Advances in MicroRNAs: Implications for Gene Therapists

Abstract

MicroRNAs (miRNAs) are a class of small regulatory RNAs that are thought to regulate the expression of as many as one-third of all human messenger RNAs (mRNAs). miRNAs are thought to be involved in diverse biological processes, including tumorigenesis. Analysis of miRNA levels may have diagnostic implications. Evidence shows that numerous viruses interact with the miRNA machinery, and that a number of viruses encode their own miRNAs. It seems likely that miRNAs will be implicated in many human diseases. Manipulation of miRNA levels by gene therapy provides an attractive new approach for therapeutic development. This review focuses on approaches to manipulate miRNA levels in cells and in vivo, and the implications for gene therapy. Furthermore, we discuss the use of endogenous miRNAs as scaffolds for the expression of RNA interference (RNAi) as well as competition between exogenous RNAi triggers and endogenous miRNAs. Because short interfering RNAs can also act as miRNAs, seed matches with the 3′ untranslated regions of genes should be avoided to prevent off-target effects. Last, we discuss the use of miRNAs to avoid immune responses to viral vectors.

BACKGROUND

MICRORNAs (miRNAs) ARE SMALL, ∼22-nucleotide noncoding regulatory RNAs involved in RNA-mediated gene silencing (Lagos-Quintana et al., 2001). At present, more than 500 human miRNAs have been identified. miRNAs are important gene regulators during processes that include differentiation, proliferation, apoptosis, and insulin secretion as well as heart, brain, and skeletal muscle development (see Bartel [2004], Harfe [2005], Kloosterman and Plasterk [2006], and Krutzfeldt and Stoffel [2006] for comprehensive reviews). The miRNA pathway involves several processing steps to produce the mature miRNA used for gene silencing (Fig. 1). Long primary miRNA transcripts (pri-miRNAs) transcribed by polymerase II or III are cleaved by a complex of Drosha and DGCR8 (DiGeorge critical region-8) into the pre-miRNA. The pre-miRNA is exported from the nucleus to the cytoplasm by Exportin 5, where it undergoes a second cleavage step by Dicer/TRBP (HIV TAR RNA-binding protein) to produce a short, ∼20-nucleotide duplex RNA with two base 3′ overhangs. Usually, a single strand of this duplex RNA is incorporated into the RNA-induced silencing complex (RISC), although a subset of miRNAs incorporate both strands individually into distinct RISCs. The mature miRNA strand in RISC is called the guide strand, and the opposite strand not incorporated is referred to as the passenger strand. The minimal RISC is composed of an Argonaute (Ago) protein, Dicer, and TRBP, although it is likely that RISC interacts with numerous accessory proteins (Peters and Meister, 2007). Once the mature miRNA guide strand is loaded into the RISC, it binds to the 3′ untranslated region (UTR) of its mRNA target and induces silencing. It is currently thought that the specificity of the miRNA for its target mRNA is primarily (although not exclusively) specified by the “seed sequence” (nucleotides 2–8 of the guide strand; Fig. 2). One implication, evidenced by the short length of the seed sequence, is that miRNAs are remarkably promiscuous. It has been suggested that some miRNAs can target hundreds of mRNAs (Farh et al., 2005; Lim et al., 2005; Stark et al., 2005). miRNAs repress their target mRNAs by a variety of mechanisms. miRNAs with perfect complementarity to their target mRNAs cause direct mRNA cleavage, although this is thought to be rare in mammals. More often, miRNAs with partial complementarity to mRNA 3′ UTRs direct translational repression, decapping, or deadenylation.

FIG. 1.

The miRNA/RNAi pathway. miRNAs are transcribed from Pol II or Pol III promoters as long transcripts. These pri-miRNAs are processed by Drosha/DGCR8 to form pre-miRNAs. Pre-miRNAs are exported to the cytoplasm by Exportin 5, where they are further processed into a short duplex by Dicer/TRBP. One strand of the duplex (the guide strand) is incorporated into RISC and the other strand is destroyed. RISC mediates gene silencing by inhibition of translation, decapping, deadenylation, or direct target cleavage. shRNAs or siRNAs enter the miRNA/RNAi pathway and mediate target cleavage. Ago, Argonaute protein; DGCR8, DiGeorge critical region-8; miRNA, microRNA; Pol, RNA polymerase; pri-miRNA, primary miRNA transcript; RISC, RNA-induced silencing complex; shRNA, short hairpin RNA; siRNA, short interfering RNAs; TRBP, HIV-1 TAR RNA-binding protein.

FIG. 2.

The miRNA seed sequence determines specificity. The seed sequence (nucleotides 2–8 of the miRNA, boxed and in boldface) is the primary determinant of miRNA specificity. The predicted base pairing of the hepatitis C virus 5′ UTR with miR-122 is shown. [Adapted from Jopling et al., Science 309:1577 (9/2/2005). Reprinted with permission from AAAS.]

The use of RNA interference (RNAi) for gene silencing has become widespread, both for functional studies and for therapeutic purposes in a gene therapy setting. Commonly, RNAi is initiated by transfection of cells or animals with synthetic short interfering RNAs (siRNAs) or introduction of plasmids or recombinant viruses expressing short hairpin RNAs (shRNAs). It has become apparent that exogenous synthetic siRNAs can silence genes because they resemble the endogenous miRNAs generated by Dicer/TRBP, and thus can be loaded into the RISC (Fig. 1). Likewise, shRNAs resemble the pre-miRNA. Exogenous siRNA and shRNAs are designed to be perfectly complementary to their target mRNA, and therefore they direct cleavage. However, they can also act as miRNAs. The consequences for design of siRNAs/shRNAs are described below. These exogenous siRNAs and shRNAs also have the potential to overload the miRNA-processing pathway.

The last few years has witnessed an explosion of publications on miRNAs. This review is not intended as a comprehensive survey of the miRNA literature. Rather, it highlights key issues of importance to researchers who seek to integrate gene therapy approaches with our emerging understanding of miRNAs.

miRNAs and cancer: diagnosis, prognosis, and therapeutic potential

High-throughput strategies, such as miRNA microarrays (Liu et al., 2004), bead-based flow cytometric miRNA expression profiling (Lu et al., 2005), and quantitative polymerase chain reaction (qPCR) (Chen et al., 2005; Jiang et al., 2005; Murakami et al., 2005; Raymond et al., 2005), have been developed for miRNA profiling. These high-throughput methods have shown that miRNA levels were dramatically shifted in various cancers (Liu et al., 2004; He et al., 2005; Iorio et al., 2005; Lu et al., 2005; Murakami et al., 2005) (refer to Calin and Croce [2006] and Esquela-Kerscher and Slack [2006] for a comprehensive review of miRNAs and cancer). Interestingly, most differentially expressed miRNAs were downregulated in tumors compared with normal tissues. These studies revealed that miRNA profiling could be used to distinguish between normal and cancerous tissues, developmental lineage, as well as differentiation. Identifying miRNA expression profiles during tumorigenesis will likely provide additional targets for gene therapy, while improving predictions for diagnosis and prognosis.

For most of the miRNAs found to be dysregulated in cancer, a causative relationship between differential expression and cancer has not been established; however, for a subset there is now convincing evidence that some miRNAs are directly involved in cancer formation. These “onco-miRs” can function as tumor suppressors (such as miR-15a and miR-16-1 [Calin et al., 2002, 2005; Cimmino et al., 2005] and the let-7 family of miRNAs [Calin et al., 2002; Takamizawa et al., 2004; Johnson et al., 2005, 2007]) and as oncogenes (such as the miR-17–92 cluster [He et al., 2005], miR-155 [Costinean et al., 2006], and miR-21 [Si et al., 2007]). A few examples are provided below.

The first tumor suppressor miRNAs were identified in B cell chronic lymphocytic leukemia (CLL) (Calin et al., 2002). miR-15a and miR-16-1 are located in a cluster on chromosome 13q14, which is deleted in more than half of all CLL cases. This correlates with the loss of expression of miR-15a and miR-16-1. It was later shown that miR-15a and miR-16-1 negatively regulated expression of the antiapoptotic gene Bcl2 (Cimmino et al., 2005). After discovering that miRNAs were located within a frequently deleted chromosomal region in CLL, researchers investigated the presence of miRNAs in “fragile sites” or chromosomal loci that are susceptible to chromosome breakage, amplification, and fusion with other chromosomes (Calin et al., 2004). They found that 50% of miRNAs were located within human fragile sites. This provided further evidence that miRNAs are involved in cancer and that they may be useful targets for therapy. Interestingly, in one report, Donsante and coworkers found that hepatocellular carcinoma associated with adeno-associated virus integration resulted from a vector insertion near a micro-RNA cluster (Donsante et al., 2007).

One of the most studied oncogenic miRNAs is the miR-17–92 cluster. The miR-17–92 cluster is at a locus that is amplified in human B cell lymphomas and therefore upregulated relative to normal tissues (He et al., 2005). Furthermore, enforced expression of the miR-17–92 cluster in the context of c-myc overexpression accelerated tumor development in a mouse B cell lymphoma model.

Another example of an oncogenic miRNA is miR-155. miR-155 is overexpressed in various types of B cell malignancies (Metzler et al., 2004; Eis et al., 2005; Kluiver et al., 2005). Costinean and coworkers produced a transgenic mouse that overexpresses miR-155 in B cells (Costinean et al., 2006). These mice exhibited polyclonal preleukemic pre-B cell proliferation followed by B cell malignancy, which strongly suggests that miR-155 is directly implicated in the initiation and/or progression of these diseases.

Although the examples discussed above are only a small subset of miRNAs implicated in cancer development, they illustrate the point that targeting aberrantly expressed miRNAs by therapeutic approaches could have a tremendous impact on future cancer treatments. Strategies for manipulating miRNA levels are discussed below.

Viruses and miRNAs

Increasing evidence suggests that viruses interact with the miRNA machinery (reviewed in Sullivan et al. [2006] and Dykxhoorn [2007]). Cloning of miRNAs from virus-infected cells revealed that many latent DNA viruses encode their own miRNAs (Pfeffer et al., 2004). To date, 11 viruses have been shown to encode their own miRNAs (see the miRBase Sequence Database [previously the miRNA Registry] at http://microrna.sanger.ac.uk/sequences/index.shtml). These include Epstein–Barr virus (EBV) (Pfeffer et al., 2004), herpes simplex virus type 1 (HSV-1) (Cui et al., 2006), human cytomegalovirus (Pfeffer et al., 2005), human immunodeficiency virus type 1 (Omoto and Fujii, 2005), Kaposi sarcoma-associated herpesvirus (Cai et al., 2005; Pfeffer et al., 2005; Samols et al., 2005), Marek's disease virus (Burnside et al., 2006), Marek's disease virus type 2 (Yao et al., 2007), mouse gammaherpesvirus 68 (Pfeffer et al., 2005), rhesus lymphocryptovirus (Cai et al., 2006), rhesus monkey rhadinovirus (Schafer et al., 2007), and simian virus 40 (SV40) (Sullivan and Ganem, 2005). These viral miRNAs represent potential antiviral targets and could also serve as diagnostic markers for viral infection or stage of infection/latency. Below, we discuss several examples of the interplay of host and viruses via the miRNA pathway.

Viruses use miRNAs in various ways, such as to inhibit viral or host transcripts or to recruit host miRNAs for viral replication. EBV and SV40 are examples of viruses that encode miRNAs that target their own viral transcripts. EBV encodes at least 18 miRNAs (Pfeffer et al., 2004; Sullivan et al., 2006), one of which targets the transcript encoding the viral polymerase to negatively regulate it later during infection (Furnari et al., 1993; Pfeffer et al., 2005). Likewise, SV40 encodes two viral miRNAs that target the viral T-antigen transcript (Sullivan et al., 2005). This regulatory mechanism may aid the virus in immune evasion.

HSV-1 provides an example of a virus that encodes miRNAs that target host transcripts. The HSV-1 latency-associated transcript encodes an miRNA predicted to target two host transcripts involved in apoptosis (transforming growth factor-β and mothers against decapentaplegic homolog-3) (Gupta et al., 2006).

Host miRNAs can also be used for viral replication as well as antiviral defense. The endogenous miRNA, miR-122, is required for replication of hepatitis C virus (Jopling et al., 2005). Figure 2 shows the interaction of miR-122 with the 5′ UTR of hepatitis C virus. As discussed below, two groups have successfully used antisense approaches to modulate the levels of miR-122 in mouse models (Krutzfeldt et al., 2005; Esau et al., 2006). Inhibition of viral replication in vivo using this strategy remains to be demonstrated. In contrast, primate foamy virus type 1 (PFV-1) has an mRNA that is the target of the endogenous host miRNA, miR-32 (Lecellier et al., 2005). Interestingly, PFV-1 encodes a protein (Tas) that may interfere with the host miRNA pathway (Lecellier et al., 2005). These examples suggest that miRNA overexpression or ablation may represent a novel antiviral therapeutic strategy.

Last, at least one commonly used gene transfer vector encodes an RNA that inhibits the miRNA pathway. The highly expressed adenoviral virus-associated type I (VAI) RNA was shown to be a competitive inhibitor of at least two steps in the miRNA pathway (Lu and Cullen, 2004; Andersson et al., 2005). It is a competitive inhibitor of Exportin 5 as well as Dicer and small VAI RNAs generated by Dicer cleavage have been shown to associate with RISC. These findings suggest that saturation of the endogenous miRNA pathway should be assessed in human clinical trials with adenoviral vectors, when possible.

Other miRNA/disease associations

Although the vast majority of associations between miRNAs and human disease are related to tumorigenesis, there is increasing evidence that miRNAs could be involved in multiple diseases. For example, 28 miRNAs were found to be dysregulated in two mouse models of cardiac hypertrophy (van Rooij et al., 2006). miR-9 and miR-128a were overexpressed in the brains of Alzheimer's patients (Lukiw, 2007) and 16 miRNAs were differentially expressed in the brains of schizophrenics (Perkins et al., 2007). Last, a mutation in the 3′ UTR of the SLITRK1 gene that is associated with Tourette's syndrome led to enhanced repression by miR-189 (Abelson et al., 2005).

THERAPEUTIC MANIPULATION OF miRNAs

As described above, miRNAs are likely to be involved in numerous genetic diseases, such as cancer, as well as some viral infections. These miRNAs represent potentially novel therapeutic targets. It would clearly be desirable to specifically modulate the levels of individual miRNAs in vivo. Fortunately, decades of research in the fields of antisense, ribozymes, and gene therapy provide a fertile tool chest that serves as a starting point for manipulating miRNA levels in vivo. Below, we discuss methods for overexpressing and ablating miRNAs and provide examples of initial efforts to apply these techniques in disease settings. Again, this is not intended to be an exhaustive review, but rather a review of important paradigms for gene therapists.

Inhibition of miRNA function through antisense and miRNA sponges

Numerous groups have now shown that antisense oligonucleotides complementary to the guide strand of miRNAs can inhibit their function in cultured cells (Esau et al., 2004; Meister et al., 2004; Poy et al., 2004; Cheng et al., 2005; Lee et al., 2005; Schratt et al., 2006), flies (Boutla et al., 2003; Hutvagner et al., 2004; Leaman et al., 2005), and mice (Hutvagner et al., 2004; Esau et al., 2006; Krutzfeldt et al., 2007). These inhibitors, referred to as “anti-miRs” or “antago-miRs,” could have significant clinical potential (Poy et al., 2004; Jopling et al., 2005; Krutzfeldt et al., 2005; Esau et al., 2006; Weiler et al., 2006). Early studies showed that 2′-O-methyl oligoribonucleotides were able to block miRNA function in cultured cells (Meister et al., 2004) and in flies (Hutvagner et al., 2004). In contrast, unmodified 2′-deoxy oligonucleotides (Meister et al., 2004) and 2′-deoxy phosphorothioate oligonucleotides (Davis et al., 2006) were unable to inhibit miRNA activity in cultured cells. Several reports suggest that anti-miRs interfere with miRNA-mediated silencing by binding to the mature miRNA guide strand in the RISC complex (Hutvagner et al., 2004; Davis et al., 2006; Krutzfeldt et al., 2007). Interestingly, it appears that anti-miR binding leads to degradation of the targeted miRNA by some previously undescribed mechanism (Esau et al., 2006; Krutzfeldt et al., 2007).

Although 2′-O-methyl oligoribonucleotides appeared to be effective inhibitors of miRNA function, more potent inhibitors have been described. Davis and coworkers conducted a limited comparison of the anti-miR activities of various oligonucleotides with different backbones and 2′ modifications (Davis et al., 2006). They found that uniformly 2′-O-methoxyethyl oligonucleotides (2′-MOEs) and oligonucleotides in which every third nucleotide was substituted with a locked nucleic acid (LNA) residue (Exiqon, Woburn, MA) were the most potent. Another report also showed that LNA-substituted oligonucleotides were effective at antagonizing miRNAs (Orom et al., 2006). Several vendors sell LNA-modified oligonucleotides (including Integrated DNA Technologies [Coralville, IA] and Exiqon [Vedbaek, Denmark]). 2′-O-Methoxyethyl oligonucleotides (Isis Pharmaceuticals, Carlsbad, CA) are not commercially available. 2′-Fluoro-oligonucleotides with a phosphorothioate backbone also showed significant activity in cultured cells (Davis et al., 2006). Oligonucleotides with central stretches of 2-deoxy residues can direct RNase H-mediated degradation of RNA/oligonucleotide duplexes. However, the inability of such oligonucleotides to antagonize miRNA function suggested that RNase H may not have access to miRNA-loaded RISCs bound to anti-miRs (Davis et al., 2006). Last, because transfection of cells with oligonucleotides with some chemistries can lead to a loss of cell viability at high doses (Davis et al., 2006), care should be taken to control for toxicity in anti-miR studies.

A significant barrier to the application of anti-miRs in vivo will concern efficient delivery of these molecules either systemically or locally. However, two groups have reported successful miRNA inhibition in vivo. Krutzfeldt and coworkers targeted several different miRNAs in mice, using antago-miRs (Krutzfeldt et al., 2005). These are 2′-O-methyl oligonucleotides with two and four phosphorothioate linkages at the 5′ and 3′ ends, respectively. The antago-miRs were conjugated to cholesterol to reduce renal clearance and enhance uptake. Administration of three injections of high doses of antago-miR (80 mg/kg) resulted in substantial decreases in the levels of endogenous miR-16 in all tissues tested except for the brain. The authors then targeted the abundant liver miRNA, miR-122. They conducted gene expression profiling and identified hundreds of genes that were upregulated in the liver on treatment with an miR-122 antago-miR. Many of these genes were involved in cholesterol biosynthesis and, indeed, miR-122 antagomiR treatment led to a 40% reduction in plasma cholesterol levels. These findings provided the proof-of-principle that anti-miRs have the potential to be therapeutic. Because miR-122 is required for viral replication, miRNA downregulation may have antiviral activity.

In a separate article, the authors found that longer antago-miRs (25 nucleotides) were more effective in vivo (Krutzfeldt et al., 2007). This was presumably because the phosphorothioate bases at the ends decreased the melting temperature of the miRNA–antagomiR duplex. Lengthening the antago-miR resulted in more pairing with the more stable interior 2′-deoxy bases, while maintaining the nuclease stabilization afforded by the terminal phosphorothioate linkages. The authors also showed that some antago-miRs exhibit mismatch specificity, although this was highly position dependent (Krutzfeldt et al., 2007).

In a similar study, Esau and coworkers reported that unconjugated 2′-MOE miR-122 anti-miRs reduced the levels of miR-122 in mouse liver after intraperitoneal injection of 12.5—75 mg/kg twice weekly for 4 weeks (Esau et al., 2006). As with the above-described study, the authors observed elevations in the levels of predicted target mRNAs and decreases in plasma cholesterol. The authors then applied this anti-miR strategy in a disease model of diet-induced obesity in mice. Treatment with miR-122 anti-miR reduced plasma cholesterol by 35%. Treated animals also had lower levels of steatosis.

Ebert and coworkers described another approach to inhibiting the function of specific miRNAs in cells (Ebert et al., 2007). They reasoned that by expressing an mRNA containing multiple binding sites for an endogenous miRNA, they could bind up the miRNA and prevent its association with its endogenous targets. They referred to these synthetic mRNAs as “miRNA sponges.” To prevent cleavage of the mRNA containing the miRNA-binding sites, they introduced central mismatches in the miRNA/mRNA duplex at positions 9–12, which were essential for activity. This should prevent release and recycling of the miRNA. They designed both polymerase (Pol) II- and Pol III-driven miRNA sponges, although they observed some nonspecific depression of luciferase expression with Pol III-driven sponges. The use of Pol II promoters should enable the creation of inducible and tissue-specific miRNA sponges. They showed that in cultured cells, their miRNA sponges were at least as effective as LNA anti-miRs. miRNA sponges were also effective at repressing multiple members of an miRNA family. Stably integrated, miRNA sponge expression cassettes yielded ∼40% as much inhibition as transiently transfected plasmids. Although it remains to be demonstrated, these results suggest that the generation of transgenic animals expressing inducible miRNA sponges may be feasible. Interestingly, one report suggests that nature uses miRNA sponges. Franco-Zorrilla and coworkers showed that in Arabidopsis, the noncoding IPS1 RNA serves as a sponge for miR-399 (Franco-Zorrilla et al., 2007). As described above, the IPS1/miR-399 hybrid contains internal mismatches that are essential for its activity.

The examples described above demonstrate the proof of principle that miRNA function can be inhibited in vivo and in cultured cells. However, there is clearly a need to refine and optimize these approaches. Creative methods that mimic natural biological processes may provide clues as to how to accomplish this.

Downregulating expression of miRNAs involved in cancer

Because miRNAs are often dysregulated in cancer, modulating their levels may have therapeutic benefit. miR-21 was an early example of how downregulating an miRNA inhibited cancer development. miR-21 is overexpressed in many different cancer types, such as breast cancer (Iorio et al., 2005; Si et al., 2007), ovarian cancer (Iorio et al., 2007), hepatocellular carcinoma (Meng et al., 2007), gliomas (Chan et al., 2005), pancreatic cancer (Lee et al., 2007), and CLL (Fulci et al., 2007). miR-21 plays a critical role in cell proliferation by targeting and, therefore, inhibiting the tumor suppressor genes TPM1 (tropomyosin 1) (Zhu et al., 2007) and PTEN (phosphatase and tensin homolog) (Meng et al., 2007). Using a xenograft carcinoma model, researchers transiently transfected MCF-7 cells with 2′-O-methyl oligonucleotides complementary to miR-21 (or negative control oligonucleotide), and then injected them into mammary pads of female nude mice (Si et al., 2007). After 28 days, tumors derived from MCF-7 cells transfected with anti-miR-21 were 50% smaller in size compared with tumors of mice receiving cells transfected with negative control oligonucleotide. The authors showed that suppression of miR-21 lasted 2 weeks. Although the model system is somewhat artificial (ex vivo treatment of tumors), this report demonstrates that knockdown of oncogenic miRNAs can inhibit tumor growth. Thus, knockdown of miRNAs by systemic or local delivery of anti-miRs could represent a novel anticancer therapy.

Overexpression of miRNAs

When miRNA levels are reduced in human disease, it may be desirable to use gene therapy-based approaches to restore miRNA expression in certain tissues. The considerable experience with nonviral and viral delivery of transgenes that has accumulated over the past few decades will certainly facilitate this goal. As is discussed below, the endogenous miRs miR-26a (McManus et al., 2002) and miR-30 (Zeng et al., 2002; Boden et al., 2004) have been used as scaffolds for the expression of RNAi triggers. Clearly, these well-characterized systems could be used to produce mature miRNA guide strands in cells. For example, overexpression of an miRNA (or miRNAs) whose expression is downregulated in cancer may be beneficial as treatment. In some cases, it may be desirable to express endogenous miRNAs in their native context with their own promoters, thus maintaining proper regulation. As more complete miRNA transcription units are characterized, this may become increasingly possible.

Reexpression of miRNAs downregulated during cancer

Loss of tumor suppressor genes is a common theme during tumorigenesis. As previously discussed, most differentially expressed miRNAs in tumors were downregulated compared with normal tissues. Exogenous expression of miRNAs downregulated during tumorigenesis may have potential as a therapeutic strategy. Researchers have demonstrated that miR-34a was downregulated in 36% of human colon cancers compared with counterpart normal tissues (Tazawa et al., 2007). In vivo reexpression of miR-34a suppressed growth of HCT 116 and RKO cells in mice. Furthermore, miR-34a functions as a potent suppressor of cell proliferation through modulation of the E2F signaling pathway. Loss of miR-34a expression could contribute to aberrant cell proliferation, leading to colon cancer development (Tazawa et al., 2007). Thus, exogenous expression of miRNAs may have application as an anticancer strategy.

Synthetic miRNAs targeting genes associated with metastasis

An alternative approach for cancer therapeutics involves creating synthetic miRNAs that target genes responsible for onco-genesis or metastasis. Loss of certain miRNAs in breast tumor cells can lead to increased invasion and metastasis of breast cancer. Inhibition of chemokine (C-X-C motif) receptor 4 (CXCR4) with siRNA inhibited metastasis of breast cells in vivo (Liang et al., 2004, 2005). Therefore, researchers designed and synthesized a pre-miRNA with an miR-155 backbone that targeted CXCR4 in order to determine whether the CXCR4/SDF-1 pathway was regulated by expression of miRNAs (Liang et al., 2007). After transfecting breast tumor cells with the synthesized miRNA, they observed a significant decrease in CXCR4 in breast tumor cells and reduced migration and invasion. Furthermore, they formed fewer lung metastases in vivo compared with control miRNA-transfected cells.

Another example of inhibiting metastatic-related genes involves PRL-3 (protein tyrosine phosphatase of regenerating liver-3), whose high expression is associated with lymph node metastasis in gastric carcinoma. This study used an artificial miRNA with an miR-155 backbone that targeted PRL-3. PRL-3 knockdown effectively suppressed the growth of peritoneal metastases and improved the prognosis for nude mice (Li et al., 2006). Therefore, synthetic miRNAs that target genes involved in metastasis serve as an alternative means for treating cancer.

Epigenetic gene therapy in cancer

Manipulation of miRNA levels may also be a means to alter DNA methylation. DNA methylation is a crucial mechanism associated with epigenetic regulation. Changes in the pattern of DNA methylation, either increased (hypermethylation) or decreased (hypomethylation), have been identified in all types of cancer cells examined so far. DNA methyltransferase inhibitors, both nucleoside and nonnucleoside analogs, are currently being tested for cancer therapy (reviewed in Brueckner et al., 2007). However, similar to most cancer treatments, there are restrictions associated with efficacy and cytotoxicity. Therefore developing a new means of targeting DNA hypermethylation is of great interest. The miR-29 family is downregulated in many cancers that contain aberrant DNA hypermethylation patterns, including lung cancer. It was shown that miR-29 targets DNA methyltransferase-3A (DNMT3A) and -3B (Fabbri et al., 2007). On enforced expression of miR-29, normal methylation patterns in lung cells were reestablished and tumor suppressors previously silenced by promoter hypermethylation were reactivated. Furthermore, expression of miR-29 inhibited tumorigenicity both in vitro and in vivo. Epigenetic gene therapy approaches can be useful across many tumor types. miRNA-mediated therapy may be useful in combination with DNA methyltransferase inhibitors that are nontoxic, but have limited efficacy. One study showed that treatment of malignant cholangiocytes with gemcitabine, a nucleoside inhibitor, alters the miRNA expression profiles (Meng et al., 2006). Furthermore, inhibiting miRNAs highly overexpressed in malignant cholangiocytes, such as miR-21, increased gemcitabine cytotoxicity. Therefore miRNAs may contribute to chemoresistance and inhibiting miRNAs overexpressed in cancer may increase the efficacy of chemotherapy.

IMPLICATIONS OF miRNAs FOR RNAi GENE THERAPY

miRNAs as scaffolds for RNAi triggers

Since the discovery of RNAi, researchers have been interested in increasing the efficacy of gene-silencing tools. One approach involves expressing RNAi triggers in the context of naturally occurring miRNAs. This can be accomplished by replacing sequences in the central stem of endogenous pre-miRNAs, such as miR-26a (McManus et al., 2002) and miR-30 (Zeng et al., 2002; Boden et al., 2004), with sequences that will direct RNAi against a target of interest (RNAi triggers). The premise for this is that endogenous miRNA scaffolds will be recognized and efficiently processed by the miRNA/RNAi machinery. When designing miRNA-based RNAi triggers, care should be taken to maintain the native sequence around the Drosha cleavage site, as that can significantly influence silencing efficiency (Zhou et al., 2005). By increasing our understanding of the processing and targeting of natural miRNAs we will be able to design a more robust silencing tool. Boden and coworkers reported that pri-miRNAs may be processed more efficiently than shRNAs, leading to 80% improvement in silencing efficiency (Boden et al., 2004). Hairpins containing loop sequences derived from pri-miRNAs were more efficiently transported to the cytoplasm than were hairpins containing artificial loops (Kawasaki and Taira, 2003). Numerous groups, including our own, have now used this approach. Silencing efficiencies of >95% can be achieved with miRNA-derived RNAi triggers (A. McCaffrey, unpublished results). Importantly, pri-miRNAs can be expressed from polymerase II promoters (Boden et al., 2004), which enables the use of tissue-specific promoters, as well as inducible expression. Alternatively, miRNA-based RNAi triggers can be expressed from polymerase III promoters for high-level expression. Thus, expression of RNAi triggers in the context of endogenous miRNAs may offer significant advantages.

Competition between exogenous RNAi triggers and endogenous miRNAs

All drugs have a “therapeutic window”; insufficient doses are not efficacious and excessive doses cause toxicity. In a sobering report, Grimm and coworkers showed that, when shRNAs are highly overexpressed in mice, they can cause acute toxicity (Grimm et al., 2006). The authors speculate that shRNAs were saturating the essential miRNA nuclear export factor Exportin 5, and thus competing for export with endogenous miRNAs. Long-term expression of the hairpins led to a depletion of endogenous liver miRNAs such as miR-122. It should be noted that high doses of adeno-associated virus (AAV) serotype 8 were used in these studies in order to achieve ∼100% transduction of hepatocytes. These results indicate that it would be desirable to design the most potent RNAi triggers possible such that subsaturating doses are efficacious. Regulated and tissue-specific expression also seems prudent. It will be essential to monitor for saturation of the miRNA machinery in preclinical and human clinical trials with RNAi, when possible.

If Exportin 5 is the only bottleneck at which RNAi triggers compete with endogenous miRNAs for the RNAi/miRNA machinery, one would predict that siRNAs (which do not require Exportin 5) would not compete with endogenous miRNAs. However, the Rossi group showed that both exogenous shRNAs and siRNAs competed with miR-21 in cultured cells, suggesting that factors downstream of Exportin 5 can be saturated (Castanotto et al., 2007). Intriguingly, an exogenous miRNA-based RNAi trigger with the same guide strand did not compete with miR-21 (Castanotto et al., 2007). Although this suggests that expression of RNAi triggers in the context of endogenous miRNAs may increase safety, further studies are required to confirm this result.

John and coworkers reported that high doses of siRNAs formulated with lipid nanoparticles transiently mediated efficient silencing of two liver transcripts in mice without interfering with the levels of three liver miRNAs (John et al., 2007). It is important to note that only two time points were examined in this experiment (2 days and 30 days). Silencing was observed only on day 2. In contrast, Grimm and coworkers observed toxicity only at late time points after chronic expression of the shRNA from viral vectors (greater than ∼25 days) (Grimm et al., 2006). John and coworkers conducted further long-term silencing studies with lipid nanoparticle-formulated siRNAs in hamsters (21 days) and observed efficient silencing of the hepatocyte-expressed sterol regulatory element-binding protein (SREBP) transcript without decreases in expression of miR-122. miR-122 is by far the most abundant liver miRNA. It would be interesting to determine whether longer term silencing (>25 days) would affect the levels of miR-122 or the levels of a less abundant liver miRNA. Competition between processing of siRNAs and endogenous miRNAs requires further investigation. Identifying the proper balance between efficacy and toxicity may be required for each individual exogenous RNAi trigger.

Seed sequence of siRNAs is critical for predicting off-target silencing

When siRNAs were first applied as gene-silencing tools, it was initially thought that they were exquisitely specific. Later, more comprehensive studies using microarray analysis indicated that some siRNAs (and presumably some shRNAs) can cause off-target cleavage of mRNAs containing closely related sequences (Jackson et al., 2003; Scacheri et al., 2004). More recently it has been appreciated that siRNAs can cause off-target cleavage of mRNAs with as few as seven nucleotide matches in the 3′ UTR (Lin et al., 2005). A bioinformatics and microarray analysis of off-target cleavage by 12 different siRNAs suggested that siRNAs can function as miRNAs during off-target silencing (Birmingham et al., 2006). Complementarity between the nucleotides 2–7 or 2–8 (of either strand of the siRNA) and sequences in the 3′ UTRs of mRNAs was associated with off-target cleavage. Nucleotides 2–8 correspond to the equivalent of the miRNA seed sequence in the siRNA. Multiple seed matches within the 3′ UTR were even more highly correlated with off-target cleavage. Homology with the 5′ UTR or coding region was not associated with off-target cleavage. Thus, when selecting siRNA sequences, special care should be taken to avoid guide or passenger strand sequences whose “seed sequence” has homology to the 3′ UTR of mRNAs. It seems quite likely that this same paradigm applies to the siRNAs produced by Dicer cleavage of shRNAs. Clearly these results have implications for the selection of therapeutic siRNA or shRNA sequences. Dharmacon/Thermo Fisher Scientific (Lafayette, CO) provides a Web tool with which to search for seed matches in siRNA sequences (see http://www.dharmacon.com/seedlocator/default.aspx). These results also underscore the importance of proper controls in siRNA screens. If a particular phenotype is indeed associated with the knockdown of a particular mRNA, then multiple distinct siRNAs targeting that mRNA should produce the same phenotype. A more rigorous approach is to demonstrate rescue with an “immunized” transgene that contains silent mutations in the siRNA-binding site, thus preventing RNAi silencing. Last, off-target effects are dose dependent (Birmingham et al., 2007), and therefore the minimal dose required to obtain sufficient knockdown should be used.

Thus, because RNAi and miRNA use overlapping machinery, care must be taken when designing RNAi triggers. Maximal potency will reduce the dose required and reduce the possibility of competition with the endogenous miRNA machinery. Care must also be taken to avoid off-target effects due to RNAi triggers acting as miRNAs.

USING miRNAs TO AVOID IMMUNE RESPONSES TO VIRAL VECTORS

Naldini and coworkers described a clever use of miRNAs for gene therapy (Brown et al., 2006). A significant hurdle to the application of gene replacement therapy is the development of transgene-specific immunity (Thomas et al., 2003). For example, clinical trials for the treatment of hemophilia have been hampered by the development of immune responses against viral vectors and transgenes (High, 2005). A major cause of transgene-specific immunity is the expression of transgenes in professional antigen-presenting cells (APCs) (De Geest et al., 2003). Tissue-specific promoters are not sufficient to prevent expression in APCs because they are leaky. In an elegant proof-of-principle article, Brown and coworkers (2006) exploited the fact that the expression of miR-142-3p is enriched in hematopoietic cells (Chen et al., 2004; Baskerville and Bartel, 2005). They placed four sites with perfect complementarity to miR-142-3p in the 3′ UTR of the lentivirus carrying the gene encoding green fluorescent protein (GFP), which under normal circumstances is a potent neoantigen (Stripecke et al., 1999). When they injected mice with a control lentiviral vector lacking the miR-142-3p sites, they observed GFP expression in hepatocytes, endothelial cells, Kupffer cells (liver macrophages), and splenocytes on day 5. However, by day 14, little or no GFP expression was observed, presumably because of immune clearance. In contrast, when they injected mice with a lentivirus carrying the GFP gene and with four miR-142-3p-binding sites in the 3′ UTR, they observed GFP expression in <1% of splenocytes, and then only in the marginal zone, which is not of hematopoietic lineage. Presumably because they avoided host immune responses, with this vector they observed long-term GFP expression (>120 days). Given the above-described report discussing miRNA sponges, one concern is that high-level expression of a transgene containing multiple binding sites for a particular miRNA could compete with endogenous miRNA targets for their cognate miRNA. Interestingly, when the authors performed the transfection with a second reporter containing four miR-142-3p-binding sites, they did not see an increase in GFP expression. This suggests that miR-142-3p was not limiting in this system. Although speculative, this may be because the complementarity of the miRNA to the target was perfect. Thus, one would predict that the miRNA would cleave the target and recycle rather than remain bound to it as is the case for miRNA sponges. In a separate report, Brown and coworkers applied the same strategy to achieve long-term expression of human clotting factor IX in hemophilia B mice (Brown et al., 2007). It seems likely that this strategy could be applied to the long-term expression of other transgenes in vivo.

CONCLUDING REMARKS

There has been an explosive increase in our understanding of the role of miRNAs in normal gene regulation and in human disease. Gene therapists are in a unique position to rapidly apply approaches developed for antisense, ribozymes, and gene transfer to miRNAs. It seems likely that miRNA expression levels can be used as novel diagnostic markers. Manipulation of miRNA levels could also have therapeutic benefit. Last, by exploiting our emerging understanding of miRNA biology, gene therapists may be able to design safer and more effective RNAi therapeutics as well as safer gene transfer vectors.