The power and perils of metaphors in science

Humans apparently cannot avoid thinking, at least occasionally, by images and parallels to already known situations. The combination of mental imagery and logical parallels produces metaphors, and metaphors have played a crucial role in the development of human thought, from religious teachings to mystical "insights," from philosophy to science.

It is with the use of metaphors in science that I am particularly concerned here. They are all over the place, from physics to biology, used in both technical papers and nature documentaries. Think of early mental models of the atom as a miniature solar system; they were useful because they recast the unfamiliar microscopic in terms of the familiar macroscopic, making it easier for countless students and researchers to make sense of the data that were coming in from the experiments. The problem is, atoms are really not miniature solar systems, not even close; and if one insists on thinking of them that way, then it becomes next to impossible to account for the advances of quantum mechanics.

But I am not a physicist, so I will instead discuss a metaphor with which I am more familiar professionally, one that has spurred a large amount of discussion and research (both theoretical and empirical) in evolutionary biology. This is the idea of an "adaptive landscape," first proposed by American biologist Sewall Wright in 1932. Wright developed a complex mathematical model to describe how evolution by natural selection (and other means) can improve the fitness of populations of organisms to the environments they encounter. Wright's math, however, was a bit too complicated for most biologists, and his advisor asked him to develop a graphical representation of his ideas before an important international meeting: without the graphical model, Wright's ideas would not have been understood, and would have probably been ignored.

Wright then produced a series of drawings that eventually became the canonized version of the "adaptive landscape" metaphor (figure 1). An adaptive landscape can be drawn in three dimensions, with two of the axes representing the frequencies of two genes that are known to affect fitness, and the vertical axis quantifying the fitness of a population characterized by those gene frequencies. Adaptive landscapes are "rugged," that is they have "peaks" of high fitness (corresponding to the genetic combinations favored by natural selection) and "valleys" (areas of low fitness due to disadvantageous gene frequencies). It became immediately clear from staring at an adaptive landscape that evolutionary theory had an interesting problem to solve: suppose a population evolves toward the top of a local adaptive peak. How can that population move away from the peak, say to reach one that is part of the same landscape, but is higher (and therefore corresponds to increased fitness)? Natural selection alone cannot do the job, because it cannot predict the future and push a population down a maladaptive valley in order to cross it and then start climbing another peak. Wright introduced an ingenious theory to solve the problem of "peak shift," as it became known, a theory that involves complementary roles of random events in small populations (so-called "genetic drift") and natural selection. The problem is that decades of research on peak shifts has led most practitioners to conclude that Wright's solution to the problem is very unlikely to work in actual biological populations; moreover, there doesn't seem to be a better alternative around.

[FIGURE 1 OMITTED]

Some biologists have always been rather cold toward the whole adaptive landscape business, for a variety of good reasons. For one, real populations evolve

in ever-changing environments, which means that the configuration of adaptive landscapes itself ought to change. What was a peak yesterday may become a valley today, and vice versa, and valleys could be traversed simply by the fact that the peaks move around, rather than the populations. Furthermore, who's to say that evolution has to allow populations to always maximize their fitness? Natural selection is not an optimizing process (i.e., one that guarantees the best possible outcome), but rather a "satisfycing" one, capable of improving the organism-environment fit only locally. As if all of that were not enough, there is the significant problem that nobody has ever measured the characteristics of a real adaptive landscape: the whole thing is a figment of mathematical imagination with little or no empirical backing.

But wait! There is another potential problem with the metaphor itself: as Wright realized, and everybody else who works in the field knows, realistic adaptive landscapes are very highly dimensional (not just 3D), because there are thousands, or even tens of thousands, of genes that affect fitness. What do these multidimensional landscapes "look" like, or rather, since we can't visualize them, what are their biological properties? Sergey Gavrilets, at the University of Tennessee, has asked that question and produced a series of papers about multidimensional landscapes. It turns out that these are "holey" (as in, with holes, figure 2)--i.e., they are characterized by extensive ridges of relatively high fitness, punctuated by areas of very low fitness (the holes). The surprise was that there are no "peaks" in holey landscapes, and populations can "slide" over ridges of equal fitness, until natural selection pushes them on a higher level.