Conformation Change of a Monoclonal Antibody in Real Time, Mimicking Low pH Hold Viral Inactivation

Good day everyone on behalf of Cambridge Healthtech Institute global web symposia series and our sponsor RedShiftBio I’d like to welcome you to Confirmation Change of a Monoclonal Antibody in real time mimicking low pH holds viral Inactivation. My name is Elizabeth Lamb and I’m the host and moderator for today’s event. Now I’d like to introduce our presenters for today First if Eugene Ma Chief Commercial Officer for RedShiftBio and our second presenter is Steven LeBrenz Scientific Director at Janssen R&D PDMS. Welcome Eugene the presenter ball is yours. Thank you this is Eugene Ma with RedShiftBio and I’m going to provide a quick introduction to the AQS3pro which is a new type of instrument that is used for biophysical characterization of proteins. This is the instrument that Steve used in his experiment, which you’ll hear about in the second part of this webinar. So as I mentioned the AQS3pro is a new type of instrument that is being used now for biophysical characterization of proteins Many tools currently exist that cover different types of measurement like CD and FTIR for structure DSC for stability SEC for aggregation and UV for quantitation, but industry is never one to sit still and demanding the ability to measure over wider concentration ranges, have greater sensitivity to see smaller and smaller changes, the extraction of better and more usable or actionable data with each measurement All this in addition to demand for simplifying the workflow, not just in prep and test but also in analysis. So we developed the AQS3pro to address these needs. It performs five key measurements in a single analysis. You see some of them here on the right aggregation, quantitation, stability similarity, and structure. It performs this with 30 times the improvement in sensitivity when compared to conventional IR methods. It has the widest concentration range of any platform. We’re routinely able to measure between 0.1 to over 200 mg/mL and it has an integrated multisample capability using a well plate design for up to 20 times savings in direct labor cost. And finally it has an automated and integrated protein analytics software suite which is able to provide all level of detail satisfying the needs of operators through scientists. The AQS3pro works in the mid IR where it probes the vibrational modes of the carbonyl bond in the protein backbone and this reveals spectral signatures which are mapped to the secondary structure of proteins. Some of those are shown on the bottom right here alpha helical structures beta sheet for example. And this technique is very powerful its been long recognized as a very effective method for the analysis of protein structure, but it hasn’t been fully exploited because of constraints in the conventional IR methods available to date particularly for aqueous solutions. So we developed a new platform a new technique called Microfluidic Modulation Spectroscopy or MMS for short and it’s really built upon three key components which are shown on the left. A tunable infrared laser microfluidic referencing is performed with a fluid cell and the third piece is our advanced analytics And so these three are integrated together and onto a platform which basically works as follows. On the right here you see a schematic that basically shows laser absorbance spectroscopy. Our source here is our tunable laser instead of say a globar so it’s extremely bright. It allows us to use a much simpler detector a t gold detector and in between we have our fluid cell. Now this cell is actually designed with two inputs one for a reference fluid and one for a sample fluid and these fluids actually modulate back and forth typically between one and five times a second and this is performed in order to achieve background subtraction. So we’re able to do this very very quickly and very timely

And this is possible because of the geometries of our microfluidic cell and all of this is choreographed with our integrated software so it coordinates the activities of our tunable laser with the modulation of the fluid and is able to construct a very accurate and a very repeatable spectral profile for all our measurements So all those put together allow us to achieve a sensitivity which is about 30 times better than conventional methods all while simplifying the actual operation we’ve tried to take the user out of the measurement as much as possible here so that measurements are very very consistent and repeatable. So to test the performance benefits we began by first modeling the performance of this platform comparing it to traditional IR methods so what you see here is a performance plot with sensitivity on the y-axis shown in mg/mL of protein concentration versus path length cell path lines on the x-axis. Shown in the blue and the black are the performance curves of conventional IR tools you can see that optimal performance there is achieved around six or seven microns path length which is why if you buy a cell today there are typically at that geometry. MMS in comparison is shown in the red curve here and what you’ll notice is that there’s about a thirty times improvement in sensitivity in addition to a much longer path length for the cell. So we go from six or seven microns to typically about 25 microns in path length. We’re able to do this because we have so much more optical power from our laser and this increased path length really improves our sensitivity to conformation change so in addition to that I mentioned that we utilize microfluidic referencing so we’re able to perform background subtraction very very rapidly and repeatedly We’re able to optimize the optical power that is on our detector because we have a tunable laser here so we’re much more efficient with the dynamic range of our detector and water vapor correction and subtraction is performed several times the second and is very very accurate. All this capability really addresses the need for an expanded concentration range Groups that we’ve worked with have cited this as one of their major motivations and one of the key features of MMS that they like they’re able now to measure samples from 0.1 mg/mL to an excess of 200 mg/mL all with the same tool so now you can look at data from discovery through formulation with the same apparatus. So you now measure at the concentration of interest there’s no need to dilute your sample this simplifies prep and workflow but more importantly you’re able to actually measure the protein conformation in the formulation of interest and that’s just better science. So these performance advantages were independently validated by Elion which is an analytics group out in Colorado and they performed experiment where they mimicked or simulated protein conformation change by spiking a beta rich protein in this case and IGG with alpha rich protein BSA and they compared MMS which is shown here on the right with traditional FTIR getting at the limit of quantitation using their metric which is a weighted spectral difference. So what you see on top here are second derivative spectra for the samples. These are multiple spectra and you can see they overlap fairly well in most places and the change by spiking with an alpha rich protein is shown here in this spectral region. Again FTIR on the left and MMS on the right. When they dive into the data they see that the limit of quantitation under FTIR measurements here the concentration I should mention is 20 mg/mL is about 22.7% in contrast MMS yields limit of quantitation much much better it’s

down to below 1% so in fact this matches up very very well with our expectation that we’d get about 30 times improvement in sensitivity What they also noted was that there were significant labor savings the average labor time for each sample measured under conventional methods was 30 minutes per sample and with MMS it was about one and a half minutes per sample and this was as I mentioned earlier not by accident this is by design when we created the AQS3pro we designed it to work with modern workflows so we integrated a well plate design with it and this slide just provides an image showing what our typical setup or workflow looks like. Samples are loaded directly into the well plates that well plate is loaded into our instrument the user is able to select a or build or select a preset test protocol press go walk away all measurements are performed automatically here and you’re able to view this at a computer remotely or or at the bench So this is typically what’s done or how we use the tool with our collaborators like Steve’s and this point I guess I will pass the mic over to him Good morning thank you very much this is Steve LeBrenz from Janssen welcome to the portion of my talk where I’m going to discuss our work around a proof of concept to develop the process to simulate viral inactivation hold and looking at molecule conformation under those conditions as part of a purification development process. So we were looking at conditions where previously in the history of working in this area we’ve had some non-conformance events in manufacturing this is many years ago and it was the closing of those non-conformist events took months to complete to plan to do the work and to close out These non-conformances can be purposeful or accidental as part of development activities and we deploy standard testing typically chromatography or mass spec and over and over again there was no indication of risks in the product We did have a separate technique which allowed us to look across number of batches involved in those non conformance and actually see differences across those and what that led us to do is identify that you know once you alter a sample out of that state that that condition where you’re mimicking you’re inferring results to the previous state and sometimes changing the sample can mask those results and your traditional techniques may not pick up what’s going on. So the purpose here was to perform a proof-of-concept assessment and look at protein stability at low pH over time in reference to the previous work that was done. To discuss what the thought process is behind the experiment we go back to a combination of colloidal biophysical chemical stability that was discussed in Yahn and Radford for his paper in 2005 and I used this figure from their paper that looks at the energy state and across the whole range of protein confirmations that can be accessed within a given energy state for a molecule different buffering conditions and what we’re looking for is ways to assess the conformation where other techniques won’t give you that information in this space And as indicated by the arrow which indicates a native state or a preferred condition it’s not necessarily the most stable that is it doesn’t have the lowest energy state across this whole funnel diagram and we are really interested in understanding how the conformation changes in these different areas in relation to possibly seeing something like an aggregation event which is on shown on the right

side of the figure as well as possibly are there are there intermediates folding intermediates that might be leading to that state or are we picking up changes in these partially folding States. So we want to look across this and be able to identify where our system is breaking down or if those changes are even relevant to the process. To go over the experiment shown here is a lot of text it outlines exactly how we prepared the experiment in brief mAb was can was buffer exchanged using the GE healthcare PD ten columns to obtain an exact buffer match between the background and the sample itself. We did some work before showing that this is a great way to generate the background match to the sample We also then worked together with RedShift and determine how much acetic acid was required to add to mAb in an acetic acid Tris buffer at pH 7 to obtain 2 PHS pH 3.5 and pH 5 and with that in hand we are able to generate samples a large bulk sample which was then dispensed across the different sample positions in the plate for the AQS3pro and assess those over time What I’m showing here is the absolute absorbance data the raw data that is from the instrument in the upper left-hand panel you see samples from fifteen minutes to three and a half hours that were at pH 3.5 the data looks very very consistent across that approximate 3 and 1/4 hour time frame. We then add the control data the pH 7 sample to that which is the upper right hand image and you can definitely see in the raw data here a change in the spectral data indicating a shift in conformation especially in the amide region. Then in the bottom left hand corner what we see is an overnight sample so the instrument was a was put into a pause mode or a wait mode and then at in the beginning of the day and the next day the sample the last sample data was collected and you see the pink line with a large change again in spectral data and there’s many items that might be represented by that change it could be a drift in sample change like background subtraction change but we believe from included in the data and discussion further that this may actually be a true change in conformation and the protein secondary structure. To further process the data you then take that raw data perform a second derivative calculation to obtain the second derivative spectra which is what people are generally used to seeing in publications because under these conditions you can the deconvolute this data and understand what different structural components are in your sample And under here what you can see again is that you have all three previously identified species the control at pH 7 the overnight sample which is in the pink and then our data from 15 minutes to three and a half hours under this what looks to be a red line there and you can definitely see changes in intensity in the second derivative spectra of the samples and these will be then deconvoluted to show you different conformational proportions indicated to beta alpha turn and helix. So you can see here the pink line which was from the control or I’m sorry the greenish line which was from the control instead of being very very different from the previous spectra which you saw in the last slide it’s actually quite aligned with the other components in here albeit different. This is another step in the analysis of data and the Delta plots here are a subtraction of the control from all the other spectra and you can designate which sample you want to subtract from everything as the baseline sample. The control is bracketed by the

dashed lines showing the variability of the control showing that is in a good stable sample and then what we see in this data is a significant difference between these samples from fifteen minutes to three and half hours as well as the overnight sample but we particularly want to examine the time data and what you see here is an indication at 1638 inverse centimeters this key signal for a beta sheet structure you do see some changes that take place over time with this. This data is deconvoluted or excuse me first we’re going to look at the time chart and what you see here is the intensities of those signals and we’re particularly looking at the 1638 signal here and what you do see is you start to see the changes occur once you go past approximately ninety minutes in the data and what we’re indicating here is that this is a conformational change over time at pH 3.5. To first glance here this may not look like a significant change but the scale on this plot is only approximately 0.02 units and we’re seeing about a point zero zero one two point zero zero two shift in the intensity so we’re looking at a five to ten percent shift in that signal it may not look like much but it actually is quite a bit. When we deconvolute the data is where some of our information comes in. In the left hand side what we’re plotting is the percent higher ordered structure for each of the individual components for a monoclonal antibody obviously we have quite a bit of beta turn as well beta sheet structure as well as turn structure and we’re low in unord and Alpha structure What you definitely see here our changes between the beta and the turn significantly from the red our control sample to all of our time samples here It’s the overnight sample that then goes back and shows you an increase of the beta sheet structure and a decrease in the turn structure over this period of time and this is further giving us insight that we believe what we’re seeing here is actually not something to do with the long term hold but is actually a confirmation or a possible colloidal interaction taking place in solution. All samples at the end of the experiment were clear and clean to the naked eye so there was no obvious precipitation of sample over time Likewise when we look at the standard deviation on the data for each of the structures it shows that the individual numbers do not overlap therefore we believe that the changes are real and statistically different. So as I mentioned we also ran pH5 and this experiment was performed in the same way that the acetic acid was added to the sample and when in bulk and the sample is submitted across the plate and this experiment shows that again over time there wasn’t a huge change in the structure of all the samples. We know for this particular mAb it’s preferred pH is approximately 6 which is demonstrated the pH 7 as our control is where we like to keep it around and that at lower pH is it has a little bit less stability but here you can see under these conditions that the lines were pretty flat we did see this overnight shift in again that 1638 beta sheet signal. When we deconvolute the data and what we have here is a higher ordered structure analysis of this data you see a very very small conformational change and again when you would come back to the numbers we do see significance the changes are real but these differences are very much smaller than what we’re seeing in the pH 3.5. As I mentioned the this the pH that this mAb prefers is pH 6 and so with less perturbation that is only going to pH 5 we don’t expect as large of changes. So what are we looking for here So under native sample conditions that is under conditions that would be representative of low pH hold in a purification process over three and a half hours at pH 3.5 there was a

conformational drift and at pH 5 there was no conformational drift or very little drift and so what we’re looking for here are conditions that tell us what might be going on in solution that is representative for purification. why is this important? it’s important because it helps us understand the time that the mAbs and how they behave over time at low pH for viral inactivation our process is typically 90 minutes and as this data shows 90 minutes seems to be pretty good under these extreme conditions that we were mimicking here with a lot of additional Tris in the buffering condition. And this also gives credence to the ability that even if you’re doing very early development as Eugene was talking about in his slides at the beginning there’s a potential here to use this to find resistance to conformational change say you have four or five six mAbs and you’re trying to determine which one to carry forward this will give you insight into developability that is which of the mAbs would be most stable against these ph denaturing conditions or ph conformational change and you could use that as a selection criterion. So why should we use the AQS3pro? Well we’ve actually followed up this experiment over the previous few months here of the year and new data from orthogonal assay indicates that this experiment can be improved. As I mentioned at the very beginning samples were prepared in acetate tris which is a normal buffer for this part of the purification process but what we were able to do is look at samples that came straight off of a protein a column only in the eluding acetic acid buffer and determine a ways to to look at that stability using this orthogonal assay and test different additives. We need to determine we saw significant changes in that screening but we need to determine if those observations are related to conformational change and are we seeing similar or different changes in the intensity of those particular wavelengths we saw under this experiment Our basic assessment that we’ve been doing cannot delineate these kind of formational flat effects from colloidal effects with an increased amount of experimentation and observation using the AQS3pro we think that we can answer this question you know conformation versus colloidal effect With the direct sample analysis as I mentioned we did this years ago but we didn’t have any secondary structure analysis we’ve actually determined compared to especially a circular dichroism instrument that we have that our sample testing time was reduced by 80% that is we could essentially do the work of a week in a day using the AQS3pro. Also the samples are analyzed under these negative conditions and that gives us 100% relevance to our purification process development. So we can take samples directly from a protein a a small protein a elution and dispense those samples across a plate and put it immediately into the AQS3pro that would allow us to perform the work essentially right there on the spot relevant to our purification process With that I’d like to thank you for your attention and open the floor up to questions. Thank you very much Steven and for your presentation as well Eugene so our first question how much optimization of parameters is needed before measuring each sample or spectrum and for each sample set?. In my experience it only took us about it only took us about morning’s worth of work on individual samples to find some optimization conditions after that it’s generally sound for the molecule that you’re examining. If you change the concentration then there’s a few parameters that you might have to quickly adjust and you can usually do a best guess to make sure that you’re collecting good data. Yeah Eugene here I’ll just quickly chime in too if the question is more about optimization of operational parameters based on say changes in viscosity concentration

things like that the instrument itself has an automatic discovery routine or process that it executes where it will change and vary the backpressure and determine what the most appropriate settings are for the sample that you happen to be testing and that’s that routine as mentioned it is automatic and typically runs very quickly usually in about two minutes or so. Excellent thank you our next question is there any water vapor subtraction applied to the MMS spectra Eugene again I’ll say I’ll take that one yes MMS as I mentioned in my part of the talk performs automatic background subtraction and typically between one and five Hertz so it’s essentially a comparator tool and so any vapor that’s present in or during the measurement of your reference fluid is also presumably present when you’re measuring your sample a half a second or shorter after and so those get background subtracted out automatically. Alright thank you the next question what are the upper and lower protein concentration limits for this technology. Eugene again I’ll I’ll answer that we routinely measure between 0.1 mg/mL to as I mentioned earlier we’ve gone as high as 230 mg/mL most of our testing is done closer to between 1 and 10 but we’ve we’ve certainly pushed those limits I suppose we could go higher than 230 but we haven’t tried yet Excellent our next question are any buffers typically found in protein formulations that are incompatible with microfluidic modulation spectroscopy? We haven’t found it yet we we’ve worked with a variety a large number of collaborators and looked at many different types of buffers that are typically used in development we haven’t yet come across anything that has been incompatible. Part of the advantage of this technique is that it’s just a straight-up differential absorbance measurement tool so if if you’re able to match your your buffer your reference to your sample having common buffer there it will do an excellent job subtracting it out so in spectrally speaking there’s no contamination or interference from excipients and we haven’t come across any buffers that have been problematic from an instrumentation point of view yet Ah right a very practical question do you demo your system? yes absolutely. Alright our next question from Korth why was aesthetic acid use for pH adjustment instead of a strong acid like hydrochloric acid it would seem like acetic acid was result in a change in OAC concentration as well as provide an oxidizing agent. Sure I’ll answer that acetic acid is used throughout our purification process not hydrochloric acid so that’s why it was chosen. I did see the question in reference to acetic acid as an oxidizer it’s our experience that acetic acid actually doesn’t lead to these oxidation events in fact using hydrochloric acid in bio manufacturing would pit the steel which would lead to oxidation events. Excellent we have a question from Lou have you ever looked at the conformation changes reversed during the neutralization step? Very good question that’s actually the next step after this proof-of-concept study so we’re expecting delivery of our AQS3pro our AQS3pro this month and those will be the first experiments going on. I guess I guess that answer is stay tuned for future results. Exactly and it’s it’s also in combination as I mentioned we have more results going on my best estimation is that we’ve probably enough new data looking at this low pH hold and neutralization step to

provide us anywhere from four to eight months worth of work on the system when it comes to our lab. Our next question has RedShiftBio done any work comparing the precision and accuracy of the AQS3pro to a CD spectra device? Yeah Eugene here yes the answer that question is we have in fact as part of our work with Elion and which I presented in one of my slides they also looked at a performance of CD. So if you go to our website you should be able to find some information on that as well. Alright and we had a follow-up question to the previous question they’re asking how will you be letting folks know about the results once you get the new unit tested? I’m guessing that might be for myself it’s possibly stay tuned we have direct interaction with RedShift regularly and we might be able to put together a white paper as well as potential to present at conferences. Our next question could you think using this technology to monitor your process at bio production scale? So I believe you could be doing that as long as you classified it as an in process control test not as anything that would be GXP reportable because of that it’s essentially you’re not at the end of either part of the process you’re not pre purification you’re not post purification that’s my opinion The beauty here is that you could do as is often can be done in manufacturing you could be pulling thieving samples from certain points in the purification process and then because of the short period of time it takes to analyze the samples you could actually generate the data before the next step is ready to go in the plant Alright and Steve since you’re on we have a question specific to one of your slides the Delta graph that you presented you can see well I’m not sure what the questioner is asking but differences can you see differences in 15 minutes at low pH so from our pH 7 sample to the pH 3.5 sample yes those differences were there if if we look we refer back to the Delta graph the first two points on the Delta graph were actually the control at pH 7 so then the severe drop we saw was the pH 3.5 sample and that first 3.5 was at 14 minutes so the control samples were run before that that 14 minute sample and you see that distinct conformational change at that time. Alright understood and then Andreia asked can you calculate melting temperatures? Yeah yes in fact we’ve done a lot of thermal stress work so I would say technically we wouldn’t calculate or determine a melt temperature we do measure and observe by a TM where the M might be a mid point temperature basically what’s done is we are able to thermally stress proteins outside the instrument for now quench it and look at the conformation that way and when we plot a spectral change at a particular signature point against say the the stress temperature we get a plot that looks very similar to what you might see for a DSC and from that you can calculate a mid point temperature. Okay and then Eugene a question for you from Prathima what is the requirement for the purity of samples for this measurement an interference of HCP? So I guess it ultimately depends on the application again when you dial everything back the the instrument itself is a comparator so if you are able to match impurities that might be in your sample also in your buffer then it’s not an issue otherwise you will like with any spectroscopic method be measuring the difference in

this case it might be it might include the impurities so if you’re able like excipient various excipient that might be in your buffer if those are captured in your reference then it’s not an issue. Alright and a follow-on question to that one has this been tried for the VI of other recombinant and fusion proteins that don’t use protein a? So we have worked with a number of collaborators looking at all various proteins a lot of mAb’s but we have looked at fusion proteins ADCs peptides as well. And another VI question for you Lou asks or says generally as part of the VI studies we apply mixing at a defined mixing rate is this capability possible within the cuvette during measuring? That’s an interesting question it’s something that has come up in discussions we’ve had with various groups so in principle yes it’s one could envision doing that the tool right now in the cell that we’ve designed for it is is not designed to perform the mixing but one could see that with some clever designs those types of applications could be addressed with this tool Excellent and then a final question from Michael have you had the chance to correlate the conformational changes seen and the formation of multimers or aggregates and subsequent DSC unit observations? Yes we have that is work that we did with an academic group it’s not yet published so I can’t yet comment on it but we have we have matched those up yes. Alright that appears to be all of our questions for today with that I’d like to thank so much our presenters for today Steven LaBrenz and Eugene Ma I’d like to thank RedShiftBio for sponsoring today’s web symposium and most of all I’d like to thank those of you who came and spent this time with us. We hope you gained some answers that will make your life in the lab a little bit easier going forward. So on behalf of Cambridge Healthtech Institute’s global web symposia series thank you all so much and have a great day