Mapping and Engineering Functional Domains of the Assembly-Activating Protein of Adeno-associated Viruses

In order to study AAP independent of the VP ORF, we generated an AAP-null construct by introducing a stop codon early in the AAP ORF without affecting the VP coding sequence and provided different AAP constructs in-trans driven by the CMV promoter (Fig. 1A and Table 1). A series of deletion mutants were generated based on AAP domains and tested for the function of AAP using four parameters: steady-state AAP levels, VP levels, vector yield, and transduction efficiency. All AAP constructs have a C9 tag derived from bovine rhodopsin at the C terminus for immunodetection (Fig. 1B). In brief, experiments were performed by transfecting pXX680, pTR-CBA-Luc, pXR1ΔAAP, and a pCDNA3.1-AAP construct into HEK293 cells. All deletion and T/S replacement constructs produced AAP at different levels. Although truncated linker 2 (TL2), truncated linker 3 (TL3), and the T/S replaced with Bombyx mori silk heavy chain repeats (Silk and SilkE1) (14) or the entire mucin-like domain from murine alpha dystroglycan mucin-like domain (Full MLD) (15) showed similar steady levels of AAP, truncated linker 1 (TL1) and T/S replaced by partial mucin-like domain (Partial MLD) produce significantly more AAP. Truncated linker 4 (TL4) shows a reduced level of AAP expression (Fig. 1C and Table 2). Multiple reports have shown that one of the observable functions of AAP is to help stabilize the expression of VP; therefore, we next determined the steady-state VP level in the transfected cells. The level of VP3 correlates proportionally with AAP protein level, except for mutants TL2 and TL3, which have a significant reduction in VP3 level compared to WT AAP, although they produced similar amounts of AAP (Fig. 1C). Despite the production of both AAP and VP proteins in all condition tested, deletion of the HR, CC, or PRR rendered AAP incapable of supporting AAV capsid assembly (Fig. 1D and Table 2). However, deletion of the entire T/S or replacing the region with other flexible protein domains does not affect any tested function of AAP (including capsid assembly), strongly suggesting the T/S is nonessential for the assembly function of AAP (Fig. 1D and Table 2). The transduction efficiency of the recombinant AAV vectors produced using different AAP constructs were similar to control (Fig. 1E). Based on these findings, we concluded that the T/S can potentially be engineered with other heterologous sequences to generate engineered AAPs (eAAPs) with new functions.

We first created a traceable AAP1 protein by replacing the T/S with enhanced green fluorescent protein (EGFP) to create AAP1-EGFP (AAP1E) for fluorescent tracking (Fig. 2A). This newly engineered AAP1E shows the same level of VP expression and capsid assembly, but a higher level of AAP expression compared to AAP1 (Fig. 2B, C, and E; Table 2). Furthermore, the AAP1E mutant localizes within the nucleolus similar to the WT AAP1, suggesting that AAP1E can be useful for further mechanistic studies (Fig. 2D). Using the AAP1E construct, we swapped the BR of AAP1E with the BR from AAP4, -5, and -9 (Fig. 2A and Table 1). The VP and AAP expression profiles of the BR substitution mutants are generally similar to AAP1E. However, 5BR and 9BR show a slight decrease in VP expression and 9BR shows a slight decrease in AAP expression (Fig. 2B and C; Table 3). Notably, BR1 and BR4 strongly localized to the nucleolus, while BR5 and BR9 only shows partial nucleolus localization as reported earlier (Fig. 2D and Table 3) (9). Intracellular localization appears to correlate with vector titers; for instance, BR5 and BR9 only restore about 50% of assembly (Fig. 2F and Table 3). In contrast, BR4 is able to restore AAP1 function to 120% compared to WT AAP1. The transduction efficiency of the rAAVs generated using different AAPs was not affected (Fig. 2F). Since it is known that AAP4 cannot rescue AAV1 capsid assembly, our data suggest that serotype-specific transcomplementation is not due to the BR but rather to other regions of AAP.

Based on the results presented above, we hypothesized that the BR can be replaced completely by other NoLS without affecting AAP function. Therefore, we engineered AAP1E by replacing the BR with other known nuclear or nucleolar localization signals (NLS/NoLS). The amino acid sequences and annotation of each NLS/NoLS are described in Table 1. In brief, we picked four different NLS/NoLS peptides from the RNase P subunit NoLS (Rpp29) (16), adaptor protein in AP-3 complex (AP3D1) (17), virus-derived simian virus 40 NLS (SV40) (18, 19), and HIV Rev NoLS domain (HIV Rev) (20) (Fig. 3A). Although AAP expression levels are comparable to AAP1E from all the BR replacement mutants, VP expression was moderately reduced for some (Fig. 3B and Table 3). This decrease correlates with assembly based on vector yield, where the various BR mutants show a spectrum of activity. SV40 is unable to restore AAP assembly function, while Rpp29, HIV-Rev, and AP3D1 restore AAP assembly function at 40, 50, and 80%, respectively (Fig. 3C and Table 3). All rAAVs generated with these constructs show similar transduction efficiencies (Fig. 3D). Since the NoLS/NLS mutants showed a spectrum of AAP function, we performed confocal microscopy to investigate trafficking efficiency compared to AAP. AAP1E shows clear colocalization with the nucleolar marker CD23 (Fig. 4). AAP1E-ΔBR shows cytoplasmic localization, as expected, and is not able to restore function (Fig. 3C and 4). AAP1E-SV40 shows nuclear localization, which only partially rescues AAV assembly. AAP1E-Rpp29 shows moderate (40%) rescue of AAV assembly and has partial nucleolar localization (Fig. 3C and 4). HIV-Rev and AP3D1 show nucleolar localization, and AP3D1 restores assembly function completely (Fig. 3C and 4). Therefore, the ability of AAP to drive AAV1 assembly correlates with its nuclear/nucleolar localization, and the BR region can be replaced by other NoLS without affecting AAP function.

Currently, there is no structural information on AAP. Based on a modular homology modeling, the N-terminal HR is the only region that can be modeled among all AAP tested and CC from AAP1, -6, -7, and -9 can be modeled. The HR is modeled as an alpha-helix, and the CC region is modeled as either a loop or a beta-strand. Deletion or point mutations in these regions completely abolish the ability of AAP to activate capsid assembly, suggesting that these secondary structures play an important role in AAP function. Furthermore, previous reports have shown that AAP5 is unable to support capsid assembly of AAV1, despite the sequence similarity between the PRR, the T/S, and the BR (9). We investigated the function of each subdomain by deletion analysis and by replacement with the corresponding domain of AAP5 (Fig. 5A). While all the constructs were able to produce AAP, the functional efficiencies were varied. The expression level of the AAP1EΔHR construct is higher than that of AAP1E; however, the level of VP is significantly reduced (Fig. 5B and Table 2). In addition, capsid formation is completely abrogated to background levels (Fig. 5C and Table 2). Replacing the AAP1 HR domain with the AAP5 HR domain shows a similar effect as the deletion, although the AAP expression level is only slightly increased compared to AAP1E (Fig. 5B and C; Table 2). Deletion of the CC region resulted in a similar phenotype, where the expression level of AAP was normal, but VP expression and capsid formation were reduced to the background level (Fig. 5B and C; Table 2). Replacing 1CC with 5CC domains restored the level of VP expression and 50% viral titer, suggesting a link between the CC region and capsid protein stability that is not serotype specific (Fig. 5B and C; Table 2). Deletion of the PRR shows a moderate effect on capsid expression and assembly. Replacing 1PRR with 5PRR only slightly affects AAP function and allows 75% viral yield (Fig. 5B and C; Table 2). All rAAVs generated with AAP show similar transduction efficiencies (Fig. 5D). Confocal microcopy studies further confirmed that the N-terminal modules localized at the nucleolus, similar to the AAP1E control (Fig. 6). These results corroborate the notion that the HR and CC domains are essential for VP interactions and maintaining stability.

Based on deletion and AAP1/5 chimera analysis, we established that capsid assembly and VP expression levels are tightly correlated. We utilized the same deletion and replacement constructs as above and replaced the BR with a human Fc domain to facilitate VP complexation and immunoprecipitation studies (Fig. 7A). The expression level of both VP and AAP varies between constructs. For instance, EGFP, ΔHR, 5HR, ΔCC, and 5CC show a reduced level of AAP expression, while AAP5E and ΔHR show an increase in AAP expression (Fig. 7C and Table 2). To verify our system, we first showed that AAP1E can pull down the AAV1 VP. Surprisingly, although AAP5E was unable to support AAV1 capsid assembly (Fig. 6C), this protein was able to pull down AAV1 VP similar to AAP1E (Fig. 7B and Table 2). Deletion of the HR or CC regions abolishes the ability of AAP to pull down VP. Replacing the HR and CC regions with AAP5 restores the ability of AAP to pull down AAV1 VP (Fig. 7B and Table 2). Deletion of PRR or replacing 1PRR with 5PRR has no significant effect on AAV1 capsid interaction (Fig. 7B and Table 2). It is worth noting that the lack of immunoprecipitation in some of these constructs cannot be explained by expression level differences. For instance, 5HR and 5CC both have low levels of AAP and VP expression but were able to pull down VP in our assay (Fig. 7B and C; Table 2). On the other hand, ΔHR and the EGFP control are unable to pull down any VP protein despite having the highest levels of AAP expression (Fig. 7B and C; Table 2). However, this interaction is either transient or weak, which is further supported by our observation that majority of the capsid protein is still found in the flowthrough fraction of the pulldown (L. V. Tse and A. Asokan, unpublished data). These results demonstrate that the interactions between AAP and VP are driven by the N-terminal modules.

We hypothesized that the T/S of AAP is the most amenable to manipulation for imparting novel/improved functionality. By introducing EGFP in place of T/S, steady-state AAP levels were likely increased probably due to increased protein stability (Fig. 2B). Thus, we attempted to further increase the stability of AAP by replacing the T/S with oligomerization domains. We picked three different oligomerization domains, including (i) Collagen (21), which forms trimers, (ii) influenza A/WSN/1933 hemagglutinin (HA), which is a coiled-coil trimerization domain (WSN), and (iii) a synthetic tetramerization domain (Tetra) (22) to replace the T/S of AAP1 (Fig. 8A). Steady-state AAP protein levels were increased among all the AAP variants. The increase was most pronounced in AAP1-Collagen and AAP1-Tetra, while AAP1-WSN shows a moderate increase in AAP level (Fig. 8B and Table 2). VP levels also showed an increase in case of AAP1-Collagen, remained the same in AAP1-Tetra, and were slightly decreased in AAP1-WSN (Fig. 8B and Table 2). Both AAP1-Collagen and AAP1-Tetra increased the rAAV vector yield by 1.5- and 2-fold, respectively, compared to WT AAP1 (Fig. 8C and Table 2). The AAP1-WSN-mediated vector yield is 60% less than that for WT AAP1, suggesting a potential functional incompatibility with AAP1-WSN, despite increased stability (Fig. 8C and Table 2). To further evaluate the impact of engineered AAPs on vector yield, we attempted to establish a dose-response relationship for natural and engineered AAP constructs in supporting rAAV vector production by varying the amount of AAP DNA during transfection. As expected, the AAP levels increased proportionally to the amount of transfected cDNA. AAP1-Collagen showed a significant increase in AAP levels compared to wild-type AAP1 (Fig. 9A). VP levels were also increased proportional to the amount of AAP DNA, with peak titers observed at 1,000 ng of AAP plasmid transfected. However, it is important to note that titers declined at higher levels potentially due to toxicity of AAP overexpression (Fig. 9A [see also Tse and Asokan, unpublished]).

HEK293 cells were maintained in Dulbecco modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher, Waltham, MA), as well as 100 U/ml of penicillin and 10 μg/ml of streptomycin (P/S; Thermo Fisher), in 5% CO₂ at 37°C. Hybridoma supernatant of anti-AAV monoclonal antibodies B1 and A20 were produced in-house and have been described earlier (27). Mouse anti-rhodopsin (1D4; ab5417) and mouse anti-actin (ab3280) antibodies were purchased from Abcam (Cambridge, United Kingdom). Mouse anti-CD23 antibody (D-6; sc-17826) was purchased from Santa Cruz Biotechnology (Santa Cruz, CA).

The amino acid sequences of AAP1 to AAP9 were used in structural prediction using SWISS-MODEL (https://swissmodel.expasy.org/). The templates used for modeling AAP is Arabidopsis G protein-coupled receptor 2, GCR2 (PDB 3T33), for the HR and CC regions. The sequence identity of the modeling region is 12.20%, and the global model quality estimation (GMQE) and QMEAN Z-score are 0.08 and −1.42, respectively. Different domains of AAP were also predicted individually; the AAP2 PRR and T/S are modeled with cellobiohydrolase (PDB 1Q9H) (28), and the QMEAN Z-score is −2.08. The AAP4 BR is modeled with HIV-Rev (PDB 4PMI) (29), and the QMEAN Z-score is −3.25. The HR domain can also be modeled with bacterial RNase ligase (PDB 4XRU) and Escherichia coli topoisomerase (PDB 1YUA), with QMEAN Z-scores of −1.81 and −2.08, respectively (30).

Sequences from AAP and VP (only the residues corresponding to those in AAP) were aligned using MUSCLE, followed by manual adjustment. Alignments were done such that residues and gaps directly correspond in AAP and VP. The percent identities and similarities of AAPs from other serotypes compared to AAP1 were calculated with the Sequence Manipulation Suite (31) using the following groups for similarity: GAVLI, FYW, CM, ST, KRH, DENQ, and P. Amino acid conservation scores at each position of AAP and VP were calculated using the prediction tool developed by Capra and Singh (32). Analysis was run using the property entropy scoring method, sequence weighting, the BLOSUM62 background and scoring matrix, and a window size of 0.

All AAP constructs were cloned into pCDNA3.1 using the EcoRI and NotI sites. Chimera constructs were cloned using either overlapping PCR or Gibson Assembly (NEBuilder HiFi; New England BioLabs, Ipswich, MA). Detailed designs for each construct are illustrated in each figure, and the amino acid sequences are given in Table 1.

HEK293 cells at 60 to 70% confluence on a six-well plate were transfected with 600 ng of pXX680 (adenoviral helper plasmid containing E2A, E4orf6, and VA RNA from human adenovirus 5), 400 ng of pTR-CBA-Luc (AAV packaging plasmid containing CBA-Luciferase flanked by ITRs from AAV2), 600 ng of pXR-AAV1-no AAP (AAV1 Rep and Cap plasmid without ITRs; the AAP ORF is terminated by introducing a stop codon at position 64), and 400 ng of pCDNA3.1 AAP constructs (different AAP substitution, deletion, and fusion constructs driven by a cytomegalovirus promoter) using polyethylenimine (PEI) as the transfection reagent. At 3 days posttransfection, cell pellets were washed two times with 1× Dulbecco-modified phosphate-buffered saline (DPBS) and lysed with 200 μl of 1× passive lysis buffer (Promega, Madison, WI) for 30 min on ice with Halt protease inhibitor (Thermo Fisher). Supernatants were collected after centrifugation at 13,000 × g for 5 min at 4°C. The sample were prepared for Western blotting in 1× LDS loading dye (Thermo Fisher) and 100 mM dithiothreitol (DTT), then boiled at 95°C for 5 min, and loaded on a NuPAGE 4 to 12% Bis-Tris SDS-page gel (Thermo Fisher). The protein bands were transferred to nitrocellulose membrane (Thermo Fisher) using a semidry Xcell Surelock module (Thermo Fisher). VP, AAP, and actin proteins were detected by using B1 hybridoma supernatant at a 1:50 dilution, mouse α-rhodopsin antibody (ID4; ab5417) at a 1:2,000 dilution, and α-actin antibodies (ab3280) at a 1:1,000 dilution, respectively; goat α-mouse antibody labeled with horseradish peroxidase (HRP) at a 1:10,000 dilution was used as the secondary antibody. A chemiluminescence reaction was initiated with enhanced chemiluminescent substrate (SuperSignal West Femto maximum sensitivity; Thermo Fisher), and the membrane was developed on an AI600RGB system (Amersham Biosciences, Little Chalfont, United Kingdom).

HEK293 cells at 60 to 70% confluence on a six-well plate were transfected with pXX680 (600 ng), pTR-CBA-Luc (400 ng), pXR-AAV1-no AAP (600 ng), and pCDNA3.1 AAP constructs (400 ng) using PEI. The transfection medium was replaced with fresh media after 24 h, and the supernatant was harvested at 5 days posttransfection. Supernatants were collected after centrifugation at 13,000 × g for 2 min. The supernatants were used directly for standard quantitative PCR (qPCR) analysis to determine the vector yield and for transduction assays. Subsequent steps involved the harvesting of media, polyethylene glycol precipitation, iodixanol ultracentrifugation, and buffer exchange. Supernatants were treated with DNase (90 μg/ml) for 1 h at 37°C. DNase was inactivated by the addition of EDTA (13.2 mM), followed by proteinase K (0.53 mg/ml) digestion for 2 h at 55°C. Recombinant AAV vector titers were determined by qPCR with primers that amplify AAV2 inverted terminal repeat (ITR) regions (5′-AACATGCTACGCAGAGAGGGAGTGG-3′ and 5′-CATGAGACAAGGAACCCCTAGTGATGGAG-3′). The relative vector yields from different AAP constructs were normalized to wild-type AAP1 unless specifically indicated otherwise in figure legend.

AAV vectors produced with different AAP constructs packaging ssCBA-Luc transgenes were prediluted in DMEM plus 5% FBS plus P/S. Portions (50 μl) of recombinant AAV vectors (1,000 to 10,000 vector genomes [vg]/cell) were mixed with 50 μl of 5 × 10⁴ HEK293 cells and added to tissue culture-treated, black, transparent bottom 96-well plates (Corning, Corning, NY). The plates were incubated in 5% CO₂ at 37°C for 48 h. The cells were then lysed with 25 μl of 1× passive lysis buffer (Promega) for 30 min at room temperature. The luciferase activity was measured on a Victor 3 multilabel plate reader (Perkin-Elmer) immediately after the addition of 25 μl of luciferin (Promega). All readouts were normalized to wild-type AAP1 or AAV1 controls.

HEK293 cells were seeded on a 12-mm-diameter poly-lysine-treated glass coverslip (GG-12-1.5-PDL; NeoVitro, Vancouver, WA) in a 24-well plate. Cells were transfected at 60 to 70% confluence with pCDNA3.1 AAP alone (250 ng) using PEI as the transfection reagent. At 2 days posttransfection, the cells were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for 15 min and permeabilized with 0.1% Triton X-100 in PBS for 10 min. The nucleolus was stained using α-C23 (D-6) antibody, followed by goat α-mouse IgG H+L Alexa Fluor 594 (Thermo Fisher). The nucleus was stained with DAPI (4′,6′-diamidino-2-phenylindole; Thermo Fisher). Coverslips were mounted onto microscope slides using ProLong Diamond mountant (Invitrogen, Carlsbad, CA). Fluorescence images were taken by using a Zeiss LSM 710 spectral confocal laser scanning microscope at the UNC Microscopy Service Laboratory.

HEK293 cells at 60 to 70% confluence on a 15-cm plate were transfected with pXX680 (3,000 ng), pTR-CBA-Luc (2,500 ng), pXR-AAV1-no AAP (3,000 ng), and pCDNA3.1 AAP constructs (7,500 ng) using PEI as the transfection reagent. At 3 days posttransfection, the cells were harvested from the plate in cold 1× DPBS, followed by more two washes in 1× DPBS. The pellets were resuspended in 400 μl of buffer D (20 mM HEPES/KOH [pH 7.9], 25% glycerol, 0.1 M KCl, 0.2 mM EDTA) and lysed on ice for 30 min with Halt protease inhibitor (Thermo Fisher). Lysates were spun at 13,000 × g for 2 min. Then, 5% of the supernatant was retained as input and prepared in 1× LDS buffer and 100 mM DTT. Next, 10 μl of Protein G Mag Sepharose Xtra beads washed three times in buffer D (GE Healthcare, Chicago, IL) was added to the remaining lysate; samples were then placed on a rotator at 4°C for 2 h. Next, the beads were washed twice for 30 min each in buffer D, followed by resuspension in 1× LDS buffer and 100 mM DTT. Samples were denatured at 95°C for 5 min and then loaded onto a precast 10% Bis-Tris gel (Thermo Fisher) and run in MOPS-SDS buffer. The protein was transferred to a 0.45-μm-pore size nitrocellulose membrane (Thermo Fisher) in a wet transfer apparatus. AAP was detected using α-human-HRP at a 1:10,000 dilution and visualized after reaction with Femto Western blot substrate (Thermo Fisher) on an Amersham AI600RGB system (Amersham Biosciences). Membranes were incubated in 30% peroxide for 30 min at room temperature and then reblocked. VP, AAP, and actin proteins were detected by using B1 hybridoma supernatant at a 1:50 dilution and α-actin antibody (ab3280) at a 1:2,000 dilution, respectively; goat anti-mouse-HRP at a 1:20,000 dilution was used as the secondary antibody. The membranes were again visualized as described above.

All error bars shown represent one standard deviation. Statistical analysis was carried out in GraphPad Prism software using an unpaired, two-tailed Student t test or two-way analysis of variance (ANOVA) in Fig. 9 (*, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001).