Friday Feature: Variable death

Error

This video doesn’t exist

It’s Friday, and what better way to get into the mood for beer hour than to talk about death?  To make sure that the mood doesn’t get too lugubrious, let’s stick to the death of entities that are unlikely to remind you of anyone you know, such as single cells.  This video shows HeLa cells undergoing apoptosis, from work described in Spencer SL, Gaudet S, Albeck JG, Burke JM, Sorger PK. 2009.  Non-genetic origins of cell-to-cell variability in TRAIL-induced apoptosis. Nature 459 428-32.  PMC2858974.  You’ll see that the greenest cells round up and bleb and die very fast, whereas the less green cells die much more slowly.  And thereby hangs a tale.

Apoptosis — programmed cell death — is one of those processes that you would think would be entirely predictable.  If you’ve triggered a cell to die, you’d think that would be the end of it.  But this turns out not to be true.  If a clonal population of cells is exposed to TRAIL (TNF-related apoptosis-inducing ligand), typically some survive; and the time to death is also very variable.   On the face of it, this variability in genetically identical cells is quite mystifying.  What makes it more than a mere curiosity is that this same variable response might be important in clinical settings, where it’s typical for cancer therapeutics to kill some cancer cells and spare others (a phenomenon called “fractional kill”).

Sabrina Spencer (who has now left the Sorger lab for post-doctoral training in the Meyer lab), set out to ask what kind of mechanism might explain this variable death.  The most traditional and obvious candidate, perhaps, is cell cycle phase; another obvious (though less traditional) hypothesis is that some yet-to-be-identified key factor is present at very small numbers per cell, so that chance fluctuations in its level create significant differences in the functioning of the death pathway.

But there is a third possible explanation.

Even if the level of the key protein is not extremely low, there could still be cell-to-cell variability in its level.   Such fluctuations would be caused not by single bursts of gene expression, but by long-term differences between cells in the rate of production or degradation of this particular protein.  If this is the explanation, then the fluctuations would be relatively slow, and sister cells should look much more similar to each other than unrelated cells.  If the fluctuations are stochastic, they would be expected to randomize on a much faster time scale than cell division, and sister cells should look no more similar than any two “unrelated” cells from the same clone.  Spencer et al. therefore set out to image the behavior of sister cells and unrelated cells in response to TRAIL, keeping track of cell cycle stage as well to address the first hypothesis.

Key findings here include: (1) cell cycle stage is irrelevant; (2) sister cells show a strong correlation in both fate and time to death; and (3) this correlation is lost after several hours.  To ask whether these observations could indeed be explained by cell-to-cell variability of protein levels, they looked at the level of five key proteins in the apoptosis pathway. The coefficient of variation for each one is in the range of 25%, which means that cells in the top 5% might have over twice as much protein as cells in the bottom 5%; so an unlucky cell might have >2x more death potential in five different dimensions.  No one protein explains all of the variability, but when you put the variation in all 5 protein levels into a model of the TRAIL-induced death pathway, what you get out is a virtual “population” of cells that mimic the variability in death seen in experiments.

Looking more closely at which step in the pathway accounts for most of the variability in the progress of the death signal, Spencer et al. determined that the main difference between cells is the time it takes for the levels of a key protein, activated Bid, to build up to the point where the cell commits to die.  In fact, although it’s really hard to predict whether a normal cell will live or die — and when it will die — modeling suggests that cells overexpressing Bid above a certain level should become quite predictable: they should die, and die quickly. And they do, as shown in the movie above of cells overexpressing Bid-GFP.

Don’t go away thinking that Bid is the only thing that matters, though.  That way of thinking is so 90’s.  It’s only when Bid is hugely over-expressed — 50x normal levels — that it becomes the primary determinant of cell death.  Normally, several factors are important.  Will this cell die? Maybe. It all depends.  The story is a lot harder to explain to your grandmother than a simple Just So story with a single cause.  But maybe if we stop imagining that one single magic bullet will solve our problems, we might find out why cancer treatments often fail.  And maybe we’ll even discover how to make better ones.

Spencer SL, Gaudet S, Albeck JG, Burke JM, & Sorger PK (2009). Non-genetic origins of cell-to-cell variability in TRAIL-induced apoptosis. Nature, 459 (7245), 428-32 PMID: 19363473