One of the central ways that cells react to their environments is to produce messenger RNA (mRNA) by a process of gene expression. Cells transcribe the genetic information, stored in DNA, into RNA molecules. Each mRNA molecule is then translated into many proteins, the molecules that carry out most biological functions.
The process is similar to a printed recipe (DNA), copied and distributed (mRNA transcription), then used by cooks again and again to make finished dishes (proteins). Just as the number of copies of a recipe being shared gives one some idea about how many times the dishes are being cooked, the number of copies of mRNA provide an estimate of how many proteins are being made.
The question remains, how good an estimate? And what is the relationship between copies of mRNA and copies of protein in a cell?
Since the introduction of large-scale mRNA and protein measurements in the late 1990s, most studies have concluded that differences in mRNA levels explained around 40 percent of the differences in protein levels. The low correlation suggested that some process acting after mRNA transcription—broadly called post-transcriptional regulation—accounted for the difference in protein levels.
But according to a new study by researchers from the University of Chicago and Harvard University, mRNA levels dictate most differences protein levels in fast-growing cells instead, when analyzed using statistical methods that account for noise in the data.
The research, published May 7, 2015 in the journal PLoS Genetics, counters those earlier studies arguing that the correlation between mRNA transcript levels and protein levels is relatively low, and that processes acting after mRNA transcription override mRNA levels. Instead, the authors argue, these conclusions result from interpreting measurement noise as biology.
“Other groups have discussed the possible effects of noise, so we decided to get serious about quantifying those effects, and the results really surprised us,” said D. Allan Drummond, PhD, assistant professor in the Department of Biochemistry & Molecular Biology at the University of Chicago. Drummond is senior author of the study, which was done in collaboration with Edo Airoldi, PhD, from the Department of Statistics at Harvard University. “Taking measurement noise into account, the biology looks quite different than what was being reported,” Drummond said.
In earlier research, post-transcriptional regulation seemed to override mRNA levels, so in some cases highly abundant mRNAs produced few proteins. To use the recipe analogy again, it is as if many widely shared recipes are rarely cooked because of some important feature (for example, requiring rare ingredients). But imprecision or inaccuracy in measurements—often called measurement “noise”—could produce many of these same effects.
“It didn’t make sense to us that the cell would go to all this trouble making lots of mRNA, then contradict itself later. That doesn’t make sense if cells are growing fast, like they were in many of these studies,” Drummond said.
In the latest study, he and his colleagues analyzed data from 24 published studies of budding yeast, a widely studied model organism on which many measurements have been made. They used techniques, some new, some over a century old, which take measurement noise into account.
The result? The team found a much stronger role for mRNA levels, accounting for 85% versus 40% of the differences in protein levels. Even more surprising was that the precise relationship was much stronger than had been reported elsewhere. High-level mRNAs produced far more protein than what would have been expected from their levels alone, indicating an amplification process.
The result was like finding that a recipe shared twice as much was cooked four times as often. It meant that post-transcriptional regulation was amplifying or magnifying the mRNA level differences, not overriding mRNA levels as other studies had concluded. Noise had prevented other groups from seeing this amplification clearly.
Drummond said their findings improve the statistical toolset available to scientists working with large sets of potentially noisy data. “We provide tools for people to work with when their measurements have noise and they’re interested in correlations and quantitative relationships,” he said. “That’s a really common case in science, and my hope is that we can help transform how a lot of people look at their data.”