Intracellular Trafficking of Adeno-Associated Virus (AAV) Vectors: Challenges and Future Directions

Abstract

In the last two decades, recombinant adeno-associated virus has emerged as the most popular gene therapy vector. Recently AAV gene therapy has been approved by the FDA for the treatment of two rare genetic disorders, namely the early childhood blindness disease Leber congenital amaurosis and spinal muscular atrophy (SMA). As is the case for the treatment of SMA, if the AAV vector must be administered systemically, very high vector doses are often required for therapeutic efficacy. But higher vector doses inevitably increase the risk of adverse events. The tragic death of three children in a clinical trial to treat X-linked myotubular myopathy with an AAV vector has thrown this limitation into sharp relief. Regardless of the precise cause(s) that led to the death of the two children, it is critical that we develop better AAV vectors to achieve therapeutic levels of expression with lower vector doses. To transduce successfully a target cell, AAV has to overcome both systemic as well as cellular roadblocks. In this review, we discuss some of the most prominent cellular roadblocks that AAV must get past to deliver successfully its therapeutic payload. We also highlight recent advancements in our knowledge of AAV biology that can potentially be harnessed to improve AAV vector performance and thereby make AAV gene therapy safer.

INTRODUCTION

After decades of research, the field of gene therapy is starting to realize its full potential for treating or even curing patients with genetic, as well as acquired diseases. In the US, the first ever in vivo gene therapy product for clinical use was approved in 2017 when the Food and Drug Administration (FDA) approved recombinant adeno-associated virus-based (AAV) LUXTURNA for correcting the genetic defect in Leber congenital amaurosis type 2 (LCA2)¹, a rare genetic form of early childhood blindness. In 2019, the FDA authorized the use of ZOLGENSMA to treat children less than 2 years of age who suffer from spinal muscular atrophy². SMA is a horrific disease, and the majority of children with the most common SMA form, type 1, die before reaching their second birthday or require mechanical ventilation³.

In addition to these approved drugs, as of December 2020 more than 200 studies using AAV vectors have been either completed or were ongoing (www.clinicaltrials.gov) for treating a diverse array of genetic diseases such as hemophilia A and B, cystic fibrosis and muscular dystrophies, among others.

AAV was originally discovered as a contaminant in adenovirus preparations more than 50 years ago⁴. Interestingly, for efficient replication AAV is dependent on a helpervirus such as adenovirus or herpes simplex virus⁵. Therefore, AAV became the founding member of the genus Dependoparvovirus in the Parvoviridae family of viruses. In the absence of a helpervirus, wild-type AAV2 (wtAAV2) can integrate preferentially into the AAVS1 locus on chromosome 19 and enter latency^6–9. Like other members of the Parvoviridae family, AAV is a non-enveloped, single-stranded DNA virus with an icosahedral capsid that has a diameter of ~22–26 nm comprised of a total of 60 capsid proteins. The wtAAV genome of ~4.7kb in length harbors only two genes, rep and cap, which are flanked by two T-shaped inverted terminal repeats (ITRs; Fig. 1). The rep gene encodes four Rep proteins (named Rep78, Rep68, Rep52 and Rep40 after their molecular weight). In the presence of a helpervirus, the Rep proteins play a central role in replication and genome encapsidation; whereas in its absence the large Rep proteins are important for the preferential insertion of the wtAAV2 genome into human chromosome 19^6–9. The cap gene expresses the three capsid protein subunits, VP1, VP2 and VP3, which have overlapping reading frames (Fig. 1). The ratio of VP1, VP2 and VP3 in the virion is approximately 1:1:10⁵. For many, though not all serotypes, the so-called assembly activating protein (AAP), which is expressed from an alternative reading frame within the cap gene, is important for efficient capsid assembly^10–12.

Figure 1: Wild-type AAV genome organization.

The wild-type AAV genome has only two genes, which are flanked by inverted terminal repeats (ITRs). Transcription from the p5 and p19 promoters, combined with alternative splicing, yields for transcripts that allow translation of four Rep proteins. In wild-type AAV, the rep proteins are responsible for genome replication and encapsidation as well as the preferential integration of the AAV genome into the human chromosome 19. The second gene, cap, encodes for the three capsid proteins VP1, VP2 and VP3, which have overlapping reading frames. The assembly activating protein (AAP) is translated from an alternative reading frame within the p40 transcripts of the cap gene. As the name indicates, AAP is essential for efficient capsid formation.

AAV vectors have a maximum packaging capacity of ∼5 kb¹³. However, one of their critical advantages is that the only cis elements required for AAV vector production are the ITRs of the viral genome, which account for a total of ∼300bp. Both genes of the wtAAV genome can be replaced with a transgene or shRNA cassette¹⁴. Since these vectors lack the rep and cap genes, they are completely replication deficient, even in the presence of a helpervirus. In addition, the lack of expression of viral proteins likely contributes to the limited immunogenicity of AAV vectors. Until recently, recombinant AAV vectors have had an unparalleled safety profile, and wtAAV has not been shown to be associated with any human disease¹⁵. Tragically, however, recently three children treated for X-linked myotubular myopathy (XLMTM) with an AAV8 vector delivering a functional copy of the MTM1 gene died (NCT03199469; now on clinical hold by the FDA). Two of the children presumably died because of hepatotoxicity and ultimately sepsis (¹⁶ and references therein), while the third patient is thought to have died as a result of gastrointestinal bleeding. At present, it is unknown what – if any – role the AAV capsid or the ITRs played in the tragic death of these children. Nonetheless, AAV has been shown previously to cause elevations of liver enzymes^{17, 18} and an anti-capsid CD8+ response¹⁹. Unavoidably, the high vector dosage used in this trial also meant the delivery of a large number of capsid proteins. It is, therefore, not unreasonable to assume that the AAV capsid proteins could have played a role in the catastrophic outcome. If proven true, this will demonstrate that ever increasing doses of available AAV vectors ultimately cannot be the answer to achieve high transduction efficiencies of the target cells. Undoubtedly, an improved understanding of AAV biology, both in vitro and in vivo, will be critical to develop approaches to allow AAV vectors to deliver their therapeutic payload more efficiently.

In this review, we will discuss the intracellular trafficking of AAV vectors - how the virions enter the cell and travel all the way to the nucleus for transgene expression. We will also highlight key cellular barriers that AAV faces along its path to transduction.

AAV Transduction

In 1982, the first AAV2 infectious clone (i.e., the entire AAV genome sequence cloned into a bacterial plasmid) was created²⁰, which formed the foundation of an enormous amount of research — yet large gaps in our understanding of biology of AAVs remain.

Studies with AAV vectors have shown that the tissue and cell-specific tropism vary widely depending not only on the AAV serotype but also the host species²¹. Furthermore, the route of administration can also have dramatic effects on the AAV tropism. For instance, Rabinowitz and colleagues demonstrated that AAV9 transduces the myocardium much more efficiently than AAV6 when the vector is injected via the tail vein²². In contrast, cardiomyocyte transduction by AAV6 is dramatically superior compared to AAV9 when the virus is injected into the left ventricle with cross-clamping of the aorta and pulmonary artery²³.

AAV2 vectors transduce cells fairly poorly, even in vitro^24–27, and the in vitro transduction efficiencies of other serotypes can be even worse. For instance, in the case of AAV9 vectors less than 1 in 100,000 vector genome containing particles are able to transduce successfully a cell²⁶. The generally poor vector performance resulted in the use of ever-increasing vector doses in clinical trials in recent years. These larger vector doses necessarily increase the risk of serious adverse events. Therefore, a deeper understanding of what the systemic and cellular barriers are to successful transduction by AAV is essential to facilitate the design of more efficient AAV vectors, which will allow the use of smaller vector doses.

When AAV reaches the cells, it attaches to receptors/co-receptors, followed by endocytosis into the cell. Inside the endocytic vesicle the virion then travels towards the trans-Golgi network (TGN). During trafficking, it goes through critical structural changes that expose a catalytic domain on the capsid^{28, 29}. The virion then escapes into the cytosol, although the exact organelle of translocation into the cytosol is not yet known. The capsid is then imported through the nuclear pore complex (NPC) into the nucleus in an importin β-dependent process³⁰ where the single-stranded genome is released from the capsid. Once released, the single-stranded DNA is converted to a double stranded DNA, and the genome can either persist as a circular episome, and linear or episomal concatemers. This is the final step of AAV transduction, which then allows the expression of the recombinant transgene. It is worth noting that, in rare cases, the recombinant AAV genome can also integrate into the host genome^31–33. In case of wtAAV, it has been suggested that “random” integration events can lead to hepatocellular carcinomas³⁴. However, this view has been challenged^{35, 36}. Maybe one of the most convincing arguments against AAV integration being a significant cause of hepatocellular carcinoma is that at least 40% of the population is seropositive for AAV2³⁷, but the prevalence of hepatocellular carcinoma is 10 per 100,000³⁶. Nonetheless, the fact that in mice “random” integration of recombinant AAV genomes occurs preferentially in genes is of potential concern³¹. Moreover, a recent, 10-year-long study in dogs aimed at the treatment of hemophilia A showed long-term expression of therapeutic levels of factor VIII. But, in 2 out of 9 dogs, it also documented clonal expansion of transduced hepatocytes with AAV genomes integrated into the host genome³². Even though there were no tumors observed, almost 44% of all integration events occurred near regions involved in cell growth³². So far, no tumors have been observed in the more than 200 clinical trials with AAV vectors (www.clinicaltrials.gov). However, as is true for the general population, patients treated with AAV will develop tumors. With our current knowledge it is impossible to estimate the risk whether these tumors might be related to the AAV gene therapy treatment or not. In the authors’ view, it will be important to determine if tumors in AAV gene therapy patients might have been caused by the AAV gene therapy treatment. Hence, a careful analysis for the presence of AAV vector sequences in tumors of patients that were treated with AAV vectors – especially patients that received high-dose peripheral injections – is prudent and warranted.

AAV Cell Surface Receptors

The receptor for AAV2, heparan sulfate proteoglycans (HSPG), was first identified by Samulski and colleagues³⁸. Summerford et al. demonstrated that the addition of heparin, a soluble form of HSPG, inhibited both AAV2 infection and binding of AAV2 to the cells. Similarly, AAV2 infection and binding of the virus to the cell surface were severely impaired in CHO cells deficient in HSPG biosynthesis³⁸.

Many viruses that bind to proteoglycans also have secondary attachment factors³⁹. Thus it is not surprising that for AAV2, a number of proteinaceous, secondary receptors have been reported to have roles in binding and transduction, such as fibroblast growth factor receptor 1 (FGFR1)⁴⁰, integrin αVβ5^{41, 42}, integrin α5β1⁴³, hepatocyte growth factor receptor⁴⁴ and laminin receptor⁴⁵. These observations prompted the idea that maybe binding to the primary receptor sets AAV in a favorable conformation that allows its binding to the co-receptors and eventual uptake. Structural studies have indeed shown that AAV2 binding to HSPG induces conformational changes in AAV2^{46, 47}, which is consistent with this model. Conversely, several studies later reported that the co-receptors fibroblast growth factor receptor 1, hepatocyte growth factor receptor and integrin are not mandatory for AAV2 entry^{48, 49}, and in 2016 an unbiased genetic screen found none of the above-mentioned proteinaceous co-receptors to be important for transduction⁵⁰. In sharp contrast, several proteins involved in HSPG synthesis were shown to play a role in AAV2 transduction⁵⁰.

On the other hand, the sequence of an AAV2-like variant isolated from children lacked the ability to bind to HSPG, most likely because in this variant two arginines that are critical for HSPG binding were absent⁵¹. This raises the possibility, which is supported by recent data by Cabanes-Dreus et al.⁵², that maybe AAV2 binding to HSPG is a tissue culture acquired trait. Moreover, AAV2 lacking the same two arginines in the HSPG binding region were still able to transduce cells in vivo, although the tissue tropism of this mutant AAV2 was different from AAV2⁵³. This implies that HSPG binding might not be an absolute requirement for AAV2 entry, but that it does influence AAV tropism. Similar observations have been made for AAV5, a sialic-acid binding serotype. If sialic acid binding of AAV5 is eliminated by mutating leucine 587 to threonine, the transduction of lung tissue was abolished, whereas the transduction of both salivary glands and muscle were increased⁵⁴.

To identify critical host factors for AAV transduction Jan Carette’s group performed a haploid genetic screen⁵⁰. The most prominent hit in this screen was a 150 kDa transmembrane protein, KIAA0319L, which the authors showed to be critical for both in vivo and in vitro transduction by AAV2 and all other major serotypes tested, except for AAV4 and AAVrh32.33^{50, 55}. Remarkably, this same 150 kDa protein was detected in a membrane lysate binding assay for AAV2 twenty years ago, but it was never identified and characterized ⁵⁶. Although KIAA0319L is essential for AAV2 transduction and is now currently referred to as AAV receptor (AAVR)⁵⁰, it is not required for binding or uptake of AAV2 to cells⁵⁵. AAVR is, in fact, largely located in the Golgi, presumably owing to its C-terminal domain⁵⁰. Taken together, these results suggest that AAVR is necessary for AAV2 transduction after endocytosis of the virus, possibly in AAV trafficking or escape of AAV into the cytosol (for additional information of AAVR’s role in transduction see below and in⁵⁷).

Mechanism of AAV2 Endocytosis

A number of studies have identified a variety of potential endocytic pathways for AAV2, and AAV2 appears to be able to be endocytosed very efficiently (~80% of all virions) via more than one pathway⁵⁸. Clathrin-mediated endocytosis of AAV2 was the first pathway of AAV that has been described^{59, 60}. Overexpression of a dominant negative form of a protein called dynamin (with an adenoviral vector), which is critical for clathrin-mediated endocytosis – among other endocytic mechanisms – decreased AAV2 uptake and transduction by up to 70%⁵⁹. However, the lack of inhibition of AAV2 endocytosis by chlorpromazine⁵⁸, or by the knockdown of the clathrin heavy chain⁶¹ suggest that this is at most a minor pathway of AAV2 uptake, at least in HeLa cells.

Sanlioglu and colleagues reported that AAV transduction is inhibited by adenovirus-mediated overexpression of a dominant negative mutant of Rac1, a known effector of macropinocytosis⁴², while overexpressing a constitutively active form of it augmented transduction⁶². These results also appear to contradict the involvement of clathrin-mediated endocytosis, since Rac1 is a well-known inhibitor of it^{63, 64}. It is worth noting, however, that adenovirus can affect many cellular pathways that are used by AAV. For instance, adenovirus E4 orf1 can bind to clathrin coated vesicles and influence Rac1 and PI3 kinase activity⁶⁵. Therefore, it is possible that overexpression of mutant proteins with recombinant adenoviruses might influence the cellular uptake of AAV, and that different results might be obtained if a protein is overexpressed by transfection.

In an attempt to reconcile the diverging results regarding the mechanism of AAV endocytosis, our lab investigated a number of endocytic pathways using both chemical inhibitors, short interfering RNA (siRNA) and overexpression (by transfection) of dominant negative mutant proteins⁵⁸. Interestingly, in our hands, neither inhibiting clathrin mediated endocytosis nor macropinocytosis significantly affected transduction of HeLa and HEK293T cells⁵⁸. Instead, the poorly understood clathrin-independent carriers and GPI-enriched endocytic compartment (CLIC/GEEC) endocytic pathway^{66, 67} was the most efficient endocytic route, leading to the most robust transduction. Membrane cholesterol, Arf1 and Cdc42 - critical components of CLIC/GEEC pathway - were each shown to play crucial role in formation of endocytic tubules/vesicles for AAV uptake. GRAF1 is the most specific mediator of CLIC/GEEC endocytosis⁶⁸. Confirming the importance of the CLIC/GEEC pathway for efficient AAV transduction, the overexpression of a dominant negative mutant of GRAF11 or knockdown of GRAF1 with siRNAs caused a strong inhibition of AAV2 uptake and transduction⁵⁸. Other groups that investigated AAV transduction with either a genome-wide siRNA screen⁴⁸ or with dominant negative mutants⁶⁹ corroborated the importance of the CLIC/GEEC pathway for AAV transduction of human aortic endothelial cells⁴⁸ and 293T⁶⁹ cells, respectively. Moreover, internalization of HSPG, the primary receptor of AAV2, also follows the CLIC/GEEC route of endocytosis⁷⁰. At the same time, while inhibition of dynamin by the drug dynasore lead to a modest (~40%) decrease in AAV2 endocytosis, transduction was unaffected⁵⁸. This suggests that, at least in HeLa cells, there are both dynamin dependent and dynamin-independent uptake routes, but that endocytosis via the CLIC/GEEC route results in the most efficient transduction⁵⁸.

Intracellular Retrograde Trafficking

Following endocytosis, the endocytic vesicles containing AAV can presumably follow different trafficking patterns, which can either lead to successful transduction or ultimately result in the degradation of the AAV capsid and the viral genome. Hours after cellular uptake, AAV2 can be seen accumulating in the perinuclear region⁶⁰. Using pharmacological intervention, genetic methods and microscopy, several groups demonstrated that for successful transduction of a cell AAV2 must be transported to the Golgi^71–75. For instance, treatment with brefeldin A^{74, 76} or Golgicide A⁷⁴ disrupts the Golgi apparatus structure and abolishes AAV2 transduction. In NIH3T3 cells AAV2 successfully binds to the cells and is efficiently endocytosed, but transduction is very low⁷⁷. Interestingly, it has been shown that intracellular trafficking of AAV2 in NIH3T3 cells differs from HeLa cells⁷⁷, and that in NIH3T3 cells the transport of AAV2 to the Golgi is severely impaired⁷⁴.

An initial dissection of AAV2 transport reported that virions, in a dose-dependent fashion, can (at least transiently) accumulate in rab7+ (late endosomes) or rab11+ (recycling endosomes) compartments⁷⁸. Partial knock-down of rab7 or rab11 reduced transduction by up to 65%, which lead the authors to conclude that AAV2 can transduce HeLa cells either via rab7+ or rab11+ endosomes – in a dose-dependent fashion. But curiously, overexpression of rab7 also reduced transduction. Furthermore, depending on the virus dose, overexpression of rab11 either had no effect or even increased transduction⁷⁸. In contrast, when our group nearly completely knocked-down rab7, rab9 or rab11 expression, or expressed dominant negative mutants of these effectors, we observed no inhibition of transduction⁷⁴.

Two canonical pathways to the TGN have been described in the literature: i) the late endosome to TGN pathway and ii) the recycling endosome to TGN pathway. Whereas the pathway through late endosomes depends on syntaxin 16, the pathway through the recycling endosomes depends on both syntaxin 16 and syntaxin 6. In line with the results of our rab7, rab9 and rab11 experiments, neither knockdown of syntaxin 6 nor syntaxin 16 inhibited AAV2 transduction⁷⁴.

Instead, we demonstrated that the transport of AAV2 to the TGN occurs via a non-canonical pathway that is dependent on syntaxin 5. In support of this, when syntaxin 5 expression was depleted by multiple siRNAs, AAV2 transduction was strongly decreased. Furthermore, when transport of syntaxin 5 from the ER to the Golgi – and presumably the plasma membrane – was inhibited by the potent drug Retro-2.1⁷⁹, AAV2 was dispersed throughout the cell in syntaxin 5 positive vesicular structures⁷⁴. Importantly, this syntaxin 5 dependent trafficking pathway has been shown to be the most efficient route for transduction by all major AAV serotypes (AAV1–9) in a number of cell lines and primary cells⁷⁴.

As mentioned earlier, loss of AAVR causes widespread loss of transduction in cells⁵⁰, although AAV can still bind to the cell surface⁵⁵. This implies a post-attachment role of AAVR and that for efficient transduction to take place AAVR itself needs to be endocytosed and travel to the TGN⁵⁵. In support of this model, experiments by Pillay et al.⁵⁰ demonstrated that the C-terminal tail region contains signals that play a role in AAVR endocytosis and trafficking. Deletion of this C-terminal tail leads to AAVR being stuck at the plasma membrane with a concomitant loss of AAV transduction. Interestingly, some low-level transduction of the AAVR-dependent serotypes can also take place in absence of AAVR⁵⁵. Furthermore, when AAVR is ectopically expressed in NIH3T3 cells, which are only poorly transducible by AAV2, transduction improves dramatically^{50, 55}.

Due to its C-terminal tail AAVR can be endocytosed (vide supra). After endocytosis, AAVR can, thanks to the same C-terminal tail, cycle between the TGN and the plasma membrane⁵⁰. Interestingly, an AAVR whose C-terminal tail was replaced with either the C-terminal tail of the LDL-receptor or the poliovirus receptor supported AAV2 transduction as well (20–40%, vs. wtAAVR)⁵⁰. This is rather surprising, since neither the LDL receptor nor poliovirus have been reported to travel from the plasma membrane to the Golgi. The cation-independent mannose 6-phosphate receptor (ciMPR), on the other hand, is known to travel to the TGN^{80, 81} and the endolysosomal system⁸². Accordingly, replacement of the C-terminal tail of AAVR with the cytoplasmic tail of the ciMPR targets AAVR to the Golgi and almost completely restores AAV2 transduction in AAVR knockout cells⁵⁰.

In the haploid genetic screen by Pillay et al.⁵⁰, GPR108, a Golgi-localized G-protein coupled receptor, was identified as the third most common hit. Recently, two independent genome-wide CRISPR screening studies validated its importance for transduction by all AAV serotypes, except AAV5^{73, 83}. Knocking out GPR108 drastically reduces AAV transduction both in vivo and in vitro⁸³, but it does not affect AAV binding and entry⁸³. Further experiments are required to determine what, if any, role GPR108 plays in the transport of AAV virions to the TGN or the escape of the virions into the cytosol (vide infra)⁷³. Notably, the effect of GPR108 knockout seems to be AAVR-independent, since AAV4 and AAVrh32.33 (both AAVR-independent serotypes) still depend on it for transduction⁸³. A potentially alternative explanation for the cellular requirement of GPR108 in AAV transduction is that it may play a role in intracellular immune modulation by activating NF-κB response and attenuating Toll-like Receptor (TLR) response⁸⁴.

Another Golgi protein that was identified in one of the genetic screens was TM9SF2⁷³. Although knocking out TM9SF2 heavily decreases transduction by all serotypes (even AAV5), its exact mechanism of action or its direct interaction with AAV have not been tested. TM9SF2 is well known for its role in HSPG biosynthesis⁸⁵, which is an important AAV attachment factor for several serotypes. On the other hand, AAV5 also depends on TM9SF2, despite its primary receptor being sialic acid⁸⁶. Together, this suggests that TM9SF2 affects steps necessary for successful transduction downstream of receptor binding. Notably, the knock-out of TM9SF2 has also been shown to dramatically reduce the toxicity of shiga toxin 1⁸⁷. Although TM9SF2 KO cells showed reduced levels of globotriaosylceramide (Gb3), the receptor for shiga toxin 1, the TM9SF2 KO cells also showed changes in endosomal trafficking of Shiga toxin⁸⁷. It is tempting to speculate that, since both AAV and Shiga toxin utilize syntaxin 5 mediated retrograde transport from the plasma membrane to the TGN, TM9SF2 is important for the retrograde transport of AAV to the Golgi.

Endosomal Processing and Escape into Cytosol

Trafficking to the Golgi is a critical step for AAV transduction, but AAV also must undergo conformational changes during endosomal trafficking^{60, 74–76}. For AAV2 and AAV8 it has been suggested that the virions are processed by the endosomal proteases cathepsin B and L⁸⁸. However, autoproteolytic processing of AAVs at low pH has also been reported to induce conformational changes⁸⁹. In most cell types^{60, 74, 76}, when acidification of endosomes is prevented by treating cells with bafilomycin A1, AAV2 loses its transducing ability, despite reaching the TGN⁷⁴. However, treatment of the highly permissive glioma cell line LN229 with bafilomycin A1 reduced AAV2 transduction only by ~1.5-fold⁷⁵. This implies that an acidic environment might not be a universal requirement for successful AAV2 transduction. Taken together, the data suggest that the structural changes are important not for TGN localization, but rather for critical downstream step(s).

Structural and electron cryomicroscopic evaluation have shown that the N-terminal regions of VP1 and VP2 capsid proteins stay hidden inside the AAV capsid and get exposed during endosomal trafficking²⁸. As is the case for autonomous parvoviruses⁹⁰, the unique region at the N-terminus of VP1 of AAVs contains a phospholipase A2 (PLA2) domain⁹¹ that aids in endosomal escape of AAV⁹². The importance of VP1 extrusion from the capsid interior for successful transduction was illustrated by Kleinschmidt and colleagues²⁹. When they microinjected intact (structurally unchanged) AAV directly into the cytosol - thus skipping the endosomal trafficking and Golgi localization steps - transduction was absent²⁹. Similarly, microinjection of A20 antibody against intact capsids, A1 against VP1 or A69 against the VP1/2 common region, either into the cytosol or the nucleus, drastically reduced transduction²⁹. These data demonstrate that for successful transduction: i) exposure of the VP1/2 N-terminal region is indispensable and ii) this reconfigured AAV needs to escape into the cytosol and travel into the nucleus intact. Interestingly, co-infection of cells with AAV with functional PLA2 activity can assist in the endosomal escape of PLA2-defective particles, i.e. there can be in trans compensation of PLA2 activity⁹².

As mentioned before, for most AAV serotypes GPR108 is a critical factor for transduction that colocalizes with AAV in the TGN⁸³. Of the serotypes tested by Vandenberghe and colleagues, only AAV5 could efficiently infect cells in a GPR108-independent fashion⁸³. Strikingly, when the AAV5 VP1 unique region (VP1u) was replaced with that of AAV2, the resulting chimeric AAV vector lost the ability to transduce GPR108 KO cells⁸³. Conversely, when VP1u of AAV2 is replaced with VP1u of AAV5, the resulting chimera could efficiently infect cells in the absence of GPR108⁸³. At present, the precise role of GPR108 in transduction by GPR108-dependent AAV serotypes is unknown. However, Dudek et al. showed that this dependency follows internalization and likely precedes nuclear import⁸³. Moreover, for GPR108-dependent serotypes, many of the internalized viral genomes are degraded⁸³. Since all AAVs, as well as AAVR and GPR108 travel through the TGN, it is possible that the interaction(s) of AAVR and/or GPR108 will influence post-Golgi trafficking, viral escape into the cytosol or capsid stability.

Nuclear Entry and Trafficking of Intact AAV

After escaping into the cytosol, AAV enters the nucleus. Although a few reports suggested that the AAV genome is released before or during nuclear entry⁹³, the de facto consensus in the field is that AAV enters the nucleus intact^{42, 60, 94–98}. This entry step has also been described as a rate-limiting factor⁶⁰ since only a very limited number of AAV could be detected inside the nucleus after 16–20 hours of helper-free infection⁹³.

A study based on super-resolution microscopy by Kelich et al.⁹⁸ also clearly showed that it enters through NPCs in a fairly quick (∼12ms) but inefficient process; only ~17% of AAV virions that dock to nuclear pores are imported into the nucleus⁹⁸. It has been reported that basic clusters 2 and 3 (located in VP1/2 common region) form a bipartite nuclear localization sequence that allows the binding of AAV^{29, 99} to importin β1 and the import into the nucleus through the nuclear pore³⁰. In an elegant study, Johnson et al. demonstrated that once AAV virions reach the nucleoplasm, they must also transit through the nucleolus to transduce cells efficiently⁹⁶. Although the exact role of the nucleolus in AAV transduction remains enigmatic, it has been shown that AAV virions interact with nucleophosmin¹⁰⁰ and nucleolin¹⁰¹. Surprisingly, despite the fact that siRNA-mediated knockdown of nucleophosmin and nucleolin expedited import and export of AAV in and out of the nucleolus, respectively, knockdown of either protein enhanced transduction⁹⁶. These data show that nucleolar localization is necessary but that it must be transient for transduction⁹⁶.

Genome Release and Second-Strand Synthesis

AAV uncoating and genome release in the nucleoplasm is poorly understood and has so far been largely studied in cell free systems and under non-physiological conditions. For instance, an analysis of genome release by atomic force microscopy showed that heat treatment of AAV8 and AAV9 virions leads to the release of viral genomes¹⁰². The release of viral DNA increases with increasing temperature and can occur either via the release of increasingly longer vector DNA segments or the release of the entire viral genome due to capsid rupture¹⁰². Interestingly, the release of viral DNA without complete capsid rupture has been described previously for the autonomous parvoviruses MVM¹⁰³ and B19¹⁰⁴.

An in vivo study by Thomas et al.⁹⁵ demonstrated that AAV2 uncoated slowly in murine hepatocytes, and intact particles could be found in the nucleus even 6 weeks after vector administration. The same genome, when packaged inside AAV8 or AAV6 capsids, was released faster from the viral capsid and transduced hepatocytes much more efficiently compared to AAV2⁹⁵. These authors also noticed that nuclear vector genome number and transduction efficiency do not correlate⁹⁵. In an elegant study, Rossi et al.¹⁰⁵ demonstrated that the insertion of a peptide into the AAV2 capsid that rendered the capsid less stable triggered i) an accelerated genome release, ii) increased transduction of dendritic cells in vitro and iii) in vivo transduction of myocytes and cells of the immune system causing a robust humoral and cellular immune response upon intramuscular injection¹⁰⁵. These studies suggest that rapid uncoating could be a rate-limiting step in transduction in a serotype and tissue-specific manner. On the other hand, a recent report by Gao and colleagues showed that a novel AAV variant, AAVv66, which has increased capsid stability compared to AAV2, transduces the CNS more efficiently¹⁰⁶. Since, DNA release from the viral capsid is a critical step in AAV transduction, more research to understand the mechanisms of genome release in cells and in vivo will be important to improve AAV vector performance.

After uncoating, single-stranded AAV DNA is converted into a double-stranded form, which has been described as a rate-limiting factor in transduction^{107, 108}. Several host proteins block this step. Phosphorylated FKBP52, for example, can bind to the 3’ end of the ITR and block complementary strand synthesis¹⁰⁹. Conversely, cellular DNA damage response proteins, consisting of the MRN complex (Mre11, Rad50 and Nbs1) can also inhibit AAV replication¹¹⁰. Moreover, a genome-wide siRNA screen identified proteins that when blocked can enhance AAV transduction up to 50-fold¹¹¹. Similarly, another siRNA screen identified the U2 snRNP spliceosome complex as a restriction factor against different serotypes¹¹². Together, these results convey the central importance of second strand synthesis in AAV transduction.

Other Intracellular Fates of AAV and the Effect of Post-Translational Modifications on AAV Transduction

So far, we largely discussed the steps of AAV trafficking that led to successful transduction. However, as illustrated by the sometimes astronomically high vector genome to transduction (vg/tu) ratios²⁴ the vast majority of AAVs fail to transduce successfully a cell. Unfortunately, our knowledge of these abortive routes is minimal. For instance, it was first reported in 2001 that AAV virions can be subjected to proteasome-mediated degradation⁷⁶. When cells were treated with the proteasome inhibitor MG-132 there was a dramatic increase in transduction. In fact, 24h post-infection with AAV2.Luc, luciferase expression increased 105 and 36-fold at a vg/cell ratio of 1 or 10, respectively. On the other hand, the number of AAV genomes increased only 4 and 7-fold, under the same conditions⁷⁶. Clearly, reduced proteasomal degradation of AAV capsids alone does not explain the dramatic increase in transduction. Nonetheless, several other inhibitors such as LLnL and bortezomib also can increase transduction¹¹³, and the effect is not limited to AAV2. In fact, these inhibitors had similar positive effects on transduction by serotypes as diverse as AAV2 and AAV5 (<60% capsid homology)¹¹⁴. Conversely, the proteasome inhibitor bortezomib did not increase cardiac transgene expression upon injection of an AAV9 vector¹¹⁵. Furthermore, modification of lysines on the surface of the AAV2 capsid to reduce potential ubiquitination resulted in a dramatic increase of liver transduction in mice; but analogous mutation on the AAV8 capsid had no effect¹¹⁶.

Mechanistically, the effects of proteasome inhibitors are not very clear. Many of the inhibitors have been reported to have pleiotropic effects on cells, such as inhibition of cellular cysteine and serine proteases¹¹⁷. As an example, when the highly specific second-generation inhibitor carfilzomib was used side-by-side with bortezomib, the in vivo effect was not as impressive as the in vitro data¹¹³. One putative explanation for the observed beneficial effects of proteasomal inhibitors could be that they induce subcellular stress¹¹⁸. Different forms of such stress (such as endoplasmic reticulum stress, inflammation and misfolded protein response) have been shown to induce AAV transduction⁷², at least in vitro.

Similar to proteasome inhibition, it has been shown that suppressing sumoylation significantly increases AAV2 transduction^{119, 120}. However, the prevention of sumoylation of VP2 of AAV2 by mutating the respective lysine residues in VP2 did not increase transduction¹¹⁹. This suggests that blocking sumoylation of cellular proteins, but not of AAV2 capsid proteins, restricts AAV2 transduction. Consistent with this model, global reduction of sumoylation by knocking down either the E1 ligase Sae1 or the E2 ligase Ubc9 increased AAV2 transduction¹²⁰. Like influenza A virus, AAV itself can trigger an increase in sumoylation of cellular proteins¹¹⁹. One potential sumoylation target that could diminish AAV2 transduction is DAXX¹¹⁹ a well-known inhibitor of viral infection¹²¹. Consistent with this hypothesis, DAXX knock-out cells are approximately 5-fold more efficiently transduced than their parent cells¹¹⁹. Furthermore, knockdown of the E2 ligase Ubc9 in HeLa or DAXX knockout HeLa cells increases transduction by approximately 5-fold and 2-fold, respectively¹¹⁹. These data support the model that DAXX can inhibit AAV2 transduction more efficiently when sumoylated.

Srivastava and colleagues hypothesized that AAV2 surface tyrosines are targets of the epidermal growth factor receptor protein tyrosine kinase, and that phosphorylation of these tyrosines could reduce AAV2 transduction¹²². They tested this hypothesis by mutating the key tyrosine residues on the AAV2 surface to phenylalanine, and they were able to demonstrate that these mutations dramatically increase AAV2 transduction in vitro¹²² as well as in vivo¹²³. They attributed this increase to a reduction in ubiquitinylation, but the differences observed in genome copy numbers in the cells suggest that alternative, unknown mechanisms might play a bigger role^{124, 125}. The same group later showed that additional mutations of threonines and serines can further increase the transduction efficiency of AAV2¹²⁶. Moreover, mutations of capsid residues that are targets for phosphorylation is not limited to AAV2; it also extends to additional serotypes (for examples see Büning and Srivastava¹²⁷). Clearly, post-translational modification of the AAV capsid, either during vector production or during viral entry, can dramatically affect transduction efficiencies.

FUTURE DIRECTIONS

In recent years, AAV vectors have emerged as the most powerful gene therapy tool for treating genetic diseases. Together, the different AAV serotypes and variants have broad tropism and are able to drive robust transgene expression over a long period in non- or slowly-dividing cells. However, in most cases when delivered systemically, high vector dosages are required for reaching clinical efficacy. Since, as Paracelsus said, “only the dose makes the poison”, the development of AAV vectors that deliver the therapeutic cargo more efficiently is paramount to reduce the risks of severe adverse events in future AAV gene therapy.

Over the past years we gained substantial insight into AAV trafficking as well as host restriction factors for AAV transduction. Much of this knowledge was gained through genetic CRISPR screens^{50, 83}, and doubtlessly more host restriction factors will be discovered in additional screens in the future. The study of the role of these factors in transduction will further our understanding of AAV biology overall. This newly gained knowledge opens up two potential avenues to improve AAV transduction and alter AAV tropism. One such avenue is the guided mutation of AAV capsids to i) alter their binding to their known receptors (see above^{53, 54}) or ii) lose (or gain) their dependence on specific host factors for transduction. That the latter is possible has already been shown by Dudek et al.⁸³. They demonstrated that AAV2 transduction of HUH7 cells and mouse embryonic fibroblasts (MEFs) is dependent on the presence of functional GPR108 whereas transduction by AAV5 is not⁸³. Strikingly, when VP1u of AAV2 was replaced with the VP1u of AAV5 this chimera could transduce HUH7 cells and MEFs independently of GPR108. Conversely, if VP1u of AAV5 was replaced with VP1u of AAV2 this chimera was now dependent on functional GPR108 for successful transduction⁸³. Hence, it should be possible to identify critical amino acid residues of VP1 that determine if an AAV variant is GPR108 dependent or independent. Targeted mutation of these amino acids should then allow to minimally alter the sequence of a specific AAV variant to gain or lose dependence on GPR108 without affecting other roles of VP1 that might influence transduction efficiency and/or tissue and cell-tropism. Of note, conferring any AAV variant to become dependent on a particular restriction factor might allow the de-targeting of an AAV variant from an off-target organ while retaining the efficiency of transduction of the target organ.

A second promising avenue to improve transduction is the identification of small molecules that modulate specific trafficking steps that are involved either in successful transduction or virion degradation. As discussed in more detail above, the inhibition of the proteasome can increase transduction of particular cell types and organs by specific serotypes. Interestingly, the increase in transduction is only partially due to a reduced degradation of AAV capsids. This demonstrates that it is possible to alter cellular processes to augment AAV transduction with small molecules. Therefore, the identification of other small molecules that affect specific processes, like AAV trafficking, offers a promising avenue to enhance AAV transduction.

Rab proteins are small GTPases of the Ras superfamily of G-proteins and are “master regulators” of membrane trafficking¹²⁸. Consequently, modulating their activity can be expected to alter the trafficking of AAV and potentially “re-direct” AAV virions from a dead-end to a productive trafficking route that leads to successful transduction (or vice versa). In this context, it is interesting to note that overexpression of a dominant negative mutant of rab11 increased transduction of HeLa cells by AAV2 by two-fold, although only one of three siRNAs against rab11 resulted in a similar increase⁷⁴. Nonetheless, modulation of rab protein activity has a great potential to increase AAV transduction. As is true for all small G-proteins, rab activity is regulated by rab-specific Guanine nucleotide Exchange Factors (GEFs), GTPase Activating Proteins (GAPs) and Guanine nucleotide Dissociation Inhibitors (GDIs). Inhibiting the function of one of the three co-factors of a particular rab protein can either activate or inhibit the activity of this rab protein¹²⁸. In fact, a large number of regulators of the proteins belonging to the Ras superfamily can be targeted with small molecules¹²⁹. Moreover, rab proteins interact with a plethora of rab effector proteins¹²⁸. Blocking or stabilizing these interactions with small molecules could possibly allow the modulation of AAV trafficking with near surgical precision.

Over the coming years, we will undoubtedly gain more insight into (cell type-specific) host restriction factors and the nature of productive and abortive trafficking routes in various cell types. We expect that the combination of this information with targeted capsid modification and the identification of small molecules that can modulate AAV trafficking pathways will further increase the already tremendous potential of AAV vectors for the treatment of inherited and acquired diseases.

Figure 2: AAV can follow multiple endocytic routes into the cell.

In an initial step AAV binds to its serotype-specific primary receptor (light blue) on the plasma membrane. It has been suggested that after this initial binding, AAV then interacts with one or more proteinaceous co-receptors (dark blue), although the importance of co-receptors remains controversial. Many serotypes bind to KIAA0319L, commonly referred to as AAV receptor [AAVR]; red/orange), although binding to the cell surface is AAVR independent, and AAVR binding to the virions might occur after endocytosis. At least three different mechanisms of endocytosis of AAV have been proposed in the literature. In addition to clathrin-mediated endocytosis and macropinocytosis, an as of yet poorly defined clathrin-independent pathway, called CLIC/GEEC, has been reported to allow AAV uptake and subsequent transduction. The mechanism of endocytosis is likely serotype and cell-type dependent, and it will be of great interest to determine what role the endocytic mechanism plays in AAV tropism in vivo.

Figure 3: Intracellular trafficking of AAV to the Golgi.

Following endocytosis, AAV must travel to the Golgi for successful transduction. In some AAV-refractory cells, such as NIH3T3 cells, the virus is endocytosed very efficiently, but the inability of AAV (serotype 2) to reach the Golgi is likely responsible for the poor transducibility of those cells. During intracellular trafficking, AAV has been shown to colocalize with a number of compartmental markers of the endosomal system. In early studies, it had been reported that AAV, depending on the vector dose, follows either the early endosome to late endosome to trans-Golgi network (TGN) or the early endosome to recycling endosome to TGN pathway. However, near complete knockdown, or overexpression of dominant negative mutants, of multiple mediators of these two pathways did not inhibit AAV transduction. Instead, it has been shown that the SNARE protein syntaxin 5 is critical for retrograde trafficking of vesicle-bound AAV to the TGN and subsequent transduction of all tested serotypes (AAV1–9) across multiple cell lines. Recently AAVR (red/orange) has been shown to be required for transduction for serotypes 1–3 and 5–9. Although AAVR appears to be dispensable for AAV2 binding to cells, overexpression of AAVR in AAV2-transduction refractory NIH3T3 cell can rescue the transducibility of NIH3T3 cells. These results are consistent with the fact that AAVR can also traffic to the Golgi from the plasma membrane. Although AAVR can bind to most AAVs, the mechanism by which AAVR facilitates the trafficking of AAV to the Golgi remains to be investigated. During its trafficking to the Golgi the AAV virions undergo significant conformational changes. The most profound change is the extrusion of the VP1 unique region (VP1u) from the capsid interior. In the vast majority of cells, the extrusion of VP1u is dependent on the lower pH in the endosomal system – as well as other factors. Dotted lines indicate routes that have been implicated in the literature, but the use of these pathways remains to be firmly established.

Figure 4: AAV must escape into the cytosol prior to nuclear entry.

Although Golgi localization is critical for AAV transduction, AAV must escape into the cytosol before it can be imported into the nucleus. During trafficking to the Golgi, AAV undergoes structural changes that expose the hidden N-terminal region of the capsid proteins VP1/2. The VP1 unique region also contains a phospholipase A2 (PLA2) domain, which is essential for viral escape into the cytosol. It should be noted that the exact organelle/location from where the virus escapes into the cytosol remains to be determined. The modified, but still intact AAV then enters the nucleus through the nuclear pore complex. Some, likely defective, AAV virions are ubiquitinated and degraded by the proteasome.

Figure 5: Steps following nuclear import that lead to transgene expression.

Remarkably, after reaching the nucleoplasm, AAV has to pass first through the nucleolus, and only after the AAV capsid is exported back into the nucleoplasm can the AAV genome be released from the virion. How the vector DNA is released from the AAV capsid is currently unknown; but two principal mechanisms have been proposed: 1) The vector DNA is released upon capsid disassembly or 2) Similar to MVM¹⁰³ and B19¹⁰⁴ the viral DNA is extruded through one of the pores at the five-fold symmetry axis of the AAV capsid, similar. Following its release from the virion, the ssDNA is then converted into double stranded DNA to allow transcription. The current hypothesis is that this occurs through synthesis of the complementary strand rather than via annealing of strands of opposite polarity.

Recombinant AAV vectors cannot efficiently integrate into host chromosome because they lack the rep gene. Instead, homologous recombination can result in circular episomes (and linear concatemers) that can persist in non-dividing cells for a long time and drive transgene expression.