To study transcriptional kinetics, Singer's lab developed a second imaging approach, albeit one that requires some significant genetic manipulation. The gene of interest is tagged on either the 3′ or 5′ end with multiple copies of a stem-loop sequence carrying the binding site for a viral capsid protein called MS2 and then introduced into cells. To make the nascent transcripts light up, the cells are further modified to express a fusion of MS2 and a fluorescent protein. As the transcript is produced, the labeled protein binds to its cognate hairpins, generating a fluorescent signal that can be counted and tracked over time.
Using this approach Singer and his then-postdoc David Grünwald (now an assistant professor at the University of Massachusetts Medical School in Worcester) showed, among other things, that mRNA movement in the nucleus occurs by diffusion rather than as an active process and that export through nuclear pores takes less than two-tenths of a second. As a postdoc in Singer's lab, Larson used the technique to show that transcription initiation is also largely random, and again, driven by diffusion. “This is precisely what people mean when they talk about noise in gene expression,” explains Larson.
Unlike chemical systems, gene expression involves just handfuls of molecules, meaning that—at the level of transcription—each individual molecule counts. Thus, variables like travel time from point A to point B become significant. “In these limits, you're talking about the probability of an event happening,” Larson says. “It becomes very non-deterministic. And synonymous with non-deterministic is stochastic, essentially random.”
Earlier in 2013 Singer, with then-graduate student Sami Hocine, added a second capsid protein/stem-loop pair called PP7 to the MS2 technique. Using two hairpin and capsid protein sets, Hocine and Singer showed that the expression of 2 alleles of the same MDN1 gene varies significantly, correlating by just 15%.
The technique can also be used to label a single transcript at either end with different colors—a trick that Singer says can be used to measure the translation kinetics of mRNAs by watching the appearance and disappearance of the two colors as the ribosome passes through each hairpin region. The conventional wisdom, he says, is that RNAs are passively translated, but, “I've become leery of our biases.”
Indeed, the deeper they look, the more researchers find that, in the world of transcription, nothing is as it seems. For instance, another bit of conventional wisdom in transcriptional biolog y is that the cell can rapidly remove transcripts as needed. How, though, is that possible in this new stochastic model? In 2011 Singer, Larson, and Singer's then-graduate student Tatjana Trcek, used FISH to identify one possible answer.
Cer ta in cel l cycle-regulated transcripts get co-transcriptionally saddled with a protein complex called Dbf2p, which can assemble a complex that will rapidly destroy the transcripts at a precise moment during the cell cycle. “It's sort of like a bomb that can be set to go off at a certain time,” Singer explains, describing Dbf2p-mediated transcript degradation as “a very elegant process, because it makes a random event deterministic.”Single-cell RNA-Seq
Elegant as imaging-based studies are, the power of modern DNA sequencing technology beckons. Take Harvard University chemist X. Sunney Xie, for instance.
Xie has been studying single-molecule biochemistry for years, mostly using fluorescent strategies to watch as proteins are translated one by one. In 2010, his team used both FISH and live cell techniques to compare the transcriptional and protein output of 1,018 individual E. coli strains, validating the first approach for investigating bacterial translation at the single-molecule level.
Still, Xie's team has now gravitated toward sequencing-based approaches. “With sequencing, it's just much easier to see the entire transcriptome,” he explains.
Several such approaches have been described, RNA-Seq variants such as SMART-Seq, CEL-Seq, Quartz-Seq, and STRT-Seq. But none, Xie says, can really yet sample the single-cell transcriptome in its entirety, first, because cDNA synthesis is not 100% efficient, and second, because “the transcriptome has a huge dynamic range,” spanning five or more orders of magnitude.
Last December Xie's team, led by Chenghang Zong and Sijia Lu, described a new genome amplification strategy they call MALBAC, which combines linear and exponential amplification steps to better sample the genome of single cells for sequencing. Using that method, Zong and Lu sequenced the entire genome of a single human cell, even detecting SNPs and copy-number variations. Now Xie's team is working to apply the approach, which already allows them to sample up to about 80% of the cell's mRNA, to the whole transcriptome.
By all accounts, the data arising from these and other single-cell and single-molecule studies are redefining the models researchers have long used when thinking of transcription. But that's not to say everything the scientific community thought it knew was wrong. Rather its assumptions were, perhaps, overly simplistic.
“The same biochemical laws still apply,” says Larson. “You can still talk about first-order RNA decay, you can still talk about Michaelis-Menten kinetics. But all of those basic mechanisms have to be recast in a form that takes into account the discrete nature of macromolecules.”
Meanwhile, it might be time for another update to Current Protocols.