We have several lines of evidence providing proof of principle that the RNAi machinery can act on L1s and limit their activity in a transient retrotransposition assay. First, long dsRNA transcribed in vitro from an RC-L1 clone is efficiently processed into functional siRNA by the RNase III enzyme DICER. Second, both synthetic and in vitro ‘diced’ L1 siRNA inhibit the expression of an L1 hybrid transcript by RNAi-mediated degradation. Third, siRNA targeting different regions of the L1 can limit the activity of an RC-L1 in a cultured cell retrotransposition assay.
We initially decided to target the 5′-UTR of the L1 retrotransposon because this region is presumably part of most, if not all, remaining RC-L1s that populate the human genome. In addition, the L1 5′-UTR contains both a sense promoter, as well as an antisense promoter (ASP), and thus might play a larger role in the transcriptional regulation (43
) through the production of L1 dsRNA. Furthermore, because the 5′-UTR contains the L1s internal promoter with transcription beginning at or near the first nucleotide of the L1 at an unconventional start site (5′-GGGGG-3′), we reasoned that it would be easy to test the efficacy of siRNAs targeting the 5′-UTR by creating a fusion construct with a measurable reporter gene such as FF luciferase (38
). Indeed, we found that the 5′-UTR-driven reporter construct achieved high levels of luciferase expression in both HCT116 and HeLa cells, consistent with the data from other studies examining the promoter activity of the L1 5′-UTR, suggesting that the factors necessary for efficient transcription are present in many somatic cells types (38
). Even with the robust expression of the 5′-UTR driven constructs, we observed a sharp decrease in the luciferase activity from the hybrid transcript in the presence of either in vitro
‘diced’ or synthetic siRNA targeting the 5′-UTR. The decrease in the luciferase activity was not due to non-specific targeting of the luciferase coding region, as a control construct expressing FF luciferase from the human U1 snRNA promoter showed no decrease in luciferase activity in the presence of 5′-UTR siRNA.
While the in vitro
‘diced’ siRNA represents a pool of many 21–23 nt siRNAs spanning the 910 nt 5′-UTR region, and thus it is not possible to know which fraction of the pool can actually function to target the homologous regions of the transcript for RNAi-mediated degradation, we have demonstrated that the RNAi effect occurs only with transcripts that include the 5′-UTR and not the luciferase coding region itself. In addition, in vitro
‘diced’ irrelevant siRNA from the β-galactosidase coding region did not reduce the expression of the hybrid transcript, providing further evidence that the significant reduction in FF luciferase activity observed with the pool of 5′-UTR assay is due to sequence-specific targeting of the hybrid transcript. Furthermore, the reduced luciferase mRNA levels detected by semi-quantitative RT–PCR provide additional evidence that the decreased luciferase activity is due to reduction in the hybrid transcript, and not due to altered translation of the hybrid reporter gene. Thus, our data is in agreement with two other published reports that in vitro
cleavage of long dsRNA by recombinant DICER is a viable method to produce sequence-specific siRNAs for gene silencing experiments (46
). Interestingly, the use of a single siRNA (5′-UTR #749) provided a more potent silencing effect than a similar concentration of diced 5′-UTR siRNA. Furthermore, the silencing effect of the synthetic 5′-UTR siRNA was very pronounced even at low concentrations, but we did not observe the same silencing effect with diced siRNA at these lower concentrations. One speculation for the increased potency of the synthetic siRNA is that the functional 21mers in the diced siRNA pool compete with the many non-functional 21mers for target binding and active RISC formation. Indeed, we have observed a similar phenomenon when targeting endogenous genes; namely that two weakly functional siRNAs do not perform together as well as a single efficient siRNA at lower concentrations (H. S. Soifer, unpublished data).
Although we demonstrated that siRNA homologous to the 5′-UTR could target a hybrid reporter transcript for RNAi-mediated degradation, it was not readily apparent that the full length L1 transcript would be susceptible to RNAi. This concern grew from the fact that the L1 RNA exists in the cytoplasm as RNPs with multimerized p40 protein encoded by its own ORF1 (3
). Although the function of the human p40 has not been assigned, one possibility is that the RNPs protect the L1 from degradation by cytoplasmic RNAses. Despite the co-localization of L1 RNA to RNPs, however, siRNA targeting the 5′-UTR produced >80% reduction in L1 retrotransposition activity in HCT116 cells at concentrations of ~10 nM. We observed a greater reduction of retrotransposition activity in HeLa cells with the same concentrations of diced 5′-UTR siRNA relative to the no siRNA control. The increased response of HeLa cells to 5′-UTR siRNA might result from the hyper-triploid karyotype of HeLa cells, thus providing more RNAi components compared with HCT116 cells, which possess a near diploid karyotype (48
). Since we have little understanding for the role of human p40 in the retrotransposition process, we can only speculate that the siRNA is working at a stage before the RNP assembly with L1 RNA. If the siRNA is capable of targeting the L1 RNA after p40 multimerization and RNP assembly, this would require association of the large RNP with RISC producing a very large complex approaching 800 kDa. While p40 does contain a leucine zipper that could mediate interactions with other proteins, there is no evidence to suggest that the human L1 p40 interacts with the RISC machinery.
In addition to targeting the 5′-UTR, we also produced in vitro
‘diced’ siRNA from the ORF1 region of the RC-L1 clone L1RP
. The level of inhibition achieved with 10 nM of diced ORF1 siRNA was significantly greater than the inhibition observed with a similar concentration of 5′-UTR siRNA. A number of possibilities exist to explain the more pronounced decrease in L1 retrotransposition in response to ORF1 siRNA. The most obvious explanation is that the ORF1 siRNA pool contains more functional siRNA molecules than the pool of 5′-UTR siRNA. While there is some indication as to the thermodynamic characteristics that distinguish a functional siRNA duplex from a non-functional siRNA, such as 5′-end duplex stability, GC content, and internal Tm, we are still unable to predict which set of 21mers from a given long dsRNA will be effective siRNAs (50
). Additionally, we cannot identify the population of ‘diced’ siRNA based on the sequence of the starting long dsRNA. Although biochemical evidence suggests that DICER recognizes the stem loop structure of both shRNA and micro RNA precursors and measures from the end of the stem to cleave a 21–23 nt duplex, the manner in which DICER cleaves long dsRNA ex vivo
has not been established (52
) (H. S. Soifer, unpublished data). Another possibility is that the stable secondary structure of the 5′-UTR resulting from higher GC content (58% 5′-UTR versus 42% ORF1) poses a barrier to RNAi. Indeed, strong localized secondary structure surrounding the target site of the mRNA is known to negatively affect RNAi (53
In summary, we have presented evidence that the L1 retrotransposon is susceptible to RNA interference. While our study does not provide direct proof that RNAi functions to control the activity of the remaining RC-L1s, we have provided proof of principle that the RNAi machinery can act on an active RC-L1 and limit its retrotransposition activity using a cell culture retrotransposition assay. Direct evidence that RNAi limits the activity of L1s awaits the cloning of endogenous LINE-1 siRNAs from human cells. Efforts to clone the small RNA population from cultured human cells failed to detect L1 siRNA, suggesting that if endogenous L1 siRNAs are produced, they may be present only during specific developmental stages (54
). One requirement for the production of L1 siRNA would be transcription of antisense L1 RNA that could hybridize with L1 sense RNA to form either dsRNA or siRNA. In fact, there is experimental evidence demonstrating that large quantities of both sense and antisense L1 RNA of variable size are present in the total RNA of a human teratocarcinoma cell line (55
). As the human genome is populated by more than 500
000 L1 sequences, it is possible that a subset is transcribed by a nearby promoter into ds- or siRNA. This arrangement, however, would still require the existence of two opposing promoters to transcribe both sense and antisense L1 sequence. Alternatively, the production of sense/antisense L1 dsRNA might take advantage of a unique feature of the L1 5′-UTR, namely the existence of an internal promoter that transcribes L1 sense RNA and an antisense promoter within nucleotides +400 to +600 of the 5′-UTR that transcribes L1 sequence in the opposite direction. In cell lines where the 5′-UTR sense promoter shows transcriptional activity, the L1 ASP is also transcriptionally active, albeit at lower levels (43
). A recent report demonstrates that transcription from the L1 ASP is dependent on the transcription factor RUNX3 (44
). In addition to characterizing endogenous L1-derived siRNAs that may arise from the L1s 5′-UTR, the establishment of conditional RNAi knock-out mice and derived fibroblast lines will permit more extensive analysis of the retrotransposon activity than can be achieved with the current DICER and AGO2 null mouse models (31