Drought reduces floral resources for pollinators

Climate change is predicted to result in increased occurrence and intensity of drought in many regions worldwide. By increasing plant physiological stress, drought is likely to affect the floral resources (flowers, nectar and pollen) that are available to pollinators. However, little is known about impacts of drought at the community level, nor whether plant community functional composition influences these impacts. To address these knowledge gaps, we investigated the impacts of drought on floral resources in calcareous grassland. Drought was simulated using rain shelters and the impacts were explored at multiple scales and on four different experimental plant communities varying in functional trait composition. First, we investigated the effects of drought on nectar production of three common wildflower species (Lathyrus pratensis, Onobrychis viciifolia and Prunella vulgaris). In the drought treatment, L. pratensis and P. vulgaris had a lower proportion of flowers containing nectar and O. viciifolia had fewer flowers per raceme. Second, we measured the effects of drought on the diversity and abundance of floral resources across plant communities. Drought reduced the abundance of floral units for all plant communities, irrespective of functional composition, and reduced floral species richness for two of the communities. Functional diversity did not confer greater resistance to drought in terms of maintaining floral resources, probably because the effects of drought were ubiquitous across component plant communities. The findings indicate that drought has a substantial impact on the availability of floral resources in calcareous grassland, which will have consequences for pollinator behaviour and populations.

One of the major aspects of climate change is predicted increases in the occurrence and intensity of drought (periods of abnormal precipitation deficit) across many regions worldwide (Dai, 2013;IPCC, 2014).
Drought has been identified as a major threat to pollinators and pollination (Brown et al., 2016), and will act primarily through changes in the availability of floral resources upon which pollinators rely (Thomson, 2016). Broadly, a reduction in water availability will affect photosynthetic rate (Pinheiro & Chaves, 2011), leading to fewer resources available to plants for investment into reproduction and flowers.
Drought has been shown to reduce flower size (Halpern, Adler, & Wink, 2010), the number of flowers per plant (Burkle & Runyon, 2016), result in flowers that produce less pollen (Waser & Price, 2016) and a lower proportion of viable pollen grains (Al-Ghzawi, Zaitoun, Gosheh, & Alqudah, 2009), and affect floral volatiles, which can influence the attractiveness of flowers to pollinators (Burkle & Runyon, 2016). In general, water availability has been found to affect nectar in terms of volume (Carroll, Pallardy, & Galen, 2001;Gallagher & Campbell, 2017;Halpern et al., 2010;Lee & Felker, 1992;Villarreal & Freeman, 1990) and sometimes also sugar concentration (Waser & Price, 2016;Wyatt, Broyles, & Derda, 1992;Zimmerman & Pyke, 1988). Changes in nectar volume or sugar concentration are likely to affect pollinator foraging behaviour because flower selection can be influenced by subtle differences in these factors (Cnaani, Thomson, & Papaj, 2006). Furthermore, changes in nectar will affect the energy intake rate of pollinators (Schweiger et al., 2010), which is optimized at intermediate sugar concentrations (Borrell, 2007). Effects of drought on other floral traits have been shown to result in fewer visits by bees (Al-Ghzawi et al., 2009;Gallagher & Campbell, 2017), although Burkle and Runyon (2016) found that the response of other pollinator groups was both plant and pollinator species-specific, with visitation rate increasing in some circumstances. More broadly, changes in the overall availability of floral resources will affect pollinators at the population level (Baude et al., 2016;Carvell et al., 2006Carvell et al., , 2017Roulston & Goodell, 2011).
There is likely to be much variation in the responses of floral resources of different plant species to drought, depending on aspects of their life history. For example, long-lived species may be adapted to respond to drought by reducing their investment in floral resources, due to trade-offs between survival and reproduction (Galen, 2000), whereas short-lived species may maintain a high level of investment in floral resources during drought to ensure reproduction over their lifetime. The reproductive system of a plant is also likely to be important because plants with mixed mating systems can switch reproductive strategy in response to changes in environmental conditions (Goodwillie, Kalisz, & Eckert, 2005). For example, plants may move from outcrossing towards selfing when under environmental stress (Levin, 2010) such as drought (Kay & Picklum, 2013), although the opposite may also be true (Bishop, Jones, O'Sullivan, & Potts, 2017). Plants without a mixed reproductive strategy, or with a greater dependency on outcrossing, are therefore expected to maintain greater investment in floral resources under adverse conditions. As a result, changes in the production of floral resources by a plant can be either an adaptive response or a negative consequence of drought stress.
Given that responses to drought are often plant species-specific, the impacts on plant communities are likely to depend on species composition (Grime et al., 2000). For example, a greater diversity of plant strategies may provide greater community resilience to drought events (Zwicke, Picon-Cochard, Morvan-Bertrard, Prud'homme, & Volaire, 2015), including in the provision of floral resources.
Although the overall provision of floral resources by a plant community may not change if some drought-resistant species replace drought-sensitive ones, changes to the diversity and quality of floral resources may affect individual pollinators (Kaluza et al., 2017;Vaudo, Tooker, Grozinger, & Patch, 2015) and the diversity of the pollinator community (Ghazoul, 2006). Differences in the functional traits of a plant community may also influence the resilience of that community and of its ecosystem functions to drought (Isbell et al., 2015;Oliver, Isaac et al., 2015). In particular, functional traits that relate to water uptake, water use and water retention, such as deep rooting structures, are likely to enable drought tolerance and therefore affect the overall performance of plants that are subjected to drought. This is likely to result in a broad range of benefits, including in terms of a greater production of floral resources. Similarly, greater diversity of functional traits may provide benefits to the plant, in terms of overall performance and resistance to drought, due to niche complementarity (Gross, Sudings, Lavorel, & Roumet, 2007;Gubsch et al., 2011), for example if different root types are collectively able to utilize a greater proportion of available water.
Existing studies of the response of floral resources and pollinators to drought have commonly been conducted on plants of arid and semiarid regions (Al-Ghzawi et al., 2009;Takkis, Tscheulin, Tsalkatis, & Petanidou, 2015), where drought events are relatively common, or in the laboratory (Villarreal & Freeman, 1990). There are few such experiments in temperate regions where drought is expected to increase in frequency and severity due to climate change (Dai, 2013;IPCC, 2014).
Ecosystems in these regions may be more severely affected because drought has not previously been an important environmental factor (Chen, van der Werf, de Jeu, Wang, & Dolman, 2013). To develop a more complete understanding of the potential impacts of drought over the coming decades, it is essential to investigate these impacts on a variety of plant communities in more temperate regions.
In this study, we investigated the impacts of an experimental summer drought event on floral resources in calcareous grassland in order to better understand potential effects of climate change on insect pollinators. Specifically, we tested: (1) how drought affects the availability of floral resources at the flower and community level; and (2) whether responses to drought vary among communities of different plant functional composition. To do so, we used four different experimentally sown plant communities of calcareous grassland.
The plant communities were derived from three sets of species grouped using database-derived information on their functional traits, particularly root traits and specific leaf area, which are likely PHILLIPS ET AL.

| 3227
to affect resistance to drought (Buckland, Grime, Hodgson, & Thompson, 1997;Cantarel, Bloor, & Soussana, 2013;Comas, Becker, Cruz, Byrne, & Dierig, 2013). In this way, we were able to explore how responses to drought varied across plant communities based on their functional trait composition, and control for the effects of species composition to ensure the general applicability of the results.
We hypothesized that: 1. Drought (reduction in water availability) leads to lower photosynthetic rate, resulting in reduced sugar concentration in nectar.

| Experimental design
The experiment was conducted in Wiltshire, UK (50.991207°N, À2.069834°W) using sown plant communities in ex-arable calcareous grassland (Fry et al., 2018). The plant communities consisted of typical calcareous grassland species (from UK National Vegetation Classification Community CG3a Bromus erectus grassland; Rodwell, 1992) and represented realistic community structures for the region. Four plant communities with contrasting functional trait compositions were sown onto bare soil in May 2013. The traits used to differentiate species were hypothesized to exert differing effects on soil carbon and nitrogen cycling, but are also likely to be related to water acquisition, water use and resistance to drought (Buckland et al., 1997;Cantarel et al., 2013;Comas et al., 2013). Functional group 1 (FG1; 16 species) consisted of species with variable longevity, deep tap or stoloniferous roots, and large, thin leaves, which we hypothesized to exhibit low nutrient cycling and poor resistance to drought (Buckland et al., 1997;Gould, Quinton, Weigelt, De Deyn, & Bardgett, 2016). Functional group 2 (FG2; 15 species) consisted of long-lived species with a shallow tap root and small rosettes, which we hypothesized to exhibit low nutrient cycling but with fairly good resistance to drought. Functional group 3 (FG3; 20 species) consisted of long-lived species with shallow, fibrous roots and thick, fleshy leaves, consistent with high nutrient cycling and fairly good resistance to drought. The fourth plant community contained all three functional groups (FG123; 51 species).
The plant species within each functional group are listed in Table S1.
Plant communities were sown into 8 9 8 m plots, separated by 2 m guard rows. The site was divided into six rows, forming experimental blocks, and each plant community was randomly allocated to a single plot in each block (n = 24), in order to control for spatial, edge and neighbouring effects ( Figure S1). The number of seeds applied to each plot was determined by the mean seedbank density for each species (from the LEDA Traitbase; Kleyer et al., 2008), the mean seed weight for each species (from the Kew SID; Royal Botanic Gardens Kew, 2016), and scaled for number of functional groups per plot.
Each of the 24 plots contained three subplots (1 9 1.5 m) which were at least 1 m from each other and from the edge of the plot.
Each subplot was given one of three treatments: (1) Drought (D), covered with a transparent roof to exclude rain, simulating drought; (2) Control (C), not covered with a roof; and (3) Roofed control (R), covered with a transparent roof with 5 cm holes, allowing rain to pass through, but controlling for possible roof effects such as increased temperature and decreased light intensity (Vogel et al., 2013). Control and roofed subplots received ambient rainfall during the 6-week period. Drought shelters were in place for 6 weeks, in two successive years, between 28th May and 11th July in 2015 between 6th June and 13th July in 2016. The 6-week period represents a one in one hundred year drought event, and was simulated

| Flower and raceme scale
Three plant species, which were sown as part of the experimental study, Lathyrus pratensis (from FG2), Onobrychis viciifolia (from FG3), Prunella vulgaris (from FG2), were selected based on their abundance, cover, floral traits, flowering period and ease of nectar extraction.
L. pratensis is hermaphrodite and normally cross-pollinated though has some capacity for selfing (Fitter & Peat, 1994). O. viciifolia is hermaphrodite (Fitter & Peat, 1994) and considered to be obligate cross-pollinated (Hanley, Franco, Pichon, Darvill, & Goulson, 2008); self-pollination is possible but results in lack of vigour and few if any viable seeds (Hayot Carbonero, Mueller-Harvey, Brown, & Smith, 2011). P. vulgaris is gymnomonoecious and either cross-or self-pollinated (Fitter & Peat, 1994), with a high capacity for autonomous selfing (Ling et al., 2017). Whilst it would have been preferable to have one plant species from each functional group, none of the plant species in FG1 met these criteria at the time of the study. Subplots which contained only FG1 were therefore only used for the community-scale studies (see below). Racemes of these species were randomly selected from all available flowering racemes across subplots.
The number of racemes tested per subplot varied according to their availability. Racemes were selected from different individual plants if possible, but this was not always clear, due to the vegetative spread of some species. Racemes were covered with a fine mesh bag for 24 hr to prevent flower visitation by invertebrates and labelled to ensure they were only used once. After 24 hr, bags were removed and the number of flowers on the raceme was recorded. Up to three flowers per raceme (if available) were randomly selected to measure nectar volume and sugar concentration, following standard protocols (Corbet, 2003). As nectar was not removed from flowers before applying bags, the amount of nectar in flowers after 24 hr represented a combination of standing crop and 24 hr accumulation. Nectar volume (ll) was measured using glass microcapillary tubes (sizes 0.5, 1, 2, and 5 ll microcaps, Drummond Scientific, Broomall PA, USA). Nectar sugar concentration (mg/mg) was measured using a When a concentration reading was absent because nectar volume was too small, a value was used that was the mean of measurements from flowers of the same plant species and in the same treatment.

| Community scale
Surveys were carried out in 1 m 2 quadrats in the centre of each subplot of all treatment plots. All flowering plants were identified to species level and the number of floral units was recorded for each species. A floral unit was defined as one or multiple flowers that can be visited by an insect without having to fly between them (following Baldock et al., 2015). Surveys were completed at each subplot on two occasions, on different days, between 18th and 22nd July.
The survey order of plots was randomized. Subplots within each plot were surveyed consecutively but in a randomized order. drought treatment and their interaction, with a priori pairwise contrasts used to examine differences between control, roofed control and drought treatments within each plant species ('lsmeans' package; Lenth, 2016). Plot and raceme identity were included as a random effect for flower scale measurements. Response variables were transformed where required in order to meet model assumptions (see Table S2). The proportion of flowers containing nectar was analysed as above, but using a generalized linear mixed effects model (GLMM) with binomial error structure ('lme4' package; Bates, M€ achler, Bolker, & Walker, 2015). Likelihood ratio tests (LRT) were used to assess if the main effects improved the GLMM fit.

| Statistical analyses
For community-scale analyses, LMMs or GLMMs were used as above. The data from the two survey periods were summed, because the time between survey periods was short. For the species richness of floral units, a LMM was used with the number of insect-pollinated plant species present in the plot, plant community, drought treatment, and the interaction between plant community and drought treatment as explanatory variables. For the number of floral units, a negative binomial GLMM was used (Bates et al., 2015), due to overdispersion of count data. Explanatory variables were plant community, drought treatment and their interaction, with a priori pairwise contrasts used to examine differences between control, roofed control and drought treatments within each plant community. In all cases, random variables were plot, nested within row, and the significance of main effects tested using LRTs. Full details of the statistical analyses can be found in Table S2.

| Flower and raceme scale
Across the three plant species, 437 racemes were selected; of these, 372 had flowers remaining after 24 hr. Nectar was collected from the flowers of between 37 and 44 racemes per plant species per drought treatment. There were large differences between plant species in terms of nectar volume per flower (F 2,274 = 225.07, p < .001), nectar sugar concentration per flower (F 2,216 = 6.86, p = .001) and weight of sugar in nectar per flower (F 2,274 = 184.80, p < .001) ( Table 1). Flowers of L. pratensis had by far the greatest volume of nectar, followed by P. vulgaris, and then O. viciifolia (Table 1). There was no interaction between drought treatment and plant species (Table S2) Table S2). Sample size (n) refers to the number of individual flowers tested and the number of racemes tested.

| Community scale
F I G U R E 1 The effect of drought treatment on the mean proportion of flowers per raceme that were found to contain nectar AE SE, 24 hr after bagging, for each plant species (Lathyrus pratensis, Onobrychis viciifolia and Prunella vulgaris). Treatment refers to Control (C), Roofed control (R) and Drought (D). Sample sizes (n) are indicated by numbers within bars. Levels of significance between a priori contrasts are indicated by symbols ( + p < .06, *p < .05, **p < .01, ***p < .001) scales. Primarily, there were fewer flowers overall, and fewer of those flowers contained nectar. However, the mechanisms by which this occurred differed among both plant species and plant communities.
Among plant species, L. pratensis and P. vulgaris responded by reducing the proportion of flowers that contained nectar on each raceme, whilst O. viciifolia responded with a reduction in the number of flowers per raceme. The roofed control treatment appeared to have a similar, but lesser effect to the drought treatment on all three plant species, suggesting that some of this response was due to other microclimatic effects of the roof such as increased temperature and decreased light intensity. Producing a greater proportion of nectarless flowers could be a mechanism for conserving resources without reducing reproductive potential, because it may be less costly for pollinators to visit these flowers than to discriminate between those that are secreting and nonsecreting (Bell, 1986). Similarly, a previous study showed that nectarless flowers can be produced in response to environmental stress (Petanidou & Smets, 1996; but see Takkis et al., 2015). Alternatively, diverting resources into fewer flowers may allow nectar volume and sugar concentration to be maintained in those flowers, as we observed for O. viciifolia. The difference in response of O. viciifolia, compared to the other two species, may be due to differences in breeding system. O. viciifolia is obligate cross-pollinated (Hanley et al., 2008), so maintaining nectar in fewer flowers may be important in ensuring pollinator visitation. In contrast, L. pratensis has the capacity to self-pollinate (Fitter & Peat, 1994) and P. vulgaris has a high capacity for autonomous selfing (Ling et al., 2017). P. vulgaris and L. pratensis may therefore have a lower reproductive cost of stopping nectar production in some flowers as those flowers may still be able to self-fertilize without pollinator visitation.
For flowers that did contain nectar, drought had no effect on the volume or sugar concentration of that nectar. This is contrary to our first and second hypotheses and contrasts with many previous studies that have demonstrated changes in nectar volume in response to water availability (Carroll et al., 2001;Gallagher & Campbell, 2017;Lee & Felker, 1992;Villarreal & Freeman, 1990;Wyatt et al., 1992;Zimmerman & Pyke, 1988). It is possible that the reduction in soil The effect of drought treatment on the number of flowers per raceme, 24 hr after bagging, for each plant species (Lathyrus pratensis, Onobrychis viciifolia and Prunella vulgaris). Treatment refers to Control (C), Roofed control (R) and Drought (D). Sample sizes (n) are indicated by numbers above boxplots. Levels of significance between a priori contrasts are indicated by symbols ( + p < .06, *p < .05, **p < .01, ***p < .001) The effect of drought treatment on the four experimental plant communities (FG1, FG2, FG3, FG123) in terms of (a) floral species richness, and (b) floral abundance. Each box represents the six subplots of that treatment type for each plant community. Treatment refers to Control (C), Roofed control (R) and Drought (D). Levels of significance between a priori contrasts are indicated by symbols ( + p < .06, *p < .05, **p < .01, ***p < .001) moisture content in this experiment (approximately 10% reduction in soil moisture content immediately after drought period) was not great enough to induce nectar volume changes. This may have been partly due to plants accessing moisture through root systems that extended beyond the subplot or because some rain was reaching the subplots as runoff. Additionally, calcareous grasslands are relatively resistant to drought (Grime et al., 2008). Sugar production per flower was lower in our study than in other habitats for L. pratensis and P. vulgaris, the two plant species for which other data are available (Baude et al., 2016),  Contrary to our fourth hypothesis, the differences in functional traits of plant communities did not appear to confer any differential resistance to drought. The community which showed the greatest magnitude of response to drought was FG3, with shallow rooted, fleshy leaved species, which is partly explained by the additional impacts of the experimental rooves on this community. The magnitude of reduction between the other plant communities was similar, suggesting that none of the selected functional traits were able to provide greater resistance to drought. As calcareous grassland plant communities are relatively resistant to drought (Grime et al., 2008), the main strategy across plant communities may be to conserve root biomass and leaf biomass at the expense of reproductive structures.
Contrary to our fifth hypothesis, the functionally diverse plant community (FG123) did not exhibit greater productivity or resistance to drought in terms of floral resources. Greater overall functional diversity of a particular set of traits is expected to result in greater resistance of related functions (Isbell et al., 2015;Oliver, Isaac et al., 2015), for example due to niche complementarity between species (Gross et al., 2007;Gubsch et al., 2011), which may reduce the competition for limited resources. In this case, greater diversity of root traits was expected to result in greater overall availability of water to communities, resulting in more water resources available for floral displays. Our results suggest that the impacts of drought were too ubiquitous across component plant communities for there to have been a benefit in the functionally diverse community. However, functional diversity may still provide longer term benefits that were not measured in this study, for example in terms of recovery.
Given these findings, we can infer multiple impacts on both pollinators and pollination. Firstly, impacts at the flower scale are likely to affect pollinator foraging behaviour due to changes in the reliability of nectar reward. Secondly and most importantly, changes in the overall availability of floral resources, which affects the amount of food that is available to pollinators, will certainly have consequences at the population level (Baude et al., 2016;Carvell et al., 2006Carvell et al., , 2017Roulston & Goodell, 2011). However, determining population level effects on mobile species such as pollinators is difficult. To do so, it would be necessary to simulate drought across the foraging range of pollinator species, and to follow this through multiple years.
In reality, this is only plausible by using real drought events (Oliver, Marshall et al., 2015;Thomson, 2016), or through modelling of population dynamics in response to landscape scale alteration in resources (Becher et al., 2014;Horn, Becher, Kennedy, Osborne, & Grimm, 2015).
Other studies have assessed the impacts of drought on floral resources using laboratory experiments (Villarreal & Freeman, 1990) and field experiments in arid or semiarid regions (Al-Ghzawi et al., 2009;Takkis et al., 2015). The advantages of our study are that it involves intact experimental plant communities based in situ in a temperate region. However, this does come with disadvantages, for example difficulties in differentiating between the effects of drought and other effects of the experimental roof. An additional limitation of our experiment is the use of only a single level of drought. Future research would benefit from assessing multiple levels of water availability, which would elucidate possible critical limits for different plant species and plant communities, and would help to isolate the effects of water availability from other roof effects.
To our knowledge, this is one of the first studies to assess the impacts of drought on floral resources for pollinators at the community level, particularly in a temperate region where the risk of drought is projected to increase under climate change. Drought was found to affect the availability of floral resources in calcareous grassland, despite this being a relatively drought-resistant habitat (Grime et al., 2008). Importantly, effects were consistently observed across a range of plant species and across a range of plant communities with different functional trait compositions. Given that species-rich calcareous grasslands are an important refuge habitat for pollinator species in the United Kingdom (Baude et al., 2016), which provide ecosystem services in nearby farmland (Woodcock et al., 2013), this result suggests that they may support lower pollinator populations in the future under current climate change scenarios.