Hello, everyone. My name is Tom Witkos, and today we're together with Hadley Myers from JMP. We'll be presenting to you about speeding up assay development with automated workflow for assessing plate bias of microtiter plate assays. First, I would like to start with acknowledgements of several team members from AstraZeneca, Deepika, and Barbara, who has helped us with explaining certain concepts and also testing the JMP plugin.
During the presentation, I would briefly introduce to you what the potency testing of biologics mean, how we use microtiter plates in our work, and also statistical analysis such as plate bias and uniformity that are very important for assay development. U ltimately, we will present to you the JMP Add-in that Hadley and I have written that allows us to perform this analysis in a much more streamlined and more powerful manner.
First, I would like to put potency assays into drug R&D perspective. There are a number of biological assays used both at the pre- clinical, clinical, and commercial stages. First, we typically have screening and selection assays at the pre- clinical stage. Those types of type of assays are trying to answer the question, which molecule has a biological activity? I dentify candidates for drugs.
When the candidate drug has been already identified, we would move to something which we would call potency assays. Here we only work with a certain drug candidate, only one. We manufacture many different lots of this molecule that are going to be used in clinical trials and ultimately as well, lots that would be made post- approval of the drug in the commercial setting. H ere, the question that we are asking is, is the biological activity at a similar level between lots? A ll we care about is the lot- to- lot consistency, whether it's preserved or not.
A s such, potency testing is a critical property of the drug, and it's part of something which we'll call CMC, Chemistry Manufacturing Process and Controls. Analytical testing, such as the biological activity tests that our team is in charge of developing, are quite important to feed into the optimization of the manufacturing process at pretty much every stage. B oth at the upstream and downstream stages. W hen the biologics are being manufactured and then further purified, as well as at the stages of developing formulation and also packaging for these drug molecules.
Screening and potency assays both measure biological activity, but they do it in a slightly different manner. Within the screening bio assays, we have quite a few compounds that are not going to have any biological activity or very little. Some with moderate and also a few compounds with quite high biological activity. Because we are focusing on identification of candidate drug molecules, the differences that we see in the biological activity are going to be very high.
Basically, our assays need to resolve lock scale differences between different drugs regarding their biological activity. For the CMC potency assays, we already have the candidate drug nominated. W hile making different lots of the same molecule, we would expect when the manufacturing process is well optimized and controlled, that the differences that we observe in the biological activity are going to be small. W e need to quite precisely resolve small differences within the twofold change between different lots of the same drug.
There are a number of different potency assays set ups. Without going into too big details, ultimately, what we try to do is measure biological activity by quantifying the target binding, m ainly the protein that our drug interacts with. T hat can be either as part of the recombinant protein or being presented by the cells that express a certain target on the surface, as well as quantification of effect of that target binding. I n this case, that will be ultimately cell- based report essays where we will be looking at the activation or incubation of certain signalling pathways due to the interaction of the drug with a certain receptor.
A s you can see, regardless of the assay types, in all of these settings, we use micro titer plates. Most commonly, we use 96- well plates, but a higher number of well plates are also possible to be used. Within one assay run, so within one experiment, we use either one or multiple plates, and that would depend on in-plate and place- to- plate variability. Ultimately, within one microtiter plate, every single well can be seen as a separate mini reaction.
In this particular case, we have a very common protein target where our drug interacts to. T hen the amount of the bound drug is detected with the secondary antibody. In this example as well, every component is the same across all the wells. The only difference being a gradient in the concentration of our given molecule. T hat will be delivered in something which we call dilution series. O n each plate, we'll have dilution series of reference material and also of our tested samples of unknown biological activity.
By having a dilution series, we would expect to see a dose- response curve. W hat you can see here in case of the binding, what we would expect is that the more drug we put, the higher the response we should be able to observe. We would then fit that data into a model. Most likely it's going to be a non-linear- 4PL model. T hen we would work with these fits, and we do a pairwise comparison of fits between the reference standard and our tested molecules, where we would share the lower and upper asymptotes and the growth rate, so the slope of those curves.
Given that the curves look similar to each other, we would be able to calculate the inflection point. In this case, it's going to be either EC 50 or IC 50, depending whether we are looking at the activation or inhibition. And that difference in the EC 50 or IC 50 between these two curves would then can be mathematically converted into something we term % relative potency. T hat's our ultimate readout. Within potency assay development, there are a number of experiments that we need to perform in order to achieve an assay that can be used for biologics testing and can ultimately be validated.
For example, here we would start with screening reagents and assay conditions, both done in the one factor at the time and the Screening DoE fashion. Then we would establish a proof of concept dose- response curve. We would look at the plate layout and then lock it together with its experimental conditions. A ll of these steps can be analyzed in JMP. F or today's presentation, I'm going to focus on quite crucial part of this assay development step, which is establishing the plate layout. L ooking at some time through the term plate uniformity and plate bias.
First, let's talk about plate bias. Let's step back for a moment and think of a situation where we have have every single well being identical. Meaning that we even have a constant concentration of our drugs across both rows and columns. I n ideal world, we would expect to see no variability between the wells, as you can see here on the plate map generated by JMP. In reality, we know that some variability is inevitable. However, we really need to understand its sources and the variability needs to be controlled in order to have a precise and accurate measurement of the biological activity.
Scale really matters in these types of plate maps that can be very easily generated using JMP. H ere I created two different data sets and then visualize them using JMP. A s you can see, just by looking at the plate maps, you could see a very random pattern in response both across columns and wells. Here I'm showing you an analysis by the plate row, but then the moment you scale it up, you can clearly see that the upper essay is much more accurate and precise and have lower variability compared to what we could see in the lower part example.
JMP offers really this great graphical and statistical tools that when combined together are very useful for this type of analysis. Patterns of variability can really be non- random as well. Here, following in our example, we still have a constant concentration of our drug, and we can see the changes in both either average responses or in variability. In the top example, you can see a drift in the average response when we move across plate rows.
I n the bottom example, you see that even though the average variability across the plate, row wise, is still the same, we can clearly see that with the plate row changes, we also see a magnitude of the changes in terms of the variability within the row. T hat's something that we really need to pay a big attention to because ultimately on our plates, we always have to have certain wells where we are going to deliver our dilution series of reference material and our tested samples.
F or the plate uniformity, I'm going to present you briefly a case study with a non- cell based binding assay where we have a recombinant target. T hat's a recombinant protein which is coded on the microtiter plates. W e come up with our drug, again, in the form of dilution series. I n this case, the biologic is a monoclonal antibody. W e are able to use a detection reagent coupled with an enzyme, where after addition of the clear substrate, the substrate gets catalyzed by the enzyme in presence of our drug in order to yield a colorful product.
A s you can see, there are multiple steps and multiple washes alongside. Many binding assays would have similar set up, but they can differ. T hese differences can be, for example, in liquid volumes, incubation times, buffers, reagent conditions, and plate types used. B ecause of that, plate uniformity has really been conducted for every assay we develop. Because there are many potential sources of variability, including the operator.
We already had a workflow in place in our team to do that, but that was really cumbersome, which I hope you would be quite clear when we dive into this slide. What we had to do is manually export the plate response data from our plate reader into the Excel sheet. We had to manually rearrange the data into dilution series and then break it down per plate. We were calculating EC 50 or IC 50 in a different third party software. Those measured inflection points had to be then manually re-exported from that software back into Excel.
We had to manually arrange the data back in the Excel in a format that would be digestible for a simple Fit Y by x analysis in JMP. T ogether with Hadley, we were able to develop a JMP plugin, which is able to do everything and even more than that in just one step without the need of using any pieces of software whatsoever. A s you can see here, the whole workflow is broken down into three steps.
First is the data import into JMP. We are able to import multiple plates at once. We are able to generate plate heat maps just to visually export the data and potentially detect and eliminate outliers that would otherwise skew our data analysis and potentially lead to misconclusions. T hen we are able to look at the curve bias analysis going by plate rows or plate columns. I f anything looks potentially suspicious or we want to do a deeper dive, we fully employ JMP interactiveness. Just by hovering over a certain data point, we're able to look at the curves and that particular infliction point calculation comes from.
I hope it will be much clearer in a moment when Hadley is going to demonstrate the JMP plugin in JMP. Hadley, the floor is yours.
All right, thank you very much. Thomas and hello, everyone. I'm going to go ahead and share my screen. Before I jump directly into the ad, I'd like to show you that the ad takes advantage of a number of features in JMP. First one is the Fit Curve platform available from Specialized Modeling. How the Fit Curve works. Actually, maybe I'll show it in bio assays a bit better. How Fit Curve works is that it allows you to calculate information about your known curve shapes. For example, the growth rate, or in our case, the one that we were interested in was the inflection point, fit a variety of different types of curves. For example, the sigmoidal curves, which is what we were using here.
The other one that it makes use of is the map shapes. Map shapes tells the software to recognize a certain value in a column as a figure on a map and to reproduce that figure in Graph Builder. You can see that here. JMP comes preloaded with many default shapes, for example, country names, and there are some others as well. The microtiter wells are not included with JMP by default. If anybody is interested in using them, you can find them over on the community. I think that there's a link to this location where you can download these well shapes, these map shapes in our presentation on the same page.
Without further ado, let's jump into some of what the Add-in does. If I were to open up this example here where you could see the name of the cell, the plate that it came from, the data, and then the plate row column, and then the concentration value. The first thing we'd like to do is to visualize this. The Add-in allows us to do that by clicking on the Generate Cell Maps button where I can put in the data, the plate name, and the cell name, press OK.
Here you can see all the different map shapes and use these to very quickly identify outliers or any special thing that may be happening over the plate that you can recognize visually. Right away, we've gotten some value of this. What I'd like to say, though, is that the data, when it comes out of the machine at a day's end, I was told this in quite such a nice format, it actually looks like this. It's these CSV files, and you can see all of that information here. The very first step is to take all of this data and to get it into JMP. Of course, JMP can read CSV files, so that's not such a big problem. There it is.
Once we've done that, we need to grab the individual data from the plates themselves. There's a number of plates here. Here we can use the Add-in for some help. I'm going to grab this. By the way, for anyone watching this, if you have similar data or similar needs and are interested in developing these Add-ins yourself, you can always reach out to your local JMP team for help. They'd be happy to show you how to do these and work with you to do them as well. N ow that I've highlighted all of these, I've selected them in the table, I'm going to go ahead and import this data. N ow it's in the right format and I can go ahead and generate my map shapes.
The other thing that I can do is calculate all my EC 50 values or anything else that I'd like to calculate. In order to do that, I need to grab the concentrations. As I understand that the concentrations that are used change from example to example, so there wasn't an option to automate this. But the people running these know what they are for their specific example, so they can grab them and then import them like that. N ow what we'll do is we'll go ahead and generate all of our EC 50 values. Generate inflection point table. H ere's my data. Here's my concentration. Here's my plate, and here's my row information. I'll press OK. Now I've got a table of all of the EC 50 values.
This is very nice. I can do some analysis JMP if I want to. The other thing you'll notice is that it's created a Fit Group script here, which I can use to explore the estimates by plate row, and by plate themselves. This is plate row. If we're looking at it by row, if we're looking at by column, it would be plate column. This is very nice, but what it does is it takes advantage of another JMP feature called Graphlets, which allows you to generate or open up and run any any platform in JMP from any platform in JMP, either that platform or another platform.
In this case, we'd like to run Fit curve from our Fit Y by x Fit Group report. We can do that simply by hovering over any point that captures our interest. When we do that, we generate the inflection point. The whole curve, we can look at it, or we can dive in a little closer by clicking that button. That takes us right into Fit curve with the sample that generated that shape. Different things that we can do with this data. Perhaps one other thing that I'll show you that the Add-in can do is to explore this a little bit more closely to explore the variability from sample to sample.
We can do that by pressing this button and then exploring the standard deviation of all of the samples that in theory would be the same, but there's going to be some variability. We could see what that looks like over here. I hope you found that interesting. I hope that's given you some ideas about how you can build similar Add-ins yourself or automate processes that typically take a long time. Thomas, how long would this have taken without the use of the Add-in?
Probably something around 3-5 hours. I think the conclusions that could be drawn may not necessarily reflect the reality. Simply because certain tools were not available for us. Definitely much of a drastic change in the ways how we work. Thank you very much, Hadley, for that presentation. Just to finish off, I will just summarize everything. Hopefully, you can see my slides now.
I would just want to reiterate the benefits of the new tools. It is really a one stop- shop for plate bias and uniformity analysis with most models which are already chosen for users. That was quite important for us as well. Just to ensure that we have this full compliance within our team that the best models are being used. The plate heat map generation option allowed us to remove any of the obvious data outliers in order to focus on the true variability analysis.
Let's think for a moment that we have an outlier in one of the five plates in one of the wells in row H that not necessarily means that the variability of the whole row H across all the assay runs would always be much higher compared to the other rows. As reiterated also at the very beginning through that question that Hadley has asked, we really were able to remove the manual and error prone copy paste of the data.
What I really personally very liked is the interconnected connectivity of curve analysis with plate raw and column variability analysis. T hat's these types of graphs that Hadley has demonstrated to you. More in depth analysis of uncertainty in the calculation of the inflection point is possible. That's something that we are exploring further as well to perhaps extend that JMP Add-in even further. A ll in all, it definitely aids and speeds up development of robust potency assays in our team going forward.
T hank you very much for your attention. Both Hadley and I would be willing to take any questions just by contacting us at JMP User forum. Thank you very much.