Efficient Nuclease-free HR by Clade F AAV Requires High MOIs with High Quality Vectors

Main Text

Two papers, by Rogers et al.¹ and Dudek and Porteus,² in the October issue of Molecular Therapy claim that homologous recombination (HR)-based genome editing by Clade F adeno-associated virus (AAV) is inefficient and virtually undetectable in the absence of double-stranded DNA breaks. These claims contradict our findings that Clade F AAVs mediate HR efficiently in primary cells and in vivo in the absence of double-stranded DNA breaks induced by exogenous nucleases.³ We showed that chromosomal insertion of donor sequences by hematopoietic stem cell-derived AAV (AAVHSC) vectors required flanking homology arms and was BRCA2-dependent. Since BRCA2 is an essential mediator of HR,⁴ we concluded that Clade F AAV-mediated editing was accomplished by mediated HR. Importantly, we showed that high MOIs were required to observe augmented HR by Clade F AAV. AAVHSC-mediated HR-based genome editing was found to occur efficiently in vivo and was persistent. Neither Rogers et al.¹ nor Dudek and Porteus² actually replicated the studies described in Smith et al.³ There were significant discrepancies between our studies and those of Rogers at al.¹ and Dudek and Porteus.² These included the MOIs, AAV capsids, transduction, culture conditions, and vector quality. Rogers et al.¹ and Dudek and Porteus² used donor vectors encoding a promoter containing a GFP gene, in contrast to the promoterless editing vectors used in our studies. Both papers only tested in vitro editing with no actual sequence confirmation or in vivo evaluation. Due to these critical discrepancies between their studies and ours, we find their conclusions regarding the ability of Clade F AAV to mediate HR to be flawed.

The most significant difference between Rogers et al.,¹ Dudek and Porteus,² and Smith et al.³ lies in the MOIs. In Smith et al.,³ we clearly showed that efficient in vitro HR-based editing in the absence of nuclease-induced double-stranded DNA breaks was dependent upon the use of high MOIs. 150,000 vector genomes (vg)/cell or higher were required for efficient HR. Accordingly, our studies were performed at MOI: 150,000 or higher.³ In contrast, Dudek and Porteus² used MOI: 5,000, and Rogers et al.¹ used an MOI: 10,000, 30-fold and 15-fold lower doses, respectively, than that required for efficient HR in absence of nucleases. Thus, it is not surprising that they failed to observe editing in the absence of nucleases at these MOIs. In fact, their studies simply replicated the low-dose data shown in Smith et al.,³ which clearly showed that little to no HR was observed at these MOIs with any AAV tested.

Curiously, both groups observed significant loss of viability at higher MOIs. Rogers et al.¹ showed 66% toxicity at MOI: 50,000, with 50,000 vg/cell being their maximum tolerated dose for AAV6. Similarly, Dudek and Porteus² were unable to use MOI >5,000 due to toxicity. This raises important concerns regarding the quality of their vectors and suggests the possible presence of toxic contaminants. Such high toxicities make it difficult to draw meaningful conclusions regarding the editing capacities of their vectors and raises serious questions regarding the validity of their conclusions.

The toxicity observed by Rogers et al.¹ and Dudek and Porteus² is in marked contrast to our data. As shown in Smith et al.³, we saw no difference in the viability of AAV-transduced and untransduced cells over a range of MOIs from 10,000 to >400,000 vg/cell. Interestingly, Rogers et al.¹ also used a completely different method for packaging and purification of their AAV6 vectors as compared with their Clade F vectors, thus negating AAV6 as a true control.

Although unaddressed in either Rogers et al.¹ or Dudek and Porteus,² it is important for editing vectors to have intact inverted terminal repeats (ITRs) to mediate HR. These high G:C content palindromic termini of AAV genomes are important for recombination but are also prone to deletion and rearrangements.⁵ Thus, stringent screening for intact ITRs is necessary for the production of high quality editing of AAV vectors.³

Rogers et al.¹ and Dudek and Porteus² compared chromosomal insertion of AAV donor sequences by homology-dependent repair (HDR) after induction of double-stranded DNA breaks by either zinc finger nucleases (ZFNs)¹ or CRISPR/Cas9 (Dudek and Porteus²). Both groups used AAV6, AAV9, and mutagenized capsid variants of AAV9, while we used naturally occurring Clade F AAVs isolated from CD34⁺ cells⁶ along with AAV2, AAV6, and AAV8 controls.³ Rogers et al.¹ erroneously claimed that AAVHSC13 is identical to AAVHSC17 and that AAV9 G505R represents both vectors. AAVHSC17, however, has an additional silent mutant at amino acid 60 (Smith et al.⁶). Dudek and Porteus² compared HDR by AAV9 A68V, AAV9 G505R, AAV9 S345A, AAV9 S345A G505R, AAV9, and AAV6. Of these, only a single capsid, AAV9G505R (AAVHSC13), is the same as one of the 14 AAV capsids tested in Smith et al.³ These include vectors that were not tested in Smith et al.³ and thus cannot be used to either substantiate or refute our data. Although Dudek and Porteus² additionally list AAVHSC15 and AAVHSC17 among the AAV Clade F capsids tested, they do not show any data with these AAVs. Since AAVHSC15 and AAVHSC15 were used for many experiments in Smith et al.,³ it would have been reasonable to use these vectors to replicate the studies in Smith et al.³ Both groups failed to acknowledge the presence of silent mutations. AAVHSC7 has a silent mutation at amino acid 308. It is well recognized that silent mutations are known to play important roles in protein folding, evolution, adaptation, and even disease pathogenesis7, 8, 9 and therefore may not be inconsequential.

Dudek and Porteus² also discounted the possible role of the amino acid A68 in AAVHSC7 due to its location in the unresolved VP1 unique region.⁶ They additionally argue that the internal location of T346A and the external location of G505R of AAVHSC17 (Smith et al.⁶) do not suggest a role for these amino acids in HR. Since the role of capsid elements in HR remains unknown, there is no basis for such conclusions.

Although the AAV vectors used in Smith et al.³ were packaged with herpes simplex virus 1 (HSV1), we are aware of successful editing studies with AAV vectors manufactured using adenovirus helper genes and triple transfection.¹⁰ Our AAV vectors were purified through two rounds of cesium chloride density gradient centrifugation and were never heat inactivated. Dudek and Porteus² raised the possibility of a role for HSV as a helper virus in elevating HR and of heat inactivation having biologic effects. Since all AAV vectors tested in Smith et al.³ were packaged and purified in an identical fashion, it is unlikely that HSV played a direct role in augmenting HR only in Clade F AAV. If that were the case, elevated levels of HR would have been seen across the board, not only for Clade F viruses.

Dudek and Porteus² additionally invoke the possibility of aberrant immune activation, which may lead to nuclease-free recombination or altered cellular responses to AAV in the context of the HSV1 helper in BRCA2^−/− cells. However, we clearly showed that BRCA2 was absolutely required for AAVHSC HR.³ No recombination was observed in the absence of BRCA2, despite efficient transduction.³ Since no AAVHSC-mediated HR was observed in BRCA2^−/− cells, this explanation offered by Dudek and Porteus² does not apply.

Dudek and Porteus² raised additional concerns regarding the toxicity of potential residual HSV carried over into the final vector batches. Injection of purified AAV vectors packaged with HSV1 helper virus into immune-deficient NOD.CB17-Prkdc ^scid/NCrCrl mice did not result in toxicity or morbidity, as would be expected from HSV contamination.^3,6 Moreover, Smith et al.³ documented a lack of toxicity in vitro over a range of MOIs from 10,000 to >400,000, indicating the absence of carryover of residual active HSV1, which might have confounded the results.

In Smith et al.,³ the editing assays were designed with a high degree of stringency. The editing vectors were designed to insert promoterless reporter genes into introns of target genes such that expression would be driven from the chromosomal promoter. A splice acceptor and T2A sequence were included to allow independent expression of the reporter genes. This ensured that expression from episomal AAV genomes did not confound the results, unlike what was shown in Rogers et al.¹ and Dudek and Porteus.² The absence of reporter expression in controls lacking homology arms indicated that possible minimal transcription from the ITRs did not contribute to reporter expression.³ In contrast, both Rogers et al.¹ and Dudek and Porteus² only used donor constructs with heterologous promoters driving reporter gene expression. Their donor vectors were designed such that expression could occur from both episomal vectors as well as edited chromosomes. Both studies then analyzed GFP expression in long-term cultures of AAV-transduced cells with the assumption that expression at 12–17 days implied editing, with no evidence offered to support this assumption.

In Smith et al.³, we showed that the vector genome copy number quantitated in purified nuclei was higher in cells transduced with Clade F AAVs, followed by Clade A, then Clade E, and finally Clade B. Based on this observation, we suggested improved nuclear transport/entry as one possibility for augmentation of HR by Clade F AAV. In contrast, Rogers et al.¹ evaluated whole cell-associated vector genome copy numbers, potentially including vector particles attached to the external cell surface and in the cytoplasm in addition to intranuclear copies. They correlated GFP expression on day 14 post-transduction, with vector genome copy number on day 2. It is unclear why a correlation between these values would be expected, since GFP and vector genomes copy number were assessed 12 days apart. Regardless, they showed that, with AAV6, they observed high levels of GFP expression and vector genome copies only in the presence of ZFN, and no GFP- or cell-associated vector genome copies were observed with AAV9 and AAVHSC13 (Rogers et al.¹). Since their AAV6 was packaged and purified by a completely different method than their AAV9 and AAVHSC13, these results could simply reflect the quality of their vector batches. Poor quality AAV9 and AAVHSC13 would likely not transduce or edit cells efficiently, even at high MOIs. Their observations of toxicity at MOIs >10,000 for AAV9 and AAVHSC13 versus MOI: 50,000 for AAV6 suggests that their AAV6 vector batches were of higher quality than their AAV9 and AAVHSC13 batches and supports this conclusion.

The culture conditions for CD34⁺ cells used in both Rogers et al.¹ and Dudek and Porteus² were different from those used in Smith et al.³ We have previously shown that AAV transduction of CD34⁺ cells is optimal for slowly dividing cells.11, 12, 13 For retrovirus and lentivirus transduction, CD34⁺ cells are often prestimulated and placed in culture at high cytokine concentrations, since rapid cycling promotes better transduction. However, for AAV transduction, we have reported that low cytokine concentrations is better,11, 12, 13 similar to that used in Smith et al.³ Rogers et al.¹ prestimulated their CD34⁺ cells, and both studies used 5–10-fold higher cytokine concentrations than were used in Smith et al.³

Neither study provided convincing molecular proof of editing. The lack of sequence confirmation of editing in both studies is troubling. Dudek and Porteus² provided no molecular evidence of editing, while Rogers et al.¹ only performed a PCR-based targeted integration assay without sequence confirmation. Rogers et al.¹ additionally lacked important controls for the specificity of their targeted integration PCRs. Sequence confirmation is critical for any conclusions regarding successful editing.

Both studies made the assumption that GFP expression at 12–17 days post-transduction indicated successful editing. However, it is well known that infection with single-stranded AAV initiates a cellular DNA damage response, checkpoint activation, and cell cycle arrest.^14,15 AAV-induced cell-cycle arrest is associated with activation of checkpoint proteins ATR, Chk1, and H2AX.^14,15 In 12–17-day culture experiments performed in Rogers et al.¹ and Dudek and Porteus,² transduced cells containing single-stranded AAV genomes that would preferentially undergo mitotic arrest, while untransduced cells would proliferate normally. Therefore, over time in culture, a reduction of the proportion of AAV-transduced cells would be expected and is actually observed.³ With AAV transductions performed at MOI: 5,000–10,000 vg/cell, as in Rogers et al.¹ and Dudek and Porteus,² an excess of intracellular single-stranded AAV genomes is expected even after HR or HDR-mediated insertion of one or two copies into the genome, leading to activation of checkpoint proteins and mitotic arrest. It is precisely for this reason that long-term in vitro experiments with AAV vectors do not yield useful information.

For these reasons, the evaluation of AAV vectors is best addressed in vivo. Unfortunately, neither Rogers et al.¹ nor Dudek and Porteus² performed in vivo evaluation of editing. In contrast, in Smith et al.³, in addition to in vitro studies, we showed long-term stable in vivo editing by AAVHSC of the murine Rosa26 locus by serial bioluminescent assays and Southern blot and sequence analyses.

Notably, we never questioned the ability of AAV6 to transduce CD34⁺ cells or K562 cells. In fact, we have previously published on AAV6 transduction of human CD34⁺ hematopoietic stem progenitor cells (HSPCs) in vitro and, importantly, in vivo.¹⁶ In this collaborative study, we showed robust, stable long-term transduction of xenografted CD34⁺ HSPCs up to 6 months post-transplantation.¹⁶ Thus, we have no doubt that AAV6 is a good vector for the delivery of donor constructs for HDR-mediated insertion at sites of double-stranded DNA breaks induced by ZFNs or CRISPR/Cas9. In the presence of intranuclear single-stranded AAV donor templates, nuclease-mediated double-stranded DNA breaks are often repaired by the insertion of the AAV donor. In contrast, the BRCA2-dependant editing by AAV described in Smith et al.³ does not require the presence of prior double-stranded DNA breaks. Single-stranded AAV has long been known to mediate HR in the absence of DNA breaks.^17,18 In Smith et al.,³ we simply reported that Clade F AAVs mediate BRCA2-dependent HR at higher efficiencies. We showed editing by 14 different AAVs in primary cells in short-term cultures in vitro and by AAVHSC15 and AAV8 long-term in vivo. In each case, we found editing by Clade F AAVs to be more efficient.³

In closing, while there are numerous differences among Smith et al.,³ Roger et al.,¹ and Dudek and Porteus², we believe that their use of very low MOIs is the primary reason for their inability to observe augmented HR by Clade F AAVs. Moreover, their use of AAV vectors that were toxic, even at low doses, further impeded their ability to observe augmented HR by Clade F AAV.

Finally, in order to successfully observe HR-based editing by Clade F AAV, it is critical to (1) ensure the presence of intact and complete inverted terminal repeats; (2) use highly purified batches of AAV vectors; (3) use AAV vectors with single-stranded genomes; and (4) target cells that are not cycling too fast to be efficiently transduced by AAV. Augmented HR by Clade F AAVs requires high MOIs. This in turn requires the use of high-titered, high quality, non-toxic AAV vectors. The inability of Rogers et al.¹ and Dudek and Porteus² to transduce cells at MOI: 150,000 or higher in the absence of accompanying toxicity or to accurately replicate our studies clearly impacted their ability to make meaningful conclusions regarding the HR capacities of Clade F AAVs. Meanwhile, we remain committed to advancing this technology to better comprehend the underlying mechanisms and explore its potential for use in correcting genetic diseases.

Conflicts of Interest

S.C. is co-founder of and has an equity stake in Homology Medicines Inc.