The last decade of LNP platform development has been, by almost any measure, a hepatic delivery story. The ionizable lipid designs that carried COVID-19 mRNA vaccines into hundreds of millions of arms — and the siRNA formulations that silenced PCSK9 and TTR in the liver — were engineered around a specific biological context: fenestrated endothelium, ApoE-mediated hepatocyte uptake, and a pKa window optimized for the pH of hepatocyte late endosomes. Those assumptions are baked deeply into the canonical formulation rationale. They are also largely incompatible with delivering cargo to the central nervous system.
This article is not a criticism of liver-targeted LNPs. They are exceptional at what they were designed for. The problem arises when teams attempt to repurpose them for CNS programs — or when platform claims don't distinguish between hepatic efficiency and CNS utility.
The Three Assumptions That Break Down at the BBB
1. Fenestrated endothelium and hepatic sinusoidal access
Liver sinusoidal endothelial cells are fenestrated — they contain pores of approximately 100–150 nm in diameter that allow nanoparticles to pass directly from the bloodstream into the space of Disse, where hepatocytes are accessible. This architecture is the reason LNPs accumulate preferentially in the liver after intravenous administration. The LNPs don't "target" the liver in any active sense; they fall through the fenestrae by passive filtration.
Brain endothelial cells are the opposite. The cerebrovascular endothelium that forms the blood-brain barrier is characterized by continuous tight junctions — claudin-5, occludin, and ZO-1 complexes that seal the paracellular space to molecules larger than roughly 0.5 nm. There are no fenestrae. A 90 nm LNP cannot cross the BBB endothelium through passive mechanisms, period. Any formulation that relies on passive access — which is to say, any formulation whose liver tropism depends on passive fenestral filtration — will not reach brain parenchyma in meaningful quantities after systemic administration.
This distinction is not a detail of scale — it is a categorical difference in access mechanism. In vivo biodistribution studies of standard LNPs consistently show brain-to-liver delivery ratios below 0.01:1 after intravenous dosing, with the brain signal often indistinguishable from background fluorescence at therapeutic doses. The liver signal, by contrast, is typically the dominant fluorescence peak across the whole-body imaging. This is what passive hepatic access looks like in practice — and it is the starting point that CNS formulations must overcome, not assume away.
2. ApoE-mediated hepatocyte uptake
After an LNP enters the liver sinusoid, its surface adsorbs serum apolipoproteins, particularly ApoE. This protein corona is not an artifact of poor formulation — it is the intended mechanism. ApoE presentation on the LNP surface drives recognition by the LDL receptor (LDLR) and related receptors on hepatocytes, mediating endocytosis. The interaction has been called "endogenous targeting" because the liver naturally expresses high levels of LDLR family members, and ApoE is abundant in serum.
The CNS context differs in two important ways. First, the BBB endothelium is not the target cell — it is the barrier that must be traversed. LDLR is expressed on brain endothelial cells, but LDLR-mediated endocytosis does not reliably result in transcytosis to the parenchymal side; much of the material is recycled back to the luminal surface or degraded. Second, the neurons that are the ultimate editing target express substantially different receptor profiles than hepatocytes. ApoE is produced within the CNS by astrocytes and microglia, but neuronal ApoE receptor expression and its contribution to endosomal uptake differs from the liver context in ways that have not been adequately characterized for LNP delivery purposes.
There is a further subtlety worth noting. Serum ApoE circulates at concentrations of roughly 30–60 µg/mL in healthy adults, meaning it is abundant enough to rapidly saturate LNP surface adsorption sites within seconds of IV injection. A liver-optimized LNP that relies on this ApoE corona for hepatocyte recognition is therefore dependent on circulating ApoE abundance — a physiological variable that changes with patient metabolic state, genotype (APOE ε4 carriers have different ApoE levels and kinetics), and disease. Designing CNS delivery to rely on a different, more controllable surface chemistry is partly motivated by the need to escape this dependence on a variable serum protein.
3. pKa selection and endosomal escape in neural cells
Ionizable lipid pKa is one of the most intensively studied parameters in LNP design. The working model is well-established: ionizable lipids should carry near-neutral charge at physiological pH (pH 7.4) to minimize non-specific interactions and extend circulation time, then become protonated in the low-pH environment of endosomes to destabilize the endosomal membrane and release cargo to the cytoplasm.
The canonical pKa range for liver-optimized LNPs is 6.2–6.5. This range is calibrated to the pH at which hepatocyte late endosomes acidify — typically around pH 5.5–6.0 — and to the kinetics of ApoE-mediated endosomal entry. In neural cells, the endosomal pathway differs. Neuronal late endosomes may acidify to pH values that differ from hepatocytes, and the kinetics of endosomal maturation in post-mitotic neurons are slower. Our own screening work has indicated that ionizable lipids performing optimally in Hep3B or HepG2 cell assays consistently underperform in primary cortical neurons and SH-SY5Y differentiated cells. The pKa shift is modest — we are working in the 5.8–6.2 range — but the effect on endosomal escape efficiency in neural cells is not modest.
The galectin-8 (Gal8) recruitment assay illustrates this concretely. Gal8 is a cytoplasmic lectin that is recruited to damaged endosomal membranes, providing a quantitative readout of endosomal disruption events. When we run Gal8-GFP reporter assays in SH-SY5Y differentiated cells versus HepG2 cells using the same ionizable lipid at pKa 6.4 (a standard liver-optimized value), the neuronal Gal8 puncta count per cell drops by roughly 60–70% compared to the hepatocyte line. The same lipid at pKa 6.1 narrows this gap significantly. This is not a trivial optimization — it translates directly to editing efficiency in post-mitotic cells where there is no cell division to dilute unescaped cargo.
What "Repurposing" Actually Costs
When a gene therapy team evaluates an existing LNP platform for a CNS indication, the conversation often starts with encapsulation efficiency and particle size. These metrics look similar regardless of whether the formulation is liver-optimized or not — and that is precisely the problem. An LNP with 80% sgRNA encapsulation efficiency, PDI of 0.12, and 85 nm z-average diameter may fail completely at the BBB for reasons that standard quality metrics do not reveal. The characterization panel does not measure BBB transcytosis, neuronal uptake, or endosomal escape kinetics in neural cell lines.
The practical consequence for CNS gene therapy programs is that teams spend months running in vitro experiments with cells that are not the actual barrier (often using HEK293 or HeLa transfection controls), conclude that the LNP "works," and then encounter BBB failure when they move to in vivo CNS models. The BBB problem was never interrogated by the assay panel.
Consider a scenario that is realistic for an early-stage gene therapy program targeting HTT knockdown: the team encapsulates a CRISPR-Cas9 RNP targeting the mutant HTT allele into a commercially licensed ionizable lipid system with an EE of 82%, PDI of 0.14, and z-average of 91 nm. Transfection in patient-derived iPSC-derived cortical neurons shows 45% indel frequency — encouraging. The team scales to a mouse HTT knock-in model, delivers IV at 2 mg/kg, and sees no detectable HTT reduction in striatal tissue. The liver, meanwhile, shows >50% HTT knockout efficiency. The LNP reached the liver perfectly. The BBB stopped it completely. No aspect of the original characterization panel predicted this outcome — because no aspect of that panel interrogated the BBB.
What CNS-First Formulation Actually Requires
Building an LNP for CNS delivery from the formulation up means starting with different design constraints than liver delivery:
- pKa screening against neural cell models: Primary cortical neurons and differentiated SH-SY5Y cells, not hepatocyte lines. Endosomal escape efficiency measured by the Gal8 assay in the correct cellular context, not inferred from hepatocyte performance.
- Surface engineering for BBB transcytosis: Active targeting to receptors expressed on brain endothelium that are competent for transcytosis — TfR1 (transferrin receptor 1) and LRP1 are the most characterized. Without active transcytosis, systemic LNPs do not reach the brain parenchyma in therapeutically relevant quantities.
- Protein corona management: In systemic circulation, LNPs adsorb a protein corona that determines their fate. CNS-targeting ligands must retain function after corona formation. This requires surface engineering that is compatible with circulation — not just with naive in vitro binding assays run in serum-free conditions.
- Helper lipid and cholesterol optimization for neural membranes: The lipid composition beyond the ionizable component — helper lipid (DSPC, DOPE, or analogues), cholesterol content, and PEG-lipid mole fraction — affects membrane fusion efficiency and cargo release in a cell-type-dependent way. Neural cell membranes have distinct lipid compositions (higher sphingomyelin, ganglioside content) that interact differently with LNP formulations than hepatocyte membranes.
- Characterization of the full CNS delivery path: Size, PDI, and EE are table stakes. A CNS formulation needs additional characterization: transwell BBB model uptake with TEER validation, apparent permeability (Papp) quantification, neuronal transduction efficiency in primary cortical neurons, and endosomal escape measured in neural cells.
The False Comfort of High Encapsulation Efficiency
Encapsulation efficiency measured by RiboGreen assay tells you how much of your nucleic acid cargo is inside a lipid particle. It does not tell you whether that particle can cross the BBB, whether it will escape endosomes in neurons, or whether the cargo will be intact after endosomal processing. A 90% EE LNP that delivers no functional cargo to neural cell nuclei is a 0% functional LNP, regardless of what the RiboGreen plate reader says.
The field has become comfortable conflating formulation quality (size, PDI, EE) with delivery quality. For liver delivery, this conflation is partially justified because the downstream biology — hepatocyte uptake via LDLR, endosomal escape — follows somewhat predictably from formulation parameters. For CNS delivery, the relationship is broken at the BBB.
There is an analogous danger with cryo-TEM morphology data. A cryo-TEM showing well-formed, electron-dense particles with uniform spherical morphology is visually compelling and confirms that the formulation process worked. It says nothing about surface ligand functionality, receptor binding capacity after protein corona formation, or transcytosis competence. Cryo-TEM of an LNP with non-functional targeting ligands looks identical to cryo-TEM of one with fully functional ligands. The structural assay cannot substitute for the functional assay.
N/P Ratio and Its Underappreciated Role in Neural Cell Compatibility
The N/P ratio — the molar ratio of ionizable lipid nitrogen atoms to nucleic acid phosphate groups — is a formulation parameter that receives less attention in LNP literature than pKa, but matters considerably for CNS applications. At high N/P ratios (above approximately 5:1 for many ionizable lipids), cationic character at physiological pH increases, which can trigger non-specific cell membrane interactions and endosomal routing toward degradative lysosomes rather than productive cytoplasmic release. Hepatocytes tolerate a broader N/P range; neurons are more sensitive to excess cationic character, likely because of their high membrane potential and distinct endosomal trafficking dynamics.
During microfluidic mixing — the standard manufacturing method for LNPs — N/P ratio is set by the relative concentrations of lipid and nucleic acid in the mixing streams. Optimization of N/P for neural cell applications typically sits in the 3:1 to 4:1 range for CNS-targeted formulations, lower than the 4:1 to 6:1 ranges commonly used in liver LNP work. This is not a universal rule — different ionizable lipid chemistries have different N/P optima — but it is a parameter that CNS programs should explicitly optimize rather than inherit from hepatic formulation protocols.
A Note on Intracranial Injection as a Workaround
Some CNS gene therapy programs bypass the BBB entirely using intracranial stereotaxic injection, intrathecal administration, or convection-enhanced delivery. This is a legitimate approach for certain indications and has produced meaningful clinical data for AAV-based CNS gene therapies. But it is not a scalable solution for the full landscape of CNS rare diseases — particularly for diseases with diffuse neuronal involvement (Huntington's disease, CLN3 Batten disease), late-onset progressive conditions requiring periodic dosing, or programs where the disease-relevant cell population is distributed across multiple CNS compartments. Systemic delivery that can cross the BBB is not an academic exercise; it is the path to treating diffuse CNS disease at scale.
The point is not that intracranial delivery is wrong for programs where it fits. The point is that a CNS gene therapy delivery field that only works when you drill through the skull is not a delivery solution for most CNS indications. Intrathecal delivery reaches the spinal cord and, with some spread, portions of the brain — but achieving widespread cortical and deep nuclei distribution requires either very high CSF volumes or repeat dosing schedules that are themselves demanding. Systemic IV delivery that can cross the BBB addresses the distribution problem at the source, if the BBB crossing mechanism can be made reliable.
Where This Leaves the Field
The current situation in LNP CNS delivery is not primarily a chemistry problem — the ionizable lipid chemical space is large enough to find CNS-compatible candidates. It is primarily a design-framing problem. Platforms built for hepatic delivery are evaluated by hepatic delivery metrics, and those metrics do not interrogate the barriers specific to CNS access.
CNS-targeted LNP programs that take the BBB seriously as a design constraint — not a problem to be addressed later or avoided via injection — require formulation work, surface engineering, and characterization protocols that are different from what the mainstream LNP platform space currently provides. The ionizable lipid pKa needs to be re-optimized for neural endosomes. The surface chemistry needs to encode active BBB transcytosis. The characterization panel needs to include functional CNS assays, not just biophysical quality metrics. None of this is conceptually difficult. All of it requires starting from a CNS-first design frame rather than adapting a hepatic frame.
