NAME: Nikolaus Rajewsky
POSITION: Professor, systems biology, Max Delbruck Center for Molecular Medicine
Assistant professor, biology/mathematics, New York University — 2003-2006
Postdoc, Rockefeller University — 1999-2002
Postdoc, Rutgers University — 1998-1999
PhD, theoretical physics, University of Cologne — 1997
A team of researchers from the Max Delbruck Center for Molecular Medicine in Berlin and the UK’s University of Glasgow has developed a new proteomic approach to examine how either microRNA transfection or endogenous miRNA knockdown affects protein synthesis in human cells.
The technique, described this week in Nature, is a variant of SILAC, or stable isotope labeling with amino acids in cell culture, and revealed that a single miRNA can repress the production of thousands of proteins, although this effect is typically mild.
This week, RNAi News spoke with Max Delbruck investigator Nikolaus Rajewsky, a senior co-author on the paper, about the findings.
Could you give a snapshot of the understanding of microRNAs and their impact on protein levels prior to this work?
There are two different mechanisms by which microRNAs can act on their targets. One is in place to degrade or destabilize the target messenger RNA, and the other, which can be independent, represses translation of the messenger RNA.
The first microRNA targets that were [identified], for example, lin-41 in C. elegans, were found to [repress their targets] primarily on the translational level and only in part [through] the messenger RNA-degradation pathway.
There was a Nature paper in 2005 where [the authors] over-expressed microRNAs in HeLa cells and used microarrays to study their effects on messenger RNAs in a more genome-wide fashion, rather than concentrating on individual targets. They found that microRNAs have widespread effects on messenger RNA stability — microRNAs directly destabilize or degrade many, if not most, of their targets.
But what was unknown, and what was maybe the most important question, was what effect microRNAs have on a more general level on translation because protein levels are decisive for phenotype, not messenger RNA levels.
People were [speculating that while] the effects on the messenger RNAs are mild — this had been found in the Nature study in 2005 — maybe the effects on the protein level are much stronger. This was an open question and this is what we were trying to address in our study.
Before you developed the [SILAC] technique detailed in the paper, what were the methods used to examine the effect of microRNAs on protein levels?
[The approach used] in the overwhelming number of studies [involved] individual reporters or genes. To study effects on protein synthesis, you would take a 3’ UTR that you think is a target for a microRNA and play around with this reporter. In most cases, these were what we refer to as artificial constructs. There was one [group that] used SILAC to look at protein levels in a more general [fashion], but this study was compromised by the very small number of changed proteins detected in the experiments.
With this one exception, there were virtually [no studies] that tried to look [at miRNAs’ impact on proteins] on a large-scale. Almost everything [else used] individual reporter constructs that were mutated, studied, and so on.
Can you talk a little about the variant of SILAC that was developed?
The standard SILAC, which was invented about five years ago by Matthias Mann [at the University of Southern Denmark], is a method that allows you to measure changes in protein concentration.
MicroRNAs repress protein translation, so you don’t really want to look at changes in protein concentration, you want to look at changes in the number of newly synthesized proteins when a microRNA is activated or knocked down. Standard SILAC doesn’t allow you to do this, therefore [Max Delbruck researcher and co-author of the Nature paper] Matthias Selbach developed pulsed SILAC, or pSILAC, which [uses] pulse-labeling with an additional label to allow you to look at changes in protein synthesis.
This is the first time that changes in protein synthesis have been measured in a genome-wide scale in human cells.
When you applied the pSILAC approach, what did you find?
We studied cases where we either induced individual microRNAs or we knocked them down. In both cases, we found that the effects on protein synthesis are largely mild, which is maybe contrary to what a number of people were thinking.
I would like to stress that although the [overall] effects are mostly mild, for each individual microRNA we also detected a number of cases where the effects were strong. I want to caution that it’s not like [the impact on protein synthesis is always] mild. In all cases, we found a dozen or so proteins that were regulated strongly. Nevertheless, the big picture is of mild regulation.
I should also say that we not only did the pulsed SILAC approach, [but also] in parallel, for the same samples, we measured transcript levels by doing standard microarrays. Then we could subtract from the pSILAC measurements the changes on the messenger RNA level and therefore really get at changes in how translation is regulated. Because pSILAC measures changes in protein synthesis, which is a complex function of changes in messenger RNA levels … [and] translation rates, by measuring in parallel the transcriptomics, we could compute the direct effects on translation only.
What we found was that microRNAs do directly regulate hundreds of target genes on the translational level, in addition to the messenger RNA level.
You did some additional work looking at target sequence characteristics. What were your findings there?
We found that a major determinant of translational regulation is the consecutive Watson-Crick complementarity between [positions 2 through 7 from the] 5’ end of the microRNA, or the seed, and the target message. In short, we found that the seed is the predominant determinant for [regulation], but this doesn’t explain everything. In a number of cases, you have seed sites but you don’t see an effect on the target on either the message level or translation level.
In past years, there have been a number of purely computational approaches where people try to predict what kind of targets a microRNA has, and this is mostly done on a sequence level. There have been a number of papers published in prominent journals that look at all kinds of things, for example, [whether] the seed is conserved … [and] accessibility of the target … to the microRNA.
We computed how well the published algorithms correlate with our data and found that some correlate very well, but others don’t. The ones that correlate are the ones that really make use of seed sites — conservation of seed sites, [for example] — but not much more.
So we did not find a good correlation with algorithms that use secondary structure information and things like that. But I want to stress that we still do not entirely understand what makes a target a target. We know some things that give you a good prediction, but there are always cases where the prediction is simply not true. Hopefully, our data are going to valuable for the future because they will be analyzed by a large number of people who will try to [solve] this mystery.
Can you touch on the work you did looking at let-7b and its effect on Dicer?
What we found is that when you over-express and knock down let-7b independently, you observe a very significant negative correlation on protein synthesis. The astonishing thing to us was that not only did the presumed direct targets go in opposite directions when … let-7b [is modulated], but the vast majority of all assayed genes were negatively correlated between the two experiments.
It is as if when you turn up the microRNA, the synthesis rates of thousands of proteins [go up or down]. If you turn the microRNA [down] … the synthesis rates go [down or up]. It is as if you can use the microRNA as a [rheostat] to regulate the protein production of thousands of genes.
When you knock down let-7, one of the strongly changing genes is Dicer, which goes up … more than four-fold. The beauty is that it goes up on the protein synthesis level, so more Dicer is synthesized in the let-7 knockdown, but you see only minor changes on the message of Dicer. This is a case where you would have trouble finding Dicer as an important effector of let-7 by just doing transcriptomics. This is a case where you clearly need to do proteomics.
It is well known that Dicer regulates the maturation of all microRNAs, so it looks like let-7 either directly or indirectly regulates Dicer, which in turn likely regulates hundreds of microRNAs. This might explain why we see such a tight negative correlation between thousands of proteins in these let-7 experiments.
Of course, this awaits further experimental evaluation, but it is a possibly very interesting feedback loop that comes out of the data.