so this uh this work that tam and i are presenting actually is the work of many individuals within uh biology chemical engineering the center for biomedical innovation and the coke and here's the team so tony as listed there in biology scott shaw tam and me who are presenting for the group uh paul yorgos museum stacy jackie caleb and andy and that's basically our group and so we've been doing a series of work in the viral vector space and this is some of that work we're presenting today all funded uh this part is funded by fda so conventional rav production workflow results in a high ratio of empty capsid in crude harvest and so the factories of course are cells you can see them in the bioreactor there and you add some plasmids to them and then you know hopefully what comes out is aav now there's a variety of ratios but it's quite common to have ratios of uh full to empty so the full is shown up there on the left and the little part of the pie um so on the order five to thirty percent and then i get a lot of empties and it's been reported 70 to as much as 95 empties and that's not so great because those aren't providing a payload for your gene therapy and so what we'll talk about is actually the use of mechanistic models um and so normally in this space everybody would be applying uh just a data driven type model so that'd be the bottom of this pyramid sometimes i call this the knowledge pyramid this one we got from world health organization and they'd be sort of at the bottom of the pyramid and doing a lot of experiments a lot of trial and error and maybe they're doing some fractional factorial design to try to optimize um so we took a very different approach here which is we just took the literature and wrote down mechanistic model it was the first mechanistic model actually for triple transfection aav and from that model we went ahead and made predictions by analyzing the model and then designed experiments and then significantly increased the ratio of fold to empty cap system so that's basically what the talk is going to be about today so what i'll do is i'll hand this over now to the to the expert so the lead graduate student on the project who will continue to discuss about the mechanistic model that we developed and its application for increasing the the ratio of full to empty capsids so thanks for surprise for the introduction so i'm going to introduce and walk you through how we went about constructing the first model for av manufacturing um using this mechanical model so we constructed the first of its kind model for production of rav via triple transfection and we use knowledge from biology cell biology and process dynamics and we utilize experimental data that we generate ourselves and also leverage from literature data to make sure that when you know less bias in that aspect to estimate these model parameters and with those data we uh guide experiments and use experiment to iteratively improve upon the developed model so we expect that the analysis from this mechanics model will review the trajectory and the bottlenecks of rav production and improve our capabilities for process designs so how do we do it so we just got through literature and found that there are actually models for plasmid delivery into the cells for the plasma into the cell nucleus so that we can start gene expression but uh for the process of three plasmid at the same time and how these blasphemous gene expression come about to make your final products capsules and full capsules are have not yet been modeled so we adapted a single transfection model and extend it to triple transfection hypothesis is that all plasmids have the same delivery kinetics so in the next step we build a recombinant viral vector production network by adapting the knowledge from wild type av synthesis and constructing a viral virtual network from the three plasmids once they're successfully delivered to the nucleus so the model includes elementary reactions that describe major connective events where you have the red protein synthesis the viral dna and the viral proteins that are the building blocks of the capsules and how they are assembled and the viral dna are packaged in the nucleus and cigarette nucleus so one notable difference between the anti-capsule phone caps that is that um they can't be packaged on the nucleus because they have a virus and once they're secured down nucleus they will be anti-capsule and result in the product and quality impurity that we ultimately don't want so we leveraged single transfection data from the literature to estimate delivery kinetic parameters so on the left hand side figure you see that our mother was able to capture the dynamics of cellular plasmid content over time and on the right hand side is a prediction of the blasphemous content in the nucleus so it showed that our model was able to predict the quality of nuclear plasmid and next we estimated viral production parameters using data from our in-house experiments so you know in this space we deal with small data and not well characterized process so a brush was different so we characterized samples for each of the time point in the process so we measured and we characterized the samples in terms of total quality of capsules over time number of full casters over time and also quantity of all viral dna species over time and this would include uh the vector plasmid like the aav vector gene of interest plasmids the replicated vna and also encapsulated dna that are inside the phone phone capsid so on the left hand side we see the total replicated dna and that in in plaque and the red in the case of dna that are actually incapacitating the capsid and you see that at the first 24 hours most of the replicated dna is actually in the capsids however as time goes on you see that there's a lot of dna that are produced but they're not incapacitated at all in the in the full in in the capsids so that like that intuitions like let us to think that perhaps is the capsule production that is limiting and you can see that that showed on the right hand side figure that over the first 24 hours that that capstone dynamic production happened most of it in the in this phase and the capsule production for both total like uh total plasmid and phone uh total caps that unfolded but told uh right after that so it could very well be that the reaction where you have the viral packaging into the capsid the capsid production the capsules number in the nucleus that are the limiting step so before we did analysis we benchmarked our model to the productivity at different blast mid ratios input to verify that our model has somewhat predictive capability and this mod this this data is from the literature uh so but then uh they use a quite similar production platforms we were able to compare these two uh and then you could see that for the very standard baseline case where you have three different plasmids at the constant at the same ratio molar ratio uh that is the one poland one here and we could see that i was able to predict which one would give the best improvements compared to the base case uh this is when you have a balance between the helper plasmid and the packaging plasmid so to identify where the bottleneck of the process are is we did a simulation of on the model species and also similarly the reaction rates over time and we found something uh that is interesting that you could see here uh so what the capsid and synthesis and dna replication dynamics there's a there's a mismatch here that most of the capsid synthesis is produced over the first phase of the process and i believe this is line up to before 24 hours and later the dna replication really peaked really later on when you already have all the capsid tenses here and the reason for the calc synthesis taper off because red protein is known to down regulate its own gene expression which is on the red cap last minute and we have like way fewer available free packaging plasmid to reduce capsule later on so the result suggests that capsule production at a later time faces the limiting step and that extending the capsule protection production timeline could improve the phone and empty capsid ratio and you we could have more full capsules at later phase so with that in mind we designed a kind of an unconventional way to reduce capsid uh using plasmid transfection so the conventional transcendence transfection method involves dosing the culture with plasmid only once at the beginning of the culture so we call this the multi-stage transfection method in which we feed the cells with a plasmid multiple times over the time over the production timeline with the goal of reducing the empty particles and it's quite it's um okay sorry about that so it's to the this is an outline of our sprinter design so we use the constant total amount of plasmid only that we divide them up to different uh amounts so you this is the whole dose at the same time and then we just split the doses in half and then transfect them at different times and then we split the dose in in three and then transfect them at three times so we split them equally and in this experiment we actually involved more measurements and even the last time and we measured that red protein we measured three different plasmids we measure total replicated dna and obviously the total calcium forecast which is the critical brother quality so uh here's the result first i wanted to emphasize about the cell viability and cell density which is very crucial for producing because you know only bio viable cells life cells can produce viruses and this experimental design gave us the opportunity to characterize the transfection dependent dynamics of cell growth and death which in turn affect the abrasion capability as the transfection reagent is known well known to induce cytotoxicity to the cells so in these figures one stage transfection is color coded in blue two stage in purple and three stages in green and it will stay the same for all the other figures you will see so ex as expected spreading out transfection in small dosage resulted in greatly improved cell density and viability as you can see it didn't drop as much uh for for dosing the control all at once you see that the viability dropped immediately and it just it reached 60 by day five whereas the viability for the three-stage inspection culture remain quite robust until even they fight it stay around eighty percent and because it has been uh in in in in analysis in data it uh i found that the net cells actually uh have compromised a membrane that make them lose their ability to bind with a transaction reagent so it wasn't to no surprise that the multi-state transfection experiment was able to produce more to we could see more blasphemous uptake later on in the cell country because we have more life cells and we have more plasmid uptake so that's very encouraging results so now uh once the plasma delivered and the viral components are starting to express and it starts with the red protein and you see that the multi stage inspection has a delayed peak and lower peak than the when you dose on the plasmas at once and despite having more plasmid copies in the cells compared like as shown it's in in previous in the previous slide figure as so we rationale this by the the fact that even though you have more plasmid this culture has already gotten to its later stage and there's a quite a lot of product my product built up and notably ammonia and those actually came back to inhibit the ability of the cells to express to translate their to produce proteins later in the culture and then uh the protein then triggers the replication of viral dna hence the name so unlike the dynamics of red routines synthesis shown in the previous slide the amount of viral dna is quite comparable across transfection methods so you could see that even though the total copies of replicated viral dna there are you know differences from different number of stages but with errors bars accounted for we don't see a lot of differences so they're quite comparable so this is results suggests that although the red proteins is essential throughout dna replication the kinetic is not sensitive to red protein concentration once a certain amount of breath is already there to trigger the the the reaction now uh i'm going to move on uh to maybe give a bit of background how we might we think about the process and model it so once we next to red protein is obviously viral proteins and capsid assembly once we have the capsid the red protein is also known as being a catalyzed catalyst for viral capsule packaging and now we have empty capsules and and phone capsules and they have they're in the nucleus now they have to go to the cytosol and they have to secure to the media so they have to go through a few steps to get outside and what's interesting is that these viral capsules even though their aav is known to be quite known to be quite stable but they can be degraded in the cytosol uh by cellular nuclease protease sorry about that um so the longer they stay in the cells that the more prone to degradation that they could be so in our experiment we measure the number of total capsules the number of phone capsules and also the number of security capsid so these allowed us to see what's really going on and constrain our model so moving on i'm going to show you the data for total capsules produced in total and also outside the cells so you could see that there's a clear differences between the total capsules produced by three different experiments and you could see that this mirror somewhat the level differences in the red protein production so because of the protein the gene expression capability differences that i've discussed so we don't see a like a immediate level of of the number of capsules compared to red protein this reason being that assembled capsules are less prone to degradation compared to individual proteins and also this kinetics is also affected by the number of red proteins available in the cell because then dead deportings are known to upregulate the p40 promoter of cap gene and you could see that there's a clear visible decline in total capsule being produced so this is what would point back to my point that assembled caps that can degrade and this would be a quite um valuable indicate when we should harvest our products because i i and that has some questions about how long you should run the batch for and i think this is quite indicative of batch optimization and operating operation optimization so now we will focus on the final product where um the viral dna is packaged inside the capsid so you can see that uh despite some minor differences the total amount of phone phone caps that we achieve over across three different experiments are quite comparable but what is really the improvement here is the phone to total cassette ratio so you see that we saw a lot of total capsule produced for the one stage for the conventional method but we don't see a lot of differences in the phone caps that's being produced so the bottom line here is that we were able to achieve a greatly improved phone to total capacity ratio without compromising the viral genome tighter and that our model was actually able to capture on differences across transfection method and that's bring me to the conclusion that we developed a mechanistic model that captures rav production dynamic changes across one stage transfection or even three transfer methods and we did design and demonstrated that a novel and unconventional way to do transfection that actually improved phone to total capsid ratio in harvest without compromising the viral genome titer and this would pave way for the improvement of viral product quality not just in terms of optimization but in different ways that we can play around with transfection to really leverage the bottleneck that's impeding the the process and industry and with that i'd like to thank you for your attention and i'll take any question [Applause] okay it looks like we have two questions um i'll read them aloud so our live stream audience can hear what changes can you make to enhance the production or stability of the full capsids capsules oh this is a great question so we are under uh we have some ideas in mind that we're actually carrying out so uh so from the indications model that you could think about aligning these timeline of viral production components actually would greatly improve your product so one way that you could think of you could play around with adding these splatters so we add all the plasmid in three different stage but you could actually actively select which components are lacking at each stage and add those at appropriately so i believe that using model guided experiments can actually be beneficial in in these experiments yeah that's a good a good answer um one one sort of may be implicit there is why did we focus on the ratio right why didn't we focus on just increasing the the total to begin with um i mean sorry the the full to begin with and actually the reason is that downstream from this process there's a separation that has very poor yields which is united in exchange to separate the two because the capsids are very similar so we focus on this part first as a first case study because that gave us you know if we could get it high enough the ratio you could actually skip that other step which has like 50 yields and get rid of unit operations that's why we sort of focused on that but of course now we're getting progress there now we're switching our focus to increasing total so well increasing total full i guess so the total amount of full [Laughter] okay yeah have you done cost analysis of transfecting once and getting 30 filled capsids versus transfecting multiple times to increase the percent filled capsids thinking about the cost of actually carrying out the multiple versus the single oh the the cost would be the same like the only cost that would be additional is labor cost because uh i think there might be some point that maybe didn't get through is that we use the same number of total caps the total blast mets which just divide them up to different smaller doses every time yeah the cost is basically the same and we um you know you could have labor but in our lab including with collaboration with cbi and elsewhere is that you know we do fully automated systems all the time so then you can get rid of that excess labor cost so even better um how do you ramp up the multi-stage transfection to a commercial scale you i believe that this is a great question because i i do a lot of bench scale so this is a maybe a naive thing to say but i've imagined you could add plasmid at different time is can you imagine that we do do the same thing we did basically do a larger scale yeah um practical implications of implementing triple process in production i think you answered that question would you observe a similar threefold increase in full to empty ratio with three staging transfection approach if you adjust your model parameters to get to the higher end of typical full to empty ratio i'm not sure i understand that but um i think it would be so the thing is we don't we don't we're doing our optimization of the process right we're not optimizing model parameters model parameters coming from the biology you know fit infinite data so so we optimize is optimize operational parameters such as the way we do the the feeding for the plasmids so um so don't fully i guess you know we're not like trying to do model parameters to get some optimization because that's not what matters in the real system the real system of that is just the operational parameters that's what we're focused on so okay and is this model true for all serotypes [Laughter] and so i i did discuss this in the the the later part of the published model that it serves as a framework and it's not guaranteed to be true for the stereotypes because the secretion rate is different you know some certainty doesn't really secrete at all aav2 so you would have to tune your parameters for your purpose and just sell light but i think it would serve as a robust framework for you to apply yeah i know it's a good answer i think the thing is is um like tam was saying is that you're going to see some differences but what we saw in this case right so we wrote when we did the first model there's a lot of uncertainty in the parameters right it's the first model it isn't like there's huge amounts of data we had access to but by doing a simulation the model to see where there was a mismatch of production of two intermediates that was needed to give full capsids that was enough information to have us think of a particular type of operational procedure that we had pretty good confidence would would be in the right direction so instead of doing a lot of trial and error experimentation or fraction factorial we did a very different kind of design motivated by a model and then implemented just took a big step right away so and so that's really the strategy we try to do in all you know in all of our projects that's always the goal right do as much as we can on the computer get as much as we can from what's known in cellular biology and day that's available take all that learning into a model and then make the kind of decisions appropriate at that level of uncertainty and take a step forward now we're going to switch to another stereotype what you would do is uh there's this thing called transfer learning right so you could start this with a basis and then say oh by the way some stereotype you can't you can't excrete okay well we'll just set a zero to that parameter there'd still be a lot of uncertainty but there's only a lot of uncertainty in our model too right and we got a lot of value out of it so well thank you very much professor bratz and tam newen [Applause]
2022-07-31