Adeno‑associated virus (AAV) vectors achieve tissue selectivity (or the lack thereof) through layered interactions between viral capsids and host biology: protein receptors and glycans on cell surfaces, extracellular and anatomical barriers, and tissue‑specific immune milieus. Superimposed on this are human‑made design choices—serotype, capsid engineering, promoter and payload constraints, dose and route—and manufacturing constraints that alter what can be delivered at scale. Synthesizing evidence retrieved from 2015–2026, this review distills the determinants that most strongly influence AAV transduction in the liver, skeletal muscle, retina, and central nervous system (CNS), and converts them into a pragmatic indication‑selection framework for gene therapy development. Where clinical evidence was sought in trial databases, no AAV gene therapy trials were found in the retrieved materials; insights below therefore rely on preclinical and translational studies, including a 13‑year clinical follow‑up in hemophilia B and domain‑specific reviews 10820.
The determinant landscape: receptors, glycans, barriers, dose/route, and manufacturability
AAVR, a multiserotype entry receptor, governs capsid‑specific docking via distinct Ig‑like polycystic kidney disease (PKD) domains. AAV2 depends predominantly on PKD2 (PKD1 accessory), AAV5 engages PKD1, and AAV1/AAV8 optimally engage both PKD1 and PKD2. AAVR glycosylation is not required for AAV2 binding/transduction, underscoring proteinaceous (not glycan) recognition at this step 2.
Heparan sulfate (HS) binding is a route‑ and tissue‑context‑specific enhancer. In the eye, HS binding enhances accumulation at the inner limiting membrane (ILM), boosting intravitreal (Ivt) transduction, yet HS binding weakly influences cell‑type tropism. HS‑deficient AAV2 drops ~300‑fold in Ivt transduction but subretinal delivery restores robust photoreceptor transduction independent of HS 139.
Anatomical barriers shape effective tropism. At the BBB, AAV9 penetrates via an active, cell‑mediated process while preserving barrier integrity and without provoking inflammatory activation in primary human brain microvascular endothelial models 14. In the retina, ILM and related structures limit Ivt spread; focal disruption (laser photocoagulation) enhances RPE transduction across serotypes (AAV2/5/8), emphasizing barrier rather than serotype limits 7.
Capsid engineering changes entry, intracellular handling, and spread—but effects are serotype‑ and tissue‑dependent. Tyrosine‑to‑phenylalanine substitutions on AAV2 enhance neuronal transduction and volumetric spread in striatum/hippocampus, while analogous changes on AAV5/AAV8 do not; ablating HS binding in AAV2 increases spread in CNS parenchyma 17. The 7m8 peptide insertion increases retinal expression intensity by improving cellular entry (not spread), benefiting AAV9 in retina and both AAV2 and AAV9 in brain, again underscoring tissue‑dependent effects 11.
Multi‑trait learning uncovers epistatic design rules. The Fit4Function framework integrates production fitness, hepatocyte binding/transduction, liver biodistribution and in vivo efficacy to optimize liver‑directed capsids under manufacturability constraints. Charged residues (Arg/Lys) tend to increase liver targeting, Cys/Trp reduce production fitness, and models trained on mouse in vivo plus human in vitro data predict macaque liver biodistribution—enabling cross‑species extrapolation from one‑round library screens 1.
Manufacturing matters. An adaptive ML purification workflow increased yields from ~70% to ~99% and reduced host‑cell impurities 230–400‑fold across AAV2, AAV5, and AAV9 while preserving activity—addressing a serotype‑spanning bottleneck with clinical impact on dose and cost of goods 15.
Payload size constraints are real. Oversized AAV8 vectors can deliver large hepatic genes (e.g., CPS1) but necessitate careful serotype‑payload trade‑offs for liver indications 4.
Table 1. Key determinants and where they matter most
| Determinant | Core insight | Tissue/route salience | Evidence |
|---|---|---|---|
| AAVR PKD domain usage | AAV2→PKD2; AAV5→PKD1; AAV1/AAV8→PKD1+PKD2; glycosylation not required | Broad; influences serotype choice per tissue AAVR profile | 2 |
| HS binding | Enhances ILM accumulation and Ivt efficiency; weakly affects tropism | Retina (intravitreal) | 139 |
| Anatomical barriers | BBB crossing by active transcytosis (AAV9) with intact barrier; ILM limits Ivt | CNS (systemic), Retina (Ivt) | 147 |
| Capsid engineering | Y→F boosts AAV2 CNS efficacy; HS ablation increases spread; 7m8 increases entry (retina) | CNS, Retina; serotype‑specific | 1711 |
| Multi‑trait ML | Predicts cross‑species liver performance; balances manufacturability and function | Liver | 1 |
| Purification optimization | 99% yield; 230–400× impurity reduction across AAV2/5/9 | Cross‑tissue (dose, safety, cost) | 15 |
| Payload capacity | Oversized AAV8 feasible for hepatic delivery | Liver | 4 |
Liver: durable clinical efficacy, cross‑species design, and manufacturability constraints
The liver remains the best‑validated AAV target in humans. In a 13‑year follow‑up of scAAV2/8‑LP1‑hFIXco for severe hemophilia B, single intravenous doses of 2×10^11, 6×10^11, and 2×10^12 vg/kg yielded stable factor IX activity medians of 1.7, 2.3, and 4.8 IU/dL, respectively, with a 9.7‑fold reduction in annualized bleeding and 12.4‑fold reduction in factor IX concentrate use 10. Safety consisted mainly of transient aminotransferase elevations; no inhibitor development, thrombosis, or chronic liver injury was observed. A 10‑year biopsy confirmed transcriptionally active hepatocyte expression without fibrosis/dysplasia. Notably, neutralizing antibodies to AAV8 remained high throughout, signaling a major barrier to vector readministration 10.
Designing beyond serotype, Fit4Function demonstrates that robust liver targeting demands balancing production fitness with hepatocyte traits and in vivo biodistribution. By uniform sampling of production‑fit sequence space, sequence‑to‑function models predicted macaque liver biodistribution from mouse in vivo and human in vitro data; 88.4% of validated variants met all six criteria (production plus five hepatocyte traits), versus 2.6–7.0% for controls. Dosing examples include 1×10^12 vg/mouse and 4.6×10^12–1×10^13 vg/kg in macaques. Production yields of validated capsids matched AAV9, and tissue trade‑offs were evident (e.g., lower enrichment relative to AAV9 in brain or kidney for some variants), highlighting the value of multi‑model consensus to balance cross‑tissue performance 1.
Manufacturing and payload constraints shape feasibility. Adaptive ML increased purification yield to ~99% and sharply reduced impurities across AAV2/5/9, maintaining activity—supporting liver programs that require scalable, high‑dose lots 15. Oversized AAV8 can accommodate large hepatic genes (e.g., CPS1), but packaging capacity must be weighed against capsid choice and manufacturability 4.
Implications for indication selection: Liver‑directed, durable, single‑dose therapies (coagulation factor deficiencies, select metabolic disorders) align with existing clinical durability, the ability to model cross‑species performance preclinically, and robust manufacturing pathways; however, re‑dosing barriers must be anticipated 10115.
Retina: route dictates the rules—HS at the ILM, barrier disruption, and targeted delivery options
In the eye, the delivery route is as determinative as the capsid. For Ivt delivery, HS binding is the dominant enhancer of vector accumulation at the ILM: HS‑ablated rAAV2 shows ~300‑fold lower Ivt transduction than HS‑binding AAV2, an effect replicated in human ex vivo retina. Introducing HS binding into AAV1 or AAV8 increases Ivt transduction by boosting retinal accumulation, yet HS binding by itself does not set cellular tropism; chimeric capsids with dual‑glycan usage further increase transduction and shift tropism 13. Mechanistic dissection confirms that while HS is critical for AAV2 Ivt delivery, subretinal delivery bypasses HS dependence: AAV2 variants lacking HS binding still efficiently transduce photoreceptors subretinally, showing that outer retinal entry uses other receptors 9.
When anatomical barriers limit efficacy, mechanical or anatomical solutions help. Laser photocoagulation primes RPE transduction after Ivt injection across AAV2, AAV5, and AAV8, underlining that barrier disruption can override serotype differences near the lesion 7. Suprachoroidal delivery of AAV8 expressing an anti‑VEGF‑A fragment demonstrated an excellent safety profile and biological activity, with wide biodistribution and outpatient feasibility; compared with Ivt, this route can broaden coverage and reduce injection burden for vascular indications 12.
Capsid engineering for the retina requires tissue‑context validation. The 7m8 peptide increases transduction intensity by enhancing cellular entry rather than spread; in retina, only AAV9 benefited from 7m8 among tested serotypes, whereas in brain both AAV2‑7m8 and AAV9‑7m8 improved expression—reinforcing that a single capsid tweak has serotype‑ and tissue‑specific outcomes 11. Receptor usage also varies by serotype: AAV5’s reliance on AAVR PKD1, versus AAV2’s PKD2 dependence, offers a lever for pairing serotypes to tissues with differential AAVR domain presentation 2.
Implications for indication selection: For inner retinal targets via Ivt, favor HS‑binding capsids or HS‑engineered variants and consider barrier‑modifying adjuncts; for photoreceptors, subretinal delivery bypasses HS dependencies; for choroidal/retinal vascular diseases, suprachoroidal AAV8 offers broader biodistribution with outpatient safety 139712211.
CNS: BBB biology, species translation, and biased cell‑type profiles in human tissue
Systemic CNS delivery pits capsids against the BBB and species‑specific receptor biology. AAV9 traverses human brain endothelial barriers through an active, cell‑mediated process while preserving tight junctions, electrical resistance, and a non‑inflamed state—evidence that efficient BBB crossing need not disrupt barrier integrity 14. In neonatal mice, systemic AAVrh10 outperforms or matches AAV9 across brain regions, with statistically significant advantages in medulla and cerebellum and a clear edge in dorsal spinal cord and lower motor neurons at low doses; superiority attenuates with dose escalation as AAV9 improves 5. Yet mouse‑optimized capsids may not translate: AAV‑PHP.B’s dramatic BBB penetrance in mice did not carry to marmosets, where neuronal (0–2%) and astrocytic (0.1–2.5%) transduction resembled AAV9; instead, strong dorsal root ganglia transduction suggested altered PNS tropism. This species‑dependence underscores the need to validate CNS capsids in primates before clinical steps 19.
Cell‑type selectivity in human brain remains limited for common vectors. In ex vivo living human brain tissue, AAV2 and AAV9 produced broad, nonselective transduction with expression most apparent in astrocytes, suggesting that refined capsid or regulatory designs will be needed to achieve precise neuronal versus glial targeting in human CNS 16. Rational capsid edits can help but remain serotype‑specific: AAV2 tyrosine/threonine mutants enhanced neuronal transduction and spread in striatum/hippocampus, while analogous edits on AAV5/AAV8 were not beneficial; ablating HS binding increased spread, decoupling distribution from entry affinity 17.
Two design frontiers expand options. First, localized delivery strategies (intrathecal, intrastriatal, intraparenchymal) can reduce systemic dose and off‑target exposure—an approach broadly supported across vectors and indications 20. Second, AAV‑nanoparticle chimeras can “escort” nanomaterials into AAV‑specified cells: coupling brain‑tropic CAP‑B10 to nanoparticles achieved ~10% injected dose per gram of brain while avoiding liver accumulation, transferring serotype tropism to non‑viral cargos 6.
Implications for indication selection: For widespread CNS needs (e.g., pediatric diffuse disorders), systemic AAV9 or AAVrh10 warrant consideration with dose‑dependent trade‑offs; for focal CNS targets or when cell‑type precision is required, localized delivery and serotype‑tuned or engineered capsids are preferable; always validate BBB‑crossing capsids in non‑human primates and, where possible, in human tissue models given broad astrocytic bias of common vectors 514191617206.
Skeletal muscle: systemic dose, immune liability, and translational uncertainty
Compared with liver and eye, skeletal muscle poses intertwined distribution and immunological hurdles. Clinical commentary from a Phase 3 Duchenne muscular dystrophy (DMD) program highlights systemic delivery needs to cover large muscle masses, dose‑limiting toxicity (including immune responses), and the difficulty of achieving uniform expression across muscle groups—challenges distinct from liver and retina 18. While direct skeletal muscle receptor biology is less resolved in the retrieved set, lessons from related tissues inform strategy: AAVR domain dependencies vary by serotype 2; dose and route shape reach; and localized delivery can mitigate systemic exposure 20. In the enteric nervous system—a peripheral neural compartment—AAV6 and AAV9 performed best after local colonic injection, with AAV9 favoring neurons and AAV6 transducing both neurons and enteric glia; coverage scaled with titer/time, a single injection covered ~47 mm^2, and no vector‑related immune response was detected—evidence that local delivery can enhance safety and targeting in peripheral tissues 3.
Implications for indication selection: For whole‑body myopathies, weigh the need for high systemic doses against immune risk; when anatomy allows (e.g., limb or compartmental disease), localized delivery and serotype choice tuned to target cell profiles (AAV6 vs AAV9) may improve benefit‑risk 18320.
Cross‑species translation and model selection
Translation hinges on species‑appropriate validation. A formal framework stresses selecting animal models with relevant receptor and immune profiles, recognizing that single‑dose AAV therapies can be durably curative and thus demand rigorous, predictive non‑clinical evidence 8. The AAV‑PHP.B primate failure 19 and Fit4Function’s success predicting macaque liver biodistribution from mouse+human data 1 bookend best and worst cases: data‑driven, multi‑context training improves translation for the liver, whereas single‑species optimization can mislead in the CNS. Human ex vivo tissue models (e.g., brain slices) provide additional, cell‑type‑resolved insights 16.
From mechanisms to choices: an indication‑selection framework
Table 2 synthesizes tissue‑specific determinants into design choices, doses, and translational flags.
Table 2. Indication‑selection matrix by target tissue
| Target tissue | Dominant biological determinants | Preferred serotypes/strategies | Routes and dose exemplars | Key engineering/manufacturing levers | Translational flags |
|---|---|---|---|---|---|
| Liver | Hepatocyte entry; cross‑species biodistribution; immune durability | AAV8 (pseudotyped AAV2/8), AAV9; ML‑designed capsids (Fit4Function) | IV single dose; 2×10^11–2×10^12 vg/kg in hemophilia B; macaque doses 4.6×10^12–1×10^13 vg/kg; 1×10^12 vg/mouse in screens 101 | Multi‑trait capsid design; purification yield/purity optimization; oversized AAV8 for large genes 1154 | High neutralizing antibodies post‑dose impede re‑administration 10 |
| Retina (inner layers/RPE, Ivt) | ILM barrier; HS‑mediated accumulation | HS‑binding AAV2 or HS‑engineered AAV1/AAV8; barrier disruption adjuncts | Intravitreal; HS enhances ILM accumulation; laser pretreatment boosts RPE transduction 137 | Glycan retargeting; peptide insertions (7m8) validated per serotype/tissue 1113 | HS binding weakly affects tropism; human ex vivo validation desirable 139 |
| Retina (photoreceptors) | Outer retinal entry via non‑HS receptors | AAV2/5/8 subretinal | Subretinal; HS binding not required for photoreceptor transduction 9 | Capsid edits validated in ocular context; consider AAVR domain preferences (AAV5→PKD1) 211 | Surgical delivery trade‑offs vs coverage |
| Retina (vascular) | Broad chorioretinal coverage | AAV8 suprachoroidal | Suprachoroidal; wide biodistribution; outpatient safety; biological activity shown 12 | Dose‑sparing via route; durable anti‑VEGF delivery 12 | Competes with established biologics; long‑term data needed |
| CNS (diffuse) | BBB crossing; species‑specific receptor biology; broad cell transduction | AAV9 (active transcytosis), AAVrh10 (low‑dose efficiency in neonates) | Systemic IV; AAVrh10 superiority at low dose; AAV9 improves with dose 514 | Capsid edits (AAV2 Y→F) for local parenchyma; avoid mouse‑optimized PHP.B without primate validation 1719 | Human brain shows astrocyte‑biased, nonselective transduction with AAV2/9; primate validation essential 1619 |
| CNS (focal) | Local parenchymal barriers; cell‑type precision | Engineered AAV2 variants; localized delivery; AAV‑nanoparticle chimeras | Intracerebral/intrastriatal/intrathecal; nanoparticle escort achieves ~10% ID/g brain with CAP‑B10 and spares liver 17206 | Design for spread vs entry as separable traits; payloads beyond genes via chimeras 176 | Translate in human tissue models for cell‑type profiling 16 |
| Skeletal muscle | High mass coverage; systemic immunity | AAV9 for systemic; AAV6 for local compartments | Systemic for DMD‑like indications; local injections where feasible; ENS analog shows AAV6/AAV9 differences and no local immune response 183 | Immune‑mitigating regimens and localized delivery to lower dose burden 2018 | Systemic dose‑limiting toxicity; variable muscle‑group efficacy 18 |
Evidence gaps in retrieved clinical datasets
A structured clinical‑trial search returned no AAV gene therapy trials in the retrieved materials, limiting head‑to‑head clinical comparisons of serotypes, dose‑response, and immune modulation across tissues. Where such data are absent, this review emphasizes preclinical and translational studies and flags the need for primate and human tissue validation, especially for CNS and skeletal muscle indications 81619.