Predators in Higher Trophic Levels Affect Selection on Floral Traits by Altering Plant-Pollinator Interactions by Amanda Decker Benoit A Thesis Presented to The University of Guelph In partial fulfillment of requirements for the degree of Master of Science in Integrative Biology Guelph, Ontario, Canada © Amanda Decker Benoit, August 2016 PREDATORS IN HIGHER TROPHIC LEVELS AFFECT SELECTION ON FLORAL TRAITS BY ALTERING PLANT-POLLINATOR INTERACTIONS ABSTRACT Amanda Decker Benoit Advisor: University of Guelph, 2016 Professor Christina M. Caruso Interactions between plants and pollinators generate selection on floral traits. These interactions, and the resulting selection, may be affected by predators in higher trophic levels that consume and alter the behavior of pollinators. The effect of predators on selection on floral traits should depend on the predator’s hunting-mode. I examined the effects of active-pursuit predators (dragonflies) and sit-and-wait predators (ambush bugs) on selection on floral traits of the bumblebee-pollinated wildflower Lobelia siphilitica. Contrary to my predictions, I found that while selection did differ between dragonfly treatments, it was not caused by decreased plantpollinator interaction strength. Consistent with my predictions, I found that ambush bugs prefer plants with larger daily displays, and significantly decrease the strength of selection on daily display size. My results suggest that predators in higher trophic levels may be an underappreciated cause of selection on plant traits. iii ACKNOWLEDGEMENTS I would like to sincerely thank all who inspired, guided, and encouraged me throughout the process of completing this thesis. I especially appreciate the thoughtful mentorship of my advisor, Dr. Christina Caruso. Her expertise, passion, and practicality made this work possible. I am also indebted to my committee members, Dr. Shannon McCauley and Dr. Brian Husband. I would like to thank them for their time, enthusiasm, and helpful suggestions. This work would not have been possible without the countless hours of field and laboratory work performed by Katie Brown, our work study students: Emily Williams, Aaron Hudson, Ann Lee, Cassia Dal Bello, Heather Van Den Diepstraten, and Lauren Paulson, and our wonderful volunteers, Stephanie Otto and Tracy Yuen. I also greatly appreciate the hard work and dedication of the Phytotron staff, as well as the staff and researchers at Koffler Scientific Reserve. Especially, Stephan Schneider for being a phenomenal manager. I would also like to thank Dave Punzalan for providing information about ambush bug identification. Last, but certainly not least, I would like to thank members of the Caruso-Maherali lab, and Daniel Siksay, for thought provoking discussion, and helpful feedback. iv TABLE OF CONTENTS Abstract……………………………………………………………………………………………ii Acknowledgements……………………………………………………………………………....iii Table of Contents…………………………………………………………………………………iv List of Figures………………………………………………………………………………….v-vi List of Tables………………………………………………………………………………..vii-viii Introduction …………………………………………………………………………………….1-6 Methods………………………………………………………………………………………..6-19 Study Species…………………………………………………………………………....6-7 Germination and Growth………………………………………………………………..7-8 Experimental Design……………………………………………..…………..…..……8-11 Trait Data Collection…………………………………………………………………11-13 Fitness Data Collection……………………………………………………………….13-14 Statistical Analysis…………………………………………………………………...14-19 Results………………………………………………………………………………………..19-22 Discussion…………………………………………………………………………………….22-27 Literature Cited……………………………………………………………………………….28-34 v LIST OF FIGURES Figure 1. Map of field site locations at the Koffler Scientific Reserve. Field sites 1 and 2 are less than 100m from ponds. Fields sites 3 and 4 are greater than 500m from ponds. Field sites 1 and 3 were found to have high dragonfly abundance, and field sites 2 and 4 were found to have low dragonfly abundance…………………………………………….35 Figure 2. Floral traits of Lobelia siphilitica: a) corolla length was measured from the base of the bracts to the corolla tube opening, b) width of the corolla tube opening was measured from the bottom of the corolla tube to the top at the point where the corolla tube opening begins, c) banner petal size was measured as the total area of the top and bottom petals outlined in white, d) banner petal shape was measured as the ratio of length to width of the bottom three banner petals, which serve as a landing platform, e) the nectar guide stripes were measured as the white stripes (arrow 1) and the purple background (arrow 2), nectar guide contrast was measured by subtracting the modal gray value of purple background from the modal gray value of the white stripes, and f) the area of the corolla underside was measured as the area outlined in white, nectar guide size was measured by dividing the area of the underside of the corolla tube (f) by the area of the nectar guide stripes (e. arrow 1)……………………………………………………………………...36 vi Figure 3. Relationship between daily display size and fruits per plant for Lobelia siphilitica in low- and high-dragonfly abundance treatments. Fruit per plant is a female fitness measure, and was relativized for each treatment by dividing each plant’s fitness by the mean plant fitness within the treatment. Daily display size was standardized by subtracting the treatment mean daily display from each plant’s daily display size, and then dividing by the treatment standard deviation, resulting in a mean of zero and a variance of one…….37 Figure 4. Relationship between daily display size, and fruits per plant (a, b) and seeds per plant (c, d) for Lobelia siphilitica in ambush bug absent and ambush bug present treatments. Fruit per plant, and seeds per plant are female fitness measures. They were relativized for each treatment by dividing each plant’s fitness by the mean plant fitness within the treatment. Daily display size was standardized by subtracting the treatment mean daily display from each plant’s daily display size, and then dividing by the treatment standard deviation, resulting in a mean of zero and a variance of one.……………..38 vii LIST OF TABLES Table 1. Split-plot analysis of variance model of the effect of dragonfly treatment (low abundance vs. high abundance), pollination treatment (hand vs. open), and the interaction between the dragonfly and pollination treatments on three female fitness components of Lobelia siphilitica plants………………………...………………………………………………39 Table 2. Mean values for three female fitness components ± 1 SE (N) for hand- and open-pollinated Lobelia siphilitica plants.………………….………………40 Table 3. Mean values for three female fitness components ± 1 SE (N) for open-pollinated Lobelia siphilitica plants in low- and high-dragonfly abundance treatments.41 Table 4. Linear selection gradients ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants in the low- and high-dragonfly abundance treatments. F compares the strength of selection between the low- and high-dragonfly abundance treatments. …………………………………………………………………..42 Table 5. Linear selection differentials ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants in the low- and high-dragonfly abundance treatments. F compares the strength of selection between the low- and high-dragonfly abundance treatments..................................................................................…43 viii Table 6. Linear selection gradients ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of hand- and open-pollinated Lobelia siphilitica plants. F compares the strength of selection between the hand- and open-pollination treatments.………………………………………….……..44 Table 7. Linear selection differentials ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of hand- and openpollinated Lobelia siphilitica plants. F compares the strength of selection between the hand- and open-pollination treatments..……………………….45 Table 8. Mean trait values ± 1 SE (N) for open-pollinated Lobelia siphilitica plants with ambush bugs present and absent. t compares mean trait values between plants with ambush bugs present and absent..……………………………………..46 Table 9. Linear selection gradients ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants with ambush bugs present and absent. F compares the strength of selection between plants with ambush bugs present and absent..………….47 Table 10. Linear selection differentials ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants with ambush bugs present and absent. F compares the strength of selection between plants with ambush bugs present and absent.48 INTRODUCTION: Mutualists and antagonists that interact with plants can generate selection on plant traits (Benkman 2013; Vanhoenacker et al. 2013). Examples include pollinators selecting on floral traits such as spur length (Johnson and Steiner 1997) and herbivores selecting on plant defense traits such as secondary metabolites (Coley and Barone 1996). However, plants and their mutualists and antagonists are embedded in larger food-webs, and therefore, predators in higher trophic levels may affect the interactions between plants and their mutualists and antagonists – and the resulting selection. Predators in higher trophic levels can alter plant-animal interactions by consuming, and altering the behavior of plant mutualists and antagonists. Predators of plant mutualists and antagonists have been shown to alter plant fitness, plant productivity, and plant community composition (Hairston et al. 1960; Paine 1980; Pace et al. 1999; Preisser et al. 2005; Knight et al. 2006; Romero et al. 2011). Less is known about how predators of plant mutualists and antagonists affect selection on plant traits. But logic suggests, if predators in higher trophic levels alter plant-animal interactions, and plant-animal interactions generate selection, then predators in higher trophic levels should affect the strength of selection on plant traits (Estes 2013). The effects of predators in higher trophic levels on mutualist- and antagonist-mediated selection on plant traits is important to consider for two reasons. First, if predators indirectly affect the strength of selection, then fluctuation in predator abundance may explain observed variation in the strength of mutualist- and antagonist-mediated selection on plant traits (Thompson 2005; Siepielski et al. 2009, 2013; Bartkowska and Johnston 2015; but see Morrissey and Hadfield 2012). Second, understanding the evolutionary consequences of topdown effects of predators on plants may allow us to better predict the consequences of trophic- 1 downgrading – the expatriation and extinction of top predators (Estes 2011). By understanding how predators in higher trophic levels affect plant ecology and evolution, we can make betterinformed conservation and management decisions for areas affected by trophic-downgrading. My goal is to determine whether predators of pollinators alter the strength of pollinatormediated selection on plant traits. Predators, however, are not all equal; they can have different hunting-modes that range from sit-and-wait to active-pursuit hunting. Therefore, I do not expect all pollinator predators to affect the strength of selection on plant traits in the same way. Predators with different hunting modes should affect selection via different mechanisms. Some pollinator predators, such as dragonflies (Odonata e.g. Aeshnidae: Anax junius), birds (Aves e.g. Hirundinidae: Hirundo pyrrhonota), and bee-wolves (Sphecidae e.g. Philanthus bicinctus), have an active-pursuit hunting mode (Dukas 2005; Meehan et al. 2005; Knight et al. 2005). Active-pursuit predators should increase the strength of pollinator-mediated selection by weakening interactions between plants and pollinators. Active-pursuit predators are highly mobile, conspicuous, and have high prey capture rates (Pianka 1966; Schoener 1971; Huey and Pianka 1981). Because of their mobility and high prey capture rates, these predators should decrease the abundance of pollinators through consumptive effects (Hairston et al. 1960). Furthermore, because these predators are conspicuous, they are expected to alter pollinator behavior (Schmitz et al. 1997; Preisser et al. 2005; Romero et al. 2011). For example, in response to predator cues, prey (i.e. pollinators) may forage less (spending more time hidden and inactive), leave the patch in search of a new safer patch, or forage on different resources (Schmitz et al. 1997; Preisser et al. 2005; Romero et al. 2011; Belgrad and Griffen 2016). As a result, active-pursuit predators are expected to decrease the number of foraging pollinators in a patch (Lima 1998; Guariento et al. 2014; Keiser et al. 2015). Therefore, active-pursuit predators 2 should decrease plant-pollinator interaction strength (Preisser et al. 2005; Knight et al. 2006; Trussell et al. 2006; Romero et al. 2011). Interaction strength can have multiple definitions (Menge et al. 2004; Wootton and Emmerson 2005), but here, I use the same definition as Benkman (2013): interaction strength is the proportion of a population that is either directly benefited or harmed by an interaction. This definition takes into account both the individual per capital effect of the interaction as well as population size (Menge et al. 2004; Benkman 2013). Weaker plant-pollinator interactions, i.e. when a smaller proportion of the plant population benefits from interactions with pollinators, should decrease plant fitness. As mean plant fitness decreases, and the variance in fitness either stays constant or increases, the opportunity for selection will increase (Benkman 2013). Opportunity for selection is the maximum strength of selection possible in a population (Crow 1958; Arnold and Wade 1984; Benkman 2013). The opportunity for selection increases as mean fitness decreases because the opportunity for selection is the variance in fitness over mean absolute fitness squared (Benkman 2013). Therefore, by weakening plant-pollinator interactions, active-pursuit predators should increase the opportunity for strong pollinator-mediated selection. In contrast, other predators, such as crab spiders (Thomisidae) and ambush bugs (Phymatinae), have a sit-and-wait hunting mode (Balduf 1939; Morse 1979). Sit-and-wait predators differ from active-pursuit predators in that they are primarily stationary, cryptic, and have low prey capture rates (Pianka 1966; Schoener 1971; Huey and Pianka 1981). Another key difference is that sit-and-wait predators hunt at the plant level, while active-pursuit predators hunt at the patch level. This means that while active-pursuit predators should be randomly distributed in relation to plant phenotypes, sit-and-wait predators are expected to be more common on plants with phenotypes that increase prey (pollinator) attraction (Greco and Kevan 3 1994, 1995; Morse 1999; Bhaskara et al. 2009; Hanna and Eason 2012; Defrize et al. 2014). In contrast to active-pursuit predators, it is unlikely that behavioral changes caused by sit-and-wait predators will decrease the abundance of pollinators in a patch. This is because pollinators can use cues to detect and avoid specific flowers or plants that harbor cryptic sit-and-wait predators (Abbott 2006; Dawson and Chittka 2012; Dawson and Chittka 2014), and pollinators can avoid these predators by moving to a neighboring plant (sit-and-wait predators do not pursue their prey). Instead of decreasing pollinator abundance, it is more likely that behavioral changes caused by sit-and-wait predators will cause decreased pollinator visitation to some plants and increased visitation to neighboring plants, such that overall visitation at the patch level remains unchanged. Again in contrast to active-pursuit predators, sit-and-wait predators are unlikely to greatly reduce the abundance of pollinators through consumptive effects, because sit-and-wait predators have low prey capture rates (Morse 1979; Mason 1986; Elliott and Elliott 1991, 1994). Given that sit-and-wait predators are not expected to decrease pollinator abundance through either behavioral or consumptive effects, it is unlikely that sit-and-wait predators will decrease the strength of plant-pollinator interactions. In contrast, sit-and-wait predators are expected to affect the strength of pollinator-mediated selection by increasing or decreasing the frequency of pollinator visits to certain plant phenotypes. Sit-and-wait predators should be more common on plants with traits that attract prey (Greco and Kevan 1994, 1995; Morse 1999; Bhaskara et al. 2009; Hanna and Eason 2012; Defrize et al. 2014) causing predation risk to be unevenly distributed across plant phenotypes. In response, pollinators should visit safer plants more frequently, and riskier plant less frequently. Plants with traits preferred by sit-and-wait predators should receive fewer pollinator visits and should decrease in fitness relative to plants with traits not preferred by predators. This should weaken the strength of selection on traits preferred by sit- 4 and-wait predators. For example, when predators preferentially hunt on plants with large nectar rewards, pollinators shift to foraging on less rewarding flowers (Jones 2010; Jones and Dornhaus 2011). Such a shift should increase the fitness of the less rewarding flowers relative to more rewarding flowers, causing weaker selection on rewarding traits. To determine if and by which mechanism predators affect pollinator-mediated selection on plant traits, I examined the effects of active-pursuit predators and sit-and-wait predators on selection on inflorescence and floral traits of the bumblebee-pollinated wildflower Lobelia siphilitica. The active-pursuit predators I studied were a community of dragonflies (Odonata), including several species of darners (Aeshnidae), and other large-bodied dragonfly species found in southern Ontario. Because not all species of dragonflies are activity pursuit predators, this study focused only on families that are (Aeshnidae and Corduliidae), excluding families which have a perch and pursue hunting mode (Libellulidae). Active-pursuit dragonflies fly continuously, circling above fields, often in groups, intercepting and consuming other flying insects (Olberg et al. 2005; Jones et al. 2008; personal observation). They are fast, agile fliers and have high prey-capture success rates (Olberg 2000; Combes et al. 2013). I predict that dragonflies will decrease the number of foraging pollinators in a patch. This will weaken plantpollinator interactions, causing plant reproduction to become pollen limited (Ashman et al. 2004), resulting in decreased plant fitness. Decreased plant fitness will cause the strength of pollinator-mediated selection to increase (Benkman 2013). The sit-and-wait predators I studied were ambush bugs (Phymatinae: Phymata pennsylvanica americana) (Balduf 1939, 1941). Ambush bugs perch on flowers or leaves, and use their raptorial claws to capture and hold insects that come within reach (Balduf 1939; Mason 1986). They select a plant to hunt on, and if prey visitation is low they fly to a different plant. After settling on a plant, ambush bugs can 5 remain on the plant as long as prey availability is high (Balduf 1939, 1941; Greco and Kevan 1994, 1995). Because ambush bugs choose hunting sites based on prey availability, and pollinators are their main prey (Balduf 1939), they should be more common on plants with traits that attract pollinators. By increasing predation risk on plants with preferred traits, ambush bugs should decrease pollinator visitation to those plants, and increase pollinator visitation to plants with non-preferred traits. Thereby, ambush bugs should weaken selection on their preferred plant traits. I measured phenotypic selection via female fitness on ten traits of L. siphilitica that were (1) open-pollinated with low and high dragonfly abundance, (2) open-pollinated with ambush bugs absent and present, and (3) hand-pollinated with supplemental pollen. I used these data to answer the following questions: (1) Do dragonflies increase the strength of selection on floral and inflorescence traits by decreasing the strength of plant-pollinator interactions?; and (2) Do ambush bugs affect the strength of selection on floral traits by changing predation risk across plant phenotypes? METHODS: Study species Lobelia siphilitica (Campanulaceae) is a short-lived, perennial wildflower native to wetlands and woodlands in eastern North America (Johnston 1991a and references therein). It produces three-centimeter-long, tubular flowers, ranging from blue to purple, on a single racemose inflorescence, and occasionally on axillary branches (Parachnowitsch and Caruso 2008; Caruso et al. 2010). In Ontario, L. siphilitica flowers from August to early October, and 6 fruits dehisce from mid-September to November (Johnston 1991a). Lobelia siphilitica is gynodioceous; plants can be female or hermaphroditic, but in northern populations females are rare (<10%) (Caruso and Case 2007). Hermaphroditic L. siphilitica plants are self-compatible; however, they are protandrous with non-overlapping male and female phases. Therefore, selfing in this species must be pollinator mediated (Johnston 1991b). Lobelia siphilitica is primarily pollinated by two species of bumblebee, Bombus vagans and B. impatiens (Flanagan et al. 2011). Germination and Growth The L. siphilitica plants used in this experiment were germinated from seeds collected in 2012 from 49 open-pollinated plants at the T.J. Dolan Natural Area, Stratford, Ontario, Canada (43.370N, 80.997W). Two methods were used to break seed dormancy. In the first method, seeds were placed on moist filter paper in a Petri dish, wrapped in Parafilm, and stratified at 4˚C for eight weeks (Johnston 1992). In the second method, seeds were treated with a 16:1:1 solution of distilled water, 70% ethanol, and bleach, with four drops of Triton X-100 detergent added per two liters of solution (Dudle et al. 2001). The seeds were then rinsed with distilled water and stored at 4˚C overnight. Cold stratification began on January 27, 2015 and the bleach treatment was performed on February 18, 2015. After breaking dormancy, approximately 90 seeds per family were planted into plug trays filled with Sunshine Mix (Sun Gro Horticulture Canada Ltd., Vancouver, British Columbia, Canada). The trays were placed in standing water on benches at the University of Guelph Phytotron, Guelph, Ontario, Canada. Five to seven weeks after sowing, 1-50 seedlings from 37 families with adequate germination were transplanted (N = 1397 plants) from plug trays into 635 7 mL3 Deepots (Stuewe and Sons Inc., Tangent, Oregon, USA) filled with Sunshine mix (Sun Gro Horticulture Canada Ltd., Vancouver, British Columbia, Canada). Each plant was randomly assigned a position on the greenhouse bench. Plants were watered as needed, exposed to supplementary light for 16 hours per day, and fertilized with eight prills of Nutricote total 13-1313 with micronutrients (Plant Products, Brampton, Ontario, Canada). Experimental design The week of June 8, 2015, plants which had bud primordia, or were ≥ 8 cm tall were assigned to be experimental plants (N= 300), pollen donors (N=80), or extras (N=118). Plants which were already flowering, or that had multiple racemes, were assigned to be pollen donors. All other plants were randomly assigned to be experimental plants, pollen donors, or extras. The experimental plants, pollen donors, and extras were then transported to Koffler Scientific Reserve (KSR), King City, Ontario (44.03N, 79.52W) where the field experiment was conducted. The experimental plants were transplanted into 8.7 liter pots filled with Sunshine Mix, and remained potted for the duration of the experiment. Pollen donors and extras remained in Deepots and were housed in the field station’s greenhouse. From the old fields of the reserve, I selected four fields similar in size, exposure (full sun), and plant community composition (typical old field communities including Solidago spp., Daucus carota, Asclepias syriaca, and Vicia cracca). Within each selected field, I constructed a 10m × 10m fence that was 2.4m tall. Within each fenced site, I placed 75 randomly assigned, potted L. siphilitica plants. They were arranged in arrays, with half a meter between each plant and its nearest neighbor. The plants were placed into the field sites on June 18 and 19, 2015, and 8 remained in the field for eight weeks. For the duration of the experiment, vegetation within the fence was trimmed to below the rims of the pots, and flowers from plants other than L. siphilitica were removed. All plants were sexed upon flowering. Female plants (N = 6) were replaced with randomly selected hermaphrodites from the extras group for the first three weeks of the experiment. Plants discovered to be female after three weeks were excluded from analysis (N = 4). Three additional plants were replaced with extras because they were found to be Lobelia cardinalis × siphilitica hybrids. Dragonfly abundance was not manipulated; instead, the locations of the field sites utilized natural variation of dragonfly abundance. I selected two sites adjacent to ponds (<100m), where I predicted dragonfly abundance would be high, and two fields greater than 500m from ponds, where I predicted dragonfly abundance would be low. Dragonfly abundance was measured on 31 days over the course of the eight-week experiment, approximately every second day as weather allowed. Dragonfly abundance was measured by conducting point counts, as done in other studies (e.g. Knight et al. 2005a). During these point counts, an observer stood at each corner of each field site for five minutes looking continuously in one direction and counting the number of dragonflies that entered their field of view. Each dragonfly was classified as small, medium, or large. Because small and medium dragonflies were rarely observed, and are unlikely to prey upon bumblebees, my analysis is restricted to the abundance of large dragonflies. However, the inclusion of small or medium dragonflies does not change the classification (low vs. high dragonfly abundance) of the field sites (data not shown). Contrary to my expectations, dragonfly abundance was not related to pond proximity. Of the two field sites adjacent to ponds (Fig. 1), site one had high dragonfly abundance (mean ( 1 SE) dragonflies/20 minutes = 23.8 (8.53); Range = 0-214), and site two had low dragonfly 9 abundance (mean (1 SE) = 2.5 (0.51); Range = 0-11). Of the two field sites greater than 100m from ponds (Fig. 1), site three had high dragonfly abundance (mean (1 SE) = 10.6 (3.04); Range = 0-68), and site four had low dragonfly abundance (mean (1 SE) = 2.6 (0.60); Range = 0-11). Consequently, site two (<100m from a pond) was reclassified as a low-dragonfly site, and site three (>500m from a pond) was reclassified as a high-dragonfly site (Fig. 1). In all four field sites, I allowed naturally occurring ambush bugs from the surrounding fields to colonize the potted L. siphilitica plants. The first adult ambush bug was observed on July 21. After July 21 the plants were inspected every second day for ambush bugs. From July 21 to 28 plants were classified as having ambush bugs either present to absent. From July 28 to August 12 the number of ambush bugs on each plant was counted. If ambush bugs were found on a plant for one or more days, then for analysis the plant was placed in the ambush bug present category; if ambush bugs were never observed on a plant, then the plant was placed in the ambush bug absent category. Of the 195 open pollinated plants, 127 were in the ambush bugs present category and 68 were in the ambush bugs absent category. During the experiment, ambush bugs were observed both attempting to prey upon and successfully consuming bumblebees and other floral visitors, including butterflies and syphid flies (personal observation). The majority of plants in the ambush bug present category hosted ambush bugs on two or more of the days that plants were examined. Of the plants in the ambush bug present category, 48 plants had ambush bugs present on only one of the nine days they were examined, 33 plants had ambush bugs present on two of the nine examination days, 17 plants had ambush bugs on 3 of the 9 days, 15 plants had ambush bugs present on 4 of the 9 days, and 15 plants had ambush bugs present on 5 or more of the examination days. The number of ambush bugs on a plant at a 10 given time ranged between zero and nine. The mean number of ambush bugs observed on each plant in the ambush bug present category was 1.6. Other studies have demonstrated that the mean residence time for ambush bugs on a plant is 5.2 days with a range of 1-14 days (Yong 2005). The ratio of plants with and without ambush bugs was similar for field sites two, three, and four. However, field site one had fewer ambush bugs than the others. In field site one, ambush bugs were present on 13 plants and absent on 37 plants. In field site two, ambush bugs were present on 40 plants and absent on 9 plants. In field site three, ambush bugs were present on 41 plants and absent on 8 plants. In field site four, ambush bugs were present on 35 plants and absent on 14 plants. To determine if selection on floral traits of L. siphilitica was pollinator mediated, 25 plants at each field site were randomly assigned to be hand-pollinated. To hand pollinate plants, pollen was collected from at least five haphazardly-chosen pollen donors, and applied with a paintbrush to the stigmas of plants in the hand-pollination treatment (Caruso et al. 2010; Parachnowitsch and Caruso 2008). Plants in this treatment received supplemental pollen every second day (as in Parachnowitsch and Caruso 2008; Rivkin et al. 2015). Comparing selection between open- and hand-pollinated plants determines whether selection is the result of interactions with pollinators (Ashman et al. 2004). Trait data collection To measure floral morphology, I scanned one of the first five flowers from each plant. Using one of the first five flowers controls for within plant variation in floral traits along the raceme (Diggle 1995). Each flower was open for a minimum of two days, to ensure that it was 11 fully expanded, before collection for scanning. Collected flowers were stored on ice and transported to the University of Guelph, where they were scanned using WinRhizo© (Regent Instruments Inc., QC, Canada). Using these scanned images and ImageJ (Abramoff et al. 2004), I measured four floral morphology traits. Corolla length was measured as the distance from the tube opening to where the bracts attached (Fig. 2a). Width of the tube opening was measured as the length from the base of the tube to the top of the tube at the tube opening (Fig. 2b). Banner petal size was measured as the total area of the top and bottom banner petals i.e. the petals visible to an approaching pollinator (Fig. 2c). Banner petal shape was measured as the ratio of length to width of the three lower banner petals, which bumblebees use as a landing platform (Fig. 2d). I measured four floral color traits. The first two traits, nectar guide contrast and size, were measured by analyzing scanned images in ImageJ. These measurements quantify variation in the white and purple striping on the underside of the corolla tube. Nectar guide contrast was calculated by subtracting the modal gray value of the purple background from the modal gray value of the white stripes (Fig. 2e). Nectar guide size was determined by dividing the area of the underside of the corolla tube (Fig. 2f) by the area of the nectar guide stripes (Fig 2e). The second two traits, chroma and brightness, were measured using an Ocean Optics USB2000 spectrophotometer (Dunedin, FL) with a deuterium-tungsten halogen light source (Caruso et al. 2010). I measured the proportion of light reflected from L. siphilitica petals at wavelengths between 400 and 800 nm at a 0.3-nm interval, standardized against black and white reference standards (Caruso et al. 2010). Principal component analysis was used to reduce the spectra into two of the three fundamental components of color: brightness and chroma, the first and third principal components respectively (Grill and Rush 2000). Brightness is the amount of light 12 reflected: bright flowers are light purple, while less bright flowers are dark purple. Chroma is the amount of gray mixed with hue: flowers with high chroma are vibrant purple, while flowers with low chroma are faded purple. Lastly, I measured two inflorescence traits: inflorescence height and daily display. Inflorescence height was measured after fruits matured as the distance from the soil to the top of the raceme. Daily display size was measured as the mean number of flowers open on a plant each day. Flowers were counted daily at the start of the experiment, and every fourth day after July 13, 2015. Fitness data collection On August 17 and 18, 2015, the plants were transported to the University of Guelph Bovey Greenhouse, where fruits were allowed to mature. Before dehiscence, between 6 and 29 fruit per plant were collected (mean (1 SE) = 15.4 (0.21); N = 257 plants and 3965 fruit). To ensure that I collected fruits from along the length of the raceme, that were representative of the pollination environment for the full timeframe of the experiment, I marked the raceme at the topand bottom-most open flower approximately every 10th day for the duration of the experiment. From each marked section of the raceme, I collected three randomly selected fruit. Three additional fruit were collected from fruiting axillary branches, when present. I measured three female fitness components: fruit per plant, seeds per fruit, and seeds per plant. To measure fruit per plant, I counted the total number of fruit produced by each plant during the experiment (fruit produced after the end of the experiment were not counted). To measure seeds per fruit, I separated the seeds from the fruit capsule, and weighed them using a 13 microbalance (Sartorius LA120S, Sartorius Corporation, Bohemia, NY). I then found the average mass of 50 seeds from a random sample of 62 fruit. To estimate the total number of seeds per fruit, I divided the mass of all the seeds in the fruit by the mean mass of 50 seeds, and multiplied by 50 (Johnston 1991b). To measure seeds per plant, I multiplied the mean number of seeds per fruit by the number of fruits per plant. Because, cleaning and weighing seeds was time consuming, seeds per fruit and seeds per plant data were measured for a randomly selected subsample of 254 plants. Statistical analysis Q1: Do dragonflies increase the strength of selection on floral and inflorescence traits by decreasing the strength of plant-pollinator interactions? I predicted that dragonflies would decrease interactions between plants and pollinators, causing plant reproduction to become pollen limited. To determine if the magnitude of pollen limitation was greater in high dragonfly abundance sites than in low dragonfly abundance sites, I used a split-plot analysis of variance (ANOVA) model. Dragonfly treatment (low vs. high) was included as the between-plot fixed factor. Pollination treatment (hand vs. open) was included as the within-plot fixed factor. The model also included dragonfly × pollination treatment as a fixed factor, and field site nested within dragonfly treatment as a random factor. Separate models were run for each of three fitness measures: fruit per plant, seeds per fruit, and seeds per plant. If the pollination term is significant, and plants in the hand-pollinated treatment have higher fitness than plants in the open-pollinated treatment, I can conclude that plant reproduction is pollen limited. If the dragonfly term is significant, and plant fitness is lower in the high-dragonfly 14 abundance treatment, I can conclude that dragonflies decrease plant fitness. If the dragonfly × pollination term is significant, and plant fitness is lowest in the high-dragonfly abundance/openpollination treatment combination, I can conclude that dragonflies increase the magnitude of pollen limitation by weakening plant-pollinator interactions. To test whether dragonflies strengthened selection on floral and inflorescence traits of L. siphilitica, I calculated directional selection gradients and differentials for open-pollinated plants in the low- and high-dragonfly abundance treatments. Selection differentials estimate both direct and indirect selection on a trait, whereas selection gradients estimate only direct selection (Lande and Arnold 1983). To calculate directional selection gradients, I used multiple regression. The model included relativized fitness as the dependent variable. Fitness was relativized for each treatment by dividing the individual fitness value by the treatment mean fitness value (Lande and Arnold 1983). The ten standardized traits were included in the model as continuous independent variables. Traits were standardized for each treatment by subtracting the treatment mean from the individual value, and then dividing by the treatment standard deviation, resulting in a mean of zero and a variance of one (Sokal and Rohlf 1995). Data from the same treatment but different field sites was pooled. Field site was included in the model as a fixed factor to control for differences in fitness or traits caused by field site. Ambush bug presence/absence was also included as a fixed factor to control for any effect of ambush bugs. The same model was used to calculate directional selection differentials, but only one trait was included per model. Separate models were run for three fitness measures: fruit per plant, seeds per fruit, and seeds per plant. After I had estimated selection on floral and inflorescence traits of plants in the low- and high-dragonfly abundance treatments, I tested whether selection gradients and differentials for open-pollinated plants differed between dragonfly treatments using a split-plot analysis of 15 covariance (ANCOVA). Dragonfly treatment (low vs. high) was included as the between-plot fixed factor. Ambush bug treatment (absent vs. present) was included as the within-plot fixed factor. The 10 standardized traits were included as continuous independent variables. The interactions between dragonfly treatment and each trait were also included in the model. Field site nested within dragonfly treatment was included as a random factor. Relativized fitness was included as the dependent variable. To determine if the strength of linear selection differentials differed between the low- and high-dragonfly abundance treatments, I used the same model, but with only one trait included per model. Separate models were run for each of three fitness measures: fruit per plant, seeds per fruit, and seeds per plant. If the dragonfly × trait term is significant, then I can conclude that selection on that trait differs between the low- and highdragonfly abundance treatments. In comparing selection between low- and high-dragonfly abundance treatments, I assumed that dragonflies were affecting pollinator-mediated selection on floral traits. However, not all selection on floral traits is pollinator-mediated. To test whether selection of floral traits of L. siphilitica was pollinator-mediated, I calculated directional selection gradients and differentials for plants in the hand- and open-pollination treatments. To calculate linear selection gradients, I used multiple regression. The model included relativized fitness as the dependent variable, the ten standardized traits as continuous independent variables, and field site and ambush bug presence/absence as fixed factors. The same model was used to calculate directional selection differentials, but only one trait was included per model. Separate models were run for three fitness measures: fruit per plant, seeds per fruit, and seeds per plant. Fitness was relativized and traits were standardized for each treatment. After I had estimated selection on floral and inflorescence traits of plants in the hand- and 16 open-pollinated treatments, I tested whether selection gradients and differentials differed between pollination treatments using analysis of covariance (ANCOVA). The model included relativized fitness as the dependent variable, the ten standardized traits as continuous independent variables, and pollination treatment (hand vs. open), ambush bug (absent vs. present), and field site as fixed factors. The model included interactions between pollination treatment and each trait. To determine if the strength of linear selection differentials differed between the hand and open-pollination treatments, I used the same model, but with only one trait included per model. If the pollination treatment × trait term is significant, then I can conclude that selection on that trait is pollinator-mediated. Q2: Do ambush bugs affect the strength of selection on floral traits by changing predation risk across plant phenotypes? To determine if ambush bugs have preferences for plant traits, I used two approaches. First, I used an independent samples t-tests to compare trait means between plants with ambush bugs absent and present. The t-tests included ambush bugs (absent vs. present) as the independent variable, and the ten floral and inflorescence traits as the dependent variables. If a ttest is significant, then I can conclude that ambush bugs have a trait preference. Second, I used a generalized linear model with a Poisson distribution and a log link function. The model included the maximum number of ambush bugs found on a plant as the dependent variable, field site as the predictor factor, and the ten traits as continuous independent variables. If a trait term is significant, then I can conclude that ambush bugs have a preference for that trait. Because the results of this model were consistent with the results of the t-tests, only the results of the t-tests are included here. 17 To test whether ambush bug preferences for traits of L. siphilitica affect selection on those traits, I calculated directional selection gradients and differentials for plants with ambush bugs absent and present (Lande and Arnold 1983). To calculate selection gradients, I used multiple regression. The model included relativized fitness as a dependent variable, the ten standardized traits as continuous independent variables, and field site as a fixed factor. Fitness was relativized, and traits were standardized separately for each treatment. The same model was used to calculate selection differentials, but with only one trait included per model. Separate models were run for each of three fitness measures: fruit per plant, seeds per fruit, and seeds per plant. After I had estimated selection on floral and inflorescence traits of plants in ambush bug absent and present treatments, I tested whether selection gradients and differentials differed between the treatments using analysis of covariance (ANCOVA). To test if gradients differed, the ANCOVA model included relativized fitness as the dependent variable, the ten standardized traits as continuous independent variables, and ambush bug (absent vs. present) and field site as fixed factors. The model included interactions between ambush bug treatment and all traits. The same model was used to determine if selection differentials differed, but only one trait was included per model. Separate models were run for each of three measures of fitness: fruit per plant, seeds per fruit, and seeds per plant. If the ambush bug × trait term is significant, I can conclude that ambush bugs alter the strength of selection on that trait. I tested assumptions of normality and homogeneity of variance for all models. To test for normality, I compared unstandardized residuals to a normal distribution using a KolmogorovSmirnov test. A significant test would indicate that the assumption of normality was violated. To test for homogeneity of variance, I estimated the Spearman correlation between the predicted 18 values and the residuals. A significant correlation would indicate that the assumption of homogeneity of variance was violated. All analysis was done with IBM SPSS Statistics for Windows, Version 23.0 (Armonk, NY: IBM Corp.). Twelve plants were excluded from analysis because they died during the experiment, were female, or were missing trait data. RESULTS: Overall, I found significant selection on both inflorescence traits and all eight floral traits of L. siphilitica in at least one treatment. I consistently found significant selection for larger daily display size, taller inflorescences, wider corolla tube openings, larger banner petals, longer/narrower banner petal shape, increased nectar guide contrast, and lower chroma (Tables 4, 5, 6, 7, 9, and 10). For three traits, I found that the direction of selection was more variable: I found significant selection for both shorter and longer corolla tubes (Tables 6 and 7), larger and smaller nectar guides (Tables 6 and 9), and brighter and less bright flowers (Tables 4, 6, and 9). Generally, I detected more significant selection via fruit per plant and seeds per plant than by seeds per fruit. Q1: Do dragonflies increase the strength of selection on floral and inflorescence traits by decreasing the strength of plant-pollinator interactions? In the split-plot analysis of variance model, which tested whether the treatments affected plant fitness, no terms were significant (Table 1). The pollination term was not significant for 19 any of the three female fitness components of L. siphilitica (Table 1). The magnitudes of the fitness differences between hand- and open pollinated plants were very small. There were ~0.01% fewer fruits per plant, ~0.08% more seeds per fruit, and ~0.06% more seeds per plant in the hand-pollinated treatment than in open-pollinated treatment (Table 2). The dragonfly term was not significant for any of the three female fitness components of L. siphilitica (Table 1). The magnitudes of fitness differences between the low- and high-dragonfly abundance treatments were small. There were ~0.09% more fruit per plant, ~0.08% fewer seeds per fruit, and ~0.03% more seeds per plant in the high-dragonfly abundance treatment than in the low-dragonfly abundance treatment (Table 3). Furthermore, the dragonfly × pollination term was not significant for any of the three female fitness components (Table 1). Although dragonflies did not affect mean plant fitness, selection on two floral traits, daily display size and petal brightness, differed between the low- and high-dragonfly abundance treatments. Direct selection via fruit per plant on daily display size was ~53% stronger in the high-dragonfly abundance treatment than in the low-dragonfly treatment (Table 4). The selection differential estimated via fruit per plant for daily display size was ~46% stronger in the highdragonfly abundance treatment than in the low-dragonfly abundance treatment (Fig. 3 and Table 5). Direct selection via fruit per plant was for brighter flowers in the low-dragonfly abundance treatment, but the strength of selection decreased by ~332% resulting in selection for less bright flowers in the high-dragonfly abundance treatment (Table 4). For all other traits, there was no significant difference in the strength or direction of selection between the low- and highdragonfly abundance treatments (Tables 4 and 5). The strength of selection on inflorescence and floral traits of L. siphilitica in the handand open-pollinated treatments differed for one trait. When selection was calculated using seeds 20 per plant, hand-pollinated plants experienced direct selection for larger nectar guides, while open-pollinated plants experienced direct selection for smaller nectar guides that was ~15% weaker (Table 6). For all other traits, selection was consistent between the hand- and openpollination treatments (Tables 6 and 7). Q2: Do ambush bugs affect the strength of selection on floral traits by changing predation risk across plant phenotypes? One of ten trait means differed significantly between plants with ambush bugs present and absent. Plants with ambush bugs present had a mean daily display 16% larger than plants without ambush bugs (Table 8). Means of the other nine traits did not differ between plants with and without ambush bugs (Table 8). The strength of selection differed between plants with and without ambush bugs for three traits, daily display size, banner petal shape, and banner petal size. The selection gradient for daily display size estimated via fruit per plant was ~47% weaker when ambush bugs were present than when they were absent, and when estimated via seeds per plant it was ~57% weaker when ambush bugs were present than when they were absent (Table 9). The selection differential for daily display size estimated via fruit per plant was ~41% weaker when ambush bugs were present, compared to when they were absent (Fig. 4 and Table 10). The selection differentials estimated via fruit per plant showed selection for larger banner petals when ambush bugs were absent, but for smaller banner petals when ambush bugs were present (Table 10). The selection differentials estimated via seeds per plant showed selection for longer/narrower banner petals 21 when ambush bugs were absent, but for shorter/wider banner petals when ambush bugs were present (Table 10). DISCUSSION: My results support the hypothesis that ambush bugs decrease the strength of selection on the plant traits they prefer, by increasing predation risk and shifting pollinator visitation away from plants with preferred traits. Ambush bugs were more abundant on plants with larger daily displays than plants with smaller daily displays (Table 8). Furthermore, when ambush bugs were present, selection for larger daily display size was significantly weaker than when ambush bugs were absent (Tables 9 and 10). From these results, I can infer that the bumblebee pollinators of L. siphilitica shifted from visiting plants with large daily displays to visiting plants with smaller daily displays to avoid predation risk (Jones 2010; Jones and Dornhaus 2011; Wang et al. 2013), and that this shift weakened selection on daily display size. These results are consistent with previous studies, which have found that bumblebees can recognize and avoid cryptic sit-and-wait predators (Abbott 2006; Ings and Chittka 2009; Wang et al. 2013) and that the presence of sitand-wait predators on flowers decreases pollinator visitation frequency and duration (Dukas and Morse 2003, 2005; Suttle 2003; Robertson and Maguire 2005; Reader et al. 2006; GoncalvesSouza et al. 2008; Higginson et al. 2010; Romero et al. 2011). In contrast, my hypothesis that dragonflies increase the strength of pollinator-mediated selection by decreasing plant-pollinator interactions (Benkman 2013) was not supported. While I did find stronger selection on daily display size in the high-dragonfly abundance treatment, and a shift from selection for brighter to less bright petals (Table 4 and 5), plants did not have lower 22 mean fitness in the high-dragonfly abundance treatment (Tables 1 and 3). Furthermore there is not consistent evidence that dragonfly abundance affected bumblebee abundance (Appendix A). My results showing stronger selection on daily display size, and a directional shift from selection for brighter to less bright flowers in the high-dragonfly abundance treatment, have two possible explanations. First, it is possible that dragonflies did affect selection, but through a mechanism other than altered interaction strength. For example, it is possible that dragonflies altered bumblebee behavior and preferences without decreasing their abundance. Second, it is possible that a factor other than dragonfly abundance caused the differences in selection between the lowand high-dragonfly abundance treatments. To determine which of these two possibilities cause selection to differ between the dragonfly treatments, a follow-up study could be done which manipulates dragonfly abundance. Then, if selection differs between the dragonfly treatments, I could conclude that dragonflies are affecting selection through a mechanism other than interaction strength. While I found no evidence that active-pursuit predators affect interaction strength or selection via interaction strength, previous studies have found that active-pursuit predators, including dragonflies, decrease mean plant fitness by decreasing interactions between plants and pollinators (Knight et al. 2005a; Dukas 2005; Meehan et al. 2005). Differences in the strength of the top-down effects of active-pursuit predators may be due to differences in the body size of prey (Romero et al. 2011). In a system reliant on small-bodied pollinators, including sweat bees (Agapostemon spp.) and flies (Syrphidae and Bombyliidae), Knight et al. (2005a) found that dragonflies exerted a strong top-down effect, which significantly decreased plant fitness. In contrast, in a system reliant on large-bodied pollinators, bumblebees, I found that dragonflies had a weak top-down effect and did not affect plant fitness. Dragonfly prey-capture success 23 decreases as prey size increases (Combes et al. 2013). Thus, while dragonflies do consume bumblebees (Beatty 1951; Wright 1944; Warren 1915), they may not pose as great a risk to bumblebees as they do to smaller-bodied pollinators (Rodriguez-Girones and Bosch 2012). If the strength of the top-down effects of predators depends on prey body size, then predators should have a stronger top-down effect on plants pollinated by small-bodied pollinators, such as sweat bees, and a weaker top-down effect on plants pollinated by large-bodied pollinators, such as bumblebees. My results suggest that sit-and-wait predators, by altering predation risk across plant phenotypes, have a stronger top-down effect on selection on plant traits than active-pursuit predators. This suggests first, that predators, particularly sit-and-wait predators, have the ability to affect the evolution of plant traits. And second, that interaction strength, contrary to popular belief, may not be the most important mechanism mediating the strength of selection. These findings may be generalizable beyond my study system. Many sit-and-wait predators use direct and indirect cues to choose hunting sites with high prey availability (Morse 1988, 1993, 1999, 2000; Greco and Kevan 1994, 1995; Heiling et al. 2004; Bhaskara et al. 2009; Hanna and Eason 2012; Defrize et al. 2014). Because sit-and-wait predators use cues to select hunting sites, they should not be randomly distributed across plant phenotypes. Rather, sit-and-wait predators should be more common on plants with traits that attract pollinators. Sit-and-wait predators, despite low prey capture rates (Morse 1979), can have a significant effect on pollinator visitation frequency and duration (Dukas 2001; Suttle 2003; Dukas and Morse 2003, 2005; Robertson and Maguire 2005; Abbott 2006; Knight et al. 2006; Llandres et al. 2013; Dawson and Chittka 2014). Therefore, by altering predation risk across plant phenotypes, sit-and-wait predators may be common, yet underappreciated, agents of selection on plant traits. 24 In addition to finding that predators of pollinators can have top-down effects on selection on floral and inflorescence traits, I also found that the majority of selection on floral and inflorescence traits was not pollinator-mediated. By comparing the strength of selection on traits of hand- and open-pollinated plants, I found that while there was significant selection on the majority of traits, selection was pollinator-mediated for only one trait (Tables 6 and 7). These results, along with results from other studies, demonstrate that the absence of pollinator-mediated selection does not imply that there is no selection on floral traits. They also demonstrate the inverse: that finding significant selection on floral and inflorescence traits does not imply that selection is pollinator-mediated (Parachnowitsch and Caruso 2008; Caruso et al. 2010). While ambush bugs preferred plants with larger daily displays, and thereby decreased the strength of selection on daily display size, selection on two other traits, banner petal shape and banner petal size, also differed between plants with ambush bugs present and absent (Tables 9 and 10). However, there is no evidence that ambush bugs had preferences for these traits (Table 8). Given the lack of predator preference, one possible explanation is that banner petal shape and size have a functional role in pollination that is affected by ambush bugs. Banner petal shape measures the ratio of length to width of the bottom three banner petals that serve as a landing platform for visiting bumblebees. This landing platform has a vital functional role; without these three petals, successful bumblebee visitation is reduced from ~90% to ~27% (Owen and Bradshaw 2011). Previous studies have found that pollinators can have strong preferences for, and exert selection on petal shape (Møller 1995; Galen and Cuba 2001; Gomez et al. 2008). If banner petal shape and size affects the efficiency of visiting bumblebees (Owen and Bradshaw 2011) or their risk of being attacked, then selection on banner petal size and shape may have shifted in response to ambush bug presence. Shorter/wider banner petals may allow bumblebees 25 to find and enter the corolla tube opening more quickly. This could decrease the opportunity for a hidden ambush bugs to attack visiting bumblebees, thereby allowing plants with shorter/wider banner petals to receive more successful visits. Similarly, if ambush bugs hide behind banner petals to conceal themselves from approaching pollinators, and smaller banner petals offer less opportunity to for concealment, then pollinators may be safer visiting plants with smaller banner petals. If banner petal shape and size do have functional roles which affect the safety of visiting pollinators, then this could explain why there was selection for shorter/wider and smaller banner petals when ambush bugs were present. One limitation of my study is that I did not manipulate predators. Because of this, I cannot rule out the possibility that other agents of section influenced my results (Wade and Kalisz 1990). I elected to utilize a natural gradient of dragonfly abundance because building dragonfly enclosures or exclosures at such a large scale was infeasible. In contrast to dragonfly abundance, ambush bug presence could have been manipulated, but I elected to allow ambush bugs to choose plants freely in order to determine if they had trait preferences. While this decision limits my ability to make causal claims, if I had manipulated ambush bug presence, I would not have observed their preference for plants with larger daily displays, and I would not have been able to test my hypothesis that sit-and-wait predator preferences affect the strength of selection on plant traits. A follow-up study which manipulates ambush bug presence/absence would be informative. If ambush bugs were randomly assigned to plants, then they should not affect the strength of selection. My results suggest that predators in higher trophic levels may be an underappreciated cause of selection on plant traits. Given that since Hairston, Smith, and Slobodkin published their classic paper (1960), evidence has mounted that predators produce cascading ecological effects, 26 it is not surprising that predators can also have cascading evolutionary effects (Estes 2013). The cascading evolutionary effects of predators in higher trophic levels have received little attention, because most studies of selection have been restricted to one or two trophic levels (e.g. Benkman 2013). However, my results demonstrate the importance of incorporating additional higher trophic levels when studying selection on plant traits. My results also suggest that anthropogenic trophic downgrading (Estes 2011) in addition to damaging ecological consequences, may have unforeseen cascading evolutionary consequences. 27 LITERATURE CITED Abbott, K. R. 2006. Bumblebees avoid flowers containing evidence of past predation events. Canadian Journal of Zoology-Revue Canadienne De Zoologie 84:1240-1247. Abrámoff, M., P. J. Magalhães, and S. J. Ram. 2004. Image processing with ImageJ. Biophotonics International 11:36-42. Arnold, S. J. and M. J. Wade. 1984. On the measurement of natural and sexual selection - theory. Evolution 38:709-719. Ashman, T. L., T. M. Knight, J. A. Steets, P. Amarasekare, M. Burd, D. R. Campbell, M. R. Dudash, M. O. Johnston, S. J. Mazer, R. J. Mitchell, M. T. Morgan, and W. G. Wilson. 2004. Pollen limitation of plant reproduction: ecological and evolutionary causes and consequences. Ecology 85:2408-2421. Ashman, T. L. and M. T. Morgan. 2004. Explaining phenotypic selection on plant attractive characters: male function, gender balance or ecological context? Proceedings of the Royal Society B-Biological Sciences 271:553-559. Balduf, W. V. 1939. Food habits of Phymata Pennsylvanica americana Melin (Hemiptera). Canadian Entomologist 71:66-74. Balduf, W. V. 1941. Life history of Phymata pennsylvanica americana Melin (Phymatidae, Hemiptera). Annals of the Entomological Society of America 34:204-214. Bartkowska, M. P. and M. O. Johnston. 2015. Pollen limitation and its influence on natural selection through seed set. Journal of Evolutionary Biology 28:2097-2105. Beatty, G. H., III. 1951. Odonate bionomics: I. Notes on the food of dragonflies. Odonata vs. ants and bees. Bulletin of the Brooklyn Entomological Society 46:29-38. Belgrad, B. A. and B. D. Griffen. 2016. Predator - prey interactions mediated by prey personality and predator hunting mode. Proceedings of the Royal Society B-Biological Sciences 283. Benkman, C. W. 2013. Biotic interaction strength and the intensity of selection. Ecology Letters 16:1054-1060. Bhaskara, R. M., C. M. Brijesh, S. Ahmed, and R. M. Borges. 2009. Perception of ultraviolet light by crab spiders and its role in selection of hunting sites. Journal of Comparative Physiology a-Neuroethology Sensory Neural and Behavioral Physiology 195:409-417. Caruso, C. M. and A. L. Case. 2007. Sex ratio variation in gynodioecious Lobelia siphilitica: effects of population size and geographic location. Journal of Evolutionary Biology 20:1396-1405. 28 Caruso, C. M., S. L. Scott, J. C. Wray, and C. A. Walsh. 2010. Pollinators, herbivores, and the maintenance of flower color variation: a case study with Lobelia siphilitica. International Journal of Plant Sciences 171:1020-1028. Coley, P. D. and J. A. Barone. 1996. Herbivory and plant defenses in tropical forests. Annual Review of Ecology and Systematics 27:305-335. Combes, S. A., M. K. Salcedo, M. M. Pandit, and J. M. Iwasaki. 2013. Capture success and efficiency of dragonflies pursuing different types of prey. Integrative and Comparative Biology 53:787-798. Crow, J. F. 1958. Some possibilities for measuring selection intensities in man. American Anthropologist 60:1-13. Dawson, E. H. and L. Chittka. 2012. Conspecific and heterospecific information use in bumblebees. Plos One 7:e31444-e31444. Dawson, E. H. and L. Chittka. 2014. Bumblebees (Bombus terrestris) use social information as an indicator of safety in dangerous environments. Proceedings of the Royal Society BBiological Sciences 281:20133174-20133174. Defrize, J., A. L. Llandres, and J. Casas. 2014. Indirect cues in selecting a hunting site in a sitand-wait predator. Physiological Entomology 39:53-59. Diggle, P. K. 1995. Architectural effects and the interpretation of patterns of fruit and seed development. Annual Review of Ecology and Systematics 26:531-552. Dudle, D. A., P. Mutikainen, and L. F. Delph. 2001. Genetics of sex determination in the gynodioecious species Lobelia siphilitica: evidence from two populations. Heredity 86:265-276. Dukas, R. 2001. Effects of perceived danger on flower choice by bees. Ecology Letters 4:327333. Dukas, R. 2005. Bumble bee predators reduce pollinator density and plant fitness. Ecology 86:1401-1406. Dukas, R. and D. H. Morse. 2003. Crab spiders affect flower visitation by bees. Oikos 101:157163. Elliott, N. B. and W. M. Elliott. 1991. Effect of an ambush predator, Phymata-americana melin, on behavior of insects visiting Daucus carota. American Midland Naturalist 126:198202. 29 Elliott, N. B. and L. M. Elliott. 1994. Recognition and avoidance of the predator phymataAmericana melin on Solidago odora ait. by late-season floral visitors. American Midland naturalist 131:378-380. Estes, J. A., J. S. Brashares, and M. E. Power. 2013. Predicting and detecting reciprocity between indirect ecological interactions and evolution. American Naturalist 181:S76-S99. Estes, J. A., J. Terborgh, J. S. Brashares, M. E. Power, J. Berger, W. J. Bond, S. R. Carpenter, T. E. Essington, R. D. Holt, J. B. C. Jackson, R. J. Marquis, L. Oksanen, T. Oksanen, R. T. Paine, E. K. Pikitch, W. J. Ripple, S. A. Sandin, M. Scheffer, T. W. Schoener, J. B. Shurin, A. R. E. Sinclair, M. E. Soule, R. Virtanen, and D. A. Wardle. 2011. Trophic downgrading of planet earth. Science 333:301-306. Flanagan, R. J., R. J. Mitchell, and J. D. Karron. 2011. Effects of multiple competitors for pollination on bumblebee foraging patterns and Mimulus ringens reproductive success. Oikos 120:200-207. Galen, C. 1997. Rates of floral evolution: adaptation to bumblebee pollination in an alpine wildflower, Polemonium viscosum. Evolution 51:1025-1025. Galen, C. and J. Cuba. 2001. Down the tube: pollinators, predators, and the evolution of flower shape in the alpine skypilot, Polemonium viscosum. Evolution 55:1963-1971. Gomez, J. M., J. Bosch, F. Perfectti, J. D. Fernandez, M. Abdelaziz, and J. P. M. Camacho. 2008. Spatial variation in selection on corolla shape in a generalist plant is promoted by the preference patterns of its local pollinators. Proceedings of the Royal Society BBiological Sciences 275:2241-2249. Goncalves-Souza, T., P. M. Omena, J. C. Souza, and G. Q. Romero. 2008. Trait-mediated effects on flowers: artificial spiders deceive pollinators and decrease plant fitness. Ecology 89:2407-2413. Greco, C. E. and P. G. Kevan. 1994. Contrasting patch choosing by anthophilous ambush predators - vegetation and floral cues for decisions by a crab spider (Misumena vatia) and males and females of an ambush bug (Phymata americana). Canadian Journal of Zoology-Revue Canadienne De Zoologie 72:1583-1588. Greco, C. F. and P. G. Kevan. 1995. Patch choice in the anthophilous ambush predator Phymata americana: improvement by switching hunting sites as part of the initial choice. Canadian Journal of Zoology-Revue Canadienne De Zoologie 73:1912-1917. Grill, C. P. and V. N. Rush. 2000. Analysing spectral data: comparison and application of two techniques. Biological Journal of the Linnean Society 69:121-138. 30 Guariento, R. D., B. Luttbeg, T. Mehner, and F. d. A. Esteves. 2014. The effect of predation pressure and predator adaptive foraging on the relative importance of consumptive and non-consumptive predator net effects in a freshwater model system. Oikos 123:705-713. Hairston, N. G., F. E. Smith, and L. B. Slobodkin. 1960. Community structure, population control, and competition. American Naturalist 94:421-425. Hanna, C. J. and P. K. Eason. 2013. Juvenile crab spiders (Mecaphesa asperata) use indirect cues to choose foraging sites. Ethology Ecology & Evolution 25:161-173. Heiling, A. M., K. Cheng, and M. E. Herberstein. 2004. Exploitation of floral signals by crab spiders (Thomisus spectabilis, Thomisidae). Behavioral Ecology 15:321-326. Higginson, A. D., G. D. Ruxton, and J. Skelhorn. 2010. The impact of flower-dwelling predators on host plant reproductive success. Oecologia 164:411-421. Huey, R. B. and E. R. Pianka. 1981. Ecological consequences of foraging mode. Ecology 62:991-999. Ings, T. C. and L. Chittka. 2008. Speed-accuracy tradeoffs and false alarms in bee responses to cryptic predators. Current Biology 18:1520-1524. Ings, T. C. and L. Chittka. 2009. Predator crypsis enhances behaviourally mediated indirect effects on plants by altering bumblebee foraging preferences. Proceedings of the Royal Society B-Biological Sciences 276:2031-2036. Johnson, S. D. and K. E. Steiner. 1997. Long-tongued fly pollination and evolution of floral spur length in the Disa draconis complex (Orchidaceae). Evolution 51:45-53. Johnston, M. O. 1991a. Natural-selection on floral traits in two species of Lobelia with Different Pollinators. Evolution 45:1468-1479. Johnston, M. O. 1991b. Pollen limitation of female reproduction in Lobelia cardinalis and L. siphilitica. Ecology 72:1500-1503. Johnston, M. O. 1992. Effects of cross and self-fertilization on progeny fitness in Lobelia cardinalis and L. siphilitica. Evolution 46:688-702. Jones, C. D., A. Kingsley, P. Burke, and M. Holder. 2008. Field Guide to the Damselflies and Dragonflies of Algonquin Provincial Park and the surrounding area. Friends of Algonquin Park. Jones, E. I. 2010. Optimal foraging when predation risk increases with patch resources: an analysis of pollinators and ambush predators. Oikos 119:835-840. 31 Jones, E. I. and A. Dornhaus. 2011. Predation risk makes bees reject rewarding flowers and reduce foraging activity. Behavioral Ecology and Sociobiology 65:1505-1511. Keiser, C. N., J. B. Slyder, W. P. Carson, and J. N. Pruitt. 2015. Individual differences in predators but not producers mediate the magnitude of a trophic cascade. Arthropod-Plant Interactions 9:225-232. Knight, T. M., J. M. Chase, H. Hillebrand, and R. D. Holt. 2006. Predation on mutualists can reduce the strength of trophic cascades. Ecology Letters 9:1173-1178. Knight, T. M., M. W. McCoy, J. M. Chase, K. A. McCoy, and R. D. Holt. 2005a. Trophic cascades across ecosystems. Nature 437:880-883. Knight, T. M., J. A. Steets, J. C. Vamosi, S. J. Mazer, M. Burd, D. R. Campbell, M. R. Dudash, M. O. Johnston, R. J. Mitchell, and T. L. Ashman. 2005b. Pollen limitation of plant reproduction: pattern and process. Annual Review of Ecology Evolution and Systematics 36:467-497. Lande, R. and S. J. Arnold. 1983. The measurement of selection on correlated characters. Evolution 37:1210-1226. Lima, S. L. 1998. Nonlethal effects in the ecology of predator-prey interactions - What are the ecological effects of anti-predator decision-making? Bioscience 48:25-34. Llandres, A. L., E. De Mas, and M. A. Rodriguez-Girones. 2012. Response of pollinators to the tradeoff between resource acquisition and predator avoidance. Oikos 121:687-696. Mason, L. G. 1986. Free-loaders, free-lancers and bushwhackers - sexual dimorphism and seasonal-changes in prey-capture behavior of ambush bugs. American Midland Naturalist 116:323-328. Meehan, T. D., H. M. Lease, and B. O. Wolf. 2005. Negative indirect effects of an avian insectivore on the fruit set of an insect-pollinated herb. Oikos 109:297-304. Menge, B. A., C. Blanchette, P. Raimondi, T. Freidenburg, S. Gaines, J. Lubchenco, D. Lohse, G. Hudson, M. Foley, and J. Pamplin. 2004. Species interaction strength: Testing model predictions along an upwelling gradient. Ecological Monographs 74:663-684. Møller, A. P. 1995. Bumblebee preference for symmetrical flowers. Proceedings of the National Academy of Sciences of the United States of America 92:2288-2292. Morrissey, M. B. and J. D. Hadfield. 2012. Directional selection in temporally replicated studies is remarkably consistent. Evolution 66:435-442. Morse, D. H. 1979. Prey capture by the crab spider Misumena calycina (Araneae, Thomisidae). Oecologia 39:309-319. 32 Morse, D. H. 1988. Cues associated with patch-choice decisions by foraging crab spiders Misumena vatia. Behaviour 107:297-313. Morse, D. H. 1993. Choosing hunting sites with little information: patch-choice responses of crab spiders to distant cues. Behavioral Ecology 4:61-65. Morse, D. H. 1999. Choice of hunting site as a consequence of experience in late-instar crab spiders. Oecologia 120:252-257. Morse, D. H. 2000. Flower choice by naive young crab spiders and the effect of subsequent experience. Animal Behaviour 59:943-951. Olberg, R. M., A. H. Worthington, and K. R. Venator. 2000. Prey pursuit and interception in dragonflies. Journal of Comparative Physiology a-Sensory Neural and Behavioral Physiology 186:155-162. Owen, C. R. and H. D. Bradshaw. 2011. Induced mutations affecting pollinator choice in Mimulus lewisii (Phrymaceae). Arthropod-Plant Interactions 5:235-244. Pace, M. L., J. J. Cole, S. R. Carpenter, and J. F. Kitchell. 1999. Trophic cascades revealed in diverse ecosystems. Trends in Ecology & Evolution 14:483-488. Paine, R. T. 1980. Food webs - linkage, interaction strength and community infrastructure - the 3rd Tansley Lecture. Journal of Animal Ecology 49:667-685. Parachnowitsch, A. L. and C. M. Caruso. 2008. Predispersal seed herbivores, not pollinators, exert selection on floral traits via female fitness. Ecology 89:1802-1810. Pianka, E. R. 1966. Convexity desert lizards and spatial heterogeneity. Ecology 47:1055-&. Preisser, E. L., D. I. Bolnick, and M. F. Benard. 2005. Scared to death? The effects of intimidation and consumption in predator-prey interactions. Ecology 86:501-509. Reader, T., A. D. Higginson, C. J. Barnard, and F. S. Gilbert. 2006. The effects of predation risk from crab spiders on bee foraging behavior. Behavioral Ecology 17:933-939. Rivkin, L. R., A. L. Case, and C. M. Caruso. 2015. Frequency-dependent fitness in gynodioecious Lobelia siphilitica. Evolution 69:1232-1243. Robertson, I. C. and D. K. Maguire. 2005. Crab spiders deter insect visitations to slickspot peppergrass flowers. Oikos 109:577-582. Rodriguez-Girones, M. A. and J. Bosch. 2012. Effects of body size and sociality on the antipredator behaviour of foraging bees. Oikos 121:1473-1482. 33 Romero, G. Q., P. A. P. Antiqueira, and J. Koricheva. 2011. A meta-analysis of predation risk effects on pollinator behaviour. Plos One 6:20689-20689. Schmitz, O. J., A. P. Beckerman, and K. M. Obrien. 1997. Behaviorally mediated trophic cascades: effects of predation risk on food web interactions. Ecology 78:1388-1399. Schoener, T. W. 1971. Theory of feeding strategies. A Rev Ecol Syst 2:369-404. Siepielski, A. M., K. M. Gotanda, M. B. Morrissey, S. E. Diamond, J. D. DiBattista, and S. M. Carlson. 2013. The spatial patterns of directional phenotypic selection. Ecology Letters 16:1382-1392. Sokal, R. R. and F. J. Rohlf. 1995. Biometry: the principles and practice of statistics in biological research. Third edition. Suttle, K. B. 2003. Pollinators as mediators of top-down effects on plants. Ecology Letters 6:688-694. Thompson, J. N. 2005. The geographic mosaic of coevolution. The geographic mosaic of coevolution. 1-443. Trussell, G. C., P. J. Ewanchuk, and C. M. Matassa. 2006. Habitat effects on the relative importance of trait- and density-mediated indirect interactions. Ecology Letters 9:12451252. Vanhoenacker, D., J. Agren, and J. Ehrlen. 2013. Non-linear relationship between intensity of plant-animal interactions and selection strength. Ecology Letters 16:198-205. Wade, M. J. and S. Kalisz. 1990. The causes of natural-selection. Evolution 44:1947-1955. Wang, M.Y., T. C. Ings, M. J. Proulx, and L. Chittka. 2013. Can bees simultaneously engage in adaptive foraging behaviour and attend to cryptic predators? Animal Behaviour 86:859866. Warren, A. 1915. Dragonflies and their food. Proceedings of the Hawaiian Entomological Society 3:72-82. Wootton, J. T. and M. Emmerson. 2005. Measurement of interaction strength innature. Annual Review of Ecology Evolution and Systematics 36:419-444. Wright, M. 1944. Some random observations on dragonfly habits with notes on their predaccousness on bees. Journal of the Tennessee Academy of Science 19:295-301. Yong, T. H. 2005. Prey capture by a generalist predator on flowering and nonflowering ambush sites: Are inflorescences higher quality hunting sites? Environmental Entomology 34:969-976 34 Figure 1. Map of field site locations at the Koffler Scientific Reserve. Field sites 1 and 2 are less than 100m from ponds. Fields sites 3 and 4 are greater than 500m from ponds. Field sites 1 and 3 were found to have high dragonfly abundance, and field sites 2 and 4 were found to have low dragonfly abundance. 35 Figure 2. Floral traits of Lobelia siphilitica: a) corolla length was measured from the base of the bracts to the corolla tube opening, b) width of the corolla tube opening was measured from the bottom of the corolla tube to the top at the point where the corolla tube opening begins, c) banner petal size was measured as the total area of the top and bottom petals outlined in white, d) banner petal shape was measured as the ratio of length to width of the bottom three banner petals, which serve as a landing platform, e) the nectar guide stripes were measured as the white stripes (arrow 1) and the purple background (arrow 2), nectar guide contrast was measured by subtracting the modal gray value of purple background from the modal gray value of the white stripes, and f) the area of the corolla underside was measured as the area outlined in white, nectar guide size was measured by dividing the area of the underside of the corolla tube (f) by the area of the nectar guide stripes (e. arrow 1) 36 Figure 3. Relationship between daily display size and fruits per plant for Lobelia siphilitica in low- and high-dragonfly abundance treatments. Fruit per plant is a female fitness measure, and was relativized for each treatment by dividing each plant’s fitness by the mean plant fitness within the treatment. Daily display size was standardized by subtracting the treatment mean daily display from each plant’s daily display size, and then dividing by the treatment standard deviation, resulting in a mean of zero and a variance of one. 37 Figure 4. Relationship between daily display size, and fruits per plant (a, b) and seeds per plant (c, d) for Lobelia siphilitica in ambush bug absent and ambush bug present treatments. Fruit per plant, and seeds per plant are female fitness measures. They were relativized for each treatment by dividing each plant’s fitness by the mean plant fitness within the treatment. Daily display size was standardized by subtracting the treatment mean daily display from each plant’s daily display size, and then dividing by the treatment standard deviation, resulting in a mean of zero and a variance of one. 38 Table 1. Split-plot analysis of variance model of the effect of dragonfly treatment (low abundance vs. high abundance), pollination treatment (hand vs. open), and the interaction between the dragonfly and pollination treatments on three female fitness components of Lobelia siphilitica plants. Fitness measure Fruit per plant Seeds per fruit Seeds per plant Source Fdf P Dragonfly Pollination Dragonfly × Pollination Dragonfly Pollination Dragonfly × Pollination Dragonfly Pollination Dragonfly × Pollination 0.7011,3.4 0.2341,44.5 0.1471,44.5 0.2721,3.6 3.2111,82.4 0.0751,82.4 0.0521,3.6 1.6631,81.7 0.4191,81.7 0.458 0.631 0.703 0.633 0.077 0.785 0.833 0.201 0.519 39 Table 2. Mean values for three female fitness components ± 1 SE (N) for hand- and openpollinated Lobelia siphilitica plants. Fitness measure Hand-pollinated Open-pollinated Fruit per plant Seeds per fruit Seeds per plant 88.73 ± 3.84 (93) 633.23 ± 18.55 (86) 56496.71 ± 2811.04 (86) 89.54 ± 2.72 (195) 581.15 ± 14.86 (168) 53079.06 ± 2200.91 (168) 40 Table 3. Mean values for three female fitness components ± 1 SE (N) for open-pollinated Lobelia siphilitica plants in low- and high-dragonfly abundance treatments. Fitness measure Fruit per plant Seeds per fruit Seeds per plant Low-dragonfly abundance 85.041 ± 7.843 (97) 605.803 ± 47.441 (84) 52,274.34 ± 8821.595 (84) 41 High-dragonfly abundance 93.908 ± 7.843 (98) 556.493 ± 47.441 (84) 53,883.777 ± 8821.595 (84) Table 4. Linear selection gradients ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants in the low- and highdragonfly abundance treatments. F compares the strength of selection between the low- and high-dragonfly abundance treatments. Fitness measure Fruit per plant Trait Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening Low-dragonfly abundance -0.072 ± 0.038 -0.014 ± 0.053 0.044 ± 0.039 -0.074 ± 0.037* -0.028 ± 0.039 0.140 ± 0.037** 0.059 ± 0.037 -0.029 ± 0.037 -0.014 ± 0.039 0.034 ± 0.047 N = 97 High-dragonfly abundance 0.001 ± 0.038 -0.003 ± 0.044 -0.102 ± 0.039* -0.106 ± 0.039** -0.055 ± 0.043 0.298 ± 0.042** 0.007 ± 0.042 0.061 ± 0.044 0.029 ± 0.040 0.028 ± 0.045 N = 98 Fdf 2.2031,170NS 0.0021,172NS 7.5951,172** 0.1301,172NS 0.1991,171NS 10.2251,171** 0.9441,171NS 2.8521,168NS 0.6231,171NS 0.0191,171NS Seeds per fruit Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.055 ± 0.036 0.022 ± 0.057 -0.079 ± 0.037* 0.049 ± 0.035 0.023 ± 0.038 0.013 ± 0.035 0.057 ± 0.036 0.040 ± 0.035 -0.070 ± 0.038 0.057 ± 0.047 N = 84 0.026 ± 0.041 0.025 ± 0.047 -0.040 ± 0.042 -0.017 ± 0.042 0.033 ± 0.046 -0.016 ± 0.045 0.052 ± 0.044 0.022 ± 0.050 0.008 ± 0.044 -0.009 ± 0.050 N = 84 0.4861,143NS 0.0041,153NS 0.3811,144NS 1.2941,144NS 0.0571,144NS 0.3521,143NS 0.0501,143NS 0.0661,145NS 1.1241,143NS 0.7791,143NS Seeds per plant Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening -0.043 ± 0.051 -0.031 ± 0.080 -0.039 ± 0.053 -0.007 ± 0.049 -0.010 ± 0.054 0.163 ± 0.049** 0.084 ± 0.051 0.027 ± 0.050 -0.062 ± 0.054 0.162 ± 0.067* N = 84 0.034 ± 0.056 0.001 ± 0.064 -0.154 ± 0.057** -0.101 ± 0.057 -0.030 ± 0.062 0.296 ± 0.060 ** 0.094 ± 0.060 0.149 ± 0.067* 0.026 ± 0.059 -0.019 ± 0.067 N = 84 1.0921,143NS 0.0501,143NS 2.4031,144NS 1.3471,144NS 0.0461,143NS 2.6231,143NS 0.0011,143NS 2.8781,145NS 0.8801,143NS 3.2371,143NS * P < 0.05, ** P ≤ 0.01 42 Table 5. Linear selection differentials ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants in the low- and high-dragonfly abundance treatments. F compares the strength of selection between the low- and high-dragonfly abundance treatments. Fitness measure Fruit per plant Trait Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening Low-dragonfly abundance -0.037 ± 0.037 0.028 ± 0.039 0.035 ± 0.038 -0.063 ± 0.038 0.000 ± 0.038 0.142 ± 0.036** 0.095 ± 0.036** -0.032 ± 0.037 -0.007 ± 0.038 0.050 ± 0.038 N = 97 High-dragonfly abundance 0.006 ± 0.047 0.028 ± 0.047 -0.016 ± 0.046 -0.062 ± 0.047 -0.015 ± 0.046 0.261 ± 0.039 ** 0.089 ± 0.046 0.011 ± 0.049 0.049 ± 0.046 0.032 ± 0.047 N = 98 Fdf 0.6481,188NS 0.0241,190NS 0.7801,189NS 0.0401,190NS 0.0891,188NS 6.9591,189** 0.1831,188NS 0.6091,190NS 0.9401,188NS 0.1471,190NS Seeds per fruit Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.032 ± 0.034 0.057 ± 0.035 -0.011 ± 0.034 0.038 ± 0.034 0.052 ± 0.034 0.030 ± 0.034 0.056 ± 0.033 0.044 ± 0.033 -0.048 ± 0.35 0.073 ± 0.033* N = 84 0.004 ± 0.038 0.038 ± 0.038 -0.042 ± 0.037 -0.024 ± 0.038 0.049 ± 0.037 -0.009 ± 0.038 0.050 ± 0.037 0.004 ± 0.040 0.017 ± 0.038 0.018 ± 0.038 N = 84 0.3481,162NS 0.3051,162NS 0.4631,162NS 1.3121,163NS 0.0011,161NS 0.6861,161NS 0.1201,161NS 0.6091,163NS 1.1481,161NS 1.1311,162NS Seeds per plant Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening -0.035 ± 0.051 0.106 ± 0.052* 0.016 ± 0.052 -0.012 ± 0.052 0.048 ± 0.051 0.178 ± 0.048** 0.134 ± 0.049** 0.019 ± 0.051 -0.036 ± 0.053 0.158 ± 0.049** N = 84 0.024 ± 0.062 0.039 ± 0.063 -0.048 ± 0.062 -0.043 ± 0.063 0.004 ± 0.062 0.266 ± 0.055** 0.161 ± 0.060** 0.074 ± 0.065 0.023 ± 0.063 0.028 ± 0.062 N = 84 0.6061,161NS 1.0061,16NS 0.6151,161NS 0.0861,162NS 0.3771,161NS 1.3581,161NS 0.0011,161NS 0.7071,163NS 0.2661,161NS 2.4501,162NS * P < 0.05, ** P < 0.01 43 Table 6. Linear selection gradients ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of hand- and open-pollinated Lobelia siphilitica plants. F compares the strength of selection between the hand- and open-pollination treatments. Fitness measure Fruit per plant Seeds per fruit Seeds per plant Trait Hand-pollinated Open-pollinated Fdf Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.003 ± 0.038 0.061 ± 0.047 0.013 ± 0.038 -0.112 ± 0.038** -0.116 ± 0.042** 0.255 ± 0.043** 0.009 ± 0.042 -0.024 ± 0.036 0.087 ± 0.037* 0.043 ± 0.045 -0.018 ± 0.027 0.009 ± 0.034 -0.032 ± 0.028 -0.082 ± 0.027** -0.044 ± 0.029 0.206 ± 0.028** 0.038 ± 0.028 0.010 ± 0.029 0.002 ± 0.028 0.032 ± 0.033 0.1331,262NS 0.4801,262NS 0.5931,262NS 0.0821, 262NS 1.256 1,262NS 1.071 1,262NS 0.313 1,262NS 0.5561,262NS 3.5781,262NS 0.018 1,262NS N = 95 N = 198 -0.026 ± 0.033 -0.018 ± 0.040 -0.046 ± 0.033 -0.009 ± 0.032 0.021 ± 0.035 0.000 ± 0.036 0.013 ± 0.037 -0.011 ± 0.031 -0.012 ± 0.032 0.087 ± 0.038* 0.040 ± 0.026 0.028 ± 0.034 -0.053 ± 0.027* 0.015 ± 0.026 0.024 ± 0.028 -0.002 ± 0.027 0.057 ± 0.027* 0.024 ± 0.028 -0.043 ± 0.027 0.028 ± 0.033 N = 86 N = 168 0.025 ± 0.046 0.067 ± 0.057 -0.027 ± 0.047 -0.145 ± 0.045** -0.102 ± 0.049* 0.256 ± 0.052** 0.013 ± 0.052 -0.019 ± 0.044 0.084 ± 0.046 0.089 ± 0.054 0.017 ± 0.037 0.016 ± 0.048 -0.100 ± 0.038** -0.048 ± 0.037 -0.027 ± 0.040 0.218 ± 0.039** 0.093 ± 0.039* 0.062 ± 0.039 -0.044 ± 0.039 N = 86 N = 168 Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening * P < 0.05, ** P < 0.01 44 0.077 ± 0.046 2.4131, 228NS 0.8371, 228NS 0.0011, 228NS 0.1561, 228NS 0.0031, 228NS 1.1 x 10-51, 228NS 0.8031, 228NS 0.6401, 228NS 0.5671, 228NS 1.1331, 228NS 1.17 x 10-41,228NS 0.2481,228NS 0.6291,228NS 1.5541,228NS 0.7371,228NS 0.1241,228NS 1.2841,228NS 1.5791,228NS 4.3881,228* 5.8 x 10-51,228NS Table 7. Linear selection differentials ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of hand- and open-pollinated Lobelia siphilitica plants. F compares the strength of selection between the hand- and open-pollination treatments. Fitness measure Fruit per plant Trait Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening Handpollinated -0.078 ± 0.044 0.099 ± 0.044* 0.036 ± 0.048 -0.108 ± 0.044* -0.016 ± 0.046 0.234 ± 0.041** 0.044 ± 0.046 -0.005 ± 0.045 0.058 ± 0.044 0.059 ± 0.045 N = 95 Open-pollinated Fdf -0.012 ± 0.030 0.027 ± 0.031 0.007 ± 0.030 -0.062 ± 0.030* -0.006 ± 0.030 0.201 ± 0.027** 0.093 ± 0.029** -0.010 ± 0.031 0.015 ± 0.030 0.040 ± 0.031 N = 198 1.4911,280NS 0.9111,280NS 0.3471,280NS 0.1191,280NS 0.0121,280NS 0.3141,280NS 0.5541,280NS 0.0101,280NS 0.7011,280NS 4.72 x 10-41,280NS Seeds per fruit Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening -0.017 ± 0.030 0.044 ± 0.029 -0.039 ± 0.031 -0.002 ± 0.031 0.037 ± 0.030 0.014 ± 0.032 0.038 ± 0.030 -0.009 ± 0.029 -0.013 ± 0.030 0.081 ± 0.028** N = 86 0.016 ± 0.025 0.048 ± 0.026 -0.026 ± 0.025 0.008 ± 0.025 0.050 ± 0.025* 0.011 ± 0.025 0.052 ± 0.025* 0.026 ± 0.025 -0.025 ± 0.026 0.048 ± 0.026 N = 168 0.7911, 246NS 0.0041, 246NS 0.2911, 246NS 0.0601, 246NS 0.0641, 246NS 0.0301, 246NS 0.2731, 246NS 0.7521, 246NS 0.1701, 246NS 0.5121, 246NS Seeds per plant Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening -0.064 ± 0.053 0.127 ± 0.050* -0.002 ± 0.056 -0.135 ± 0.051* -0.007 ± 0.052 0.239 ± 0.049** 0.075 ± 0.053 -0.008 ± 0.052 0.054 ± 0.052 0.104 ± 0.051* N = 86 0.000 ± 0.040 0.070 ± 0.041 -0.018 ± 0.040 -0.028 ± 0.041 0.027 ± 0.040 0.223 ± 0.037** 0.149 ± 0.038** 0.044 ± 0.0± 0.040 -0.012 ± 0.041 0.097 ± 0.041* N = 168 1.0291,246NS 0.3631,246NS 0.0011,246NS 1.6841,246NS 0.2471,246NS 0.0911,246NS 1.5811,246NS 0.6361,246NS 1.2101,246NS 0.0511,246NS * P < 0.05, P <0.0 45 Table 8. Mean trait values ± 1 SE (N) for open-pollinated Lobelia siphilitica plants with ambush bugs present and absent. t compares mean trait values between plants with ambush bugs present and absent. Traits Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening * P < 0.05 Ambush bugs Absent 1.402 ± 0.019 (68) 1.510 ± 0.044 (68) -0.124 ± 0.108 (68) 0.189 ± 0.121 (68) 1.345 ± 0.010 (68) 15.327 ± 0.689 (68) 46.087 ± 1.321 (68) 23.265 ± 1.986(68) 6.977 ± 0.752 (68) 0.587 ± 0.008 (68) Ambush bugs Present 1.416 ± 0.016 (127) 1.532 ± 0.031 (127) 0.044 ± 0.090 (127) -0.022 ± 0.086 (127) 1.336 ± 0.010 (127) 17.801 ± 0.622 (127) 47.104 ± 0.823 (127) 22.126 ± 1.350 (127) 6.014 ± 0.206 (127) 0.588 ± 0.006 (127) 46 tdf P 0.510193 0.414193 1.147193 -1.438 193 -0.615193 2.502193* 0.686193 -0.485193 -1.23577.186 0.109193 0.610 0.680 0.253 0.152 0.539 0.013* 0.493 0.628 0.221 0.913 Table 9. Linear selection gradients ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants with ambush bugs present and absent. F compares the strength of selection between plants with ambush bugs present and absent. Ambush bugs Fitness measure Fruit per plant Trait Absent Present F1, 143 Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.039 (0.046) 0.006 (0.056) -0.018 (0.050) -0.108 (0.050)* 0.000 (0.051) 0.308 (0.053)** -0.031 (0.052) 0.058 (0.056) 0.025 (0.050) 0.034 (0.052) N = 68 -0.061 (0.034) -0.020 (0.045) -0.047 (0.034) -0.079 (0.033)* -0.053 (0.037) 0.163 (0.033)** 0.070 (0.034)* -0.023 (0.036) -0.036 (0.036) 0.054 (0.043) N = 127 2.466 0.332 0.339 0.139 0.630 7.848** 2.944 0.187 0.408 0.017 Seeds per fruit Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.113 (0.049)* 0.018 (0.058) -0.092 (0.056) 0.019 (0.053) 0.048 (0.056) -0.016 (0.057) 0.057 (0.056) 0.024 (0.059) -0.015 (0.053) 0.018 (0.056) N = 109 0.009 (0.031) 0.066 (0.044) -0.034 (0.031) 0.016 (0.030) -0.008 (0.035) 0.015 (0.031) 0.066 (0.031)* 0.001 (0.033) -0.070 (0.034)* 0.023 (0.041) N = 109 4.292* 0.811 1.088 0.112 1.428 0.695 2.61 x 10-4 0.281 0.324 0.023 Seeds per Plant Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.124 (0.061)* 0.018 (0.073) -0.149 (0.07)* -0.084 (0.067) 0.035 (0.07) 0.377 (0.072)** 0.028 (0.071) 0.127 (0.074) -0.011 (0.067) 0.013 (0.071) N = 68 -0.041 (0.046) 0.019 (0.066) -0.09 (0.047) -0.043 (0.045) -0.068 (0.052) 0.164 (0.046)** 0.138 (0.046)** 0.002 (0.050) -0.078 (0.051) 0.112 (0.062) N = 109 4.953* 0.002 5.85 0.155 1.902 5.414* 1.588 2.121 0.554 1.198 * P < 0.05, ** P < 0.01 47 Table 10. Linear selection differentials ± 1 SE estimated via three female fitness components for 10 floral and inflorescence traits of open-pollinated Lobelia siphilitica plants with ambush bugs present and absent. F compares the strength of selection between plants with ambush bugs present and absent. Ambush bugs Fitness measure Fruit per plant Trait Absent Present F1, 161 Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.051 (0.054) 0.079 (0.055) 0.074 (0.054) -0.029 (0.055) 0.017 (0.055) 0.280 (0.045)** 0.070 (0.054) 0.024 (0.056) 0.021 (0.057) 0.080 (0.055) N = 68 -0.044 (0.036) -0.010 (0.038) -0.025 (0.036) -0.078 (0.036)* -0.013 (0.036) 0.166 (0.033)** 0.110 (0.035)** -0.020 (0.037) -0.016 (0.037) 0.026 (0.038) N = 127 0.850 4.615* 0.039 1.545 0.497 5.962* 0.027 0.155 0.863 1.876 Seeds per fruit Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.087 (0.045) 0.033 (0.046) -0.075 (0.046) 0.005 (0.046) 0.047 (0.046) -0.003 (0.048) 0.027 (0.046) -0.017 (0.048) 0.010 (0.048) 0.024 (0.047) N = 59 -0.007 (0.030) 0.065 (0.030)* -0.003 (0.029) 0.01 (0.030) 0.042 (0.029) 0.029 (0.029) 0.069 (0.028)* 0.035 (0.030) -0.043 (0.03) 0.054 (0.03) N = 109 3.512 1.008 2.318 0.006 0.127 0.998 0.738 0.583 0.146 0.365 Seeds per plant Banner petal shape Banner petal size Brightness Chroma Corolla length Daily display Inflorescence height Nectar guide contrast Nectar guide size Width of tube opening 0.138 (0.072) 0.109 (0.073) 0.025 (0.075) 0.023 (0.073) 0.042 (0.075) 0.337 (0.061)** 0.091 (0.072) 0.060 (0.076) 0.002 (0.077) 0.112 (0.074) N = 59 -0.054 (0.049) 0.055 (0.051) -0.033 (0.049) -0.057 (0.049) 0.014 (0.049) 0.187 (0.045)** 0.187 (0.045)** 0.027 (0.050) -0.018 (0.051) 0.092 (0.050) N = 109 5.168* 0.247 0.321 0.864 0.200 2.582 1.525 0.109 0.031 0.015 * P < 0.05, ** P < 0.01 48 Appendix A Bumblebee Abundance Bumblebee abundance at each field site was determined during the dragonfly point counts. During each point count in addition to counting the number of dragonflies, the observers noted the number of bumblebees that entered their field of view. While bumblebees were not identified to species level, the majority of bumblebees counted were actively foraging on Lobelia siphilitica flowers (personal observation). Sites one through three had similar bumblebee abundance, mean (1 SE) 8.4 (1.79); range: 0-51, 9.5 (1.75); range: 1-40, and 7.8 (1.07); range: 226) bumblebees per 20 minutes of observation, respectively. Site four however, consistently had higher bumblebee abundance, (mean bumblebees/20 minutes of observation (1 SE) 17.8 (1.52); range:10-41). Field site 4 was far from ponds and had low dragonfly abundance (Fig. 1). 49
© Copyright 2026 Paperzz