Eppendorf Customer Webinar: Precise Bioprocessing Through Modelling

[00:00:11]Hello and welcome to this Eppendorf Customer Webinar where experts share their knowledge and vision for their work in bioprocessing.

[00:00:20]Today we are joined by Diogo Peres Dos Santos, a bioprocessing engineer at the National Institute for Bioprocess Research and Training (NIBRT), which is based in Dublin, Ireland.

[00:00:31]Diogo will tell us about his work on how modeling can help to achieve more precise bioprocessing approaches.

[00:00:39]If you have any questions to Diogo after watching this webinar recording, reach out to him or to us.

[00:00:46]The contact information will be available at the end of the talk.

[00:00:49]And now without further ado, I hand over to Diogo for his presentation.

[00:00:57]So hello, as Stefan said, my name is Diogo.

[00:00:59]I am a Bioprocess Engineer here at NIBRT.

[00:01:03]So, we are going to show you a bit about our research, how I am doing precise bioprocessing through modeling here at NIBRT.

[00:01:10]So first, I would like just to showcase that my name is Jee-oh-go.

[00:01:17]So a bit different because it is a Brazilian name, but that is how it is pronounced, and Stefan did great.

[00:01:23]So, now a bit about myself.

[00:01:26]So, as you know, I am a biotechnology engineer with a master degree and PhD in chemical engineering.

[00:01:32]I was working at a Brazilian animal pharmaceutical company here at the south part of Brazil, so right down with the red dot.

[00:01:42]And then what happened? I crossed the Atlantic Ocean and came here to Ireland.

[00:01:46]So, how did that happen?

[00:01:48]What happened for me to get here? So, I was recruited by Professor Sakis Mantalaris and Nicki Panoskaltsis, and they had a grant about metabolism driven precision biomanufacturing of cellular therapeutics.

[00:02:03]That was granted by Research Ireland, their logo is right down there.

[00:02:08]So, they needed a bioprocess engineer to develop this approach of modeling in their grant.

[00:02:13]So, now that you know who I am, what I am doing here, and how I got here to Ireland, let's talk about a bit about our work.

[00:02:25]So, this grant is about an interesting interdisciplinary approach to address heterogeneity by understanding and directing metabolic change in the cell.

[00:02:35]So, this is composed of four main pieces, that's one for bioinformatics, that's going to construct a metabolism database, one for cell biology that's going to characterize this heterogeneity by single cell analysis or population analysis.

[00:02:51]Then, there is bioprocess engineering, that is going to increase precision through modeling, using models to help with that.

[00:02:58]And then, there is also analytics that going to define metabolic biomarkers.

[00:03:03]So, all these areas will work together to make a interdisciplinary approach to tackle this challenge of directing heterogeneity that is happening in all cellular therapeutic products nowadays.

[00:03:18]So, and what is my task?

[00:03:20]My task here is being the bioprocess engineer.

[00:03:23]I am working with the task about bioprocess engineering.

[00:03:26]So, I am trying to increase precision through modeling, making models for these cell types.

[00:03:33]So, what do I do? I develop these models, and the mathematical model is just equations that can predict cell behavior.

[00:03:42]And then, we use it to optimize two main factors.

[00:03:47]One is the culture media, so think about the food that the cells need, the nutrients that they need to grow and have the quality for their theraupetic activity and then the feeding schedule as well.

[00:04:02]So, you need to think when and how much of that nutrition you need to give the cells in the right amount, in the right time, so that the cells are not overfed or starving as well.

[00:04:17]So, everything we do is thinking about the critical quality attributes or, as you know, the things that make the cells have their therapeutic potential, and reach their therapeutic dose and then have the activities that we know they are capable of.

[00:04:36]So everything we do here is thinking about metabolism, because even the best cells are not going to thrive in a environment that doesn't feed them enough.

[00:04:46]So, metabolism is the key to understand what they need and how they need it, and how it is going to control their fate in the bioprocess.

[00:04:57]So, and how am I doing this modeling?

[00:05:00]So, basically I'm going to show you a basic bioprocess.

[00:05:04]So, if I have a bioprocess, I have a bioreactor here, normally, one of the main things we do monitor is the viable cell number or concentration and the dead cell number or concentration, that is a key process variable in my bioreactor and we know that they are controlled by the growth rate.

[00:05:31]So, how fast the cells can grow it is going to determine how many cells we have in the bioreactor and the death rate is going to control how many dead cells we are going to have, as well.

[00:05:45]So, they (variables) are controlled by this parameter that I created.

[00:05:50]And we know that these parameters are dependent on some nutrients.

[00:05:54]So, think about, growth rate is going to be dependent on glucose, that is the main source of energy for the cells.

[00:06:00]And the death rate is going to depend on some kind of toxic metabolite, that's lactate.

[00:06:06]So if the lactate is super high, then cell death is going to happen, it is going to increase and then dead cells are going to start appearing in the bioprocess.

[00:06:16]But then we can have a different effects as well, negative effects.

[00:06:21]So, if you think that the glucose is high, then it is going to affect negatively the death rate.

[00:06:27]And if lactate concentration is low, it is going to be good for the growth rate.

[00:06:32]So, then you are going to start making connections between the nutrients and the rates that you have, the parameters that you have.

[00:06:41]So, we are going to keep building that map, that's going to control everything.

[00:06:47]And these dots, the connection dots, are going to be equations.

[00:06:51]So, I am going to explain the equation that's going to control that.

[00:06:56]And then, the simulation can try to calculate it and then reach the actual growth rate, actual death rate, and then reach the actual viable cell number and dead cell number.

[00:07:10]So, this is just a simple model.

[00:07:13]So, in the final model you are going to have several more nutrients, like all the amino acids, and more toxic metabolites, like ammonia and other things.

[00:07:23]So, going to this map is going to be way more complicated, but that is just a basic model, so you can understand what we are doing.

[00:07:33]So then, if I know everything, I am going to first digitalize my system.

[00:07:39]So, I am going to put in my bioreactor and my feeds that I'm going to use. Then Im going to try to control how long it's going to happen.

[00:07:49]Here is a image from the gPROMS software.

[00:07:54]So, what I start to do, I start to write these equations.

[00:07:57]We are going to keep adding more equations to the system to explain every metabolite, to explain every growth parameter I have.

[00:08:05]So, everything is going to be inserted the model in the end.

[00:08:10]So, in a final model, it can have like hundreds of equations, some of them are going to be normal equations, some of them going to be mass balance and differential equations.

[00:08:22]So everything can be done here and then after I have everything set up so I can run the simulation.

[00:08:30]But now, you are probably thinking: the cells are complex systems that can be really hard to understand.

[00:08:37]We have been doing it for centuries and still don't understand everything that cells can do and how they do it.

[00:08:43]So, how can a bunch of equations, the simulation, be realistic and not just assumptions or a kind of hallucination.

[00:08:53]So, how can we trust that the equation is going to be a real prediction, predicting cell number and other things.

[00:09:03]So, what we are going to do, we are going to use experimental data.

[00:09:06]We are going to use real data, real experimental data to train this model.

[00:09:11]So, after being trained, we can calculate some kind of statistical significance, so we can check if the model is really predicting cell behavior.

[00:09:23]So, that is how we know that the model is good.

[00:09:27]And then, once we prove that it is realistic, we now have a digital copy of our cell that can be run as many times as possible.

[00:09:37]So, it is really useful, but now, you know that I'm going to need experimental data.

[00:09:44]So, how do I do it?

[00:09:47]So for us, we chose the DASbox® Mini Bioreactor System as our platform because it is a parallel bioreactor system.

[00:09:54]So, we can run a lot of bioreactors at the same time.

[00:09:58]And then, we can monitor all the common variables that we are going to have, pH, O2, temperature, CO2. And we can even put feeding, a flow of media going into the system and a removal pump as well, so we can take media out of the system (perfusion bioprocess run).

[00:10:18]So, we chose that because we can now run multiple experiments, and we can create feeding and perfusion runs, making the bioprocess optimization possible.

[00:10:27]So, we are going to use tissue flask as well to start.

[00:10:31]But then, we want to use a high throughput system, so we can generate data faster and then we can really optimize the bioprocess and it can be really scalable as well.

[00:10:46]So, that is why we chose it, that is why we are using it.

[00:10:50]And then, what kind of data are you going to be needing from this DASbox run?

[00:10:59]So, I am going to need to understand the cellular heterogeneity.

[00:11:03]That is why the cell biology is going to be very important.

[00:11:07]So, I am going to try to understand the metabolic signature.

[00:11:10]So, what is going to be driving cells to be different?

[00:11:15]I am going to be needing some kind of data about cell composition because normally there is some cells that differentiate or are modified in some way.

[00:11:26]So, I need to know what's going to be the effect of that modification as well.

[00:11:32]And basically I am going to need to know the growth kinetics as well.

[00:11:36]So, I am going to need to understand how fast they grow, how much glucose they consume, how much amino acid they consume.

[00:11:43]So, everything I am going to need to know from the whole population and some subsets, because then you can have some specific part of the cells that can be more consuming than the whole population.

[00:11:55]So, I am going to need to try to understand that as well and metabolism in general.

[00:12:01]But, I am thinking here as well about the energy metabolism, because energy is the main thing, so if the cells cannot generate energy, then they are going to die.

[00:12:12]So, that is why it is really important for us, the ATP generation.

[00:12:17]So, that is basically the data that we are going to need and now we are going to see how we can develop this model.

[00:12:25]So, the pipeline is based on experimental modeling.

[00:12:30]Some things are in vitro and some things are in silico.

[00:12:35]So, then we are going to use this approach being that some days we are going to be in the lab prepping bioreactors, and some days going to be sitting on the computer running simulations.

[00:12:45]So first, we can start with the model development.

[00:12:48]We are using a software like gPROMS to make the model.

[00:12:52]We then need to go to the lab to the bioreactor to do the first bioprocess run, so we can get the initial parameter values, calculate everything, so we can start using the values in the model.

[00:13:09]Then we can test the model, see if the structure is correct, if it is showing growth of cells and then dead phase or growth phase and everything else we expect and that we have seen in the lab data as well.

[00:13:23]Then we can run what we call sensitivity analysis that is going to try to make a statistical relevance of every parameter.

[00:13:34]So, it is going to then say to me that one parameter was really a key factor in determining the cell growth.

[00:13:42]So, then I am going to need to go back to experiments to really be sure that I capture the right number.

[00:13:49]Then we can go back to the lab to make more experimentals runs and then optimize it.

[00:13:56]And after doing that, I can do even more parameter estimations, try to correct predictions, train the model with the data that we now have and then try to finalize it.

[00:14:09]Then, as usual, we go back to a lab, do more experimental runs.

[00:14:16]Those ones are not going to be used to train the model, just to verify the model, to make a validation analysis.

[00:14:25]So, it (the model) is going to try to use the data, the parameters that it were generated from other runs to try to predict that run.

[00:14:33]And then, I can do some calculations, then it is going to say that the growth is validated, that is going to be predicting correctly the viable cell number, dead cell number and the other parameters that I find useful for us, that is how we get to a validated model.

[00:14:52]So, but now I think maybe you understand what I am doing.

[00:14:57]But do you understand why do I do it?

[00:15:01]Why do I need to have a validated model?

[00:15:04]Why would I want a digital copy of a cell?

[00:15:09]How can I use it?

[00:15:11]So, that is the most interesting part.

[00:15:13]So I am going to show you a real use case scenario that was made by my group.

[00:15:19]A few years back, it was using mouse myeloma cells that produce a specific antibody that detects colorectal cancer.

[00:15:31]What they did back then.

[00:15:33]So first, you have the experimental data that is the blue squares.

[00:15:37]So you see a cell, that is growing over time and then normally reaches a maximum value and then starts to die.

[00:15:46]Then, we have this experimental data and now we can try to make our model.

[00:15:51]So, try to do the development pipeline, try to construct everything and train the model.

[00:15:56]And then, we get this line, this dotted line.

[00:16:00]So, now we have this simulated data as well.

[00:16:03]So then, we have a simulation that is not perfect, but captured the trend line.

[00:16:10]So, cells were growing, then stopped growing.

[00:16:13]So, then we have now to think about that.

[00:16:17]It is now a validated model that is a digital copy of our cells.

[00:16:22]So, what we are going to do with that? Now we can start arguing with it, instead of going back to the laboratory and test new experiments, like start feeding some new nutrients to the cells, I can just, instead of going to the lab, ask the model: “So, what can I do to make the cells grow more, stay healthy and reach a quality attribute that I have now?”

[00:16:49]So, I can start testing a lot of scenarios and then I can even ask them: “So, try to maximize the viable cell number before the toxic metabolite accumulates as well.”

[00:17:02]So, then I can get this red dotted line.

[00:17:06]So, it is the answer of the model that says if you start feeding, you are going to be able to grow the cells into this data point, but then they are going to die eventually because of another toxic metabolite.

[00:17:23]So, then we have now a thing that did not cost to run.

[00:17:30]It is just a simulation.

[00:17:31]So, now it is going to saying to me that there is a way to optimize it.

[00:17:37]So, then what I need to do is to go back to the lab and test this theory that the simulation, the digital copy, said that was possible.

[00:17:46]So, that's how we got the red circles as well.

[00:17:51]You see that applying the feeding strategy that the model said that would be good, I could do what he predicted.

[00:18:00]So, I could grow cells for longer than expected and I even a bit more than the model predicted, but with a close estimation.

[00:18:10]So, we did optimize the run with just a few runs.

[00:18:15]Normally these things, take a lot of experimentation, a lot of tries and errors, so it takes a long time, but using modeling, we can do it faster and easier than the traditional approach of trial and error.

[00:18:31]So you see, if I have a digital copy of a system, you can optimize it in a way that is going to be way less cost-intensive.

[00:18:43]That is why we do digital copies of our cells.

[00:18:47]And then, we can even have different approaches of how to optimize it.

[00:18:51]And we can even link the experimental data with the modeling and then the modeling is going to try to correct the bioprocess run in real time.

[00:19:04]So, if you connect the model with the real physical system, so that's what can be done.

[00:19:13]But that was done in the past.

[00:19:15]So, what are you doing now?

[00:19:16]So now, our current focus is in cell therapy.

[00:19:22]We are using 5 cell types, that are really interesting, being a real, being really deeply researched and desired because they all have like really huge impact in cell therapy.

[00:19:40]So, I think cells that are being heard about like CAR-Ts, MSCs or iPSCs, cells that are being used in cell therapy, but they all need some kind of optimization or some kind of understanding how their metabolism works.

[00:19:59]So then, what we want to do is to be able to use this approach of modeling so we can scale up this production, make the cells grow faster in shorter time, shorter time frame of development because they are already being used but they are all really costly and if they stay in this price range, they are not going to be broadly available for the population.

[00:20:29]So, that is why, that is our driving motive.

[00:20:33]And then, if you can do this (modeling) we are going to drive down the treatment cost because there are cells going to be able to reach cell number faster, while keeping their quality as well as their therapeutic potential.

[00:20:50]So that is all, that is our goal.

[00:20:52]and that is why the group is focusing now.

[00:20:58]So that is why I am doing it. I started with NK cells, but we are going to be eventually doing it for all the cell types.

[00:21:08]So that's it, I think.

[00:21:11]I hope you understand now how we can use modeling to optimize a bioprocess, to help with the development of a bioprocess.

[00:21:20]I hope you liked the presentation.

[00:21:23]And if you do bioprocess, well, think if you could do a simple modeling as well, to try to predict it.

[00:21:30]So prediction is a really useful tool to be prepared for when you need to stop the bioprocess, when you need to go and continue it for longer.

[00:21:41]So I think modeling is getting really, I think, more easily and more easy to approach than it was back then.

[00:21:50]So now they have a lot of softwares as well and coding is getting simpler as well.

[00:21:55]So I think if you think about it, think if you can apply it as well in your bioprocess.

[00:22:02]And if you want to know more about our research, check our website, BSEL.ie and follow us in LinkedIn in X as well.

[00:22:13]So, I hope you will see the fruits of our labor soon enough.

[00:22:19]Thank you and have a nice day, guys.

[00:22:24]Thank you, Diogo.

[00:22:25]I think I like the presentation very much, so I can answer your question already there.

[00:22:29]Thank you for this very insightful talk and thank you, the audience, also for watching the recording.

[00:22:37]And, as said, if you have any questions, reach out to us or Diogo.

[00:22:40]For more information about Eppendorf bioprocess, you can also visit our website via the displayed link.

[00:22:49]Here, you can also see again the e-mail where you can contact us.

[00:22:53]We want to thank you for your interest and hope to see you soon in the next presentation.

[00:22:59]Again, thanks Diogo for your nice talk.

[00:23:02]Goodbye and have a good day everyone.