System and study sites
Aconitum gymnandrum Maxim. (Ranunculaceae) is an annual herb widely distributed in alpine meadows (1600–3800 m) on the Tibetan Plateau, China. Plants commonly bloom from June through August. Individual plants usually produce one to several erect racemes consisting of 2–30 blue-purple zygomorphic flowers, which typically open sequentially from bottom to top (generally 4–7 flowers open at once), and are pollinated by bumblebees (at the study site mainly by Bombus consobrinus and B. sushkini) [30]. Each flower has 6–14 separate carpels (each with 8–14 ovules) surrounded by 30–90 stamens [57]. The galea (or hood), formed from one of the petaloid sepals, contains two stalked petals with nectaries inside, and two other petals extend and cover the stamens and carpel. A. gymnandrum is self-compatible and strongly protandrous like other species in the genus. The anthers dehisce over 4–5 days and stigmas become receptive 1–2 days after the end of anther dehiscence [58]. Fruit maturation requires 20–30 days.
We studied a population of A. gymnandrum located in the Alpine Meadows and Wetland Ecosystems Research Station of Lanzhou University (E102°53′, N34°55′, 2950 m a.s.l) on the eastern Tibetan Plateau, Gansu Province, China. A. gymnandrum is a native plant species and not endangered, collection of samples for scientific purposes was permitted by local legislation. Research permission was obtained for this project from the State Key Laboratory of Grassland and Agro-Ecosystems, School of Life Sciences, Lanzhou University. Dr. Zhigang Zhao undertook the formal identification of the samples. A voucher specimen has been deposited at the State Key Laboratory of Grassland and Agro-Ecosystems, Lanzhou University (voucher No. ps − 20,060,618–001).
Flowering phenology and floral longevity
From July to August 2013, we characterized seasonal variation in flower longevity, flower size, sex ratio (female vs. male flowers), and plant phenotypic sex, and quantified pollinator visitation. Temperature and precipitation data were obtained from Hezuo weather station, less than 1 km away from our field station. The study population consisted of 586 individuals. Prior to anthesis we randomly tagged 100 plants and recorded the flowering day of all flowers of main inflorescences of individuals. A total of 1085 flowers were marked, and flower longevity was estimated as the duration of combined male and female phases (Figure S1). Anther dehiscence usually occurred soon after flower opening in this species. The male phase began as one anther dehisced and ended when all anthers dehisced and withered. At the end of the male phase, the stigma generally becomes receptive, and we regarded this as the start of the female phase. The end of the female phase was recorded when two stigma branches of each carpel curved down and the stigma surface became brown, accompanied by at least one sepal wilting. On each plant, we sampled two flowers from each of the bottom, middle, and top of the main inflorescence to measure flower size. The size of these six flowers was estimated by measuring the galea height (vertical distance from the base of the flower to the top of galea) using digital calipers (Mahr Federal 16 ER Digital Calliper, Germany) when they individually entered the middle of the male phase (i.e., half of anthers dehisced). In order to estimate the association of the anther number with single flower lifespan, 55 of the 100 plants were randomly selected, and the anther number of the two bottom flowers was counted.
We used flowering time and floral longevity data to calculate the daily population sex ratio and the phenotypic sex, and to examine possible causes associated with floral longevity and duration of the male and female phases.
Pollen deposition and seed set
The six flowers sampled for estimating longevity on each plant were also used to determine components of female reproductive success. The first flower at each of three positions was used to estimate natural seed production (female reproductive success) when fruits matured (N = 296 fruits). The other flower from each position was used for recording stigmatic pollen loads. At the end of the female phase, the stigmas were cut off and mounted on slides and taken back to the lab. Aconitum pollen morphology was obviously different from other co-flowering species and easy to distinguish from interspecific pollen. The deposited pollen was counted directly under a microscope (Olympus BX53) on a total of 298 flowers; interspecific pollen grain loads (mainly Pedicularis kansuensis and Taraxacum mongolicum) were lower than 5%.
We used pollen deposition per flower and seed set data to test the associations of floral longevity, phenotypic sex, and flower phenology on female reproductive success.
Effects of pollen removal and pollen receipt on the duration of male and female phases
We performed artificial pollinations to test whether high pollination levels can affect the duration of the male and female phases. Twenty plants of a similar size were marked near the studied population. For each plant, four flowers at the bottom of the main inflorescences (some individuals had 1–3 lateral inflorescences) were chosen: the first and third flowers received the artificial pollination treatment, while the second and fourth served as controls. As the flowers in the artificial pollination treatment came into the male phase, we gently brushed the dehisced anthers using a fine hairbrush twice every day (in the morning and afternoon). When they entered the female phase, we used dehisced anthers from other plants to touch receptive stigmas twice every day (in the morning and afternoon) until the female phase ended. Flowers were bagged prior to treatment to prevent pollinator access.
Statistical analysis
All statistical analyses were conducted in R version 3.2.3 [59].
Temporal variation in sex ratio and phenotypic sex
The daily population sex ratio was estimated as the number of male divided by the number of female flowers. As the daily floral sex ratio does not consider the mating opportunities for female and male success (which depend on the population floral sex ratio [60, 61]), we estimated the standardized phenotypic sex that incorporates the relative availability of pollen and ovules in the whole population as [60]:
$$ {G}_{i,t}=\frac{f_{i,t}}{f_{it}+{m}_{i,t}{E}_t}, $$
Where Gi, t is the phenotypic sex of plant i on day t, fi, t and mi, t are its numbers of female- and male-phase flowers, respectively, of plant i on day t, and \( {E}_t=\frac{\sum \limits_i{f}_{i,t}}{\sum \limits_i{m}_{i,t}} \). Et shows the relative opportunity of male-phase flowers to sire seeds for a plant when considering population context, as an equivalence factor. Mean phenotypic sex for each plant over the whole season can be estimated as:
$$ {\overline{G}}_i=\frac{\sum_t{G}_{i,t}{n}_{i,t}}{\sum_t{n}_{i,t}}, $$
where ni,t is the number of flowers opened by plant i on day t.
Potential causes affecting floral longevity and the duration of sex phases
We used Generalized linear mixed model to estimate the magnitude of potential factors affecting floral longevity and duration of the male and female phases. In the models, predictors included the first flowering date of individuals (FFD), the day of flower opening relative to the plant’s day of flowering onset (RFD), flower size (galea height) and flower number per plant, the mean air temperature, and precipitation experienced by that flower over the days it was open. Since our dataset had a nested structure and flower-level data because response variables were not fully independent, we considered variation within plants by including plant ID within the population as a random effect in the models. The effect of hand pollination on duration of male and female phases was examined by ANOVA.
Effects of flower longevity, phenology and phenotypic sex on female reproductive success
We conducted generalized linear mixed models at the flower and individual levels to determine the relative effects of flower longevity and other display traits, phenology, and phenotypic sex on the pollen number deposited per flower and seed set. All traits were standardized for a comparison between regression coefficients in the models. In the flower level analysis, when pollen deposition per flower was the response variable, we included the standardized duration of the female phase, flower size, RFD, FFD, and flower number as predictors, and plant ID as a random effect. When seed set was used as the response variable, we included the standardized flower longevity, flower size, RFD, FFD, and flower number as predictors, and plant ID as a random effect. In both models, Poisson error distribution was assigned to response variables. The analyses included 295 flowers on 99 plants. In the plant level analysis, we used mean pollen deposition per flower and mean seed set per fruit as response variables, both with Poisson error distribution. In each model, we used the standardized mean phenotypic sex of individuals, mean flower longevity, mean flower size, and flower number as predictors, and plant ID as a random effect. As the flowers used to score pollen deposition on stigmas were not the same as those used for seed production, we further estimated the relationship between mean pollen deposition per flower and mean seed set of individuals by a regression analysis.