Go with the Flow Expert Panel Discussion | Hosted by Tony Hunt CEO - Repligen Corporation (06/29/21)

Yeah, thanks Tony for that question. When we were looking at flow VP, the challenges we were really trying to solve were just the extent of how well we understand the process that we have day-to-day in our manufacturing facility. Traditionally, we have a lot of offline sampling—we take a sample from this part of the process, we take a sample from that part of the process—so we're just capturing a very small percentage of what is really going on. Having the ability to have online real-time monitoring is really the challenge we wanted to solve. For us specifically, where we looked at flow VP was that we were running a clinical facility, and very often it's the first time that product actually comes to a GMP facility or to that specific scale of manufacturing. What you're looking for, besides manufacturing your product, is to understand how the product is behaving through the different process unit operations and collect as much understanding on the process and the product as possible, so you can optimize your process and get ready for commercial manufacturing. That was really the challenge we were trying to solve: how do we get all the data we're looking for in the best way? For us, the answer was to go into PAT, and specifically for protein concentration, we ended up going for flow VP.

Thanks, Tony. I'll be brief because I know we have a lot of questions to talk about, and people can ask if they want me to follow up. Simply, it was the high-concentration UFD processes that we started working on in development about four years ago, in 2017 at Bristol Myers. They always require a concentration range during the process that is very wide—between about 10 and up to over 200 grams per liter. There's not really any other instruments available that can measure concentration in real time. Most people in industry do it by mass, which gives large errors if you're using mass and extrapolating based on the starting mass and concentration. It's much better to measure the concentration directly. The flow VP was really the only instrument available that could measure that large dynamic range at the time, and I think it still is. It's relatively simple to set up too, unlike Raman or SFD, which require modeling. This is a pretty fast system to set up.

Yeah, I feel there's no reason why we cannot apply it to different molecules. We have used it successfully for different protein molecules, for mAbs, bispecifics, fusion proteins. I don't see any reason why we cannot apply these to other technologies as well. Even though the molecules are different, the basic unit operations we are using for purification—since I'm a downstream person, I'm talking about purification—we are still using chromatography, we are still using UFD. Everywhere where we are going to use a UV spectrophotometer, that can be easily applied. Talking about viral vectors for gene therapy and similar stuff, we probably still use affinity chromatography for the capture. We probably cannot use that for quantifying the vectors, but we can definitely try to have an application for impurity measurement, concentration measurement. I definitely see that we can extend the applications of this beyond what we are currently doing.

Yes, as Rashmi said, there is a lot of potential with the technology. Not to repeat what she already said, I think another advantage is the fact that it's quite applicable to connect with other sensors as well. That can be quite interesting, especially with A280, which can be challenging to detect. Maybe you can couple the signal from flow VP also with Raman or with other spectroscopic methods. I think there is a lot of promise there.

I don't know if I'm the most adequate person for that question. Maybe someone in manufacturing can better answer that. But the impression I get is that if it's worked so far, from a project point of view, if you have to get a new technology into your project development and you have your timelines and your risks, then you are taking more risk than you want to, than you have to. If it's worked so far, then why do you have to do it? So maybe the question is about the motivation, from a project management point of view.

Maybe I can add to that, Tony. From a pure manufacturing perspective, especially in a drug product fill-finish, people are always looking at removing things from the process to maintain sterility. If you're going to start introducing additional pieces, that puts the thrill in somebody from a QA group. I think that's for sure a challenge we are generally facing. On top of that, with PAT, we are in between two traditional functions. You have the QC group that has all the knowhow and expertise on analytical testing, and you have the operational group that has all the knowhow and expertise on the manufacturing shop floor. You're bringing those two worlds together, and there is a lot of ambiguity on who's going to combine that skill set into a single person or into a single group. Companies are now starting to tackle how to train operators to have additional skills they didn't have before, or how to give QC staff more operational expertise. That probably adds to why we're seeing a slow implementation of these new technologies.

Yeah, I guess you call it advanced, or I call it enhanced, which is kind of the same thing. I think it speaks to the fact that these are pretty complex molecules—biologics—especially if you are measuring CQAs. Compared to a small molecule, where the API has a molecular weight under say 500 and is almost completely pure, a biological molecule like a monoclonal antibody is extremely heterogeneous, with hundreds of thousands of chemical bonds and atoms. That requires much more sophisticated analytical instrumentation to measure the CQAs. Protein concentration is a great place to start with the flow VPX. What we're trying to do in our enhanced PAT group is take some of these tools that were first developed for small molecules and try to utilize technologies that we could theoretically use in a GMP facility in the relatively short term, like a five-year time span. Technologies that are not mature—you're not going to take something completely brand new and try to implement it in PAT at this point. We're going to try to take the most rugged, proven technologies, such as this VP, which has been around for over a decade, and put that into the GMP world. That's sort of the enhanced PAT we've been working on: using proven, reliable, robust, accurate, precise technologies for our enhanced PAT, and then go from the most simple to more complicated, like mass spectrometry, which can probably come at the very end because that's the most complex of all analytical tools.

Sure, thanks for the question. First of all, for batch, there's definitely an advantage for this technology because it can reduce the lead time for each one of your batches. As Raf talked about in his last talk, at some point we can have automated dilution process control based on real-time measurements, where you are controlling the concentration during UFD and subsequent dilution to the drug substance. Right now, we are at the stage where we want to do process monitoring with the VPX. For batch processing, this is going to replace the offline measurements by doing inline measurements, so that's slowly going to reduce lead time. Once we get to the point where we can measure real-time inline and do real-time in-process release, then I think true continuous could actually be something that's realistic and possible, at least for the UFD part. True continuous is going to be difficult, linking from upstream all the way through downstream. But the way our process development group has gone about the problem is to do elements of continuous. They were looking at a continuous protein A capture, continuous polishing steps. I think we're on the cusp that maybe in five years we could look at a continuous UFD process, and I don't think it would be possible without inline A280 monitoring.

Yeah, that's a very good question. Over the years, there has been so much improvement in upstream processes, and continuously the input to the downstream processes is changing. When we are talking about operating protein A on a large scale in continuous cycles, it's very necessary that we understand that some type of changes can take place in the process. There is a variability which can occur maybe in the upstream process, or there may be errors in the titer determination, so we cannot stick to a fixed value of binding capacity, for example. We are continuously loading to a fixed value, but if we can implement this flow VPE downstream of our columns, we can actually control the start and stop of our loading based on what we are getting. This can also provide us a warning and indication that our resin is degrading, that it has completed its lifetime. It can definitely be applied for enhancing the capture process to mitigate the variability which can occur upstream of the actual capture step. For continuous chromatography, I definitely see that happening, since as the capacity decreases with time, we can capture it real-time during the manufacturing process.

I think something that could benefit in general from the concentration measurement would be, in some cases for example Raman, that you have an increase in background or resonance because of the increased protein concentration. Then maybe there's a possibility to scale down the signal that you get by taking into account the flow VP information. I think that would be the application that comes foremost to mind. Maybe you could also use your concentration measurement if you have a varying sample going into an HPLC, so that you can determine how much volume you want to inject or how you want to treat your sample before doing an HPLC on that.

It's a mixture of both. On the one hand, we're looking for applications where we take an existing tool and try to see what that tool, in addition to what it has been made for, can be used for. For instance, flow VP is predominantly or almost exclusively used at 280 nanometers, but it comes with the option to go from 190 up to 1,100 nanometers, so you have access to that whole UV-Vis range that contains a lot of additional information. That's for sure an area where we are looking as well. In addition to that, it's the combination of multiple PAT tools where the standard multivariate analysis—PAT tool number one combined with PAT tool number two—individually they might not give you something you're looking for, but the combination, if both of them are trending in a certain direction, could give you additional feedback on what is happening in the process. For sure, we're looking at both of those routes.

I'm smiling because it's a typical question you always get: what's the business case? As a side note, I always think that if we keep on using the same processes that we use for a business, then we will never invest in innovation; we will always keep on investing in safe bets. But it is a reality that the business case question always comes back. I think it's a mix. On the one hand, you do have tangible benefit. You can calculate the return on investment for flow VPE. If I'm going to implement flow VPE, which costs that amount in US dollars, if I don't pull the sample anymore and I get a certain lead time reduction, or I just have no person in the QC lab coming in at night anymore, that's obviously the benefit I get, and you can really quantify a certain business value out of that. Typically, that comes out positive, so that's at least enough to convince the base level of people that would need to sign off on that. But what we have been really trying to do is convince people of the added, less tangible benefits. If we have this additional data that could be used to better understand the process, that's going to lead to a better manufacturing process at launch of that product. It's very difficult to quantify that value, but it is something we try to emphasize a lot. Can we reduce our tech transfer timelines or scale-up efforts based on the tool? If we have a tool collecting data, we can remove a lot of the offline sampling during a PPQ approach. It's a mixture of both, but I do agree that there is still the reality of the business that you need to provide that business case value.

Yeah, I can speak from my experience. We've had companies where Ramsey came in a couple years ago with the flow VP to collect data, and that data was crucial in deciding whether to go with flow VP. Similarly, we've had companies that were unable to provide a demo instrument for us to test, and that led to us not purchasing the technology because we don't really want to purchase something of which we have no data. That data piece is absolutely crucial in justifying the value it brings to the business.

Not yet, but I would just add: if anyone has any questions on the line, please type them in, and I'll relay the message, and we can have a discussion about whatever topics you'd like to hear from the panel.

It's absolutely true. People get much more skeptical when you say 'I have this tool, I just want to monitor what happens in the process and I'll leave you to it' versus 'I'm going to use this data to actually drive GMP decisions.' It's the same tool, the same GMP process you're implementing it into, but the applications are completely different. That comes with a bunch of additional requirements. From a purely analytical perspective, you need to be 100% sure that the data you're capturing is correct. If you were monitoring simply and an operator had a wrongful assembly of the sensor or a calibration that's out of date, that's okay because you might just be interested in looking at a certain trend. But if you're really using that in a GMP process, a lot of those actions become very important. The analytical method needs to be spot on, fully validated. Your data integration, the signal coming from the flow VP, has to go into a data historian for verification, so that connection needs to be spotless as well. Then you have the whole quality piece, the rules and responsibilities around that to make sure your instrument is maintained to the right level, with regular checkups by the supplier or regular calibration, and all of that documentation. That's a lot of things that need to come into place from within the company implementing the tool. On top of that, you have external acceptance. You're using this tool now to make GMP decisions, so you need to not only convince your own operators that it's reliable, but you need to convince health authorities as well. To do that, you need to have a good data package showing them that this is equivalent to or better than the current control strategy you're applying for your process. That could be a comparability study or a comparability protocol, but there is indeed a huge amount of data that needs to be collected to fully convince the health authorities that this is equivalent or better.

I had a question from the line if I can break in for a second. I have a couple actually. Maybe this is open to everyone: what is the focus of your PAT strategy? Is it to replace IPC endpoint release testing, add knowledge about your process, or something else? I think we kind of touched on that, but maybe as part of the justification, when you first bring in the technology to get that data, what was the original point or the main focus? I guess we could start with Rashmi there.

As I mentioned before, the dynamic binding capacity of the resin is important, and if it's changing over time, application of PAT is definitely going to be useful there. To make sure that we are not underloading or overloading our resin, we are getting consistent performance. Maybe we have an output of the flow VP or any of the PAT instruments attached to some algorithm that is doing the control for you. In your case, the main strategy here was to get better process understanding. It is also to handle the variability which can inherently happen in any of the process, so that we have a control strategy that, irrespective of whatever variability, we are still going to have consistent performance.

For us specifically, our initial challenge that we wanted to solve was that increased process understanding. But from the moment that you have that tool, you're seeing a lot of data that you never saw before, and then it goes up to the point that you can start using it for other applications. Now we're using that for an IPC, and in the future, as I talked about last week, for actively controlling the process, and who knows in the future potentially for release. It is something that is not static. You might implement flow VP for one particular purpose, and then you potentially see all of the applications you could use it for in the future, and it kind of takes off from there.

There's another question here. Maybe Jay, this would be a good one for you. In the GMP environment, who owns the data and the equipment—the analysis, QC, or the process? I guess the question is really who—and it may make sense to give the BMS perspective because it's a big company—who owns the equipment or who's the owner of the analysis in the case of an inline system?

I can chime in and give the Jansen perspective on that. It is indeed kind of in between two worlds. Ultimately, the process and the whole equipment, since it's embedded into manufacturing, the owner of the equipment should be manufacturing. All of the qualification exercises or actions that revolve around that should be owned by the people that normally own the other equipment on the manufacturing side, whether that's the engineering group as well. If it's about data, we look at what the application is. If the data is used for IPC purposes, in our case that data is owned by the QC group. That also means that the data coming from that PAT tool is owned by the QC group, which will also be responsible for the review and the actual sign-off on that data. If it's only for monitoring purposes, then it's generally the development group that would own that data. For IPC, it would be QC. That's how we have settled that duality: it's an analytical tool but on the operational shop floor. From a hardware perspective, it's really owned by operations, but the data coming off of it would be owned by QC.

Can you hear me now? Yeah. We haven't gotten as far as what Raf is in terms of in-process control, so I can't give you that. I know the equipment is owned by manufacturing, and in-process testing is simply owned by manufacturing. For in-process controls, that's something I can't answer definitively how it's going to be owned by quality or not. I guess quality is typically thought of for the offline measurements. Maybe Raf is right that we need to have quality involvement if it's going to be in-process control. I just want to go back to a couple of other questions that were asked before while I was muted. There was a question about how we choose our PAT. To be honest, we started actively working on PAT in biologics probably only about four or five years ago at Bristol-Myers Squibb. We've really gone after the low-hanging fruit because there is some low-hanging fruit. The process has undergone a lot of evolutionary changes in the past five years: the titers are much higher, we introduce elements of continuous, and really the analytics haven't kept up. As Tony was talking about continuous manufacturing, we really need PAT tools to make the continuous capabilities fully capable, because there are still some quite long steps during the continuous—we're waiting for a QC test to come back to start the process. Those are the ones we targeted first: to decrease lead time for those semi-continuous processes, we can eliminate a quality offline method, and also improve process consistency. The titers are getting much higher, and the UFD final concentrations or drug substance are getting much higher, which has introduced its own issues with process consistency. As we've learned, the protein concentration with a monoclonal antibody itself can affect excipient concentrations—it's something called the Donnan effect. Even a small difference in protein concentration, say 210 versus 200, can significantly affect the concentrations of your excipients during processing, so your end product is going to have significantly more variability. All of a sudden, protein concentration becomes far more critical in terms of maintaining the consistency of that and its importance on overall process consistency and quality.

Yeah, I mean, over the years I've definitely heard companies talk about over-concentrated or under-concentrated. Raf, when you're talking about the final formulation, it's awful when you know it's the wrong concentration and you have to go back and figure out how you're going to reprocess a batch. Getting to control using flow VPX as a control for the UFD step—how do you see that evolving at Bristol?

That's kind of where we are right now. I'm trying to sell the value proposition to our process team of process control for UFD. I think we have a pretty strong argument. The process consistency if you are monitoring by mass, you're talking about plus or minus 10%. As Raf has shown and we've demonstrated ourselves, you can decrease that probably to plus or minus 3% by using inline A280. That's a significant gain, not just in terms of protein concentration, but also the critical quality attributes of your excipient concentrations. I also wanted to go back quickly to which instruments you'd like to combine PAT with the flow VPX. One suggestion was online LC, which I think Andre may have mentioned. For a continuous process, say you're measuring a CQA like aggregation or your charge species, in order to have an accurate mass balance of your entire product pool, you need to have an accurate concentration. An LC system typically can do a separation in two to five minutes to determine your CQAs, but your protein concentration is continuously changing. You take snapshots of windows, and you can basically come up with a model that can give you an aggregate of what the CQAs are based on the mass and their values. That's something we're working on, so that you can have a continuous system where you are continuously monitoring your CQAs and what they would be if you start or stop the collection at a certain point. An accurate mass balance is critical in terms of determining the CQA for a given continuous batch.

I'll go real quick. I think it's absolutely critical as part of the value proposition that we sell this. That's actually something I want to investigate when we're purchasing new instruments from the vendor: make sure that the quality is very high. Once we establish that, we can have evidence that we can quickly transfer between sites—much more quickly. A given process that relies on an A280 meter, like a complex high-concentration process, would be much simpler to transfer between sites and locations. I believe it will be.

We haven't really used the flow VPA in the GMP environment; we have been using it in the PD labs. But as Jay said, if we have data to prove it, we can definitely transfer it. I can also think of a couple of simple applications. For example, we had a process where the flush for the viral filtration was dependent upon the concentration—we reach a particular concentration, we stop the flush. Things like that, I think I can see that easily being transferred if we have data that it is working well in the PD labs.

Sure. One aspect that we think is quite promising is to follow the efficiency of a column. We know with a digital twin how it behaves during time, and you can see when we're losing some binding capacity and maybe it is time to either change the process parameters or take a look if we need a new resin. In that regard, the digital twin can be quite valuable for the process.

I fully agree. The concept of digital twins is getting more and more routinely used. An important prerequisite for that is you need to have the data in order to build your digital twin. Online PAT tools such as flow VP are exactly there to provide that data so that it can flow into the models underlying the digital twin. I don't think you can really establish digital twins by having offline sampling approaches and offline QC lab experiments. It's a very important part of establishing digital twins.

For tech transfer, one important thing is that if you really move from one site to another, sometimes there is a lot of need for engineering runs or development work at scale to be done in order to fully understand how this product is behaving at the specific site or scale you want to manufacture it at. One approach would be if we all had the same equipment, where we would transfer from one site to another and it would be a 600-liter tank from the same brand at the CMO, so we wouldn't need any development work at all. But in the absence of that, online tools will really give us that understanding in real time. Otherwise, we would have to do an engineering run to get data. It might be data we can get directly as part of PPQ during a tech transfer. A very concrete example: mixing studies. If you go from a 5-liter tank to a 50-liter tank, you would have to characterize how your mixing is going. But having your online protein concentration or any other PAT tool, you can track how your mixing is going. It might not take five minutes like it would on the 5-liter tank, but at least you can monitor and say maybe after eight minutes my mixing is completed. It's this kind of information that otherwise would require a lot of development work that you can get in real time without any need for at-scale tests.

Actually, there is. I'm glad you reminded me. It says: 'Has anyone presented data on the flow VPX to the authorities, and if so, what were the learnings?' I guess that means the health authorities like FDA or the EU.

Silence could be 'no.' That's rough. So we haven't shown any data as of yet, but we're in the process of doing so. We haven't done that so far.

That's all I have from the line right now.

My personal experience about using this technology is that it's a very powerful technology, but it took me some time to understand its full potential. If I had to advise, I would say understand the instrument, understand the technology very well. Once we do that, we can just extend the application to no end.

I would say, along the lines of what Rashmi is saying, be very careful when you set up your methods and data collection so that the data is useful. It looks like a very simple tool from the start, and you can just plug it in and start collecting data. That's actually one of the biggest issues we had early on in the days of PAT: people would just get a PAT tool and plug it in and not go through proper analytical methodical system suitability testing. Every week, every day, just checking the instrument is working perfectly and you're using the most accurate method for that molecule. Otherwise, you get data that doesn't look so good, or the data hasn't been time-stamped correctly, and it's really difficult to go back and make anything meaningful out of it. Be very careful, especially if you're going to use it in a pilot or anything, to really carefully set the thing up. Don't generate bad data in the start, which can cause more headaches than anything else.

Maybe to add to what has already been said, I would suggest to think about how you connect and collect your data. Whether you are using the electrical connection or the OPC server, think beforehand how you can get the data, because the idea is to get it as fast as possible. Once you have your instrument, a lot of the times you have to wait until you develop something that enables you to get the data, and it can be a bit frustrating because you can see that you're getting your data but there is a delay before you can really use it to its full potential. Invest beforehand a little bit in thinking 'this is how I can access my data in real time and how can I unlock this capacity.'

I agree with what everyone else said. Maybe one thing I can add is that if somebody is about to implement flow VP or any PAT tool, tying that back to what we said in the beginning of this discussion, it needs to be done by a dedicated team. There will be challenges to overcome, and there will be a lot of pushback from regulatory quality—people that in general are afraid of implementing new things into a process of which they know that it runs very well, and why should we change that? Having a dedicated team with people that have that mindset that they want to change something to even better is very crucial. Looking back at the last couple of years that we have been implementing flow VPE, if you would have had a team that would do this maybe on the side of their day-to-day job, I think it would have taken much longer, or maybe it wouldn't have worked at all. We really had people that were dedicating in a dedicated way working on implementing this, and I think that's one of the success keys to having reached that implementation so far.

You're absolutely right, Tony. There are two things that come to mind. One was what I presented last week: filter absorption. To get a data set like that would be nearly impossible with traditional sampling approaches, or you would have to do a very elaborate dedicated study and use a lot of material. Having that refresh rate of data at 10 seconds—with some optimization you can even bring it down to five seconds—that's a lot of data that was never there, a lot of insight into the process. Before we actually plugged it into our process, we weren't even thinking that this is the data we were going to see. That was a very important learning, and that learning also translated into us moving from monitoring to control. Before monitoring, we were saying 'let's put this online protein concentration tool in there and we can replace the offline sampling, so we might shave off two hours waiting time for the QC lab analysis.' But then having all the data, we thought 'why don't we use it in real time to control the process?' That thinking only came when we actually saw the data implemented into our process. We didn't really think about that before we had it available.

As Raf said, with this fast sensor information, you open the door to monitoring so you can better develop your process, and for control as well. You can actually control your process based on real information and take into account the variability that's happening in your process, whether you want it or not. For our specific case, the combination with MALS was before the flow VP. We were able to do it at the analytical concentration, but if we hadn't had access to this fast concentration information, this application would have remained in the analytical realm and we wouldn't have been able to use it for a process stream.

I wanted to mention that they had some excellent points. It opens up a whole other world when you get high-quality data in seconds. The other side of the coin is what I call the fear of data. I don't know how you—maybe Raf is smiling—I've heard this. It almost sounds counterintuitive, but it's actually not. Based on reality, a lot of people are afraid of this data. They think 'oh my God, it's going to show that our process is inconsistent.' I think people are a little afraid of that. I'm like 'well, the FDA is going to want this. You can't hide anymore. You can't just say see no evil.' That's the argument I use: we can't hide anymore. We have to show them. As long as these tools produce really high-quality data, hopefully the worst fear—people think that every time there's a bad result we have to investigate—is not the case. The other side is overcoming this fear that we're going to expose ourselves if we collect all this data and see that there's actually a lot more noise than we thought. Just being okay with that, because that's reality. It's like making sausage.

Before we implemented flow VP for DBC determination, we used to get a breakthrough curve with like 105 points. Once we implemented flow VP, we realized that we are getting nice and smooth breakthrough curves. That opened up the potential application for studying resin lot-to-lot variability as well as resin lifetime studies. We have used it for resin lot-to-lot, where we compare a resin from the GMP and from the process development labs, and we just overlay the DBCs. Even minor differences can be caught. We definitely are going to explore this for resin lifetime studies, where we can see that the difference in the DBC curves—the breakthrough curves—is a very important tool to tell us how good our resin is performing.

No one wants to run resin lifetime studies; it just takes too long. I can see it being used as well to put in the cleaning cycles on chromatography resins and determine what the right time frame is to do that—is it every five cycles, every 10 cycles? You can probably use this technology to help determine what the right timing is.

No, I think we can wrap up.

Sure. Final comments? No, just thanks a lot. Thanks everyone for all the hard work you've done over the years on this product. I know it's not always easy, and there are challenges. We're here to help too. Anyone on the line who's going down this road, reach out, and we're happy to help get you through any tough times with the product.

Thank you. Thank you. Take care, everyone. Yeah, thanks guys. Thanks for the invitation. Yeah, no problem.