One of the challenges to understanding the relationship between genotype and phenotype is that complex phenotypes, such as learning and memory, emerge from interactions amongst groups of genes. Despite its widespread relevance, the nature of multi-gene interactions are ill understood, in part, because most genetic studies are limited to pair-wise studies. To investigate this question, a novel approach was developed and implemented in Drosophila, using the biologically important and clinically relevant cAMP pathway as a model. I used selective breeding to evolve combinations of alleles capable of suppressing the learning defect of mutations in the rutabaga adenylyl cyclase gene. Unlike a classical suppressor screen, the use of experimental evolution allowed me to explore the potential impact of gene interactions among more than two loci. And unlike a classical selective breeding experiment, the genetic variability was constrained to a set of 23 known loci, providing access to the underlying causal alleles. After 41 generations a clear response to selection was observed. Remarkably, selected groups had performance at levels approaching that of wild-type despite the fact that all animals were homozygous for a null allele of rutabaga. High throughput genotyping and multivariate analyses lead to identification of loci underlying the selection response. Using independent genetic experiments, I then exhaustively tested the effects of each of the identified loci as well as of all di-allele combinations. One of the 8 loci partially but significantly suppressed rutabaga on its own. Simulations of the lab evolution experiment indicate that combinations of up to 5 loci could feasibly have been selected. Interactions involving 6 or more loci likely could not. Taken together, the findings in this thesis support the idea that multiple combinations among even a limited set of loci are capable of bypassing the requirement for a central player such as rutabaga. This speaks to the remarkable flexibility of gene networks. Understanding how gene networks are modified in response to a selective pressure can help to model complex phenotypes, including those associated with human disease.