Guest Column | November 20, 2026

Continuous AAV Polishing Takes On The Empty-vs-Full Problem

A conversation between Julia Müller, University of Applied Sciences and Arts Northwestern Switzerland, and Life Science Connect's Jon O'Connell

3d İnfographic circular arrows-GettyImages-2162579770

Researchers are exploring whether continuous twin column chromatography can help fix one of AAV purification's most vexing issues: separating full particles from the empty ones and keeping only the material of interest.

Julia Müller, a downstream processing scientist at the University of Applied Sciences and Arts Northwestern Switzerland, wanted to know whether a technique proven on other product classes could break the empty/full trade-off. Her team applied multicolumn countercurrent solvent gradient purification (MCSGP), a continuous twin-column method that recycles the mixed fraction onto a second column, to anion exchange polishing of AAVs.

The result is a process whose value rises with titer. Higher loading widens the very overlap MCSGP is built to resolve, making the recycling process increasingly worthwhile as upstream processes mature.

Müller answered some of our questions about the role of partially filled particles, the recycling strategy, and the control and resin-reuse questions standing between bench-scale results and manufacturing. Our conversation is edited for clarity.

AAV purity is a persistent challenge. Why is separating full from empty capsids so difficult, and why does it matter?

Müller: Beside the full variants, not only empty capsids, but also partially filled and overfilled capsids are produced. All these variants, they still can be taken up by the cells, but do not deliver a functional therapeutic cargo. So, they add up to the overall capsid dose, without really contributing to the therapeutic effect, increasing the risk for immunogenic side-effects. Accordingly, the goal is to have as many full particles in your final dose as possible.

The problem is that what is inside the capsid changes the surface charge only minimally. From the outside, they look identical and are genuinely hard to separate. This tiny charge difference can still be used in anion-exchange chromatography to separate the species, but with a normal batch gradient elution, the empty and full particles overlap. So, you are forced into a trade-off: collect only the narrow high purity fraction of your product and lose a lot of full capsids in the overlap or collect more widely but also include more empty particles. That typical purity-yield trade-off is what we aimed to break.

What led you to believe that a continuous twin-column chromatography approach would deliver better outcomes?

Müller: This continuous twin column polishing process that we’re using, called MCSGP, has proven to be very efficient for other product classes. We asked ourselves: would this also be helpful for AAVs? The core idea is simple but elegant. Instead of discarding the overlapping fraction, you recycle it onto the second column and separate it again in the next cycle.

For our AAV2 study, we focused on empty particles that do not contain the transgene. To a lower extent, you also have partially filled and overfilled particles, but we did not consider them in this first proof-of-concept study.

So, we thought: Why not try it with MCSGP, which is built to address this type of challenge? You collect only the fraction that meets your purity spec in every cycle, but the product sitting in the mixed fraction isn’t discarded but recycled onto the second column, and the product can still get cleanly separated in a subsequent cycle. You get high purity and high yield at the same time, instead of trading one for the other. The empty/full separation is essentially a textbook case for what MCSGP is built to do.

And what did the process actually deliver?

Müller: We were really pleased. In the final MCSGP pool we increased the full-capsid content from about 30% to nearly 70% over six cycles.

But the result I’m proudest of is that we broke the purity-yield trade-off directly. At a purity comparable to a conventional batch step, MCSGP recovered about 20% more full product — roughly plus-20% by ddPCR and plus-18% by cryo-TEM. On top of that we saw around 23% higher productivity and 27% less buffer consumption than batch. In other words, you’re not choosing between purity and yield anymore. You recover more of your valuable product and run the step more efficiently, all at the same purity.

Part of your design was to eliminate the strong-side recycling and reduce process time. Do you sacrifice any product when discarding the strongly adsorbed fractions?

Müller: That was a deliberate simplification, and it worked out really well. A classic MCSGP cycle recycles overlaps on both, the weak and the strong side of the product peak. In our AAV2 feed, we didn’t see any strongly adsorbing impurities nor overfilled particles. That’s then also an analytical boundary. You can only set a meaningful recycling window around species you can detect. At the strongly-adsorbing end we saw no distinguishable peak, unlike the clear peak you see for the empty particles, so there was nothing meaningful to recycle. They’re probably not completely absent, but if present, only to a very low extent. In the end, we collected the whole elution window at the tailing end, and I would say we didn’t sacrifice anything.

By dropping that strong-side recycling we rearranged the cycle into a leaner, binary separation and cut the cycle time by about 30%, with no loss of product or purity. It’s a nice example of tailoring the method to the actual separation instead of running it by default - keeping the recycling where it pays off and dropping it where it doesn’t.

The polishing study used a relatively low loading density. Is there a feed titer threshold at which this continuous process no longer outperforms batch methods?

Müller: The loading density impacts how much overlap you get, and a higher loading widens the overlap between empty and full. Therefore, I’d expect that if you load the column higher, the continuous polishing would probably be even more beneficial because you have even more of these empty particles overlapping with your full fraction. So the trend works in the method’s favor as processes scale up.

It’s inherently a high-throughput, high-value technology. Because you need two columns for MCSGP, you should have sufficient throughput to justify them. The technique is therefore best matched to the high-titer, high-value processes the field is increasingly moving toward. These continuous MCSGP processes are very productive and pay off when you have enough material to run a meaningful number of cycles and reach a steady state. If you don't have enough AAVs, it makes no sense to do it continuously.

In your work, we see peak height increase over cycles, suggesting the AAVs were accumulating. Was that intentional, or does it signal aggregation risk?

Müller: It was expected, and we understand why, but it was not intentional. If you think of different species — the partial particles charge-wise fall exactly into this recycling window. As a result, their retention time falls between the empty and the full particles, leading to their repeated recycling over time. Ideally, you would reach a steady state where the amount of particles removed equals the mass you recycle and refeed, reaching an equilibrium after a certain number of cycles. Usually, you would reach a steady state from the third cycle on. In our case, because these partial particles kept getting recycled, we did see this accumulation. But that is the nature of the recycling and not really a red flag in that particular case.

In a follow-up study, we optimized recycling and largely prevented accumulation, which also confirms this partial-particle explanation. As AAV2 is prone to aggregation, we can’t completely rule it out, but the evidence clearly points to the recycling of partial particles, and we hope to confirm that in the next few months. So, it’s a well-understood effect that we are already resolving, and even so, we were still able to deliver the yield gains.

I'm always thinking about the next steps to reach commercial scale. What are your biggest concerns when you scale this to manufacturing-size columns with longer run times?

Müller: We’re tackling that now in the follow-up study. In this first example that we published, the control of the different windows was purely time-based, which can be quite risky. With AAV2, the particle concentration, and consequently the UV-signal, was very low, which made a UV-based control hard. So, the time-based approach was the right call at that time.

In the meantime, we moved to AAV5 with a more productive upstream process that allowed us to load more material, resulting in higher UV-signals. This allowed us to implement a UV-based control. Process control could have been a constraint at a larger scale, but we’ve already solved it in that next step. Resin reuse over many cycles is the other open question, since we only tested a limited number of cycles. But this is not a roadblock, it’s the next question to answer as we move toward manufacturing.

About The Expert:

Julia Müller is a downstream processing scientist at the University of Applied Sciences and Arts Northwestern Switzerland, where her PhD thesis work focuses on continuous chromatography methods for AAV and plasmid DNA. Previously, she held downstream process development roles at Lonza, CSL Behring, and Thermo Fisher Scientific. She received a master's degree in molecular and cellular biology from the University of Bern.