The birth and death of stem cells

Jeremy Gunawardena suggested that it would be appropriate to write about one of Ernest McCulloch’s papers, given the news that he passed away last week.  McCulloch, together with James Till, was responsible for the original demonstration that stem cells actually exist.  He was 84 when he died, two weeks short of the 50th anniversary of the publication of his most famous paper (Till and McCulloch, 1961, Radiation Research A direct measurement of the radiation sensitivity of normal mouse bone marrow cells 14, 213).  McCulloch was a physician and Till a biophysicist, and together they did the kind of experimentally-driven quantitative modeling that we today consider to be super-modern.  [I wonder if there’s a fashion cycle in science as well as in clothes, and if so whether they’re synchronized?  If one could predict based on the fashionability of, say, Markov models in biology that miniskirts would be making a comeback in the next few years… well, actually I can’t think why that would be interesting or useful, so never mind.]

Jeremy, who is more familiar with the dusty areas of the library stacks than most, suggested a fractionally more recent paper for today’s post: Till, J.E., McCulloch, E.A. and Siminovitch, L. 1963, A stochastic model of stem cell proliferation, based on the growth of spleen colony-forming cells.  Proc. Natl. Acad Sci. USA 51 29-36.

IBM 7090. Image from http://www.computer-history.info.

Take a look at the title — see what I mean?  You’d expect that title to be attached to an article published in this century, not the last one.  It gets better.  The authors are trying to understand single-cell behavior (and infer the behavior of populations), using Markov processes and Monte Carlo simulations, painstakingly performed on an IBM 7090 computer. I’ve given up trying to get people interested in when systems biology started, but if I hadn’t this would certainly be a contender in the systems biology ancestry stakes.

The starting point for this article is the then-recent discovery (for which McCulloch and Till never did get the Nobel Prize) that mouse hematopoietic tissue contains a class of cells that can divide rapidly, producing large numbers of cells.  These cells can be detected because their progeny form colonies in the spleen of irradiated mice: one “colony-forming cell” produces one colony.  At the time, the colony-forming cells could not be observed directly, and so everything known about colony-forming cells was deduced by observing and measuring the colonies they form.

They had interesting properties.  A single cell could give rise to over a million progeny, showing that they had lots of proliferative potential.  Although they were capable of self-renewal — since colony-forming cells could be found within the colonies — most of the cells in the colonies were differentiated.  Thus, Till et al. reasoned, the development of a colony involves processes of differentiation acting on some of the progeny of a single cell, but not others.

Are these mixed fates (differentiated vs. stem cell) programmed, or random?  The authors measured the number of colony-forming cells in individual colonies, in a separate paper, and found that the distribution of such cells was extremely heterogeneous.  This led them to suspect that the process of differentiation was stochastic, governed by a random “birth-and-death” process, and they set out to perform an experimental test of this idea.

Till et al. measured the distribution of colony-forming cells per colony and find that the statistical model that fits it best is a gamma distribution. One way such a distribution can arise is if the underlying process involves a single entity (the colony-forming cell) giving rise to random “births” (division) and “deaths” (differentiation, after which division is no longer possible).  They therefore model the process of colony formation in this way.  They point out that it’s quite possible that the colony-forming cell may divide asymmetrically, giving rise to one cell that is still a stem cell and another that differentiates (but see also evidence against asymmetric division).   But this need not affect their model, since it’s equivalent to adjusting the probability of “birth” vs “death”.

The authors then simulate the growth of a colony using the Monte Carlo method, assuming a random occurrence of “births” and “deaths” as above, and come to the conclusion that it indeed fits the data well.  So, they conclude, the decision to differentiate or remain a stem cell cannot be tightly regulated.  How can this be reconciled with the rather orderly behavior of hematopoietic cell populations?  Perhaps, just as the decay of a large number of radioactive atoms follows a predictable curve even though the behavior of an individual atom is random, large populations of cells may behave predictably even though individual cells do not.

It must have been so exciting to tease these results out of the uncooperative, incomplete data available at the time.  Imagine what these authors could have done with modern tools.  Go, and do ye likewise.

Till, J., McCulloch, E.A., & Siminovitch,L. (1964). A Stochastic Model of Stem Cell Proliferation, Based on the Growth of Spleen Colony-Forming Cells Proceedings of the National Academy of Sciences, 51 (1), 29-36 DOI: 10.1073/pnas.51.1.29