More microRNA mysteries
December 3, 2010 § Leave a comment
Since we only recently found out that microRNAs and small interfering RNAs (siRNAs) exist, and modulate gene expression, it’s perhaps not surprising that many aspects of their function are still puzzling. One mysterious feature is the fact that the efficacy of microRNAs and siRNAs in silencing mRNAs is rather unpredictable. For example, a single microRNA that targets many mRNA transcripts — binding to the same or very similar sites on each one — may shows a very different degree of silencing on different targets. And some targets seem to be resistant to silencing no matter how hard you try. In the past, we’ve looked at this as a problem with the design of microRNAs and siRNAs; perhaps we don’t know how to do it right, yet. But a new paper from Debbie Marks and collaborators (Larsson et al. mRNA turnover limits siRNA and microRNA efficacy. doi:10.1038/msb.2010.89) now shows that at least part of the problem lies in the target, not the silencer: mRNAs that turn over rapidly are far harder to silence significantly than mRNAs that turn over slowly.
Larsson et al. provide a simple mathematical argument to show why they started exploring this idea, and I will reproduce part of this in a minute. But for those who aren’t used to thinking about the mathematical analysis of steady states, here’s an analogy. Suppose that you’ve finally been pushed over the edge by the wasteful, ear-shattering use of leaf blowers in your neighborhood, and you’ve taken the necessary time away from your normal concerns to invent and build a Roomba equivalent for the garden. Now the leaves are falling, and you’re testing it.
What determines the steady-state level of leaves in your garden while the Groomba is operating? If the leaves are falling at a constant rate, the steady state will be determined by how fast the Groomba picks up leaves: when the Groomba is picking them up at the same rate that they’re falling, you have a steady state. But the rate the Groomba picks leaves up is determined by how many leaves are on the ground — the denser the leaf coverage, the easier it is for the Groomba to pick up 10 leaves in one scoop. So as the leaves fall, the total number of leaves in the garden gets larger and larger until the leaf density gets to the point where the Groomba can finally keep up. If you had two Groombas running, the leaf density at steady state would be lower; alternatively, the rate of leaf fall you could keep up with for a given steady state would be higher.
Eukaryotic mRNA levels are subject to many kinds of regulation, mediated for example by special stem-loop structures, or sequences in the 3′ untranslated region such as AU-rich elements. Any given mRNA might be considered to have the equivalent of at least one Groomba maintaining its steady state at a particular level. But some mRNAs turn over much faster than others, and we could therefore think of them as having, say, 10 Groombas instead of one. The main point of Larsson et al.’s mathematical analysis, paraphrased, is that once you have 10 Groombas running it is hard for the 11th one to make much difference. [I should note parenthetically that Larsson et al. take it as settled that microRNAs, like siRNAs, obtain most of their effect by destabilizing their target mRNA — which has been controversial in the field, as I’ve discussed before.]
It takes much less time to say this in equations, which of course is why (some) people like using them. If the level of mRNA is y, the production rate is A, the endogenous decay rate is B and the additional decay rate imposed by your siRNA or microRNA is C, then in the absence of the siRNA we have:
dy/dt = A – By
at steady state, dy/dt = 0 so A = By, or y = A/B
When the siRNA is present, and assuming for the moment that the two forms of regulation don’t interfere with each other, we have:
dy/dt = A – By -Cy, or dy/dt = A – (B+C)y
at steady state, dy/dt = 0 so A = (B+C)y, or y = A/(B+C)
When you look at the difference your siRNA has made to the steady state level, you look at the new level, A/(B+C), as a proportion of the old level, A/B. If we divide the new level by the old level, the A’s cancel out and we have
Relative change = B/(B+C).
Obviously, the larger B is the harder it is for C to make a large difference. If you start with one Groomba and add a second, the steady state drops to 1/2 of where it was, but if you start with 10 and add an 11th the change is only from 1 to 0.9. Allowing the two forms of regulation to interact (they could be synergistic, or antagonistic) does make a difference to the analysis, but the overall picture remains similar.
So much for the theory. Does it translate to reality? Larsson et al. created a set of mRNAs that were identical except for the number of destabilizing AU-rich elements in their 3′-untranslated region, and tested to see how their steady state level was changed by an siRNA that targets a site in the middle of the coding region. The mRNAs carrying several AU-rich regions had significantly higher turnover rates than mRNA with no destabilizing elements. And the theoretical prediction held up: a 2-fold change in turnover rate was enough to give a large difference (55%) in the residual signal after siRNA targeting. Since the variation in turnover rate in mRNAs in vivo is a lot larger than 2-fold, this effect could indeed be significant.
How much of the observed variation in targetability of mRNAs does this effect explain? Perhaps quite a lot. Larsson et al. used literature data (a microarray time series of mRNA decay after actinomycin D inhibition) to develop an estimate of the half-lives of individual mRNAs in HeLa cells. Then they took new data on the efficacy of >2500 siRNAs targeting these same mRNAs, and asked whether there is a correlation between mRNA half-life and siRNA efficacy. Even though the siRNAs were not individually optimized for the mRNAs, they still see a significant correlation: long-half-life mRNAs (those that turn over slowly) were significantly easier to target than short-half-life mRNAs. A similar approach gave qualitatively similar results for microRNAs.
This doesn’t mean that you’re completely out of luck if the mRNA you want to target is rapidly turned over — there’s a lot of scatter in the data. But perhaps it does suggest that if you have a choice between targeting a fast-turnover mRNA and a slow-turnover one, you could be better off picking the latter.
[PS: the word “Groomba” appears to have been already used — though not extensively — to refer to a monkey-like robot that grooms your facial hair and brushes your teeth while you’re still sleeping. So I may need to think up a new name when I go into Groomba production mode; suggestions welcome.]
Larsson E, Sander C, & Marks D (2010). mRNA turnover rate limits siRNA and microRNA efficacy. Molecular systems biology, 6 PMID: 21081925