Episode 1: Transcript

Codina, Anna
Welcome, Marina, to this series of conversations called “PAT Chats with Marina Kirkitadze”. We are delighted to have you with us and to benefit from the wealth of experience you have gained over the years. I am sure your insights will make a fantastic contribution to the discussions we are going to have. In our first episode, we will talk about PAT with a particular focus on its applications: where PAT works well, where it struggles, and why it matters. So, Marina, let's start with the fundamentals. What does PAT really mean? What is PAT, and perhaps just as importantly, what is not PAT?

Marina Kirkitadze
Sure. First of all, thank you very much, Anna. It's a pleasure to be here today and have a conversation about PAT, which is a very dear topic to me because I worked in that field, and I see its evolution. I'm coming from a biophysics background, since at Tbilisi State University that was my first step in education, and during my career I always wanted to see biophysical tools converted into tools for monitoring and controlling manufacturing, but also to see them in quality control. So, PAT is Process Analytical Technology, and it is a kind of wonderful bridge that we all drive towards, to see as you produce and release at the same time as you manufacture the material. PAT spans the full pharma value chain in R&D, development of the process, manufacturing, and it also aims for QC. I guess there's still a lot of work to do, but I think we see it best for now in many places, at least in biologics, since R&D and process development, piloting the process space. Of course, in spaces like small molecules it went further and landed in the manufacturing space for blending, drying, coating, and crystallization, and that's where PAT's success story is. For biologics, it shines in upstream manufacturing. It also supports downstream operations, but it's more challenging at present. Nevertheless, efforts are being made, especially with online probes, to see it happen. And yes, of course, it doesn't work everywhere. So, when does PAT not work? When the signal is not strong enough, the signal-to-noise ratio is not appropriate, then the PAT method is not reliable, meaning that damage happens over CIP or SIP cycles (cleaning in place or steam in place). So those things affect it as well. For example, sometimes coating happens and the optical window is closed, so we don't get the signal we desire. This shows that some technology has to be further improved, or conditions adjusted so that it can perform to its full power. And of course, the ultimate topic, as mentioned, is maturity.  The organization must be ready for real-time, knowledge-driven manufacturing. The readiness of the team is also important.

Codina, Anna
Yes, absolutely. Marina, you commented on the importance of real-time measurements: we're measuring while the process evolves, instead of waiting until the end. Would this differentiate PAT from QC end testing? Would that be a fair statement?

Marina Kirkitadze
Yes, it would be. With PAT, we want to see real-time measurements and we basically want to shift away from end-product testing. With end-product testing, whether it's an intermediate, a final product, or a filled product, you still have an end result. At that point, it's very hard to change something and sometimes regulators do not permit it. So, when PAT is in place and real-time in-process monitoring happens, it transforms the development, and it also allows us to set ranges for automated adjustment. Basically, the whole purpose of it is to have this closed loop, or feedback loop, to adjust automatically. For example, if you have an infrared probe inserted that reports the characteristics of your product, let’s say the concentration of a substrate, then you can see a gradient (the intermediate), and then the final values for the product at this stage of manufacturing. If at one point of the process the concentration is low or is decreasing, it can be adjusted to the normal level at that time, as the process goes, no need to wait until the end. Then this becomes more proactive. First, the teams can learn from the process. They can detect variability or deviations faster, and find the root cause of the event, which accelerates investigations for them as well. So definitely, PAT means a shift from end-product testing (what we traditionally see in QC or an analytical lab at the manufacturing stage) to real-time monitoring and control. PAT is more like a movie, as opposed to snapshots in time (photographs), a paradigm shift from off-line to real-time insights. Overall, PAT improves quality because you know more, you have huge data sets, you can see the whole process holistically. This leads to more robust, data-driven process improvements, and shortens the timeline for development as well.

Codina, Anna
I love the movie comparison! Marina, when you mentioned feedback loop and adjusting as we go along, are you referring to commercial manufacturing or pre-manufacturing? I'm wondering whether this is actually allowed to make adjustments while the commercial manufacturing process is ongoing.

Marina Kirkitadze
Definitely, definitely. It's a perfect question because it opens up situations that I think best apply to research and development, in particular to the development of the product.  During this stage, you can adjust and polish the process, so to speak. If you validate within a range for the process specifications (the critical process parameters), then yes, this feedback loop would be allowed to adjust. Of course, it takes working with regulators to demonstrate that this range is robust. It's indeed to ensure the process stays within the range. And we all know that sometimes it drifts. So, what to do when drifting happens, how to mitigate that? That will be the question. But for well-known processes—and of course we see it more for small molecules than for biologics—but even for biologics at the upstream fermentation stage, or when we use bioreactors, these things can be achieved because the substrate concentration has to be at a certain level. For example, you may see metabolites surfacing. It's not predictable, but it's measurable at what concentrations these intermediates will remain. So these things can be added into the validation of the process. And the continuous data helps to define what is repeated from one run to another, and what happens occasionally and should not be used in the validation either. So, with that, I would say R&D is a perfect space for it. For manufacturing, once you know the process and it's really robust, then yes, we can validate within the range. And then this feedback loop will serve to adjust to maintain that range. For example, temperature and pH in fermentation are monitored and adjusted, that's already happening to keep it in a certain range. You heat it up so it's within the range of 36.9 to 37.2 degrees. That range is carried through the entire round of fermentation. So, so yes. What you know well and where the process is robust, that definitely helps.

Codina, Anna
And the beauty of the power of being able to adjust, so then you can correct and mainly avoid batch failure. That is the ultimate sustainable goal, right?

Marina Kirkitadze
It can correct, yes, that's right.

Codina, Anna
This drives us to the next question: what's working versus what's not working currently? What potential hiccups can we have when trying to adopt or implement PAT?

Marina Kirkitadze
I think what's working is the maturity of the technology. So the sensors have to be mature, the technology has to be mature. And also, on the other hand, the process has to be well understood and well known. So, for example, if you use UV-visible for concentration tracking and impurities, it works really well. You can measure concentration related to UV-Vis. With the same respect, you can do it with infrared, but UV-Vis is often sufficient to proceed. pH, dissolved oxygen, conductivity—it’s present in upstream bioprocessing, but not only. For example, if you happen to produce in-house adjuvant, such as aluminum-based adjuvants, then pH and conductivity are used in the tanks of the raw materials for mixing and synthesizing the aluminum adjuvant in another tank. So with that, you have real-time measurement continuously, and it gives you the full picture of what happens during the mixing. Then, for example, mass flow, pressure, and temperature—again, for continuous manufacturing—are often used, and they are reliable because these processes are well known. And IR, we also see it, by the way. In some companies, they use it for upstream fermentation and upstream bioprocessing, but they also use it for blending to demonstrate uniformity. And of course, for dry formulations, for drying, granulation, and for some liquid processing. So again, the key is that the tools are stable and well understood, and on the other hand, the process is also well known and well established. So, this works well. What else do we see? I already mentioned real-time monitoring for upstream fermentation. Raman is there, but not in many places, but it nevertheless finds its way into the upstream. Again, dissolved oxygen (DO), pH, and capacitance probes are used as well to see the live cells, right, versus non-live, and to check off-gas. Sometimes off-gas analytics are linked to ELISA results, and there are some efforts towards that to give more meaning to how to link them. Still, more work needs to be done to see that it is real and robust in terms of modeling. But again, the process is understood. Small-molecule reactions—chemistry is well defined—so spectroscopy works well, and also Raman and NIR for reaction endpoints, crystallization, and so on. Change of solvent, of course, changes the spectrum. You see the diminishing of one solvent and the appearance of the other. UV and IR, and Raman actually, can also use mid-infrared probes to check that. It's very visible, and for small molecules, it's easy to adopt. Easy, I mean, in terms of technical and scientific value. Now regarding what is not working so well, we can mention inline high field NMR that gives excellent data but is expensive and maintenance heavy. On the other hand, low-field NMR is definitely trying to find its way as a PAT tool, if not on the manufacturing floor, then maybe in the QC lab, where it can be connected and measured in parallel with the processes, especially processes like mixing or fermentation, the usual low-speed processes. We have recent work that compared near-infrared results with low-field NMR results, and Gabriella Gerzon is the first author on the paper . She is currently a graduate student at York University and has worked on some of it. We started together, so she is close to her finish line, writing her thesis at present. So that’s how the recent paper came out this year. Another example is inline mass spec: it’s sensitive but requires expert knowledge and support. Another issue is the models. We need to validate the model and then see whether the process and the points we collect match that model. And we all know that chemometric-heavy models often drift. So, what do you do if the model drifts? How do you incorporate that into the regulatory landscape? How do you show regulators that if it drifts, what we do ensures that the product meets the same specifications as before? These things are challenging and not easy to address. Another challenge comes from the biologics themselves having both heterogeneity and variability. Additional difficulties arise from spectra overlapping and low sensitivity.  One compound has peaks in the same range as another compound. And you know, peptides and sugars, for example, sometimes overlap—not identically, but they nevertheless mask the spectrum. So, it's hard to dissect which one is which. That's where the difficulties actually occur. Even if you have individual spectra collected for each component, you still have difficulties translating them when they all come together at different concentrations and different types. These limit PAT’s capacity in terms of replacing end-product testing for biologics. But nevertheless, things progress, and hope is there. More and more technologies getting into the PAT space will find their way into process development, improvement, and manufacturing. Another challenge I want to mention is integration and data infrastructure. Data silos (how the data is collected, how it's curated, meaning metadata), whether all these PAT probes are interlinked or standalone, also introduce difficulty because we need systems that bring them all together as one organism. If we look at manufacturing as a plant, just like the human body is a continuously working plant where everything is interconnected and the brain monitors everything, we want PAT as an active participant, with software linking together, to operate this complex organism as one unit. So here I would mention synTQ and ZONTAL—those programs are trying to bring it together and close the gaps. These platforms definitely give hope to see when these systems are all intertwined and connected.

Codina, Anna
Perfect. And so far, the benefits of implementing PAT seem obvious. You just explained so many different examples and techniques that we could be using, and ultimately how beneficial this is. However, we are now in 2026. The guidance from the FDA was issued in 2004, and we've spent many years in a kind of lethargy, with maybe incomplete adoption in the pharmaceutical industry. What are the adoption barriers that made this quite slow?

Marina Kirkitadze
It's an interesting question. You're right: why? Because on one hand, there are a lot of proponents for PAT. And of course, if you are on the R&D or process improvement side, you definitely see the value. But manufacturing is stricter, because they have to comply with regulatory rules, and it’s a legal environment, not just technical and scientific. So, validation becomes a big thing. What can be validated has potential to go through, but overall, the validation burden and its cost are often what stop manufacturers from adopting PAT. Plus, there is the readiness of the teams to adopt. It's not just technical and scientific readiness, but also readiness to take ownership, the so-called long-term ownership. Quality assurance, manufacturing, pilot teams, manufacturing technology teams, and R&D need to work together to make it happen. Not just, okay, this is R&D and it doesn’t go any further because we're not ready. From the beginning, whether it’s a new product, an existing product, or an improvement, or a new plant, there is an opportunity to work together and implement PAT, even if it is in small steps. Teams can start with pH probes, something simple compared to infrared, Raman, or even low-field NMR, and then add more. We need to see if, in this collaborative environment, we can identify the roadblocks. If these are identified collectively from the beginning, there is a better chance to collectively win the budget needed to do the validation step and to see PAT come to fruition. So, I would say it’s really ownership and commitment. Of course, objections can be raised such as: “the technology is not mature enough, the models are not robust enough, they drift over time, what can we do?”, but all these questions can be asked, and then the team can go through whether it’s feasible or not. Out of ten probes, maybe three or four go through and that may be just enough to control the process. You don’t need too many; you just need critical ones that perform well and give you a continuous picture of what happens in these stainless-steel tanks or even in single-use systems. The digital world allows us to “see”, to give a kind of “digital vision” of what happens inside the process. This exercise is worth it, even though it may appear heavy or overly enthusiastic. The benefit is that, overall, companies can reduce the cost of quality and generate readouts that become release tests, release as you manufacture. I think that was the actual call from the FDA in 2004. But the technology and the complexity of biologics sometimes do not allow you to move fast enough. When there are weak connections, energy dissipates. So, we need to carry this energy through. It’s a commitment to get through the entire route and see this materialize at the other end. I would say long-term ownership by the teams themselves can help see it through. And I think there are companies, often CDMOs, that can be exemplary in introducing PAT and using it in their processes. Lonza is one of them. They use PAT in their processes, and that gives them a significant advantage.

Codina, Anna
It is certainly a differentiator and an advantage. And Marina, to conclude, if there was one piece of advice that you would give to those who are still relying on snapshots instead of the movie, and who are hesitant or thinking about implementing PAT but see the mountain to climb, what would you say?

Marina Kirkitadze
I would say the vision of the future is often a brave step. In manufacturing, it's often calculations, budget, and regulators. I think also it’s about how risk is perceived. In many places, PAT is associated with risk. Usually, you cannot convince people to take risks, but if you help them find alternatives (“if PAT doesn’t work, what am I going to do to release my product?”), then there is a better chance to introduce PAT. If this mitigation is embedded in the process, teams have a better chance to find solutions.
What I mean is, for example, if PAT doesn’t work, what do you have? You may have two probes sitting in a fermenter or mixing tank, both giving you a readout showing the process behaves as expected. At the end, you get a product that matches the specification range. Fine. But what if both probes fail? What remedy do you have? You may have a small piece of equipment—same principle, for example pH—at the bench, or in a nearby room. You can immediately check what happens. Some equipment is allowed in manufacturing suites and is robust enough to sustain CIP, cleaning, and sterilization. If you have offline equipment nearby, that’s your fallback in case something doesn’t work. If you satisfy the de-risking needs of the people involved in manufacturing, knowing the responsibility they carry to deliver and commercialize the product, then the proponents of PAT have a better chance to bring probes into the manufacturing suite and see them succeed. These aspects—risk and responsibility versus novelty and advancement—must be addressed together. If PAT works most of the time, teams gain far more insight than before. If something doesn’t work, it’s not a showstopper. There is offline testing available right at the manufacturing suite, also validating the inline measurement. It’s additional investment and time, but it’s doable. Manufacturing continues without interruption. If inline probes don’t work, the offline instrument reports the data. With that, I think crossing the bridge or climbing the mountain becomes like climbing Everest: you have camps along the way, step by step, supported by many people, until a few reach the top. In the same way, with support all along the way, there is a better chance for the probes to make it into the manufacturing space. My suggestion is to respect both sides. Both have value, and often they cannot agree because they don’t hear each other. But if both needs are satisfied, success is possible.

Codina, Anna
Excellent, fantastic, Marina. And with this, we conclude the first episode of “PAT Chats with Marina Kirkitadze”. Join us for the second episode, which we will dedicate to software and how important software is in PAT. Thank you very much, Marina. Thank you all.

Marina Kirkitadze
Thank you, Anna. Thank you. Thank you all. Bye.

Reference

Evaluation of Low-Field NMR as a PAT Technology for Upstream Bioprocess Monitoring. Gerzon G, Fischer C, Pennestri M, Hunter HN, Anklin C, Misra R, Wilson D, Sheng Y, Kirkitadze M. 2026. s.l. : Pharm Res. , Mar;43(3):891-903. 2026.