Florivory indirectly decreases the plant reproductive output through changes in pollinator attraction

Abstract Species often interact indirectly with each other via their traits. There is increasing appreciation of trait‐mediated indirect effects linking multiple interactions. Flowers interact with both pollinators and floral herbivores, and the flower‐pollinator interaction may be modified by indirect effects of floral herbivores (i.e., florivores) on flower traits such as flower size attracting pollinators. To explore whether flower size affects the flower‐pollinator interaction, we used Eurya japonica flowers. We examined whether artificial florivory decreased fruit and seed production, and also whether flower size affected florivory and the number of floral visitors. The petal removal treatment (i.e., artificial florivory) showed approximately 50% reduction in both fruit and seed set in natural pollination but not in artificial pollination. Furthermore, flower size increased the number of floral visitors, although it did not affect the frequency of florivory. Our results demonstrate that petal removal indirectly decreased 75% of female reproductive output via decreased flower visits by pollinators and that flower size mediated indirect interactions between florivory and floral visitors.


| INTRODUC TI ON
Myriad species interact indirectly with each other in nature, which is known as "indirect effects." Indirect effects are classified into two categories: density-and trait-mediated. There is increasing evidence that trait-mediated indirect effects can link multiple interactions (Ohgushi, 2005;Ohgushi, Schmitz, & Holt, 2012). In general, organismal traits are variable among individuals due to genetic, stochastic, and environmental effects (Doebeli, 1996;. Also, some traits, such as plant-induced defenses may be altered through interactions with other organisms (Poelman & Kessler, 2016). To what extent does the trait determine the strength of these trait-mediated indirect interactions? This question has been explored by theoretical (Luttbeg, Rowe, & Mangel, 2003;Holt & Barfield, 2012), and empirical studies in predator-prey systems (Ovadia & Schimitz, 2002;Matassa & Trusell, 2014;Gravem & Morgan, 2016) and in plant-herbivore systems (Ohgushi, 2005;Freeman, 2006;Sendoya & Oliveira, 2015). These studies suggest that trait can change the strength of indirect interactions in a wide variety of systems.
To tackle this critical issue, flower-insect interactions provide an excellent system. This is because flowers interact with not only pollinators but also floral herbivores (i.e., florivores), and flower traits such as flower size are critical to the strength of their interaction (McCall & Irwin, 2006). Furthermore, different pollinators and/or florivores respond to the floral traits in a species-specific manner (Tsuji & Sota 2011, Tsuji & Sota, 2013Antiqueira & Romero, 2016), and these floral visitors can affect each other (Romero, Antiqueira, & Koricheva, 2011;Fukano, Tanaka, Farkhary, & Kurachi, 2016), suggesting that the effect of floral traits on the floral visitors may be altered by species identity of floral visitors. Damage to floral tissues results in changes in flower size, shape, and nectar production (Krupnick & Weis, 1999;Strauss & Whittall, 2006). In this context, florivory can decrease fruit and seed set, either directly or indirectly via decreasing pollinator attraction (Krupnick & Weis, 1999;Mothershead & Marquis, 2000;Leavitt & Robertson, 2006;McCall & Irwin, 2006;Strauss & Whittall, 2006;Sánchez-Lafuente, 2007;Carezza et al., 2011). Among a wide range of flower traits, flower size has been well studied showing that flower size can alter the strength of flower-florivore and flower-pollinator interactions (for florivores: McCall & Irwin, 2006;Teixido, Mendez, & Valladares, 2011;McCall & Barr, 2012;for pollinators: Willson, 1979;Bell, 1985;McCall & Irwin, 2006;Lobo, Ramos, & Braga, 2016;Sletvold & Agren, 2016). To examine whether flower size alters the strength of flower-florivore-pollinator interactions, we studied the interaction of Eurya japonica plants and its associated pollinators and florivores. As E. japonica has a long bud period of several months, florivory will likely occur before pollination. Florivores attacking E. japonica often consume stigmas and all petals (personal observation). Florivory on stigmas may directly decrease fruit and seed production and flower with a smaller size by florivory may indirectly decrease the reproductive output via decreasing pollinator attraction. Antiqueira and Romero (2016) reported that florivory lost floral symmetry of Rubus rosifolius and decreased pollinator attraction. However, it is unlikely to occur in E. japonica, because florivory to folded petals of flower buds does not cause asymmetry damage.
We first examined whether artificial damage before pollination directly and/or indirectly decreased fruit and seed production. Then, we examined whether bud size determined flower size because florivory occurred in bud period and floral visitors were attracted in blooming period. We examined whether the flower size affected florivory level and/or the number of visitors on blooming plant, based on the fact of a strong positive correlation between bud size and flower size.
This shrubby plant produces flower buds in summer, blooms in early spring, and bears mature fruits in autumn. Flowers are attacked by moth larvae including Ourapteryx nivea, Alicis angulifera, Somena pulverea, and Chloroclystis excise during bud and blooming stages (Tsuji & Sota, 2013), and these moth caterpillars damage approximately 3% of male flowers and <0.1% of female flowers (Tsuji & Sota, 2013).

| Field experiments
We conducted a field survey and experiments at Kozagawa in Wakayama prefecture, Japan. To examine how florivory affects fruit and seed production, we applied artificial florivory to female flowers for two reasons. First, female fitness can be easily and accurately estimated than male fitness (Conner, 2006)

| An experiment testing direct effect of floral damage on fruit set
Seven plants have four treatments designed in a twig unit (Figure 1).
On 29 January 2012, before blooming, we bagged all twigs with flower buds using nylon nets (mesh size: 0.46-0.59 mm). As damaged buds usually retain most of the ovules but not the stigmas, we removed stigmas to simulate florivory (i.e., artificial florivory [AF]).
To examine how AF affects pollination, we set two treatments with artificial pollination (AP) from 2 to 16 March 2012 (i.e., one treatment had AP only and the other had AP after AF), using a brush and mixed pollen collected from five male plants around the female plants. AF and AP were conducted after flowers bloomed because of the difficulty of bud manipulation. We kept untreated and bagged twigs from pollination as a control (C). Then flowers on twigs were pollinated artificially (AP) and naturally (NP). Besides these three treatments, we set another treatment to evaluate the effect of artificial florivory before pollination (AF before AP). Four treatments were set as follows ( Figure 1a To evaluate fruit set, the number of fruits and dropped flowers in the nets was counted. On the same day, we measured the length of fruit on the treated twigs and another untreated twig (N) using a caliper.

| An experiment testing direct and indirect effect of floral damage on fruit and seed sets
In 2013, we bagged treated twigs on 9 February, just before flowering. To examine whether petal damage affects fruit set via changes in pollinator attraction, we removed all petals leaving

7.
Removal of all petals (PR) exposed to natural pollination (NP): the nets of twigs were removed, followed by removal of all petals of flowers on the same day. Thereafter, the uncovered twigs were exposed to natural pollination for the same period for each twig as treatment six did, after which they were covered with the net again.
To examine fruit production, we counted the number of fruits and dropped flowers in the nets from 6 to 18 August 2013. During the same period, to examine seed production, we counted the number of developing seeds and dead ovules in fruits, and measured fruit length on the treated twigs and fruits on another untreated twig (N).

| Measurement of bud and flower size
To confirm whether bud size determines flower size, we measured the size of buds and blooming flowers. We tagged 11 buds each of 15 male and 13 female plants using different color strings fastened on stalks on 4 January 2014 and then measured bud length using a caliper. Furthermore, we measured perianth length as flower size of the tagged flowers on the next day after flowers opened during the period from 13 February to 7 April 2015.

| Record of flower damage
To examine whether bud size determines the proportion of buds with florivory, we checked bud damage and measured bud size of 13 male and 8 female plants on 2 January 2014. We counted the number of intact and damaged buds on ten 10 cm twigs of each plant.
Finally, we measured the length of five buds of each plant using a caliper.

| Sampling of floral visitors
To

| Statistical analysis
To compare fruit and seed set among treatments with considering individual variation, we applied generalized linear mixed models (GLMMs) with a negative binomial distribution and a log link function, as dispersion was too large to use Poisson distribution. We used the number of fruit/developing seeds as response variables.
Model predictors included treatment and log-transformed number of flowers/ovules. Plant individual was used as a random effect.
To test whether the fruit size was affected by treatments, we applied GLMMs with a gamma distribution and a log link function, followed by a chi-square test and pairwise tests. In the 2012 experiment, fruit length, treatment, and plant individual were used as the response variable, explanatory variable, and random effect, respectively. In the 2013 experiment, we measured fruit size from 6 to 18 August 2013, and we included the measurement date as a fixed variable. To compare fruit size among treatments, GLMMs were followed by likelihood ratio test and pairwise tests. We also examined the association between developed seed number in a fruit and the size of fruit using a GLMM with a gamma distribution and a log link function followed by a chi-square test. Fruit length, number of developed seed, and plant individual were the response variable, fixed explanatory variable, and random effect, respectively.
To examine the association between bud size and blooming flower size, we applied a GLMM with a gamma distribution and a log link function, followed by likelihood ratio tests. We used perianth length as the response variable. Bud length, plant sex and their interaction were explanatory variables, as sex significantly affected flower size (df = 2, Likelihood ratio = 22.8, p < .0001). Plant individual was included as a random effect.
To examine the effects of bud size on florivory, we used a generalized linear model (GLM) with a negative binomial distribution due to overdispersion and a log link function, using the MASS package (Venables & Ripley, 2002). The number of damaged buds was the response variable, and the number of observed buds, average bud length, plant sex, and their interactions were fixed explanatory variables. To test the significance of fixed variables, the GLM was followed by an F-test using the car package (Fox & Weisberg, 2011). To
The most abundant floral visitor was Diptera (99.1% and 75.5% in total visitors in 2012 and 2013, respectively), followed by Hymenoptera (0.7% and 21.6%), and Coleoptera (0.1% and 8.6%). The number of Diptera, including flies, was positively affected by flower size in 2012 (Table 3a). In 2013, the number of  Hymenoptera increased with increasing flower size, although the number of Diptera was not significantly affected by flower size (Table 3b) (Table 3a; F 34,1 = 6.7, p = .01 for Diptera).

| D ISCUSS I ON
Petal removal decreased both fruit and seed set by approximately 50% under natural pollination (Figure 2). This implies that petal removal results in 75% reduction ((1 − 0.5 × 0.5) × 100) in female reproductive output under natural pollination, although it did not decrease fruit and seed set when pollen was artificially supplemented ( Figure 2). This suggests that petals are necessary to attract pollinators. As florivores badly destroy petals, damaged petals mediate the indirect interaction between florivores and pollinators. Furthermore, flower size increased the number of floral visitors (Table 3), although it did not affect the intensity of florivory ( Table 2). As a result, flower damage by florivory decreases pollinator attraction and thus also female reproductive output.

| Florivory indirectly decreases reproductive output
In the artificial florivory treatment, flowers that received stigma damage before pollination failed to produce fruits. However, flowers with stigma damage set fruit in 2012. This is because we may not have imposed sufficient damage to the stigmas. In 2013, as all stigmas were carefully removed, there was no fruit production by flowers without stigmas, which is consistent with the previous finding that damage of stigmas in Chamerion angustifolium inhibited pollination and thus decreased fruit set (Ladio & Aizen, 1999;Sheffield, Smith, & Kenan, 2005;Buchanan & Underwood, 2013). Thus, stigma damage of unpollinated flowers is a direct negative effect of florivory on fruit set. Interestingly, our results also suggest an indirect effect of florivory on fruit set; petal removal decreased fruit set when flowers were naturally pollinated, although it did not occur when flowers were artificially pollinated.
As well as fruit set, stigma damage directly and petal damage indirectly decreased seed set in 2013. Although we did not count seed in 2012, damaged flowers can be expected to have less seed. This is because damaged flowers resulted in smaller fruits than intact flowers in both 2012 and 2013, and the number of matured seed was positively related to fruit size.

| Size effect of florivory and floral visitor attraction
Recent studies on how flower size affects florivory have suggested that larger flowers receive greater floral damage (McCall & Irwin, 2006;Teixido et al., 2011;McCall & Barr, 2012). In contrast, we did not detect the effects of bud size, which was positively correlated with flower size, on florivory. This implies that bud size does not affect the strength of flower-florivore interactions.
On the other hand, this study showed that flower size was positively associated with the number of floral visitors. Also, the visitor number was significantly greater in male than female plants, and plant sex and the interaction between plant sex and flower size significantly affected the number of dipteran visitors (99.1% of total visitors) in 2012. These findings suggest that larger flowers are more preferred by insect pollinators, which is consistent with studies on other plant species (Willson, 1979;Bell, 1985;McCall & Irwin, 2006;Lobo et al., 2016;Sletvold & Agren, 2016).

| CON CLUS ION
Our study demonstrated that flower damage decreased female reproductive output via weakened flower visitor attraction, and this is consistent with studies showing that florivory decreases fruit and seed set indirectly via decreasing pollinator attraction (Krupnick & Weis, 1999;McCall & Irwin, 2006;Strauss & Whittall, 2006). In E. japonica, the petal removal decreased flower size and clearly decreased natural pollination. Furthermore, the low-level of flower visitation can be caused by reduced petal size. Thus, the flower size is a critical trait that determines the strength of trait-mediated indirect interactions among florivores, flowers, and pollinators. Also, note that different taxa responded to flower size differently, and their responses differed between years. As abundance and species composition of flower visitors can differ among years (Price, Waser, Irwin, Campbell, & Brody, 2005;Buide, 2006;Sgolastra et al., 2016), the indirect interaction mediated by flower size would temporally vary in its intensity, due to abundance and/or species composition of floral visitors.

ACK N OWLED G M ENTS
We thank Toyohei Saigusa for identification of pollinators. Takashi Ida, Anu Valtonen, and Andrew Letten provided comments. This research was supported by JSPS Grant-in-Aid for JSPS fellows (25451) to K. T. and JSPS KAKENHI (B-25291102) to T. O.

CO N FLI C T O F I NTE R E S T
The authors have declared that no competing interests exist.

AUTH O R CO NTR I B UTI O N S
KT conceived and designed this study, conducted the experiment, and analyzed the data. KT and TO contributed to the writing of the manuscript.