In this webinar, Senior Process Development Scientist, Alisdair Hearn outlines the common approaches to removing residual toxins from antibody drug conjugates (ADCs), how the scale and phase impact the options available, and discuss a case study on the emerging technology of carbon purification.
The development of an ADC can be split into three broad stages – R&D, process development and manufacturing. Each of these stages has different goals and therefore different requirements for the control of an ADC’s quality attributes.
At the R&D stage, we want our drug to antibody ratio to be about right, we want a good level of monomeric purity, and we need to have minimal residual toxin, payload, payload-linker.
This material will be going to studies to determine antigen binding cell killing, efficacy and tolerability. The payload may be novel, or it may be a well understood payload being used in a different way.
As we move from R&D into process development, we’re increasing the control of our key quality attributes, as the goal becomes to establish a repeatable and robust process.
Target DAR specifications become tighter, monomeric purity requirements increase, and the methods for removal of residual impurities, such as the toxin and solvent, need to be developed with scalability in mind. By the time an ADC reaches the manufacturing stage, it needs to have established absolute control of all quality attributes with repeatable and robust process. As I talked through a few of our approaches to toxin clearance, we need to keep in mind the stage in the process and the goals for the project.
To compare the purification of options available in the ADC field, we needed some model conjugates. We selected models for this work that are commercially interesting and have different challenges when it comes to residual payload clearance. We looked at three payload classes.
MMAE, an auristatin based payload used successfully in Adcetris, Polivy, and Padcev. MMAE is typically considered an easy to clear payload.
The industry has had a lot of interest in camptothecin derivatives since the success of Enhertu with deruxetecan and Trodelvy with SN38. For this work, we selected DXd. DXd provides us with a great model for a hard to remove residual payload. PBDs, in this case we use SG3249, have seen commercial success in Zynlonta, and typically fall somewhere in between auristatin and camptothecin in terms of ease of residual toxin removal.
Having selected the payloads we wanted to showcase, we needed to make some ADCs. Conjugates were prepared with a Humax TAC mAb by partial reduction for MMAE and SG3249, or full reduction for DXd. This is followed by alkylation with an excess of the payload. The target DAR for each was selected based on existing commercial ADCs and our own experience with these payloads.
We use standardised reverse phase, or hydrophobic interaction HPLC analytical methods to calculate DARs, and size exclusion chromatography to confirm the conjugates were sufficiently monomeric.
The residual payload limit as a quality attribute evolves as an ADC progresses through the development stages. For our model conjugates, we set an R&D appropriate limit of 5% residual as a percentage of the total free and bound payload.
By the time an ADC reaches the clinical stage, the residual payload specification will be scientifically informed. A standardised approach was applied to all three payloads. This removes the need for time consuming analytical development when working with R&D scale materials, and provides a basis for optimising the assay during process and analytical development programs. The protein and bound payload is crashed with a methanol salt participation, and the supernatant was analysed by reverse phase HPLC. The method has a limit of quantification of 1 to 2% for dilute samples, and around 0.2 to 0.5% for more concentrated samples.
In this talk, I’ll cover the most common method for residual payload removal. That’s desalting, size exclusion chromatography, bind./elute chromatography, and tangential flow filtration. I’ll discuss the benefits of each, the limitations, and then show some of our data on an emerging alternative using activated carbon.
I’ll start with a look at desalting columns and size exclusion chromatography. At an R&D scale, many protocols recommend using desalt in columns. Our data in this graph here shows that a single pass through a desalt column had minimal impact on the residual payload in all three conjugate models. This ties into our prior experience that, while very useful for buffer exchanging, desalting columns are typically not suitable for residual payload purification.
Size exclusion was much more successful at removing residual payload. In all model conjugates shown here, the residual payloads were reduced to below or close to the limit of quantification. The typical use for SEC at the R&D stage of ADC development is to improve monomer. Given that slightly salty buffers required for SEC are often suitable for the end use of R&D materials, SEC chromatography may be the only processing step required post conjugation at small scale.
A less commonly employed benefit of SEC is the fine tuning of DAR. The graph shown here shows SEC purification of our HuMax TAC-vcMMAE model conjugate fractionated the elution peak and measured the DAR. Higher DAR species eluted earlier than the lower DAR species. This doesn’t always work, but if you need to make slight adjustments to the DAR of a conjugate, it may be worth looking at.
Although undoubtedly useful at the R&D scale, size exclusion chromatography columns can only be scaled up to a point and can become prohibitively expensive. As such, it is rare to find manufacturing scale processes using size exclusion chromatography.
Now I’ll look at bind and elute loop chromatography. We used cation exchange chromatography to purify our model conjugates, but the principles are broadly applicable to other bind elute methodologies.
We adjusted the PH for the crude conjugates to facilitate direct binding of the sample to the column. The residual payloads didn’t bind under these conditions in any of our three model conjugates. Shown here is the AKTA trace for the HuMax SG3249 conjugate. You can clearly see the residual payload eluting off the column in the no salt loading buffer.
We then elute off by increasing the salt concentration. We use the same gradient on all three model conjugates. In all three cases, the ADC eluted with only minimal amount of salt required. Like SEC in the R&D stage, this slightly salty buffer is often suitable as a formulation and thus avoids the additional purification steps. At larger scales and with a more developed process, a step elution is typically preferable to a gradient, and an additional buffer exchange step would be included post purification.
Bind elute strategies such as ion exchange protein A or G, hydrophobic interaction, or CHT chromatography are especially helpful for stubborn residual payloads where additional wash steps, particularly solvent containing washes, can be added as needed. Bind wash elute chromatography methods are typically scalable, although some may become cost prohibitive due to resin and hardware costs.
TFF is the gold standard for ADC manufacturing. It is used in all current manufacturing processes to clear residual impurities, including the residual payloads. The graph shown here is taken from a Seattle Genetics presentation at the fourth ADC World Summit in 2012. They compared the clearance of residuals in the Adcetris process against the ideal clearance, that is, the rate of clearance if we achieve 100% pass through.
They showed that while the cosolvent and water soluble impurities were cleared ideally by TFF, the residual payload, in orange, deviated significantly from the ideal clearance curve. This non-ideal clearance increased the difiltration volume requirements even for a toxin considered easy to remove, like MMAE.
So how do our model conjugates perform? Remember here that all three payloads are used in commercial products, with TFF as a purification step.
Our MMAE data is similar to the data of Seattle Genetics showed. It clears well enough, but requires more dye volumes than the ideal. The PPD, SG3249 clears worse than MMAE. Whilst the clearance of is really poor, this is a trend observed in other payloads we’ve worked on with camptothecin payloads, typically exhibiting especially poor clearance by TFF, though there are always exceptions.
Improving residual payload clearance by TFF can be approached in a few different ways. You can screen cassette; the architecture, membrane and supplier can all have an impact. The graph here shows a comparison of roughly equivalent cassettes from two suppliers, purifying the DXd and vcMMAE conjugates. While there is an observable difference between these cassettes, the improvement is minimal and the development cost test multiple cassettes is high. For most processes, the improvements won’t be significant enough to justify the cost and time commitment.
Excipients or process aid, such as maintaining some cold solvents or adding sugars, can have a major impact on improving the clearance of residual payload during TFF. Unfortunately, the cost of adding process aids or maintaining excipients through the TFF is often prohibitive at manufacturing scales.
It’s also worth bearing in mind that process aids, added to improve TFF clearance, will need to be removed by additional TFF, and the improvements may be lost versus just extending the diavolumes in the first place. When optimising TFF parameters, keep in mind that the conditions best for flux are not necessarily the best for residual payload clearance. The formation of gel layers and payload interactions will influence clearance rates.
When screening TFF parameters costs can be high, and this is a situation where experience is key to making sure that resources can be applied in a most efficient way. The majority of ADC processes include a step following conjugation, where excess residual payload linker is quenched with the addition of a small molecule. For conjugates, using the common maleimide chemistry.
The most common addition is N-Acetylcysteine. This paper by Nadkarni, however, shows that NAC may have a negative impact on the clearance of some payload linkers, and that other options, in this case they use cysteine, improved clearance much more substantially.
Now we’ve covered the current standard practices. I’m going to move on to an emerging method for residual payload removal. Some of the earliest available examples of this method come from Seattle Genetics work with activated carbon for the purification of PBD based ADCs. Here you can see the patent filed in 2013.
More recently, Gilead presented data at world ADC San Diego in 2024, showing their work developing carbon filtration processes for the removal of residual impurities in Trodelvy, a deruxtecan based ADC. The published work for both of these companies ties in well with our own observations and experience over the past decade or so, utilising activated carbon in R&D and process development settings.
This is the purification of our exemplar conjugates. Here we use a standardised batch mode protocol we developed for the purification of DXd payloads at R&D scale. Crude samples were first passed through a desalting in column to buffer exchange into formulation buffer without any excipients. They were then incubated with powdered carbon. At this scale, the carbon is removed by filtration through a syringe filter.
You can see the excellent depletion of residual MMAE and DXd in this graph here. The residual PBD was reduced in this experiment as does not meet specification. As I mentioned on the previous slide, the Seattle Genetics patent was the purification of PBD with carbon, and our own experiences that carbon can be used very successfully with PBDs. The takeaway here is the different payloads can require different conditions to maximise clearance using activated carbon.
The batch mode protocols we use for R&D materials use powdered carbon, and as such, the optimal ratio of carbon to ADC needs to be determined. Here we look at a camptothecin based payload; less carbon to ADC at this end, more carbon to ADC at this end. You can see in blue that less ADC to carbon means less residual payload but there is a compromise of the yields. ADCs are generally pretty poorly absorbed to carbon, but there will be some and it will vary between payloads and antibodies. I should point out that the step yield here is still around 90%, for R&D materials that’s not bad. The other key considerations when developing this process are the carbon source and contact time, which needs to be optimised on a case by case basis.
For the last few slides, I’m now going to move away from our model conjugates. This is a case study of a novel camptothecin-based payload. The toxicology data required a 0.5% limit on residual payload to bound payload, shown in the blue line on this graph. As is typical for this type of project, we started with a program to develop a TFF protocol. However, as you can see from the clearance data here, the residual payload was not reduced to a sufficient extent despite extensive diafiltration.
With TFF proving to be non-viable, we switched our approach to focus on alternative purification strategies. Small scale work had shown that incubation with activated carbon powder cleared the residual payload extremely well, but we needed to find a scalable option that could be used in manufacturing.
Powdered carbon would leave the question of how you ensure it’s removed from the drug substance. To work around this, we used a commercially available cartridge where the carbon is bound to a membrane.
The final data I’d like to show you is a comparison from the case study we’ve been discussing. In this reverse phase HPLC trace from our residual payload assay, the blue trace is crude conjugate, the red trace is from the optimised TFF protocol, and the green trace is from an optimised process using a normal flow carbon filtration protocol with a short TFF for buffer exchange at the end. The different peaks, which correspond to impurities, cleared at different rates in the base protocol, but are completely removed by the carbon.
There are a lot of commercially available sources of carbon. These include powdered, carbon filters, and scalable cassette based options. These all have their uses, but for anyone wants to start using carbon in their own labs, I would give a caveat that we found not all carbon to be created equally.
Just want to give a quick overview of what we’ve talked about in this presentation. In summary, the G25 desalting columns did essentially nothing for residual payload removal. Prep SEC was effective, requires very little in the way development, and can be run at small scale with minimal material, but it’s just not scalable to GMP. Bind elute chromatography is effective, scalable, and GMP-able, has additional polishing effects depending on the type you use. But the development effort can be high, though good scaled down models and experience can reduce the development effort significantly.
Loose carbon. The effectiveness does vary but is typically very good. I’ve not mentioned this yet, but we see in some cases carbon can also be used to remove endotoxin. The scaled down models may be a development effort and GMP requirements low, but there is a question of whether this will be useable in a GMP setting. A carbon filtration, like the loose carbon, is effective with good scale down models, but the development effort is higher and more material is required. TFF effectiveness obviously varies, but as we’ve seen from all of the approved ADCs, it is clearly GMP-able.
Thanks for listening to this talk on residual toxin removal from ADCs. If you want to discuss anything in this talk, please do feel free to reach out to me.
