RNA-LNP Scale-Up: From Screening to GMP Manufacturing

What LNP development actually needs from the formulation side

Preclinical development asks a formulation approach for something fundamentally different than GMP production does.
The first development step of an RNA-LNP product is screening. LNP screening is a mix of rational design and empirical testing: known lipid structure-activity relationships and ratios borrowed from approved products narrow the field early through DOE, but the link between composition, mixing, and biological performance is still too complex to predict from first principles alone. So it becomes a numbers game — a campaign spanning dozens to hundreds of lipid ratios and flow conditions, each competing for RNA (or lipids) that’s expensive and often available only in milligram quantities. What screening needs above all is the ability to make a small, information-rich batch without spending down the RNA budget on a single data point.

That means working volumes in the hundreds of microliters, fast turnaround between conditions, and fine control over flow parameters. Microfluidic mixing — laminar flow inside a fixed channel — is built for exactly this, which is why it’s the default at the development stage: not because it’s more sophisticated, but because it’s the only approach that efficiently works at screening volumes at all.

Requirements don’t jump straight from screening to production, though — they shift gradually as a program moves through preclinical development. As candidates advance from in-vitro screening into in-vivo testing, the number of formulations under test drops from hundreds down to a handful, while the volume needed per condition grows, from a few hundred microliters to milliliter-scale doses per animal. By the time a single candidate is locked ahead of IND-enabling studies and clinical supply, volumes have grown again — into the hundreds of milliliters, and eventually tens or hundreds of liters — while the number of formulations left standing has dropped to exactly one.

What LNP GMP production actually needs from the formulation side

Once that candidate is locked, the challenge changes from finding the right formulation to carrying it intact into a different reality. Clinical and commercial supply need volumes ranging from around 100 mL to tens or hundreds of liters, produced under GMP conditions, at a throughput that makes a clinical trial timeline realistic. Precision on a single 200 µL batch is no longer the constraint; consistent output across a multi-liter run is.

Microfluidic formulation of LNP

Microfluidics doesn’t make that jump easily. Its precision comes from a fixed, narrow channel (200 to 300µm) running in laminar flow — the same feature that makes it so controllable at low volumes caps how much can pass through it. Pushing more volume through means risks of clogging and fouling due to the narrow channels and the natural tendency of nanoparticles to agglomerate, extremely long runs, which kills throughput, or raising the flow rate, which breaks the formulation conditions and hence does not make the same nanoparticles. Running many chips in parallel to compensate — scaling out rather than up — is extremely challenging due to the pressure flow imbalance that can arise from clogging, lead to high chances of batch failures. Above roughly 50–100 mL, microfluidics stops being a viable production tool, not for lack of a bigger version, but because the property that makes it precise at low scale is the same property that caps its throughput.

Impingement jet mixing (IJM) and T-mixer formulation

IJM and T-mix— colliding two high-velocity streams under turbulent flow — is what the industry has converged on to meet that need instead. It is fast, it scales to continuous operation, and it is genuinely well suited to large-volume LNP manufacturing. But it comes with its own rigid requirements: flow rates typically need to sit above 10s of mL/min – meaning large minimum volume and major losses – to sustain the turbulence that makes the mixing work, it is extremely sensitive to even small flow rate changes, and — critically — the flow rate ratio between the two phases is generally locked close to 1:1, since deviating from it pushes mixing outside the chamber and degrades particle quality. Below the volumes screening & preclinical testing actually needs, IJM is not a smaller version of the same tool — it does not function at all in the regime that matters for development.

This isn’t unique to LNPs, for what it’s worth. Biologics manufacturing runs into the same problem with bioreactors: conditions inside a bench-scale vessel don’t simply carry over to a commercial-scale one [1]. “Just use a bigger vessel” is one of the more persistent myths in bioprocessing — matching a single engineering parameter across scales doesn’t guarantee the same biological outcome. RNA-LNP scale-up runs into its own version of that lesson: matching final lipid composition and N/P ratio across scales does not guarantee the same particle, because the mixing conditions that shaped it changed underneath it.

So development and GMP production are not the same job at different volumes. They are two different engineering problems, each solved today by technology that is structurally unsuited to the other’s regime.

What happens when you switch mixing technologies

So development and GMP production don’t use the same mixing technology. But what actually happens when a program switches from one to the other?

Why the same particle doesn’t survive the switch

The short answer: the same nanoparticle doesn’t reliably survive the switch. Partly because the relationship between mixing conditions and every particle properties is multiparametric rather than linear — change the flow regime and several CQAs move at once, not just one, and not in ways that cancel out. And partly because several of the properties that matter most can’t yet be measured reliably at all, which makes it hard to even confirm whether a “successful” transfer actually preserved what mattered.

“When you transition from preclinical small scale to large scale, you have to switch the mixing technologies, which costs time and money. You have to make sure that when you’ve changed the manufacturing process, you haven’t changed the drug itself — the nanoparticle, or how much of the mRNA gets into the LNP. It has to be the same.”

— Alex Aust, LNP Consultant, PharmaSource Podcast [2]

Part of why this happens is that an LNP is not a stable object you specify once and reproduce anywhere. It forms in milliseconds through nanoprecipitation, and even after formulation it stays metastable — the particle sits in a kinetically trapped state and keeps evolving. That is precisely why mixing conditions leave such a lasting mark: the process is not incidental to the particle, it is constitutive of it. Formulation method sits upstream of nearly every CQA that follows — particle size, polydispersity (PDI), encapsulation efficiency, morphology, RNA integrity — and those CQAs in turn govern biodistribution, stability, and therapeutic performance, as the figure above lays out.

It’s worth being honest about a separate, unresolved issue here: several of the parameters that matter most — RNA-LNP internal organization, RNA copy number per particle, the ratio of full-to-empty LNPs, endosomal escape efficiency — still lack standardized and simple measurement techniques. That is a field-wide characterization gap, and no formulation technology closes it on its own. What a formulation technology can address is a related but distinct problem: keeping the process itself consistent enough that whatever CQAs you can measure stay stable across scales. That’s the process-continuity gap, and it’s the one this review is really about.

“A main barrier to mRNA vaccine development and manufacturing is achieving reproducible LNP formulation and scale-up while maintaining key particle quality attributes across batches.”

— Robert Langer, MIT, via Nature Reviews Bioengineering [3]

The switch happens earlier than most teams realize

The same problem shows up in miniature even before a formal scale-up decision is made. Teams that hand-mix or pipette-mix early formulations — a common shortcut at the very earliest discovery stage — are, often without realizing it, already setting themselves up for the same disconnect:

“What many people end up doing is to just pipette-mix or hand-mix their components, which means that the LNPs you’re making are very different than the LNPs you’re going to be making farther down in your development process during scale-up. If you don’t think about that early on, you might end up optimizing for something that doesn’t end up being quite as good later on.”

— Colin McKinlay, Nutcracker Therapeutics, Pharmaceutical Technology (2024) [4]

What the data actually shows

The mechanism behind this is straightforward once you separate the two mixing regimes. As one industry analysis puts it:

“Scale-up often involves a change in the mixing technology itself — moving from microfluidic chips to impingement jet mixers or T-mixers. This shift in hydrodynamics can alter particle size distribution, encapsulation efficiency, and morphology, forcing teams to re-optimize formulations that were thought to be finalized.”

— Creative Biolabs, Scalability Challenges in mRNA-LNP Manufacturing (2026) [5]

A 2026 cryo-EM review of the field/i [6] shows this happening in practice: particles that look identical under DLS — same size, same PDI — can still produce markedly different morphologies and transfection outcomes, because it’s the mixing conditions during formation, not just the final composition, that determine what gets built inside. Switching only the mixing geometry, lipids and RNA held identical, shifts the share of one structural subtype (bleb-forming particles) from roughly 90% to roughly 30% of the population. Mixing technology alone is enough to change the particle, hence it biological behavior, its toxicity, biodistribution….

A practical benchmark makes the same point in vivo: eleven different mixing techniques applied to the same mRNA-LNP formulation (same lipids, RNA, downstream purification…) produced bioluminescence expression ranging from roughly 10% to 250% relative to hand mixing — a 25-fold spread from the mixing method alone [7].

What this costs in practice

That IJM’s flow rate ratio is essentially fixed near 1:1, as noted above, is exactly the kind of rigid process constraint that forces a full re-optimization campaign when a program makes the switch. Beam Therapeutics’ own account of its CRISPR-LNP development, shared at the 2024 NanoDDS conference in Orlando [8], put a number on it: the switch from microfluidic screening to IJM production alone took roughly 18 months of iteration in their program — before accounting for the rest of the development timeline.

The hybrid workaround: how biotech teams cope today

The scale of the bottleneck is easy to underestimate. As of mid-2026, only seven RNA-LNP drug products have reached approval anywhere in the world: four from Moderna (Spikevax, mRESVIA, mNEXSPIKE, and the recently approved flu-COVID combination mCOMBRIAX), one from BioNTech/Pfizer (Comirnaty), one from Arcturus/CSL (Kostaive, approved in Japan), and one from Alnylam (Patisiran, the first LNP drug ever approved, back in 2018) [9]. Against a pipeline of hundreds of candidates in development, that is a strikingly small number — and it keeps moving, which is itself telling: Moderna alone expects a fifth approval within months.

As introduced above, Beam Therapeutics’ 18 months CMC transfert is not an outlier; it’s close to the norm.; it’s close to the norm. In the absence of a single method that spans the full pipeline, most biotech and pharma teams have settled into a hybrid workflow by necessity rather than design: microfluidics for screening and early in-vivo work, a transition to IJM or T-junction mixing somewhere around the 50–100 mL mark (or at the end of the preclinical development), and a formal re-characterization campaign at every transition. In our experience working with formulation teams, that transition can take up to two years to complete end to end, largely because of exactly the CQA discrepancies described above — and anecdotally, something like a third of microfluidic-optimized formulations don’t transfer cleanly, requiring significant re-optimization once they leave the scale they were designed at. Neither figure is published data; they reflect what we consistently hear from teams living through the transition, not a formal study.

That repeated re-optimization is not free, and it is not just a manufacturing inconvenience — it is a direct constraint on how much confidence a program can have in its own delivery vehicle at the point it matters most.

“Even more than funding, the quality of the delivery vehicle is what is critical. If there’s any toxicity or issues with it, you have to start from scratch all over again — having an efficient, non-toxic, but also scalable and manufacturable vector is the most critical piece.”

— Fátima Bosch, Scientific Advisor, Kriya Therapeutics, at a Gene Therapy Summit 2026 roundtable [10]

Why LNP GMP scale-up is getting harder, not easier

It’s tempting to treat this as solved — mRNA-LNP vaccines already reached billions of doses. But that happened under emergency authorization, before any dedicated LNP quality framework existed. That framework is now being written, and the next generation of RNA-LNP products — repeat-dose therapeutics, gene-editing payloads, non-vaccine indications — won’t get the same latitude.

Three regulators are converging on the same ask. The European Pharmacopoeia adopted new general texts on mRNA-LNP products in 2024, in force since July 2025 [11]. The EMA has a dedicated mRNA-LNP quality guideline in development, already calling for batch-level consistency data [12]. The USP’s analytical procedures for mRNA vaccine quality have expanded twice since 2022 to add LNP-specific methods [13]. All three point to the same standard: proof that a process change hasn’t changed the product — borrowed from small-molecule pharma, and a poor fit for LNPs.

“The statement that ‘if you can show a process change does not alter the product, you do not have to re-prove everything’ may be tricky for LNPs: many potentially relevant structural features are not routinely measured — and maybe cannot even be measured yet. Consequently, demonstrating true product equivalence following a process change may be more difficult than for conventional pharmaceuticals.”

— Cédric Devos, MIT [14]

That’s exactly the requirement microfluidic-to-IJM transitions struggle to satisfy. A hybrid workflow that “worked” under emergency authorization is a much harder case to make under a framework that expects justified, batch-characterized consistency by design.

How can LNP scale-up actually move forward?

None of this means LNP scale-up is unsolvable — it means asking a different question. Instead of assuming the mixing technology has to change between screening and production, the more useful question is what a mixing approach would need to do differently to avoid that change altogether.

That is the specific gap NanoPulse, developed at Inside Therapeutics, is engineered around — not the characterization gap described earlier (no technology closes that by itself), but the process-continuity gap. NanoPulse is built on high-frequency alternating injection of two liquid phases into a shared channel, rather than on either laminar microfluidic flow or turbulent jet impingement. Because mixing happens at the interface between successive alternating fringes rather than through a volume-dependent flow regime, the same mixing conditions apply whether the total volume is a few hundred microliters (so only a few fringes injected) or several liters delivered continuously — without a flow-regime switch, and without the flow-rate-ratio constraint that locks IJM into a single operating point.

In practice, that has meant CQAs — size, PDI, encapsulation efficiency — held consistent from 0.25 mL to 60 mL, with matching in-vitro and in-vivo performance across that range.
Continuous production has been demonstrated at volumes above 4 L, and in-vivo performance has been shown equivalent to microfluidic TAMARA formulations in independent testing at the University of Strathclyde (Yvonne Perrie’s group). Against the limited commercially available options for scale-up today, NanoPulse offers a single platform spanning the full screening-to-production range, built around one principle: the mixing stays identical across scales, so the particle does too.

None of this replaces the need for good characterization; the open questions about full-to-empty ratios, bleb or not bled and endosomal escape efficiency remain open regardless of which mixing technology produced the particle. What changes is whether the particle characterized at screening scale is still the particle being made at production scale — which, for now, is the part of the problem that’s actually solvable.