We have manipulated genetic expression in Drosophila to produce a first-order approximation to the genotype-phenotype map of wing shape, and we have dubbed this effort the dictionary of genetic effects. Starting from the principles (1) that both genotypic and phenotypic variation are complex and multidimensional, making their mapping a very cumbersome endeavor, and (2) that the mapping of macro-effects (e.g., gene knockouts) informs little about the causes of natural variation, we have implemented a system that allows us to introduce gradual, controlled knockdowns in gene expression, and measure their phenotypic effects as functions of the magnitude of the knockdowns. To date, we have phenotyped wing shape effects associated to misexpression of over 150 genes.
     The critical contribution of the Dictionary goes beyond showing which-genes-have-what-effects; our dictionary is producing a catalog of wing shape traits with known causal origin. Unlike traditional measurements (e.g., length, area, or shape measurements), which are rarely chosen based on their biological meaning, dictionary traits comprise the phenotypic targets of changes in gene expression, and in principle contain the full (or measurable) set of pleiotropic effects of this change.

Least-squares approximation of the local effect of expression knockdown of the gene engrailed on Drosophila wing shape (i.e., "engrailed-ness").

     As an example, consider the phenotypic effect of knocking down the gene engrailed. By regressing shape onto a surrogate of en transcription, we characterize this effect as a vector, illustrated above as an animation that relates perturbed and unperturbed phenotypes. We can call this vector, for lack of a better name, engrailed-ness, which we can use alongside other dictionary vectors to score a sample of wings. We can envision many scenarios where having a good idea of what developmental process is involved in producing a particular direction of variation can boost the explanatory power of the data. Consider a geographical gradient where engrailed-ness was found to be correlated to some environmental variable; the trait itself would offer en or a related gene as a candidate causal factor for the observed variation, which can then be targeted in genetic analyses.
     At present, we are exploring individual and collective variational properties of dictionary vectors, including their evolvability, modularity, and pleiotropy. We are also using dictionary vectors as traits in a GWAS, as selection gradients in artificial selection experiments, and as input parameters in a developmental model. We are interested in exploring the uses of these vectors in ecology, biogeography, ecophysiology, and similar applications.
     We are also quantifying phenotypic responses to gene expression perturbations as non-linear multivariate functions, focusing on two general aspects of non-linear behavior. First, we are cataloging the distinct pattern of non-linearity associated to each gene, e.g., threshold-limited, asymptotic, diminishing returns. This pattern, which we refer to as the phenotype of the gene in itself can be interpreted as a trait of the underlying regulatory system. Second, we are combining all of these functions in a common phenotype space centered on wild phenotypes (dubbed the spaghetti bowl, illustrated below), whose structure we are beginning to explore at multiple scales to search for causal links between genetic interactions and phenotypic covariation.

Sample of gene expression trajectories in phenotype (wing shape) subspace spanned by PCs 1,2, and 5. Each trajectory represents local responses to gene expression knockdown; parrot plots describe shape changes implied by each dimension.