In vitro and in vivo gene therapy vector evolution via multispecies interbreeding and retargeting of adeno-associated viruses

Abstract

Adeno-associated virus (AAV) serotypes differ broadly in transduction efficacies and tissue tropisms and thus hold enormous potential as vectors for human gene therapy. In reality, however, their use in patients is restricted by prevalent anti-AAV immunity or by their inadequate performance in specific targets, exemplified by the AAV type 2 (AAV-2) prototype in the liver. Here, we attempted to merge desirable qualities of multiple natural AAV isolates by an adapted DNA family shuffling technology to create a complex library of hybrid capsids from eight different wild-type viruses. Selection on primary or transformed human hepatocytes yielded pools of hybrids from five of the starting serotypes: 2, 4, 5, 8, and 9. More stringent selection with pooled human antisera (intravenous immunoglobulin [IVIG]) then led to the selection of a single type 2/type 8/type 9 chimera, AAV-DJ, distinguished from its closest natural relative (AAV-2) by 60 capsid amino acids. Recombinant AAV-DJ vectors outperformed eight standard AAV serotypes in culture and greatly surpassed AAV-2 in livers of naïve and IVIG-immunized mice. A heparin binding domain in AAV-DJ was found to limit biodistribution to the liver (and a few other tissues) and to affect vector dose response and antibody neutralization. Moreover, we report the first successful in vivo biopanning of AAV capsids by using a new AAV-DJ-derived viral peptide display library. Two peptides enriched after serial passaging in mouse lungs mediated the retargeting of AAV-DJ vectors to distinct alveolar cells. Our study validates DNA family shuffling and viral peptide display as two powerful and compatible approaches to the molecular evolution of novel AAV vectors for human gene therapy applications.

FIG. 1.

Generation of an AAV capsid library via DNA family shuffling. (A) Phylogram tree (created using PhyloDraw [http://pearl.cs.pusan.ac.kr/phylodraw/#test]) showing the eight AAV serotypes used as parents for DNA family shuffling (numbers denote lengths of capsid genes, in nucleotides). Branch lengths are proportional to the amounts of evolutionary change, calculated in ClustalW (http://www.ebi.ac.uk/clustalw/#). CAAV, AAAV, and BAAV, caprine, avian, and bovine AAVs, respectively. (B) Individual steps for generation of the library (scheme). Full-length cap genes were PCR amplified and subcloned for further amplification (1) and then isolated (2) and DNase I digested (3). Two consecutive PCRs without (4) or with (5) conserved primers were performed to reassemble shuffled full-length cap genes. These genes were inserted (6) into a plasmid carrying AAV-2 ITRs and a rep gene. The transfection of 293 cells with the resulting plasmid library (7) together with an adenoviral helper resulted in a viral library. One possible selection scheme used in this study was the coinfection of cultured liver cells (8) with the library and helper adenovirus under stringent conditions, resulting in the amplification of specific AAV capsids. Viral DNA can then be isolated (9) and cloned into an AAV helper plasmid carrying the AAV-2 rep gene without ITRs (10) for subsequent vector production. Ad5, helper adenovirus type 5. (C) Examples of shuffled cap genes in an initial small-scale library. DNA was extracted from 24 randomly chosen clones, and 5′ and 3′ ends of the individual cap genes were sequenced (using T3/T7 primers binding in the plasmid backbone). Shown, per end, are six representative alignments with the eight parents.

FIG. 2.

Molecular evolution of AAV vectors via DNA family shuffling. (A) The AAV capsid library was serially amplified on primary or transformed human liver cells. Purified human Igs (IVIG) were added to increase the selection pressure and to force vector evolution. Each scheme yielded a distinct pool of viral capsids (pools A to C). The alignment of ≥96 clones per pool with the sequences of the eight parental viruses confirmed the enrichment with specific sequences in association with increasing selection pressure. 5x, five times. (B) First 217 amino acids of the VP1 capsid protein for each pool. Colors represent the relationships to the parental strains (serotypes 2, 4, 5, 8, and 9), as also shown and detailed in Fig. S3 in the supplemental material. Arrowheads represent point mutations. Start codons for all three capsid proteins are shown. Pool C contained a single clone, designated AAV-DJ. (C) Putative atomic structure for each pool (the previously reported AAV-2 structure [Protein Data Bank file 1LP3 {http://www.rcsb.org}] was used as the basis for modeling; this structure lacks the residues represented in panel B). Thin green lines indicate sequence homology among the AAV-2, AAV-8, and AAV-9 parents. Residues shown as colored balls were derived from a subset of parental strains (see panel B for color codes; note that beige symbolizes AAV-5 in pool A here). AAV capsid symmetry axes (pool A) and four of the five loops (pool B) are shown. The location of two arginines as part of the conserved HBD (37) at the tip of loop IV is shown for AAV-DJ (pool C). (D) Capsid protein sequence of AAV-DJ. The three parental viruses are shown as thin lines above the sequence (AAV-2, red; AAV-8, blue; AAV-9, orange). Locations of the capsid loops, VP start codons, and the first residue of the atomic structure are indicated. A20 and B1 epitopes are boxed in blue and red, respectively (two mutations in the A20 epitope are shown by blue asterisks). Two recently identified immunogenic AAV-2 peptides (47) are boxed in yellow (AAV-DJ carries three point mutations, indicated by asterisks). Residues in green boxes form the conserved AAV-2 HBD (asterisks denote two arginines mutagenized in this study). Red asterisks denote residues previously discovered using other methods and believed to determine AAV-2 immunogenicity (see the text).

FIG. 3.

In vitro analysis of selected shuffled capsids. Following five consecutive amplifications of the AAV library on primary human hepatocytes, viral DNA was extracted from 10 randomly chosen clones (with the exception of the clone corresponding to lanes A, which was recovered twice from pool A), and the cap genes were subcloned into an AAV helper plasmid. (A) Western blot (using B1 antibody) showing differences in the expression levels and sizes of the individual VP proteins compared to those of wild-type AAVs (wtAAV). (B) Results from titration of infectious gfp-expressing particles. The helper plasmids described above were used to package a gfp-expressing AAV vector plasmid, and titers of recombinant particles in crude cell extracts were determined (n = 3) as detailed in Materials and Methods. All shuffled clones gave higher titers than the AAV-8 or AAV-9 helpers. No linear correlation between VP protein expression levels (A) and titers (B) could be made, suggesting that the various chimeras differed in their packaging efficacy, infectivity, and/or other parameters.

FIG. 4.

Presentation of epitopes on the AAV-DJ capsid. Shown at the top are the putative AAV-2 epitope for the monoclonal antibody A20 (a conformational epitope comprising three distinct peptides [see also Fig. 2D]) and the corresponding sequences in AAV-DJ, AAV-8, and AAV-9 (amino acid changes compared to the sequence of AAV-2 are highlighted in red [AAV-DJ] or orange [AAV-8 and AAV-9]). The bottom panels show results from immunofluorescence studies of cells cotransfected with the various helper constructs and an adenoviral helper plasmid (to boost gene expression from the AAV plasmids). The two amino acid changes in AAV-DJ were already sufficient to abolish the binding of the A20 antibody, validating and narrowing down the A20 epitope (80) and thus exemplifying the potential of DNA family shuffling as a reverse-genetics tool. AAV-DJ cross-reacted with both the polyclonal anti-AAV-2 and anti-AAV-8 sera, as expected from its chimeric structure. Similar cross-reactivity with these sera was also observed for wild types 2, 8, and 9. The fact that AAV-DJ, AAV-8, and AAV-9 were detected by the B1 antibody (initially raised against AAV-2 capsid proteins) was not surprising considering the high degree of conservation of its epitope in natural AAVs (see Fig. S4 in the supplemental material). All mono- and polyclonal anti-AAV antibodies were described previously (78, 79), except for the polyclonal rabbit anti-AAV-8 antiserum. 303.9, anti-Rep; B1, anti-VP; αAAV-2, anti-AAV-2 VP serum; A20, anti-AAV-2 capsids; αAAV-8, anti-AAV-8 VP serum.

FIG. 5.

Protein sequence alignments for AAV-DJ and 10 clones from pool A. Shown are full protein alignments for the 10 clones described in the legend to Fig. 3. (The clones correspond to lanes A to J in Fig. 3 and are listed in order; e.g., clone A4 corresponds to lanes A, and clone A11 corresponds to lanes B, etc.). AAV-DJ served as the standard to highlight the different degrees of evolution between pools A and C (in which clones were selected with [pool C] or without [pool A] IVIG [Fig. 2A]). The clone A4 sequence is shown as a full sequence, as clone A4 was recovered twice from pool A. Residues identical in A4 and AAV-DJ are colored in red, while changes are shown in yellow. Only amino acids divergent from AAV-DJ are shown for the other nine clones (for the origin of these residues, see the wild-type sequences in Fig. S4 in the supplemental material and also see the text). The horizontal bars indicate the capsid loops (see Fig. S4 in the supplemental material). Note that many changes in the 10 clones from pool A were clustered in these loops, as could be expected. Moreover, the gene regions corresponding to six of these cluster regions were identical to 6 of the 12 previously reported HVRs (HVRs 1, 3, 4, 5, 11, and 12 among the previously described HVRs 1 to 12, identified by green boxes) in the AAV capsid gene. Intriguingly, our alignments also identified several further regions of sequence diversity not described before, especially in the VP1 and VP2 N termini. The dotted orange lines mark a highly conserved phospholipase 2A domain in the VP1 N terminus; note the unique amino acid change (D97H) in clone C8 (yet the capsid was infectious [see Fig. 3]). The three purple boxes show the A20 epitope; note that almost all clones had fully maintained the respective consensus sequence from AAV-2 (H in the first and V in the second part of the epitope) but that AAV-DJ had not (compare Fig. 4). The blue box shows the location of two of the five residues (two arginines) which constitute the HBD; they were fully conserved in all clones.

FIG. 6.

In vitro analyses of AAV-DJ and HBD mutants. (A) Two arginine residues (numbers refer to positions in AAV-2) in AAV-2, AAV-8, AAV-9, or AAV-DJ were mutagenized to eliminate or introduce an HBD (37). (B) Western blots (using B1 antibody) confirming correct VP protein expression from all HBD mutants. AAV-8 and AAV-DJ (wild types and mutants) expressed proteins more strongly than AAV-2 or AAV-9, for reasons unknown. +, with; −, without. (C) Titration of infectious particles on 293 cells confirmed the role of the HBD in infection in culture. The mutation of the HBD in AAV-2 or AAV-DJ reduced infectivity, measured the ratio of total to infectious AAV particles, by 2 to 3 logs. However, including an HBD in AAV-8 and AAV-9 did not further increase the infectivity of these vectors. (D) Results from cell binding assays confirming the role of the HBD in attachment to cultured cells (HeLa or Huh-7). The drop in binding with the AAV-2 and AAV-DJ mutants correlated well with the transduction data presented in panel C. Surprisingly, the HBD-positive AAV-8 and AAV-9 mutants bound severalfold more efficiently than AAV-2 on HeLa cells and, in all cases, far better than wild types 8 and 9 but transduced much less efficiently. Cell attachment and transduction thus do not necessarily correlate, suggesting that additional intracellular factors and steps contributed to the superior transduction efficiency of AAV-DJ. (E) AAV particle digestion with the endosomal proteinase cathepsin B (cath. B) (2) yielded distinct patterns for the individual serotypes in a Western blot analysis using polyclonal anti-AAV-2 VP serum. AAV-DJ showed a hybrid pattern with bands from AAV-2 and AAV-8 (white and black arrows, respectively), further supporting the idea that its properties resulted from synergistic or additive effects from its parents (cell binding from AAV-2 and rapid uncoating from AAV-8).

FIG. 7.

hFIX expression from AAV-DJ in mice. (A) Dose-dependent and liver-specific hFIX expression. C57BL/6 mice (n = 3 to 8) were infused with all four hFIX-expressing vectors via peripheral tail vein injection. Gray shading indicates the range from 1 to 100% of normal hFIX levels in humans (0.05 to 5 μg/ml). Levels over 1% are considered to be therapeutic in hemophiliacs. Note that AAV-8, AAV-9, and AAV-DJ vectors exceeded the 100% level already at the lowest dose, whereas AAV-2 required a 20-fold-higher dose. (B) hFIX expression from the AAV-DJ HBD mutants (n = 3 per group). Shown are results from two representative doses; there was no significant difference from the results for AAV-DJ. (C) In contrast, the AAV-2 or AAV-8 HBD mutants expressed less hFIX than the corresponding wild types (n = 3 per group). (D) AAV-DJ showed unique transduction kinetics at a maximum dose of 7 × 10¹² particles. The onset of gene expression was delayed compared to that from AAV-8 or AAV-9, yet hFIX levels became similar after ∼40 days (n = 3 per group). The AAV-DJ HBD mutants showed intermediate kinetics; stable hFIX levels were eventually also similar to those from AAV-8 and AAV-9 (and AAV-DJ). pi, postinjection.

FIG. 8.

Vector DNA biodistribution and dose response. (A) Genomic DNA extracted from nine tissue types (li, liver; lu, lung; h, heart; k, kidney; s, spleen; b, brain; p, pancreas; g, gut; and m, muscle) was analyzed for the presence of hFIX-expressing vector DNA. The results and the reference standard shown are representative of data for the two highest doses used here. The AAV-DJ transduction pattern was more restricted to liver, heart, kindey, and spleen tissues than those of AAV-8, AAV-9, and the HBD mutants. At the highest dose (7 × 10¹² particles), AAV-DJ spillover into nonhepatic tissues was also less obvious than that of the other vectors. The HBD-negative AAV-2/8 mutant gave increased heart transduction compared to wild-type AAV-2, confirming previous data (37) (an unknown production deficiency prevented evaluation at the highest dose). (B) Comparison of vector DNA levels in liver following transduction with increasing particle doses (from left to right, 5 × 10¹⁰, 2 × 10¹¹, 1 × 10¹², and 7 × 10¹² particles). AAV-DJ showed a blunted response at the highest dose, likely correlating with its slower onset of gene expression (Fig. 7D).

FIG. 9.

In vivo and in vitro neutralization of AAV-DJ and wild-type AAVs. (A) Mice (n = 4 per group) passively immunized with IVIG (4 or 20 mg) were injected with hFIX-expressing AAV. Plasma hFIX levels per virus and time point are shown as percentages of corresponding levels in control mice (those receiving PBS instead of IVIG). (B) Mice (n = 4 per group) immunized with the higher IVIG dose were also injected with the AAV-DJ HBD mutants. AAV-2, AAV-9, and AAV-DJ were included as controls. hFIX expression from the HBD mutants was marginal, comparable to that from AAV-2. (C) Mice (n = 4 per group) were injected with PBS or 10¹¹ particles of hAAT-expressing AAV-2, AAV-8, AAV-9, or AAV-DJ (x axis), and 3 weeks later, they were reinjected with 10¹¹ particles of hFIX-expressing viruses (5 × 10¹¹ for the least efficient AAV-2, due to the enzyme-linked immunosorbent assay detection limit of ∼10 ng/ml). Shown are stable hFIX levels for each group as measured 6 weeks after the second injection. (D) Sera were taken from the mice described in the legend to panel C at the time of reinjection (bars H [higher dose]), as well as from a parallel group injected with a lower dose (bars L) of 2 × 10¹⁰ particles. Titers of neutralizing antibodies (NAb) against the wild-type AAVs or AAV-DJ were determined as detailed in Materials and Methods. pi, postinjection.

FIG. 10.

Analyses of in vivo transduction with peptide-displaying AAV-DJ capsid variants. (A) Wild-type FVB mice were nasally infected with luciferase-expressing AAV vectors as described in the text. Shown are representative examples of lung-directed luciferase expression (numbers below the image are photon counts) 7 days after the inoculation. The black arrow highlights an example of very occasionally observed vector spillover into the stomach (an artifact from the installation procedure). DJ, parental AAV-DJ; NSS, NSSRDLG mutant; MVN, MVNNFEW mutant; ΔHBD, HBD-negative AAV-DJ/8 mutant; 8, wild-type AAV-8. (B) In vitro imaging of isolated lungs confirmed the similarity of luciferase expression levels among all vectors and the mild trend toward higher numbers with the NSSRDLG variant. C, control. (C) Representative examples of results for wild-type FVB mice 7 days after peripheral injection (via the tail vein) with all luciferase-expressing vectors as described in the text. Note the slight (two- to threefold) drop in expression from the MVNNFEW mutant compared to that from the other vectors. (D) Histological analyses (by X-Gal staining) of murine lungs 2 weeks after nasal infection with β-galactosidase-expressing AAV-DJ variants (5 × 10¹⁰ particles per mouse). Shown are various representative examples of different sections at different magnifications (×100 or ×200) for each mouse. Black arrows highlight the predominantly transduced cell type for each capsid: (putative) alveolar type II cells for the NSSRDLG variant and alveolar macrophages for the MVNNFEW peptide. The arrows in the images corresponding to the AAV-DJ/8 mutant-infected mouse (labeled ΔHBD) point at occasionally observed positive alveolar cells, while the main transduction target was pulmonary endothelial or smooth muscle cells (top frame), identical to the target of the parental AAV-DJ capsid (the middle frame corresponding to the AAV-DJ-infected mouse shows a vessel cross-section). Sections from an uninfected control mouse (rightmost frames) showed no background signals, with the exception of very faint macrophage staining that was also found in sections from the other mice (not visible here, except in those from the MVNNFEW mutant-infected mouse, which had stronger and more clustered signals than sections from the other mice).

FIG. 11.

Biodistribution of AAV capsid libraries following peripheral delivery (via tail vein injection). Wild-type FVB mice were infused with 5 × 10¹¹ particles of the shuffled (A) or AAV-DJ-based peptide display (B) library, and 1 week later, all major organs were harvested for the preparation of total genomic DNA. AAV DNA genomes were detected via PCR using primers flanking the entire cap gene (∼2.2 kb; arrows). Numbers (in kilobases) on the left refer to a DNA size marker. Shown are results from two representative mice per injection protocol. Note that AAV DNA signals could be detected in all analyzed tissues, including brain tissue, highlighting the potential for the biopanning and evolution of AAV capsids in all major organs in vivo.

FIG. 12.

Model of an AAV VP3 trimer (panel A, top view down the threefold symmetry axis; panel B, side view) created using Swiss-PdbViewer (www.expasy.org/spdbv/text/getpc.htm) and the VIPER oligomer generator (viperdb.scripps.edu/oligomer_multi.php), with the following parameters: T=1 capsid structure; selected matrices A5, A6, and A17; and Protein Data Bank file no. 1LP3 (Fig. 2C) for the AAV-2 sequence. Sequence motifs that may contribute to AAV receptor binding are colored as follows (serotypes with the highest degree of conservation are shown in parentheses): purple, HPD; red, motif with similarity to the NSSRDLG peptide (in AAV-DJ and AAV-2, 534-NGRDSL-539; numbers refer to the AAV-DJ sequence) (compare Fig. 2D); blue, motif with similarity to the NDVRAVS peptide (in AAV-DJ and AAV-2, 505-RVS[KT]SADNNNS-516); and yellow, another motif with (partial) similarity to the NSSRDLG peptide (in AAV-DJ and AAV-9, 277-SGGSSNDN-284). Note how all four motifs are exposed on the capsid surface (B) and how pairs of motifs are located in close proximity to each other (yellow and red motifs or blue and purple motifs), as well as near the HBD (purple), suggesting that they may act cooperatively in receptor binding. Intriguingly, AAV-DJ has the only capsid combining all four motifs in one sequence, perhaps contributing to its high level of efficacy (see the text).