Adeno-associated Virus 9 Structural Rearrangements Induced by Endosomal Trafficking pH and Glycan Attachment

ABSTRACT

Adeno-associated viruses (AAVs) are small nonenveloped single-stranded DNA (ssDNA) viruses that are currently being developed as gene therapy biologics. After cell entry, AAVs traffic to the nucleus using the endo-lysosomal pathway. The subsequent decrease in pH triggers conformational changes to the capsid that enable the externalization of the capsid protein (VP) N termini, including the unique domain of the minor capsid protein VP1 (VP1u), which permits the phospholipase activity required for the capsid lysosomal egress. Here, we report the AAV9 capsid structure, determined at the endosomal pHs (7.4, 6.0, 5.5, and 4.0), and terminal galactose-bound AAV9 capsids at pHs 7.4 and 5.5 using cryo-electron microscopy and three-dimensional image reconstruction. Taken together, these studies provide insight into AAV9 capsid conformational changes at the 5-fold pore during endosomal trafficking, in both the presence and absence of its cellular glycan receptor. We visualized, for the first time, that acidification induces the externalization of the VP3 and possibly VP2 N termini, presumably in prelude to the externalization of VP1u at pH 4.0, which is essential for lysosomal membrane disruption. In addition, the structural study of AAV9-galactose interactions demonstrates that AAV9 remains attached to its glycan receptor at the late endosome pH 5.5. This interaction significantly alters the conformational stability of the variable region I of the VPs, as well as the dynamics associated with VP N terminus externalization.

IMPORTANCE There are 13 distinct Adeno-associated virus (AAV) serotypes that are structurally homologous and whose capsid proteins (VP1 to −3) are similar in amino acid sequence. However, AAV9 is one of the most commonly studied and is used as a gene therapy vector. This is partly because AAV9 is capable of crossing the blood-brain barrier and readily transduces a wide array of tissues, including the central nervous system. In this study, we provide AAV9 capsid structural insight during intracellular trafficking. Although the AAV capsid has been shown to externalize the N termini of its VPs, to enzymatically disrupt the lysosome membrane at low pH, there was no structural evidence to confirm this. By utilizing AAV9 as our model, we provide the first structural evidence that the externalization process occurs at the protein interface at the icosahedral 5-fold symmetry axis and can be triggered by lowering the pH.

INTRODUCTION

Adeno-associated viruses (AAVs) are single-stranded DNA (ssDNA) T=1 icosahedral viruses belonging to the Dependoparvovirus genus of the family Parvoviridae (1). The vast majority of the Dependoparvovirus genus members are capable of successful replication in nondividing (non-S-phase) cells due to helper activity provided by concurrent larger DNA virus infection (1, 2). Currently, 13 human and nonhuman primate AAV serotypes have been classified, with the exception of AAV5, to the species Adeno-associated dependoparvovirus A. In addition, more than 100 other variants have been isolated from both human and nonhuman primate tissues and samples (3–6). The Adeno-associated dependoparvovirus A species encodes capsid proteins (VPs), which exhibit a minimum of 50%, typically >80%, sequence identity (7). Despite the high conservation of the VP sequence, AAVs have been shown to differ in host serum protein interactions, cellular receptors, and trafficking and transduction efficiency in a broad range of tissues (5, 8–10). In addition, AAVs are readily capable of packaging genomic and nongenomic ssDNA yet have not been associated with any pathology (7, 11, 12). Taken together, these characteristics have led to the development and use of AAVs as gene therapy vectors for several monogenetic diseases, resulting in AAV-mediated gene therapies for clinical use (13–15).

AAV9 is one of the most promising serotypes for gene therapeutic applications because of the wild-type (WT) and recombinant construct capsid characteristics (16–19). AAV9 transduces a wide range of tissue types, including cardiac and skeletal muscle, liver, pancreas, and eye (18, 20–24). Furthermore, AAV9 can cross the blood-brain barrier and targets the central nervous system with greater efficiency than other AAV serotypes (24–27). Also, in contrast with serotypes AAV1 and AAV2, two of the other frequently used therapy vectors, AAV9 has significantly lower seroprevalence in the human population, which makes it an even more desirable candidate for therapeutic applications (28).

The AAV T=1 icosahedral capsid is assembled using 60 subunits of the structural virus proteins (VPs) VP1, −2, and −3, in an approximate ratio of 5:5:50 (29, 30). The VPs are all products of the structural protein-encoding open reading frame of the genome, designated cap, VP1 (∼82 kDa) and VP2 (∼73 kDa), which are the minor capsid proteins, and VP3 (∼61 kDa), the major capsid protein. Due to the utilization of both alternative splicing and leaky scanning, when expressed, the individual VPs share a C terminus that encompasses the entire VP3, while VP1 and VP2 are N-terminal VP3 extensions. VP1 and VP2 share a region of ∼73 amino acids (aa), which is extended by an additional ∼137 amino acids in VP1, designated the VP1 unique region (VP1u) (31). The AAV VP1u, like most parvoviruses, contains a phospholipase A2 (PLA2) domain, a calcium-binding loop, and a nuclear localization signal, all of which are essential for infection (32, 33).

The capsid structure of AAVs has been widely studied using both X-ray crystallography and cryo-electron microscopy (cryoEM) (34–43). Due to the lower copy numbers and possible disorder, the VP1u, VP2 common region, and first N-terminal ∼15 amino acids of the VP3 have so far not been observed in AAV capsid structures. The ordered VP3 comprises an eight-stranded antiparallel β-jelly roll core (βB to βI), which is N-terminally complemented by an additional βA strand, consistent with the general parvovirus capsid architecture (44). The β-strands are connected by loops of various lengths, designated by the two β-strands making the connection (e.g., connecting loop between the βD and βE stands, designated the DE loop). Previous studies, have further identified these loops to contain nine variable regions (VRs), responsible for most of the sequence and surface structural variance among the different AAV serotypes (29, 35, 44). Apart from the VRs, the AAV surface morphology is highly conserved, displaying a depression at the icosahedral 2-fold, finger-like protrusions at the 3-fold, and a pore, extending into the interior of the capsid, at the 5-fold axes (44).

The AAV capsid, apart from packaging, protecting, and delivering the viral genome, also serves as the attachment surface to the viral host cell receptors, which is responsible in part for determining the cell and tissue tropism of AAVs. Many AAV serotypes have been shown to interact with glycosylated cell surface proteins for viral absorption, which indicates cell surface glycans are the capsid’s first contact receptors. Although the majority of AAVs are heparan sulfate proteoglycan binders, e.g., AAV2, AAV3, and AAV13 (45–48), sialic acid binders, e.g., AAV1, AAV4, and AAV5 (49–51), or even both, e.g., AAV6 (50, 52), AAV9 has been shown to be a unique terminal galactose binder (53, 54) and has also been described to transduce cells in the presence of the 37- to 67-kDa laminin receptor (55). Recently, however, a heavily glycosylated multiserotype AAV receptor (AAVR) was identified using a genome-scale genetic screen, which directly interacts with the AAV capsid via immunoglobulin-like (Ig-like) polycystic kidney disease (PKD) domains present in the AAVR ectodomain (9, 56). This AAVR has also been demonstrated to be essential for AAV9 transduction in vivo in a knockout mouse model (56).

AAVs have been shown to enter the cytoplasm enclosed in clathrin-coated endocytic vesicles (57). Consequently, the capsids remain trapped inside the early endosome and traffic through the endosomal-lysosomal network (58, 59). The activity of the enzymatic PLA2 domain is required to disrupt the lysosome membrane prior to capsid release into the cytoplasm. Exposure to the acidification of the endosome has been deemed essential for AAV nuclear entry, which is followed by capsid uncoating and eventually replication (59–61). In the case of AAV8, amino acid conformational changes in the capsid structure have been shown to occur as the pH is reduced (62). It has been detected that the pH reduction of the environment triggers the externalization of the N-terminal VP regions, including the PLA2 domain and the nuclear localization signal (NLS) of VP1u (59, 63–65). These significant conformational changes, however, have never been directly structurally observed.

Here, we characterize the AAV9 capsid’s structural changes, which probably occur during trafficking, through the endosomal-lysosomal system in vitro. To this end, we have resolved the near-atomic three-dimensional (3D) structure of the AAV9 capsid using cryoEM at four different pHs, modeling the endosomal acidification stages from cytoplasmic entry through early and late endosome, to mature lysosome. Moreover, we have structurally characterized, for the first time, the AAV9 capsid interaction with terminal galactose. We have also resolved the galactose-bound AAV9 capsid structure, by cryoEM, at both the cytoplasmic and the late endosomal pH. This is the first study to characterize, in an in vitro setting, how AAV cytoplasmic trafficking is affected by the presence or absence of its potential receptor, a glycolyzed peptide.

RESULTS AND DISCUSSION

CryoEM reconstruction of the AAV9 capsid structure at four different pHs.

WT AAV9 virus like-particles (VLPs) were purified and dialyzed to four different pHs, mimicking those encountered during endosomal maturation as follows: (i) near neutral, cytoplasmic pH of 7.4, modeling the moment of endocytic compartment formation, (ii) pH 6.0, corresponding with the mild acidic environment of the early endosome, (iii) pH 5.5, mimicking the late endosomal acidic environment, and (iv) pH 4.0, the highly acidic environment of the mature endosome. This experimental setup has been successfully applied previously to investigate pH-induced structural changes in AAV8, as well as to examine AAV capsid-associated protease activity (62, 66, 67). VLPs of each pH stage remained intact and, consequently, could be vitrified and subjected to data collection and structure determination by cryoEM.

We have obtained near-atomic, high-resolution 3D structures of the WT AAV9 capsid at the four pHs examined, with the comparable resolutions of 2.8 (pH 7.4, PDB ID 7MT0), 2.7 (pH 6.0, PDB ID 7MTG), 2.8 (pH 5.5, PDB ID 7MTP), and 3.0 Å (pH 4.0, PDB ID 7MTW) (Fig. 1). The processing and refinement statistics of the data collection, image reconstruction, and structural modeling are summarized in Table 1. Although AAV9 capsid structures have been investigated at high resolution before, using X-ray crystallography (PDB ID 3UX1) (37), this is the first time that high-resolution cryoEM structures have been obtained for this serotype.

FIG 1.

Structural studies of the AAV9 capsid at four different pHs, representing the acidification stages of endosome maturation. All four capsid structures were resolved using cryo-electron microscopy and three-dimensional image reconstruction. The results of each reconstruction are presented as a surface view (left) as well as a cross-section view of the AAV9 empty particle (right), orientated at the 2-fold symmetry axis as demonstrated by the icosahedral line diagram. An asymmetric unit of the AAV9 capsid is marked in the case of the pH 7.4 structure. The luminal opening of the 5-fold channel, located at the 5-fold symmetry axis and covered by the basket-like density at pHs 6.0, 5.5, and 4.0, is indicated by an arrow. Each map is radially colored, shown at a contour of 1ơ, and is corrected with a B-factor of 50 Ų. A piece of example density, encompassing the atomic model of the AAV9 capsid, is shown below each reconstruction at a ơ of 4. The 5-fold symmetry axis is marked by a pentagon, the 3-fold axis by triangles, and the 2-fold axis by an ellipsoid, in case of the pH 7.4 structure.

TABLE 1.

Reconstruction processing and model refinement statistics

Processing and refinement parametersa	Data for:
	AAV9 pH 7.4	AAV9 pH 6.0	AAV9 pH 5.5	AAV9 pH 4.0	AAV9 pH 7.4 + galactose	AAV9 pH 5.5 + galactose
Total no. of micrographs	1,411	1,405	1,502	1,325	1,044	1,445
Reconstruction software	Auto3DEM	Auto3DEM	Auto3DEM	Auto3DEM	cisTEM	cisTEM
Defocus range (μm)	0.90–4.50	0.86–4.14	0.80–4.38	0.08–4.48	0.57–2.57	0.56–2.58
Electron dose (e−/Å2)	60	60	60	60	63	63
Frames/micrograph	50	50	50	50	50	50
Pixel size (Å/pixel)	0.974	1.087	1.087	1.086	1.049	1.048
Starting no. of particles	188,074	119,322	110,989	91,392	77,436	134,697
No. of particles used for final map	150,469	107,405	99,899	82,290	56,370	110,690
B-factor used for final map (Å2)	50	100	100	100	20 (postcutoff B-factor)	20 (postcutoff B-factor)
Resolution of final map (Å)	2.82	2.67	2.79	2.99	2.43	2.68
PHENIX refinement statistics
Residue range (VP1)	219–736	219–736	219–736	219–736	219–736	219–736
Map correlation coefficient	0.6528	0.629	0.6115	0.7147	0.7126	0.7718
RMSD (bonds) (Å)	0.010	0.008	0.011	0.009	0.012	0.011
RMSD (angles) (Å)	0.852	0.788	0.863	0.772	0.921	0.910
All-atom clash score	12.19	8.43	12.30	9.19	7.74	8.39
Ramachandran plot
Favored (%)	97.1	96.7	96.7	97.3	96.7	97.7
Allowed (%)	2.9	3.3	3.3	2.5	3.1	2.1
Outliers (%)	0	0	0	0.2	0.2	0.2
Rotamer outliers (%)	0	0	0	0	0.4	0
C-β deviations	0	0	0	0	0	0
RCSB PDB ID	7MT0	7MTG	7MTP	7MTW	7MTZ	7MUA

^aRMDS, Root mean square deviation.

Similarly, for all AAV cryoEM capsid structures reconstructed to date, the ordered residue range corresponded with VP1 residues 219 to 736, with the first 16 amino acids of the VP3, the VP1-VP2 common region, and all of VP1u not visible, and therefore not modeled in the structures (44). Given the resolution, the density was adequately ordered to model and reconstruct side chain orientations of the VP structure (Fig. 1).

However, two regions were disordered, VRI and the apex of the BC loop. Previously, VRI was reported to be associated with the blood-brain barrier (BBB) crossing ability of both AAV9 and AAVrh10 (26, 68). Although these VRIs significantly differ in conformation between the two capsids, the conservation of both Ser368 and Ser369 suggests these amino acids to be the key residues of the BBB crossing ability (40). VRI was consistently associated with “poor” density quality, regardless of pH, making the assignment of side chain orientation uncertain. With the density of the VRI main chain visualized either at ơ of 1.9 (pH 7.4, 5.5, and 4.0) or lower. The high number of Gly and Ser residues of the poorly ordered amino acid string (STSGGSS), which are associated with high flexibility and therefore most likely a mobile region, possibly explains the disorder.

The second region of similar disorder was located at the apex of the DE loop, comprising VRII, corresponding with the exterior tip of the 5-fold channel. In this region, again, independent of pH, backbone density could only be visualized at a ơ of 2.5, as a result of missing densities for Asp327 to Val331 (DNNGV) at values higher than 2.5. At ơ values of <2, most likely as a result of structural movement, the density became diffuse and made it difficult to be confident about the conformation of the backbone.

The overall surface morphology of all four structures displayed a highly similar appearance, with differences limited exclusively to the area directly surrounding the 5-fold pore (detailed below). The pH 7.4 capsid structure was superimposable to the previously determined AAV9 crystal structure (37).

The 5-fold symmetry axis of the AAV9 capsid changes as pH becomes acidic.

The parvoviral icosahedral 5-fold symmetry axis consists of five radially orientated DE loops, forming a cylindrical channel and a pore, which is large enough to allow solvent movement in and out of the capsid. This portal is believed to be the route of genomic DNA packaging and uncoating, as well as the location of VP1u externalization during endo/lysosomal trafficking after cell entry (59, 64, 65, 69, 70).

Examination of the cross-section view of the AAV9 capsid maps reveals a “basket-like” extension of density in the interior of the capsid at the base of the 5-fold axis (Fig. 1). The basket contributes a significant density mass covering the entire floor of the 5-fold axis pore which extends from the first ordered VP1 residue of each 5-fold subunit. Interestingly, this density exhibited different morphologies for each of the three acidic pH structures, which is absent in the neutral, pH 7.4, capsid structure (Fig. 1, Fig. 2A). Moreover, correlating with the presence of the “basket,” the 5-fold channel itself becomes occluded with a “column-like” density (Fig. 2A).

FIG 2.

pH-dependent structural dynamics of the AAV9 5-fold axis. (A) Close-up cross-section view of the AAV9 5-fold symmetry axis, displaying various extents of the column-like density inside, and of the basket-like density below the 5-fold channel, depending on the pH of the capsid environment. Density maps are radially colored, following the same scale as in Fig. 1 and are displayed at a ơ of 1. (B) Reconstructed atomic model of the AAV9 partial VP3 and VP2 N termini, modeled into the column-like density at pH 4.0 at the residue range from Thr201 to Asp213 (VP1 numbering). The channel-related density of each map is indicated by a mash of different colors for each reconstruction, visualized at the ơ of 2.5. (C) Surface view of the AAV9 5-fold symmetry axis, encompassing the density map of the HI and the DE loops. The density map is shown with the same parameters as in panel A.

The channel structure of the pH 7.4 capsid, similarly to that of the X-ray crystal structure, only exhibits a small, elongated “blob-like” unmodeled density, which can be visualized only at a contour level of 2.5 ơ and lower. Consequently, the pore base density is also minor, just occupying the space between the channel “blob” density and the first ordered N-terminal residue. In contrast, in the pH 6.0 capsid structure, the 5-fold channel is fully occupied with the column-like density and is visualized at a contour ơ level of 3.5, attached to and directly extending from a broad and prominent inward protruding basket. The 5-fold pore column density is still present in the pH 5.5 structure, though not continuous and less prominent, possibly as the consequence of movement and flexibility. Finally, at pH 4.0, the column structure is at its longest expansion, with its apex reaching the top of the DE loop, and consequently, the basket narrows and appears to have moved inward into the channel (Fig. 2A).

The presence of column-like density occupying the 5-fold channel has been observed before in both AAV crystal- and cryo-EM structures, all of which were determined at neutral pH (34, 39, 40). In the case of AAV8, AAVrh.8, and AAVrh.39, the presence of a density-filled 5-fold channel was associated with genome-packaged particles, yet AAVrh.10 displayed the same characteristic in the absence of a packaged genome. In all cases, this density has been interpreted as the highly flexible Gly-rich N-terminal region of the VP2, overlapping the VP3 N terminus. Based on the morphology and elongation of the density, the amino acid stretch of Thr201 to Asp213 (TMASGGGAPVAD) was modeled into the AAV9 pH 4.0 channel density (Fig. 2B). This loosely corresponds to the region previously assigned to the similar observed column density in the case of AAVrh.8, AAV8, AAVrh.10, and AAVrh.39 (34, 39, 40). Our results suggest that the AAV9 capsid, even in the absence of a packaged genome, undergoes conformational changes when exposed to an acidic pH, which results in the externalization of the VP3 N terminus along with the VP1-VP-2 common region. It is yet to be determined if this corresponds with the simultaneous externalization of the VP1u, encompassing the PLA2 domain. It should be noted, nevertheless, that the 1:1:10 incorporation ratio of the VP1, 2, and 3, respectively, also implies that, on average, six VP1 subunits are assembled into an AAV9 capsid; hence, the VP1u is expected not to conform to the icosahedral nature of the capsid and averaged density map. The reduced order, in the case of the pH 5.5 capsid structure channel density, is intriguing, as it could potentially mean that additional capsid conformational changes might occur at acidic pH, leading to the externalization of the VP1u. The observation that the AAV lysosomal escape and infectivity require exposition to low pH corroborates this concept (59, 63). The 5-fold channel density, in the case of the pH 4.0 capsid structure, is slightly more extended, wider, and more ordered than in the pH 6.0 structure, which could be the consequence of the VP1u externalization (Fig. 2A and B). Interestingly, this is the first time that potential VP1/2 N-terminal externalization may have been observed in an AAV capsid in vitro; however, no such observation was noted in the case of the AAV8 or AAV2 structure (59, 62).

The basket-like density, at the base of the 5-fold channel, was first reported in VLP structures of members from another Parvovirinae genus, Bocaparvovirus (71, 72). Since then, it has also been observed in AAVs, including the empty AAVrh.39 and the antigen-binding fragment (FAB)-AAV5 VLP complex capsid structures (40, 73). In these cases, the presence of the basket facilitated the modeling of additional N-terminal residues for as many as three amino acids (40, 72, 73). In the case of AAVrh.39, the basket was not present once genome was packaged into the capsid, but was present with the column-like density inside the 5-fold channel (40). The presence of the AAV9 basket was not accompanied with increased order of additional N-terminal residues. As a result of the presence of additional N-terminal residues being associated with the observed density inside the 5-fold channel, it is a possibility that the condensation of the observed basket density is the consequence of protein accumulating at the base of the 5-fold symmetry axes as an effect of the flexibility and movement of the VP N-terminal regions. The variability in its appearance in the case of these acidic AAV9 pH structures corroborates this rationale.

The pH-dependent structural changes at the 5-fold symmetry axis were also accompanied by the observation of increased disordering associated with the VRII region of the DE loop as the pH decreased (Fig. 2C). Interestingly, this did not affect the conformation of the HI loop, the other conserved loop associated with the parvoviral 5-fold axis. The extent of the broadening density, reaching its largest diffusion at pH 4.0 and covering the 5-fold pore completely, positively correlated with that of the density changes observed in the interior of the channel. This suggests that the extra density present is the result of (i) movement of the flexible DE loop to accommodate the VP N-term externalization, (ii) the movement of the VP N termini on the apex of the DE loop, following externalization, or (iii) perhaps both. A similar phenomenon can be observed in the case of the AAV8, AAVrh.10, and AAVrh.39 genome-packaged capsids compared to empty capsids (40). In both the AAV8 and AAVrh.39 structures, this observation is accompanied by the presence of the observed column-like density inside the 5-fold channel.

pH-mediated conformational changes of the AAV9 capsid.

Apart from the observed density pH-dependent differences linked to the 5-fold channel and associated regions, the ordered region of the structural proteins also displayed some subtle conformational changes that could be localized in the vicinity of the icosahedral symmetry axes.

Superimposition of the Cα backbone from the ordered region of the AAV9 VP3 pH structures indicated that the backbone itself did not undergo significant conformational changes in response to the pH changes, similarly to what was observed in the crystal structure of AAV8 (62) (Fig. 3A). In the case of the four AAV9 structures, the root mean square deviation (RMSD) of the superimposed Cαs was ∼0.4 Å, with the pH 4.0 structure being the least similar to the neutral pH structure with a Cα RMSD of 0.5 Å. These differences in RMSD could be ascribed to three capsid regions, corresponding to VRs I, II, and VI. As these regions are associated with poorly ordered density, these differences could be attributed to the uncertainty of model building and do not represent conformation differences at the Cα level.

FIG 3.

pH-related conformational changes of the AAV9 capsid. (A) Superimposition of the AAV9 capsid monomer ribbon diagrams, reconstructed at four different pHs. The arrows indicate dynamic regions associated with poorly ordered density. (B) Atomic model and associated density of the AAV9 pH quartet residues, comparing sidechain density strength and sidechain conformation changes as the pH declines. (C) Side chain dynamics of AAV9 luminal histidines. Homologues of H229 and H630 harbor pH-sensitive side chain conformation in the case AAV8. The shift of the H423 side chain is unique for AAV9.

Previously, the 2-fold symmetry axis of AAV8 was shown to be a pH-sensitive interface, identifying four residues that undergo conformational changes with pH, designated the “pH quartet” (62). The pH quartet amino acids are conserved throughout all the AAV serotypes and include residues Arg392 and Tyr707 facing toward His529 and Glu566 on the 2-fold-neighboring subunit (AAV8 numbering). In the AAV8 capsid, at neutral pH, Glu566 (Glu564 in AAV9) forms hydrogen bonds with Arg392 (Arg391 in AAV9) and Tyr707 (Tyr705 in AAV9) at the 2-fold interface. In AAV8, as the pH becomes more acidic, the Glu566 side chain density becomes disordered, as it displaces these hydrogen bond interactions, and at pH 4.0, Glu566 completely shifts conformation to interact with the side chain of His529 (His527 in AAV8) and regains its side chain density. Consequently, the now uncoordinated side chain of Tyr707 rotates to face away from the interface. This conformational shift significantly reduces the 2-fold interface buried area of AAV8 at pH 4.0 (62).

Interestingly, the AAV9 pH quartet also exhibits a pH-induced conformational change (Fig. 3B) but differs from that reported for AAV8 (not shown, refer to reference 62). At pH 7.4, the side chain density of Glu564 is disordered, which suggests the lack of AAV8-like hydrogen bonds seen in Glu566. As the pH decreases, however, Glu564 gains side chain density, becoming more ordered at pH 6.0, which probably indicates a more stable interaction at the interface. The density for Glu566 is again disordered at pH 5.5 yet is reacquired at pH 4.0, as Glu566 hydrogen bonds with His527, similarly to its counterparts dynamics in AAV8. Tyr705, however, does not flip its side chain conformation as seen for Tyr707 in AAV8 and instead adopts a dual conformation. This implies that Tyr705 is contributing to stabling interaction only 50% of the time. Table 2 summarizes and compares the surface-buried area of the AAV9 and 8 capsid interfaces at pH 4.0 and 7.4, indicating that the magnitude of the 3- and 5-fold interfaces are conserved in both serotypes. However, the AAV9 2-fold interfaces differ by ∼100 Ų in interacting area size, depending on the distal or proximal conformation of Tyr705. The decrease in the 2-fold interface surface area of AAV8 is suspected to be a transition to facilitate uncoating (62). It is intriguing, nonetheless, why AAV9 has evolved to partially maintain the integrity of the 2-fold interfaces and suggests maybe serotype-specific adaptation in intracellular trafficking.

TABLE 2.

Buried surface area (Ų) involved in the AAV9 and AAV8 multimer interactions

Parameter	AAV9			AAV8
	pH 7.4	pH 4.0		pH 7.4	pH 4.0
Dimer	1,370	1,370a	1290b	1,530	1,460
Trimer	4,870	4,860		5,010	4,820
Pentamer	2,390	2,380		2,380	2,420

^aTyr705 side chain oriented toward the 2-fold interface.

^bTyr705 side chain oriented away from the 2-fold interface.

The AAV capsid has been shown to possess a protease enzymatic activity, which is capable of cleaving both an external substrate and the AAV capsid proteins (66, 67). However, this activity is diminished in AAV2 upon the substitution of Glu564 to alanine (66). The difference in the pH quartet conformational changes of AAV8 and 9 might therefore also be related to differences in protease activity between the AAVs, though this requires further experimental studies to confirm. In addition, both Glu564 and Tyr704 have been shown to be responsible for a nonstructural function of the AAV2 capsid, as their presence is required for both successful transcription and second-strand synthesis during AAV2 replication (74). Given the differences in the AAV9 pH quartet structural dynamics and AAV8, it is possible that this nonstructural activity harbors slightly different capsid characteristics as well.

Proximal to the 5-fold axis, facing toward the interior of the capsid, all the AAV serotypes have a conserved histidine, His229 in case of AAV9. In concordance with the pH sensitivity of the 5-fold axis, His229 appears to adopt dual conformations, as the pH is reduced to 6.0 (Fig. 3C). However, the equivalent residue was shown not to be pH sensitive in AAV8, as it possessed the dual side chain conformation at all four pHs (62). Dual conformations were also observed in the case of the empty AAVrh.10 capsids as well, yet this shifted back to a single side chain orientation in the genome-packaged particles (40). His229 has been shown to cause assembly-deficient mutants when changed to alanine (75). Considering these observations, it is possible that His229, given its luminal position, may play a role in AAV9 pH-dependent DNA metabolism, which warrants further investigation.

His629 of AAV9 is the homologue of AAV8 His632, situated in a small luminal pocket directly under the 3-fold symmetry axis. His632 in AAV8 is pH-sensitive and adopts dual side chain conformation as the pH declines (62). In AAV8 VLPs, moreover, His632 and Pro422 coordinated additional density, which was modeled as a dinucleotide consisting of a purine and a pyrimidine. These ordered nucleotides have been observed in multiple AAV serotypes, regardless of being ssDNA full, empty, or VLP capsids (34–36, 38–41, 62). As His632 undergoes a pH-induced conformational change, the ordered nucleotide density fades in the case of AAV8. This homologous pocket in the AAV9 cryoEM capsid structures is void, regardless of pH, similar to what was observed for the AAV9 crystal structure (37). His629, however, displays the dual conformation in each structure from pH 7.4 to 4.0 (Fig. 3C). Interestingly, the nucleotide-binding pocket is completely conserved between the AAV9 and AAV8 structures, the side chain orientation of His632 being the only difference. It is possible that His229, due to its mobility, makes it inaccessible to bind DNA, and hence the lack of ordered nucleotide density.

Also, in the proximity of the 3-fold axis, another luminal histidine was seen to be pH sensitive (Fig. 3C). His423 is the neighboring residue of Phe422, a pattern similar to His629 and Phe628 (AAV9 numbering, and seen in all AAV serotypes). Although His423 and Phe422 are also conserved in AAV serotypes, neither has been associated with a specific function thus far. His423 adopts a dual conformation at pHs 6.0 and 5.5, and it is a possibility, given its position, that it also plays a role in capsid-DNA interactions, similar to the His629 homologues.

Characterization of the galactose-bound AAV9 capsid reveals receptor binding-induced structural changes.

To determine how binding to the terminal glycan of its receptor would affect the AAV9 capsid structure, purified AAV9 VLPs were complexed with terminal galactose in the form of N-acetylgalactosamine (Gal-NAc). To investigate the pH-dependent dynamics of receptor-bound AAV9, the conjugated VLPs were dialyzed into buffers of pH 7.4 and pH 5.5, similarly to the VLPs used for the previous pH experiments detailed above. pH 5.5 was selected to represent the acidic pH for two reasons; (i) at this pH, the most dynamic structural changes were observed in the AAV9 capsid structures, and (ii) the AAV9 capsid had maximal stability at this pH, based on previous studies (76). Following data collection of both AAV9-galactose complex populations, the high-resolution 3D capsid structure of both were obtained, namely, 2.7 Å resolution at pH 7.4 (PDB ID 7MTZ) and 2.4 Å resolution at pH 5.5 (PDB ID 7MUA) (Fig. 4A). The overall surface morphology of both capsids appeared to be highly similar to that of the corresponding native structures. Interestingly, neither of these structures appeared to possess the basket-like density under the 5-fold channel (Fig. 4A).

FIG 4.

Structural analysis of the AAV9 capsid in complex with terminal galactose. (A) surface (left) and cross-section (right) views of the galactose-bound AAV9 3D capsid structure, reconstructed at neutral and acidic pH, representing the environment of cytoplasm and late endosome, respectively. (B) Close-up view and comparison of the AAV9 galactose-binding pocket at the respective pHs with that of the non-galactose-bound, native AAV9 cryo-EM structures. Side chain atomic models are displayed for the residues previously identified to be essential for AAV9 glycan binding. The terminal β-galactose molecule, inserted in the corresponding density, is shown in cyan. The map density is contorted at a ơ of 1. (C) Corresponding density and ribbon diagram, in the presence and absence of AAV9-bound galactose, of the BC loop peak, which includes variable region (VR) I. Galactose binding alters the ordering of VRI from that of the corresponding native structures. Density is visualized at a ơ of 3 in the case of the pH 7.4 and at a ơ of 4 in the case of the pH 5.5 structures.

Previously, using site-directed mutagenesis and cell-binding, five residues of the AAV9 VP3 were identified which contributed to and were necessary for galactose binding (77). These residues, Asn470, Asp271, Asn272, Tyr446 and Trp503, are located in a pocket in the proximity of the 3-fold axis, on the surface wall of the 3-fold spikes. Upon examination of this pocket in both the pH 7.4 and pH 5.5 AAV9-galactose complex structures, it was seen to be occupied by a small unmodeled density, coordinated by the five residues identified in the previous study (Fig. 4B). In contrast, the corresponding pocket was void of any residual density in of all the native structures. Although the density was poorly defined, its size and shape were in agreement for a ring-shaped β-galactose molecule to be fitted to both the pH 7.4 and pH 5.5 capsid structures (Fig. 4B). The density profile indicated an interaction between the galactose molecule and Trp503; Trp503 is the only residue to be in direct contact with the glycan. These findings suggest that the AAV9 capsid binds the cell surface receptor and uses it for its cellular entry (pH 7.4) and is still possibly attached during the late endosome (pH 5.5), targeted toward the trans-Golgi network (78). Although these are the first AAV structures to be complexed with galactose, the sialic acid-bound capsid structure has been resolved for serotypes AAV1 and AAV5 (79, 80), as well as the heparin-bound capsid structure of AAV2 (81, 82). In both cases of the high-resolution sialic acid- and the low-resolution heparin-bound structures, the associated glycan had a significantly more ordered disposition. This suggests that the AAV9-galactose interaction might be flexible and transient compared to other AAV-glycan interactions. This might have significance during receptor attachment and internalization, or alternatively, a stabile interaction requires the entire receptor protein to be present, not just the terminal galactose.

In contrast to the less ordered nature of the galactose density, the glycan-binding significantly stabilized the conformation of the flexible VRI, which is directly followed by two residues of the galactose binding pocket, namely, Asp271 and Asn272 (Fig. 4C). At pH 7.4, continuous, well-ordered density appears for the residues from Ser263 to Ser269 at 3.5 ơ, in contrast to the weak corresponding density of the pH 7.4 native structure. Although at pH 5.5 the VRI density is better ordered in the case of the native structure, galactose binding renders the VRI density ordered at ơ of 4.0 when subjected to the pH 5.5 environment (Fig. 4C). Considering the significance of VRI, especially Ser369, in crossing the BBB (26, 40, 68), these observations suggest that receptor attachment might be important for this process in order to stabilize the structure of the otherwise flexible VRI. Given this is the case, the flexibility of VRI might have been selected for during AAV9 evolution to ensure that this serotype can only be internalized if their terminal glycan stabilizes the structure of the VRI-containing loop, resulting in transduction specificity to cell types with intact and accessible terminal galactose only.

Terminal galactose binding to the AAV9 capsid alters pH-dependent dynamics at the 5-fold axis.

Upon examining the 5-fold symmetry axis in both the pH 7.4 and pH 5.5 galactose-bound AAV9 structures, similar to the native structures, it also undergoes conformational changes as the pH is reduced (Fig. 5). Interestingly, however, these changes were different from the response of the non-galactose-bound, i.e., native, AAV9 capsid 5-fold dynamics. Most notably, despite a large and substantially sized basket under the channel of the pH 5.5 native structure, the galactose-bound pH 5.5 structure lacks the basket-like density almost completely (Fig. 5A).

FIG 5.

The attachment of terminal galactose to the AAV9 capsid alters the dynamics of the 5-fold symmetry axis as pH declines. (A) Comparison of the 5-fold channel cross-section density maps in the presence and absence of bound galactose under corresponding neutral (pH 7.4) and acidic (pH 5.5) environments. The maps are radially colored, identical to that of Fig. 1, and are corrected with a B-factor of 50 and visualized at a ơ of 1. (B) Direct comparison of aligned map densities in the presence and absence of bound galactose at both pHs of resolved AAV9 galactose-bound and native structures. The density is displayed as a mash of different colors. The maps are corrected with a B-factor of 100 to visualize regions that are more or less ordered and are contorted at a ơ of 1. The DE loop, forming the 5-fold channel, is shown as ribbon diagrams.

The column-like density of the galactose-bound capsids displays different structural states at the 5-fold channel compared to each other and their native capsid structure counterparts. The galactose binding resulted in the presence of a longer and wider density column in the 5-fold channel of the pH 7.4 structure, which diminishes into pieces of unordered density toward the surface opening of the channel, unlike the small blob-like density of the native pH 7.4 structure, which is completely void further outwards (Fig. 5A and B). At pH 5.5 the galactose complex capsid structure channel is fully filled with density, which is very similar in its appearance to that of the native pH 4.0 structure, and not the native pH 5.5 structure. Interestingly, the ordering of the apex of the DE loop, comprising VRII, is very similar in both the galactose-bound structures, harboring the attached extra density, as seen in case of the native pH 5.5 structure. Ascribed to this, the 5-fold pore is almost completely closed in both structures, in contrast to the open 5-fold pore in the native pH 7.4 structure (compare Fig. 2 and 5).

Taken together, these findings suggest that the receptor attachment changes the dynamics of the AAV9 5-fold in response to decreasing pH, rendering the 5-fold channel density more ordered. Considering the observation that the AAV9 capsid might remain attached to its receptor via the terminal galactose throughout its endosomal trafficking stages, it is a possibility that this interaction is in fact required to ensure the correct time point and conformation of the VP N terminus externalization, at lower pH. The mechanism of how this change in 5-fold-related structural dynamics is mediated by receptor attachment is currently unresolved, though it most certainly remains an intriguing new aspect of AAV host-virus interactions. Interestingly, changes in the nature of the basket density have been observed previously in the case of the AAV5 capsid, at neutral pH. AAV5 only displays a 5-fold-associated basket-like density if attached to the FAB of the IgG molecule (73), which is in contrast to the absence of basket-like density in case of the galactose-attached AAV9 VLPs. In prospective, AAV capsid interactions with cell surface receptors and molecules of the host immune system might have a wider influence modulating capsid-related trafficking dynamics, even after internalization and en route to the nucleus.

Conclusions.

AAV9 is one of the most investigated AAV serotypes in a gene therapy context, yet little is known about its intracellular trafficking properties, especially when pH-mediated structural transitions are concerned, as a consequence of exploiting the endosomal-lysosomal pathway to reach the site of replication. These fundamental aspects of AAV9 biology, however, are essential for the understanding of how potential AAV9-based vectors behave in interactions with the targeted host cell.

By investigating it in an in vitro setting that highly mimics the biological environment, we have structurally characterized, for the first time for any AAV serotype, that the lowering of pH results in the externalization of the VP3 and VP2 N termini, in concordance with previous indirect experimental findings (63–65). The most dynamic 5-fold-related structural movement appeared to be at pH 5.5 and structurally confirmed the previous finding that VP1 N-terminal externalization is associated with low pH, consistent with the environment of the late endosome (59, 63–65). Moreover, the AAV9 capsid shell itself appears to undergo conformational changes, as a result of the changing pH, which differ from those seen in case of AAV8, the only AAV serotype thus far for which pH-mediated structural changes have been characterized (62). These findings suggest that there might be more variation in evolved trafficking strategies between AAV serotypes than previously thought.

We have structurally characterized, for the first time, the interactions of a galactose-binding AAV serotype with its receptor, terminal galactose, and established that the interaction is probably more dynamic than those of other AAV serotypes with sialic acid or heparin (79–82). Furthermore, the interaction between the AAV9 capsid and its receptor appears to be maintained even at acidic pH, while the capsid is trafficking enclosed in the late endosome. Lastly, our findings suggest that AAV9 receptor attachment is important in determining the conformation of certain regions of the AAV9 capsid, including VRI and the 5-fold-related dynamics, implying that the glycan interaction harbors a wider scale of capsid-related functions than virus absorption and internalization.

Taken together, our results shed light on new aspects of AAV endosomal trafficking, which opens up new challenges and issues to be considered when developing gene therapeutic applications, whether utilizing AAV9 or any other serotype.

MATERIALS AND METHODS

Virus capsid production and purification.

To express the recombinant AAV9 virus-like particles (VLPs) the Bac-to-Bac expression system (Invitrogen, Thermo Fisher Scientific) was used. Bacmids including the complete AAV9 capsid protein-encoding gene (cap) were transfected into Sf9 (ATCC CRL-1711) cells and seeded at a density of 8 × 10⁵ cells per well on a 6-well culturing dish, by lipofection with Cellfectin II reagent (Invitrogen). Cells were checked daily for signs of cytopathic effects (CPE), and the whole culture was collected when 70% of the cells showed granulation. This was followed by three cycles of freeze-thaws on dry ice, and 200 μl of this passage 1 (P1) stock was transferred to 25 ml of fresh Sf9 suspended cell culture in polycarbonate Erlenmeyer flasks (Corning) at a density of 2.5 × 10⁶ cells/ml, to create the P2 stock, cultured in serum-free SF900 II medium (Gibco).

The P2 stocks were incubated and monitored for CPE every third day. When at least 70% of the cells showed signs of CPE, the culture was collected and centrifuged at 3,000 × g. The pelleted cells, after resuspension in 1 ml of 1× TNTM pH 8 (50 mM Tris pH 8, 100 mM NaCl, 0.2% Triton X-100, 2 mM MgCl₂) were disrupted by three cycles of freeze-thaws on dry ice and centrifuged again. Supernatant was mixed back together with the cell culture supernatant and was subjected to treatment with 250 units of Benzonase nuclease (Sigma-Aldrich) per every 10 ml. The liquid was mixed with 1× TNET pH 8 (50 mM Tris pH 8, 100 mM NaCl, 0.2% Triton X-100, 1 mM EDTA) in a 1:1 ratio and concentrated on a cushion of 20% sucrose in TNET, using a type 60 Ti rotor for 3 h at 4°C at 45,000 rpm on a Beckman Coulter S class ultracentrifuge. The pellet was resuspended in 1 ml of 1× TNTM pH 8 and, following overnight incubation at 4°C, purified on sucrose step gradient of 5 to 40% for 3 h at 4°C at 35,000 rpm on the same instrument in an SW 41 Ti swinging bucket preparative ultracentrifuge rotor. The visible single band that formed at the 20 to 25% sucrose interface was then collected by a needle puncture and a 10-ml volume syringe.

Dialysis, galactose binding, and pH treatment of the purified AAV9 particles.

Following purification, the VLPs were dialyzed to 1× phosphate-buffered saline (PBS) at pH 7.4 in three rounds of incubation at 4°C, with the complete dialysis buffer renewed after a minimum of 5 h of incubation. Purified particles were concentrated to a density of 2 mg/ml using an Apollo 20-ml concentrator with a 150-kDa molecular cutoff. A 100 μl fraction of the concentrated AAV9 particles was subjected to another round of dialysis for 48 h in 1× universal buffer (20 mM HEPES, 20 mM MES, 20 mM sodium acetate, 0.15 M NaCl, 3.7 mM CaCl₂) with the buffer pH adjusted to either 7.4, 6.0, 5.5, or 4.0. Upon collecting each pH-modified fraction, the correct adjustment of pH was confirmed by pipetting 5 μl of each on a broad-range pH strip. Prior to pH adjustment, two additional 100-μl fractions of 2 mg/ml concentrated AAV9 particles were mixed with 4 M solution of N-acetylgalactoseamine in a 1:240 capsid to glycan ratio, which is the equivalent of four glycans for each AAV9 galactose binding site. The particle-glycan mixtures were then dialyzed to pH 7.4 and 5.5, respectively, as described above for the case of the native, non-glycan-treated particles.

Structural studies.

Three-microliter aliquots of the dialyzed AAV9 VLPs following pH adjustment and with or without terminal galactose attachment were applied to glow-discharged 2/2 C-flat holey carbon grids (Quantifoil Micro Tools GmbH) with a thin layer of carbon (Protochips, Inc.) and vitrified using a Vitrobot Mark IV instrument (FEI Co.) at 95% humidity and 4°C. The quality and suitability of the grids for cryo-data collection were determined by screening with a 16-megapixel charge-coupled device camera (Gatan, Inc.) in a Tecnai G2 F20-TWIN transmission electron microscope operated at 200 kV under low-dose exposure (∼20 e−/Ų) prior to data collection.

High-resolution data collection was carried out at two locations, namely, Florida State University (FSU), where the native data sets were collected at four different pHs, and the National Institutes of Health Intramural Research Program (NIH IRP) Cryo-EM Facility, where the two AAV9-galactose complex data sets were collected. At both locations, a Titan Krios electron microscope (FEI Co.) was used operating at 300 kV, equipped with a DE64 DED (Direct Electron) at FSU and a Gatan K2 summit camera at NIH. At NIH IRP, the scope also contained a Gatan bioquantum energy filter and a zero-loss energy slit width of 20 eV. Movie frames were recorded using the Leginon semiautomated application (83) at FSU and serialEM at the NIH (84). Images were collected at 50 frames per 10 s with a 63e−/Ų electron dosage at both locations. Movie frames were aligned using the MotionCor2 application with dose weighting (85).

To reconstruct the 3D structure of the native AAV9 VLPs from the micrographs, AAV9 capsids were extracted using EMAN2 interactive boxing (86) and the AUTO3DEM software suite (87). Individual particle image normalization and apodization were performed using the AUTOPP subroutine of AUTO3DEM, with options F and O. Estimations of the defocus values for the micrographs used the CTFFIND4 subroutine (88) in AUTO3DEM (option 3X) to enable correction of the microscope-related contrast transfer functions (CTFs). Initial models at low resolution (30 Å) were generated from the images of 100 capsids using an ab initio random-model method, applying icosahedral symmetry. Orientations and origins of each particle were determined based on the initial model, and the final map was obtained after a number of iterations (15 to 31) by AUTO3DEM, until the resolution could not be further improved. A noise suppression factor was applied in AUTO3DEM to avoid amplification of noise in the density maps.

To reconstruct the 3D structure of the two galactose-bound AAV9 data set micrographs, we used the cisTEM software for single-particle image reconstruction (89). Micrograph quality was assessed by CTF estimation using a box size of 700 pixels. Capsids included in the analysis were automatically picked by the particle selection subroutine, at a threshold value of 2.0. Boxed capsids were subjected to 2D classification, imposing icosahedral symmetry at 35 classes. Particles of classes which failed to display a clear 2D-reconstructed image of the AAV9 capsids were eliminated from the reconstruction. Startup volume generation was carried out in 40 iterations, imposing icosahedral symmetry, following an ab initio routine as well. The obtained startup volume was subjected to automatic refinement under icosahedral constraints and underwent iterations until reaching a stabile resolution. The final map was achieved by sharpening at a postcutoff B-factor of 20.

The resolution of each cryo-reconstructed map was calculated based on a Fourier shell correlation (FSC) of 0.143. The atomic model of the previously reconstructed AAV9 crystal structure (37) was fitted into the density map using the UCSF Chimera visualization system (90). Each map was resized to the voxel size determined in Chimera using the e2proc3D.py subroutine in EMAN2 and then converted to the CCP4 format using the program MAPMAN (91). The atomic model was fit to and refined into the CCP4 density map using Coot (92). The reconstructed model was refined against the map utilizing the rigid body, real space, and B-factor refinement subroutines in PHENIX (93).