Is increased male flower production a strategy for avoidance of predispersal seed predation in andromonoecious plants?

Abstract Floral gender in angiosperms often varies within and among populations. We conducted a field survey to test how predispersal seed predation affects sex allocation in an andromonoecious alpine herb Peucedanum multivittatum. We compared plant size, male and perfect flower production, fruit set, and seed predation rate over three years among nine populations inhabiting diverse snowmelt conditions in alpine meadows. Flowering period of individual populations varied from mid‐July to late August reflecting the snowmelt time. Although perfect flower and fruit productions increased with plant size, size dependency of male flower production was less clear. The number of male flowers was larger in the early‐flowering populations, while the number of perfect flowers increased in the late‐flowering populations. Thus, male‐biased sex allocation was common in the early‐flowering populations. Fruit‐set rates varied among populations and between years, irrespective of flowering period. Fruit‐set success of individual plants increased with perfect flower number, but independent of male flower number. Seed predation by lepidopteran larvae was intense in the early‐flowering populations, whereas predation damage was absent in the late‐flowering populations, reflecting the extent of phenological matching between flowering time of host plants and oviposition period of predator moths. Seed predation rate was independent of male and perfect flower numbers of individual plants. Thus, seed predation is a stochastic event in each population. There was a clear correlation between the proportion of male flowers and the intensity of seed predation among populations. These results suggest that male‐biased sex allocation could be a strategy to reduce seed predation damage but maintain the effort as a pollen donor under intensive seed predation.


| INTRODUC TI ON
Diverse sexual systems and gender expression in plants are thought to maximize reproductive success through pollen donation and seed production (Barrett & Harder, 2017). Sex allocation theory predicts that resource allocation to male and female functions is determined by the relationship between resource investment in a given sexual function and the fitness gain from that sexual function (Charlesworth, 1991;Charnov, 1982), which results in the evolution of diverse sexual systems. Many studies have reported phenotypic gender variation within and between populations of single species (Barrett, 2002;Charlesworth & Charlesworth, 1981;Lloyd, 1984).
Various ecological factors, such as plant size, resource condition, selfing rate, pollinator availability, and plant architecture, affect sex allocation (reviewed in Barrett, 2002;Barrett & Harder, 2017).
Furthermore, recent studies have reported that not only pollinators (mutualists) but also herbivores (antagonists) can affect the sexual system and phenotypic gender of plants (Clarke & Brody, 2015;Johnson et al., 2015;Wise & Hébert, 2010), although empirical studies demonstrating the significance of herbivory as an agent of selection on floral gender are limited (e.g., Wise & Hébert, 2010).
Because of the sessile nature of plants, phenological events in plants are strongly influenced by abiotic factors within local habitats. This often restricts phenological regulation in plants responding to a selective force acting on specific phenological events, such as flowering and fruiting time. In snowy alpine ecosystems, lingering snow patches create mosaics of local environments in which the time of snowmelt strongly affects the growing period and reproductive schedule of alpine plants (Kudo, 1991). Because flowering time progresses sequentially along snowmelt gradients during the summer, alpine ecosystems provide opportunities to test the ecological significance of phenological variation on reproductive outcomes (Kudo, 2006). Previous studies revealed that pollination success and outcrossing rates in a single species varied greatly among neighboring populations along snowmelt gradients, reflecting the extent of phenological matching between flowering period and pollinator activity (Kameyama & Kudo, 2009;Kudo & Hirao, 2006;Kudo et al., 2011). The intensity of predispersal seed predation may also vary among neighboring populations along snowmelt gradients, but to date, no studies have investigated this possibility. In the present study, we aim to explore how intensity of predispersal seed predation varies seasonally and how alpine plants respond to the local variation in seed predation stress.
Several studies reported that herbivorous damages of flowers (florivory) affect floral gender and breeding system of plants (Ashman, 2002;McCall & Irwin, 2006;Wise & Cummins, 2007). In contrast to florivory that decreases the fitness through both male (pollen donation) and female (seed production) functions, predispersal seed predation (frugivory) intensively decreases the female fitness since frugivores consume developing fruits (Marshall & Ganders, 2001). Because a large amount of resources is wasted by predispersal seed predation, plants are expected to regulate reproductive allocation when the risk of seed predation is high. One possibility is a decrease in resource allocation to female function by which plants can reduce the predation damage during a reproductive event. On the other hand, plants may keep or increase the resource allocation to male function. When the seed predation rate is high and unpredictable in spatial terms, wide pollen dispersal may help to reduce the impact of predation of sired seeds. This theory is analogous to the colonization hypothesis of seed dispersal (Howe & Smallwood, 1982), which postulates that wider seed dispersal results in a higher probability that some seeds will encounter a safe site for survival and establishment.
In the present paper, we conducted a field survey to reveal how floral sex allocation varies among populations of single species under various seed predation damage. For this purpose, we selected an andromonoecious alpine herb, Peucedanum multivittatum (Apiaceae), that is a perennial iteroparous species with obligate outcrossing mating system. Andromonoecy, existence of male (staminate) and perfect (hermaphroditic) flowers within a plant, is a sexual system in which flexible regulation of resource allocation between female and male functions occurs (Lloyd & Bawa, 1984;Spalik, 1991). Male flowers sometimes show higher pollen dispersal ability than perfect flowers in andromonoecious plants (Dai & Galloway, 2012;Schlessman et al., 2004), and siring success often increases as male flower number increases (Elle & Meagher, 2000; but see also Tomaszewski et al., 2018). Therefore, we predicted that the proportion of male flowers would be higher in populations suffering from intensive seed predation than in populations in which such predation damage is less common.
The aim of this study was to detect the expected linkage between floral sex allocation and seed predation pressure in an andromonoecious species across local populations along natural snowmelt gradients. We compared gender expression, fruit-set success, flowering phenology, and predispersal seed predation among local populations of P. multivittatum in areas where the timing of snowmelt differs. We compared both fruit-set success and seed predation damage with respect to flowering phenology. Because pollinators and seed predators are counteracting selective agents (Altan et al., 2010;Elzinga et al., 2007;Johnson et al., 2015;McCall & Irwin, 2006), variation in sex allocation among populations needs to be interpreted in terms of both pollination success and avoidance of predation damage. The questions addressed in the present study are as follows: (a) How do fruit-set rate and seed predation rate vary among local populations having different flowering phenology? (b) How does floral sex allocation, that is, the proportion of male flowers, vary among local populations along snowmelt gradients? (c) Is the pattern of floral gender variation related to fruit-set success and/or risk of seed predation, or does it simply reflect plant size? Based on the results, we discuss the possibility that plants employ a sex allocation strategy to avoid predispersal seed predation.

| Plant material
Peucedanum multivittatum Maxim. (Apiaceae) is an alpine herb that inhabits snow meadows in alpine regions of Japan. Like many other apiaceous species, P. multivittatum is andromonoecious; it is characterized by a terminal umbel composed of male and perfect flowers with a few lateral umbels composed of only male flowers (Kudo, 1997). Protandrous flowering is common in perfect flowers, in which anthers dehisce soon after opening and pistils develop after the anthers are shed in the terminal umbel (Kudo, 1997). Flowering of male and perfect flowers occur simultaneously within a terminal umbel on which male flowers and perfect flowers at male phase are mostly identical in size and shape. The flowering of lateral umbels usually begins after the anthesis of the terminal umbel within a plant.
Thus, dichogamous flowering, in which male and female phases occur at different times, is sequentially deployed within a plant.
Flowering in P. multivittatum usually occurs between mid-July and late August, depending on the time of snowmelt, and the flowering period within a population is about 10-14 days (Kudo, 1997;Kudo & Hirao, 2006). Fruits usually mature two weeks after flowering. This species is an obligate outcrosser, and major pollinators are dipteran insects, such as syrphids, and other flies. Developing seeds are often predated by specific fruit-mining lepidopteran larvae (Phaulernis fulvijuttella; Epermeniidae). Our preliminary observation revealed that the predator moth deposited a few eggs on peduncles of terminal umbels during the flowering period (Figure 1), and major oviposition period was from mid-to late July in the study site.

| Study site
This study was conducted in an alpine area of the Taisetsu Mountains, Hokkaido, northern Japan (43º32-33′N, 142º51-53′E).
This mountain area is characterized by cold, snowy winters and mild, wet summers. The treeline is at around 1,500-1,600 m elevation.

| Statistical analysis
All statistical analyses were conducted using R version 3. Flowering onset time means a period in which flowering started in each plot expressed by rank at 5-day intervals since July 1, ranging from rank 1 (July 1-5) to 12 (August 26-30).
Predation damage to fruits was compared among plots and between years using a GLM postulating a negative binomial error distribution to reduce overdispersion because many zero values were included in the data. In this GLM, the number of damaged fruits was a dependent variable, plot and year were explanatory variables, and the number of fruits was an offset term after logarithmic transformation. When significant differences were detected among populations, Tukey's post hoc test was performed between individual plots using pooled data across years. Furthermore, the effects of flowering time and floral gender on seed predation rate were analyzed across plots and years using GLMM postulating a negative binomial error distribution using the "ADMB" package. In the GLMM, the number of damaged fruits was a dependent variable, male flower number, perfect flower number, and flowering onset time were explanatory variables, the number of fruits was an offset term after logarithmic transformation, and plot nested by year was set as a random factor.

| Flowering phenology
The flowering period of P. multivittatum ranged from mid-July to early September, reflecting the progress of snowmelt across populations ( Figure 3, Table 1

| Effects of plant size on flower and fruit productions
Generalized linear models and multiple comparison tests revealed significant size differences among five populations (Figure 4).

| Floral gender variation across populations
There were significant differences in the numbers of male and perfect flowers per umbel among populations (Figure 6a

| Fruit-set success across populations
The mean fruit-set rate in each population ranged from 0.24 to 0.63 across plots and years. Although significant variation in fruit-set rate was detected among plots (p < .001) and between years (p < .001) by  (Table 2).

| Effects of flowering time and gender variation on predispersal seed predation
There were significant differences in the predation damage to fruits among plots (p < .001) and between years (p < .001) by the GLM (Figure 8a). Predation damage was intense at the early-snowmelt plots (HA, PK, KE, HL), ranging from 0.29 to 0.88, whereas there was no predation damage at the late-snowmelt plots (KD, KL, KT).
The GLMM to test factors affecting predation damage revealed that predation rate was negatively related to flowering time, while both male and perfect flower productions were independent of predation damage (Table 3) There was a significant correlation between the risk of predation damage (mean predation rate in each plot) and the proportion of male flowers across populations and years (r 2 = .64, p < .0001; Figure 9). This indicates that the proportion of male flowers was higher in the populations suffering from continuous severe predation damage.

| D ISCUSS I ON
The major flowering period of individual populations ranged from mid-July to late August, depending on snowmelt time, and seed predation damage was most intense in early-snowmelt populations.
Floral sex allocation varied significantly along the snowmelt gradient, and male flower production was positively related to the risk of seed predation at population level. These results support our prediction that male-biased sex allocation is selected in andromonoecious species subject to intensive seed predation. To our knowledge, this is the first report of the linkage between floral sex allocation and predispersal seed predation among local populations.
Intensive seed predation in the early-snowmelt populations was Basal diameter (mm) No. fruits  (Figure 8b), indicating a stable oviposition period between years. When predispersal seed predation is severe in early season, later flowering trait has evolved as a predator avoidance strategy in several species (Ehrlén, 2015;Pilson, 2000;Sercu et al., 2020;Valdés & Ehrlén, 2017). However, the regulation of flowering phenology may be less effective in alpine snow meadows because flowering time is strongly determined by snowmelt time and actual snowmelt time highly varies from year to year (Kudo, 2006).
Actually, seed predation rates of HL and HC were two times higher in the year when snowmelt occurred earlier than usual years (Figure 8).
In contrast to the trend in predation damage, there was no clear trend in fruit-set success among populations irrespective of large variation in flowering time. Major pollinators of this species are dipteran insects that are most common throughout the summer (Mizunaga & Kudo, 2017). Thus, difference in flowering period may be less important for pollination success in this species. Both total and perfect flower productions increased with plant size, and the number of perfect flowers was positively related to fruit-set success.
This means that plants having large umbels can have higher fruitset success, probably due to higher attraction of dipteran insects (Inouye et al., 2015). On the other hand, variation in floral sex allocation among populations was independent of pollination success.
Relationships between size and sex allocation have been reported previously in many animal-pollinated plant species, where female-biased sex allocation is common in larger plants (e.g., de Jong & Klinkhamer, 1989;Klinkhamer & de Jong, 1997).   Figure 6). Thus, the variation in floral sex allocation among local populations cannot be explained by plant size.
There are only a few studies on the relationship between floral gender and floral herbivory in andromonoecious plants. In two apiaceous species (Heracleum lanatum and Pastinaca sativa), for instance, herbivorous damage to first umbels resulted in increased production of perfect flowers in late-developing umbels by the regulation of resource allocation within plants (Hendrix, 1984;Hendrix & Trapp, 1981). In another andromonoecious herb (Solanum carolinense; Solanaceae), the proportion of male flowers was negatively correlated with the intensity of weevil florivory among populations, because the production of many ovaries is important to maintain seed production under intensive florivorous conditions (Wise & Cummins, 2007). These compensative responses to floral herbivory indicate an opposite trend of sex allocation to that observed in the present study. This discrepancy means that populations are subject to different selective forces depending on whether they suffer from floral herbivory or predispersal seed predation. In the case of floral herbivory, resource investment in fruit development usually occurs after herbivory damage. In the case of seed predation, however, a large amount of resources has been invested in fruit development when seeds are predated; thus, compensative responses after seed predation may be difficult.
Predispersal seed predation results in an intensive reduction in female success in terms of seed production. The risk of seed predation was simply determined by flowering time of individual populations, and floral sex allocation of individual plants was independent of predation rate within a population (Table 3). It suggests that seed predation is a stochastic event in each population irrespective of floral gender, probably because predator moths do not discriminate the proportion of male flowers of individual umbels at the time of oviposition. When the risk of seed predation is high, plants may reduce the waste of resources due to seed predation by reducing fruit production during a single reproduction.
As mentioned before, P. multivittatum is a perennial iteroparous species, and about 70% of fruiting plants do not produce flowers next season, indicating large cost of fruit production. By saving F I G U R E 8 Comparison of predation rates of developing fruits among plots and between years (a), and the relationship between flowering onset time and predation rate shown by logistic regression (b  the resource investment in excess fruit production in each reproductive event, plants may be able to increase the frequency of reproductive events during the lifetime. Furthermore, plants can produce more male flowers at the expense of fruit production. Improvement of male success in terms of pollen donation could be a possible strategy against seed predation, that is, a spatial avoidance strategy by pollen dispersal. Although we did not evaluate the relationship between the number of male flowers and pollen donor success in this species, previous studies have reported that male flowers in andromonoecious plants have wider pollen dispersal and higher siring success than hermaphroditic flowers (Dai & Galloway, 2012;Elle & Meagher, 2000;Schlessman et al., 2004). Therefore, increased male flower production could be an effective avoidance strategy against intensive predispersal seed predation.
The present study suggests that predispersal seed predation can be a selective agent for floral sex allocation at a local scale. Flowering of snow-meadow plants progresses sequentially across local populations along snowmelt gradients, resulting in the restriction of gene flow via pollination process among local populations (i.e., phenological isolation; Hirao & Kudo, 2004, 2008. Because most seeds of P. multivittatum are dispersed by gravity around mother plants, long-distance seed dispersal seems to be rare. Thus, local adaptation of defense strategies may exist in snowy alpine ecosystems. To test this prediction, clarifications of genetic differentiation among populations and the relationship between male flower production and success as a pollen donor under different seed predation situations are required.

ACK N OWLED G M ENTS
We are grateful to Tetsuo Kohyama for the identification of moth species and Kota Onizawa for help in the field survey. This study is supported by JSPS KAKENHI (No. 15H002641; 17K07551).

CO N FLI C T O F I NTE R E S T
We declare that there is no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data set used for the analyses is archived in Dryad doi: (https:// doi.org/10.5061/dryad.b5mkk whcq).