Search tips
Search criteria 


Logo of plosonePLoS OneView this ArticleSubmit to PLoSGet E-mail AlertsContact UsPublic Library of Science (PLoS)
PLoS One. 2010; 5(10): e13119.
Published online 2010 October 1. doi:  10.1371/journal.pone.0013119
PMCID: PMC2948504

RNAi Experiments in D. melanogaster: Solutions to the Overlooked Problem of Off-Targets Shared by Independent dsRNAs

Michael Polymenis, Editor



RNAi technology is widely used to downregulate specific gene products. Investigating the phenotype induced by downregulation of gene products provides essential information about the function of the specific gene of interest. When RNAi is applied in Drosophila melanogaster or Caenorhabditis elegans, often large dsRNAs are used. One of the drawbacks of RNAi technology is that unwanted gene products with sequence similarity to the gene of interest can be down regulated too. To verify the outcome of an RNAi experiment and to avoid these unwanted off-target effects, an additional non-overlapping dsRNA can be used to down-regulate the same gene. However it has never been tested whether this approach is sufficient to reduce the risk of off-targets.


We created a novel tool to analyse the occurance of off-target effects in Drosophila and we analyzed 99 randomly chosen genes.

Principal Findings

Here we show that nearly all genes contain non-overlapping internal sequences that do show overlap in a common off-target gene.


Based on our in silico findings, off-target effects should not be ignored and our presented on-line tool enables the identification of two RNA interference constructs, free of overlapping off-targets, from any gene of interest.


Genes can be silenced using RNA interference (RNAi). This powerful method is widely used to study biological consequences induced by the down-regulation of selected genes [1], [2], [3], [4]. Since its discovery, a great amount of valuable information has been collected using this technology. However, RNAi technology also has some drawbacks such as off-target effects [5], [6], [7], [8], [9], [10], [11], [12]. Off-target effects are caused by short stretches of sequence similarity between the RNAi molecule and one or more genes other than the target. Because of high success rates, the fly and worm (D. melanogaster and C. elegans) model systems generally use large double strand RNAs (dsRNAs) of 300–800 bp. From (large) dsRNAs, numerous siRNAs are generated by the action of DICER and each of these can provoke an RNAi response and exert their gene down-regulating action [13]. Although this results in a favourable synergistic RNAi response towards the target gene, it may in theory also increase the number of off-target possibilities.

A straightforward method to reduce off-target effects, is to use 2 independent and non-overlapping dsRNAs to down-regulate a specific target. Because these dsRNAs are different in sequence composition, their individual off-targets are also assumed to be unique while they both silence the same on-target gene. Consequently, it is reasonable to assume that any shared phenotype which is observed after the independent use of both dsRNAs is an effect of down-regulating the on-target gene (ure 1A). Although, this line of reasoning is rational, hypothetically it is possible that different non-overlapping siRNAs may actually target different sequences within one identical off-target gene (illustrated in Figure 1B). In such an unfortunate case, a shared off-target effect induced by 2 independent dsRNAs may be misinterpreted as an on-target effect. It has never been investigated what the occurrences are of shared off-target effects when dsRNA are randomly chosen. Here, we present a detailed analysis, based on sequence similarity and a randomized trial which suggest that most genes have independent dsRNA-spanning sequences showing sequence similarity with the same off-target gene. In addition, we present an on-line tool that allows to scan Drosophila gene sequences for the occurrence of off-target overlapping regions and to design dsRNAs that have a reduced likelihood to induce identical off-target effects.

Figure 1
Hypothetical off-target events.

Results and Discussion

Statistical analysis on a randomized genome shows that it is likely that 2 distinct 21 nt sequences from the same gene can map closely elsewhere on the genome (see Supporting Information S1). This hypothetical event (illustrated in Figure 1B) may cause distinct dsRNAs to have common off-targets and that particular combinations of dsRNAs should therefore be avoided. These calculations are based on a non-organized genome containing random sequences, while the Drosophila genome is highly functional and far from ‘randomized’. To evaluate the risks of our hypothetical event more pragmatically, we used the following approach. First, we picked a dataset of 99 random chosen genes (see Supporting Information S1) from the D. melanogaster genome. We investigated the occurrence of independent dsRNAs derived from one gene to have shared off-targets. dsRNAs are often derived from cDNA so for our analysis only the cDNA of the 99 genes were considered. Because the complete cDNA can be used to design dsRNAs from, and the dsRNAs are split into siRNAs of approximately 21nt by the RNAi machinery, we first created a list of all possible siRNA sequences that can be obtained from the cDNA sequences of each of these 99 genes. This complete list was subsequently reduced using established scoring rules to exclude 21-bp siRNAs that are most likely non-active (see Supporting Information S1). We like to stress that this assumption will only underestimate our findings. Next we calculated the occurrence of all siRNA derived from one cDNA to have a shared off-target with another siRNA derived from the same gene. For this analysis we included pre-mRNA sequences of the complete D. melanogaster genome because of the following published results: 1) It has been demonstrated that the RNAi machinery can target pre-mRNAs in C. elegans [14]. 2) RNA silencing components have been shown to localize in the nucleus in other organisms (including human) [15], [16], [17], [18], [19], [20], [21], [22], further suggesting that pre-mRNAs can be targetted by the RNAi machinery. 3) RNAi constructs can be complementary to miRNAs which are often derived from introns [23] and might act like antagomirs [24]. We therefore analysed the filtered list of siRNA sequences against both mature and pre-mature RNA sequences to map all possible off-targets with up to 3 mismatches in their sequence alignments with the use of a new tool (see Methods and By doing so, a list of potential off-targets for each of the individual genes was generated. Next we analyzed whether there was overlap between the potential off-targets of siRNAs derived from the same gene (see Methods). We used the term cot-group (Common Off-Target group); a cot-group consists of 2 or more siRNAs, derived from a single gene, that map to the same off-target gene (also illustrated in Figure 2; the lines represent members of cot-groups) . The generated siRNA lists of all 99 genes were scanned individually for the presence of cot-groups. The occurrence of cot-groups appeared to be present in all genes, with sometimes excessive high frequencies (Figure 3). As expected, the number of cot-groups are highly correlated with the length of the cDNA of the gene; the larger the sequence, the more cot-groups are formed (Figure 4).

Figure 2
Non-overlapping cDNA sequences of randomly chosen genes have a high prevalence of sharing off-targets in pre-mature sequences of other genes.
Figure 3
Common Off-Target (COT) groups are present in all 99 analized genes.
Figure 4
The number of cot-groups correlate with the length of the cDNA.

We then looked at the general profile of the cot-groups for each gene separately and tried to deduce the required number of dsRNAs to strongly reduce the event of common off-targets. If for example the number of members within the cot-groups does not exceed 2, than this implies that at most 2 siRNAs within the same gene map to a common off-target. For that particular situation, the use of 3 or more non-overlapping dsRNA will always generate bona fide data as there is no possibility for all 3 of them to share a common off-target (bases on sequence similarity). Unfortunately, most genes have cot-groups with at least 3 members (Figure 5, also depicted by red lines in Figure 2 and in Supporting Information S1) or even 4 members (Figure 5, also depicted by blue lines in Figure 2 and in Supporting Information S1). This finding demonstrates that just using multiple non-overlapping dsRNAs is not sufficient to exclude off-target events (see also insert Figure 2), even if the number of independent dsRNAs is 3 or more. We therefore developed a bioinformatics approach to design dsRNAs that do not contain predicted off-targets. Our freely available website presents such a tool at This web based tool accepts a gene name as input and presents a number of choices each containing a combination of 2 unique dsRNAs that lack overlapping –based on sequence similarity- off-targets (Supporting Information S1).

Figure 5
Most genes have cot-groups with at least 2 members.

Next, we repeated the above analysis, but now only considering off-targets targeting mature RNA sequences, because these are maybe be more active in RNAi [25]. Filtering out the intron off-targets, causes much less off-targets to be found in general per cDNA (Figure 6). Overall, both the sizes and occurrences of the COT groups are smaller (Figure 3B, Figure 5B). Nevertheless, there is still a significant number of overlapping potential off-targets to be expected in >74% of the genes. In 24% of the analysed genes there is at least 1 COT groups present of size 3 (Figure 5B), meaning that there are at least 3 areas within the cDNA that target the same off-target. Therefore, even if only mature RNA sequences are considered and 2 randomly chosen non-overlapping dsRNA's are used, the experimental outcome can be obscured by off-target effects. This further underscores the utility of our tool.

Figure 6
Non-overlapping cDNA sequences of randomly chosen genes have a significant prevalence of sharing off-targets in exon derived sequences of other genes.

Although not exclusively, we observed a strong tendency for overlapping off-targets to occur at the end of genes (see graphical illustrations of the common off-targets in Figure 2 and in Supporting Information S1), corresponding to the untranslated regions (UTR). The UTR sequences are less unique in the genome as compared to the coding region and therefore preferably should be avoided when dsRNA constructs are designed. Our tool includes an option to avoid UTR sequences to minimize off-targets when designing dsRNAs of interest.

Our analysis demonstrated that most genes in the D. melanogaster genome contain 2 or more (distinct) sequences that show sequence similarity (containing 3 or less mismatches) to the same off-target gene. The potential consequence of these overlapping occurrences is that 2 dsRNAs which are generated to down-regulate a specific target gene, may possess a common off-target gene as well. In case these 2 distinct dsRNAs are used, their common phenotype induced by down-regulation of their shared off-target gene may lead to misinterpretation of the experiment. We present a method to identify 2 distinct dsRNAs from a gene of choice that do not show any off-target overlap, -based on sequence similarity- by performing a thorough off-target overlap analysis. This tool is freely available at and can be used by the Drosophila community where dsRNAs are generally used for gene down-regulation.


Genomic data (build 45–43b for Drosophila) were downloaded from the Ensembl website ( The data from Ensemble and its derived seed tables were processed, stored and indexed in a MySQL database, version 5.0, running on top of Ubuntu 6.06. The on-line available RNAiSelect program ( was written in C#.NET and performs a comprehensive sequence alignment against the input genome for up to 3 mismatches.

The RNAiSelect algorithm was specifically designed for finding relationships between short nucleotide sequences. It has a high performance and usability for short-sequence studies, including siRNA (off-)targets. The complete source code and documentation for a standalone version of this algorithm may be downloaded from The algorithm is based on the following assumption:

“An example sequence TTTTAATTTGGGCCCGGG consists of 18 nucleotides and may be split into two 9-nt child sequences; TTTTAATTT and GGGCCCGGG. By plain observation, we know that the sequence GGGCCCGGG is exactly 9-nt separated from TTTTAATTT in the original sequence.”

For the RNAiSelect algorithm to work, we first wrote a program that generates a seed table which holds the exact genomic location(s) for every possible 9-nt sequence (4∧9, or 262.144 sequences). Generating such an index is a general strategy used by many algorithms to rapidly look-up any sequence of fixed length for its positions in the genome. The used algorithm however uses a novel method to calculate the positional relationship between indexed seeds, instead of performing string-to-string comparisons for every nucleotide after a hit has been found. In other words, by searching 9-nt subsequences of the whole query sequence for consecutive matches of locations, it will find hits larger than 9 nt without performing actual DNA comparisons. This following example, in layman code, shows how to find an 18-nt sequence in the genome by first splitting the sequence into its two 9-nt subsequences and comparing these sequences with the available index table with a word size of 9.

  • 1 SPLIT QUERY SEQUENCE(18 nt) INTO dnacode_left(9 nt) AND dnacode_right(9 nt)
  • 2 EXTRACT LOCATIONS FROM index_table FOR dnacode_left AND STORE IN seedtable_left
  • 3 EXTRACT LOCATIONS FROM index_table FOR dnacode_ right AND STORE IN seedtable_ right
  • 4 SELECT ALL HITS WHERE (LOCATIONS seedtable_left + 9) EQUALS (LOCATIONS seedtable_right)

This example merely demonstrates how to find an exact 18-nt hit not allowing any mismatches. However, mismatches may be added by expanding the seed searches with variations so that all possible combinations will be found. We thus included variations of the 9-nt sub-sequences and then compared the distance relationship between the original locations of the seed hits, which has to be exactly 9. This may considerably increase the number of seed searches, but because these are relatively cheap in terms of processing time, the overall performance is very high while it guarantees that every possible alignment is evaluated.

The cDNA sequences (meaning only sequences derived from the mature RNA) from each of the 99 genes (Supporting Information S1) were first analyzed for potentially active sequences as might be produced by endogenous DICER. A scoring schema was used (Supporting Information S1) during this analysis to estimate and extract the most potential sequences. Each extracted sequence was analyzed for potential off-targets. The combined output from all these potential off-targets was cross-referenced with each other to map areas on the original cDNA sequence that are predicted to have overlapping off-targets. At the same time, regions can be identified that lack these areas and dsRNA sequences can be extracted that are completely devoid of off-targets. The results are presented and 2 or more dsRNA are indicated that originate from the same gene and that are predicted to lack overlapping off-targets. From the indicated areas, dsRNA can be designed. An identical analysis can be done for every Drosophila gene of interest through a web-interface at which presents the output in a user friendly interface.

Supporting Information

Supporting Information S1

Statistical analysis on a randomized genome shows that it is likely that distinct sequences from the same gene map closely elswhere on the genome, demonstrating that non-overlapping dsRNA constructs derived from the same gene can share a common off-target. 99 randomly-chosen genes were analysed for these -based on sequence similarity- off-target effects using defined scoring rules. Our results show that non-overlapping cDNA sequences of randomly chosen genes have a high prevalance of sharing off-targets in pre-mature and in exon sequences of other genes. A freely available website is provided to avoid these off-target effects and to identify “clean” combinations of dsRNA constructs derived from the same gene to down regulate any gene of interest.

(6.68 MB DOC)


Competing Interests: The authors have declared that no competing interests exist.

Funding: This work was supported by a VIDI grant from the Netherlands Organization for Scientific Research (NWO; 971-36-400) to O.C.M.S. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.


1. Moazed D. Small RNAs in transcriptional gene silencing and genome defence. Nature. 2009;457:413–420. [PubMed]
2. Castanotto D, Rossi JJ. The promises and pitfalls of RNA-interference-based therapeutics. Nature. 2009;457:426–433. [PMC free article] [PubMed]
3. Dietzl G, Chen D, Schnorrer F, Su K-C, Barinova Y, et al. A genome-wide transgenic RNAi library for conditional gene inactivation in Drosophila. Nature. 2007;448:151–156. [PubMed]
4. Morris K, Simon WLC, Jacobsen S, Looney D. Small Interfering RNA-Induced Transcriptional Gene Silencing in Human Cells. Science. 2004;305:1289–1292. [PubMed]
5. Qiu S, Adema C, Lane T. A computational study of off-target effects of RNA interference. Nucleic Acids Res. 2005;33:1834–1847. [PMC free article] [PubMed]
6. Kulkarni M, Booker M, Silver S, Friedman A, Hong P, et al. Evidence of off-target effects associated with long dsRNAs in Drosophila melanogaster cell-based assays. Nat Methods. 2006;3:833–838. [PubMed]
7. Moffat J, Reiling JH, Sabatini DM. Off-target effects associated with long dsRNAs in Drosophila RNAi screens. Trends in pharmacological sciences. 2007;28:149–151. [PubMed]
8. Fedorov Y, Anderson EM, Birmingham A, Reynolds A, Karpilow J, et al. Off-target effects by siRNA can induce toxic phenotype. RNA. 2006;12:1188–1196. [PubMed]
9. Ma Y, Creanga A, Lum L, Beachy P. Prevalence of off-target effects in Drosophila RNA interference screens. Nature. 2006;443:359–363. [PubMed]
10. Doench JG, Petersen CP, Sharp PA. siRNAs can function as miRNAs. Genes Dev. 2003;17:438–442. [PubMed]
11. Saxena S, Jonsson Z, Dutta A. Small RNAs with imperfect match to endogenous mRNA repress translation. Implications for off-target activity of small inhibitory RNA in mammalian cells. The Journal of biological chemistry. 2003;278:44312–44319. [PubMed]
12. Jackson A, Burchard J, Schelter J, Chau B, Cleary M, et al. Widespread siRNA “off-target” transcript silencing mediated by seed region sequence complementarity. RNA. 2006;12:1179–1187. [PubMed]
13. Hannon GJ. RNA interference. Nature. 2002;418:244–251. [PubMed]
14. Bosher J, Dufourcq P, Sookhareea S, Labouesse M. RNA Interference Can Target Pre-mRNA: Consequences for Gene Expression in a Caenorhabditis elegans Operon. Genetics. 1999;153:1245–1256. [PubMed]
15. Pal-Bhadra M, Bhadra U, Birchler JA. RNAi Related Mechanisms Affect Both Transcriptional and Posttranscriptional Transgene Silencing in Drosophila. Molecular Cell. 2002;9:315–327. [PubMed]
16. Verdel A, Jia S, Gerber S, Sugiyama T, Gygi S, et al. RNAi-Mediated Targeting of Heterochromatin by the RITS Complex. Science. 2004;303:672–676. [PubMed]
17. Langlois M-A, Boniface C, Wang G, Alluin J, Salvaterra P, et al. Cytoplasmic and Nuclear Retained DMPK mRNAs Are Targets for RNA Interference in Myotonic Dystrophy Cells. J Biol Chem. 2005;280:16949–16954. [PubMed]
18. Matzke MA, Birchler JA. RNAi-mediated pathways in the nucleus. Nat Rev Genet. 2005;6:24–35. [PubMed]
19. Robb GB, Brown KM, Khurana J, Rana TM. Specific and potent RNAi in the nucleus of human cells. Nat Struct Mol Biol. 2005;12:133–137. [PubMed]
20. Weinberg MS, Barichievy S, Schaffer L, Han J, Morris KV. An RNA targeted to the HIV-1 LTR promoter modulates indiscriminate off-target gene activation. Nucleic Acids Res. 2007;35:7303–7312. [PMC free article] [PubMed]
21. Lin S-L, Kim H, Ying S-Y. Intron-mediated RNA interference and microRNA (miRNA). Frontiers in bioscience : a journal and virtual library. 2008;13:2216–2230. [PubMed]
22. Politz JC, Hogan EM, Pederson T. MicroRNAs with a nucleolar location. Rna. 2009;15:1705–1715. [PubMed]
23. Lin SL, Miller JD, Ying SY. Intronic MicroRNA (miRNA). J Biomed Biotechnol. 2006;2006:26818. [PMC free article] [PubMed]
24. Krutzfeldt J, Rajewsky N, Braich R, Rajeev KG, Tuschl T, et al. Silencing of microRNAs in vivo with ‘antagomirs’. Nature. 2005;438:685–689. [PubMed]
25. Elbashir SM, Lendeckel W, Tuschl T. RNA interference is mediated by 21- and 22-nucleotide RNAs. Genes & development. 2001;15:188–200. [PubMed]

Articles from PLoS ONE are provided here courtesy of Public Library of Science