SELECTION ON FLOWERING TIME: AN ADAPTIVE FITNESS SURFACE FOR NONEXISTENT CHARACTER COMBINATIONS

The purpose of this study was to determine how natural selection might have shaped the flowering phenologies currently observed in natural populations by examining the potential mechanisms of selection and generating a hypothetical fitness surface that includes phenotype combinations that do not occur in natural populations. To this end I tagged >3500 individual flowers to track seed set on a per‐day basis, performed a hand pollination experiment, and quantified pollinator visitation throughout the flowering season in a population of Lythrum salicaria. I then used the seed production data to construct a hypothetical adaptive surface for two highly correlated characters: date of initial flowering and flowering duration. I found evidence of temporal variation in seed set that can be explained by temporal patterns of pollinator visitation. There were significant effects of date (linear and quadratic) and number of flowering stems on the number of pollinator visits per patch. A significant date × number of stems interaction indicated that the effect of number of stems was dependent on flowering date. Early in the flowering season pollinator visitation was dependent on the density of plants in flower. However, late in the season pollinators stopped visiting the patch for reasons unrelated to flowering plant density. A hand‐pollination experiment demonstrated that the paucity of pollinators late in the season resulted in late‐season pollinator limitation. Late‐season pollinator limitation could explain the shape of the fitness surface. The shape of the estimated fitness surface indicates saturating directional selection on flowering duration and a combination of directional and stabilizing selection on date of flowering initiation. The estimated fitness surface has a single peak located at the earliest date of flowering initiation and the longest flowering duration. This fitness peak is at the same point as the point of highest fitness in the actual fitness distribution. Areas of the phenotype space representing nonexistent phenotype combinations were the locations of fitness valleys on the hypothetical fitness surface. This suggests that the correlation between date of initial flowering and flowering duration is the result of past selection.