A Multifaceted Approach to Optimizing AAV Delivery to the Brain for the Treatment of Neurodegenerative Diseases

Abstract

Despite major advancements in gene therapy technologies, there are no approved gene therapies for diseases which predominantly effect the brain. Adeno-associated virus (AAV) vectors have emerged as the most effective delivery vector for gene therapy owing to their simplicity, wide spread transduction and low immunogenicity. Unfortunately, the blood-brain barrier (BBB) makes IV delivery of AAVs, to the brain highly inefficient. At IV doses capable of widespread expression in the brain, there is a significant risk of severe immune-mediated toxicity. Direct intracerebral injection of vectors is being attempted. However, this method is invasive, and only provides localized delivery for diseases known to afflict the brain globally. More advanced methods for AAV delivery will likely be required for safe and effective gene therapy to the brain. Each step in AAV delivery, including delivery route, BBB transduction, cellular tropism and transgene expression provide opportunities for innovative solutions to optimize delivery efficiency. Intra-arterial delivery with mannitol, focused ultrasound, optimized AAV capsid evolution with machine learning algorithms, synthetic promotors are all examples of advanced strategies which have been developed in pre-clinical models, yet none are being investigated in clinical trials. This manuscript seeks to review these technological advancements, and others, to improve AAV delivery to the brain, and to propose novel strategies to build upon this research. Ultimately, it is hoped that the optimization of AAV delivery will allow for the human translation of many gene therapies for neurodegenerative and other neurologic diseases.

Keywords: adeno-associated virus (AAV); blood–brain barrier disruption; capsid engineering; gene therapy; genetic vectors; intra-arterial (IA) delivery; intra-thecal drug delivery systems; neurodegenerative disease.

FIGURE 1

Methods for optimizing at each of four stages of AAV delivery to the brain. In order to get to the brain AAVa are (1) introduced into the body via a delivery route. Routes which avoid systemic delivery such as IT-CM and IA with BBB crossing should be optimal. (2) If intravascularly delivered, a strategy for enhancing BBB delivery should be chosen. (3) Target cell entry (i.e., neurons for NDDs). This can be accomplished by capsid engineering to enhance specific cellular tropism. And (4) transgene expression. Promoters can be used to target transgene expression to specific cells and/or conditions, in this example, a promoter which drives gene expression in the context of proteostatic stress is used. This should reduce transgene toxicity. Created with Biorender.Com.

FIGURE 2

Meta-analysis of 14 papers reporting quantitative metrics of transduction efficiency using different routes and AAV serotypes. (A) Liver: brain ratio by route and serotype in mice. 26 data points from 12 studies in mice. Each used qPCR to report viral genome concentration in both liver and brain. Viral concentration in liver was divided by viral concentration in brain after injection of a single dose via IV, IT, or lA routes. (B) Liver: brain ratio in non-human primates. Eight data points from four studies in non-human primates. Calculated using same method as “A.” For bath panels (A,B) the greater the ratio, the less efficient the route/serotype. combination. (C) Percent transduction in NHPs normalized to a median dose of 10¹³. Percent transduction was calculated either by% GFP positive cells reported by authors or by estimating the percentage of brain cells transduced multiplying the number of viral genomes per cell calculated by authors via qPCR by 100 (i.e., if the ratio of viral genomes to normal cells is 1 then it was estimated that all cells or 100% of cells were transduced). This was then normalized to the median dose of l × l0¹³ vg by dividing the % transduction by the ratio of actual dose/median dose. Data was extrapolated from 10 data points from 4 studies in non-human primates. (D) % transduction (calculated by the same method as in panel C) vs. viral dose for different combinations of routes and serotypes. In the event that there the same route/serotype was used at different concentrations, the linear regression of these values is displayed as a dotted line with this line labeled, rather than each data point. This data was extrapolated from 31 data points from 11 studies was used. In panels (A–D) if there were multiple data points for a given route/serotype combination (and in panel (D) route/serotype/dose combination) the average of all data points was used.

FIGURE 3

Organization of potential routes of AAV delivery based on biodistribution and invasiveness. Ideal routes for the treatment of NDDs would transduce the entire brain but minimize systemic delivery, and would the least invasive possible. The dashed circle denotes where this ideal treatment would lie. Although no delivery method is perfect, the closest would be intravenous or infra-arterial with a strategy to cross the BBB which works in humans, intra-cisternal or intra-lumbar for spinal cord diseases. Created with Biorender.Com.

FIGURE 4

A method of directed evolution to optimize AAV delivery to the brain, in humans. First, properties of an idea delivery system are identified. Next, a delivery route is chosen which optimizes as many of these properties ill and of itself. Capsid engineering can be used to acquire as many of the remaining properties as possible. To do this, random mutagenesis is used and a ML model trained to select only variants with high packaging fitness can be used to diversify the library. Next these variants are introduced into an in vitro model which selects for a desired property, for example a model containing all of the main cell types which the virus could transduce. The DNA is then harvested from the target condition and sequenced using next generation sequencing. In this example variants which transduced neurons are identified for positive selection, while variants which transduced APCs are identified for negative selection. The ML model is then trained with the lists of successful variants. This process is repeated multiple times for a single property until novel high yield variants are no longer discovered in successive rounds of evolution. Once one property is complete, this process can be repeated for new properties using a different in vitro model for selection. Finally, the ML algorithm which has been trained with all of the data can identified variants which have the greatest fitness for all selected properties. These variants are then introduced into NHPs in vivo to confirm that they are function in a live environment and that they are safe. Finally, the variants are ready for human trials.