Selection favours high spread and asymmetry of flower opening dates within plant individuals

Natural selection on traits expressed repeatedly by individuals is usually investigated with a focus on mean values, although within‐individual trait distributions often differ also in other aspects, such as their spread and shape. In plants producing multiple flowers during a season, there might not be a single optimal flowering time, but rather an optimal distribution of flower opening dates. This optimal distribution might depend on both resource allocation patterns and interactions with the abiotic and biotic environment. In this study, we quantified mean, variance, skewness and kurtosis of 495 individual flowering schedules (5287 flowers) over 3 years, and assessed phenotypic selection on these aspects of the within‐individual distribution of opening dates in the perennial herb Lathyrus vernus. We also explored how selection on within‐individual variation in flowering schedules was related to effects on two fitness components: fruit set and the proportion of seeds escaping pre‐dispersal predation. Within‐individual variation in phenology was larger than, or at least similar to, among‐individual variation in all years. We found phenotypic selection on several aspects of individual flowering schedules. In 1 year, selection favoured plants with higher variance in opening dates, and this coincided with a higher fruit set in plants with an increased spread of the flowering schedule. In two of the study years, selection favoured a higher asymmetry of the flowering schedule, and plants with more right‐skewed distributions had higher fruit set and higher proportions of seeds escaping predation. Both fruit set and seed predation increased with an earlier mean flowering, resulting in no net selection on mean flowering date. Synthesis: Our results suggest that phenotypic selection on the spread and shape of flower opening date distributions might be at least as important as selection on the mean flowering date. In a broad sense, this implies that we should consider the entire trait distribution if we aim to understand the evolution of traits that are expressed multiple times within individuals.