Gene therapy programs targeting the CNS have two primary vehicle options: adeno-associated virus (AAV) and lipid nanoparticles (LNPs). These platforms are not interchangeable. They differ in cargo capacity, immunogenicity profile, redosing potential, manufacturing complexity, and — most importantly for the CNS — mechanism of delivery across the blood-brain barrier. Choosing between them, or evaluating whether a hybrid approach makes sense, requires a clear-eyed comparison of what each platform actually does well and where each has unresolved limitations.
This article is not a promotional comparison. Biopathio is an LNP company, but the goal here is to present the technical landscape accurately. AAV has a clinical track record in CNS that LNPs do not — that is a fact that belongs in any honest comparison. The question is whether that track record translates to programs targeting the full spectrum of CNS rare diseases, particularly those requiring systemic delivery, large cargo, or the possibility of redosing.
AAV for CNS Gene Therapy: Genuine Strengths
Established CNS tropism and clinical data
Several AAV serotypes — particularly AAV9, AAVrh10, and PHP.B — have demonstrated CNS biodistribution after systemic administration or CSF delivery in rodents, non-human primates, and in some cases humans. AAV9 crosses the BBB after intravenous administration in neonatal mice and primates (less efficiently in adults), and intrathecal delivery of AAV9 achieves broad spinal cord distribution. The clinical approvals of AAV-based CNS therapies — including onasemnogene abeparvovec for SMA (spinal muscular atrophy) delivered intravenously — provide proof that AAV can reach motor neurons from the bloodstream in pediatric patients.
This clinical validation is meaningful. It means the AAV9 CNS delivery mechanism is not hypothetical — it has been demonstrated in humans, with efficacy signals in a serious neurological disease indication. No LNP CNS delivery platform has this record. Any comparison that does not acknowledge this asymmetry is not a serious comparison. PHP.B achieves even more efficient CNS transduction than AAV9 in adult mice, but has shown lower CNS delivery efficiency in non-human primates — an important reminder that rodent models do not reliably predict primate CNS delivery, a caveat that applies equally to LNP CNS data.
Long-duration gene expression from episomal DNA
AAV vectors delivered to post-mitotic neurons can provide years of transgene expression from episomal DNA that does not integrate into the host genome at significant frequency. For loss-of-function conditions requiring stable protein replacement — spinal muscular atrophy, CLN3 Batten disease, AADC deficiency — single-dose durable expression is a clinically important property. AAV's episomal persistence in non-dividing post-mitotic cells is one of its most valuable attributes for CNS applications, and distinguishes it from mRNA-based LNP delivery where expression is intrinsically transient.
This persistence is not without caveat. In dividing cells — which are present in the CNS in the form of neural stem cells in the subventricular zone and dentate gyrus — episomal AAV DNA is diluted with each cell division. For CNS programs targeting predominantly post-mitotic neurons, this is not a significant concern. For programs where editing of neural progenitors is relevant, episomal dilution matters more. And in a very small fraction of integration events, AAV genomes can integrate near growth-control genes — a consideration that has been raised for liver-directed AAV at high doses, though the clinical significance for CNS delivery at typical doses remains unclear.
Serotype tropism selection
Different AAV serotypes show distinct cellular tropisms in the CNS, established through decades of preclinical characterization. AAV9 and PHP.B show broad neuronal transduction after systemic delivery; AAV-PHP.eB shows preferential cortical and hippocampal neuron transduction; AAV5 has strong striatal tropism in several species. AAV1 transduces cerebellar granule neurons efficiently. This serotype library provides a degree of cell-type selectivity that LNP surface engineering is still working toward — selecting the right AAV capsid can enrich delivery toward the neuronal population of interest without requiring surface engineering from scratch.
Capsid engineering has extended this further: directed evolution and rational design of novel capsid variants continue to expand the tropism toolkit. For CNS programs where cell-type specificity matters and the target cell population is one for which a well-characterized AAV capsid exists, this is a real practical advantage.
AAV for CNS Gene Therapy: Significant Limitations
Cargo capacity ceiling (~4.7 kb)
AAV packaging capacity is approximately 4.7 kb of single-stranded DNA, including inverted terminal repeats. This limits the gene payloads that can be delivered as single-vector constructs. Dystrophin (DMD, ~11 kb coding sequence) requires truncated microdystrophin strategies. Many CRISPR base editors and prime editors exceed 4.7 kb when accounting for promoter, editor coding sequence, and guide RNA expression cassette. Dual-AAV approaches using split intein protein trans-splicing are being explored for larger cargoes but introduce additional manufacturing complexity and reduce editing efficiency relative to single-vector delivery, since both vectors must transduce the same cell for reconstitution of the full editor.
CRISPR base editors — ABE8e (adenine base editor) and CBE4max (cytosine base editor) — are among the most clinically relevant advanced editing tools for CNS point mutation disorders. ABE8e's coding sequence alone is approximately 4.4 kb; with promoter, polyadenylation signal, and sgRNA expression cassette, a single-vector ABE8e construct exceeds 4.7 kb. Delivering a full-function base editor with a guide RNA in a single AAV is not currently feasible without truncation strategies that may compromise editor efficiency or specificity. For programs targeting point mutations — which describes a substantial fraction of rare neurological disease genetics — this constraint directly limits AAV applicability for advanced editing modalities like base editing and prime editing.
Pre-existing immunity and immunogenicity
A significant fraction of the human population has pre-existing neutralizing antibodies (NAbs) against common AAV serotypes due to natural exposure to wild-type adeno-associated viruses. Published prevalence estimates for NAbs against AAV9 in adult populations range from 20–40% depending on geography and assay threshold, with higher prevalences reported for AAV2 (above 50% in some populations) and geographic variation in AAV5 NAb prevalence. Patients with pre-existing NAbs above a defined titer threshold are typically excluded from AAV gene therapy trials because NAb-mediated neutralization dramatically reduces the number of viral particles that reach target cells.
This is not merely a safety concern — it is an efficacy filter that systematically excludes a defined fraction of any patient population from treatment eligibility before the program begins. For rare CNS diseases where patient populations are already small — Huntington's disease has approximately 30,000 affected individuals in the US; CLN3 Batten disease has under 500 newly diagnosed cases per year in the US — NAb-based exclusions reduce the treatable population further in ways that affect both trial feasibility and eventual commercial reach. Immunosuppression regimens to enable AAV treatment in NAb-positive patients are being evaluated but add clinical complexity.
Single-administration constraint in most programs
The strong immune response triggered by initial AAV administration — including capsid-directed cytotoxic T lymphocyte responses and humoral responses generating high-titer anti-capsid antibodies — makes re-administration of the same serotype in most clinical settings impractical. For programs where a single lifetime dose is an acceptable clinical model (SMA, LCA10, some Batten disease strategies), this constraint is manageable. For progressive CNS diseases where neurodegeneration continues after an initial edit, or where the editing efficiency of a single dose is insufficient and a second treatment would provide clinical benefit, the re-administration limitation is a significant constraint on program design.
Heterologous serotype switching (re-dosing with a different AAV capsid) is in clinical investigation as a redosing strategy, but cross-reactive immunity between serotypes limits how many re-dosing cycles are feasible, and each serotype switch requires its own manufacturing process development and dose characterization. This is not an insurmountable problem, but it is a real constraint that LNP-based delivery does not share in the same form.
Manufacturing complexity and scalability
Clinical-grade AAV manufacture requires baculovirus-insect cell systems, HEK293 triple transfection, or other specialized production platforms, followed by extensive downstream purification to remove empty capsids, residual host cell proteins, and process impurities (residual baculovirus, benzonase, process-related DNA). Analytical characterization of the final drug product — full-to-empty capsid ratio, genome titer, infectious titer, capsid protein purity — requires specialized assays. Yields vary substantially by serotype and production process. Scale-up from research-grade (1–10 × 10¹³ viral genomes) to clinical-grade (10¹⁵ –10¹⁶ viral genomes) adds considerable process development time and manufacturing cost.
For rare disease programs with small patient populations, AAV manufacturing economics are tractable and have been demonstrated commercially. The constraint becomes more significant for programs that would require repeated dosing, larger patient populations, or multiple manufacturing campaigns per patient — all of which are plausible scenarios in CNS rare diseases where single-dose efficacy may be incomplete.
LNPs for CNS Gene Therapy: Genuine Strengths
Cargo flexibility and size independence
LNPs are not constrained by packaging capacity in the same way AAV is. mRNA cargo up to at least 10 kb has been encapsulated in LNPs with acceptable encapsulation efficiency, though formulation optimization for large mRNAs requires N/P ratio and lipid composition adjustment to maintain both EE and particle quality (PDI, aggregation resistance). CRISPR base editor mRNAs (4.5–5.5 kb for ABE8e or CBE4max), prime editor mRNAs (typically 6–7 kb for PE2 variants), and pegRNAs can all in principle be co-encapsulated. The co-encapsulation of mRNA and guide RNA as separate cargo species in the same LNP is an active formulation area — the biophysical challenge is that mRNA and sgRNA have different sizes, secondary structures, and optimal N/P ratios, which complicates co-formulation.
For RNP (ribonucleoprotein) delivery — Cas9 protein pre-complexed with sgRNA — LNPs can encapsulate the protein-RNA complex with EE values that vary substantially (typically 30–70% for RNP versus 70–90% for nucleic acid alone) depending on ionizable lipid chemistry and N/P ratio optimization. RNP delivery has a distinct kinetic advantage: because the editing complex is pre-formed, the delay between endosomal escape and nuclear editing activity is shorter than for mRNA, which requires cytoplasmic translation before the editor is active. For post-mitotic neurons where time-in-nucleus access before cell division is not a relevant parameter, the kinetic advantage of RNP may matter less than for dividing cells.
Redosing potential
Unlike AAV, LNPs do not generate durable anti-vehicle immune responses in the same way that anti-capsid immunity limits AAV re-dosing. The ionizable lipid and PEG components are metabolized and cleared without leaving persistent immunological memory against the particle structure. This creates the potential for repeat dosing — a meaningful advantage for progressive CNS diseases where an initial dose achieves partial but not complete editing efficiency, or for epigenetic editing approaches (CRISPRa for frataxin reactivation, for example) where periodic maintenance dosing may be required as epigenetic marks are re-established over time.
Redosing with LNPs is not immunologically unconstrained. The nucleic acid cargo can trigger innate immune activation via toll-like receptors and cytoplasmic RNA sensors (RIG-I, MDA5), which must be managed through chemical modifications — pseudouridine substitution in mRNA to reduce TLR7/8 activation is the most established approach, applied in all approved mRNA LNP therapeutics. Anti-PEG antibodies can develop after repeated LNP dosing, potentially accelerating PEG-lipid shedding or triggering complement activation. These considerations require characterization in any repeat-dosing protocol, but the fundamental constraint is different from the AAV re-administration problem in kind, not just degree.
Manufacturing determinism
LNP manufacturing by microfluidic mixing is a deterministic chemical process: mix defined lipid components in ethanol with nucleic acid in aqueous acetate buffer at a controlled flow rate ratio, temperature, and total flow rate; then remove ethanol and exchange buffer by tangential flow filtration (TFF). The process is highly reproducible batch-to-batch, characterization-friendly (size by DLS, PDI, and EE by RiboGreen can be measured within hours of production), and scalable without the biological variability inherent in cell culture-based virus production. Microfluidic chip-based production scales from nanoliter research volumes to clinical-batch scales by increasing chip parallelization rather than changing process chemistry — a significant advantage for GMP production consistency. This manufacturing profile has been validated extensively through vaccine (COVID-19 mRNA vaccines) and siRNA LNP production, providing a process development foundation that CNS LNP programs can build on.
LNPs for CNS Gene Therapy: Significant Limitations
BBB crossing requires active engineering — and remains unvalidated clinically
This is the central limitation and the core problem that BBB-engineered LNP platforms are designed to address. Unmodified LNPs after intravenous administration accumulate predominantly in liver, spleen, and lung via passive hepatic sinusoidal access and phagocytic uptake. They do not cross the BBB in therapeutically meaningful quantities through passive mechanisms. Achieving CNS biodistribution requires active surface engineering — targeting ligands for TfR1, LRP1, or other BBB-expressed transcytosis receptors — and this surface engineering has not been validated in CNS delivery studies in adult non-human primates at therapeutic doses as of this writing.
This is the single most important distinction between AAV and LNP for CNS gene therapy programs evaluating their options today. AAV9 CNS delivery after IV administration is clinically validated, with human efficacy data in a pediatric neurodegenerative disease. LNP CNS delivery after IV administration is not. Preclinical rodent data for BBB-targeted LNPs is encouraging and mechanistically plausible; the translation to primate and human CNS delivery remains to be demonstrated. Any program evaluating LNPs for CNS delivery should hold this uncertainty clearly rather than treating rodent transcytosis data as predictive of human brain delivery.
Transient cargo expression — asset for editing, liability for replacement
LNP-delivered mRNA is inherently transient — in neural cells, mRNA half-life is typically hours to a few days depending on sequence and cell type. For editing applications (CRISPR nuclease, base editing, prime editing), transient expression of the editor is actually the desired property — it limits the window of potential off-target nuclease activity, which is a safety consideration for editing programs. A brief pulse of base editor expression that achieves the desired base conversion and then disappears is preferable to sustained editor expression that continues to modify both on- and off-target sites over weeks.
For protein replacement applications requiring continuous expression — frataxin restoration in Friedreich's ataxia, where frataxin must be present continuously for normal mitochondrial iron-sulfur cluster assembly — mRNA delivery from LNPs is not appropriate for a single-dose strategy. A repeat-dosing model for protein replacement requires frequent IV administrations to maintain therapeutic protein levels; the clinical, logistical, and regulatory complexity of frequent intravenous dosing for CNS protein replacement is substantially higher than for AAV single-dose gene replacement. This is a genuine limitation that should inform platform selection for each specific indication.
Off-organ accumulation and the brain-to-liver ratio problem
Even with CNS surface targeting, systemic LNP administration results in substantial liver and spleen uptake because passive hepatic sinusoidal access operates simultaneously with receptor-mediated BBB transcytosis. The brain-to-liver delivery ratio for CNS-targeted LNPs in published preclinical studies using TfR1 or LRP1 targeting is typically far below 1:1, meaning substantially more cargo reaches the liver than the brain even in optimized formulations. For CRISPR payload delivery, this means off-target editing in hepatocytes and peripheral immune cells — a safety consideration that must be characterized in any IND-enabling toxicology package.
The practical implication is that dose selection for CNS efficacy will be constrained by hepatic off-target editing thresholds, not just by BBB crossing efficiency. If a therapeutic CNS editing dose requires 50 mg/kg LNP systemically, and at that dose the liver receives 20× higher cargo exposure than the brain, the hepatic indel frequency will be the safety-limiting parameter. Engineering higher brain-to-liver delivery ratios — through combinations of improved transcytosis efficiency, reduced passive hepatic uptake (PEG density optimization, lipid composition changes), and active hepatocyte-avoiding surface chemistry — is an active area of research.
A Framework for Platform Selection
Neither platform is universally superior for CNS gene therapy. The selection depends on specific program parameters that should be evaluated explicitly:
- Cargo size below 4.7 kb, durable expression needed, pediatric patient population with manageable NAb prevalence, and no clinical requirement for redosing: AAV has the clinical track record and should be the default starting point unless a specific immunogenicity, manufacturing, or delivery-route concern drives evaluation of alternatives.
- Cargo size above 4.7 kb — base editor, prime editor, or large gene replacement — or patient population with high NAb seroprevalence, or progressive disease where dose-escalation or redosing would be clinically beneficial: LNP becomes more attractive on cargo and redosing grounds. The BBB crossing limitation must be confronted as the primary development challenge, not treated as solvable by using a standard liver LNP platform.
- Diffuse CNS neurodegeneration requiring systemic IV delivery as the only practical route: Both platforms have genuine work remaining. AAV9 BBB crossing efficiency decreases substantially in adults relative to neonates, and large adult primate data for AAV9 IV delivery shows CNS delivery that may be insufficient for some therapeutic applications. LNP CNS delivery is unvalidated in humans. This is the frontier where developing BBB-engineering science is most urgently needed, and where no current platform can claim clinical readiness for systemic delivery to the adult human brain.
Where We Are
The honest summary is that AAV is the more clinically proven CNS gene therapy vehicle and should be evaluated first for programs where it is technically feasible. LNPs are the more chemically flexible vehicle — no cargo size ceiling, redosing potential, deterministic manufacturing — but the BBB crossing limitation is real and unresolved in humans. Biopathio's platform work is specifically directed at that gap: building the preclinical evidence base for BBB-crossing LNPs capable of delivering advanced CRISPR cargo — base editors, prime editors, RNP — to CNS neurons. That work is not finished, and we are not claiming it is. This comparison is meant to clarify where the genuine technical decisions lie for any program facing the CNS delivery problem — not to declare a winner before the evidence is in.
