It’s all about energy

Over the last 10 years or so — thanks to tools that allow us to study the behaviors of individual cells — we’ve become increasingly aware of and interested in cell-to-cell variation in genetically identical populations.  For example, in response to a challenge, some cells may live and others die (I’ve written before about examples of this in both mammalian cells and bacterial cells).  There seem to be many possible reasons for this variability, including an inherent randomness in the expression of individual genes that leads to individual cells having distinct complements of proteins.  Here’s a new one, which personally I find quite surprising: according to a recent paper (Pires des Neves et al., 2010, Connecting variability in global transcription rate to mitochondrial variability, PLoS Biology 12 e1000560, doi:10.1371/journal.pbio.1000560) the overall level of transcription may vary from cell to cell due to variations in the level of ATP, which are in turn caused by individual differences in mitochondrial mass as a result of uneven partitioning at cell division.

This paper is written more like a detective story than are most scientific papers: instead of starting with a hypothesis and setting out to test it, they start with the corpus delicti and set out to discover the perpetrator.  The evidence that there is indeed variation in the overall transcription rate in individual cells — not just the rate of transcription of individual genes — comes from experiments in which they add bromo-uridine (BrU) to cells, and track the incorporation of BrU into RNA transcripts using anti-BrU antibodies and imaging.  There’s considerable cell-to-cell variability in the amount of BrU-labeled RNA, and thus we can infer that the rate of RNA transcription varies from cell to cell.  This experiment was done on a variety of mammalian cell lines, including — importantly — primary cell lines.  The authors ruled out two obvious explanations right away: the variation is not due to cells being in different stages of the cell cycle, nor is it due to a major variation in the total number of RNA polymerase molecules.

If the number of RNA polymerase molecules is the same, but the number of transcripts they produce is different, is there a difference at the level of the polymerase activity? The authors added a GFP-labeled pol II protein to a cell line carrying a temperature-sensitive mutant of pol II, generating a cell line in which only the GFP-labeled version of RNA polymerase is active at the non-permissive temperature.  Using photobleaching, they could then study the rate of movement of RNA polymerases from an unbleached area of the nucleus to a bleached area. The overall rate of elongation gets read out as an average half-life of the fluorescence representing GFP-labeled polymerases; and this half-life measure varies widely from cell to cell.  When the same cells are then used for BrU incorporation studies, the measured RNA polymerase half-life (≅ elongation rate) correlates nicely with the rate of incorporation of BrU-labeled RNA.  A second measure of elongation rate, based on the displacement of GFP-labeled histones, provides similar results.

So by this point in the paper, we know that different cells show variability in their average RNA elongation rate.  The variation in the amount of elongation is not due to differences in the total amount of RNA polymerase.  Nor is it due to differences in the rate of individual polymerases: all the polymerases in an individual nucleus appear to be elongating their RNA at the same rate, within experimental error.  What is the factor that determines how fast the polymerases can work?

An important clue: the mysterious factor is soluble.  When the authors fused the cytoplasm of two cells, producing a single cell with two nuclei, both nuclei showed the same transcription rate.  So whatever the factor is that determines elongation rate, it can quickly diffuse through the cytoplasm: it’s not something bound to DNA, and it’s probably not a protein.  Is it one of the nucleotide substrates needed for RNA polymerization?

By performing elongation assays in permeabilized cells, the authors were able to determine the effect of nucleotide concentration on the rate of RNA polymerization.  For UTP, CTP and GTP, they see a hyperbolic curve that reaches a maximum polymerization rate at around 250 µM; since the normal concentration of these nucleotides in the cell is much higher than this (approximately millimolar), it appears that we can rule them out as the likely source of the variation.  But the curve for ATP is very different.  It looks sigmoidal, not hyperbolic (my re-creation of their data is rather rough and amateurish, but you get the idea); in other words, it looks as if it’s allosterically regulated.  Furthermore, it looks as if the normal cellular level of ATP (about 1 mM) isn’t enough to saturate the enzyme, so small variations in ATP concentration might give significant variations in enzyme activity.

Now, this is a surprise; and you don’t usually expect surprises of this kind with an enzyme as well-studied as RNA polymerase.  The authors suggest that this is because previous studies have been performed in vitro, or lacking the near-physiological salt concentrations and macromolecular crowding agents that they use to maintain in vivo-like conditions.  One of the differences between the cellular context and in vitro assays would be that in cells, RNA polymerase needs the activity of chromatin remodeling factors, such as the SWI/SNF complex that helps get nucleosomes out of the way.  This complex, like RNA polymerase, depends on ATP for the energy it needs to work.  Could it be that the sigmoidal dependence of polymerase activity on ATP is the result of combining several processes that all depend on ATP?  It’s possible to cause chromatin to decondense, opening up access of RNA polymerase to DNA, using a brute force method, hypotonic shock; when Pires des Neves et al. do this, the sigmoidal dependence of RNA polymerase on ATP goes away, leaving a hyperbolic curve.

When you have eliminated the impossible, whatever remains, however improbable, must be the truth: variation in ATP levels is the cause of the cell to cell difference in global transcription rate.  That might be enough for Sherlock Holmes; but the authors wanted more direct evidence that ATP levels actually do vary in individual cells, and — if so — they wanted to know why. As a functional measure of ATP levels, they used a reporter system that the cell itself uses to measure its energy state: the mTOR pathway acts as a homeostatic ATP sensor, and phosphorylation of the ribosomal protein S6 downstream of mTOR depends on ATP levels.  You can manipulate the energy state of a cell by incubating it with different concentrations of deoxyglucose, which allowed them to change S6-P levels over a range of almost 100-fold in individual cells; transcription rates correlate with S6-P levels extremely well, in fact so well that I wrote “wow” next to the figure (Figure 3D); but perhaps what this implies is that global transcription level is itself somehow controlled by the mTOR pathway.

Why do ATP levels vary?  Since ATP production largely happens in mitochondria, the authors looked at the mitochondrial content of individual cells; they found that mitochondrial mass varies widely from cell to cell.  They used FACS to sort cells with high mitochondrial content from cells with low mitochondrial content, and found that low mitochondrial content correlates with low transcription.  And they used imaging of YFP-labeled mitochondria to ask whether the variation in mitochondrial content comes from asymmetry in the partitioning of mitochondria between daughter cells at mitosis, and find evidence that it does.  Indeed, the partitioning of mitochondria is so asymmetric that it leads to significant differences in cell cycle times between sister cells.

If it’s generally true that there’s significant cell-to-cell variability in mitochondrial content, this would have quite a lot of implications, and I’m not entirely convinced.  This particular set of experiments is done only in HeLa cells, which could easily be unusual; and you’d expect mammalian cells to care about what happens to their mitochondria on cell division (or I would, anyway).  ATP levels could also vary for reasons other than sheer mass of mitochondria; the authors show a correlation between transcription rate and mitochondrial membrane potential, which could vary independently of mitochondrial mass.

But isn’t it fascinating that random variations in something as basic as ATP concentration can lead to cell-to-cell differences?  Marc Kirschner points out that this could have much more general implications: if ATP levels modulate the overall rate of transcription, then perhaps a wide range of cellular functions such as cell division, morphogenesis and differentiation could be coordinated by the level of ATP (or, more probably, the ATP/ADP ratio).  This could explain how the overall process of development is coordinated even at different temperatures or nutrient levels; instead of each separate function being similarly tuned to the correct level by temperature, they could all depend on ATP and thus automatically work together. A very interesting speculation.

das Neves RP, Jones NS, Andreu L, Gupta R, Enver T, & Iborra FJ (2010). Connecting variability in global transcription rate to mitochondrial variability. PLoS biology, 8 (12) PMID: 21179497