We discovered an unexpected mode of bimodal distribution of stable and plastic traits, which was consistent for homologous traits of 32 varieties of seven species both in well-irrigated fields and dry conditions. We challenged archived genetic mapping data for 36 fruit, seed, flower and yield traits in tomato and found an unexpected bimodal distribution in one measure of trait variability, the mean coefficient of variation, with some traits being consistently more variable than others. To determine the degree of conservation of this distribution among higher plants, we compared 18 homologous phenotypes, including yield and seed production, across different crop species grown in a common 'crop garden' experiment. The set included 32 varieties of tomato, eggplant, pepper, melon, watermelon, sunflower and maize. Estimates of canalization were obtained using a 'canalization replication' experimental design that generated multiple estimates of the coefficient of variation of traits, as well as their reaction norms in optimal and water-stressed field plots. A common pattern of bimodal distribution of stable and plastic traits was observed for all the varieties and for a wild weed (Solanum nigrum). We propose that canalization profiles of traits in a variety of taxa were ancestrally selected to maximize adaptation and reproductive success.

We present a complementary resource for trait fine-mapping in tomato to those based on the intra-specific cross between cultivated tomato and the wild tomato species Solanum pennellii, which have been extensively used for quantitative genetics in tomato over the last 20 years. The current population of backcross inbred lines (BILs) is composed of 107 lines derived after three backcrosses of progeny of the wild species Solanum neorickii (LA2133) and cultivated tomato (cultivar TA209) and is freely available to the scientific community. These S. neorickii BILs were genotyped using the 10K SolCAP single nucleotide polymorphism chip, and 3111 polymorphic markers were used to map recombination break points relative to the physical map of Solanum lycopersicum. The BILs harbor on average 4.3 introgressions per line, with a mean introgression length of 34.7 Mbp, allowing partitioning of the genome into 340 bins and thereby facilitating rapid trait mapping. We demonstrate the power of using this resource in comparison with archival data from the S. pennellii resources by carrying out metabolic quantitative trait locus analysis following gas chromatography-mass spectrometry on fruits harvested from the S. neorickii BILs. The metabolic candidate genes phenylalanine ammonia-lyase and cystathionine gamma-lyase were then tested and validated in F populations and via agroinfiltration-based overexpression in order to exemplify the fidelity of this method in identifying the genes that drive tomato metabolic phenotypes.