The vast variation in floral traits across angiosperms is often interpreted as the result of adaptation to pollinators. However, studies in wild populations often find no evidence of pollinator-mediated selection on flowers. Evolutionary theory predicts this could be the outcome of periods of stasis under stable conditions, followed by shorter periods of pollinator change that provide selection for innovative phenotypes. We asked if periods of stasis are caused by stabilizing selection, absence of other forms of selection, or by low trait ability to respond even if selection is present. We studied a plant predominantly pollinated by one bee species across its range. We measured heritability and evolvability of traits, using genome-wide relatedness in a large wild population, and combined this with estimates of selection on the same individuals. We found evidence for both stabilizing selection and low trait heritability as potential explanations for stasis in flowers. The area of the standard petal is under stabilizing selection, but the variability is not heritable. A separate trait, floral weight, presents high heritability, but is not currently under selection. We show how a simple pollination environment coincides with the absence of current prerequisites for adaptive evolutionary change, while heritable variation remains to respond to future selection pressures.

The methods to collect the datasets are described below. R code used for analysis is also included in a separate file.

Pollinator censuses. To quantify the diversity of floral visitors and visitation rates, we ran multiple three-minute pollinator censuses at different times of the day, for up to five hours of observations per site, on two separate days during peak blooming in 2014 (plus extra censuses in two sites in 2013). Each census recorded the number and identity of visitors to patches of flowers on haphazardly chosen shrubs, including but not limited to 40 tagged individuals. We counted the number of flowers surveyed in each census and the number of flowers visited to estimate the per-flower visitation rate.

Floral phenotypes. We collected five haphazardly selected flowers from each individual plant for phenotypic characterization of two floral traits that function as proxies for flower showiness and flower size. The area of the upwards-facing petal, or standard, plays a key role in flower showiness. We removed standards from all flowers when fresh, and pressed them flat individually in a plant press. We then used scanned images of the standards to measure their area. Flower mass reflects the size of the flower. We estimated size as the dry weight of flowers (calyx and corolla) after removing the standard petal and the pedicel, and brushing off all pollen grains. Flowers were pressed and oven-dried at 40ºC for 48 hours and weighed to the nearest 0.01 mg.

Fruit set. We estimated fruit set in the 40 individual plants in each of the six sites as a proxy for female reproductive success. We labelled a representative flowering twig per plant during peak flowering and collected it a few weeks later when fruit capsules were beginning to brown. In the laboratory, we measured 10 cm of twig to count a) the number of fruits developing normally, and b) scars left by all flowers produced by the twig, clearly visible under a dissecting microscope. From this, we calculated fruit set as the proportion of flowers that develop into a fruit.

Plant genotyping. Fresh twigs were collected from each tagged individual plant and dried in silica gel. After DNA extraction we used a Genotyping-by-Sequencing (GBS) protocol to identify single nucleotide polymorphisms (SNPs) across the genome (Elshire et al. 2011). Two libraries were built for each individual after separate digestions of genomic DNA with PstI and EcoT22I, followed by HiSeq 2000 Illumina sequencing. We implemented SNP calling using the UNEAK pipeline (Lu et al. 2012) in the TASSEL v.3 software package (Bradbury et al. 2007), designed for data sets without a reference genome.

SNP-based relatedness. Pairwise relatedness between all pairs of the remaining 225 individuals (after quality filtering) was estimated from the similarity of their SNP genotypes. To estimate the genome-wide relatedness matrix among all pairs of individuals, we used the kin function of package synbreed in R (Wimmer et al. 2012). Relatedness values are a measure of excess allele sharing compared to unrelated individuals. As a consequence, negative values can be common and correspond to individuals sharing fewer alleles than expected given the sample.

References

• Bradbury, P. J., Z. Zhang, D. E. Kroon, T. M. Casstevens, Y. Ramdoss, and E. S. Buckler. 2007. TASSEL: software for association mapping of complex traits in diverse samples. Bioinformatics 23:2633-2635.
• Elshire, R. J., J. C. Glaubitz, Q. Sun, J. A. Poland, K. Kawamoto, E. S. Buckler, and S. E. Mitchell. 2011. A robust, simple genotyping-by-sequencing (GBS) approach for high diversity species. PloS one 6:e19379.
• Lu, F., J. Glaubitz, J. Harriman, T. Casstevens, and R. Elshire. 2012. TASSEL 3.0 Universal Network Enabled Analysis Kit (UNEAK) pipeline documentation. White Paper 2012:1-12.
• Wimmer, V., T. Albrecht, H. Auinger, and C. Schoen. 2012. synbreed: a framework for the analysis of genomic prediction data using R. Bioinformatics 28:2086-2087.

All data and script files are *.csv or *.txt and can be opened in any text editor or in R.