Delivering genes across the blood-brain barrier: LY6A, a novel cellular receptor for AAV-PHP.B capsids

The engineered AAV-PHP.B family of adeno-associated virus efficiently delivers genes throughout the mouse central nervous system. To guide their application across disease models, and to inspire the development of translational gene therapy vectors for targeting neurological diseases in humans, we sought to elucidate the host factors responsible for the CNS tropism of the AAV-PHP.B vectors. Leveraging CNS tropism differences across 13 mouse strains, we systematically determined a set of genetic variants that segregate with the permissivity phenotype, and rapidly identified LY6A as an essential receptor for the AAV-PHP.B vectors. Interfering with LY6A by CRISPR/Cas9-mediated Ly6a disruption or with blocking antibodies reduced transduction of mouse brain endothelial cells by AAV-PHP.eB, while ectopic expression of Ly6a increased AAV-PHP.eB transduction of HEK293T and CHO cells by 30-fold or more. Importantly, we demonstrate that this newly discovered mode of AAV binding and transduction can occur independently of other known AAV receptors. These findings illuminate the previously reported species- and strain-specific tropism characteristics of the AAV-PHP.B vectors and inform ongoing efforts to develop next-generation AAV vehicles for human CNS gene therapy.


Introduction
With the boom of gene replacement, knockdown, and editing technologies, the number of diseases that are potentially treatable by gene therapy is rapidly expanding. AAV vectors are proving to be safe, versatile vehicles for in vivo gene therapy applications [1][2][3][4] . However, delivery challenges impede the application of gene therapy, particularly in the context of the brain, which is protected by the blood-brain barrier (BBB). To improve gene delivery across the central nervous system (CNS), our group and others have engineered AAV capsids using in vivo selection and directed evolution [5][6][7][8] . Our group has focused on engineering AAV9 variants, such as AAV-PHP.B 6 and its further evolved, more efficient variant, AAV-PHP.eB 5 , to cross the adult BBB and enable efficient gene transfer to the CNS. Since their invention, AAV-PHP.B and AAV-PHP.eB have been applied across a wide range of neuroscience experiments in mice 5,9,10 , including genetic deficit correction 11,12 and neurological disease modeling 13 .
A critical question has been how the AAV-PHP.B vectors cross the BBB and whether this mechanism can be translated to other species and, ultimately, humans. The enhanced CNS tropism of AAV-PHP.B and AAV-PHP.eB appears to extend to rats 14,15 , whereas studies testing AAV-PHP.B or related capsids in nonhuman primates (NHPs) have yielded differing outcomes [16][17][18] . Surprisingly, the enhanced CNS tropism of AAV-PHP.B 6,9,[12][13][14][15]19,20 is starkly absent in BALB/cJ mice 16 . These findings indicate that the ability of AAV-PHP.B to cross the BBB is affected by genetic factors that vary by species and mouse strain. In this work, we leverage this strain-dependence to identify LY6A as the cellular receptor responsible for the enhanced CNS tropism exhibited by the AAV-PHP.B capsid family. We demonstrate that LY6Amediated transduction is independent of known AAV9 receptors and is a unique means by which AAV-PHP.B capsids cross the mouse BBB. This has widespread implications for guiding the use of AAV-PHP.B capsids in disease models, as well as ongoing efforts to engineer next-generation AAVs that cross the BBB in other species.

Results
Ly6 genetic variants associate with the CNS tropism of AAV-PHP.eB The dramatic difference in the CNS tropism of AAV-PHP.B in C57BL/6J versus BALB/cJ mice 16 extended to AAV-PHP.eB (Fig. 1A) and two other AAV-PHP.B capsids (AAV-PHP.B2 and AAV-PHP.B3; Supplementary Fig. 1). Consistent with the reduced transduction phenotype in BALB/cJ mice, we observed a loss of the CNS-specific enhanced accumulation of AAV-PHP.eB, relative to AAV9, along the vasculature and within the brains of BALB/ cJ mice (Fig. 1B, C). These findings suggest that a cellular factor present on the brain endothelium is responsible for the efficient CNS-wide transduction mediated by the AAV-PHP.B capsids, and that this factor is absent or nonfunctional in BALB/cJ mice.
In light of this disparity, we sought to test for the enhanced CNS tropism of the AAV-PHP.B capsids across a panel of mouse lines, harnessing the natural genetic variation between mice to identify the genetic variants and, subsequently, candidate gene(s) responsible for the difference in CNS transduction by AAV-PHP. eB. Diverging from conventional diversity outbred or genetic linkage studies requiring the inter-breeding of mouse lines, we leveraged the open-source software Hail 21 to perform a genomewide association study (GWAS) utilizing the Mouse Genome Project database spanning 36 mouse lines 22 . We aimed to rapidly identify variants whose alleles segregate across mice with the observed phenotype (permissive or non-permissive to enhanced CNS transduction by AAV-PHP.eB). Starting from millions of genetic variants comprised of single-nucleotide polymorphisms (SNPs) as well as insertions and deletions (indels), we narrowed our analysis to variants predicted to affect expression, splicing, or protein coding regions (Supplementary Table 1). Using a statistical simulation framework, we estimated that 12 mouse lines would be sufficient to narrow our search to ~10 high/medium impact variants ( Supplementary Fig. 2, Supplementary Table 1), which could feasibly be experimentally interrogated for the enhanced AAV-PHP.eB CNS tropism. Based on this estimation, we acquired mice from 13 commercially available lines, including C57BL/6J and BALB/cJ, and administered 10 11 vector genomes (vg)/animal of AAV-PHP.eB, which packaged an AAV genome encoding a green fluorescent protein (GFP) with a nuclear localization signal (NLS-GFP). Intravenous administration of AAV-PHP.eB resulted in GFP expression in the brain of permissive lines such as C57BL/6J, but not those of nonpermissive mice such as BALB/cJ; we identified seven permissive and six nonpermissive lines ( Supplementary Fig.  2).
The GWAS reduced the number of high/medium impact gene variants to missense SNPs in the Ly6a and Ly6c1 genes (Fig.  1D). RNA sequencing data from sorted mouse brain cells (www. BrainRNAseq.org) 23 indicates that Ly6a and Ly6c1 are expressed in brain endothelial cells (Fig. 1E). Intriguingly, the mouse Ly6 locus is linked to susceptibility to mouse adenovirus (MAV1) 24 , which possesses an endothelial cell tropism that causes fatal hemorrhagic encephalomyelitis in C57BL/6 but not BALB/ cJ mice 25 30 . Despite these promising links between Ly6 genes and various virus susceptibilities, the mechanism remains elusive and has not been leveraged in the development of gene therapy vectors.
We first validated whether protein expression or localization differences arising from genetic variation within Ly6a or Ly6c1 is associated with the differential AAV-PHP.eB tropism across mouse lines. Immunohistochemistry (IHC) assays revealed that LY6A is abundant within the CNS endothelium of C57Bl/6J mice but not in BALB/cJ mice ( Fig. 2A-C). In contrast, Ly6c1 was expressed on the CNS endothelium of both lines (Fig. 2D). The reduced level of LY6A, relative to that observed in C57BL/6J mice, correlated with the nonpermissive AAV-PHP.eB transduction phenotype across five of six mouse strains ( Supplementary Fig. 3). This suggested that reduced LY6A levels may contribute to, but is unlikely to be the only factor responsible for the absence of AAV-PHP.eB transduction. LY6A was evident as multiple bands by western blotting, but only the more slowly migrating band was detectable at low levels in BALB/cJ mice (Fig. 2E), suggesting that LY6A maturation or post-translational processing differs between the mouse lines. We noted that the Ly6a V106A, Ly6c1 P133L (Fig.  1D), and several noncoding SNPs in the Ly6 locus perfectly segregate with the nonpermissive phenotype (observed in six out of six nonpermissive mouse strains). Together, these results suggest an association between Ly6 gene variants and permissivity to transduction by AAV-PHP.eB.
Ly6a is necessary for the enhanced CNS transduction phenotype of AAV-PHP.eB We sought to determine whether LY6A and/or LY6C1 are necessary for the ability of AAV-PHP.eB to bind and transduce CNS endothelial cells. To achieve this, we performed Ly6a and Ly6c1 knockout experiments in brain microvascular endothelial cells (BMVECs) from C57BL/6J mice, which express both genes ( Fig. 3A) and are more efficiently bound and transduced by AAV-PHP.eB than by AAV9 (Fig. 3B, C). We used CRISPR/ SaCAS9 31 and Ly6a-or Ly6c1-specific sgRNAs to disrupt each gene. Using three different sgRNAs to target Ly6a, we achieved a ~50% reduction of LY6A ( Supplementary Fig. 3) and a 50% reduction in binding by AAV-PHP.eB, but not AAV9 ( Fig. 3D and Supplementary Fig. 4). A similar reduction in transduction by AAV-PHP.eB was observed (Fig. 3E); AAV9 transduction of BMVECs was inefficient and not included for comparison. None of the sgRNAs targeting Ly6c1 affected AAV-PHP.eB or AAV9 binding to the BMVECs (Fig. 3D). Collectively, the reduced AAV-PHP.eB binding resulting from Ly6a disruption in BMVECs, the high level of Ly6a expression within the CNS endothelium of permissive mouse lines, and the association of Ly6a SNPs with the nonpermissive phenotype, suggest that LY6A functions as a receptor for AAV-PHP.eB.

Ly6a expression enhances transduction by AAV-PHP.eB in human cells
We next asked whether ectopic Ly6a expression is sufficient for increased binding and transduction by AAV-PHP.eB. We transiently transfected HEK293T cells with cDNAs encoding C57BL/6J Ly6a or Ly6c1, and evaluated the effects on binding and transduction by AAV-PHP.B capsids. Remarkably, Ly6a expression resulted in a >50-fold increase in binding by each of the AAV-PHP.B capsids to HEK293T cells, but did not increase binding by AAV9 (Fig. 3F). As expected, expression of Ly6a, but not Ly6c1, enhanced transduction by AAV-PHP.eB by 30-fold compared to the untransfected control (Fig. 3G).
Binding between LY6A and AAV-PHP.eB is disrupted by a V106A variant of LY6A To determine whether AAV-PHP.eB directly binds LY6A and whether either of the missense SNPs (D63G or V106A) in the BALB/cJ Ly6a gene (Fig. 1D) affect binding to AAV-PHP.eB, we performed virus overlay assays 32 . AAV-PHP.eB bound a protein that co-migrates with LY6A ( Fig. 3H) from HEK293T cells transfected with the C57BL/6J or D63G Ly6a cDNAs, but not from cells expressing Ly6a from the BALB/cJ or V106A cDNAs. Interestingly, the V106A variant is located near the predicted cleavage and GPI anchoring site (ω), and is predicted to reduce the likelihood of GPI-anchor modification 33 (Supplementary Table 2).
LY6A enhances AAV-PHP.eB transduction independently of known AAV9 receptors To determine whether LY6A acts solely as a primary attachment factor or has additional roles in promoting the internalization and trafficking of AAV-PHP.eB, we explored whether AAV-PHP. eB binding and transduction are dependent on known receptor interactions. AAVs typically use multiple cellular receptors for attachment and internalization and intracellular trafficking 34 ; AAV9 utilizes galactose as an attachment factor 35 , and, like most AAVs, requires the AAV receptor (AAVR) for intracellular trafficking and transduction 32 .
We first tested the effect of varying levels of galactose on cell surface glycoproteins on LY6A-mediated AAV-PHP.eB binding. Using Pro5 Chinese Hamster ovary (CHO) cells and derivatives that expose excess galactose (Lec2 cells) or are unable to add galactose to its glycoproteins (Lec8 cells) 36 , which were previously used to map the galactose binding site on the AAV9 capsid 35 , we assessed whether AAV-PHP.eB-LY6A interactions are influenced by galactose levels. As expected, AAV9 and AAV-PHP.B similarly bound and transduced Lec2 cells more efficiently than Lec8 or Pro5 cells (Fig. 4A-C), showing that AAV-PHP.B also utilizes galactose for cell attachment. In contrast, ectopic Ly6a expression selectively increased AAV-PHP.eB, but not AAV9, binding and transduction (Fig. 4A, B) to Pro5 and Lec8 cells. Ly6a expression did not increase binding of AAV-PHP.eB to Lec2 cells (Fig. 4A), potentially due to the high levels of binding driven by excess galactose. Interestingly, Ly6a expression enhanced AAV-PHP.eB transduction of Pro5, Lec2, and Lec8 cells (Fig. 4B). The finding that Ly6a expression rendered Lec8 cells as receptive to AAV-PHP. eB transduction as Pro5 cells indicates that LY6A functions as an attachment factor for AAV-PHP.eB independently of galactose. Furthermore, Ly6a expression enhanced AAV-PHP.eB transduction of Lec2 cells without increasing binding, suggesting that LY6A mediates the internalization and/or trafficking of AAV-PHP.eB.
In this vein, we reasoned that this interaction may render Ly6aexpressing cells permissive to AAV-PHP.eB transduction in the absence of AAVR, which is essential for the intracellular trafficking of nearly all AAV capsids including AAV9 37 . To test this possibility, AAVR WT and KO FVB/NJ mice 37 were injected with AAV-PHP.eB, and their brains were collected two hours later for capsid detection. We detected AAV-PHP.eB along the vasculature and in the brain lysates of both AAVR WT and KO mice (Fig. 4D,  E). Three weeks post-administration, AAV-PHP.eB transduction of neurons and astrocytes, which do not express Ly6a, was nearly absent in the brains of AAVR KO mice. In contrast, AAV-PHP. eB transduced Ly6a-expressing endothelial cells throughout the brains of both AAVR WT and KO mice (Fig. 4F). Collectively, these results demonstrate that the AAV-PHP.eB virus uses a unique and unexpected endothelial cell receptor, LY6A, to gain access to the CNS.

Discussion
Our development of AAV-PHP.B capsids with dramatically improved tropism for the mouse CNS provided a proof-of-concept that AAVs can be engineered to cross the BBB 5,6 . In this publication, we leveraged Hail 21 , the Mouse Genome Project dataset 22 , and 13 commercially available mouse lines to rapidly identify a single amino acid change, out of a starting pool of millions of genetic variants, that impacts AAV-PHP.eB permissivity; the code was implemented and run end-to-end on whole-genome sequencing data within hours, harnessing Hail's ability to scale computation across a large compute cluster, and the in vivo screening was completed in three weeks. The speed and reduced number of animals required for this approach is unprecedented compared to the conventional approaches of using diversity outbred lines or breeding generations of mice to determine the approximate genomic loci that segregates with a given phenotype.
Our identification of LY6A as the receptor engaged by the AAV-PHP.B capsids has important translational value. We demonstrate that LY6A facilitates both binding and internal trafficking of AAV-PHP.B capsids. Other cellular factors that share LY6A's properties such as abundant luminal surface exposure on brain endothelium, localization within specific lipid micro-domains through GPI anchoring, or specific recycling/intracellular trafficking capabilities, may be prime molecular targets for gene delivery vectors in mice, NHPs, and humans. Although Ly6a is absent in primates, other LY6 proteins with homologs in primates are present within the CNS endothelium, and can potentially be explored and harnessed for AAV capsid engineering. Directing capsids and/or other biologicals that target these receptors can open up new therapeutic avenues for treating a wide range of currently intractable neurological diseases.

Materials and Methods
Mouse strain permutation analysis Access to whole-genome sequencing data for 36 commercially available mouse lines 22 made it possible to estimate the number of lines necessary to produce a shortlist of candidate variants. Starting from the known permissive C57BL/6J and nonpermissive BALB/cJ strains, we used the genomic analysis software Hail 21 to simulate permissivity phenotypes for other mouse lines and compute the number of candidate variants in both the high and high-and-medium predicted functional impact classes. First, we sampled the probability p that a random mouse line is permissive from a Beta(ɑ=2, β=2) distribution, based on the two known mouse phenotypes. Second, we sampled a subset of additional commercially available lines (Supplementary Table 3). Third, we simulated a permissivity phenotype for each line in the subset by flipping a p-coin. Finally, we calculated the number of variants with perfect allelic segregation between the permissive and nonpermissive lines. We ran 500 iterations of this model for each of ten subset sizes (3 to 12 mice), providing a distribution over the number of candidate variants at each mouse sample size. This simulation informed our decision regarding the number of mouse lines to order and test in parallel.

Virus production and purification
Recombinant AAVs were generated by triple transfection of HEK293T cells (ATCC CRL-11268) using polyethylenimine (PEI) and purified by ultracentrifugation over iodixanol gradients as previously described 6 .

Western blotting and virus overlay assays
The virus overlay assay was performed as previously reported 32 with some modifications. Briefly, protein lysates were separated on Bolt 4-12% Bis-Tris Plus gels and transferred onto nitrocellulose membranes. After incubation with AAV9 or PHP.eB at 5e11 vg/ mL, the membranes were fixed with 4% PFA at room temperature for 20 minutes to crosslink the interaction between the capsid and its target protein, followed by 2M HCl treatment at 37ºC for 7 minutes to expose the internal capsid epitope for detection. The blots were then rinsed and incubated with anti-AAV VP1/VP2/VP3 (1:20; American research products, Inc., cat. #03-65158) followed by incubation with a horseradish peroxidase (HRP)-conjugated secondary antibody at 1:5000. The detection of the HRP signal was by SuperSignal West Femo Maximum Sensitivity Substrate using a Bio-Rad ChemiDoc TM MP system #1708280.  37 . Recombinant AAV vectors were administered intravenously via the retro-orbital sinus in young adult male or female mice. Mice were randomly assigned to groups based on predetermined sample sizes. No mice were excluded from the analyses. Experimenters were not blinded to sample groups.
In vivo vector and capsid biodistribution 5-to 6-week-old C57Bl/6J mice, BALB/cJ mice, AAVR WT FVB/ NJ, or AAVR KO FVB/NJ mice were injected intravenously with 10 11 vg of AAV vector packaged into the indicated capsid. One or two hours after injection, the mice were perfused with PBS and tissues were collected and frozen at -80°C. Samples were processed for AAV genome biodistribution analysis by measuring the number of vector genomes (vg) in the sample using absolute qPCR for the GFP sequence. The values are reported as vg/mouse genome after normalizing to the number of copies of host cell genomes using primers recognizing mouse Gcg or human GCG as previously described 6 . To visualize the capsid distribution, mice were perfused with 4% PFA after dosing with AAV vector and brains were sectioned into 100 µM and labeled with the indicated antibodies.

Microscopy
Images were taken on an Axio Imager.Z2 Basis Zeiss 880 laser scanning confocal microscope fitted with the following objectives: PApo 10x/0.45 M27, Plan-Apochromat 20x/0.8 M27, or Plan-APO 40x/1.4 oil DIC (UV) VIS-IR or on a Zeiss Imager.M2 (5x tiled, whole sagittal section images). All images that were compared within an experiment were acquired and processed under identical conditions.
In vitro binding assays Ly6 family members (0.5 µg/well) were transfected into HEK293T cells (3x10 5 /well) using PEI or into CHO cells (1.5x10 5 /well) with lipofectamine 3000 reagent (ThermoFisher, L3000001) in 24-well plates. 48 hours later, the cells were chilled to 4°C and the media was exchanged with fresh cold media containing the indicated recombinant AAV (10 5 vg/cell). One hour later, cells were washed three times with cold PBS, then fixed with 4% PFA for IHC or lysed for genomic DNA extraction and qPCR analyses. Vector genomes bound were normalized to the number of cell genomes by using primers specific to Gcg or GCG. For BMVECs, 2x10 4 cells/well were seeded in 12-well plates the day before exposure to virus. The assay was performed as above except that the AAV vectors were added at 10 6 vg/cell.

Statistical analysis and experimental design
Microsoft Excel and Prism 8 were used for data analysis. For the comparison between AAV9 and AAV-PHP.eB biodistribution, a group size of 6 per group (3 males and 3 females) was used based on prior data that indicated a large effect size. No animals or samples were excluded from the analysis. Supplementary Fig.  1 images are representative of two animals per group. To evaluate AAV-PHP.eB in the 13 mouse lines, AAV9 (n=1, 10 11 vg/animal) or AAV-PHP.eB (n=2; one per dose at 10 11 and 10 12 vg/animal). LY6A IHC in Supplementary Fig. 3 is representative of two animals/line. In vitro transduction and binding experiments display the means from three independent experiments. In Fig. 3D and Fig. 3E, each data point represents a different sgRNA, each averaged from three independent experiments. Data were normalized separately for each AAV capsid to cells transduced with SpCas9 vectors without an sgRNA. Supplementary Fig. 4 presents the same data as