Recombinant adeno-associated virus (AAV) vectors have enabled efficient, durable in vivo gene transfer in numerous therapeutic areas, with clinical successes in inherited retinal diseases, spinal muscular atrophy, and hemophilia (Shen et al., 2022; Wang et al., 2019). AAV vector can be used in the context of gene replacement (Mendell et al., 2017), RNA interference (Amado et al., 2025), or gene editing (Ou et al., 2019). Despite favorable safety profiles and scalable manufacturing (Liu et al., 2024), AAV translation remains challenging because vector behavior differs substantially across species. Rodents are indispensable for discovery and proof-of-concept studies, yet their immune systems, receptor repertoires, and tissue architecture often diverge from those of non-human primates (NHPs) and humans (Byrne et al., 2025). Even among non-human primates, patterns of AAV susceptibility, tissue tropism, and biodistribution vary by clade: engineered AAV variants can show neuron-biased transduction in infant Old World monkeys yet vasculature-biased transduction in adult New World marmosets, highlighting clade-specific differences in how AAV vectors interact with host biology (Chuapoco et al., 2023).

Species barriers manifest at multiple levels: capsid binding to glycans or protein receptors, intracellular processing and innate sensing, vector dissemination across tissue barriers, and adaptive immune responses that differ in magnitude and specificity (Cao et al., 2024; Wang et al., 2024; Keeler et al., 2025). Importantly, such effects are not isolated phenomena but recur across tissues and vector classes. A prominent example is the strain-dependent enhancement of central nervous system transduction by AAV-PhP.B, which depends on LY6A expression in select C57BL/6J mouse strains and fails in other strains and in NHPs (Deverman et al., 2016; Hordeaux et al., 2018; Huang et al., 2019). In the liver, the capsid AAV-LK03 exhibits efficient transduction of human hepatocytes yet markedly reduced activity in murine livers, revealing fundamental interspecies differences in hepatocyte entry and intracellular processing (Lisowski et al., 2014). Similarly, muscle-directed capsids such as MyoAAV-1A and MyoAAV-2A demonstrate robust performance in mice but altered or diminished transduction profiles in NHPs, underscoring limitations in extrapolating muscle tropism across species (Tabebordbar et al., 2021). At a systemic level, whole-body biodistribution studies further show that AAV tropism and clearance patterns following intravenous delivery differ substantially between mice and NHPs, affecting both target engagement and off-target exposure (Fang et al., 2025).

This review summarizes molecular and physiological determinants of AAV species barriers, strain and species differences in vivo, and strategies—including machine-learning-guided capsid design and advanced human-derived models—that enhance translational predictability.

Rodent strains exhibit profound differences in susceptibility to individual AAV variants. The most striking case is PhP.B, which provides high-level central nervous system (CNS) transduction in C57BL/6J mice following intravenous injection but fails almost entirely in BALB/c mice, other mouse strains, NHPs, and humans (Deverman et al., 2016; Hordeaux et al., 2018; Huang et al., 2019; Matsuzaki et al., 2019). This phenotype depends on expression of LY6A on brain endothelium, a protein absent from many mouse strains and primates. PhP.B thus illustrates how strain-specific genetics can create misleading impressions of cross-species transduction potential. Beyond the CNS, hepatic and immune traits also vary among rodent strains. Differences in Kupffer cell abundance, complement activation, and sinusoidal fenestration could influence vector uptake and clearance.

NHPs provide a closer physiological and immunological approximation to humans than rodents and are therefore central to translational AAV development. Old World monkeys, particularly rhesus and cynomolgus macaques, are widely used for systemic and CNS studies because key aspects of their glycan landscapes, complement systems, and immune repertoires more closely parallel those of humans (High and Roncarolo, 2019). Nevertheless, even within macaques, inter-colony variability in pre-existing immunity remains substantial. Neutralizing antibody titers against AAV1, AAV2, AAV8, and AAV9 can range from undetectable to high depending on geographic origin, husbandry, and viral exposure history (Boutin et al., 2010), complicating cross-study comparisons and translational interpretation. New World monkeys, such as the common marmoset, are also widely used in AAV research and often reproduce major features of vector biodistribution observed in macaques, particularly for CNS- and vasculature-targeting capsids. However, marmosets possess distinct immunogenetic and developmental features that can influence immune-mediated readouts and vector interpretation. These include reduced diversity and structural differences in major histocompatibility complex (MHC) haplotypes, frequent hematopoietic chimerism, and differences in endothelial and neurodevelopmental biology relative to Old World primates and humans (Antunes et al., 1998; Sweeney et al., 2012; Seki et al., 2017). Consistent with this, engineered AAVs can show species- and age-dependent shifts in CNS cell-type bias between marmosets and macaques (Chuapoco et al., 2023). As a result, while marmosets are valuable for assessing feasibility, developmental neurobiology, and ophthalmic delivery, caution is warranted when extrapolating performance directly to humans.

Human clinical data underscore the inadequacy of rodent-only and even NHP-only predictions. Pre-existing neutralizing antibodies remain a major barrier to effective AAV transduction across indications (Li et al., 2022). High-dose intravenous AAV9-like vectors have caused severe hepatotoxicity and dorsal root ganglion injury in NHPs and piglets, phenomena not fully anticipated by rodent studies (Hinderer et al., 2018). MicroRNA-regulated expression systems can mitigate these toxicities in NHPs (Hordeaux et al., 2020), illustrating how species-appropriate refinements can directly improve safety.

Beyond molecular determinants, tissue microanatomy strongly influences vector performance. In the retina, the inner limiting membrane (ILM) is considerably thicker and structurally more complex in NHPs and humans than in mice. Dalkara and colleagues showed that many capsids able to cross the murine ILM and transduce photoreceptors after intravitreal delivery perform poorly in larger species, necessitating subretinal delivery or specialized engineered variants (Dalkara et al., 2013). Human stem cell–derived retinal pigment epithelium and photoreceptors display distinct capsid preferences relative to murine tissues, emphasizing the need for human-relevant models (Gonzalez-Cordero et al., 2018). In the liver, sinusoidal endothelial cell fenestration and scavenger endothelial cell biology differ markedly across species (Shetty et al., 2018). These differences affect not only hepatocyte access but also opsonization and sequestration by non-parenchymal cells. The blood–brain barrier (BBB) also differs structurally between rodents and primates. Rodent endothelial cells possess distinct transporter and glycocalyx profiles compared with macaques and humans, contributing to discrepancies in CNS penetration of systemically delivered AAV. PhP.B’s failure to cross the primate BBB illustrates this point clearly (Hordeaux et al., 2018). Age and disease state further modulate BBB permeability, potentially affecting AAV CNS tropism (Foust et al., 2009). These structural differences highlight the need for model systems that recapitulate human tissue barriers, particularly when vectors are intended for intravitreal, intrathecal, or systemic delivery.

Despite significant advances in capsid engineering and human-relevant model systems, species barriers remain a major limitation in AAV gene therapy. Several key challenges remain unresolved. First, receptor conservation alone does not ensure translational success. While receptor-defined capsids improve cross-species performance, downstream processes such as intracellular trafficking, uncoating, and nuclear entry remain poorly understood and may differ substantially between species. Clarifying these post-entry mechanisms in human-relevant systems remains a critical need. Second, immune-mediated species differences are difficult to predict and model. Variability in innate sensing, complement activation, and pre-existing humoral immunity can strongly influence vector clearance and toxicity (Li et al., 2022), yet these responses differ across species and even among non-human primates. Improved immune-competent humanized systems and standardized immunoprofiling will be essential for improving predictability. Third, dose-dependent and tissue-level effects remain underexplored. Many species barriers emerge only at clinically relevant doses, where nonlinear effects such as scavenger saturation or endothelial injury become apparent. Preclinical studies that explicitly address dose scaling are needed to bridge this gap. Finally, no single experimental model captures all relevant species barriers. Rodent, non-human primate, humanized, and organoid systems each provide partial insights, underscoring the need for integrated, multi-model evaluation strategies. Together, these limitations highlight that overcoming species barriers is a systems-level challenge that will require coordinated advances in capsid design, immune characterization, and human-relevant modeling, rather than further optimization within isolated preclinical contexts.

In summary, AAV species barriers arise from a combination of molecular, cellular, anatomical, and immunological factors that differ significantly across rodents, NHPs, and humans (Figure 1). These differences can lead to both false positives and false negatives in preclinical studies and therefore must be carefully considered during vector selection and study design. Integrating multi-species data, machine-learning-guided capsid engineering, and human organoid and explant systems offers an opportunity to improve translational reliability (Figure 1). As the field advances, embracing species and strain diversity as informative rather than inconvenient will be essential to design vectors that are safer, more predictable, and more effective for human patients.