than the transcriptional part so the proteins will first be equilibrated and only then we will take a next step in the mRNA now we tested again that network by doing a number of analysis and these are typically canalization so you initialize your network 100,000 times with different combinations of values and then you see where it goes to and we had three attractors which is a non-state which represents either apoptosis or something that is not in the scope of the model we have a Sox9 state and a Runix2 state and so you can see from the 100,000 analysis how many go to a Runix2 state how many go to a Sox9 state how many go to a non-state and that tells you something about the attractor basin of certain genes or certain states you can do a perturbation where you basically start from a Sox9 state and then I get a push one of the growth factors in the network is upregulated and then you see whether I go to another state or I stay in my state so that tells you something about the stability under perturbation of this network one of the things that we were able to do with the additive network that we weren't able to do with the Boolean network is to have dose effects what you see here is results for increasing doses of the growth factors that are mentioned here so for growth factors you see what is the response of the network under increasing doses of these growth factors and here I have to say we start with all the models that will end up or all the combination of parameters that will end up in a Sox9 attractor and what you see here is for all these combinations of initial states what happens if I increase the doses so for BMP what happens is when I increase the doses at some point part of the initial states will no longer lead to a Sox9 state but to a non-state whereas for IGF1 for instance it happens that the Sox9 attractor is decreasing so the height of the Sox9 expression is decreasing and then at some point if you increase IGF1 further you exit the Sox9 attractor entirely so those are things they might seem a little bit abstract but we tested this on for instance a network for T helper cells and we were able by having this difference in time resolution we were able to reach steady states that we were not able to reach if you just had a pure Boolean model so and those states actually did exist so Boolean models were not able to get there but these models were the additive framework was able to get there and at the same time it still had the simplicity of not having any parameters so all the interactions they are added but there is no weight that is assigned now how are we using this we are we have done this big study where all of these boxes in the network we upregulated and we downregulated and then looked at their effect on the size of the attractor basin for Ronex2 and Sox9 the way you use this is for instance by saying if I here upregulate the FGF R3 so the receptor 3 of FGF then I will induce the Ronex2 state so I will lose my Sox9 state and I will increase the attractor basin for the Ronex2 so likely my culture will end up in a Ronex2 positive state but at the same time you see if you then regulate FGF R3 nothing really happens so you have no effect so you can induce a state but you cannot use the FGF R3 to keep the cell in that specific state for that you will then need to add other things to your culture medium now we had to validate in some way this model and that was a bit problematic because by shifting from the Boolean to the additive framework we couldn't use some of the basic tools for validation of the network that Chiron already mentioned that were developed by Mendoza and others because we no longer fitted those definitions so we had to come up with another way but it's so complex that you cannot just intuitively look at is it right or not so we developed a number of things we are doing a number of things one is an ensemble approach so we tried to see all the different combinations that would fit the data and extract some commonalities from that and test them in the lab another one is inference so we took the data driven approach we looked at microarray studies that were published in the literature and then we tried to infer the network from that and because we didn't really have a strong preference for any of the available bioinformatics methods we just took an ensemble approach or a consensus network where we used a whole bunch of inference methods and we took the network that was actually given or all the edges that were given by all of those so that doesn't give us the best network that you can get from the data but the least bad so at least all the techniques agreed on that network and then we compared that to our literature derived or qualitatively derived network and see how it did and we compared that to just a randomly derived network and if we compare our literature derived network with this inferred network and we used a receptor operator curve we actually got this concave shape meaning that the similarities between the inferred network and our literature derived network are more than coincidence because if it would have been coincidence we would have been sitting here so we had we had a better fit than randomly would have been the case if we then use our literature derived network as a prior for the inference of the networks we can get this area under the curve up to a quite high value so that is one way to validate it's not an absolute validation but it gives some more confidence in that what we are doing seems to be working we've also taken this model and expanded it with a different pathway so that it no longer is only valid for the growth plate but also for osteoarthritis so for the stable cartilage phenotype and we're exploring in that direction as well so that was for the cells if we then look at the carriers we'll go from the endocondral part to the intramembranous ossification for a bit so we take calcium phosphate based carriers which are also which is a track in the lab and the question then is if we put our stem cells in these scaffolds what happens to the cells what happens to the whole thing in vivo when we implant it we know that it's degrading but it's quite hard to control and what the influence is of the release of these for instance calcium ions from these scaffolds on the cells is something that we need to quantify or that we want to quantify here we had quite a lot of data from screening experiments that we had in the lab so basically the MOA of those experiments are we take a calcium phosphate either with or without collagen scaffold that is used in the clinics we do our own analysis because the data sheet that you get from the suppliers aren't always so reliable so we do our own materials characterization a lot of it is done using CT analysis and then we combine it with our cells we implant it on the back of a nutmice because our cells are human cells and then after a certain time we take it out again and we analyze it looking at gene protein expression but also at morphology in the nano CT what we can do with the nano CT this is just a small example here you see a whole mount skeletal staining we can recreate that using our nano CT where we have specific contrast agents that will in this case negatively stain the cartilage this was hexa bricks and so you see that there is you can make out the bone and the cartilage very well in this developing mouth so these things have now this technology has been validated so we use that instead of histology because it does go quite a bit faster than histology and it gives you a 3D image of the the tissues developing in the scaffold so what is the set of data we had in the end we had so all the pre-implantation data of the scaffolds we had 3-day explants 12-day explants the gene expression and protein data for those because there is no bone to be seen then and at 8 weeks we knew the amount of bone that was forming here you see a series of 5 scaffolds this is we usually cut them in cylinders of 3 millimeter by 3 millimeter that's what you see here is section and you see here in the white those are the calcium phosphate grains in the gray you see the new bone that is forming inside those scaffolds so basically then we try to see what in all of this pre-implantation data or 3 and 12-day data is representative of the 8 week data just a very quick test we used a partial least square regression it's linear so it's by no means very accurate but it could be a first test for ourselves so we saw that actually the data at day 3 so if you implanted for 3 days and then you take it out and you analyze it looking at the phosphorylation of the BMP pathway mainly the BMP and wind pathways well then you could be able to say something predictive of the amount of bone formed at 8 weeks now this seemed very very simple we tested a number of other scaffolds also calcium phosphate based and for most of them they seem to be within the predictive range there was one that was completely off and I was actually happy for that because this was the only one where there was no collagen present so this clearly indicated to my experimental collaborators that there is a limit to what these simple models actually can do that you need to put in a bit more intelligence if you want to have a model that is valid for a wider range of scaffolds so the model that we are now currently developing to have a bit more intelligence in there is a model well it has two components one we need to have a model that will simulate the dissolution of calcium for instance from these scaffolds and then we need to have a model that will look at the effect of the biology of this calcium ions on the biology the way we do the dissolution of the calcium is using a level set method that was already commented on yesterday during the imaging talks so what you see here in white are the calcium phosphate grains and in the colors you will see the calcium the local calcium concentration as the calcium phosphate grains are dissolving this is something that you cannot measure at the moment these local calcium concentrations and they are interesting because well this is where the cells are actually are the only thing that you can do in vitro are these dissolution experiments where you take a whole scaffold you put it in a medium or in PBS and you just you know you measure how much calcium is present in the PBS over time this is what we did to validate the model so you see over time the calcium release in the in vitro dissolution tests and so we have validated that also using micro nano CT so we can see the degradation of the different grains and calibrator model for that now if you want to link that to the biology we had a simple model linking calcium ions or the concentration of calcium to the differentiation of stem cells towards osteoblasts you have their proliferation and then the production of non mineralized bone and mineralized bone and of course there is also the production of growth factors that we implemented in the same software framework and what you get is here three times a similar scaffold but with different release characteristics of the calcium and you see that it has a profound effect on the formation of bone so here you see bone is being formed in red close to the grains here you see bone in the open spaces in between the grains and here you see bone surrounding the scaffold now the reason I'm showing this is not because I believe the model is correct it is not but these are the three locations we do see bone formation in reality but instead of only being dependent on the calcium it is actually a combination of the BMP that we put on it so the growth factor and the calcium so we need to change our equations but I found it hopeful that we had these three types and locations of bone formation with the framework so one of the ways we are validating this is by having the amount of bone formation and related to the calcium release rate because that is a parameter in the model as well as something that we can measure so if you put all the scaffolds out in terms of their calcium release rate we simulated what their bone formation was and we used one calibration point and then we saw that the other scaffolds that we had on this side we're doing well on this side we're doing well but in between we seem to have missed something we have a problem with a bunch of other scaffolds but unfortunately they were all on the left hand side so now we are producing our own scaffolds here somewhere in the interval so that we can check what it is that we've been missing in the model how do we want to use that well ultimately we want to use those scaffolds to determine or these models to determine optimal scaffolds for each patient that's a lot to say but for different patient categories for all patients we know that we have very few stem cells that we can get from them for younger patients you might have an abundance of cells and you see that when you have an abundance of cells that you can put on your scaffold well then it really doesn't matter that much which type of scaffold you use because you will get bone formation everywhere even though there is some optimization that you can do but if you have very little cells to begin with that have stem cell properties well then it really matters what type of scaffold you use and you need to tailor your scaffold for your patient's population so that was the story for the carriers we then go to the culture well the culture or the bioprocessing bench as we call it in the lab what they are doing is actually they are taking the cells not from the marrow but actually from the periost they put it on scaffolds in this case titanium scaffolds because that eliminates all the chain scaffolds that are degrading over time which makes our study more complex so in this case we just settled for a fixed scaffold and they put it in a perfusion bioreactor system so this is perfusion and the fluid is being pumped through the system and it is perfused through the scaffold like this that then can be a construct that we can either harvest the cells from so that we can use the cells and seed them on a calcium phosphate or we could just implant it as such now the questions we have here are mainly to do with quality control what is going on on the inside of the bioreactor process, take it out, image it every time to look at what is going on you want to be able to measure what is going on but the only thing you can access is the medium and some sensors that you can put here at the beginning and there at the end but that doesn't tell you anything about what is going on inside so how did we go about trying to develop our model that would answer those questions well we started very simply with a static model just the cells on the scaffold and what we saw is that there is this curvature dependent whatever scaffold we had in the end if you put the cells on the scaffold you always ended up with these rounder shapes so they liked the corners best so we decided to use the curvature controlled cell growth which was something that was put forward in the literature by other groups and to use that as a basis to simulate tissue growth inside our scaffold and we use the level set approach to have the interface between the new tissue that we will always represent in green and then the empty zone or the free flow zone that you have and the level set is basically the function that tells you how this interface will move and we made that velocity of motion dependent on the curvature that you see here so this is actually the description of the level set and here you see that the speed of movement or the movement is dependent on the curvature on the local curvature so we calibrated that by using six scaffolds that we had with different shapes so we had a hexagonal square and a triangular triangular shape in two different sizes and we used one calibration point and no kidding to my surprise we actually were able to also fit this data which I had not expected because of this big jump that we saw in the simulations but apparently there must be some truth in this curvature based growth because it does go from a more square representation to a more roundish and that is explaining the jump that is taking place there so the added value of doing it in a level set because there were multiple models already in the literature that were doing the same thing using the model of the curvature based growth but using the level set method that we had allowed us to do complex shapes so we didn't have to worry about when two surfaces came together what would happen there it naturally actually solved the whole scaffold until complete filling so that is the static part we need to enter the flow the first thing we did is develop a flow model of the bioreactor and what I show in this slide is actually a bit of an embarrassing example of what a model can do in the lab for consistency purposes we always put the scaffold at the bottom of the bioreactor chamber so this whole thing is the bioreactor chamber the scaffold is only this high but the first people that started using the bioreactor wanted to make sure that they always would put them at the same location so they thought why not put it at the bottom if you look at this graph of course that was a bad idea but at the time we just never realized it we did see that there was an influence of the entrance so at the bottom there was always less tissue than at the top but then we did actually the calculations using the CFD and there is a critical length of course after which the flow is reasonably well developed and you have a more homogeneous situation which improved our results as I said it's quite embarrassing because this is very intuitive but it took the development of the model to actually realize that there was a better choice to make so how do we then simulate the fluid flow because we have the free flow zone but where the neotissue that's the porous medium so there's also flow through the porous medium we actually use the combination of Darcy here in the porous part and Stokes equation here in the free flow zone and we use a kind of penalty method where you had this parameter that would be very high when we wanted to get rid of this influence and then you would just have the Stokes or this would be actual value of the permeability when we wanted to have the Darcy from that we could calculate based on the flow we then calculated the shear stresses both at the interface between the neotissue and the free flow zone but also inside the neotissue using a number of models that we got from the literature and what we actually saw is that the wall shear stress which is often used in tissue engineering as a kind of measure for optimizing your scaffold or looking at the influence of the flow on the cells was actually way lower than the shear stress inside the neotissue where you also have the cells that will respond to that shear stress so this is something that we have to reconsider whether we still want to use a wall shear stress which should not transfer to the inside shear stress so what we now did is we had the static model we had the influence of the growth of the neotissue on the flow well we may have to couple back so we have to look at what is the effect of the flow on the growth of the neotissue so basically we made the velocity here instead of only depending on the local curvature we also made it dependent on the local shear stress because we were at it we added the oxygen tension at the same time so now the velocity is a function of the curvature the shear stress and that was again taken from the literature where we have an optimal region so that's optimal stimulation and then you have a zone where it is detrimental so there is no growth anymore and oxygen well you basically need a certain level of oxygen to have optimal growth of the cells that we have again you see here the chamber the flow chamber in the bioreactor the scaffold that we have and then we went we tested the different situations that we saw in the lab so we put the scaffold at the bottom and in the middle for the same flow rate and we indeed saw that there was no tissue developing here at the bottom whereas you got a more homogeneous feeling for the other case so here you see the movies what you see here is the fluid flow and it's actually changing but very little so you don't see it and you see here there's no flow so that there is a minimum there is some tissue growing and here the flow is simply too high to get anything done when you are in the middle of the scaffold you see that it's far more homogeneous throughout the scaffold now you can play around with the flow which flow is optimal and what you see here is a very low flow rate which is also very good and actually when you have such a low flow rate by the time you reach the upper part of the scaffold all the oxygen is gone so you have that problem of too little oxygen nutrients at this side of the scaffold because all of it has been consumed by the scaffold and also the stimulation by the flow is not optimal for that flow rate so at this point we are measuring the model with glucose and lactate not because we like to extend the model and make it as complex as possible but glucose and lactate are very specifically used in bioreactric cultures to measure the state of the cell there are studies that show that by measuring the lactate you can actually steer the process and you can say something about the state of the cells and the number of cells that are inside but obviously this is all some kind of integrated value because this is one lactate value for the entire scaffold it doesn't tell you anything about the local lactate concentration and cells they don't care about the global lactate concentration they care about the local lactate concentration so that is why we are adding them so again it's the same speed of growth but now we also added the glucose and the pH into the equation and we run the simulations on two different scaffolds you have a gyroid geometry scaffold and here this is called D cup these are triple periodic surfaces they can be described with a combination of sinuses and cosinuses so it's easy to model and we can 3D print them in titanium anyway so we did the cell growth and we saw that the gyroid was doing much better than the D cup and we saw similar results in the simulations only that we were overestimating still the growth in the D cup so here you see the growth sorry here you see the growth so the filling of the new tissue there is hardly anything going on in the D cup but we were overestimating the lactate production and that was because we didn't take into account the seeding density so we just started with a layer of cells and then we had it grow but the seeding density on this cell was fairly low so if we adjusted the seeding density we were getting closer to the experimental values now this seems to more or less calibrate in some way the model back to the questions then so the questions that we had is what happens inside the bioreactor, can we use it for quality control and for optimization so what happens inside the bioreactor is that now with the model you can have a view like this so you can look at at a particular time during the analysis that is the local concentration of glucose lactate, oxygen at different locations in the scaffold and that will give you an idea at how many cells what is the percentage of cells that will experience a certain lactate concentration that will have an effect on their proliferative or differentiative capacities afterwards so you can look at niches in your scaffold and you could even change your scaffold geometry to optimize these niches how can you use that for control one of the things that we use in bioreactor control is the pressure drop so if you see that the pressure drop is going up we assume that the scaffold is actually filling with the tissue now if you look at the simulation results you see that by day 10 we're pretty much full so the scaffold is pretty much full and it's only at that point in time that we see the pressure drop going up so pressure drop is not sensitive enough to actually capture these early phases of filling of the scaffold so if you can combine pressure drop with the model you can actually be more sensitive in the control of your bioprocess and in terms of optimization here you see just a number of designs that we're playing with in silico to look at their effect on the tissue growth so the filling of the scaffold over time under specific cultural conditions and you see that certain scaffolds are behaving really badly and others are way better so that was for culture for clinics it's not so long I just wanted to show you what we're doing there mostly the questions are looking at adverse fracture healings and ways to solve it this is the result of a long series of studies including my own PhD where we have developed a partial differential equation model that looks at cells, tissues, growth factors and individual blood vessels that simulate the fracture healing process so it's a multi-scale model where we have a tissue level and we have an individual cell level for the blood vessels what you see here is a difference between a normal fracture and a large defect where you see that at some point it actually stops the blood vessels stop growing and correspondingly also the bone tissue will stop growing after a while and even after 90 days is more or less what we see here also in the experiment so if you take a non-union after 8 weeks you will have a mouse you will have this capping off so there will be a bit bone forming closing off the bones and there will be a large gap of course there is still imperfections that we had in the if I can go back this is too much bone that has been formed and we have been able to solve it why is this happening? you have a drop in the oxygen in the centre because the blood vessels aren't there and then it does go up because the blood vessels keep on growing because they get signals to grow into the centre because of the hypoxia but by the time they reach the centre and there is sufficient oxygen in the centre all the cells that were there have died because they cannot be even the chondrocytes cannot be without oxygen for that long so one of the things that we are now trying out is the oxygen releasing particles that we could put in a scaffold or that we could inject into the fracture side which could keep the oxygen at a certain minimal level so these are beads that would release oxygen and using the model we have actually come up with values saying if you have at least a certain percentage of oxygen or so many beads you should be able to cover that first period and you should get full healing we are actually now at the stage where we are testing the release of the beads and then we can implant them with this fracture healing model we have done a lot of things we have looked at mechanical loading we have looked at different injection of stem cells at different times and places so we are getting fairly confident that there must be something that is right about this model because we can do so many things without really changing it and I just wanted to come back to my first slide with this with the neurofibromatosis we took the model and we adapted it for the neurofibromatosis case where from the literature there were like eight factors that were changing dramatically in these cells affected by neurofibromatosis and we adapted them in the model now the problem is this is a very rare disease there is no way of getting these parameter values so we were left with just a purely academic exercise where we used an experiment approach we decided to go for 200 combinations of these eight parameters somewhere between hugely affected and normal and then just run the simulations and see if we could recapitulate the clinical observations so what you see here is for two of the parameters 200 dots representing the combinations that we tested but we used an eight-dimensional space so the eight parameters were varied all at the same time so what you get then here is for a result for one of the specific parameters you see the evolution between zero so completely bad and the normal situation and you see that you have a non-union because a high value for this complexity index is a combination of a non-union and a lot of fibroblasts present which is indicative of neurofibromatosis so a non-union for the entire range and unions only from a certain value onwards so you need a minimum level of this parameter in order to have healing to have healing possible if you wait longer so instead of three weeks which is normal enough in mice in mice to have healing if you wait five weeks you do see that a number of these dots are dropping so you get healing for lower values but it takes longer time so what we see is that there is a wide variety you get anything from complete non-union to union but a lot of fibro-stiches so very prone to re-fracture now we don't know how to link this to patient characteristics so that is still an open question but since we were doing the modeling we extended it just a little bit by adding a BMP treatment which is what is often used in the patients so we took those same 200 patients and we gave them a BMP treatment the growth factor treatment and these are the results so what you have here is the complexity index without BMP so these are the non-healing patients that we had before if you give them the BMP we look at how this complexity index changes so if it goes up the situation is now better if it stays zero the situation is the same and if it is negative the situation is worse so these are the non-healing patients which have a high difference so they are actually improved by the BMP treatment these is a group of non-responders so they had a non-union but they don't respond to BMP which is something we do see in the literature or we know from not in the literature we know from clinical experience that BMP treatment doesn't work with all of the patients but we don't know why but we do see this population popping up in the experiments we also have the group that were healing before and they do not respond to BMP which is a good thing but we seem to have not so many of them but a fourth category which we're healing before but if you add BMP you make it worse which is possible because BMP is not an undivided success it triggers certain pathways and if it just is triggering the pathway that was too much affected by the NF1 it can cause an adverse effect so now we have to somehow come up with some kind of justification for this and this is really tricky because you will never know what happens to a patient when they receive a BMP treatment what would have happened if they didn't have that BMP treatment that's only something you can do in in silico and not in reality so we're kind of struggling with that but I just wanted to show you because it still gives some interesting insights I think now to wrap up what we did is we used modeling along our tissue engineering pipeline we mainly use it as quality control but as all models it can help in the 3Rs and in personalization and I just had two more few more slides asking the question whether this is utopia or reality it's not just that we have things like the VPH so groups that are working on this so you're not alone it's also the FDA and Blanka already mentioned it FDA actually allows now models to be used as part of a preclinical dossier and finally finally we actually got EMA to agree to also do that so that is a good evolution and that will mean that companies will also be more interested in developing these models FDA is actually a step ahead of us and they already have this set of guidelines if you use a CFD model what do you need to have in your preclinical dossier to be able to admit it to FDA so that is also an interesting document to look at and then finally I know it doesn't mean much maybe but if industry is getting involved it must mean that we're onto something that might someday become really a reality so that's it for me I would like to thank all the people whose work I've shown here and that's mostly the underlined people my funding sources and you for your attention thank you Liz for this great talk so questions yes there is evidence in the literature that non-steroid anaergetic therapies are reducing the bone formation process if you are planning to design these scaffolds and these will be implemented during surgeries it will be I think a very important aspect to show this effect or somehow to simulate and calculate because these patients will receive after the surgery a large amount of non-steroid anaergetic therapies so maybe from clinical side it has to be a change but far today this is the first line of drugs but the patients are receiving you're absolutely right that this inflammation is critical and it's something we have not looked at neither in our models nor in our animal models actually because we're using nude mice as I said because we have human cells we see that if we use mouse cells in wild type mice we already have a problem with the inflammation response due to the scaffold so if you then add that to patients that have a problem in that area as well it's not so guaranteed that it will be an undivided success we didn't see any problems with inflammation in our large animal studies but if we then go to patients which have problems with the inflammation or which receive these anti-inflammatory drugs we don't know at this point the effect is so perhaps modeling that that would be good but we start from right after the inflammation phase at this point any more questions? maybe a bit more political question it's like when you look at all the things that you're doing so you have these gene networks you do computational flow you do experimental stuff you do like bioreactors it's like if you want to do research in this field do you go to mechanical engineering do you go to biology do you go to medicine what's your experience? it's a very good question maybe I should enlighten my complicated affiliation system in Leuven I'm affiliated to mechanical engineering because they have a biomechanics division but if I'm there physically I'm in the hospital and I run the wet lab for my colleague who is a clinician so he doesn't have time to run the wet lab so I'm responsible for that so I'm not paid to do that but I'm still there but that gives me an access and some very tiny level of control over experimental work as well in Yesh my education is affiliated to the mechanical engineering department for the master of biomedical engineering but my research is affiliated to the inter-faculty group for novel technologies in medicine let's say so I am kind of in between a lot of chairs and I don't know if the affiliation is necessarily the biggest problem but it is somehow translated in the way most funding institutes work and that is my bigger problem that when I go to typical engineering panels with my research they will say yeah you know I cannot really judge this but I don't think it's original that's literally a quote that I got from the evaluations so I have more problems in that direction I was lucky enough to have an ERC in an engineering panel but they really try to be aware of the interdisciplinary aspect and make sure that they don't penalize it but locally I've not been so very successful but partially related to that first is like locally when you see that you're in a place and then maybe you get some colleagues that say this is not the most interesting for a department what's your reaction then you say like no you're going to fight for your position or you say like let me look for people which are more open to things and do that in a way and go there it's about the people really if you I'm not like James yesterday he's hiring all of them I also hire biologists but I still feel more comfortable if I have a colleague or a biologist that is looking over my shoulder so for me I need to have these people that I collaborate with and I've been lucky enough to land in a department in Liège where they were interested in me because they didn't know anything about this research so that was fine so they don't it's not to say that they don't care what I do but they know that what I'm doing is something that they don't know much about so in that sense I was lucky I would say if you're in a department where they don't really see the value of what you're doing and you're not in a position to change that I think it's better to go look for people who do see the value of what you're doing and try your luck there so it's a matter to really look for groups and places that are interested in what you're doing rather than what they already do or like how things are going so I was convinced of that I wanted to do modeling and I wanted to have a close link to the biologist and when at some point in one of my departments I was at they said well this doesn't interest us well then I knew enough I couldn't stay there in the end when you're successful they might change their mind anyway but it takes some time so that's important I think also for the younger students it's like really if you believe in what you do you will find a place where they appreciate it but you have to look for it you have to go actively for a place in order to do that but this is really really important maybe another thing since we have quite a lot of young researchers here so you also as reasonably young researcher you have an ERC which is of course the thing that in Europe is I think the thing that everybody should try to do at least in order to be at the front edge of research maybe you can say especially specifically for this type of research and the modeling what do you think is your approach to get successful applications there so that would be successful in applications for ERC because I've gone with the same research to regional funding and that has not worked always but with ERC what I proposed was really the program that I wanted to do and the thing with ERC is they like high risk high gain so I just proposed something that I would not propose regularly but in this case it was products and process engineering but I still came with half of my grant which is mostly experimental so it's half modeling half experimental and what they did ask me in my interview at the panel was okay what experiments are you going to do what they are for so they wanted to see if that was really me who had written everything and that I was able to explain everything but after that they then only asked questions about the engineering part but at least they realized that this combination for biomedical engineering was absolutely necessary so for me that was the interesting thing with ERC that I could just have a program that was a bit bigger than just one PhD student or one company or something like that just really if I could choose what did I want to do well that was the thing that I presented there but especially if you propose a project which is like partially computational partially mechanics partially experimental how do you convince reviewers that you are able to do these experiments because I mean you have a different background or that you already prove that you could do the experiments or how did you convince them practically this ERC I had an additional participant so it's a two institute ERC so I had in Leuven I had Prometheus Group and I indicated that I would be collaborating for the experimental parts with Frank Leut who is my colleague in Leuven and he has experience in those kind of things at the same time I also showed by means of publications where I was co-author that I had been involved in experimental research and the second one that I applied for that I will have to defend soon in September I did the same thing only now I made myself the external collaborate or the additional participant in Leuven so I took responsibility for both the experimental and the modeling part and by arguing that I've been more and more and more involved in the experimental part and that now I would know enough to be able to supervise it myself in collaboration with others but you explicitly said that you were at two places so one place with this background one place with that so that's maybe also an interesting message for some of you that are trying because it might be tricky but some of the experiments of other people that we know is that you say like in one place okay that place has knowledge in that but this collaboration maybe if you do it more explicit saying okay I'm affiliated with both then might indeed. But it's important to show that you're in control because that was also a question I got the first time if you have this additional participant and he's world renowned in a certain field how are you going to become a leader in your own grand so you need to show that you are a leader that's what they're looking for apparently so you need to give them arguments as to why you would be able to do that and that's very important and it doesn't need to be just science it can be anything like if you're involved in your scientific groups or whatever you're organizing things you're taking the lead somewhere doesn't need to be just a number of papers that you've published so there's a lot of ways to show that you are maturing and that you're becoming a leader. Any other questions? Yes I have one so you've been working a lot in tissue engineering so nowadays many people question the real value of tissue engineering and even research in tissue engineering because it costs it costs a lot of money it's been affordable for most of the companies and the results after several decades of trials the results are still at the beginning still incipients and there is nothing really powerful so could you already have through your collaboration a guess on how your modeling work can either shorten the time of studies or the cost of studies in order to progress towards successful strategies? At this point the closest we are there to answer the last part is with the bioprocessing so I think if we want to make it somehow viable and robust because there are two problems amongst others in tissue engineering one is the cost which is huge and the other one is that the treatments are very much not robust so it's a lot of manual interventions if you go to a company that is producing these cells basically it's just a web lab you have a flow cabinet and that's it so they're still doing everything by themselves there's no technology there so by entering technology and all the things that we've built up in terms of knowledge of pharmaceutical production in chemical engineering we would be able to reduce some of the costs and increase the robustness and I think that would be a big step forward and there in order to have good control algorithms afforded by processes these type of models that I showed can help in another way there is the definition of unmet medical needs I think tissue engineering solutions are so expensive you shouldn't use them for everybody at least not the allogeneic approach you should really have an unmet medical need for which there is no treatment which can justify the slightly higher cost in that sense so these would be some elements of an answer your last question since we still have a couple of minutes do you think do you believe that in silico results can contribute to kill a paradigm so for example you were showing results on BNP and indeed it was not that nice in terms of clinical in terms of clinical safety and actually in the clinical world BNP are more and more criticized but when you look into research it's still a very strong target of research people keep on trying using BNP in order to treat patients so do you think that in silico simulations can influence on the shift from one paradigm to another I would say that it might help in identifying the patients for which you would want to use BNP and for which you don't want to use BNP at this point the simulations that we did there's too little data so we need to do a lot more simulations if we want to be able to assess why in this fourth category of patients it has adverse effects if we would know what kind of what causes these adverse effects and if we would be able to trace that back to patient characteristics we would be able to do a patient triage basically and say okay with this profile of patients you would be helped with BNP but this profile of patients we need to be careful not killing a paradigm but being more careful in where and when to apply one or the other I would say of influence the power of influence well I think that companies have every interest now in being precise about the patients they know that blockbusters are that something that everybody know so they know that they have to go to more careful identification of patient groups so if you can show that well they are becoming interested in modeling also in the pharmaceutical industry but yeah the power it depends on again if you have a good collaboration with a company if you can convince them that you're serious about your data if you have some data some experiments to back it up I think if you have a good package and good results they will listen to you because it's in their best interest well since we are not producing the BNP so I think it will be even to population and health care because the companies they are not happy not to give drugs to people they try to give it to as many as possible because of course that's where their money comes from but as you said trying to go to personalized medicine will be more important but I think it's much more health care and costs that will put at cost efficiency so I think that's quite important okay I think that was it for today for this morning as today we will start with the practical sessions again in the same places starting from around 2.30 and additionally today is our day of our kind of social event which would be like a beer tasting we will take place here in the courtyard starting from 6.37