The structure of adeno-associated virus serotype 3B (AAV-3B): Insights into receptor binding and immune evasion

Abstract

Adeno-associated viruses (AAVs) are leading candidate vectors for human gene therapy. AAV serotypes have broad cellular tropism and use a variety of cellular receptors. AAV serotype 3 binds to heparan sulfate proteoglycan prior to cell entry and is serologically distinct from other serotypes. The capsid features that distinguish AAV-3B from other serotypes are poorly understood. The structure of AAV-3B has been determined to 2.6Å resolution from twinned crystals of an infectious virus. The most distinctive structural features are located in regions implicated in receptor and antibody binding, providing insights into the cell entry mechanisms and antigenic nature of AAVs. We show that AAV-3B has a lower affinity for heparin than AAV-2, which can be rationalized by the distinct features of the AAV-3B capsid. The structure of AAV-3B provides an additional foundation for the future engineering of improved gene therapy vectors with modified receptor binding or antigenic characteristics.

Adeno-associated viruses (AAV) have emerged as leading gene therapy candidate vectors. Cited advantages include lack of disease associated with the virus, the ability to transduce non-dividing cells, absence of an adverse immunological response, and long-term expression of therapeutic payload (Buning, Braun-Falco, and Hallek, 2004; Raake et al., 2008). Despite these attributes, AAVs have broad cell and tissue tropism, low transduction efficiency, and are neutralized by a pre-sensitized host immune system (Halbert et al., 1997), limiting the therapeutic potential of vectors based on natural serotypes. Improved AAV-based vectors can be designed more efficiently with an understanding of the structure and functions of natural virus capsids.

Adeno-associated viruses are small (~250Å) non-enveloped parvoviruses with a 4.7kb single-stranded DNA genome packaged inside a protein capsid of T=1 icosahedral (60-fold) symmetry (Chapman and Agbandje-McKenna, 2006). Parvoviruses of the genus Dependovirus, such as AAV, depend in replication upon co-infection with a helper virus, such as adenovirus (Bowles, Rabinowitz, and Samulski, 2006). The AAV genome contains two genes, encoding capsid and replication proteins. Multiple start codons and alternate mRNA splicing give rise to 3 overlapping capsid proteins (VP1, VP2, and VP3). Capsid proteins differ only at their N-termini and are present in a 1:1:8 ratio. VP1 contains a unique N-terminal region of ~137 amino acids, followed by a 65 amino acid region, common to VP1 and VP2. The VP3 protein contains ~534 amino acids (depending on the serotype), common to all 3 capsid proteins.

12 AAV serotypes (and hundreds of natural variants thereof) have been identified (Gao et al., 2004; Schmidt et al., 2008). In spite of the high sequence identity of their genomes (53–93%), AAV serotypes are antigenically distinct from one another and use a range of different receptors for cell binding and entry (Akache et al., 2006; Asokan et al., 2006; Qing et al., 1999; Rutledge, Halbert, and Russell, 1998; Seiler et al., 2006; Summerford, Bartlett, and Samulski, 1999; Summerford and Samulski, 1998), which contribute to widely different cell specificities among serotypes. AAV serotype 3 was isolated from a human clinical specimen (Muramatsu et al., 1996) and has somewhat different cell tropism compared to other serotypes (Handa et al., 2000; Liu et al., 2005; Rabinowitz et al., 2002). AAV-3B is a minor variant of AAV-3 (also known as AAV-3A), with 6 amino acid differences in the VP1 capsid protein (Rutledge, Halbert, and Russell, 1998), 4 of which are in VP3. Cell entry by AAV-3B is not inhibited by neutralizing rabbit anti-AAV-2 serum, demonstrating that these 2 serotypes are antigenically distinct (Rutledge, Halbert, and Russell, 1998). Although anti-AAV-2 serum does not efficiently neutralize AAV-3 cell entry, some conformation-specific monoclonal antibodies raised against the AAV-2 capsid are cross-reactive with AAV-3 (Wobus et al., 2000). This suggests structural conservation of some epitopes on the two capsids. Other monoclonal antibodies (such as C37-B) specifically inhibit receptor binding by AAV-2, and are not cross-reactive with other serotypes (Wobus et al., 2000). The effect of structural differences on antibody cross-reactivity among AAV serotypes is not fully understood.

In spite of high sequence identity between some of the AAV serotypes, there are important functional differences beyond their antigenic properties. Comparative studies therefore have the potential to shed important light on, for example, the determinants of cell tropism / specificity. AAV-3B is most closely related in sequence to AAV-2, sharing 87% capsid identity to the type species. AAV-2 is the most widely characterized serotype, and its crystal structure was the first to be determined among AAVs (Xie et al., 2002). Subsequently, crystal structures of AAV-4 (Govindasamy et al., 2006) and -8 (Nam et al., 2007) and a cryo-EM structure of AAV-5 (Walters et al., 2004) have been determined. Crystal structures of serotypes 1, 5–7, and 9 are also in progress (DiMattia et al., 2005; Miller et al., 2006; Mitchell et al., 2009; Quesada et al., 2007; Xie et al., 2008). All serotypes have a core β-barrel subunit fold, which is highly conserved among all parvoviruses of known structure (Chapman and Agbandje-McKenna, 2006). The AAV capsid surface topology arises from loops that connect the core beta strands of each subunit, generating distinctive features including protruding spikes surrounding the 3-fold symmetry axis; a cylindrical protrusion surrounding the pore near the 5-fold symmetry axis; and a dimple near the 2-fold symmetry axis. Differences in the intervening loops of AAV-2, -4, and -8 lead to unique capsid structural features, which presumably contribute to serotype-specific receptor interactions, immune recognition, cell tropism and transduction efficiency.

AAV-2 and -3 require heparan sulfate proteoglycan (HSPG) for cell attachment (Handa et al., 2000; Rabinowitz et al., 2002; Summerford and Samulski, 1998), while serotypes 1, 4, 5, 6 bind to specific sialic acid derivatives prior to cell entry (Kaludov et al., 2001; Wu et al., 2006). AAV-6 can also bind heparan sulfate, but it is sialic acid binding that facilitates cell entry. AAV-3 has a weaker affinity for heparin than AAV-2, as higher concentrations of heparin are required to inhibit cell entry (Handa et al., 2000). This is due, presumably, to incomplete conservation of the heparin binding site identified on AAV-2 (Kern et al., 2003; Levy et al., 2009; Lochrie et al., 2006; O'Donnell, Taylor, and Chapman, 2009; Opie et al., 2003). Following attachment to HSPG, AAV-3 then binds to a co-receptor before entering the cell through an endosomal pathway. FGFR1 and the 37/67kDa laminin receptor (LamR) have been identified as co-receptors for both AAV-2 and AAV-3 (Akache et al., 2006; Belur et al., 2008; Blackburn, Steadman, and Johnson, 2006; Qing et al., 1999), and the proposed binding motif on AAV-2 for another co-receptor, α₅β₁ integrin, is conserved in serotypes 1–3 and 6–10 (Asokan et al., 2006). In spite of this apparent overlap in co-receptor usage, AAV-3 did not compete with AAV-2 in cell entry assays, indicating that there are distinct preferences in the receptors for cell entry (Mizukami, Young, and Brown, 1996). In fact, a 42kDa receptor that binds AAV-3 has been identified that does not bind AAV-2 (Handa et al., 2000). Little is known about how structural differences between serotypes contribute to heparin binding affinity or co-receptor selectivity among serotypes.

The functions of natural AAV capsids can be better understood through their structures, which can also guide the future design of novel recombinant gene therapy vectors. In an effort to more fully understand receptor binding and antigenic determinants of AAVs, the X-ray crystal structure of AAV-3B was undertaken. Although there is high conservation of the protein fold etc., at a detailed level, critical regions of the capsid surface have distinctive physical-chemical properties. These differences suggest that the viruses have faced selective pressures to find a variety of means of binding to cellular receptors and evading immune neutralization.

Materials and Methods

AAV-3B was produced and purified as described (Lerch et al., 2009), following procedures optimized for AAV-2 (Xie et al., 2004). A detailed description of the crystallization, data collection, processing and preliminary analysis of AAV-3B crystals has been reported (Lerch et al., 2009). Briefly, two partial diffraction data sets were collected from AAV-3B crystals at the Cornell High Energy Synchrotron Source (CHESS) F1 beamline. The crystals belong to space group R3, diffracted X-rays to 3.0Å and 2.6Å resolution, but cannot be merged well due to different levels of merohedral twinning. Crystal twinning was characterized for each data set using the program phenix.xtriage (Adams et al., 2002). The crystal used to collect a 12% complete 3.0Å data set was partially twinned (twin fraction of 0.15), while a 32% complete 2.6Å data set came from a crystal that was near perfectly twinned (twin fraction of ~0.48), both by the twin law (h, -h-k, -l). With long exposures needed to record intensities that are weak (in proportion to the volume of the unit cell) and progressive radiation damage it is common that virus crystals provide only a fraction a complete data set. Given that a number of virus structures have been determined with 10–30% complete data sets (Badger et al., 1988), the varying level of twinning and therefore the intractability of merging different data sets, it was decided to determine the structure with our most complete (32%) data set, although the lesser twinned data set would be helpful in early steps. Data collection and refinement statistics are shown in Table 1.

Table 1.

Summary of diffraction data and refinement statistics

Parameter	Crystal 1	Crystal 2
Space group	R3	R3
Unit cell dimensions (Å; hexagonal setting)	a = b = 257.8, c = 607.0	a = b = 257.7, c = 603.8
Resolution range (Å)a	14–3.0 (3.07–3.00)	15–2.6 (2.66–2.60)
Twin fraction	0.15	0.48
Rmerge (%)b	6.7 (9.4)	6.6 (42.2)
Completeness (%)	10.9 (10)	28.1 (10)
cRcryst	0.23 {0.24} (0.42)	0.19 (0.38)
cRfree	0.26 {0.27} (0.46)	0.22 (0.36)
RMSD bond lengths (Å) / RMSD bond angles (°)	0.003/0.694	0.003/0.746
No. protein/nucleotide atoms	4154/22 (× 20)	4154/22 (× 20)
Avg. B-factor (Å2)	105.9	90.6
Estimated error (Luzzati, Å)	0.95	0.57

^aNumbers in parentheses are values in highest resolution shell

^bR_merge = ∑_hkl ∑_i |I_i(hkl) - <I(hkl)>|/∑_hkl ∑ _i I_i(hkl), where I_i(hkl) is the ith observation of a symmetry equivalent of reflection hkl. The highest resolution shell has no symmetry equivalent reflections, and therefore R_merge was calculated in medium resolution shells (4.7–4.3Å for crystal 1 and 2.9–2.8Å for crystal 2).

^cR^cryst/R^free in brackets were determined without the twin law (h, -h-k, -l). R^free was calculated with 2822 (4.4% of data from crystal 1) and 5095 (4% of data from crystal 2) reflections, selected from identical thin resolutions shells from each data set. Reflections neighboring test set resolution shells were omitted to reduce bias in R^free. As a result, the highest resolution shell for which R^free was calculated was 3.23–3.17Å for crystal 1 and 2.8–2.75Å for crystal 2.

Test set reflections were selected from thin resolution shells, to avoid bias due to high (20-fold) non-crystallographic symmetry (NCS) (Fabiola, Korostelev, and Chapman, 2006) and twinning. The width of the shells varied from 0.005Å (at higher resolution) to 0.042Å (at lower resolution), to provide comparable numbers of test reflections in each shell. For the 3.0Å data set a total of 2822 reflections (4.4%) were selected from 6 different resolution shells. To avoid bias when alternating between data sets, the same 6 shells were used for the higher resolution data set, plus an additional 2 shells between 3.0 and 2.6 Å. The test set for the higher resolution data contained the 5095 reflections (4%). In the presence of non-crystallographic symmetry, significant bias in R^free can result from correlations between test reflections and their neighbors in reciprocal space unless the latter are also excluded from refinement (Fabiola, Korostelev, and Chapman, 2006). Thus, reflections were excluded from wider shells enclosing each test shell – a total of 10,395 for the 3.0Å data and 18,711 for the 2.6Å data.

As described previously (Lerch et al., 2009), the orientation of AAV-3B in the unit cell was determined from the rotation function, using the less twinned 3.0Å data set and the program GLRF (Tong and Rossmann, 1997). One-third of the capsid (20 subunits) is contained in the asymmetric unit, and the starting model consisted of AAV-2, oriented in the AAV-3B unit cell. The position of the particle was constrained, with the icosahedral and crystallographic 3-fold symmetry axes co-incident, and the subunit orientation / position was optimized by rigid body refinement using the program CNS (Brünger et al., 1998).

Starting phases for icosahedrally averaged maps were calculated at 3.75Å resolution from the starting AAV-2 model and extended to 3.0Å resolution with the Rave/CCP4 program suite (Collaborative Computational Project Number 4, 1994) and the 3.0Å data set, disregarding the twinning. Initial electron density maps from the 3.0Å data exhibited side chain and other structural features unique to AAV-3B, and omit maps for regions omitted from the initial phasing model were clearly interpretable, together verifying that the NCS averaging was of sufficient power to obtain unbiased density from the partial data sets (Lerch et al., 2009). The AAV-3B sequence was manually inserted and an initial model built using the program O (Jones et al., 1991). Coordinate, B-factor, and nucleotide occupancy refinement against both 3.0Å and 2.6Å resolution data was performed using the program phenix.refine (Adams et al., 2002), taking into account twinning of the 2.6Å data. Subsequent refinement of the 3.0Å model also took twinning into account, although the twin fraction was likely negligible. The full content of the asymmetric unit (20 subunits) was refined with non-crystallographic symmetry (NCS) tightly restrained. Further model building was performed using icosahedrally averaged detwinned electron density maps (2.6Å resolution) with sections of the model omitted to remove bias. Final modeling (searches for water molecules, addition of nucleotide, geometry optimization) was performed using the program Coot (Emsley and Cowtan, 2004), and the model quality was assessed with Procheck (Laskowski et al., 1993).

Comparison of the AAV-3B structure to AAV-2 (Xie et al., 2002; PDB id 1LP3), -4 (Govindasamy et al., 2006; PDB id 2G8G), and -8 (Nam et al., 2007; PDB id 2QA0) structures was performed by structural alignment using the secondary structure matching (SSM) algorithm (Krissinel and Henrick, 2004) in Coot. Regions that could not be aligned structurally (C_α atomic distance > 5Å) are defined as variable.

Electrostatic potentials of AAV-2 and -3B were calculated using the programs PDB2PQR (Dolinsky et al., 2004) and APBS (Baker et al., 2001). A course grid covered the entire capsid, while a fine grid (80Å × 80Å × 50Å) was centered near a 3-fold symmetry axis, covering one set of 3-fold symmetric spikes. The respective potentials were mapped onto the molecular surfaces of AAV-2 and AAV-3B, which were rendered in PyMOL (DeLano, 2002) with color ramps between −5 and +5 k_bT/e_c. Images of individual AAV subunits and full capsids were made using PyMOL (DeLano, 2002) and Chimera (Pettersen et al., 2004) software, respectively.

The AAV-3B structures at 2.6 and 3.0 Å have been deposited in the Protein Data Bank (PDB) with accession codes 3KIC and 3KIE, respectively.

Results and Discussion

The structure of AAV-3B

The structure of AAV-3B was determined to 2.6Å resolution, in an effort to elucidate features that are distinctive to this serotype and how these features contribute to cell entry and host immune response. The structure was determined from two crystals, which differed in levels of merohedral twinning, preventing merging of the data sets. Merohedral twinning is a condition in which a crystal contains two domains that differ in molecular packing orientation. Twinning can be difficult to detect, as the diffraction patterns of the two domains overlap exactly. A hybrid approach used each data set at different points during the structure determination process (Lerch et al., 2009). A preliminary structure was determined using the data set that was less complete (~11%) and of lower resolution (3.0Å), but that came from a crystal with a lower twin fraction, making structure determination tractable. This structure was then iteratively refined and modeled using the 28% complete 2.6Å resolution data set, which was nearly perfectly twinned (twin fraction of 0.48). Electron density maps calculated using the 3.0Å data, which were processed as if untwinned, were sufficient to build an initial model, even though partially twinned (twin fraction of ~0.15). Subsequent refinement of the model against this data set did take twinning into account. This is reflected in the refinement statistics shown in Table 1. In spite of incomplete data and twinning, the 2.6Å electron density map used to build the final AAV-3B model was of high quality (Figure 1A) as a result of 20-fold non-crystallographic symmetry (NCS) averaging. The AAV-3B model contains amino acids 217 to 734 (VP1 numbering), representing all but the first 14 residues of the VP3 protein, and one nucleotide molecule. The N-terminal residues are likely disordered or distinct between subunits, as noted in prior parvovirus structures (Govindasamy et al., 2006; Nam et al., 2007). Similarly, regions unique to the N-termini of VP1 or VP2 proteins were not observed. These N-terminal extensions are absent in ~83% of the subunits, thus structural information is lost during NCS averaging. (Furthermore, there is no evidence that the outer surface of the capsid reflects the locations of VP1-unique extensions inside the virus. Therefore, the crystal lattice contains many orientations of the VP1-unique region that are consistent with 60-fold symmetric packing interactions. The diffraction data inherently reflect the average of the many VP1 orientations.) The final R^cryst/R^free values are 0.24/0.27 for the 3.0Å data (R^cryst/R^free = 0.23/0.26 when refined as twinned). For the 2.6Å (twinned) data the final R^cryst/R^free values are 0.19/0.22 after restrained refinement of coordinates, individual restrained B-factors for protein atoms and an occupancy for the nucleotide (Table 1). Test reflections for the calculation of R^free were selected from thin resolution shells (identical in both data sets) to reduce bias that occurs in the presence of twinning and high non-crystallographic symmetry (NCS) (Fabiola, Korostelev, and Chapman, 2006). Reflections that neighbor the test reflections were also removed prior to refinement, as these are most strongly correlated with the test set and bias R^free in the presence of NCS. The structures determined from two different crystals are very similar. The root mean squared deviation (RMSD) between the structures is 0.5Å, which is similar to the estimated coordinate error (0.6Å) of the 2.6Å resolution structure. Thus, the structures can be considered the same, and, forthwith, the 2.6Å resolution structure of AAV-3B is described, except as noted.

Figure 1.

The structure of AAV-3B. (A) A portion of the 2mF_o-DF_c electron density surrounding part of the AAV-3B model shows the quality of the map calculated at 2.6Å resolution. (B) The jellyroll barrel subunit architecture of AAV-3B is conserved throughout the Parvovirus family. A cartoon representation of a single subunit is shown, surrounded by the transparent surfaces of neighboring subunits of the assembled capsid. Arrows mark the locations of the 5-, 3-, and 2-fold symmetry axes and are labeled. Loops are named after the core β-strands that they connect. The DE loop lines the interior of the 5-fold pore and forms the cylindrical protrusion surrounding the pore of the intact capsid (also shown in C). The HI loop covers part of the surface on the adjacent subunit (Chain E), lining the floor of a depression surrounding the cylinder. Two sections of each GH loop and two loops from neighboring subunit C comprise the prominent spikes that surround each 3-fold symmetry axis of the capsid. (C) Sixty AAV-3B subunits assemble to form the icosahedrally symmetric virus capsid. The viral asymmetric unit is bounded by 5-fold (pentagon), 3-fold (triangle), and 2-fold (oval) axes. The features on the capsid surface are formed by the subunit loops, shown in panel B.

The AAV-3B subunit fold (Figure 1B) is a jellyroll barrel conserved in all parvoviruses. The eight core β-strands (βB through βI) are bridged by loops, which are variable among parvoviruses and AAV serotypes. These loops are named after the β-strands that they connect and some contain additional secondary structural elements (Chapman and Agbandje-McKenna, 2006). 60 identical copies of the AAV-3B subunit form the icosahedrally symmetric capsid (Figure 1C), where loops from different subunits come together to form unique capsid surface features. Like other AAV serotypes (Govindasamy et al., 2006; Nam et al., 2007; Xie et al., 2002), the AAV-3B capsid surface has several notable features: three prominent spikes surrounding each icosahedral three-fold symmetry axis; a pore surrounded by a more subtle cylindrical protrusion at the 5-fold symmetry axis; and a depression (or dimple) at the 2-fold symmetry axis. The 3-fold symmetric spikes are formed by two GH sub-loops from one subunit (AAV-3B VP1 residues 447–462 and 581–593) intertwined with two sub-loops (residues 490–508 and 544–558) from neighboring subunit C (green surface in Figure 1B). Surrounding the 5-fold symmetry axis, the DE loop (between strands βD and βE; residues 318–334) lines the pore and forms the cylindrical protrusion. The floor of the depression surrounding the 5-fold cylinder is formed by the HI loop (residues 652–673) from neighboring subunit H (yellow surface in Figure 1B). The bottom of the depression (“dimple”) at the 2-fold axis is the region on the surface nearest to the viral center and consists of two α-helical regions from adjacent subunits (blue and pink; subunit B).

On the interior of the AAV-3B capsid, clear electron density was observed for one nucleotide, modeled as dAMP (Figure 2). Some smearing or disorder in the density is expected, because there is no constraint that the base-type be adenine in all of the sites averaged according to the icosahedral symmetry. The nucleotide is positioned within a pocket at the interior surface of the capsid near the 3-fold symmetry axis, beneath the exterior spikes. The pocket is lined by residues 628–638 and 418–419 (VP1 numbering, Figure 2). Nucleotide coordinates were refined to a fractional occupancy of ~0.7 against the higher resolution twinned data when fixing the nucleotide B-factors to the average for the capsid protein.

Figure 2.

Nucleotide binding inside the AAV-3B capsid, near the 3-fold symmetry axis. A single dAMP nucleotide was modeled in a pocket at the interior of the AAV-3B capsid The nucleotide is a fragment of the genomic single-stranded DNA that, through interactions with the capsid protein, adopts the capsid’s icosahedral symmetry. Thus, unlike most of the genomic DNA, its density remains strong through the symmetry averaging applied during the structure determination. The interior surface of one capsid subunit is shown in light pink, and the surfaces of 3-fold symmetric subunits are shown in light blue and light green. The dark grey mesh shows the 2.6Å mF_o-DF_c difference map (within 1.5Å of dAMP atoms), calculated with nucleotide coordinates omitted from the phasing model and contoured at 3.3 σ. The light grey mesh shows a 2mF_o-DF_c electron density map calculated using the 2.6Å data contoured at 1σ. Residues near the dAMP binding site are shown as sticks and are labeled. Density was also observed in an averaged map calculated using the untwinned 3.0Å data (not shown).

In AAV-3B there appears to be a preference for adenosine nucleotides, the size of the density indicating a predominance of purines over pyrimidines, and the shape consistent with a C6 amino group, but not C2. However, it is possible that the density represents a symmetry averaged mixture with a small proportion of other base-types. The viral genome is not icosahedrally symmetric. Thus, resolution of the nucleotide density, following symmetry-averaging indicates that interactions with the capsid force a small piece of the nucleic acid to conform to the capsid’s icosahedral symmetry. A nucleotide in the same position and orientation has been observed previously in the structure of AAV-4 (Govindasamy et al., 2006), but the nucleotide in AAV-8 is oriented differently (Nam et al., 2007). Residues with side-chains directed toward the nucleotide in each structure are fully conserved among AAV serotypes (compared to <65% sequence identity for the interior surface of all AAV capsids) so the site might have functional importance. It is noted that this nucleotide (and its 60 symmetry-equivalents) is a smaller fraction of the genome, and located in a different position from the oligo-nucleotides observed in the autonomous parvoviruses, CPV and MVM (Agbandje-McKenna et al., 1998; Chapman and Rossmann, 1995).

Comparing AAV-3B to other AAV serotypes of known structure

When the AAV-3B structure is compared to the structures of AAV-2, -4, and -8 (Govindasamy et al., 2006; Nam et al., 2007; Xie et al., 2002) (Figure 3), one is first struck by similarity in the backbone path that is perhaps higher than their sequence identities might suggest. Nevertheless, the structural differences between serotypes are in proportion to their sequence dissimilarity. The capsid contains three proteins (VP1, VP2, and VP3) of 736, 599 and 534 residues, in 1:1:8 ratio, differing only in their N-terminal extensions. As in the prior AAV structures, the unique regions of VP1/2 are not resolved in the icosahedrally averaged density. Interpretable density starts at VP3 residue 15 (residue 217 in the conventional VP1 numbering). Thus, structural differences are quantified as the root mean square deviation (RMSD) between paired C_α atoms for the 520 resolvable residues of VP3 (AAV-3B VP1 217–736). Each of the other serotypes was superimposed on AAV-3B using the iterative SSM algorithm. This starts with secondary structure matching, then extends the alignment using both structure and sequence (Krissinel and Henrick, 2004).

Figure 3.

Comparing AAV serotype structures. (A) Molecular surfaces calculated from the atomic structures of AAV-3B, -2, -4, and -8. The triangular icosahedral asymmetric unit is outlined and is bounded by a 5-fold symmetry axis (pentagon), two 3-fold axes (triangles), and a 2-fold axis (oval). The most distinctive surface features correspond to the greatest differences in the subunit loop structures at “VR-I” (dashed circle) and “VR-IV” (solid circle) as illustrated in panel B. (B) Capsid subunits are overlaid with the secondary structure matching (SSM) algorithm (Krissinel and Henrick, 2004): AAV-3B (blue), -2 (Xie et al., 2002; red), -4 (Govindasamy et al., 2006; purple), and -8 (Nam et al., 2007; green). The greatest differences are in two of the variable regions (VR), I and IV.

Structurally AAV-3B is most similar to AAV-2 (88% capsid sequence identity; RMS C_α difference of 0.57Å for 500 of 520 AAV-3B alignable residues). Estimated coordinate errors of ~0.6Å (Table 1) would be expected to give an all-atom RMSD of 0.8Å. As the C_α positions, used in comparing structures of different sequence, are determined more precisely, the consistency of error estimates and RMSDs indicates that much of the modest difference between coordinate sets is due to experimental error. AAV-8 (86% sequence identity) is only slightly less similar structurally, where 508 of the AAV-3B/8 residues can be aligned with an RMSD of 0.72Å. AAV-3B is the most different structurally from AAV-4 (63% sequence identity), with 472 of 520 residues aligning with an RMSD 1.0Å.

In all of the pair-wise structural alignments, the vast majority of VP3 can be superimposed. Residues that could not be aligned are at the N- and C- termini and also within a subset of nine structurally variable regions (VR I-IX), described previously as deviating by at least 1Å between the AAV-2 and AAV-4 structures (Govindasamy et al., 2006). Between the AAV-2 and AAV-3B structures, these variable regions differ significantly (RMSD > 1Å) in only two regions, VR-I and VR-IV (Table 2). Thus, for the most part, AAV-3B differs from AAV-4 in the same places as does AAV-2, with similar magnitudes of deviation. When AAV-3B is compared to AAV-8, a very similar picture emerges, that the VR loops are similar except in VR-I and VR-IV, where the differences are slightly greater than between AAV-3B and AAV-2. Within the core viral jelly-roll β-barrel, the structures of AAV serotypes are exceptionally well conserved, even between AAV-3B and AAV-4, and the sites of divergence are confined to the surface-exposed loops. In the following paragraphs, variable regions (VR) I and IV, which are the most distinct (RMSD > 1Å) between serotypes 2, 3B and 8 are analyzed in greater detail.

Table 2.

RMSDs of residues in variable regions (Å)

Variable region	AAV-3B vs. AAV-2	AAV-3B vs. AAV-4	AAV-3B vs. AAV-8	AAV-2 vs. AAV-4
I	1.3	3.0	3.2	2.6
II	0.9	1.6	0.9	1.6
III	0.8	1.8	0.7	2.1
IV	2.8	6.9	2.6	6.0
V	0.9	2.4	0.8	2.8
VI	0.8	4.3	0.8	3.7
VII	0.8	2.4	0.7	2.3
VIII	0.7	1.5	0.7	1.7
IX	0.9	2.9	1.0	2.6

Structures were initially aligned using SSM (Krissinel and Henrick, 2004). Residues that remained unpaired by SSM were aligned manually to give the lowest RMSD for the variable region of the loop.

VR-I contains residues 262–268 on the surface of AAV-3B, which are identical in the sequence of AAV-2. Part of this region in AAV-2 (corresponding to AAV-3B residues 265–271) has been implicated genetically in transduction efficiency in a step that is independent of attachment to heparan sulfate (Lochrie et al., 2006). The structural differences between serotypes 2 and 3B in VR-I appear to result from different interactions that anchor one end of VR-I to its neighbors. In AAV-3B, Asn₂₇₀ hydrogen bonds with Ser₄₇₂. By contrast, the corresponding residues in AAV-2, Asp₂₆₉ and Arg₄₇₁ form a salt bridge. The different sizes of Ser (AAV-3B) and Arg (AAV-2) affect the 269-to-471 backbone distance, resulting in interactions that are unique to AAV-2: between Asn₂₇₀ N_δ2 and Ser₄₆₈ O_backbone, and between Asn₂₆₈ N_δ2 and Gly₅₁₂ O_backbone. Within the VR-I residues, the interactions of AAV-3B are analogous to AAV-2 with Ser₂₆₂ and Ser₂₆₇ hydrogen-bonding with His₂₇₁. This is in contrast to AAV-8, the outlier of the three serotypes, in which a threonine replaces His₂₇₁. The more exposed AAV-8 loop is 2 residues longer with B-factors that are ~2-fold higher than the average (Nam et al., 2007). Thus, AAV-8 differs from AAV-3B/2 both in structure and flexibility of VR-I. In summary, the local VR-I primary structure is not distinctive between AAV-3B and AAV-2, but tertiary interactions with regions distant in primary structure differentiate the serotypes in a manner that is apparent only with 3-D structure.

Variable region IV on AAV-3B (residues 449–469) is the region that differs most from AAV-2, -4 and -8, and is located at the tip of the spike (Figure 3B). In all serotypes, the base of the loop is formed by a β-ribbon. AAV-3B has the short β-ribbon of AAV-4 and AAV-8 rather than the extended ribbon of its closest relative, AAV-2. The loop is therefore less constrained by secondary structure than in AAV-2. The loop extends outwards to form the sharp spike of AAV-8 rather than folded down in the blunter version of AAV-4 (Figure 3A). Relative to AAV-2, the spikes are slightly splayed out with C_α-C_α distances between 3-fold equivalents of 68Å at Thr₄₅₅, compared to 61Å for AAV-2. VR-IV is distinct in all AAV structures to date. Perhaps this exposed and antibody-accessible region has been under particularly stringent selective pressure to change to evade immune detection, and that its exposed location has provided more freedom to do so without collateral impact on neighboring structural elements.

To this point, the analysis has been framed mostly in terms of backbone structure. Differences in sequence add to this and modulate the surface characteristics, even where the backbone is conserved. This is important as we analyze the structural determinants of serotype-specific functional attributes.

The receptor binding site

The mechanism by which AAV-3B attaches to heparan sulfate proteoglycan (HSPG) on the cell surface is not fully understood. For AAV-2, there is now a wealth of structural and genetic evidence about these interactions at the amino acid level. Inferences into the mechanism of heparan binding by AAV-3B can be made by integrating our more comprehensive understanding of AAV-2 receptor interactions with the structure of AAV-3B. Implicit in such an approach is the assumption that binding of the primary receptor shares commonalities in the two serotypes. With 87% sequence identity, prima facie, this is not an unreasonable assumption. With this important caveat, the heparin density can be superimposed with the precision with which the structures can be aligned (~0.6 Å), but interpretation is limited by the 8 Å resolution of the AAV-2 cryo-EM, corresponding to the dimensions of 1 or 2 amino acids. In functional measures of receptor binding, serotypes 2 & 3 appear similar, but differ in detail. They both enter the cell after initial attachment to heparan sulfate proteoglycan (Rabinowitz et al., 2002; Summerford and Samulski, 1998), but AAV-3 appears to have a lower affinity for heparin than AAV-2 (Handa et al., 2000). Similar to AAV-3A, characterized in previous studies, AAV-3B also has a lower heparin binding affinity than AAV-2, based on weaker binding to a heparin affinity column (Supplementary Figure S1).

Heparin and heparan sulfate are highly sulfated, and thus negatively charged molecules. Heparin binding sites on proteins are typically positively charged, due to the presence of basic (arginine and lysine) residues (Conrad, 1998). Mutagenesis of AAV-2 has identified 5 basic residues involved in heparin binding (Kern et al., 2003; Opie et al., 2003), including arginines 585 and 588 in AAV-2 whose mutation to alanine eliminated binding to heparin altogether (Opie et al., 2003). These residues are located within VR-VIII on the side of each 3-fold proximal spike on the surface of AAV-2 (Figure 4A). The recent ~8Å cryo-electron microscopy (EM) structure of AAV-2 in complex with heparin revealed two important features of the AAV-2 heparin binding site (O'Donnell, Taylor, and Chapman, 2009). Firstly, the complex provided independent confirmation of the importance of Arg₅₈₅ and Arg₅₈₈ in contacting heparin. The strongest heparin electron density was located in contact with these residues. Secondly, the structure revealed that the overall heparin-binding footprint on AAV-2 is much broader than previously appreciated, spanning 36 total amino acids (including the 5 previously identified). This includes residues that contact heparin directly or have unobstructed (through solvent) access to heparin (Figure 4A). The surface topology of the heparin-binding region on AAV-2 is very similar to the analogous region on AAV-3B (Figure 4B). Analysis of the electrostatic surface potential in this region, however, reveals striking differences between these serotypes.

Figure 4.

Structural insights into receptor binding. Three of the sixty spikes on AAV-2 (A and C) and AAV-3B (B and D) are rendered as molecular surfaces. (A) Residues that have been shown previously to contact heparin directly (blue) or indirectly (teal) on AAV-2 are colored (O'Donnell, Taylor, and Chapman, 2009). Residues within the heparin binding footprint that are not conserved in AAV-3B are shaded lighter than conserved residues. (B) Residues from AAV-3B that align with the heparin contact residues in AAV-2 are similarly colored and shaded. The loss of the positive charge of R585 & R588 of AAV-2 might be partly compensated by the presence of R594 and/or R447 of AAV-3B. The electrostatic potential of AAV-2 (C) and AAV-3B (D) mapped onto the respective capsid surfaces is shown. Difference electron density for heparin bound to AAV-2, contoured at 7.5 σ (~5 estimated error units), is superimposed (O'Donnell, Taylor, and Chapman, 2009). The strongest density for the sulfonated heparin (negatively charged) was observed immediately adjacent to R585 and R588 of AAV-2 (C). (D) The corresponding region on the AAV-3B spike is less electro-positive. Two distinct basic patches on AAV-3B, one near R594 and the other near residue R447, are farther from the strongest heparin density seen in AAV-2, but are contained within (R594) or adjacent to (R447) the heparin binding footprint. The positive region that contains R447 is located on the opposite side of the spike relative to R585/R588 on AAV-2, and is most clearly seen here on the side of the 3-fold symmetric spike 3.

We were encouraged to analyze the surface electrostatics of AAV-3B, given the success of the electrostatics-based prediction of the AAV-2 heparan-binding site (Xie et al., 2002), which has now been confirmed experimentally (O'Donnell, Taylor, and Chapman, 2009). Potentials were calculated and mapped onto the surfaces of AAV-2 (Figure 4C) and AAV-3B (Figure 4D), highlighting the contributions of sequence differences to surface charge in these two serotypes. In the AAV-2:heparin complex (O'Donnell, Taylor, and Chapman, 2009), the strongest heparin density was centered over the most positively charged patch, anchored by Arg₅₈₅ and Arg₅₈₈. The corresponding region in AAV-3B is more neutral, as Arg₅₈₅ and Arg₅₈₈ are not conserved. It is remarkable that Arg₅₈₅ and Arg₅₈₈, which form the most intimate heparin contacts in AAV-2, are not conserved in other heparin-binding serotypes, although this could account for AAV-2’s stronger binding. Based on differences in the charge distribution in this region, it is likely that heparin binds AAV-3B somewhat differently than it does AAV-2.

Although the main determinant of heparin attachment is likely distinct on AAV-3B, we cannot exclude the possibility that, like AAV-2, heparin binds to AAV-3B in the general vicinity of the 3-fold spikes. The substitutions for Arg₅₈₅/Arg₅₈₈ are polar in AAV-3B (Ser₅₈₆ and Thr₅₈₉), residues that are also common in heparin-binding sites (Conrad, 1998). Two distinct positive patches, created in part by residues Arg₅₉₄ and Arg₄₄₇, are located near the 3-fold axis on AAV-3B and might compensate, in part, for the loss of Arg₅₈₅/Arg₅₈₈ (Figure 4D). The first site is located at the center of the 3-fold axis, where symmetry equivalents of Arg₅₉₄ from three different subunits come together (Figure 4B and D). Arg₅₉₄ is unique to AAV-3B and extends the positively charged zone into the region between the 3-fold proximal peaks. Furthermore, its bulk, relative to the alanine of AAV-2, raises the surface towards where heparin is seen to bind in AAV-2 (O'Donnell, Taylor, and Chapman, 2009). Arg₅₉₄ would not directly contact the heparin as seen in AAV-2, but it is adjacent to Arg₄₈₅ which would have solvent-mediated access. The presence of Arg₅₉₄ extends positive surface charge from just below the locations of Arg₅₈₅ & Arg₅₈₈ in AAV-2 to the 3-fold axis. Curiously, amino acid 594 is one of six that are different between AAV-3B (Arg) and AAV-3A (Gly). As AAV-3A also binds to heparin (Handa et al., 2000; Rabinowitz et al., 2002), Arg₅₉₄ cannot be the sole determinant for heparin binding.

A second possibility is that Arg₄₄₇ could make up for the absence of AAV-2’s Arg₅₈₅ / Arg₅₈₈ in AAV-3B. Arg₄₄₇ is conserved in AAV-2, but its charge is neutralized in AAV-2 by Glu₄₉₉. This is not the case in AAV-3B (and AAV-3A), where an uncharged asparagine (Asn₅₀₀) occupies the same position as Glu₄₉₉ in AAV-2. The positively charged surface at Arg₄₄₇ in AAV-3B is on the side of the spike, opposite the heparin binding site in AAV-2, and could well help compensate for the absence of Arg₅₈₅/Arg₅₈₈.

In summary, the surface topologies of AAV-3B and AAV-2 are similar, and there is more than enough positively charged surface in the general area for favorable interactions with a negatively charged primary receptor. We might expect that heparin binding be similar, but not exactly the same, as the charge distribution differs in detail. In particular, it is noted that the greatest positive charge is no longer midway up the side of the 3-fold proximal peak, but closer to the valley floor (when comparing AAV-3B to AAV-2). It is plausible that heparin might sit lower in the valley in AAV-3B to take advantage of this. With differences emerging in the details of the heparin binding sites as homologous structures are determined, it appears that evolution has found several variations on a theme that are all acceptable solutions to receptor binding. Apparently the exact details do not matter critically for the virus to enter the cell.

Antigenic regions

Insights into AAV evolution and immune evasion can be gained by comparing the structures of AAV-3B and AAV-2. With their high homology, similar receptor-binding properties, and likelihood that immune selective pressure has led to faster divergence at neutralizing immunogenic sites (Chapman and Rossmann, 1993; Rossmann and Palmenberg, 1988), many of the differences in surface characteristics are likely to be antigenic determinants. Such differences might be indicative of ways that structure could account for the inability of polyclonal anti-AAV-2 serum to neutralize AAV-3B (Rutledge, Halbert, and Russell, 1998). When AAV structures are compared, it is the three-fold symmetric spikes that have the largest changes, and, as these regions are highly exposed, they therefore likely represent antigenic hot-spots on the capsid.

Linear peptide epitopes (mimotopes) on AAV-2 have been identified for several neutralizing monoclonal antibodies (mAbs), including C37-B, which displays specificity for AAV-2 and does not bind to AAV-3 (Wobus et al., 2000). C37-B neutralizes AAV-2 at a step prior to cellular binding (Wobus et al., 2000), presumably by competing for HSPG binding. Structural differences between AAV-3B and AAV-2 offer a potential explanation for C37-B specificity (Figure 5). One stretch of amino acids (493–502 of AAV-2; Figure 5B and D) was found to be the dominant contributor to the eptitope (Wobus et al., 2000). The sequence of this C37-B mimotope differs in AAV-3B by 4 amino acid substitutions. However, differences are not restricted to these four amino acid substitutions, because the surface topology adjacent to the mimotope (in the context of the assembled virion) is noticeably different. The local topology depends on substantial (6Å) differences in a neighboring loop, VR-IV. The surface neighboring the C37-B mimotope is much more concave in AAV-2 than in AAV-3B (Figure 5A and B), perhaps forming a more intimate interface with a convex antibody FAb. Furthermore, the surface charge distribution near the mimotope is different between AAV-2 and AAV-3B. In AAV-2 it is more negatively charged, while in AAV-3B the mimotope is adjacent to a positively charged region centered at Arg₄₄₇ (described above). Typical epitopes have a diameter of ~15–30Å (Smith and Moser, 1997). Thus, it is likely that C37-B interacts not only with the mimotope peptide, but with variable regions VR-IV and VIII from the GH loop of an adjacent subunit that flank it in the assembled capsid. A 30Å-wide footprint would cover the tip of VR-VIII and one side of VR-IV, encompassing the region where surface shape and charge differ (Figure 5C and D). In AAV-2, the C37-B footprint would likely include Arg₅₈₅ and Arg₅₈₈ (Figure 5D), explaining how C37-B inhibits receptor binding. VR-IV adopts substantially different conformations in each AAV serotype, and might be the effective limitation on cross-reactivity of C37-B-like neutralizing antibodies.

Figure 5.

Specificity of MAb C37-B binding. A close-up view of a single 3-fold proximal spike on AAV-3B (A and C) and on AAV-2 (B and D) shows differences in surface topology and charge close to the mimotope of C37-B (AAV-2 residues 492–501, shown as sticks). C37-B blocks receptor binding to AAV-2, but does not bind AAV-3 (Wobus et al., 2000). Cartoon representations of two subunits (grey and yellow) are shown inside of a semi-transparent surface, colored by electrostatic potential. The mimotope is sandwiched between VR-IV and VR-VIII (protruding GH loops, labeled) from a neighboring subunit. The dimensions of a typical antibody footprint are drawn as a 30Å diameter circle, centered at N497 (AAV-3B) and N496 (AAV-2) in panels E and F, illustrating that C37-B would likely cover part of VR-IV and VR-VIII, including R585/R588 on AAV-2, perhaps explaining how C37-B blocks receptor binding to this serotype.

The limitations of peptide scanning technologies are well known, that mimotope identification relies on an isolated peptide fragment adopting a near-native conformation. It is therefore to be expected that, at best, only part of a conformational epitope will be correctly identified. As the differences in the structure of the mimotope itself are modest, it is likely that it is adjacent regions, refractory to pepscan analysis, are the greater antigenic determinants. Particularly VR-IV, that is immediately adjacent in the assembled virion, is seen in the structure to be associated with changes to the local surface topology and to the electrostatics that would be large enough to allow AAV-3B to evade C37-B binding and subsequent neutralization.

Conclusions

Consistent with high sequence similarity, the subunit fold of AAV-3B is similar to that of other AAV serotypes. However, differences in a subset of surface-exposed loops give rise to surprisingly distinct characteristics in the local surface topology and electrostatic charge of AAV-3B in critical regions that, from the structure, appear to be the determinants of the distinct receptor binding and immunogenic properties of the virus. Indeed, comparing the structure of AAV-3B with other serotypes shows the greatest structural diversity at locations implicated in antibody binding (Wobus et al., 2000) and polyclonal neutralization (Lochrie et al., 2006). Additionally, differences in the electrostatics on AAV-3B near the 3-fold symmetry axis suggest that while AAV-3B binds heparin in the same general vicinity, the specific interactions with its primary receptor are distinct from those of AAV-2.

There are broad implications from the detailed differences that are seen in the heparin binding site. Others have suggested that polysaccharide binding is used by the virus to trigger conformational changes that might be a part of cell entry (Asokan et al., 2006; Levy et al., 2009). It is difficult to see how the diversity in heparan sulfate binding interactions could be consistent with conservation of the type of precise triggering mechanisms typically found in allosteric proteins. Indeed our own visualization of an AAV-2:heparin complex indicated that there were no large conformational changes (O'Donnell, Taylor, and Chapman, 2009). In fact, the structural diversity in the attachment sites for HSPG-binding serotypes is more consistent with the use of heparan sulfate by the virus as a passive anchor to maintain proximity to the cell surface.

Studies of AAV-2 have demonstrated the influence of the heparin-binding site on tissue tropism. In vectors based on AAV-2, inclusion of the heparin binding site motif (585-RGNR-588) was required for transduction restricted mainly to the liver (Grimm et al., 2008), while substituting this peptide motif in AAV-2 with that from a non-heparin binding serotype results in systemic gene transfer to muscle tissue (Asokan et al., 2010). Moreover, insertion of peptide ligands at the heparin binding site (amino acid position 587) of AAV-2 can result in disruption of the native tropism, while simultaneously redirecting transduction to alternate tissues (Perabo et al., 2006). The same might be true for AAV-3B, where a heparin-binding site distinct from that on AAV-2 could be modified. The diversity with which AAVs are able to assemble a functioning primary receptor binding site is encouraging for those attempting to engineer changed tropism through the receptor-binding specificity of AAV vectors. The structure of AAV-3B provides a new template for the design of vectors, where antigenic and receptor binding properties are distinct from other serotypes.