Abstract
Recombinant adeno-associated viruses (AAV) are under intensive investigation in numerous clinical trials after they have emerged as a highly promising vector for human gene therapy. Best exemplifying their power and potential is the authorization of three gene therapy products based on wild-type AAV serotypes, comprising Glybera (AAV1), Luxturna (AAV2) and, most recently, Zolgensma (AAV9). Nonetheless, it has also become evident that the current AAV vector generation will require improvements in transduction potency, antibody evasion and cell/tissue specificity to allow the use of lower and safer vector doses. To this end, others and we devoted substantial previous research to the implementation and application of key technologies for engineering of next-generation viral capsids in a high-throughput ‘top-down’ or (semi-)rational ‘bottom-up’ approach. Here, we describe a set of recent complementary strategies to enhance features of AAV vectors that act on the level of the recombinant cargo. As examples that illustrate the innovative and synergistic concepts that have been reported lately, we highlight (i) novel synthetic enhancers/promoters that provide an unprecedented degree of AAV tissue specificity, (ii) pioneering genetic circuit designs that harness biological (microRNAs) or physical (light) triggers as regulators of AAV gene expression and (iii) new insights into the role of AAV DNA structures on vector genome stability, integrity and functionality. Combined with ongoing capsid engineering and selection efforts, these and other state-of-the-art innovations and investigations promise to accelerate the arrival of the next generation of AAV vectors and to solidify the unique role of this exciting virus in human gene therapy.
Introduction
The most recent authorization by the Food and Drug Administration of Zolgensma—a recombinant adeno-associated virus serotype 9 (AAV9) for treatment of spinal muscular atrophy (SMA) (1)—clearly showcases the enormous potential of this virus/vector for human gene therapy. It also exemplifies a major reason for the wide and continuous success of AAV as a therapeutic gene delivery vehicle, which is the availability of a diverse pool of capsid variants that can be harnessed to pseudotype and hence retarget recombinant vector genomes to tissues and cells of interest. Thus far, all three approved AAV gene therapy products were derived from naturally occurring AAV variants, including AAV1 (Glybera) and AAV2 (Luxturna), but overwhelming data from the past two decades additionally illustrate the power of other synthetic AAV capsid variants that were created and selected by directed molecular evolution or rational design. As the underlying principles and technologies have been reviewed extensively in the past by others and us, including in recent comprehensive articles (2–6), it may suffice to point out a few of the most exciting capsids that were reported and that exemplify the vast potential of AAV capsid engineering. These include AAV-PHP.B and related peptide display variants of AAV9 that effectively cross the blood-brain barrier in C57BL/6 mice (7), Anc80L65 that was derived by ancestral reconstruction and that is highly efficient in, among other tissues, livers of nonhuman primates (8), as well as AAV-DJ, a chimera obtained by shuffling of eight different AAV wild-types that is very potent in numerous cell types and concurrently more immunoevasive than the AAV2 prototype (one of AAV-DJ’s parents) (9).
Remarkably, while the field of AAV capsid engineering is thriving and the number of groups developing associated technologies is rapidly expanding, less attention has been devoted thus far to improvements of the second essential vector component, i.e. the recombinant genome encoding a therapeutic DNA of interest (e.g. an expressible cDNA or a repair template for gene editing strategies) flanked by AAV inverted terminal repeats (ITRs, replication and packaging signals) (Fig. 1). Perhaps the most notable is the invention of ‘double-stranded’ or self-complementary (sc) AAV vector genomes in which one of the two ITRs is truncated or mutated, causing an arrest during vector replication at an intermediate stage comprising two inverted copies of the vector-encoded transgene (10). As a result of this duplication, the ensuing AAV particle packages a palindromic, single-stranded (ss) DNA genome that can rapidly self-anneal in the transduced cell to form a double-stranded, expression-competent DNA, explaining why scAAV vectors tend to express faster and more potently (at least early after transduction) than their conventional ssAAV counterparts.
Components of a typical AAV expression vector. Expression of the transgene is driven by an upstream promoter, either constitutive or tissue/cell-specific, natural or synthetic and/or inducible. The polyadenylation signal (pA) marks the end of the transcriptional cassette. The three elements that are boxed in orange constitute the basis of any standard expression cassette. As discussed in the text, several optional cis-regulatory elements can be included to improve the efficiency and/or specificity of transgene expression. These include enhancers (Enh), cis-regulatory modules (CRM), introns (I) or post-transcriptional elements such as WPRE. Moreover, binding sites for miRNAs can be incorporated into the 3′ end to fine-tune vector expression or to turn it on or off in a tissue/cell-specific fashion (see also Fig. 2). The entire cassette is flanked by ITRs that serve as signals/primers for vector DNA replication and encapsidation. Mutation of the so-called terminal resolution site in one of the ITRs (indicated with the asterisk) creates a scAAV genotype that is characterized by encapsidation of two inverted copies of the entire recombinant DNA between the ITRs.
A second notable improvement to the design of AAV vector genomes was introduced to alleviate another bottleneck in the use of recombinant AAVs for gene transfer, which is their limited cargo size of ~ 4.7 kb (kilobases) of foreign DNA. While this suffices for e.g. expression of small RNAs such as shRNAs or gRNAs [triggers of RNA interference (RNAi) or clustered regularly interspaced short palindromic repeats (CRISPR), respectively], size quickly becomes limiting for larger transgenes such as the Cas9 nuclease from Streptococcus pyogenes, especially when they are driven by complex and longer promoters (11). Therefore, several groups invented and validated strategies to split up a longer DNA into two parts, each of which is then encapsidated into, and delivered by, a separate AAV particle. As these strategies have also been the subject of previous comprehensive reviews (12,13), we will only mention a few notable examples of split vector designs that have been implemented, such as trans-splicing constructs that reconstitute a full-length DNA in the transduced cell via homologies in the ITRs or introns (13,14) or the separation of a larger protein into two peptides that will reassemble in the target cell via intein-mediated protein splicing (15,16).
Here, we will discuss a series of additional improvements that also act on the level of the vector DNA, not the capsid, with a particular focus on most recently reported and most innovative strategies, from (i) engineering of synthetic enhancers/promoters and (ii) the use of endogenously or exogenously controlled genetic circuits that regulate AAV gene expression to (iii) enhancement of various AAV vector properties through the dissection and optimization of structural DNA elements. We note that in the first part on promoter engineering, we will also briefly recapitulate a selection of previous studies that provided the rationale for the latest generation of transcriptionally controlled AAV vectors, and we apologize in advance to the many colleagues whose pivotal work we had to omit for space reasons.
The road to next-generation, synthetic AAV enhancers/promoters
A key element in typical AAV expression vectors are promoters, i.e. cis-acting elements that enable and control transcription of the gene(s) they are associated with. They consist of a core/proximal promoter that is located ~ 100–1000 bp (base pairs) upstream of the regulated gene and carries clusters of binding sites (BSs) for transcription factors (TFs). These either stimulate or repress transcription initiation at the transcription start site, mediated by the binding of RNA polymerase II in the core promoter (17). The recruitment of widely expressed, spatially restricted and/or cell-type-specific TFs eventually defines the activity of the promoters. Thus far, exogenous (often derived from viruses and constitutively active) or endogenous (ubiquitous or tissue/cell-specific) promoters have been commonly used in AAV gene therapy vectors (18). Examples for constitutive strong promoter that were used to achieve robust, rapid and long-term transgene expression in most cell types include the spleen focus-forming virus (SFFV), the human polypeptide chain elongation factor (EF1α), the phosphoglycerate kinase (PGK), the ubiquitin C (UbiC), the cytomegalovirus (CMV) or the chicken beta-actin (CBA) promoter (19–22). Of note, CMV or CBA promoters were used in recent clinical trials for the treatment of SMA, Duchenne muscular dystrophy (DMD) or Leber’s congenital amaurosis (LCA) (23–26). In addition, they also form the basis of AAV gene therapy products, including the use of the CMV promoter to drive the lipoprotein lipase gene in Glybera (27) or of the CBA promoter to transcribe the SMN transgene in Zolgensma (1).
However, these promoters that induce strong and constitutive transgene expression also present some disadvantages that need to be overcome in next-generation AAV vectors. Firstly, they are more prone to inactivation than tissue-specific promoters (see below), exemplified by findings that extensive promoter methylation can lead to drastic transcriptional silencing (28–30). Further limitations are the risk of toxicity from (i) transgene overexpression (31–33), (ii) non-targeted transgene expression (34–37) or (iii) immune reactions partly due to inadvertent transgene expression in antigen-presenting cells (APCs) (38–42). Moreover, a notable recent study (37) highlighted a strong correlation between cis-regulatory elements and ocular toxicity following sub-retinal AAV vector injections in mice. Acute toxicity in photoreceptors and the retinal pigment epithelium was observed with all three constitutive promoters tested (CMV, UbiC and CAG) but not with photoreceptor-specific promoters (37).
The latter is an example of tissue-specific promoters that are endogenous and derived from genes solely expressed in certain cell types. Despite the fact that they typically yield lower expression levels than their ubiquitous counterparts (43–45), these promoters provide the seminal possibility of spatially controlling transgene expression (46,47). In fact, the first two clinical trials for hemophilia B harnessed factor IX (FIX) expression from liver-specific promoters, i.e. ApoE/hAAT (48) and LP1 (49,50). Notably, it was found that targeting transgene expression to hepatocytes can induce antigen-specific tolerance and thus confer an additional safety level (51–53). Since then, more and more preclinical and clinical studies now use tissue-specific promoters for transgene expression from AAV vectors (54–58).
Concurrently, different strategies have been devised to circumvent the inherent weakness of many tissue-specific promoters. In particular, the addition of a strong distal enhancer element upstream of the tissue-specific core promoter allows for strong transgene expression while maintaining its selectivity (47,48,59–61). For example, an in vivo study with AAV2 showed that the use of the CMV enhancer upstream of a cardiac muscle promoter resulted in a 50-fold increase of transgene expression in the heart compared to the CMV promoter alone (60). Another mouse study evaluated combinations of the albumin promoter with different enhancers and untranslated regions for liver-specific expression. This led to identification of an optimal design (albumin promoter juxtaposed with α-fetoprotein MERII enhancer) that drove expression of a secreted alkaline phosphatase reporter in the liver not only specifically but also as efficiently as the strong CMV promoter (61). Finally, in a most recent study, McDougald et al. (62) used a ‘dead’ CRISPR/Cas9 enzyme fused to a trans-activating domain to boost the activity of two photoreceptor-specific promoters, namely, the human rhodopsin kinase (GRK1) and cone arrestin (CAR) promoters. This latest work from the Bennett lab is particularly notable owing to its flexibility and thus broad utility, due to the ease and speed with which g (uide) RNAs can likely be identified de novo or from the literature that allow repurposing to many alternative tissue-specific promoters.
Besides enhancers, other cis-regulatory elements can be added to the expression cassette to achieve fine regulation and boost the efficiency of transgene expression (18,47). As one example, the woodchuck hepatitis virus post-transcriptional regulatory element (WPRE) has recently been shown to enhance AAV transduction in mouse liver, brain and muscles (63), as well as in mouse and human retinas (64). Another option are introns that are usually incorporated between the promoter and the transgene-coding sequence and that can also improve gene expression. It was suggested that the presence of a chimeric intron fosters the incorporation of the mRNA into the spliceosome, which, in turn, increases nuclear stability or enhances nuclear mRNA transport (65). One representative candidate is the minute virus of mice (MVM) intron, which is a particularly strong element that was reported to enhance FIX expression in the liver of mice by about 80-fold (66).
In theory, the combination of one or several of these cis-regulatory elements with a given promoter could yield very potent expression cassettes, but one needs to consider a major drawback of AAV vectors, which is their limited packaging capacity of 4.7 kb for ssAAV and 2.5 kb for scAAV vectors. Thus, many groups developed shorter promoters by deleting non-essential parts, while ideally retaining a high level of gene expression and tissue specificity. One representative pair is the 2.2 kb human glial fibrillary acidic protein (GFAP) promoter expressed in astrocytes throughout the central nervous system and its shorter version gfaABC1D with only 681 bp. The latter exhibits the same expression pattern as the full-length GFAP promoter and is even 2-fold more active (67). Different muscle-specific promoters were also compacted, such as the myosin light chain-2v (MLC) (60,68) or the muscle creatine kinase (MCK) (69) promoters. As the large size of the genuine MCK promoter of 6.5 kb makes it incompatible with AAV vectors, Wang et al. (69) created a shorter version by ligating a modified MCK enhancer to a highly truncated miniature MCK basal promoter. The resulting promoters [dMCK (509 bp), tMCK (720 bp)] showed high muscle specificity and better activity than the CMV promoter. A last example is the ubiquitous 1.6 kb hybrid promoter CAG (also called CAGGS) composed of the CMV immediate-early enhancer, the CBA promoter and the CBA intron/exon 1 (70). Early on, a 800 bp miniature version of this CMV enhancer/CBA promoter was made by replacing the CBA 5’UTR with a smaller simian virus 40 (SV40) intron (71). Since this version failed to reach the expression levels of the full-length CAGGS promoter, especially in the brain and motoneurons (72), Gray et al. (72) created a new 800 bp version, called CBh for CBA hybrid intron, by replacing the SV40 intron in the CBA promoter with a hybrid intron composed of a 5′ donor splice site from the CBA 5’ UTR and a 3′ acceptor splice site from the MVM intron.
A related major challenge in AAV promoter optimization is to establish which elements in the promoter, the distal enhancer and/or in the untranslated regions actually govern the strength and/or the specificity of expression. Luckily, over the last few years, high-throughput methods have emerged that now help to characterize mammalian promoters and to identify new enhancer activities (73–75). Concurrently, novel strategies have been implemented that capitalize on large collections of cis-acting elements and on the modular nature of promoters. Indeed, many of these elements retain their activities when taken out of their native promoter context and used as building blocks in synthetic promoters. Consequently, to date, a number of strong, synthetic and tissue-specific promoters, whose use is now well established in the AAV gene therapy field, have already been designed this way. For example, the liver-specific promoter 1 (LP1) combines the hepatic control region of the human apolipoprotein E/C-I gene locus, the hAAT promoter and the MVM intron (49,50,76). Similarly, the SPc5-12 promoter (54,77) consists of a specific arrangement of muscle-specific control elements (BSs for MEF-1/−2, SRE, TEF-1) and is superior to the CMV promoter (46). As a third example, the CK8 promoter (43,78) was identified in a screening of regulatory cassettes based on the enhancer and promoter of the MCK gene (79). Besides, multiple companies have specialized in the design of synthetic promoters based on proprietary algorithms.
Despite the promise of synthetic promoters, a challenge that has remained until very recently is that their design relies on the in vitro characteristics of regulatory elements in cell lines (80), which is usually not predictive of their in vivo performance (38,41). In addition, these approaches at promoter design often do not take into account the importance of including conserved regulatory motifs. This was now addressed by the latest computational methods, which explicitly consider evolutionarily conserved transcription factor binding sites (TFBSs) that are related to strong tissue-specific expression and which thereby represent a powerful novel strategy for generation of cell-type-specific promoters. In particular, this approach was used to identify short sequence units called cis-regulatory modules (CRMs) consisting of clusters of TFBSs (81). Already in 2003, Sharan et al. (82) were the first to design a tool named CRÈME, which relies on a database of putative TFBSs annotated across the whole genome, for identifying and visualizing CRMs in promoter regions. Other computational methods have since been described such as LogicMotif (83), POCO (84) or differential distance matrix–multidimensional scaling (DDM-MDS) (85). Of particular note is work by the Chuah and VandenDriessche group who has specialized on the identification and validation of CRMs in an AAV vector context that are specific for hepatocytes, heart and skeletal muscles (77,86–88). To devise tissue-specific and evolutionarily conserved CRMs, these authors implemented and used a clever multistep, genome-wide data-mining strategy. First, they identified highly expressed tissue-specific genes based on microarray expression data and then extracted the corresponding promoter sequences. Next, using a DDM-MDS approach (85), they identified CRMs and their corresponding TFBSs. Finally, they applied a filtration step for evolutionarily conserved TFBS clusters and open chromatin structures as defined in the ENCODE database (89). In their latest and, in our opinion, most impressive example reported to date (77), the Chuah and VandenDriessche lab successfully harnessed this pipeline to identify a potent skeletal muscle-specific CRM (Sk-CRM4, 435 bp), which, in combination with the muscle-specific desmin promoter, boosted luciferase expression up to 400-fold or up to 96-fold in skeletal muscles of neonatal or adult mice, respectively, as compared to conventional promoters. Importantly, the augmentation of transcriptional activity was specific for muscle tissues and allowed to restore the DMD phenotype in mdx mice using micro-dystrophin- and follistatin-encoding AAV9 vectors (77).
Finally, it should be noted that fine-tuning of AAV gene expression can also be achieved by the use of inducible synthetic promoters, which add a layer of exogenous control as they can be induced under specific conditions and with a variety of ectopic triggers. From a clinical perspective, particularly attractive is the ability to not only modulate transgene expression but to switch it on or off at will, as this provides a seminal safety feature. Among the currently available inducer/repressor systems permitting control over gene expression, the tetracycline (Tet)-dependent system is by far the most advanced and most widely used in vitro and in vivo (90). The first gene therapy vectors designed in a Tet-inducible manner were lentiviruses (90–92), before this system was later also successfully applied to AAV vectors (93–97). Yet, a major concern with its use in gene therapy remains the risk of immune reactions against the bacterial components. Indeed, many studies showed that the Tet-On reverse tetracycline-controlled transactivator (rtTA)-based system was rejected by the host immune system, concurrent with a rapid loss of inducible gene expression from AAV vectors (98–100). Notably, though, a most recent study reported that nonhuman primates injected with an AAV1-CMV (tetO)-rtTA/Epo vector exhibited persisting transgene expression following periodic doxycycline administrations over 5 years, without any humoral or cellular responses against the rtTA protein (101), which illustrates the great potential of inducible promoters in an AAV context in mammals and potentially humans.
Furthermore, next to the Tet operon, a large variety of inducible promoter systems have already been evaluated in the AAV context, which, due to space constraints, we can unfortunately not discuss in detail. Still, we wish to point out some of the most remarkable and most recently implemented systems, including intriguing work by Serrano-Mendioroz et al. (102) who incorporated regulatory enhancer elements into AAV vectors that respond to stimuli related to the pathology of acute intermittent porphyria (AIP), such as estrogens or starvation, and that hence tightly couple therapeutic transgene expression to the time when it is most needed in AIP patients. Other examples comprise the use of small molecules to induce dimerization of promoter-activating TF domains (103), of histone deacetylase inhibitors to activate the stress-inducible Grp78 promoter specifically in cancer cells (104) or of systemically administered kainic acid to stimulate a NF-ΚB-responsive promoter in various AAV-transduced regions of the murine brain (105). Moreover, in the next paragraph, we will provide examples for the regulation of AAV expression including CRISPR with a specific physical trigger, namely, blue light.
Last but not least, latest reports are noteworthy that have already begun to illustrate the usefulness of inducible or regulated genetic circuits to control the expression and activity of the CRISPR gene-editing system in an AAV vector context. In one very recent example, Kumar et al. (106) have demonstrated the ability to concomitantly, temporally and spatially control gRNA and Cas9 expression from AAV using a smart combination of doxycycline and Cre recombinase in vitro or in vivo. A similar proof-of-concept has been reported before by Yang et al. (107), who likewise exemplified the possibility to juxtapose the CRISPR and Cre/loxP systems in AAV as a novel, site-specific and spatially and temporally regulated toolkit for gene manipulation in mammalian cells. A third and final example is recent work by Li et al. (108) who proposed a dual ‘self-deleting’ AAV vector system in which one vector encodes Cas9 and a gRNA against a specific target, while the second vector delivers an anti-Cas9 gRNA. Hence, if both vectors are present in the same cell, this particular genetic circuit will lead to inactivation of Cas9 expression from the first vector, ideally after the intended on-target has been cleaved and edited. Of note, we have recently established a different version of AAV-CRISPR vectors that are truly self-inactivating, as they encode the anti-Cas9 gRNA on the same construct that harbors the Cas9 itself (Fakhiri and Grimm, in preparation).
Regulation of AAV transgene expression through microRNAs or light
As an alternative or complementary strategy to the use of specific or inducible promoters, cell selectivity of AAV transgene expression can also be honed by exploiting the RNAi machinery and the tissue- or cell-specific expression pattern of microRNAs (miRNAs) (Fig. 2). These are conserved post-transcriptional regulators of gene expression that act by destabilizing sequence-complementary mRNA targets, with one miRNA typically binding to hundreds of different targets. Thus, miRNAs collectively regulate the majority of all protein-coding genes and affect nearly all cellular pathways (development, hematopoietic cell differentiation, apoptosis, cell proliferation, organ development, etc.) (109). As the biogenesis and function of miRNAs has been extensively described before (110–112), it should suffice to point out that mammalian endogenous miRNA guides are typically complementary with their target along the seed sequence (nucleotides 2–8) but lack perfect complementarity along the rest of the guide sequence. As a result, the miRISC complex remains engaged with the target strands and induces partial inhibition of gene expression by translational repression and/or target deadenylation/destabilization (109,113,114). Although rarely occurring in mammalian cells, a higher degree of complementarity between the miRNA and its target can induce Ago2-mediated cleavage of the target transcript (111,115). Thus, the magnitude of repression depends on the complementarity between a miRNA and its target. For gene therapy applications, a rapid and efficient degradation of mRNA targets would be ideal, since it guarantees the specific down-regulation of the transgene expression without the risk of sequestering the cellular RNAi machinery, as observed with artificial si/shRNAs (116).
miRNA-based strategies to control AAV gene expression. Shown on top is a standard AAV vector genome (see also Fig. 1) for comparison. The presence of binding sites (BSs) perfectly complementary to a cell-type-specific miRNA in the 3’UTR of the transgene suppresses its expression by mRNA destabilization in cells/tissues expressing the cognate miRNA (switch-OFF). Variation of the degree of complementarity between miRNA and BS permits to gradually fine-regulate the mRNA level and hence transgene expression levels (FINE). Alternatively, miRNAs can be used to regulate the expression of a repressor protein that interferes with transcription of the transgene, by binding operator (Op) sequences located in the promoter (prom_Op) and by thus turning a miRNA into an activator of AAV vector expression (switch-ON). As indicated at the bottom, both strategies (OFF/ON) can also be juxtaposed in a single vector genome (and further combined with the FINE strategy, not shown), to maximize the specificity of its expression. pA, polyadenylation signal.
Luckily, many miRNAs have been identified over the last few years including several that are exclusively expressed in particular tissues (117–119), such as miR-122 that is highly abundant and active in the liver only (120) or miR-1 that is detected solely in muscle (117). Based on these findings, Brown et al. (121) were the first to exploit these tissue/cell-specific miRNA signatures as a means to regulate transgene expression in the context of a lentivirus vector. They demonstrated that the incorporation of four binding sites (miRNA-BS) for miR-142-3p (enriched in hematopoietic cells) in the 3’UTR of a gfp transgene specifically decreased expression in splenocytes but not hepatocytes (non-hematopoietic cells). In a follow-up study, the same group showed that de-targeting gene expression from APCs using miR-142-3p could prevent immune responses against a FIX transgene in FIX-deficient mice (122). Subsequently, several labs have successfully applied this concept in AAV vectors of serotype 1 as well and confirmed its ability to reduce transgene-directed immunity in mice following intramuscular vector administration and using different transgenes, including the highly immunogenic ovalbumin (OVA) (123–125). In particular, most recently (125), the Gao lab has succeeded at elucidating the mechanisms underlying this effect and identified, among others, a blunting of cytotoxic CD8+ T-cell responses and prevention of anti-OVA antibody formation as a result of AAV detargeting from APCs. Impressively, Xiao et al. (125) also found that miR-142-mediated APC detargeting even enabled redosing of the same mice with a different AAV serotype (AAV8) expressing the same transgene product, highlighting the power of this approach not only for improving the efficacy of AAV gene therapies but also for permitting multiple vector administrations.
In parallel, this ‘switch-OFF’ system whereby perfectly complementary BSs for tissue-specific miRNAs are tagged on to a transgene’s 3’UTR was successfully expanded to other miRNAs, applications and vectors, including very recent work (126–136). For example, in 2018, an AAV with BSs for miR-122 and miR-199a was reported that specifically mediates suicide gene expression in hepatocellular carcinoma (HCC) as these two miRNAs are down-regulated in HCC compared to healthy hepatocytes (127). Concurrently, we have reported an AAV vector for CRISPR/Cas9 expression in which Cas9 was tagged with BSs for miR-122, reducing its levels in hepatocytes (useful in cases where these are considered as off-targets for the vector) (11). Finally, together with the Niopek lab, we have most recently harnessed this ‘switch-OFF’ strategy to create a new generation of AAV-CRISPR/Cas9 vectors in which we specifically down-regulated an anti-CRISPR (Acr) protein in cells expressing miR-122 or miR-1, thus turning Cas9 activity on in these cells (131).
In another new modification of this system, miRNAs were harnessed to achieve precise regulation of transgene expression by using imperfect or bulged BSs and thus mimicking the endogenous mRNA destabilization mechanisms (132). To this end, high-throughput sequencing was used to study thousands of synthetic miRNA-BS (called miRNA response elements or MREs) with varying complementarity to an endogenous miRNA (for proof-of-concept, miR-17 was used). This yielded a library of miRNA-based fine tuners that enabled precise modulation of expression levels of endogenous genes, such as PD-1, a T-cell co-inhibitory receptor and a target for cancer immunotherapy, or BRCA1, a tumor suppressor involved in breast and ovarian cancers (132).
Generally, when designing such switch-OFF vector constructs, several aspects need to be considered, including possible cross-targeting by different miRNAs (129,132). For example, in the context of cardiac gene transfer, miR-206-mediated skeletal muscle-specific silencing failed due to the expression of miR-1, another member of the same miRNA family, in heart tissues and its binding to the miR-206 BS in the AAV vector cassette (129). To circumvent this, the authors introduced single-nucleotide substitutions into the miR-206-BS and searched for those that were resistant to miR-1 but remained sensitive to miR-206. Thus, one should avoid miRNAs that belong to larger families, as the same construct may perfectly bind the targeted miRNA but act as a bulged BS for another miRNA sharing the seed region.
Moreover, another important factor is the number of miRNA-BS that is required for efficient down-regulation. To date, the majority of studies suggest the use of four miRNA-BS as the strength of inhibition seems to increase with their number (128,132,136). However, it was also shown that sequestration of cellular miRNAs can occur with higher numbers of BSs (130). Besides, the number of BS needed for robust inhibition directly depends on the endogenous miRNA expression level in the cells or tissues of interest (38).
Theoretically, given the small size, it would be possible to incorporate multiple miR-BSs into a vector cassette to concurrently detarget several tissues. However, a problem may be their accessibility that could be impaired by the complex secondary structure of longer mRNAs. In addition, not all tissues or cell types exhibit a suitable miRNA signature that would be compatible with such a detargeting strategy. To circumvent this, a ‘switch-ON’ strategy was invented. Here, miRNAs are used to down-regulate the expression of bacterial repressor proteins that bind to operator sequences in the promoter, thus selectively activating transgene expression in the targeted tissue expressing the miRNA. Consequently, this system utilizes a single endogenous miRNA to indirectly drive the expression of the transgene in the targeted tissue, while expression of the latter is fully repressed in other tissues (94,95,137,138). The first to describe such a regulatory system were Pichard et al. in 2012 (95), using the TetR/tetO system. Proof-of-concept was obtained by cloning four perfect BSs downstream of the TetR coding region for miR-122, miR-142-3p or miR-133 and by showing up-regulation of TetR-regulated green fluorescent protein (GFP) expression specifically in liver-, macrophage- or muscle-derived cells, respectively. Using AAV vectors, the authors also demonstrated the efficiency of their expression cassette containing miR-133-BS in the muscle in vivo (95). Another, more recent study also investigated the switch-ON system in vivo but used the cumate-inducible repressor protein CymR (139) instead of TetR. In mice administered with plasmids containing four perfect miR-122-BS in the 3′UTR of CymR, a 8.5-fold increase of luciferase expression was measured in the liver. Moreover, a 37-, 33- and 17-fold induction of luciferase expression was observed in muscle, using four BSs for miR-1, miR-133 and miR-206, respectively (138).
Lastly, we note yet another trigger than can be exploited to regulate AAV vector gene expression in transduced cells, i.e. light. In one exemplary study by Redchuk et al. (140), an optogenic circuit was derived from a bacterial phytochrome (BphP1-QPAS1 system) and expressed from an AAV9 vector in primary rat cortical neurons, where it mediated a 2.5-fold activation of a sensitive reporter cassette after near-infrared light illumination. In addition, the Niopek lab and we have recently exploited the LOV2 domain from Avena sativa to generate a blue light-sensitive anti-CRISPR protein and shown that the resulting, so-called CASANOVA system permits tightly controlled regulation of Cas9 activity (141). This is worth mentioning here in view of the aforementioned miRNA-regulated AAV/CRISPR system that we have also recently established (131), as the two are complementary and could readily be combined in the future to maximize spatio-temporal control over AAV/CRISPR vector activity.
Identification and elimination of adverse structural elements
In this last chapter, two recent studies from the Gao lab should briefly be highlighted that used single-molecule, real-time (SMRT) sequencing technology to profile genome populations in AAV vector stocks and that identified unexpected truncations and human-vector chimeras whose characterization should inform next-generation AAV vector manufacturing (142,143).
In the first of these studies, Xie et al. (143) investigated whether DNA sequences encoding short hairpin (sh) RNAs (artificial RNAi triggers) would interfere with scAAV vector replication, owing to their inherent secondary structure that mimicks an AAV ITR (i.e. the genuine replication and packaging signal of AAV). Indeed, in particular when the shRNA sequence was in proximity to the wild-type ITR, yields of the scAAV vector were reduced. Moreover, for all vectors containing an shRNA element, in vivo infectivity in the mouse liver as measured by expression of a co-encoded GFP reporter was also impaired. As most likely explanation, the group identified intra-molecular strand switching during viral genome replication based on the palindromic shRNA sequences, resulting in truncated and partly defect AAV vector genomes. Notably, based on these findings, Xie et al. then designed so-called shAAVs in which the mutated ITR (that causes the scAAV genotype) was purposely replaced by a shRNA sequence and showed that these were fully transduction competent. Finally, using SMRT sequencing, it was found that other hairpin-like structures in AAV genomes including in a CMV or CB promoter, or in a gfp cDNA, can also act as a pseudo-ITR akin to shRNAs and hence compromise vector genome integrity as well as expression.
As a whole, the discovery in this work that various hairpin-like structures can promote AAV vector genome heterogeneity is alarming but concurrently also informative with respect to the design of innovative next-generation AAV constructs, including CRISPR/gRNA-encoding variants. Ideally, through careful design, these future vectors should avoid inherently highly structured elements and repeat-palindromic sequences, which can be achieved by e.g. optimizing codons and reducing G/C content and hence eliminating surrogate ITR structures.
In their second related study that should briefly be noted, Tai et al. (142) used essentially the same SMRT sequencing pipeline, here called AAV-GPseq, to profile packaged AAV vector genomes and to assess their integrity, regardless of the presence or absence of hairpin-like elements. Stunningly, this revealed the presence of a large fraction of heterogenous genomes containing undesired sequences from the rep-cap or adenoviral helper plasmids, as well as chimeras originating from the host-cell (HEK293) genome and often including the AAV ITR. Again, akin to the aforementioned study and together with previous work by others (144,145), this detection of encapsidation of unwanted sequences not only highlights caution but concomitantly offers clues to improve homogeneity in next-generation AAV vectors. In particular, the field may benefit from a more comprehensive analysis of potential sequence motifs that trigger mechanisms such as replication stalling and intra-molecular strand switching that potentially underlie the observed truncations. This should inform future attempts at avoiding these motifs and hence improve not only the quality of AAV vector preparations but also their efficiency and, most importantly, patient safety.
Conclusions
Surprisingly, and in contrast to the flurry of efforts that has been dedicated over the past two decades to innovative technologies for AAV capsid diversification and selection, the field has seen relatively few improvements to the engineering of the second critical AAV vector component, i.e. the recombinant genome and its various parts. Here, we highlighted a selection of respective strategies that have been reported most recently and that we find very encouraging, comprising new generations of synthetic promoters with unprecedented specificity, novel ingenious uses of biological (miRNAs) or physical (light) triggers for on- or off-switching of AAV vectors as well as for their fine-regulation and the application of one of the latest next-generation sequencing platforms for quality control and rational improvement of AAV vector genome integrity, efficiency and stability. Combined with further recent discoveries and improvements by other colleagues that we had to omit for space reasons, such as new insights into the impact of AAV ITRs on vector production and transduction (146–148), this creates substantial hope that a next generation of AAV vectors is indeed in sight that no longer should be judged only by their cover (the capsid) but that are optimized on multiple, synergistic levels and that will benefit an even larger patient population in the future.
Conflicts of Interest
D.G. is a co-founder and shareholder of AaviGen GmbH.
Funding
Smart-HaemoCare (ERA-NET E-Rare-3, JCT-2015 to C.D. and D.G.) project; KARTLE (ERA-NET NEURON, JTC 2017 to C.D. and D.G.) project; German Center for Infection Research (DZIF, BMBF; TTU-HIV 04.803, TTU-HIV 04.815 to D.G.); German Research Foundation (DFG) through the Cluster of Excellence CellNetworks (EXC81) and the Collaborative Research Centers SFB 1129 (Projektnummer 240245660 to D.G.) and TRR 179 (Projektnummer 272983813 to D.G.) as well as through the MYOCURE project.
Acknowledgements
MYOCURE has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 667751. The authors thank Julia Fakhiri for critical reading of the manuscript.
