There are few examples of successful shRNA expression using pol II promoters. Although several requirements for pol II shRNA expression have been established, optimization and direct assessment of different hairpin stem lengths, TSS and TTS have not been performed. In the current study we evaluated the relevance of these factors by positioning shRNAs targeting luciferase or ApoB at +5 or +6 relative to the CMV TSS and added the T5, pA or U1 TTS at the 3’ end of the hairpins. Additionally, the same siRNA sequences were incorporated in artificial miRNA scaffolds and direct evaluation between the silencing efficacy and processing of pol II-expressed shRNA and miRNA was performed.
The nucleotide distance between the promoter, the shRNA structure and the TTS determines the length and sequence composition of the 5’ and 3’ overhangs of the expressed hairpin. Both ends act as reference points for recognition and proper processing by Dicer [17
]. Initial rules published by Xia et al.
include shRNA location within 6 nt from the TSS and pA TTS [11
]. Later shRNA containing T5 and U1 TTS where also found to be active when substituted for pA [13
]. These settings were used in our study to express shRNA from the CMV promoter. The shLuc, shApoB1 and shApoB2 sequences have been previously validated as potent inhibitors of their targets when expressed from the pol III H1 promoter [15
]. Surprisingly, all CMV-shLuc constructs induced only a mild target inhibition when the hairpin was placed at +6 TSS and T5 or pA TTS were used. In contrast to previously published data, the use of U1 TTS did not improve silencing efficacy. By shifting the hairpin from +6 to +5 from TSS, thus minimizing the 5’ overhang, no significant improvement was achieved and the effect was not dependent of TTS used. Additionally, the CMV-shApoB1 and CMV-shApoB2 hairpins were ineffective when the initial settings of +6 TSS and pA TTS were used, indicating that those settings cannot be used as a general rule for pol II shRNA expression.
The length of the shRNA hairpin stem has been shown to play an important role in silencing efficacy as it can lead to differential processing into multiple siRNAs [8
]. In the current study we optimized the pol II shRNA expression by testing stem lengths of 20, 21, 25, and 29
bp. Slight increase in silencing efficacy was observed only when extending hairpin stem from 20 to 21
bp but not when the stem length was further extended to 25 or 29
bp. Mcintyre et al.
performed a similar study by extending the shRNA stem length from 16 to 41
bp and looked at the core placement of the shRNA [23
]. While the processing of hairpins was clearly dependent on the stem length, the activity was primarily dependent on the sequence of processed products. Consistent with their data, our results suggest that there is no strict correlation between the increase in stem length and better silencing activity of the shRNAs. Here, we focused only on pol II expression of shRNA but an alternative approach would be to express long hairpin RNA (lhRNA) where the stem length is extended up to 300
]. Using lhRNA allows generating multiple siRNA from a single transcript, which can be used for viral infections or cancer, where multiple sequences have to be targeted. To date, expression of active lhRNA from pol II promoter has not been successful due to inefficient processing of the hairpins by the RNAi machinery, and detailed studies on TSS, TTS and hairpin location, similar to those for shRNA are lacking.
An alternative to optimizing pol II expression of shRNA is to use pre-miRNA scaffolds and replace the mature miRNA sequence with siRNA targeting a gene of interest. Cellular miRNAs are naturally expressed from pol II promoters [1
]. Therefore, when shLuc, shApoB1 and shApoB2 were expressed from the pri-mir-155 scaffold the knockdown efficacy was significantly improved. Importantly, artificial miRNA have been found to be less toxic in vitro
and in vivo
compared to shRNA as they do not disrupt the endogenous RNAi pathway. McBride et al.
used this approach to target huntingtin mRNA and to avoid neurotoxicity caused by overexpression of shRNA [14
]. Similarly, incorporation of a siRNA against spinocerebellar ataxia 1 protein into an artificial miRNA abolished neuronal cell death observed with a shRNA harboring the same siRNA sequence [6
]. In a previous study we have provided evidence for the efficacy of liver-specific expression of artificial miRNA targeting ApoB in vivo
and demonstrated its advantages over the pol III-expressed shRNA (Maczuga et al.
, manuscript submitted). In conclusion, we and others have shown that pol II expression of artificial miRNA scaffolds is a more robust and promising approach than pol II shRNA expression, probably due to the specific structural characteristics of the miRNA.
An additional advantage of the use of artificial miRNAs is that several precursors can be expressed as clusters from longer transcripts, allowing simultaneous targeting of multiple genes [28
]. This feature of the miRNA is very important when mutation-prone viruses, such as HIV-1 or HCV, are targeted or when several disease-related genes need to be simultaneously silenced. Moreover, miRNAs can be linked to the 3’ untranslated region of a therapeutic gene, which is an additional advantage for therapeutic applications since it allows co-delivery of a codon-optimized gene together with a miRNA targeting the disease-causing version of the same gene. Such combinational therapy has been shown for alpha1-antitrypsin (AAT) deficiency, in which a hairpin RNA targeting mutated AAT transcript was delivered together with codon-optimized AAT gene [29
Quantification of the siRNA molecules expressed from the H1-shRNA, CMV-shRNA or CMV-miRNA constructs was highly specific and revealed differences in the amount of processed siRNA molecules per cell. As expected, highest siRNA amounts were detected when the shRNAs were expressed from the strong H1 promoter. However, when the same shRNAs were expressed from the CMV promoter, 12- to 125-fold less siRNA molecules were detected indicating that either the shRNA was not efficiently transcribed from the CMV promoter or that there was impaired siRNA processing by the RNAi machinery. Surprisingly, the amount of processed siRNA from the CMV-miRNA constructs did not correlate with efficacy. For example, CMV-miLuc yielded 1591 siLuc molecules per cell and was completely ineffective in target knockdown while CMV-miApoB2 yielded only 56 siApoB2 molecules per cell and was highly efficient. A possible explanation for this discrepancy is that the small RNA TaqMan assay detects only one variant of the guide strand and although identical predicted siRNA sequences were incorporated in the CMV-shRNA and CMV-miRNA scaffolds, they can still be processed differently. Indeed, our newest data based on Next Generation Sequencing (NGS) of small RNAs from cells transfected with H1-shApoB2 and CMV-miApoB2 indicates differential processing and different mature sequences from the two scaffolds (Maczuga et al., manuscript submitted). The siRNAs originating from the H1-shRNA were more heterogeneous in cleavage sites and length compared to the products originating from the CMV-miRNA scaffold supporting the notion that Dicer cleavage is less precise than the combination of Drosha and Dicer. Unfortunately, NGS data from pol II-expressed shRNA, which would allow verifying the processed siRNA variants and their abundance, are still not available.