Copyright © 2006 Taxman et al; licensee BioMed Central Ltd.
Criteria for effective design, construction, and gene knockdown by shRNA vectors
Debra J Taxman: Debra_Taxman/at/med.UNC.edu; Laura R Livingstone: LRL/at/med.UNC.edu; Jinghua Zhang: JHZ2580/at/yahoo.com; Brian J Conti: Brian_Conti/at/med.UNC.edu; Heather A Iocca: Heather_Iocca/at/med.UNC.edu; Kristi L Williams: Kristi_Williams/at/med.UNC.edu; John D Lich: John_Lich/at/med.UNC.edu; Jenny P-Y Ting: Jenny_Ting/at/med.UNC.edu; William Reed: William_Reed/at/med.UNC.edu
Received June 20, 2005; Accepted January 24, 2006.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
BackgroundRNA interference (RNAi) technology is a powerful methodology recently developed for the specific knockdown of targeted genes. RNAi is most commonly achieved either transiently by transfection of small interfering (si) RNA oligonucleotides, or stably using short hairpin (sh) RNA expressed from a DNA vector or virus. Much controversy has surrounded the development of rules for the design of effective siRNA oligonucleotides; and whether these rules apply to shRNA is not well characterized.
ResultsTo determine whether published algorithms for siRNA oligonucleotide design apply to shRNA, we constructed 27 shRNAs from 11 human genes expressed stably using retroviral vectors. We demonstrate an efficient method for preparing wild-type and mutant control shRNA vectors simultaneously using oligonucleotide hybrids. We show that sequencing through shRNA vectors can be problematic due to the intrinsic secondary structure of the hairpin, and we determine a strategy for effective sequencing by using a combination of modified BigDye chemistries and DNA relaxing agents. The efficacy of knockdown for the 27 shRNA vectors was evaluated against six published algorithms for siRNA oligonucleotide design. Our results show that none of the scoring algorithms can explain a significant percentage of variance in shRNA knockdown efficacy as assessed by linear regression analysis or ROC curve analysis. Application of a modification based on the stability of the 6 central bases of each shRNA provides fair-to-good predictions of knockdown efficacy for three of the algorithms. Analysis of an independent set of data from 38 shRNAs pooled from previous publications confirms these findings.
ConclusionThe use of mixed oligonucleotide pairs provides a time and cost efficient method of producing wild type and mutant control shRNA vectors. The addition to sequencing reactions of a combination of mixed dITP/dGTP chemistries and DNA relaxing agents enables read through the intrinsic secondary structure of problematic shRNA vectors. Six published algorithms for siRNA oligonucleotide design that were tested in this study show little or no efficacy at predicting shRNA knockdown outcome. However, application of a modification based on the central shRNA stability should provide a useful improvement to the design of effective shRNA vectors.
RNA interference (RNAi) is a naturally occurring phenomenon by which RNA duplexes known as short interfering RNA (siRNA) can reduce gene expression through enzymatic cleavage of a target mRNA mediated by the RNA-induced silencing complex (RISC). The ability of synthetic siRNA to inhibit targeted genes with near specificity makes it an extremely powerful tool for functional genomics that has drawn considerable interest recently [1,2]. RNAi is commonly achieved by introducing chemically synthesized siRNA 19–22 mers into cells by transfection. However, many cells and cell lines are either refractory to or adversely affected by transfection, and the transient nature of this methodology renders it unsuitable for the generation of long-term cell lines of the desirable phenotype. Two alternatives to synthetic siRNA are DNA-vector mediated RNAi production [3-5], and most recently viral-mediated siRNA synthesis [6-10]. For the latter technologies, sense and antisense strands can be expressed from different promoters . Alternatively, short hairpin (sh) RNAs, expressed from a single promoter, are processed into siRNAs by Dicer or a homologous double strand RNase .
One caveat of siRNA design is that not all 19–22 base RNA duplexes will cleave their target with efficacy, and much effort has gone towards identifying a set of rules for selecting an effective siRNA target site within a gene. Recent findings [13,14] offered the first clue towards the development of guidelines for selecting an siRNA target site. These studies showed that the RISC complex is asymmetric and favors the strand of the siRNA duplex with the least thermodynamically stable 5' terminus. Subsequently, Reynolds et al. designed an algorithm based on statistical data showing patterns of efficacy for siRNA oligonucleotides containing specific residues at defined positions within the 19-mer . A limitation of their study is that a small number of genes were tested. Several additional algorithms for designing effective siRNAs have been published since those initial reports with surprisingly disparate results, making the determination of which residues are generally favorable for siRNA efficacy a point of controversy [16-20]. Additionally, whether any of the algorithms developed for synthetic siRNA oligonucleotides apply to the design of shRNA expressed stably from a vector has not been well explored.
In the present report, we construct and analyze a set of 27 shRNAs for 11 different human genes. To our knowledge this is the largest individual set of data published for shRNA 19-mers. We describe a method for simultaneously preparing wild type and control mutant shRNA vectors that is time and cost efficient, and show that sequencing of shRNA plasmids can be quite problematic due to the intrinsic secondary structure of the hairpin. We examine several different strategies for overcoming this problem including the use of modified BigDye chemistries and the addition of agents known to relax DNA structure. The knockdown efficacy for each of the 27 shRNAs was evaluated against six published algorithms for siRNA oligonucleotide design by linear regression and ROC curve analyses. We describe a modification of three of the algorithms that provides fair-to-good prediction of shRNA efficacy, and confirm the significance of the modified algorithms using a pooled set of shRNAs from previous publications. These findings should be of general applicability in the design and construction of shRNA vectors.
Results and discussion
Design and preparation of shRNA plasmidsTo address the question of how shRNA sequence correlates with knockdown efficacy, 27 shRNA vectors from 11 different genes were designed and constructed (Table 1). Target sequences were selected in the coding region of each gene and were designed to broadly conform to the seminal studies of sequence features for siRNA oligomer efficacy [13-15]. Accordingly, sequences are low in runs and have a G/C ratio of about 50%. The shRNAs were designed to target sites that are devoid of single nucleotide polymorphisms, and correspond to all splice variants amplified by our real time PCR primer sets.
Since siRNAs can have off-target effects, it is important for functional assays to make a specific mutant with one or more base mismatch within the target recognition site as a control . To conserve time and cost, we have developed a method of making wild-type and mutant shRNA vectors simultaneously (detailed in Methods and Figure 1). Gene knockdown results for four wild-type/mutant shRNA pairs are shown in Figure 2. These results demonstrate the utility of this method in providing a point mutant shRNA vector that can serve as a loss-of-function control for gene knockdown by wild type shRNAs. Though detailed protocols have been published for construction of shRNA vectors , this is the first protocol for producing wild-type and mutant vectors simultaneously and should facilitate the implementation of highly controlled system for shRNA.
Strategy for accurate sequencing through hairpin structuresVerifying the sequence of an shRNA hairpin is essential since mismatch of even one nucleotide within the target sequence can ablate knockdown (Figure 2 and [5,23].) An issue that is frequently encountered in the preparation of shRNA vectors is that many are difficult to sequence due to the intrinsic secondary structure of the hairpin. One strategy recently proposed to overcome this issue involves engineering a restriction site within the loop/stem region of the hairpin to physically separate the inverted repeats by digestion, and then piecing together sequence using sense and antisense primers . However, the ability to achieve sequencing of shRNA constructs without modifying stem/loop sequence would be of clear advantage. To address this possibility, we evaluated modified sequencing reactions for improvement in the read-through of the hairpin secondary structure in three shRNA hairpins. Modifications include adding agents known to relax DNA structure including DMSO, Betaine, PCRx Enhancer and ThermoFidelase I; and adding increasing amounts of dGTP BigDye terminator (dGTP) chemistry to the standard BigDye v1.1 (BD) chemistry which contains dITP rather than dGTP.
Sequencing results for each of the three DNA constructs are summarized in Table 2. Read-through of the hairpin structure was measured as the ratio of the peak height about 300 bases after the hairpin structure to the signal about 50 bases before the hairpin structure. A ratio of 1 indicates no loss in signal and 0 indicates complete loss of read-though. In the absence of any additive to BD chemistry, the hairpin caused a reduction in peak height ratio for our less tightly structured hairpin, pHSPG-shmutTLR4, to 0.4, and a complete loss in read through for the other two plasmids. This can be visualized as an abrupt stop in the sequence peak profile for pHSPG-shTLR4 (Figure 3A).
Among the DNA relaxing agents, 5% DMSO, 0.83 M Betaine and 1 × PCRx Enhancer each improved the sequence read significantly for some constructs. However, the addition of 0.83 M Betaine plus 1 × PCRx Enhancer to BD chemistry was found to sequence most consistently, with peak height ratios of 0.5–0.9 (Table 2 and Figure 3B). The addition of 10:1 BD:dGTP chemistries alone also improved read through somewhat, with peak height ratios of 0.5–0.6 (Table 2 and Figure 3C). The sub-optimal peak height ratio for 10:1 BD:dGTP can be attributed to a visible step in the sequence peak profile after the secondary structure region where the signal is reduced (Figure 3C, arrow). Increasing the dGTP chemistry content to 5:1 and 3:1 BD:dGTP or using straight dGTP chemistry increased the peak height ratio and reduced the step somewhat (0.6 to 0.8 ratio). However, the mixed incorporation of dITP and dGTP resulted in worse peak broadening as the amount of dGTP used increased [see Additional file 1], and dGTP only chemistry caused severe sequence compressions (data not shown). The best overall results were observed by combining Betaine plus PCRx and 10:1 BD:dGTP mixed chemistries together. This combination reduced the step with less peak broadening and increased peak height ratios to 0.9–1.0 (Table 2 and Figure 3D). ThermoFidelase I, a DNA destabililizing enzyme that is frequently used to improve sequencing of genomic DNA [25,26], did not improve sequencing of any of the three hairpins in straight BD chemistry (data not shown), and actually reduced the peak height ratio significantly in 10:1 BD:dGTP chemistries for all three shRNA constructs, causing the reappearance of a stop at the hairpin structure (Table 2 and Figure 3E).
In summary, the combination of 10:1 BD:GTP chemistries, 0.83 M Betaine, and 1 × PCRx Enhancer provided optimal sequencing, and mixed BD:dGTP chemistries, Betaine, PCRx Enhancer, and DMSO each had some positive effects on their own. ThermoFidelase I, however, probably should be avoided for shRNA vectors with difficult intrinsic secondary structure.
Correlation between shRNA knockdown efficiency and published algorithms for siRNA designTo determine whether the efficacy of knockdown by shRNA vectors correlates with published rules for the design of effective siRNA oligonucleotides, shRNAs were evaluated for their ability to knockdown gene expression. The shRNAs were transduced stably into either THP1 or Jurkat human cell lines as detailed in Table 3, first two Columns. The average knockdown was determined from RNA collected on three or more different days and is listed for each shRNA (Column 3). Knockdown was shown to be reproducible for cell lines that were independently transduced and sorted, suggesting that knockdown is a function of the shRNA target sequence rather than features of the viral transduction [see Additional file 2]. More than one third of the shRNA vectors constructed were unable to suppress transcription (<10% in Column 3), despite comparable growth rates and long term expression of the GFP marker at high levels in these cell lines. Furthermore, grea