RNA Journal Club 7/9/09
Transfection of small RNAs globally perturbs gene regulation by endogenous microRNAs
Aly A Khan, Doron Betel, Martin L Miller, Chris Sander, Christina S Leslie & Debora S Marks
Nature Biotechnology 27 (6): 549-55, June 2009.
doi:10.1038/nbt.1543
This week’s cerebral analysis by Graeme Doran:
Investigating the application of small RNAs to destabilize specific mRNA targets, researchers have observed a variety of non-specific ‘off-target’ effects – alterations in the expression level of mRNAs that do not contain a perfectly complementary target site for the small interfering RNA transfected into cells. In a minority of cases, the target mRNA is actually stabilized!
Briefly, ‘off-target’ effects have been broadly understood to derive from 4 main sources:
a) Partial complementarity between the small interfering RNA and mRNAs in the cell.
b) Stress responses due to transfection of foreign RNA.
c) Downstream secondary gene expression changes due to silencing of a specific target.
d) Disruption of the endogenous miRNA regulatory network due to competition between the cellular miRNA machinery and exogenously supplied small interfering or short hairpin RNAs.
To date, conflicting experiments have suggested that some si/shRNAs may inhibit endogenous miRNA activity in some scenarios, particularly when the silencing RNAs are present at high levels. This may occur through saturation of either or both the RISC and DICER activities in the cell, depending on the type of small RNA used.
This study from Khan et al. attempts a broad survey of previously published small RNA transfection experiments.
The key evidence presented are:
1) Transfection of small RNAs into immortalised cell lines produces a consistent de-repression of genes that contain an endogenous miRNA target site within their 3’UTR.
2) The extent of de-repression is quantitative – that is it depends both upon the extent of endogenous miRNA repression on a specific UTR, and the concentration of siRNA used in the transfection.
3) Cellular ARGONAUTE/RISC activity is likely subject to loading competition between small RNAs within cells.
Compiling an extensive set of data leads to some brevity of description in the methods. The authors use 4-species conservation as a criterion for miRNA site prediction. One would expect that the observed de-repression effect would be independent of the seed site conservation, as it is commonly possible to see miRNA repression on non-conserved seed sites, and the authors do not make clear whether this signal was present for the non-conserved site data or not. Further, the normalization of predicted target site sets is not well described. One might envisage that the ‘baseline’ gene sets with no endogenous or exogenous seed sites would be shorter on average than gene sets with 1 or more conserved sites. Shorter UTRs have less scope for regulation by miRNAs or other
RNA binding factors, and so maybe would have less regulation to perturb.
The core data (Figure 2) indicates a consistent and significant de-repression of UTRs that contain conserved (higher confidence) endogenous miRNA target sites, but the data is correlative rather than conclusive – and would have been greatly enhanced by a simple luciferase assay to precisely determine the contribution of individual sequences to repression/de-repression in the context described. As is, the model and the data fit each other, but other modes of de-repression are not discounted experimentally. Data suggesting that the de-repression is dose-dependent seem to rely on relatively small sets of genes selected specifically for responsiveness to siRNA treatment and thus this aspect of the story is less convincing. Furthermore, there is little proof that concentrations required by ‘good’ siRNAs to silence genes (1-10nM) have a significant de-repressive effect on miRNA targets, and so the importance of the observed effect in therapeutic situations where delivery concentrations are low is still an open question.
The core findings of this paper, and the methodology employed, open up some interesting questions about the basic biology of RISC competition in cells. How, for instance, do rapidly induced miRNAs (such as miR-21 upon SMAD pathway activation) interact with the endogenous pool of miRISC? Is there present a surplus of free ARGONAUTE for such instances, or is ARGONAUTE protein concentration limiting in the cell? Do endogenous siRNAs – generated on cell stress or viral infection – displace miRNAs from their silencing role, and what is the half-life of endo-siRNA loaded RISC in vivo? Tumors commonly downregulate miRNA activity, and DICER depletion enhances tumorigenesis. Can this also be achieved by limiting the available miRNA RISC in tumor cells?
Broadly speaking this was a stimulating paper, and whilst I don’t think that it breaks new ground in considering ‘off-target’ effects of small RNAs within cells, it provides some of the most convincing evidence of miRNA target de-repression to date. And beyond this, it provides the thought provoking concept of RISC competition, new questions upon which to focus, and a validated method for analyzing miRNA target de-repression.
You'd Prefer An Argonaute 
Thanks for this interesting summary of our work. Answering a few of the questions;
Yes, we did also look at sites without conservation and still got a clear signal ( data unpublished) Happy to share this if anyone wants it. The signal was stronger using conservation and worked for single sites . Much stronger for multiple sites vs single when not using conservation for obvious reasons. Remember that all, even ‘gold-standard’ target predictions are only getting 1:8 targets correct maximally ( false positives), so its pretty amazing we got a signal at all.
As for the dose-titration data. maybe there’s a misunderstanding or we weren’t clear enough in the paper. We used unconserved sites so counted all those with 2 or more ( since so many genes have one site unconserved for cellular microRNAs) and we wanted to use the siRNA data. As you know from other work ( eg Jackson 2009) siRNA sites in target mRNAs maybe less likely to be conserved than microRNA sites. We would love to have more dose data and are in the process of analyzing some new data released 2009.
Your miR-21/SMAD pathway point is interesting. We touch on it in the paper by intimating that there may be ‘physiological’ conditions where ‘saturation/competition’ is used by the cell to down-regulate the ‘older’ endogenous microRNAs thereby up-regulate their targets. If Argonaute was limiting physiologically, this would be a cute way for the cell to switch regulatory programs before having to degrade all the ‘older’ microRNAs. That is, by sponging up the AGO. You can imagine this is how viral microRNAs might work too.
One line of evidence in support of AGO being limiting in the cell even before adding ectopic mi/siRNAs, is that we see a signal when microRNAs are inhibited. But we need more data on that.
Screams out for some experiments and a model, and there’s no substitute for knowing the real numbers.
Do we know the concentration of AGO2 ( for example) in any cell type?
Thanks, I enjoyed the paper.
I don’t know anybody that has a confident figure for AGO protein numbers yet, not sure that the antibodies available are good/comprehensive enough. There is also the issue of how different si/mi are distributed amongst the family members.
One comment that came out of the discussion of the paper that I forgot to mention – it would be nice if the de-repression signal was absent/reduced in when using siRNA in DICER knockout cells (and maybe even in the hypomoph HCT116 line), on the assumption that there is more free AGO# in these cells. Not sure whether there is array data to that end yet though.
We did indeed test the HCT116 dicer hypomorphs in this paper and we do see some saturation -like effect. This may be because these cells do have significant mature microRNA expression despite the homozygous deletion of the helicase domain of dicer. See Cummins et al., 2005 esp Figure 5.
(If you examine Fig 5B you will see that whilst many microRNAS are indeed down-regulated in the hypermorph, some clearly aren’t.)
On the other hand the absolute amounts of the expression changes should be lower, this should be clear in the original data. I will check
.