Heparan Sulfate Proteoglycan Binding Properties of Adeno-Associated Virus Retargeting Mutants and Consequences for Their In Vivo Tropism

Adeno-associated virus type 2 (AAV-2) is gaining increasing attention as a gene therapy vector. However, the wide distribution of its primary receptor, heparan sulfate proteoglycan (HSPG) (11), hampers selective transduction of target tissue. Vectors aiming to redirect AAV-2's tropism have been generated by insertion of ligands at position 587/588 of the capsid (1, 2). This is likely to interfere with the HSPG binding of at least two (R585 and R588) of the five positively charged amino acids of the recently identified HSPG binding motif (5, 7), explaining the ablation of HSPG binding of some targeting vectors (3, 4, 6, 8, 10). In some cases, however, binding was only partially affected (12) or was even restored (4, 8, 14). To investigate molecular mechanisms responsible for these differences, we applied a library of AAV-2 capsids carrying insertions of seven randomized amino acids at position 587 (8) to a heparin affinity column (13) to separate binding from nonbinding mutants. We sequenced and statistically analyzed (Table 1) at least 80 clones from (i) the original DNA-library, (ii) the viral AAV-2 display library, (iii) the flowthrough fraction (nonbinders were designated the NB-AAV-pool), and (iv) the 1 M NaCl eluted fraction (binders were designated the B-AAV-pool).

TABLE 1.

Peptides detected in B-AAV-pool and NB-AAV-pool and statistical analyses^a

Open in a new tab

^a

Panel A shows a representative example of 20 peptides detected in B-AAV-pool and NB-AAV-pool. The net charge of the insertions is provided. Parentheses indicate a weak charge considering the low pKa of histidine. Panel B shows the statistical analysis of the occurrence of amino acids. Based on the χ² test, the colors indicate a higher (red), lower (green), or expected (black) occurrence of each amino acid in the analyzed populations (P = 0.0001). For each population, >80 clones were sequenced. Briefly, viral DNA was obtained from library viral preparations or column fractions (DNeasy kit; Qiagen, Germany). Isolated DNA was then cloned into pUC-AV2 (3) and electroporated in Escherichia coli (XL1-Blue MRF′). Genomes of single mutants from single bacterial colonies were obtained and sequenced. Asterisks mark amino acids that were found at a clearly lower frequency than expected, but the stringent statistical test did not attribute statistical significance to this difference.

The DNA library showed a higher than expected presence of alanines, which originates from the oligonucleotide synthesis procedure. Occurrence of every other amino acid met statistical expectations for an unselected library.

The AAV-2 display library showed an excess of the amino acids P, G, and A and a defect of C, L, F, W, and Y (Table 1B). Since P, G, and A are three of the four smallest amino acids, whereas F, Y, and W are three of the four biggest, this bias suggests that the packaging process selects against bulky inserts which would introduce dramatic structural rearrangements and have a deleterious effect on capsid structure. In addition, prolines could favor spatial accommodation of the peptide by introducing kinks and reducing its bulkiness.

The B-AAV-pool showed a significant increase of arginine residues (Table 1B). Strikingly, arginines were particularly frequent at the seventh position of the peptide (30%). In contrast, in the AAV-2 display library and the NB-AAV-pool, the frequency of R at this position (15% and 9%, respectively) was equal to or lower than the expectation for a randomized distribution among the seven amino acid positions (14.3%).

Interestingly, B-AAV-pool insertions carrying no positive amino acids displayed an exceptionally large amount of A, G, and S (data not shown), the three smallest amino acids, suggesting a reduced impact on the wild-type capsid structure, which is less likely to interfere with binding of heparin.

The NB-AAV-pool clearly showed a greater presence of negatively charged amino acids (D and E) (Table 1A and B). Moreover, although the number of negative residues observed in the B-AAV-pool matched statistical expectations, only 2% of the clones carried a net negative charge in its insertion, while 76% carried a positive charge and 21.5% a neutral charge. In clear contrast, the NB-AAV-pool consisted of 53% negative, 8% positive, and 39% neutral net charged inserts. This bias becomes even clearer if histidine is considered neutral due to its low pKa: B-AAV-pool insertions would then be 2% negative, 64% positive, and 34% neutral, while the NB-AAV-pool consisted of 64% negative, 2% positive, and 34% neutral. These observations strongly suggest that the presence of negative charges is deleterious for functional binding of AAV-2 targeting vectors to negatively charged heparin/HSPG, while positive charges facilitate this interaction. Capsid binding to heparin/HSPG is driven by electrostatic interactions (5). Since R has the highest pKa (12.48) and therefore the strongest interaction, the high frequency with which this amino acid was selected in the B-AAV-pool (Table 1B) can be explained. Accordingly, due to their lower pKa values (10.79 and 6.04, respectively), K and H seem to play a less important role compared to that of R.

Taken together, these data suggest a model in which peptide insertions at 587 can either disrupt or conserve the capsid ability to bind heparin by different mechanisms. Insertion of a peptide between R585 and R588 (Fig. 1A and B) could cause their spatial separation or sterically block the heparin binding ability. In either case, bulky amino acids are prone to lead to one or both of these results (Fig. 1C and D). If the peptide consists of small residues, the insertion could be less invasive and the structure of the HSPG binding motif maintained functional (Fig. 1E). Insertion of positively charged peptides could lead to an HSPG binding phenotype by reconstituting a binding motif in combination with one of the original arginines (Fig. 1F) or independently from them (Fig. 1G). The proximity of R588 to the last position of the inserted peptide could facilitate reconstitution of a functional motif if an arginine is present at the latter position (Fig. 1H). It should be noted that this behavior could be due to the particular sequence of the construct we used, where the seventh position of the randomized peptide and R588 are separated by two residues, resembling the wild-type situation.

To prove that lack of heparin affinity column binding reflected lack of cellular HSPG binding, binding and nonbinding pools and AAV with wild-type capsid (RC) were produced as recombinant AAV vectors (rAAV) as previously described (9), and heparin competition studies on the cervix carcinoma cell line HeLa were performed (Fig. 2A). The addition of heparin inhibited transduction by rAAV-RC and the B-rAAV-pool, while the effect on the NB-rAAV-pool was negligible, demonstrating the initial hypothesis.

We previously described an AAV targeting vector, rAAV-MTP, that allowed systemic vascular targeting (12). Simultaneous detargeting of this vector from liver and spleen was observed. This vector showed a reduced ability to bind to heparin (12). Here, we analyzed whether the inability of AAV insertion mutants to bind heparin directly correlates with detargeting from liver and spleen. First, the ability to transduce the hepatocellular carcinoma cell line HepG2 was analyzed (Fig. 2B). In addition to rAAV-RC, the B-rAAV-pool, and the NB-rAAV-pool, a known HSPG knockout mutant (rAAV-A3) was generated by three amino acid substitutions (R585A, N587A, and R588A) (15) and was used as a control. rAAV-RC and the B-rAAV-pool showed a comparable transduction efficiency, whereas the NB-rAAV-pool showed an eightfold reduction in infectivity. In contrast, a reduction of ∼500-fold was determined for rAAV-A3. Higher transduction levels of the NB-rAAV-pool in comparison to rAAV-A3 could be due to new ligand-receptor interactions of some of the inserts displayed within the pool.

Thereafter, 4 × 10⁹ genomic particles were injected intravenously into C57/B6 mice (n = 4/group), and biodistribution studies were performed as described previously (12). Animals injected with rAAV-RC and the B-rAAV-pool showed a comparable biodistribution, with the highest vector DNA level in spleen and liver (Fig. 3). In contrast, the NB-rAAV-pool showed a 102- and 31.8-fold reduction in vector DNA level in the spleen in comparison to rAAV-RC and the B-rAAV-pool, respectively, whereas in the liver an 8.8- and 6.7-fold reduction was detected. In addition, elevated levels of viral DNA in the blood were measured for the NB-rAAV-pool consistently with the level of liver and spleen detargeting. rAAV-A3 also showed a detargeting from liver and spleen as well as the highest amount of viral DNA in the blood. Results obtained with rAAV-A3 and the NB-rAAV-pool are in agreement with previously published data obtained with other mutants impaired in HSPG binding, which showed a detargeting from liver (5, 12) and spleen (12).

This suggests an unspecific HSPG-dependent retention of rAAV-2 and of HSPG-binding rAAV targeting vectors in liver and spleen, and it provides the rationale for using the NB-AAV-pool for AAV display selections of cell/tissue type-specific AAV targeting vectors to avoid the HSPG-dependent retention in liver and spleen, thereby increasing the in vivo targeting ability of the respective vectors. Furthermore, our studies revealed different ways by which an inserted peptide is able to confer HSPG binding abilities to AAV targeting vectors, and they may help to fine tune the peptide insertion in order to ablate HSPG binding and to obtain tissue-specific vectors. This knowledge could also improve targeting mutants where R585 and/or R588 are replaced by other amino acids, since even in this case, some peptides (Fig. 1G) would restore HSPG binding.