The RNAbox: Continuous RNA Manufacturing

0:00Hi, everyone. Thank you for the opportunity to speak here. So my name is Zoltán Kis. I'm a professor of bioprocess engineering at the School of Chemical, Materials, and Biological Engineering at the University of Sheffield, and I'm leading a research group looking at developing next-gen processes for manufacturing RNA-based vaccines and therapeutics. We're looking at developing processes that enable manufacturing and also development of these products at lower cost, higher quality, and against multiple diseases. So in a manner that basically we can fight multiple diseases, including infectious diseases, cancers, genetic disorders, and autoimmune diseases. And again, we are developing the underlying technologies that enable the development and manufacturing of these products, which are RNA-based vaccines and therapeutics.

1:05Besides experimental process development, we also do process modeling, and we combine these two aspects to enable technology development in a more streamlined way. And we are also building equipment that will facilitate the development and manufacturing of these products. So, one of the main equipments that we're developing, it's called RNAbox, because this is an RNA production process in a box. And I'll talk more about that. And after that, I'll also talk about some of the key exciting innovations that we are integrating into this RNAbox. One of them looks at reducing costs and improving quality, and the other innovation is a soft sensor which looks at enabling monitoring of the process, so knowing the product quality, product amounts, and process performance in real time, which is very important for a continuous process.

2:06And at the end, I'll wrap up with some conclusions. So I'll first give an overview and a background to this technology and to the, I guess, larger, bigger picture initially, and how these things work. So firstly, I'll start with introducing the RNA technology. So as you might know, mRNA stands for messenger ribonucleic acid, and this sits in between DNA and a protein. So the genetic information is stored in the form of genes, and this is then transcribed into this messenger RNA, which is a conveyor of this genetic information from the DNA, and it takes this information to the ribosomes in the cells where the proteins are being manufactured. And basically, this is how our body makes proteins all the time.

3:02So the information or the blueprint is stored in a DNA form, and the RNA is like a temporary message that conveys that information to these machines called ribosomes that make the protein based on the information coming from DNA via RNA. And this is how all the proteins are being made in our body, including all the structural proteins, enzymes, signaling proteins, and so on and so forth. And what we do here, in case of the RNA platform, we deliver this RNA molecule, which can encode for proteins that could be of a specific function. Those could be an antigen for the vaccine to generate an immune response and train the immune system to fight a specific pathogen, or it could be a therapeutic protein, such as a protein involved in fighting cancer.

3:57And basically, with these products, what we do, we deliver this RNA into the cells of the body, and then the cells of the body will manufacture the protein. And again, that protein could be a vaccine antigen or a therapeutic protein. So it basically gives unlimited possibilities for making different proteins, and encoding proteins in RNA, and instructing the body with this temporary message to make those proteins, which would have a therapeutic or a vaccination effect. Now, the interesting thing here is that, as you can see, our therapeutic molecule that we deliver is not the final product, but it is this information molecule, which is a temporary information, the messenger RNA or mRNA for short.

4:52And the manufacturing happens in the body, and this is a manufacturing of a protein in the body. And as a consequence, the quality attributes or the properties that we want to control in the manufacturing process are of those of the information molecule, the messenger RNA. And as you can see, these include things like integrity, identity. There's a cap at one end of the RNA, there's a poly(A) tail at the other end, and a bunch of other impurities there. And also quality attributes related to the lipid nanoparticles or LNPs for short, that are encapsulating the RNA for delivery. But the important thing to note here is that most of these quality attributes are independent of the protein that you deliver, so independent of the sequence. This is quite interesting because now you can think of designing manufacturing processes that can manufacture these mRNA molecules, these information molecules with the required quality attributes, irrespective of what protein you are encoding in the RNA.

6:04So that's how we are seeing this, and that's how we are approaching this in our team. We are looking at this product as a platform product. As an information molecule that can encode different proteins, and as a consequence, we're developing manufacturing processes that can manufacture these platform products, and we develop these as platform processes in which we would want to control the product quality via these critical quality attributes. And if they're critical, that means that they impact safety or efficacy or both of the product. We also would want to control key performance indicators of the manufacturing process. These include things like manufacturing costs, manufacturing productivity, and so on. And to control these, we tweak the input material, so these are CMAs, so critical material attributes, and we also tweak process parameters or CPPs, which are critical process parameters, to obtain the desired quality attributes and KPIs.

7:12So that's how we develop our manufacturing process development work, looking at changing these inputs, the inputs are CPPs and CMAs, to optimize these outputs, which are CQAs and KPIs. And as I mentioned, we do both experimental work, so we do process development. We're developing continuous processes. That's something new and quite exciting. It's a more efficient way of manufacturing products. And we also combine this with process modeling and advanced analytics, predominantly mass spectrometry and HPLC-based, but also other methods. And we're also building software that integrates all that information and provides an easily accessible user interface to the entire process and the models that enable this manufacturing. So basically, we integrate these four components in a core technology, which has continuous processing inside.

8:12So in this technology, in terms of processing, we do RNA synthesis using an enzymatic process, so that's a cell-free enzymatic process. Then we do purification of these products, the RNA that we make with this enzymatic process. For purification, we use chromatography techniques and filtration techniques, and then we also encapsulate this RNA, the RNA molecules, into lipid nanoparticles that then help to stabilize these products and help delivery of the products. So by combining these, all of these aspects, our aim is to develop these processes as multi-products, and that comes partly just by the nature of the RNA information molecule. As I mentioned, you can make different RNA sequences to encode different proteins with the same process.

9:02And with this, you can manufacture both vaccines and therapeutics. And we use also this quality-by-design framework and the prior knowledge, because a lot of the knowledge is transferable from one RNA construct or one RNA sequence to the next one, because many aspects stay the same in terms of process parameters and materials when you make different RNA sequences. And we are looking at manufacturing these products at high quality. We are monitoring over 20 critical quality attributes or CQAs with advanced analytical methods. We're also developing real-time methods for analyzing these, and these real-time methods are very important for a continuous process where we are making this product continuously, and we would want to know in real time whether we are achieving the right quality and the right process performance, rather than having to wait a week and having to run the process in the wrong way for the whole week, which would be very inefficient because we would make the wrong product, or we could make the wrong product for the whole week without knowing.

10:08So with this real-time analysis method, we can know in real time actually what's happening in a process. As I mentioned, we are also looking at reducing manufacturing costs. Here, the main cost drivers are raw material costs, and we developed some exciting processing approaches or processes that can reduce cost by reusing raw materials, so that can reduce the costs. And we also made this process quite compact, so it can fit into a smaller footprint facility, which again, reduces costs. And we also build a lot of automation, which would reduce the number of operators needed to run this process. And we would want to make these highly productive.

10:51For that, we use continuous processing, and ultimately, this technology, the RNAbox technology that I'm talking about here, will enable rapid development and manufacturing of products and will also enable technology transfer or basically transfer of this technology to different parts of the world, including to low- and middle-income countries, as this would be just one equipment to transfer rather than having to set up a whole process from scratch and buying, let's say, 10 different equipment items and make them work together in this process. So it does simplify a lot how these products can be manufactured in the future. And again, just to reiterate that, it's a continuous process doing synthesis, purification, formulation, and then purification of these LNPs as well.

11:47And the key advantages are that it would provide rapid access to this technology, including in low- and middle-income countries, and it will enable rapid development and manufacturing of these RNA-based vaccines and therapeutics at high quality and at low cost. It would simplify technology transfer, and we are building a lot of automation and leveraging all of this prior knowledge via the quality by design framework, which will ultimately lead to fewer highly trained people required to run this process. And it will also be highly productive, and it will provide flexibility in terms of how much you can produce. So it actually can bridge scales. It can go from making products for clinical trials, phase one, phase two, phase three, and also beyond that at larger scale.

12:42And that's due to the nature largely of the continuous process that you can run for different time durations. So you can run this for, let's say, one hour per day or 24 hours per day, and then you can make broadly different amounts with the same process. Yeah. So those are in a nutshell the innovations. And for each of the unit operations that we are developing, we are building equipment and we are integrating experimental knowledge and modeling. So we do this iteratively. So we take experimental data, and with that, we develop models, and then we use those models to simulate scenarios and streamline optimization of the process.

13:32And so yeah, the experimental work and the modeling work go hand in hand, and they basically create a synergy in which both are progressing. And we publish several papers on these different topics for different unit operations that you can read about via these links here, and there's a QR code as well that you can scan. Just to give an overview of the different development that went into this so far. So we have so far built a prototype of this equipment. This is an R&D prototype. It's not fully GMP yet, and currently we are working on a GMP prototype that should be GMP compliant ultimately. But for the RNAbox prototype, we tested different reactors, different purification methods, different mixers, and different alternatives for each part in the process.

14:36As you can see a few of the different reactors we tested here on the top, and we found one that gives optimal performance in the sense that it gives high yield, high product quality, and well-controlled residence time distributions in the process, which impacts quality. And then we tested various operating conditions and various ways of controlling the process, and we compared different pumps and valves and so on, and we took the best performing ones and integrated them into this prototype that we have the RNAbox. As you can see here, this is the scale of it. This part is the IVT synthesis reactor, then we have the purification units here and the LNP formulation. The mixers are here, actually.

15:29There's pumps using syringes there as well. So that's where we are now. We have built this earlier this year, and now we are working on the GMP equipment, which is actually going to look quite different from this one. This is just an R&D rig, but the GMP is going to look quite different. I think what I'm going to talk about next is these novel, I guess, developments or novel technologies that we have developed that will be integrated into the RNAbox. One of them looks at reducing raw material costs. And again, raw material costs are the main cost driver in this process because we use things like highly purified enzymes, template DNA, capping reagents, and all the materials are highly purified, GMP grade, and free of RNases and all sorts of contaminants.

16:24So a lot of work goes into making these materials. And in my view, these materials are not used efficiently in the process. So if you look at the synthesis, for example, what happens here is you have these raw materials or substrates that get made into products. So these are the monomers, if you will, or the building blocks of the RNA. There's four of them. These are triphosphates, so there's ATP, CTP, GTP, and UTP, and the capping analog, which is not actually consumed. But these get assembled into the RNA based on the template DNA using an enzyme. And what happens is that this is conventionally done in batch processes where all of these components, so all of these raw materials and these enzymes and template DNA and these buffers and so on, they're all put together, incubated for a certain time, normally around two hours, for the reaction to happen.

17:27And after this reaction happened, the product is purified out, and the rest of these materials are discarded, are thrown away, and that's not very efficient because as you can see, many of these materials are not actually consumed in the process. The enzymes are like catalysts. They are not consumed. Template DNA is just there to provide a template, but it's not actually consumed again. And it's just not efficient to fully discard these after every process. So then what some people have been doing, they have been immobilizing these components onto solid substrates like beads and so on into the reactor to keep them in the reactor for longer, so that you can put in more of these materials and make more product and reusing these materials. But with that, the main issue is that the cap analogs, which are actually the main cost driver, are not reused in that sense. Are actually not reused, and they're just used inefficiently, right?

18:29Because they are not consumed. Basically, maybe I should say that for each RNA molecule, which could be many thousands of nucleotides, you would only add one cap molecule at the beginning and nothing else in the same molecule. So the ratio of consumption between cap and NTP is like a thousandfold difference. So basically meaning that these are actually not consumed or virtually not consumed in the process, right? Yeah, barely consumed. So we thought, is there a better way of running this process to avoid this inefficient use of raw materials, including capping reagents and also all the other components? And what we developed is a process in which we co-develop this synthesis process with the purification process in a way that they can seamlessly work together, right?

19:19And as I will explain now. So basically the process, at the beginning, you have all of these building blocks, these NTPs, the capping reagents, the enzyme, the template DNA. That's what you have at the start of the reaction. And by the end of the reaction, you would make the product by consuming those NTPs. So this shows the product with a poly(A) tail there and a cap. So that's the mRNA, right? And you still have the enzymes and template DNA, the capping reagents there. But you consume the NTPs to make that product, right? And what we did, we developed this process in a way that we can directly take this and load it directly onto an affinity column, that has oligo-dT ligands. So these oligo-dT ligands form hydrogen bonds to the poly(A) tail of the RNA, and basically this allows specifically binding the RNA and the other materials that were loaded flow through the process in an undiluted way. And as a consequence, the RNA can be eluted in a separate step. But these materials, importantly, these can be then reused or recycled to make more product. And you just have to top up the NTPs, right?

20:39And as you can see here, we can reuse the enzymes, the template DNA, and the capping reagents, which was not possible before. And just by the way this process is done, it also reduces dsRNA contaminants, which are… So dsRNA stands for double-stranded RNA, which is an undesired byproduct that everyone wants to get rid of. But with this process, we get rid of those for two reasons that I don't have the time to explain probably now. But basically, we can do this cycling several times, and we can recycle our raw materials to make several rounds of products, right? And the next slide just shows some results.

21:25So this was done across five cycles with two different buffer setups. As you can see, the yields are consistent across the five cycles, and we observed a 2.3-fold cost reduction with this recycling process compared to a highly optimized batch process. We also saw that the quality was maintained consistently high across the five cycles. We see the integrity measurements here. So this is a measurement of the RNA integrity or RNA intactness, which is essential. You want the RNA to be full length and intact for it to work, right? Otherwise, if it's degraded, it cannot encode the protein product that you would want to deliver.

22:13The capping efficiency was fairly high, although it dropped slightly. So now we are looking at improving that as well. And importantly, we saw a reduction in double-stranded RNA amounts, as I mentioned before, which is very important because this would then mean that our product is more pure and is less immunogenic, which is beneficial for these products, right? We also measure the poly(A) tail distribution, and that was as expected. The expected size was 49 in this case, and the distribution was similar to what we see for a standard process. So that's all good. Then we tested these products made by this recycling process in cultured cells, and they performed as expected, both in terms of protein expression and also in terms of immunogenicity or cytokine response, cytokine profiles.

23:08So just wrapping up this part. So we demonstrated that we can recycle materials to make more product, and this can reduce manufacturing costs. And in addition, it can also improve product quality, right? And now we are building this into the RNAbox platform that we're developing. So that's quite promising, and this will be quite important at large-scale manufacturing, because raw material costs at large scale are the main cost driver. And with this, we can reduce raw material costs. We already see a 2.3-fold cost reduction. Now we are working on further improving this. The second exciting innovation that I will briefly talk about today is this idea of a soft sensor, based on pH, that can enable real-time monitoring of the IVT reaction.

24:07So, a soft sensor stands for a software sensor, and this normally combines a physical sensor, like could be pH or UV or conductivity, with a computational model to take that measurement into model and then with the model calculate other CQAs or KPIs of the manufacturing process that were not able to be measured otherwise in real time. So just to give you a bit of a background again here. So why pH? Why we chose pH. We know that when you make RNA, when you synthesize RNA, so as shown here, this is the growing chain of the RNA molecule. So this is the heteropolymer of the different building blocks here.

25:01And when we add one of these nucleotide triphosphates, what happens is we release a pyrophosphate. So this was a triphosphate, and when it gets into the RNA, it becomes a monophosphate, so we release a pyrophosphate. But importantly, we also release a proton, and this proton release can cause a drop in pH, because as you know, pH is a measurement of the proton concentration. So proton is the hydrogen ion. And by measuring pH, we can determine the concentration of hydrogen ions there. So in a nutshell, the equation looks like this. You have the RNA. When you add one NTP, the RNA grows by one nucleotide or one nucleobase, but you release pyrophosphate and proton.

25:51So proton is the hydrogen ion, and this can lead to a pH drop. However, when you run this process, you have buffers there, and in addition, this pyrophosphate is then broken down into orthophosphate, which is, again, a buffer, or it can act as a buffer. So there's an interesting dynamic there that you create protons, but you also create buffering capacity. And this just looks at more detail at the molecular mechanism of how this proton is released. So this is the ribose sugar. This is the nucleobase. This is the RNA. So this is the three prime end of the RNA, or three prime hydroxyl group. And this is the hydrogen atom right now, which will be released as a proton from here when this bond is formed.

26:54So the way this works is that you have this nucleophilic attack of this oxygen atom to this phosphorus center. You create this pentacoordinate phosphorus center here with, as you see, five oxygens around it. And this gets stabilized by cleaving this bond. So you create then this phosphodiester bond, and you break this bond here to release the pyrophosphate, together with the magnesium. I should say that, yeah, this is a two-magnesium mechanism. I should have said that this is the catalytic core of the T7 RNA polymerase that you see here, and that's how it works. And there's lots of publications on this, and this is a well-known fact. But the proton release is kind of mostly overlooked, and not everyone looks at it.

27:47But basically, the point is that for every NTP that you incorporate, one proton is being released. And in theory, if you do this a lot, when you make RNA, for every molecule, you can incorporate thousands of NTPs. You release thousands of protons per RNA molecule, and you make many, many RNA molecules. So there's a substantial proton release there that could, in theory, cause a pH drop. And I guess the next question was, can a pH drop be measured, given that for each proton, we also release phosphates that have buffering capacity? So to test that, we just used a tiny pH probe, a fiber-optic pH probe in our IVT reactions.

28:37So as you can see, we measure pH over the course of the reaction with this pH probe in different buffers. So commonly used buffers in IVT are HEPES-based or Tris-based buffers. There's two main types. And we did this across different RNA constructs. So an RNA encoding for the enhanced green fluorescent protein, or EGFP for short, and separately for an RNA encoding for the COVID spike protein as well, which is a longer protein. So we looked at two different RNA transcripts across two different buffers, and we could always see a nice pH drop as shown by this blue line on this axis. The pH dropped from around 6.5 down to 6.2, normally 6.1, or sometimes even lower.

29:31And we could measure that with a small pH meter. And throughout this reaction, we also took samples, and we measured RNA yield. And we could see by, actually by orthogonal measurements, measured both by UV and fluorometry, that the RNA concentration was increasing as the pH was decreasing. So there's a nice correlation between pH drop and RNA yield increase. And the pH drop was obviously temperature corrected because temperature impacts our pH measurement, but that was fairly straightforward because the reaction was kept at 37 degrees. Next, then we asked the question, what can we do with this information? Yes, we can measure pH, pH drops. And what can we do with this? So in parallel, we have developed models of the process, as I mentioned before.

30:26And for the IVT, we developed a detailed model that captures the kinetics of the reaction, looks at all the different species in the reaction. We can calculate RNA concentration, which is RNA yield, consumption of these NTPs, these building blocks, generation of all of these phosphates, pyrophosphates, and all the different species there. And importantly, we also account for the proton mass balance in all sorts of forms and complexes. So in a nutshell, we have this complex model that captures the kinetic behavior of the IVT, looking at these dynamic equations that look at concentration changes over time, and we also capture all the equilibria or different equilibrium equations in this model.

31:22And that's quite a detailed model. And in parallel to that, we also developed a much simpler model, which is a model of the buffer based on the Henderson-Hasselbalch equation, which basically relates pH to buffer pKa, as you can see here, which is a much simpler model. It doesn't need kinetic rate constants, it doesn't need calibration for each template, and it's just broadly applicable for every buffer and across templates. But this model can only account for 15 species in the reaction, and these are mostly buffer species. Whereas the more detailed, full-on kinetic model can account for 40, so that's four, zero, different species in the reaction. Then you might ask, okay, we can measure pH, we have this model, what new insights can we get? So what we then do, we measure pH, and we feed this pH measurement into the model into here.

32:22And then with the model, we calculate all of these different species. So there's a lot of species that we can calculate. As I said, 40 species for the full model and 15 species for the simplified model. And as you can see here, we can predict RNA yield in real-time for the common-use buffers and two different RNA transcripts. Again, that's EGFP and COVID spike protein or CSP, with fairly high R² values of around 0.9 or even higher for the kinetic model and around 0.9, again, for the Henderson-Hasselbalch simplified model. So that looks quite promising. That's already quite promising. This was not possible before. So basically now we can measure pH, feed that information into the model, and the model calculates RNA yield in real-time. So this serves as a basically real-time way of quantifying RNA yield with the model, which was not possible before.

33:34And the dynamic model can actually even forecast ahead in time. So let's say in a 5- to 10-minute time window ahead in time, we can forecast RNA yields, meaning that we can have future measurement values of RNA yield, which is very exciting. And besides RNA yield, we can do the same for NTP consumption. Based on this pH reading, we can calculate the consumption of the NTPs, and we can calculate NTP concentration throughout the reaction, and we can even forecast it with this kinetic model. And again, at fairly high R² values of above 0.8. And I guess I should say that these R² values are… 0.8 is acceptable in my view, but this would be even higher if the measurement noise was reduced.

34:22And the measurement noise here comes from the offline assays that we use to validate. So these are like HPLC assays that, as you can see, there's fluctuations in some of these measurements. And if these were corrected, the R² would be even higher. So yeah, the R² is lower because the assays used for validation were a bit noisy. And these are the HPLC and offline assays, not the pH measurement. So far, for the yield and NTP consumption, we could measure things offline later on and compare the model prediction with the offline measurement, and that's how we validated these models. But there's also other species that we can measure and predict dynamically, or we can get future measurements for these, but we cannot validate them. So these include things like the pyrophosphate, the orthophosphate, the monophosphate, and all the various complexes of the magnesiums and NTPs.

35:26So magnesium and the four different NTPs in this case, and all sorts of other complexes and buffering species. But for these, we don't have ways of measurement and validation, so all we can rely on is these mass balances and these model predictions. And these are, again, just some more of these buffering species that we measure. So the concentration of the buffers, the conjugate bases for HEPES, and so on, acetates, and we can do it for Tris as well, and phosphates, and so on. And this just gives a list of all different IVT components or species that we can now quantify via this model. And as you can see, we can only validate for a subset of these, including the NTPs and the RNA yield.

36:25So basically, this gives us lots of insights into the RNA process, especially for this upstream part of the process where we create RNA, where we synthesize RNA. We can quantify 40 different species. At every 25 milliseconds, we can quantify them, and we can do this based on pH measurements, and we can use the pH measurements to actually course correct the models as well and improve the model predictions. So for course corrections, we use this unscented Kalman filter. So that's a filter that looks at the noise in the pH measurement and the uncertainty in the model prediction, and it averages them out so that we have a reasonable prediction of these concentration estimates for the 40 different species.

37:23And for the Henderson-Hasselbalch, again, we can only quantify 15 species, but that's again, a lot more powerful than what's currently available. And out of these species, the RNA yield and NTP consumption across the four individual NTPs is validated against offline assays. We published a paper on that, so fairly recently in May, it was published in May this year, and you can read more about it here. So this slide just sums up, gives a summary of how this works. So again, in the RNA production process, in the upstream part, you have an IVT reactor where the RNA is being synthesized. And then you can use a pH meter there, both in batch and continuous to measure pH. In a continuous process, you would need more than one pH meter because you would want to measure at different points in the reactor, right?

38:15But the point is that you see a pH drop or a pH difference, and you take that pH difference, you feed it into two different models. That could be either a simplified model of the buffer based on the Henderson-Hasselbalch correlation, or it could be a more detailed kinetic model that can forecast ahead in time. And with these models then, and based on pH reading, we can quantify RNA yields, calculate NTP consumption, monitor the reaction rate, and look at monitoring 15 species with a Henderson-Hasselbalch model or up to 40 species with this more detailed kinetic model. So that's what a soft sensor is. It takes a reading from a physical sensor, which gets processed into a model and gives us outputs that were otherwise not available.

39:10Right? So just in conclusion, we are building this RNAbox platform, as I presented, which is a continuous process in a box. It does end-to-end production of RNA from IVT, which is the upstream part. It does purification and LNP encapsulation, and we are building in two exciting innovations into this. One based on raw material recycling, which helps us reduce costs. So we observed a 2.3-fold cost reduction already, and we are further working on reducing the cost even more. And we also saw that the double-stranded RNA or dsRNA impurities were reduced threefold. The other innovation that we are building into this RNAbox now is this soft sensor, which can be used based on pH measurements to monitor the reaction at the upstream process quite extensively, with high R² values, so with high accuracy.

40:20And this provides unprecedented insights into this process. We can obtain 1,600 concentration estimates per second with this. Right? And my final slide is just the acknowledgment. You can see our research team members here, the past and current members of the team. Our research team is called the RNA Manufacturing Innovation Team or RNA MINT for short. This is our website for the research team. You can read more about our work there. But again, as I said, a lot of it is focusing on developing these next-gen processes for making RNA-based vaccines and therapeutics and doing process modeling work as well and combining all of these things, right?

41:06I'm also involved in two companies. RNA Forge is a company that provides RNA analysis and manufacturing services, and we're also providing bioprocess modeling consultancy via a separate company called Simulenta. There's websites if anyone's interested in getting bioprocess modeling consultancy from us in terms of kinetic modeling, techno-economic analysis, or things like computational fluid dynamics, CFD, or multi-physics modeling, please get in touch. And these are our funders here. And with that, I thank you for your attention, and I thank again the organizers Invert for the invitation and opportunity to speak for you. Thank you.