Mettler-Toledo International Inc. (MTD) Earnings Call Transcript & Summary

Hello, everyone. Welcome to today's Transform Productivity with Automated Sampling online seminar. My name is Wei Deng, the Market Manager at Mettler-Toledo. I'm excited for today's talk, and I would like to thank you and the presenters for your participation in today's event. A bit of housekeeping before we get started. Please submit your questions via chat. We will address them in order of when they come in. Any questions that aren't addressed in today's session will be given to the presenter and answered directly after the event. Our first talk today is from Brian Seifried with Bristol Myers Squibb. The title of the presentation is Application of EasyFrit for Crystallization Process Design and Optimization. Dr. Brian Seifried received his PhD in Chemical Engineering from MIT. His thesis work focused on creating mucin mimetic materials using a combination of synthetic biology and polymer chemistry within the Olsen research group. Currently, he is a senior scientist in the Materials Science and Engineering group at BMS, where he works API crystallization process development as well as investigates enabling technologies. Please start with the presentation, Dr. Seifried.

Hello, everyone. My name is Brian Seifried. And on behalf of my colleagues at BMS, I'd like to thank you for coming to my talk today. The title of my talk is Application of EasyFrit for Crystallization Process Design and Optimization. And I'll be telling you 3 brief little stories about this exciting new invention that Mettler-Toledo was so kind to let us beta test at BMS. And so before I get into the main topic of my talk, I'd like to make sure everybody is on the same page and talk a little bit about the solution phase in active pharmaceutical ingredient or API crystallization. So crystallization itself is one of the most commonly used isolation and purification of the active ingredient in your drugs. And this is because it provides great control over the form, purity and powder properties of your crystals. That being said, there is a whole host of information contained within the solution phase for crystallization. The way this solution phase is typically analyzed within the industry is through high-performance liquid chromatography where you take an aliquot of your solution, dilute it in some sort of diluent and then run it through a liquid column where the different components within the sample are separated based off of differences in properties. And then this is detected and can quantify the different species and solution. And this is super useful for crystallization process design because this lets you understand the desaturation or essentially, what's the concentration of your API and solution and let you know how well your crystallization is working as well as the impurity part or essentially how good is your crystallization at getting rid of your unwanted impurities. That being said, collecting the solution phase within a crystallization is not so straightforward. So crystallizations are characterized by being a 2-phased system. There's your solid phase, which is hopefully your pure crystal and then the liquid phase, which has everything else ranging from the impurities, solvents and your dissolved API. And that being said, HPLC does not handle solid-liquid mixtures very well. So what you end up having to do, at least within a standard practice within the industry is actually taking an aliquot of your reactor and then syringe filtering it. This ends up being a very time- and labor-intensive process that's very difficult to automate just because of the filtering component. And so this is where the engineers and scientists at Mettler-Toledo came up with a very incredible update to their automated liquid sampling system. They have a product known as the EasySampler, which essentially allows scientists to take liquid aliquots of their reactor vessel at predetermined times and thus automating the liquid sampling. The previous limitation to this sample though was a function of how it's designed. They can only handle homogenous liquids and did not handle 2-phase systems or crystallizations very well. This is because the way the probe works is it's inserted into your crystallization vessel, that probe is actually closed in the beginning, and at set intervals, the probe head will open up, liquid to flow into the probe and then it's diluted with various quenched or diluent solvents. However, they created a brand-new product that we were so fortunate to investigate and look at BMS, lovingly dubbed the EasyFrit. The EasyFrit system essentially is a filter frit attachment that goes on to the EasySampler probe. What happens is in solution, when the probe head actually goes out of solution, it depresses -- or extends a spring, which forces solution through the filter frit into the reaction packet and thus keeping all the unwanted solids outside the filter frit and then letting the solution phase go into your frit, thus letting you not clog up the filter -- clog up your EasySampler and still be able to collect samples from your 2-phase system for your crystallization. So like all good scientists, we had BMS got this brand-new toy and hearing how it's essentially too good to be true, had it run some validation tests. See if this is actually as good as the manual method. And so what we did is we took some acetaminophen, dissolved it in a 3:1 acetone water and measured the solubility both with manual poles as well as EasySampler poles. And we did each of these in triplicate. What we found is that the EasyFrit and manual poles are in good agreement for the solubility values and thus giving us enough of a leg to stand on for using this for our own crystallization process development experiments. So now that we got everybody on the same page, I'll get into the 3 little stories like I said. The theme of the first story is temperature. So crystallization itself typically works on taking advantage of the difference in solubility. And there's different ways of accomplishing the difference in solubility between 2 time points of your API. You can either use an anti-solvent addition, you can use evaporation or you can use a temperature-based crystallization. Just because [Technical Difficulty] called on your reaction or your vessel, it tends to come out of solution. But this creates a unique challenge when you're interested in taking liquid samples at higher temperatures. Just because if you remember, the main way of preparing your HPLC samples for understanding your liquid phase is manually filtering it. And so if your manual filter is at room temperature, but your reactor is at like 50, 60 degrees Celsius, it becomes a little bit of a challenge. The best practice within industry is typically to try to put all your syringes, filters, miles, everything in a glass oven to try to get the same temperature as your reaction vessel, and then pull things out of the oven, filter as quickly as possible and hope not too much of your API's crash out solution from the ensuing temperature drop. Just because it's essentially impossible to get temperature of the reaction vessel, as you carried across the room or into your hood. And so this became an issue for this experiment I was actually running earlier this year. I was running a cooling crystallization with a seed age at 70 degrees Celsius. And so I wanted to understand [Technical Difficulty]. However I was running this experiment with manual poles, I was as getting non-sensible data. If you direct your attention to this graph, you will notice that the solubility line are at around 120 mg API per gram solvent is a lot higher than the solubility or API concentration values I was getting when I was doing my manual poles. This made absolutely zero sense considering how the solution was a slurry and clearly supersaturated. And so I was kind of stuck. My colleagues recommended giving the EasyFrit a world and I succeeded in my experiment. When I reran the experiment with the EasyFrit, I was able to get API concentrations that actually makes sense. It's about solubility and value of the API and was slowly going down. This information has allowed me to start designing the seed age for my crystallization and essentially saved us from a lot of guesswork in super conservative estimates for seed age given the inability to collect useful data. Now the second story revolves around the concept of efficiency. So this story goes along with one of my colleagues who specializes in modeling crystallizations. One of the major tenets of modeling work is that there is a relationship between the accuracy of your model and the amount of resources and time you put into it. And typically, it's some sort of positive relationship. The more time and effort you throw into it, generally, the more accurate it is. That being said, [Technical Difficulty] models, there's some range where the results are actually acceptable if it's too far off from reality, your model is fairly useless. So the issue is that since we're engineers, we believe time is money, and we're always looking for ways of doing things better. And our end goal is always to try to improve the relationship of the resources and time we put into our model and its general efficiency. So this is a very cool to say that my colleague was able to do. She was trying to build a model for a different crystallization system. And she was looking to fit the parameter, I believe, the order in respect of supersaturation. She had begun with a very small data set. She had run 2 experiments, getting tired of setting up everything over and over again, and then she has had a little Eureka moment. She thought what if I just set up one experiment, put in the EasyFrit and just thermocycled my reactor, letting the crystals come in and out over and over again, maybe like 2 or 3 times. This way, I can get 3 extra desaturation curves for the price of setting up one experiment instead of having to come in and set up an experiment, clean up, set up in the next one and repeat. And what she found is a significant improvement in the model fit, both in terms of estimating the actual value as well as the standard deviation of the fit. That being said, she is a very responsible scientist, and she had to go and validate her EasyFrit iterative experiment by having to go and do those 3 extra experiments. But for the benefit of this talk, at least, we were able to get a head-to-head comparison and how effective her EasyFrit iterative experiment was. It turns out that her 3-for-1 deal experiment was just as good as running 3 additional experiments in terms of creating a model. And it just shows that the EasyFrit system itself gives a lot of opportunity for designing creative ways of maximizing your data output from single experiments. And so this brings us to the last story I want to talk to you guys today about. The theme of this talk is time. So a very common characteristic of crystallization is that, it typically is a very long and slow process on the order of hours, if not more, just because the way the molecules need time to orderly pack themselves in solution among other things. And so this creates some unique challenges, especially within the industrial setting for crystallizations that last longer than the workday. It's because the way that you typically run a crystallization and if it's a very long one, is you set it up in the morning, collect samples throughout the day, go home for the evening, come back in the next morning and say, "Hey, my crystallization is probably done", collect 1 or 2 last samples and just roll with it. But the issue is that, that type of schedule doesn't scale up when you go to the plant situation. Those are typically 24/7 operations, and they're going to want to know, hey, let's say, we are running to cooling crystallization. And the cooling graph is done. Do we need to hold it? Or can we just go with the isolation step? Problem is that if you're running this cooling ramp in the lab setting, you'll be like, "I have no idea. I'm at home. This is your cooling ramp finished at midnight." So we guess we need an 8-hour cooling or hold after this, but we have no data proving this. And sadly or ironically, this situation happened to one of our colleagues. So he was running a slow cooling crystallization at 0.1 Celsius per minute. And so based off of such a gentle cooling rate, we assumed that the desaturation was being reached at the end of the cooling ramp. And so his data looked a little bit like this. He set up the reaction in the morning or crystallization in the morning, the seed age would go, he'd be collecting the samples regularly, and then he just collected a couple of samples throughout the day before heading home. And then in the morning, he'd come in, collect one last sample, see it is completely desaturated. But that being said, we had no clear idea of how long this sample -- or if the sample is truly desaturated at the end of the cooling ramp, since this is happening while we are at home. Turns out when we got our hands on the EasyFrit, we reran the crystallization and were able to collect the data after the end of the cooling ramp and made a very surprising discovery. We found out that even though we're very gentle and very slow with the cooling ramp, our API is still not desaturating. And there's actually a noticeable amount of API still within solution. This was the basis of a series of follow-up experiments where we were able to collect more data overnight and determine an appropriate holding period after the end of the cooling crystallization and saved us a lot of headache when -- if there was like a major mismatch between the yield on the lab scale versus the plant scale. And so that concludes the 3 stories I wanted to tell you guys today. Just briefly recapping their highlights. The EasyFrit has shown itself that it can actually be superior to sampling at elevated temperatures just because the probe is already within the reaction fluid and at the temperature. So there's no temperature drop during sampling. My colleague was able to show that you can develop some very clever experiments for maximum data output using the EasyFrit and its automated nature. And lastly, I wanted to show that the EasySampler and EasyFrit together allows for the streamlining of data sampling, especially at odd hours and long hold points. And so with that, once again, on behalf of my colleagues at Bristol Myers Squibb, I'd like to thank you guys, the audience for your time and attention. And I'd like to open the floor to questions now.

Yes. Thank you, Brian, for the presentation and recommendations. Now we open Q&A for Brian's presentation. Brian, can you open your mic.

Yes. Can you hear me?

Yes. I can hear you, Brian. So yes, so the first question is EasySampler with EasyFrit saves resources and time compared to traditional best practice, as you mentioned in the presentation. Can you roughly quantify how many -- how much experiment is saved?

So it saves a lot of your personal time in the term of babysitting your crystallization because if you need a data point every 30 minutes, that means you're essentially tied to that reactor. I mean you can have some in between rest periods, but it's very difficult to set up something more time consuming, whereas with like the EasyFrit you can literally just get everything set up in a couple of hours in the morning and then go about your day and instead more reactions, more crystallizations, going to meetings, it frees up a whole lot of bandwidth. So it essentially turns an all-day event into a couple of hour event in the morning.

Yes. So essentially talk about it saves a lot of -- increases your productivity overall, right?

Yes.

Okay. Yes. The second question is, does EasySampler with EasyFrit meet all your sampling needs in crystallization process. That's a very general question.

I'd say it meets about 90%, 95% of the needs. There are certain situations where, I guess, intrinsic limitations exist. One of them is just like the EasySampler does have some minimum dilution levels. So if your samples or your systems has extremely low concentrations, you might be over-diluting your sample and not be able to read it with HPLC, but those are few and far between -- at that point, you can just pick a manual sample. So -- but overall, it's very good from what our experience is.

Yes. There's also another question. Is there a limit in how low of the solvability you can measure towards the end of the crystallization given the subsequent dilution by the EasySampler.

I think my previous response kind of led into this a little bit. But yes, I think the setting for the EasySampler that I was using at BMS had a minimum dilution level or dilution factor of 80. And so there is some API / HPLC limits that did -- that couldn't dilute or couldn't detect the API, when it was diluted by 84 at that point. So that was like usually the very final couple of samples. Normally, at that point, though, it's -- I can just -- I see the mother liquor and take that directly. So...

Yes. Thanks, Brian, for answering all the questions. For our audience, if you still have questions afterwards, please put in your questions in the chat box so that we can direct to Brian afterwards. Thank you, Brian.

Thank you.

Yes. So we are moving on to our next talk today is Automated Sampling and LC Analysis of Process Solutions given by Mr. Makoto Michida. Makoto Michida is a process chemist at Daiichi Sankyo for 25 years, with recent interest in lab automation and automating medium-scale synthesis experiments. His expertise and passion for automation have been recognized, and he's expected to contribute to future research in the field. Mr. Michida-san, please start with the presentation.

Hi, everyone. Thank you for inviting me to this revenue. My name is Makoto Michida from Daiichi Sankyo, Process Technology Research Laboratory. Today, I'll be talking about development of Automated Sampling and LC Analysis System of Process Solutions. Here's agenda of my presentation today. The first is background, I will provide an introduction on the motivation behind lab automation, especially medium-scale experiments. Next, we will explore automation for qualitative analysis. By connecting the EasySampler with HPLC system to automate sample preparation and injection processes. Then we want to quantitative analysis. We will discuss the challenges that need to be overcome in this area. We have been working to conduct accurate and reproducible quantitative analysis by using of the systems. Lastly, I will summarize the key points of this presentation and emphasize the potential benefits of lab automation for process chemistry. First of all, let me share with you why I am interested in automation of medium-scale experiments. There are 3 main reasons for this. Firstly, medium-scale experiments essential for scaling up all types of projects. Secondly, they take a long time to complete, especially when considering the manufacturing timelines. Lastly, they are suitable for automation due to their predetermined experimental procedures. For these reasons, we have been working on the automation of medium-scale experiment for a few years. We identified that the core technology to achieve this is making [Technical Difficulty]. And we believe that there are 4 key technologies such as addition, sampling and analysis, separation and concentration. Don't need to be developed for this purpose. In this session, let me share with you about the automation of sampling and analysis. Basically, sampling and analysis of process solution is essential for data collection and it frequently performed operation and process development. Example includes reaction monitoring, quantitation in separation and the variation of stability and the concentration condition and quantitation of losses in the filtrate during the crystallization. So you can imagine there are many opportunities in medium-scale experiments. Here is an example of a typical timeline for medium-scale synthesis experiment. As you can see, the process involves sampling, preparation, HP analysis and then data variation. Although each individual operation may just only take a minute, they must be repeated approximately 10x that makes the entire human resources inefficient and nonproductive for researchers. So we came up with the idea of automating these reserve tests. To be more specific, let me show you with the actual procedures. First, we use micropipette to sample 5 to 10 microliters of the process solution, then diluting to 1 millimeter in the vial and set it in the HPLC autosampler and run. After that, we take and analyze the results. As a navigator to somehow automate these operations, we thought about finding a way to directly connect the EasySampler to the HPLC. First, we tried the qualitative analysis by connecting EasySampler to HPLC. Originally, the EasySampler takes 20 microliters of sample, performs automatic programs, planting and division and then transferred to the solution to stop [indiscernible] in the carousel. This equipment is already very useful, but it cannot automate HPLC analysis by itself. So we tried to transfer the prepared solution to the [indiscernible] in HPLC autosampler instead of using the [indiscernible] in the carousel. The [indiscernible] mounted autosampler is provided by [indiscernible] Corporation, and it's totally used for [indiscernible] testing of direct product and flow reactors. We investigated the timing of injection from the sampling and the concentration. This image shows the concentration distribution of the sample in the tube. Initially, the sample is quenched by mixing with the quenching solution, then transfer the mixture by diluting solution to the [indiscernible]. The peak of the sample was supposed to come after approximately 100 seconds from the sampling. This slide shows the relationship between the timing of injection and concentration of the sample. Before testing, we prepared 0.5% of sample solution as a standard and then compared the peak area. If the concentration of the sample was constant, it should be around here. But the result of the distribution was widely changed according to the injection time. As a result, qualitative analysis was possible by time adjustment of injection, but large concentration distribution when low reproducibility of the peak area were observed in the [indiscernible]. Then our focus shifted to quantitative analysis. The challenge lies in how we can consistently prepare the sample solution with a constant concentration. As we mentioned before, it was confirmed that the sample can be transferred from EasySampler to HPLC. Then instead of the [indiscernible], we try to prepare the sample solution by connecting 4-milliliter [indiscernible] in autosampler through a switching [indiscernible]. The sample solution was dispensed by opening the [indiscernible] with accurate duration to prepare the constant concentration samples. The closure timing was adjusted within the range of 50 to 60 seconds, and each analysis was performed with 3x each. And we observed that the peak area closely match the standard sample. However, a few samples degraded from the standard. We attribute this to the position of the diagram pump. But as a result, most of the samples yielded consistent result, [indiscernible] as the optimized condition like this. Next, let me show you the software configuration for automating this entire [indiscernible]. We use eye control for EasySampler and lab solution for [indiscernible] HPLC. Here is an example for conducting 5x sampling and analysis every 2 hours. But first, input the number of repetition in the [indiscernible], in this case, 5x, then put the command for sampling with the dilution factor. In parallel with this, at the same time of the sampling put this command for sending a [indiscernible] signal to the HPLC for 80 seconds. This is very important because the sampler transfer starts afterwards. Then send the [indiscernible] signal to the switching [indiscernible] to transfer the sample solution to the HPLC for 85 seconds to prepare a 0.5% sample solution. Since this process is repeated every 2 hours, include a wait command like this. Regarding double solution, the program consists of 4 methods. The first method is designed to receive the signal from iControl for the following sequence. In the [indiscernible] method, the sample solution is transferred to the [indiscernible]. The [indiscernible] method is to maintain a constant concentration. And finally, the sample is analyzed as usual. Here is an example of tracking reactions every 1 hour. This reaction proceeds through multiple intermediates. So it is difficult to understand what's going on. However, by increasing the sampling frequency through automation, it was possible to follow detailed trends of each product in complex reaction systems. This is also the same reaction. We have variated tracking the reaction yields by quantitative analysis every 1 hour. By using this peak area, targeting the yield by comparing the area of the standard sample. This approach can measure the reaction yield directly without being affected by other peaks. This slide doesn't contain actual data, but this system would be effective for automating these analysis. For example, it can be utilized for quantitative stability test. Also, for the time cost monitoring of the filter rate in crystallization to confirm the endpoint by attaching a filter. In summary, the routine and repetitive operations of sampling and HPLC analysis in process research can be automated by connecting EasySampler and HPLC. Also, we can see the results through remote desktop. So all the operations can be conducted without being on site. Currently, we are actively developing the automation system for medium-scale [indiscernible] experiments. We believe that such a device would be in demand in Euro laboratories as well. If you are interested, please feel free to contact us. We will be delighted to collaborate with you and explore the possibility together. That concludes my presentation. And finally, I'd like to make one more announcement. I am currently serving as an Editorial Advisory Board member for the OPRD Journal, and we are planning to feature a special issue on lab automation. I encourage you all to actively contribute and submit your research to this upcoming issue. Your valuable insights and findings will greatly contribute to the advancement of lab automation. Thank you for your attention.

Thank you, Michida-san for the presentation and recommendations. Now we open for Q&A for Michida-san's presentation. Michida-san, can you open your mic?

Sure.

Yes. Michida-san, thank you for staying online. I know it's very late in Japan. Now we -- so the first question is what was the biggest bottleneck in this approach towards the next level of automation.

So -- what does it mean?

Yes. So what was the -- so I think the question is about, if you want to reach the next level of lab automation, as you mentioned in the presentation, what was the still the biggest bottleneck in your mind.

So -- that's a great question. But I'm currently trying to lab automation. So the biggest problem is automatic separation and solid handling. So it's very difficult.

Yes. So you say that the biggest bottleneck that in your mind is the separations in solid handling.

Yes.