Small silencing RNAs: State-of-the-art

Abstract

Over just a single decade, we have witnessed the rapid maturation of the field of RNA interference — the sequence-specific gene silencing mediated by small double-stranded RNAs — directly from its infancy into adulthood. With exciting data currently emerging from first clinical trials, it is now more likely than ever that RNAi drugs will soon provide another potent class of agents in our battle against infectious and genetic diseases. Accelerating this process and adding to RNAi's promise is our steadily expanding arsenal of innovative RNAi-based experimental tools and clinically applicable technologies. This article will critically review a selection of relevant recent advances in RNAi therapeutics, from novel asymmetric or bi-functional siRNA designs, deliberate use of small RNAs to regulate nuclear transcription, engineering of potent adeno-associated viral vectors for shRNA expression, exploitation of endogenous miRNAs to control transgene expression or vector tropism, to elegant attempts to inhibit cellular miRNAs involved in human disease. This review will also present cautionary notes on the potential risks inherent to in vivo RNAi applications, before discussing the latest surprising findings on circulating miRNAs in human body fluids, and concluding with an outlook into the possible future of RNAi as an increasingly powerful biomedical tool.

Introduction

Just slightly over a decade ago, a series of celebrated and award-winning publications have jump-started the fascinating and burgeoning research field of RNA interference or RNAi [1], [2]. In its essence, RNAi comprises a naturally occurring, evolutionary conserved, highly efficient and specific pathway by which short double-stranded (ds) RNAs trigger the inhibition of gene expression [3], [4], [5], [6], [7]. In cells of many low and high organisms, these dsRNAs are typically expressed in the nucleus in the form of long primary micro-RNAs or pri-miRNAs [8], [9], [10], [11], [12]. These are then cleaved by the nuclear microprocessor (the RNase III-like endoribonuclease Drosha and partners) into shorter precursor or pre-miRNAs, which are subsequently exported into the cytoplasm by the karyopherin Exportin-5. There, they are further processed by the enzyme Dicer into ~ 22 nucleotide (nt) long miRNAs with characteristic 2-nt 3′ overhangs. One single strand of the miRNA, the mature guide or antisense strand, will next be loaded onto the RNA-induced silencing complex RISC, which will allow it to bind to the 3′ untranslated region (UTR) of its mRNA target to induce silencing. This target binding is remarkably imperfect and promiscuous, since its specificity is primarily determined by nt two to eight (from the 5′ end) of the guide strand, the so-called seed region. As a result, a single miRNA is probably capable of regulating hundreds of cellular targets, and vice versa, each gene is likely controlled by hundreds of miRNAs, suggesting the existence of a very intricate and complex regulatory network. To date, the outcomes of imperfect binding of a miRNA to its target mRNA have been discussed highly controversially, and multiple models have been proposed, including inhibition of translation at the initiation or elongation step, co-translational degradation, re-localization to cellular P-bodies, and/or de-capping or de-adenylation. In either case, the result of miRNA targeting is post-transcriptional silencing (PTGS) of the expression of the affected gene. Further details of the mechanism of miRNA-mediated gene regulation shall not be discussed here, as they have already been covered extensively in a large number of excellent recent reviews [10], [11], [12], [13], [14], [15].

From a historic standpoint, it is curious to note that miRNAs have long been discovered (in 1993, by Victor Ambros et al.) before the 1998 Nature paper by Andrew Fire and Craig Mello, which introduced the term RNAi and was later rewarded with the Nobel prize for the two main authors [1], [16]. The latter publication was certainly distinct and unique, as it suggested for the first time that small dsRNAs could play pivotal roles in gene expression, a function that had previously been unanticipated, including in the initial miRNA studies. The authors themselves were probably surprised when they noticed that exogeneous introduction of dsRNA molecules into C. elegans (a nematode) resulted in suppression of a complementary gene, instead of an increase in its expression [1]. Albeit this suggested a potent and novel means of regulating gene expression, the direct translation into humans was initially prevented since the dsRNAs used were long (> 30 nt), making them potential inducers of an interferon response in higher organisms [2], [17].

Luckily, this concern was alleviated just a few years later, when Tom Tuschl's lab reported the feasibility to trigger efficient gene silencing in mammalian cells by introducing short, only ~ 21-nt long dsRNAs, termed small interfering RNAs or siRNAs [2], [18]. In essence, these siRNAs resemble the products of Drosha/Dicer processing of pri-miRNAs, and like those also enter the RISC complex for subsequent silencing. In contrast, however, siRNAs are typically designed to be perfectly complementary to a sequence in the target mRNA [17]. This is because a perfect match between trigger and target will induce site-specific cleavage of the latter by a particular component of the RISC complex, the Argonaute-2 protein (Ago-2). Out of the four human Ago proteins found at the heart of RISC, only Ago-2 has this remarkable capability to nick a bound mRNA, resulting in its subsequent degradation by cellular RNases [19], [20], [21], [22]. As compared to the miRNA machinery, this cleavage pathway is substantially more potent and results in superior ablation of gene expression. For reasons still not fully elucidated, only perfectly matched RNAi inducers will specifically engage Ago-2, providing the rationale for the typical siRNA design. Moreover, the fact that siRNAs are fully matched to a target should theoretically increase the specificity and efficacy of binding, as hybridization can accordingly rely on 19–21 nt, as opposed to only seven in case of miRNA seed regions.

Unsurprisingly, the general simplicity with which siRNAs can be designed and introduced into cells, coupled with their superior efficiency at gene silencing as compared to prior means (e.g., ribozymes), as well as the assumed lack of immunostimulation (due to their short length), launched an avalanche of efforts to develop siRNAs as novel potent and specific inhibitors of human gene expression. Indeed, since the initial 1998 and 2001 papers introducing RNAi and the concept of siRNAs [1], [2], a vast plethora of publications have already demonstrated the outstanding potential of siRNA technology to perform loss-of-function studies in cultured cells as well as in whole animals, allowing researchers to study gene function in a previously unprecedented way. In addition, the field experienced a second wave of enthusiasm in early 2002, when several groups demonstrated that siRNAs cannot only be delivered exogenously, but also expressed intra-cellularly from gene expression cassettes embedded in DNA plasmids or viral vector genomes [23], [24], [25], [26], [27], [28]. This immediately offered solutions to some specific drawbacks that started to become apparent from the use of siRNAs, especially in whole animals, namely, problems with their delivery and stability [29]. Besides, expression from a plasmid/vector backbone also provides the inherent and essential advantages that siRNA levels can be maintained in the cells for longer periods of time (siRNAs are usually degraded within a few days). Moreover, by utilizing appropriate promoters, RNAi expression can even be controlled in a spatio–temporal manner. Important to note, siRNAs are usually not expressed as two individual strands, but rather in the form of a short hairpin or shRNA, i.e., a molecule in which the two strands are linked via a short (typically seven to nine nt) loop sequence. Expression from an RNA polymerase (pol) II or III promoter, such as CMV, U6 or H1, and use of appropriate termination signals will then yield a dsRNA molecule with a single-stranded loop. This will then be cleaved off by Dicer, leaving basically an siRNA. Like conventional siRNAs, shRNAs are usually also designed to match their target perfectly, in order to be loaded into Ago-2 and trigger the most efficient means of gene silencing.

The availability of essentially two categories of RNAi triggers, exogenously delivered siRNAs or endogenously expressed shRNAs, each with their own assets, has spurred tremendous optimism regarding potential RNAi-based therapies for human disease. Indeed, a vast and rapidly growing number of publications (extensively reviewed in Refs. [5], [7], [30], [31], [32], [33] and elsewhere) have already provided overwhelming and strong proof-of-concept that theoretically, RNAi triggers can be engineered to silence virtually any therapeutically relevant gene with a known sequence. Prominent examples for clinically pertinent targets are genes involved in malignant transformation of cells, or playing central roles in human infection with viral pathogens, ideally those encoded by the viral genomes themselves. It is thus not surprising that already six years after the initial report of RNAi in nematodes (see above), and following a rapid series of successful pre-clinical studies in small animals, the first clinical Phase I studies in humans had been launched [31], [34]. As of today, several biotechnology companies have in fact already proceeded into Phase II and III trials, and the development of RNAi drugs has become an enormous and burgeoning market. Notably, while the vast majority of clinical trials thus far have employed siRNAs (mostly in the eye, to treat age-related macular degeneration (AMD)), the first study involving (lenti)viral vector-encoded shRNAs has also been started recently. In this unique trial, conducted by the City of Hope National Medical Center and the company Benitec, RNA pol III-expressed shRNAs directed against the HIV rev and tat exons will be ex vivo delivered into blood stem cells, a strategy that might eventually provide a future treatment for AIDS [31].

Despite the hope and excitement surrounding the impressive pre-clinical achievements and recent clinical data thus far, various adverse findings from in vitro and in vivo experiments also underscore the necessity to further improve the design and application of our present arsenal of RNAi triggers. Perhaps the greatest setback came with the discovery that siRNAs (and perhaps shRNAs as well) might not be as target-specific as originally anticipated. In fact, delivery of a single artificial RNAi trigger might inadvertently affect the expression of hundreds of cellular genes, a phenomenon that was termed “off-targeting” [35], [36], [37], [38]. This can occur via perfect binding of the guide strand to multiple mRNA sequences (a risk that is nowadays unlikely due to improvements in siRNA design algorithms), but also via much harder-to-predict and more difficult-to-eliminate miRNA-like down-regulation of cellular off-targets [36]. Moreover, off-targeting can also originate from the presumably inactive passenger (sense) strand of the siRNA, in cases where it is unintentionally incorporated into Ago/RISC. These adverse findings have sparked a flurry of efforts to optimize the rules and algorithms for siRNA design whose description would be beyond the scope of this article, so that the reader is referred to other current reviews for further information [17], [34], [39]. Notably, one of the latest and most original approaches to eliminate passenger strand off-targeting, via the synthesis and use of asymmetric interfering RNA (aiRNAs), is reviewed in more detail in Section 2.1.1 below.

A second fundamental problem that became apparent in many of the thorough pre-clinical siRNA evaluations is that these molecules are not only often less specific than expected, but also substantially more immunogenic. This was initially surprising, as the deliberate reduction in length to 19 or 21 nt should have sufficed to circumvent the recognition by typical cellular sensors of long dsRNA. One example is protein kinase R (PKR), whose inadvertent activation results in an adverse interferon response and the shutting down of global gene expression. Yet, most recently, it had to be appreciated that other sensors of foreign RNA, especially the toll-like-receptors TLR-3, -7 and -8, are also able to detect siRNAs based on particular highly immunostimulatory sequence elements [7]. In addition, structural features of certain siRNA variants can also be detected by the cellular helicase RIG-1(retinoic acid-inducible gene-1), leading to an interferon-independent immune response. Similar to the issues with unwanted off-targeting described above, numerous strategies have been developed and are currently being evaluated to ablate the immune response from first-generation siRNA triggers (for recent comprehensive reviews, see Refs. [34], [40], [41]). A very unique approach will be reviewed in more detail in Section 2.1.2 below, which basically turns the inherent disadvantage of a particular type of immunogenic siRNAs onto its head, by purposely exploiting immune stimulation as an adjuvans to RNAi-mediated silencing.

A third essential and general issue with the use of siRNAs in animals, or even more so in a clinical setting, is the need for efficient and specific delivery to target cells and organs. As naked siRNA is neither stable in serum nor readily taken up by most of the clinically relevant target cells in the body, many academic and industrial groups have developed a plethora of chemical modifications and conjugations to improve these two parameters. Again, the wealth of reported strategies is already so exhaustive that the reader is referred to more specialized reviews and references therein [34], [39]. Also, several accompanying reviews in this issue of ADDR will describe particularly promising strategies in more detail. An alternative means already mentioned above to concurrently overcome both problems is to embed siRNA-encoding sequences in the context of a recombinant viral vector, in the form of shRNA expression cassettes. The promise of this approach stems from the fact that viral vectors have already been optimized as well as evaluated (pre-clinically and clinically) for decades, providing us with a fertile tool chest to improve both efficiency and specificity of delivery [26]. Amongst the viral vectors engineered for shRNA delivery thus far, the two most outstanding candidates that have emerged are lentiviruses (reviewed in the accompanying article by Shankar), as well as vectors based on Adeno-associated virus or AAV [42], [43]. The latter holds a unique set of advantages over all other viral vector systems, including the complete lack of pathogenicity and a superior degree of versatility, and will therefore be reviewed in greater detail in Section 2.2.1 below.

Last but not least, a fourth potential problem with the clinical use of RNAi only became truly obvious with the advent of highly efficient viral vector-based expression systems, in particular the above mentioned AAV. In a seminal 2006 paper, Grimm et al. reported that intravenous injection of adult mice with high doses of shRNA-expressing AAV vectors resulted in early liver damage and ultimately organ failure and death in many animals [44]. The findings that the effect was dose-dependent and concurrently shRNA sequence-independent hinted at non-specific saturation of the endogenous RNAi machinery as the underlying cause for morbidity and mortality. Importantly, this potentially detrimental side effect from RNAi therapies is not restricted to vector-encoded shRNAs, as Castanotto et al. showed one year later that saturation phenomena can also be observed with high doses of siRNAs [45]. Because of the obvious high relevance of these potential adverse effects for future clinical RNAi applications, Section 2.2.2 below will review these two important papers in more detail and discuss potential solutions.

Concurrent with the efforts in the field to improve and optimize existing artificial RNAi triggers, many other groups have also made substantial progress in our understanding of the natural RNAi pathways, in particular related to miRNA functionality. Our emerging picture of the basic biology of miRNAs, as well as of several other classes of small RNAs found in mammals/humans, has already been reviewed extensively elsewhere [10], [14], [46], [47]. Here, Section 2.3 below will summarize several most recent attempts to exploit endogenous miRNAs for therapeutic purposes. A particularly attractive feature of miRNAs in this respect is their frequently high degree of tissue-specificity, often coupled with developmental stage- or differentiation-specific expression patterns [9], [48]. Accordingly, several groups have already reported first successes with original attempts to utilize endogenous miRNAs for specific control of transgene expression (Section 2.3.1), or of replication of oncolytic viruses (Section 2.3.2). In parallel, others have focused on the pivotal roles that miRNAs play in many aspects of the cell, including differentiation, development, proliferation or apoptosis, to name a few. Accordingly, the emerging model is that miRNA dysregulation might be critically involved in, or even causative of, a large number of human diseases, from cancer to viral infections [49], [50], [51]. Thus, increasing numbers of reported diverse strategies aim at inhibiting miRNAs that are up-regulated under certain pathological conditions. Similar to siRNAs and shRNAs, expression of these inhibitors from a viral vector backbone might hold inherent advantages, which is why this particular approach will be reviewed in more detail in Section 2.3.3 below. Finally, the last Section 2.3.4 will give an overview over a series of papers describing a most recent, very unexpected development in the miRNA field, namely, their discovery in human body fluids. Rapidly accumulating evidence suggests that not only can miRNAs be detected in extra-cellular environments, but their distribution might also mirror that within the cells from which they originate. If true, this would suggest that quantification and profiling of miRNAs in easily accessible fluids, such as blood or urine, could provide a straight-forward, highly promising avenue towards the establishment of novel biomarkers for human disease.

Fig. 1 summarizes the various topics that will be reviewed in detail in the following.