“When we go for a larger volume, everything changes — and most companies only discover that during the purification step, when it’s too late.”
— Kinkini Roy, PhD, Associate Director, Drug Product Development, Aviceda Therapeutics
Kinkini Roy, PhD, is Associate Director of Drug Product Development at Aviceda Therapeutics, with over a decade of experience leading LNP formulation, process scale-up, and technology transfer to CDMOs across nanomedicine, injectables, and RNA-based therapeutics.
The RNA therapeutics pipeline is expanding faster than manufacturing infrastructure can absorb it. As biotechs race toward IND-enabling studies and Phase 2 batches, the gap between what works in a microfluidics lab and what survives a CDMO’s TFF system at scale is slowing down programs. Kinkini spoke with PharmaSource about where those failures originate and how to avoid them.
You’re Not Transferring a Molecule. You’re Transferring a System.
Most small molecule technology transfers involve moving a defined chemical entity to a contract manufacturer who can reproduce a known process. LNPs are different.
“In small molecules, the process is very much just small-scale to larger scale. But in this case, the mixing speed we can achieve in small-scale microfluidics is not possible for large-scale mixers — and that needs to be optimized from the beginning.”
What Kinkini describes is a supramolecular system: a carrier cage containing an API (or drug substance), potentially with additional surface modifications. Every element of that system — particle size distribution, surface charge, RNA stability — is sensitive to process parameters that change non-linearly with scale.
The mixing step and purification via tangential flow filtration (TFF) are pressure points. In a lab setting, dialysis or small-scale TFF can produce a clean, high-performing LNP with manageable hold times. At manufacturing scale, that hold time can increase exponentially, and the resulting changes to distribution pattern, surface modification, and RNA stability may not surface until stability testing months later.
“Maybe we can find a formulation that is best in class for efficacy. However, when we are purifying on a larger scale, it cannot survive the longer hold time. Then it will not function.”
Specification setting cannot happen after scale-up. It has to be part of the formulation design from the first experiment.
The CDMO Evaluation Question Most Biotechs Don’t Ask
‘Can you manufacture LNPs?’ is not a useful question. Every CDMO with relevant equipment will say yes. What Kinkini recommends instead is asking for evidence of failure, and listening carefully to how the answer is framed.
“Ask the CDMO: Can you give me an example of when you started a very novel delivery system and failed on the first scale-up? From there, you will understand how they navigate failure and what solutions they bring. That portion tells you everything.”
It’s a behavioral question as much as a technical one. CDMOs that can’t describe a specific failure with a clear account of what they learned and how they adapted are either inexperienced with novel modalities or not being candid. Both are problems for a program at the clinical stage.
Beyond this, Kinkini flags a set of questions that biotech teams frequently neglect during CDMO selection: Does the CDMO have existing relationships with the lipid suppliers you’re using? Who holds the license on the ionizable lipid? If you’re performing surface conjugation or functionalization, at which step does the CDMO plan to do it — and do they understand how that choice affects product specification?
The Biggest Formulation Mistake: Treating Purification as a Later Problem
Kinkini explains that the decision most likely to cause manufacturing problems downstream is handcrafted purification at a small scale, with no parallel assessment of what happens at the TFF scale.
“We do the dialysis or small-scale TFF and get a very pure product, and say this is the highest efficacy. However, at manufacturing scale, the distribution pattern changes — and that changes efficacy, changes stability, and the agency will not approve it.”
The corrective approach she describes involves two parallel workstreams. First, specification setting with appropriate width — not so narrow that normal manufacturing variation triggers an OOS, not so broad that biologically meaningful changes pass undetected. Second, a rationale that connects the specification range to actual biological performance: if the distribution widens during scale-up, how much does that affect transfection efficiency or target engagement?
On the process side, the recommendation is to move to TFF and mixer alternatives as early as the lab allows. If RNA cost makes small-scale TFF impractical in early discovery, that’s understandable, but the transition should happen quickly, and the data generated should inform specification setting rather than be bypassed until the CDMO asks for it.
Kinkini also points to jet mixing as an underutilized option that many RNA biotechs haven’t yet adopted in-house. CDMOs are increasingly equipped with JetMixers, but biotechs developing formulations with microfluidics don’t always generate data that translates cleanly to a JetMixer process.
“The CDMO needs to communicate with R&D and biotech hubs: you guys can go to jet mixing early when setting up the lab. Many RNA companies still don’t know that this mixing technology is available from the very early stage.”
Communication Gaps That Don’t Show Up Until a Batch Fails
Kinkini’s most specific frustration isn’t with CDMO capability per se, but with the gap between what the CDMO’s instruments can detect and what’s actually changing in the product.
“There is no instrument available to detect that it is changing. We will only understand that it changed after a couple of months, when the stability falls, and by then it is very difficult to communicate with the CDMO that a failure is happening.”
This is a structural problem in advanced delivery technology transfer. The analytical methods that define product specification were often developed for established modalities. For LNPs and other novel nanoparticle systems, the specification parameters themselves may not capture all the attributes that matter for performance, and CDMOs, who are knowledgeable and experienced, may not know that their standard setup is introducing drift that the instruments aren’t measuring.
Her proposed fix is regular technical meetings at the director level — not SVP to SVP, but the scientists and engineers who are actually running the process on both sides. Those conversations, held early and consistently, are where potential drift gets flagged before it becomes a batch disposition decision.
