The results from these studies clearly illustrate the vast potential of combining AAV capsid DNA shuffling with directed evolution in the CNS. The in vivo selection process was designed to be very stringent where only those capsid clones that crossed the seizure-compromised BBB would be recovered. Indeed clones 32 and 83 exhibited transduction patterns that proved localized to brain areas that experience a seizure-compromised BBB, though clone 83 appeared to transduce substantially more cells than clone 32. As importantly, these clones did not transduce cells in areas of the brain that do not exhibit a seizure-induced compromise in the BBB, such as the striatum or hypothalamus. Another unintentional but beneficial outcome involved the biodistribution patterns. Clones 32 and 83 lost the parental tropism for non-CNS organs, such as the heart, lung, and liver. By testing the parental serotypes that compose clones 32 and 83, it appears that the shuffling and selection process combined the ability of AAV8 to gain CNS access through the seizure-compromised BBB with the peripheral organ detargeting properties of AAV1.
The success of this CNS selection process depended on a unique combination of two crucial elements. In a number of previous studies where the selection occurred in vitro
, the library clones were rescued and amplified using WT adenovirus.6,18
Even when clones were rescued in vivo
using a “biopanning” technique, animals were infected with WT adenovirus in order to rescue the AAV clones.18
This use of WT adenovirus was not practical in the CNS for two reasons. First, WT adenovirus would need to be injected directly into the CNS, but soon after direct injection adenovirus toxicity would kill many if not most of the infected cells (T.J. McCown and R.J. Samulski unpublished results). Second, even if some cells survived the WT adenovirus infection, the selection process would be extremely biased by the specific cellular tropism of adenovirus in the CNS. Thus, mutant AAV rescue from the CNS required PCR techniques. Another key element involved the tissue selection. Even though several previous peripheral shuffling/directed evolution studies have used PCR rescue techniques, these investigations extracted DNA from whole organs19
or intact tumors.20
Whole-tissue extraction in the CNS, however, would include not only all CNS cells, but also endothelial cells of the cerebral vasculature. Thus, it would be possible to select a mutant AAV that targets the cerebral vasculature. In order to circumvent this problem, cells were mechanically dissociated from seizure-sensitive areas of the CNS, and the viral DNA was extracted. This approach allowed the extraction and amplification of AAV DNA from neuron-rich, site-specific cell populations. Given that the virus DNA was amplified by PCR, however, it must be emphasized that the vectors were not selected for their ability to transduce neurons per se
, but only on their ability to penetrate the compromised BBB and bind to cells in regions modulating seizure activity. Thus, we likely recovered some clones that can bind to neurons and other contaminating cells but have defects in cellular entry, trafficking, or uncoating. For this reason, multiple clones needed to be screened rather than continuing the selection until clonal homogeneity was attained. Clearly, the present findings validate the utility of this approach.
Four previously published studies have described some level of AAV delivery to the adult rodent brain after i.v. injection of naturally occurring AAV vectors. Foust et al.16
and Duque et al.17
delivered high doses (2 × 1014
vector genomes/kg and 1 × 1014
vector genomes/kg, respectively) of recombinant AAV9 (rAAV9) to achieve widespread transduction of cells throughout the brain. Towne et al.21
injected 1 × 1013
vector genomes/kg of rAAV6 to achieve a lower level of widespread brain transduction, whereas more recently McCarty et al.22
found widespread brain transduction by optimizing the time interval between mannitol BBB disruption and subsequent i.v. rAAV2 administration (2 × 1013
vector genomes/kg). Although in all of these cases substantial transduction was achieved, the patterns of transduction lacked any regional specificity. In contrast, clones 32 and 83 only transduced cells in areas known to experience seizure-induced BBB compromise. This specific pattern would restrict gene expression to those areas of seizure-induced pathology, a property that should minimize unwanted side effects. Also, in the case of clone 83, this selective transduction was achieved using 3.5 × 1012
vector genomes/kg, a dose that contained 3- to 60-fold fewer viral particles, when compared to these previous studies.
Another beneficial property of clones 32 and 83 involves the dramatic divergence in biodistribution from that of the parental serotypes. Unlike AAV8 or 9, clones 32 and 83 exhibited virtually no tropism for the liver, heart, muscle, or lung. Also, although AAV1 exhibits a biodistribution that is similar to clones 32 and 83, the liver concentration of clone 83 was 30-fold less than the AAV1 concentration (0.12 genomes/cell versus 3.87 genomes/cell; Supplementary Table S2
). This substantial reduction in peripheral biodistribution proves to be quite important in light of studies on AAV-mediated liver transduction. In recent clinical trials for hemophilia B, when rAAV2/factor IX vector was delivered to the liver at doses of 4 × 1011
vector genomes/kg or 2 × 1012
vector genomes/kg, an asymptomatic immune response to the capsid cleared the vector and the accompanying gene expression.23,24
Furthermore, in nonhuman primates intraportal administration of rAAV7 at doses ranging from 0.7 to 3.5 × 1012
vector genomes/ kg provoked a cytotoxic T-cell response, which led to a loss of transgene expression.25
In a separate clinical trial for hemophilia B, this immune response was not observed in the muscle at doses up to 1.8 × 1012
vector genomes/kg, where the rAAV2 vector was well tolerated and transgene expression persisted for over 3 years (ref. 26
). These findings suggest that liver transduction likely poses a major barrier to systemic delivery in humans. The apparent inability of clones 32 and 83 to accumulate in the liver should minimize immune responses to these mutant AAV vectors following peripheral administration.
Tests of the parental serotypes provided important insight into the structural features of the capsid that allowed the mutant AAV vectors to cross a seizure-compromised BBB, as well as detarget peripheral organs. I.v. administration of AAV8 after kainic acid–induced seizures produced a pattern of CNS transduction that was highly similar to the pattern produced by clones 32 and 83. Thus, it appears that the AAV8 component of clones 32 and 83 conveys the ability to cross the seizure-compromised BBB. This conclusion is reinforced by the observation that clones 32 and 83 were the only clones where the AAV8 component extended to the C-terminus of VP3. Although cerebrovascular endothelial cells comprised the dominant cell type transduced by AAV1 after i.v. administration, AAV1 exhibited a biodistribution pattern that was far more similar to clones 32 and 83 than to AAV8 and 9. Thus, the AAV1 portion of clones 32 and 83 likely contributed to this peripheral detargeting. However, the concentration of clone 83 was significantly lower than that of AAV1 in all organs tested except the spleen and gastrocnemius muscle (Supplementary Table S3
). Therefore, other factors, such as PCR point mutations, also must contribute to the peripheral detargeting of clone 83.
In conclusion, the application of AAV capsid DNA shuffling and directed evolution allowed the identification of two mutant AAV clones that upon i.v. administration, selectively cross the seizure-compromised BBB and transduce cells. Coincident with this property was the unexpected but advantageous finding that these clones also exhibit a unique, favorable biodistribution. Future studies will focus on both the therapeutic efficacy of peripheral administration, as well as further mutagenesis focused on increasing the transduction efficacy. Thus, these studies not only illustrate the power of combining capsid DNA shuffling and directed evolution, but also provide a potentially novel avenue of seizure gene therapy.