Aligning the Spaces: A Comment on Polly—Developmental Dynamics and G-Matrices

Abstract

In this issue, Polly (2008) discusses the appropriateness of using metaphors and visualization techniques like the adaptive landscape and morphospace and their mathematical descriptors for understanding phenotypic evolution. I was stimulated to explore further, how these abstract visualizations relate to each other and how variation at each level of biological organization (e.g. genotype, developmental program, phenotype) is manifest at other levels. In particular, what is the nature of the adaptive and phenotypic landscapes and what is the distribution of mutational effects in those landscapes? Does the structure inherent at one level promote clustering of organisms at a higher (e.g. phenotypic) level? I prefer to view the relationships among genetic changes, phenotypes, and the adaptive landscape a little differently than in the accompanying paper. The environment interacts with an organism’s performance (e.g. running speed, food crushing efficiency) to select those phenotypes that are more successful (Emerson and Arnold 1989). Different combinations of phenotypic (especially morphological) traits can yield the same performance values. This many-to-one mapping makes for a more complex and rugged adaptive landscape when fitness is graphed against individual phenotypic traits. The impact of that selection is filtered and translated through the various levels of organization to result in changes in allele frequencies, evolution. In that context, Rice’s (2004) ‘‘phenotypic landscape’’ is placed within this hierarchy differently. Rather than inserting it between the adaptive landscape (fitness against genotype) and the phenotype, I view it as underlying the adaptive landscape (fitness against phenotype) and that Rice’s (2004) ‘‘phenotypic landscape’’ is the mapping function that (1) translates the genotype, via development and environment, up the hierarchy into the phenotype, and (2) conducts the effects of selection down the hierarchy onto the genotype. There are several types of adaptive landscapes, depending on which level of organization is being compared to fitness, and as Polly (2008) points out, they can have different topographies or textures. Adaptive landscapes are often presented where fitness is a function of one or more genetic traits (‘‘factors’’ in Polly 2008). But what is represented by the scale called, for example, ‘‘genetic factor 1’’? (Polly 2008). It is clear it must be a continuous variable, whether graphed against fitness in an adaptive landscape or against phenotype in a phenotypic landscape, because of the smooth surface where fitness values of adjacent genetic values are very similar (i.e., Moran’s I, a measure of spatial autocorrelation, monotonically approaches 1.0 with decreasing distance). I would suggest however, that the only genetic factor that could be continuous is allele frequency within a population; this is Wright’s original formulation of the adaptive landscape (Phillips and Arnold 1989; Wright 1932) and the one common in population genetics. But whenever the graph refers to individuals rather than populations, all genetic changes are discrete. There is no genetic scale that is both objective and continuous. How does a mutation at site 451 in locus X from an A to a T move an organism on such a scale? One could, for example, refer to magnitude of gene expression, but that is really a phenotypic trait. At the genetic level, all mutations are discrete with a wide range of possible phenotypic effects, from none to lethal. And no matter how you order those factors on a scale (e.g. adjacent nucleotide positions), the fitness surface of genetic factors S. J. Steppan (&) Department of Biological Science, Florida State University, Tallahassee, Fl 32306, USA e-mail: steppan@bio.fsu.edu