Premise: Evolution of separate sexes from hermaphroditism often proceeds through gynodioecy, but genetic constraints on this process are poorly understood. Genetic (co-)variances and between-sex genetic correlations were used to predict evolutionary responses of multiple reproductive traits in a sexually dimorphic gynodioecious species, and predictions were compared with observed responses to artificial selection.

Methods: Schiedea (Caryophyllaceae) is an endemic Hawaiian lineage with hermaphroditic, gynodioecious, subdioecious, and dioecious species. We measured genetic parameters of Schiedea salicaria and used them to predict evolutionary responses of 18 traits in hermaphrodites and females in response to artificial selection for increased male (stamen) biomass in hermaphrodites or increased female (carpel, capsule) biomass in females. Observed responses over two generations were compared with predictions in replicate lines of treatments and controls.

Results: In only two generations, both stamen biomass in hermaphrodites and female biomass in females responded markedly to direct selection, supporting a key assumption of models for evolution of dioecy. Other biomass traits, pollen and ovule numbers, and inflorescence characters important in wind pollination evolved indirectly in response to selection on sex allocation. Responses generally followed predictions from multivariate selection models, with some responses unexpectedly large due to increased genetic correlations as selection proceeded.

Conclusions: Results illustrate the power of artificial selection and utility of multivariate selection models incorporating sex differences. They further indicate that pollen and ovule numbers and inflorescence architecture could evolve in response to selection on biomass allocation to male versus female function, producing complex changes in plant phenotype as separate sexes evolve.

Baseline generation data come from a partial diallel crossing design of Schiedea salicaria in the greenhouse. Two generations of artificial selection were carried out to select for either high male (stamen) biomass or high female (carpel plus capsule) biomass. Traits measured included the traits directly selected (average male biomass and average female biomass) plus biomass of other flower parts, the potentially correlated traits of pollen number, ovule number, and several inflorescence traits (inflorescence length, number of flowers per inflorescence, and pedicel length). Biomass of individual parts was obtained by drying and weighing to 0.01 mg.  Pollen number was obtained from a particle counter or hemacytometers (in generation 2 only). Where possible, traits were measured separately for terminal flowers and lateral flowers. Data were entered into spreadsheets. Each line of data contains mean values for traits of an individual plant. Baseline generation data are in "Basefinal". Generation 1 data are indicated by "G1". Generation 2 data are indicated by "G2". Analysis was performed with SAS ver 9.3.

Data files are all .csv or .txt files.

A README file is provided for each data set. Missing data is indicated by "."