Harnessing analytical chemistry in ADC development

In this webinar, our experts will discuss how we harness analytical chemistry in ADC development here at Sterling.

My name is Chris Bailey. I am a Senior QC Analytical Scientist for bioassays, and I’m presenting today with my colleague Martin Feighan, who is one of our Senior QC Scientists for biochemistry.

The webinar is divided into three parts. Initially, I will give a brief introduction of ADCs and how they work. Then Martin will talk you through our analytical method development program, and finally, I will take you through a cell-based potency assay we have developed and validated.

An ADC comprises of an antibody that only binds to one particular cell type, and a cytotoxin (the drug) that indiscriminately kills all cells it comes in to contact with. These components are linked together to create an antibody-drug conjugate that only kills one particular cell type.

ADCs target cancer cells. Cancer cells differ from normal cells in a number of ways; they grow and divide much more rapidly than normal cells, and they have distinct morphologies. In particular, there are significant differences between cancer cells and normal cells at the molecular level. Cancer cells express different antigens on the cell’s surface compared to normal cells. The normal cell on the left of the slide does not interact with the ADC because it lacks the appropriate antigen. When the normal cell transitions to a cancer cell, there is a change in gene expression which leads to a change in cell surface antigens. The ADC targets one of these cancer cell specific antigens, or at least an antigen that is expressed in much higher levels, on the cancer cell compared to the normal cell.

ADC antigen binding triggers intracellular signalling that the cell tries to shut down by internalising that part of the membrane where the ADC is bound. This receptor-mediated endocytosis creates an endosome within the cell containing the complex of antigen and ADC. The endosomes fuse with lysosomes where upon the pH drops from neutral to acidic and the ADC comes into contact with lysosomal proteases. In this environment, the antigen and antibody are denatured and degraded. There are different modes for release of the toxin from the antibody within the lysosome. The linker between toxin and antibody may be cleavable by specific lysosomal proteases, it may be acid labile and hydrolysed by acid hydrolase, or it may be uncleavable and the toxin linker releases following degradation of the antibody. Once released from the antibody this free toxin then moves through the cell to its site of action.

There are several different modes of action for the toxins currently used in ADCs, as highlighted in this table. MMAE and DM1 are examples of tubulin polymerisation inhibitors, which disrupt the cytoskeletal changes that take place during cell division, leading to cell death. DNA interacting toxins, such as the DNA cleaver calicheadmicin or the DNA cross-linker PBD dimer, prevent DNA replication, again leading to cell death. DXD is an example of a topoisomerase-1 inhibitor which prevents topological rearrangement of chromosomal DNA essential for DNA replication. This table also highlights the range of different target antigens and target conditions amongst approved ADCs. It is worth noting that the field is constantly evolving. New ADCs are being developed with toxins that are DNA alkylators or protein synthesis inhibitors, as well as some toxins with immunostimulatory modes of action. Also, with respect to the cell targeting, bispecifics are now being developed for use as ADCs. Bispecifics target two antigens on the cell surface and can be used to discriminate certain cancers. In addition, non-internalising cell surface antigens are also being selected for ADC targeting. These can be used to deliver radioisotopes to specific cell types or to deliver cytotoxins to the cell surface for subsequent extracellular release.

ADCs already targets a wide range of different receptors, utilising several different modes of action and with these continuing innovations, the range of targets and modes of action is set to keep on growing for the foreseeable future.

Some ADCs have modes of action that are independent of the toxin an example is trastuzumab and tansin, with the commercial name kadcyla. Trastuzumab, the antibody of kadcyla, blocks HER2 receptor dimerisation, inhibiting the HER2 mediated intracellular signaling that triggers cell division. When bound to HER2, trastuzumab also stimulates antibody dependent cellular cytotoxicity in which immune cells are recruited to the sites of trastuzumab HER2 complexes on the target cell surface and subsequently kill the cell. Trastuzumab, originally sold commercially as herceptin, is a successful therapy for the treatment of breast cancer.

Kadcyla is a second generation drug in which an additional mode of action, provided by the toxin, has been introduced to an already approved biologic. In contrast, brentuximab vedotin, known commercially as adcetris, only has one mode of action, that provided by the toxin. Brentuximab on its own does not have cytotoxic activity or therapeutic use. Adcetris is an example of a monoclonal antibody being converted into a therapeutic by addition of a toxin. Antibody drug conjugation can be used to increase the potency or efficacy of an existing biologic drug or can be used to confer therapeutic potential to previously clinically ineffective antibodies.

If take a look at a systemic disease, for example leukaemia, we can immediately see the critical requirement for ADCs to be specifically targeted to cancerous cells. The way to treat leukaemia is to inject the drug directly into the circulatory system. If the drug did not specifically attack cancer cells, the effect would simply be to poison the patient. This lack of specificity is why traditional chemotherapy can have serious side effects. The image in the centre of the slide shows the variety of different blood cells within the circulatory system, including erythrocytes, the red blood cells, and monocytes, lymphocytes and neutrophils, the white blood cells. When a patient develops leukaemia, there is a large increase in cancerous white blood cells within the blood vessels. These cancerous white blood cells have a high level of expression of the antigen CD25 or interleukin-2 receptor alpha. There’s a much higher level of CD25 expression in cancerous cells compared to normal cells. In particular, over expression of CD25 on regulatory and activated t-cells, is associated with acute myeloid leukaemia as well as both hodgkin and non-hodgkin lymphoma.

Here at Sterling, we have produced an exemplar CD25 specific ADC named SDE-100. SDE-100 is a hybrid ADC which we have produced to demonstrate our capability to take a novel molecule from process development through to GMP manufacture. The antibody component is taken from camidanllumab tesirine, or cami, which is the second lead candidate for ADC therapeutics. Cami comprises the monoclonal antibody HuMax-TAC and the toxin PBD, a dimer that cross-links DNA. We have taken the HuMax-TAC antibody, kindly provided by ADC Therapeutics, and linked it to MMAE, a tubulin-associating cytotoxin. By combining the antibody from cami with a different drug, with a different mode of action, we have created a hybrid with different biological and different physiochemical properties to cami.

Martin will now talk about our approach to development and validation of physiochemical assays.

Thanks, Chris. As a CDMO, there are a few routes which we can take for establishing analytical methods. The first is when the client will transfer in a basic method they may already have for their ADC. Alternatively, we have our platform methods, which are in-house generic methods we have developed over years of R&D work and have been proven to be suitable for many types of conjugated material. In both these cases, optimisation of the analytical methods will be required to make it specific for the unique properties of the client’s ADC. If the platform methods are completely unsuitable and the client has no established methods, then development will need to be started from scratch. We combine the experience of the Sterling Deeside site with literature reviews of relevant publications, and we use design of experiment studies to quickly find suitable parameters for testing and then optimise from there.

The requirement for these analytical methods is driven by the required drug substance, or drug product release specifications which is set by the regulatory bodies such as the FDA. Generally these attributes can be summarised as identity, purity, impurity, potency and safety. Each of these attributes will have a range of analytical methods that are required to be put in place. In black are the requirements for a typical antibody. For ADCs, there are additional considerations due to the conjugation of the drug to the antibody, and these additions are shown in blue. Uniquely we must monitor a property called the drugs antibody ratio.

The drug antibody ratio is the number of drugs conjugated to the antibody. A DAR of two for example mean two drugs have been conjugated. Different ADCs will have different DARS, these differences can come from the properties of the antibody, the type of linker used and the type of conjugation, whether stochastic or site-specific even with the same batch particularly when produced stochastically. The DAR of two different molecules can vary. There may be a majority of ADCs with a DAR of four, however some only have a DAR of two, while others could have a DAR of six. for manufacturing purposes monitoring and controlling the DAR is often expressed as the average DAR. This will be the average of all the DAR species produced within a batch. The average star must be consistent between manufacturing batches, as the average DAR has multiple implications for the ADC, being important for potency, as low number of conjugated drugs per antibody can reduce the effectiveness of the treatment per dose, and safety complications as well. If the number is too high the dose could be toxic as there are too many drugs attached.

The total amount of drugs conjugated to the antibody is only the first level of heterogenicity of ADCs that have been produced stochastically. The second is the positions where the drugs have been conjugated to. As can be seen here for this potentially atrial conjugated example, only DAR0 and DAR8 will have one isomer. However, for DAR2 through to 6, due to the multiple locations the conjugation can occur to, multiple isomers exist. During development of the ADC, both need to be monitored and are typically done so with HPLC-based analytical techniques which we will now go through.

Hydrophobic interaction chromatography exploits the increasing hydrophobic properties as drug conjugation increases. The antibody will have low hydrophobicity and so will elute out first. As the salt concentration decreases, the lower conjugated antibodies will then elute out and eventually the higher conjugated antibodies. What this results in is a chromatogram with distinct peaks for the increasingly conjugated species. The size of the peak corresponds to how much of that species is in the sample. Using the peak area of each we can work out how much of each species is in the sample and calculate the average DAR.

The alternative approach is to use reverse phase chromatography. Again this explodes the differing hydrophobic properties of the ADC species. This time binding to a hydrophobic column resin illusion is achieved using gradient of non-polar organic solvent due to these conditions used the heavy chain and light chains of the antibodies separate out. Addition of the hydrophobic drug influences this further with more heavily conjugated chains taken longer to elute. This can provide useful information on the conjugation site distribution, seeing how many drugs are conjugated to light chains and how many to heavy chains. As with the hick approach, we can use the peak area of these peaks to calculate how much of these species there are in a particular sample and give the overall average DAR.

While both methods can provide the average DAR, they do so in distinct ways. The hick method preserves a native confirmation of the protein species, given quantitative information of the drug load distribution including naked antibody aka DAR0. Using reverse phase reduces the sample down to the heavy and light chain subunits, but this gives quantitative information on drug chain conjugation positions. Both have drawbacks as well, with the salt gradient meaning the hit cannot be linked to mass spec for further detailed information and the reverse phase method is not suitable for linkers with disulphide bonds or acid labor bonds, due to the reducing conditions. As previously mentioned, this average DAR will be a controlled characteristic which manufactured batches expected to be consistent and so it must be included in QC testing. During development of our exemplar batch, both hick and reverse phase was used. However when taken forward to QC, we decided to proceed with only the hick method. This is due to the hick method providing the breakdown of each conjugated species, including the unconjugated antibody which must be monitored as an impurity. The drug conjugation position while useful information was not needed for a theoretical batch release of our exemplar batch.

Deciding on using the hick HPLC method, we were still required to optimise the method to ensure it was accurately provide the average DAR consistently for exemplar SD-100 batch. Generally, optimisation will require focus on one or more of the following areas; column chemistry, is the resin suitable to bind about secondary interactions, the mobile phase and gradient, will elute the different species as required, the sample prep does not influence and cause unnecessary or undesired peaks, detection wavelength settings are optimal, if a column wash is required to ensure all material is removed, and if sample loading to ensure the total amount of ADC injected onto a column is not an excess to cause protein breakthrough of unbound material. For the SD-100 project, the following had to be altered for our hick method; sample prep detection, and we needed an addition of a column wash. We’ll quickly cover why and what was done in the next few slides.

Firstly, it is important to check that the reporting wavelength is optimal for accuracy. For hick, ideally only the protein absorbance will be measured as this will be consistent. If the toxin linker has absorbance as well as the antibody, it can lead to increased peak size of the highly conjugated species as there are more linkers. This results in a higher reported value for the species and overall a higher reported average DAR than is real. For SD-100, the vedotin toxin linker has a strong absorbance at 252 nanometers, plus a lesser absorbance at 280. As can be seen on the graph, the absorbance at the toxin max of 252 has a higher relative area percentage for DAR4 and DAR6, which led to a proportional drop for DAR2 and DARR3. As the average DAR is calculated from these relative area percentages, these changes will lead to an accurate final average DAR value. In this instance, it gave an average DAR of 4.23 instead of 4.09.

Even if there is known linker absorbance of specific wavelengths, it is important to measure at these wavelengths at least once as this can allude to problems currently not considered. In this instance, collecting information at 218 nanometers, as well as 214, did demonstrate another aspect of the method that required optimising. The top chromatogram is when measured at 214 nanometers and the first peak you can see is a buffer peak. This should not be present in the 280 nanometers, however when measured, their small peak is present. The cause of this is some of the ADC, likely the low DAR species or naked antibody not sticking to the column and washing straight through. This will cause the expected peak response of the DAR0 and DAR2 to be lower than expected, and causing the overall average DAR to be underestimated. To prevent this from happening, slight manipulation of the sample was required as part of its preparation. Regardless of the starting concentration, a one-to-one dilution is required in the mobile phase. This ensures there is a hydrophobic surface present before injected onto the column, enabling it to bind and not wash through.

Now, comparing the chromatograms again, we see at 218 nanometers that the peak is no longer present, indicating all the protein is binding to the column. Simple modifications to a method like this can make a big difference in ensuring accuracy. Another simple change you can make is the addition of a column wash step after the main gradient has been run, which prevents a leftover protein accumulating on the column.

Once the optimisation of the method has been complete we then move into validation of the method to gain QA approval, to have a fully validated method for GMP testing.

Method validation is a critical activity in the pharmaceutical industry. Validation data is used to confirm that the analytical procedure employed for the specific test is suitable for its intended purposes. These results demonstrate the performance, consistency and the reliability of the analytical method. The level of validation depends on what stage of the project is at. Phase one, phase two or three, but as a generalisation we consider the repeatability of the method, its precision, its reproducibility, and its robustness as the key parameters to be tested.

Using our HIC method for the SDE-100 project as an example again, this is the full list of characteristics that needed to be tested against as part of the validation study, plus the acceptance criteria it was tested against to pass the validation. Keep in mind that this is just for one method and other analytical methods may have slightly different validation requirements depending on their purpose. We’ll just focus on linearity for now, in the interest of time, and why it is important to confirm linearity to ensure an accurate average DAR.

For more simple molecules, showing total peak area or just the area of the main peak of interest is linear across concentration ranges is enough. But to ensure average DAR accuracy for ADCs, a HPLC method must have linearity for all its peaks across concentrations being measured. If any of the peaks are not linear, then there is potential for the average DAR again to be affected if testing samples at the high or low ends of the spec. A non-linear DAR8 for example could lead to underestimation of the average DAR. This slide showing the linearity of DAR8, one of our smaller DAR peaks of SDE-100, shows the method is linear at all potential loads.

With all peaks being linear, we can therefore be confident that the overall average DAR as shown here is consistent regardless of the sample concentration. Again, linearity is just one aspect of validation, all of which must pass the acceptance criteria to demonstrate the method is suitable for its analytical requirement. This will start though with choosing the best analytical method for the requirement and ensuring that thorough development and optimisation takes place, to ensure it is fit for purpose before validation.

I’ll now hand you back to Chris, who’s going to explain how we apply this analytical method establishment process for a cell-based potency assay.

Thank you, Martin. Initially I’ll describe how we handle a new cell line and the experimental procedure for performing the potency assay, then I’ll go through our assay development steps, and finally I’ll describe the assay validation.

As described earlier, cancerous white blood cells found in leukaemia patients have a high level of CD25 expression. We selected karpass 299 cells for use in this assay which is a human non-hodgkin’s large cell lymphoma cell line, that expresses high levels of CD25. Prior to using the cells in an assay, three or four weeks are spent culturing the cells, monitoring the growth rate to establish a doubling time. It is important to establish this metric because the interaction between the cytotoxin MMAE and tubulin does not result in cell death until the cell divides or tries to divide. The doubling time for karpass 299 cells was approximately 30 hours, and roughly three doublings are required to generate a significant difference between live and dead cells that can be readily detected in an assay. So this period culturing the cells provided information for an initial drug cell incubation time in the early stages of assay development. During this initial cell culture period, we also prepared a master cell bank and two working cell banks of thought for use files. By using a bank of thought for use files, we ensure that all cells used for these assays are at the same passage number and cell density.

The assay itself is performed using a 96 well plate. the karpass 299 cells are distributed evenly in all wells of the plate. The drug is then applied in a concentration series with a high concentration at the top of the plate and a low concentration at the bottom of the plate. The drug is then incubated with the cells for five days.

After the five-day incubation period, we apply the detection reagent. For detection we are using MTS. Yellow MTS is reduced to blue purple formazan in the presence of NADPH. NADPH is continually produced in mitochondria the respiratory organelle in living cells. Therefore this reduction and colour change only takes place in live cells. In simple terms, the wells with predominantly dead cells remain yellow, the wells with predominantly live cells turn blue. After a three-hour incubation we see a drug concentration dependent colour change from yellow to blue down the plate.

So how is this colour change converted to a readout we can analyse? Blue formazan absorbs light at a wavelength of 490 nanometers. The number of live cells in each well is proportional to the amount of formazan, which is proportional to the amount of absorbance at 490 nanometers. A high level of absorbance is equivalent to a high proportion of live cells in the well. A low level of absorbance is equivalent to a high proportion of dead cells in the well. By measuring absorbance at 490 nanometer across the plate, we can generate a drug concentration dependent kill curve or potency curve with live cells shown at the top of the curve where we have a low concentration of drug, and dead cells shown at the bottom of the curve where we have a high concentration of drug. Those are the basics of how the assay works. I’ll now describe how we develop the assay.

The first stage in development is to do some research. It’s important to know all modes of action associated with the ADC. As described in the introduction, a single ADC can have several modes of action associated with the antibody. Binding to the target may have a cytotoxic effect, perhaps by preventing dimerisation and activation of growth factor receptors. Also the FC tail of the antibody may recruit cytotoxic components of the immune system via antibody dependent cellular cytotoxicity, antibody dependent phagocytosis, or complement dependent cytotoxicity. In the ADC, these cytotoxic effects may be induced in addition to the drug mediated cytotoxicity. It may be necessary to develop a suite of cell-based assays such as ADCC or CDC assays in order to test the various modes of action of the ADC. In particular, where a monoclonal antibody has been shown to have a therapeutic effect, it is important to establish that linking this antibody to a cytotoxin does not alter or abrogate those therapeutic modes of action. But today we are focusing on a single potency assay assessing cell death due to intracellular MMAE activity. A search of the literature surrounding a particular ADC and cell type is also useful for establishing some starting conditions to use in the initial phases of assay development.

An important challenge to overcome when developing a cell assay is biological variation. Sources of variation between cell lines include cell doubling time, antigen density on the cell surface, toxin sensitivity and seeding density. Therefore, when developing a new assay, it’s important to optimise incubation time, the length of time the cells are exposed to the drug, cell density, and the concentration series of the drug. The better these variables are appropriately defined during the assay development process, the better the quality and reproducibility of the potency curve for a particular assay, and the more accurate and robust the assay will be.

It is also important to be clear what the characteristics of the final potency curve must be. The criteria we look for are highlighted in the table here. Firstly the data points themselves must define a curve with the four parameters of an upper and lower asymptote, a linear portion, and an easy to Define EC50. There should be a minimum three-fold response difference between the upper and lower asymptotes, it should be possible to fit a four parameter logistic model to the data with an r-squared of greater than 0.95, indicating a good fit of the raw data to the model. Finally, in order to give confidence that the sample and reference are interacting with the cells in the same way, the linear portions of the respective potency curves must be parallel, with parallelism established by both visual inspection and statistical analysis of the curves. Assay parameters that affect these characteristics are as described previously; drug concentration series, drug incubation time, and cell seeding density. Development steps varying these parameters are shown in the next couple of slides.

This slide shows a trial of several different drug concentration series applied to cells using an initial fixed cell density and incubation time this trial demonstrated that the assay we wanted to develop was feasible as we were able to generate some decent potency curves in our initial experiments. It also helped us establish an appropriate drug concentration series for use in subsequent development steps. In particular, these curves have a good sigmoidal shape with reasonably well-defined upper asymptote, linear region and lower asymptote.

Here, we’re looking at the effects of varying incubation period and cell density. The potency curves on the left show the same drug concentration series applied to varying cell densities for a five-day incubation period. The potency curves on the right show the same drug concentration series applied to varying cell densities for a six-day incubation period. You can see straight away that the quality of the potency curves after six days is much reduced compared to the five-day incubation period. One reason for this could be that after six days so many cells have died that it becomes difficult to distinguish between live and dead cells in the assay. For this assay, a five-day incubation period was found to be optimal for generating good quality potency curves. For the five-day curves, we performed statistical analysis to determine which was the best quality curve. In particular, we examined the r-squared of each curve to determine which curve best fit the corresponding data. We also examined the AD ratio, the ratio between the upper and lower asymptotes of the potency curve. We wanted the AD ratio to be as high as possible in order to generate data that provided a clear distinction between live and dead cells. Looking at the curves for the five-day incubation, it was the curve in red which had the best r-squared value, r-squared closest to one, and the highest AD ratio. Following these development steps, the concentration series and cell density were further optimised to generate the final assay that was taken forward to pre-qualification.

Our self-potency assay does not report absolute potency of each sample, but rather we report relative potency. In each assay we test a reference standard and report the potency of each sample relative to this reference. In the example on this slide, we have the reference in blue and a control sample in purple, the central potency curve. You can see that the control sample almost perfectly overlays the reference sample. We also have a low DAR sample in light blue on the left and a high DAR sample in yellow on the right. The low DAR sample has a low ratio of drug to antibody and therefore a reduced relative potency. This can be seen here with the low DAR potency curve shifted to the right compared to control the high DAR sample has a high ratio of drug to antibody and therefore an increased relative potency. This can be seen with the high DAR potency curve shifted to the left, compared to control. This data was collected using the final optimised assay setup. This capability to distinguish between high DAR and low DAR samples is another important criteria for potency assays involving ADCs where internalisation and subsequent toxin release is the mode of action. If the assay was not sensitive to DAR, there would be a question as to how the ADC was killing the cells and whether the assay was an appropriate model for drug activity in vivo.

I’m now going to go through the validation of the assay which was performed according to ICH guidelines for a phase one validation. This slide shows the results for the linearity and range of the assay. For linearity, we had statistically significant correlation between nominal and observed relative potencies for all assessments between 50 and 200 percent relative potency. Accuracy and precision of all data points between 50 and 200 percent also passed acceptance criteria. These are shown on the next slide. Meaning that the assay has a validated range from 50 to 200 percent.

Accuracy of the assessments was determined by calculating relative bias, which compares the nominal potency of a particular assessment to the observed potency. The largest percent relative bias for assessed measurements was 3.3 percent. This was observed for the 141 percent assessment, indicating that on average the observed potency at 141 percent was 3.3 percent higher than expected. Repeatability was determined by measuring relative standard deviation across six 100 percent assessments performed in the same run. The percent RSD was 3.9 percent. Intermediate precision was determined for four assessments performed across two runs by different operators using different lots of fetal calf serum in the media and performed on different dates. The percent RSD was 4.5 percent. Robustness was determined by comparing relative bias of assessments made using two different cell banks. The largest percent relative bias was 5.6 percent. All these validation criteria, looking at assessment accuracy, variability within a run, variability between runs, and variability between different cell banks, passed the validation acceptance criteria.

We also tested system suitability criteria. We examined the r-squared for the potency curves, replicate precision of the three replicates measured for each data point in the potency curve, the AD ratio, and the parallelism of the sample and reference curves used to determine relative potency. These system suitability criteria passed the acceptance criteria set for the validation and was subsequently incorporated into the data analysis for sample testing. Data generated using the validated assay must meet these system suitability criteria in order to be valid.

Specificity of the assay was examined by comparing relative potency of the ADC, SDE-100, to the relative potency of the antibody alone, HuMax-TAC, and the NAC-quenched toxin linker, vcMMAE. The potency curves on this slide show that neither HuMax-TAX in yellow, nor NAC-quenched vcMMAE in green, were toxic to the cells across the concentration range tested. The HuMax-TAC control confirms that the toxin is required to kill the cells. NAC-quenched vcMMAE is a charged form of the MMAE toxin. The lack of activity for this control shows that there is little to no extracellular processing or passive uptake of the drug by the cells, demonstrating that uptake of the ADC in the assay is antibody mediated. Another type of specificity control not performed here would be to run the asset using a cell line that does not express the target antigen.

These validation slides have shown that their assay meets criteria for a validated phase one assay described by the ICH guidelines. The assay is sensitive to DAR. The assay is also specific to the activity of the ADC. Neither antibody alone, nor toxin linker alone, exhibit potency. We have a robust system suitability criteria that are applied to sample testing.

That is the conclusion of our presentation. Thank you for listening and we would be happy to take any questions.