Reduced visitation to the buzz-pollinated Cyanella hyacinthoides in the presence of other pollen sources in the hyperdiverse Cape Floristic Region

Many plant species have floral morphologies that restrict access to floral resources, such as pollen or nectar, and only a subset of floral visitors can perform the complex handling behaviours required to extract restricted resources. Due to the time and energy required to extract resources from morphologically complex flowers, these plant species potentially compete for pollinators with co-flowering plants that have more easily accessible resources. A widespread floral mechanism restricting access to pollen is the presence of tubular anthers that open through small pores or slits (poricidal anthers). Some bees have evolved the capacity to remove pollen from poricidal anthers using vibrations, giving rise to the phenomenon of buzz-pollination. These bee vibrations that are produced for pollen extraction are presumably energetically costly, and to date, few studies have investigated whether buzz-pollinated flowers may be at a disadvantage when competing for pollinators’ attention with plant species that present unrestricted pollen resources. Here, we studied Cyanella hyacinthoides (Tecophilaeaceae), a geophyte with poricidal anthers in the hyperdiverse Cape Floristic Region of South Africa, to assess how the composition and relative abundance of flowers with easily accessible pollen affect bee visitation to a buzz-pollinated plant. We found that the number of pollinator species was not influenced by community composition. However, visitation rates to C. hyacinthoides were negatively related to the abundance of flowers with more accessible resources. Visitation rates were strongly associated with petal colour, showing that flower colour is important in mediating these interactions. We conclude that buzz-pollinated plants might be at a competitive disadvantage when many easily accessible pollen sources are available, particularly when competitor species share its floral signals.


Introduction 32
The majority of flowering plants are pollinated by animals (Ollerton et al. 2011), and most animal-33 pollinated species offer resources such as pollen, nectar, oils and scents as rewards to attract floral 34 visitors. The degree to which these resources are accessible to floral visitors varies between plant 35 species, and the accessibility of resources is often modulated through morphological restrictions, 36 such as nectar tubes or keel flowers ( The choices of floral visitors in communities where resources are available in a range of restrictions 56 levels are likely contingent on the identity of competitor species (Stout et al. 1998), abundances of 57 plant species (Kunin and Iwasa 1996), and degree of floral trait overlap between plant species 58 (Hargreaves et al. 2009, Lázaro et al. 2013). Thus, plant species that restrict access to resources can 59 potentially be at a competitive disadvantage under certain conditions, and this is likely contingent on 60 the abundance of unrestricted resources offered by the co-flowering community, as well as the 61 degree of floral trait overlap. 62 One way in which plants that offer pollen as primary reward can restrict access to pollen grains is 63 through poricidal anthers (Buchmann 1983, van der Kooi et al. 2021). Some species of bees have 64 evolved the capacity to produce vibrations (also known as floral vibrations or sonication) that 65 facilitate the removal of pollen grains from poricidal anthers (Buchmann 1983, De Luca and Vallejo-66 Marín 2013). The interaction between plants with specialised floral morphologies, such as poricidal 67 anthers, and the bee behaviour of deploying floral vibrations has given rise to the phenomenon of 68 buzz pollination (Buchmann 1983, Vallejo-Marín 2019). During buzz-pollination, bees typically grasp 69 the anthers with their mandibles, curl their bodies around the anthers, and then generate vibrations 70 that result in pollen being released from the anthers through apical slits or pores (De Luca and 71 Vallejo-Marín 2013). Using vibrations for pollen extraction is likely energetically expensive, as the 72 production of floral vibrations by bees involves rapid contraction of the same thoracic muscles that 73 power energetically costly wingbeat during flight (King and Buchmann 2003). During flight, these 74 muscles consume as much as 100 times the energy than the resting metabolic rate (Dudley 2002). 75 Floral vibrations have higher frequency and amplitude (velocity, acceleration, and displacement) 76 than flight vibrations (Pritchard and Vallejo-Marín 2020) and, therefore, it is likely that floral 77 vibrations are equally or more energetically costly as those produced during flight. Because of the 78 energetic costs associated with vibratile pollen extraction, we might expect bees to favour more 79 easily accessible pollen resources under certain circumstances. Buzz-pollination is prevalent among 80 both plants (6-8% of all plant species across 65 families (Buchmann 1983) and bees (half of all bee 81 species (Cardinal et al. 2018)), however, our understanding of how visitation to buzz-pollinated 82 plants is influenced by the availability of unrestricted pollen resources in the surrounding plant 83 community is limited. 84 Recent work has shown that competition for pollination services between buzz-pollinated individuals 85 is prevalent (Mesquita-Neto et al. 2018, Soares et al. 2020). It is likely that these plant species also 86 compete for pollinator services with non-poricidal taxa with unrestricted pollen resources. We 87 hypothesise that due to the energetic and, potentially, learning costs associated with using 88 vibrations for pollen extraction (Laverty 1980, Russell et al. 2016, visitation to plants with poricidal 89 anthers should be reduced when flowers that do not restrict access to pollen are available in high 90 abundances in a community. An alternative hypothesis is that because only a subset of floral visitors 91 in a community can use vibrations to extract pollen, plants with poricidal anthers could potentially 92 act as a private and reliable pollen resource to particular bee species, which could result in 93 consistent visitation from these bees regardless of community context. 94 The Cape Floristic Region (CFR) of South Africa is well-suited for studying the effects of variation in 95 co-flowering species composition on a focal plant species because it has sharp spatial and temporal 96 changes in the composition of flowering communities (Cowling 1992, Simmons andCowling 1996). 97 Our study focuses on buzz-pollinated Cyanella hyacinthoides (Tecophilaeaceae) in the CFR. This 98 species is widespread and thus co-occurs with a variety of other plant species, making it ideal to 99 study the effects of co-flowering species composition on visitation. Cyanella hyacinthoides is a 100 cormous geophyte endemic to the CFR that flowers during Austral spring (August to November) 101 (Manning and Goldblatt 2012). It has light blue flowers with six poricidal anthers that vary in 102 morphology (five smaller upper anthers and one larger lower anther) (Dulberger and Ornduff 1980). 103 Plants from this species can present multiple inflorescences and each inflorescence can produce up 104 to 15 flowers. Individual flowers can remain open for six or seven days if not pollinated (Dulberger 105 and Ornduff 1980), but flowers close within a few hours after pollination has occurred (JEK, personal 106 observation). Self-compatibility varies between populations, with two-thirds of assessed populations 107 exhibiting complete self-incompatibility (Dulberger and Ornduff 1980), indicating that the 108 persistence of this species mostly relies on successful pollinator-mediated reproduction. 109 Here, we contrast how visitation by bees that can successfully manipulate the buzz-pollinated C. Floral visitation observations commenced when bee activity started, that is, after 06h00, depending 164 on the weather. If the temperature exceeded 30°C and bee activity decreased, observations were 165 halted and resumed later in the day. Interactions between bees and flowers were recorded in 20-166 minute intervals, and multiple plant species were observed simultaneously. The number of flowers 167 per species that were observed per interval was noted. In total, interactions were observed for 607 168 intervals, resulting in 202.3 observation hours. Interaction strengths were calculated as the number 169 of visits per flower per 20-minute interval multiplied by 1000 and rounded to create integers. We did 170 this as some of our analyses required integers as input. 171 Our analyses (described below) relied on the interaction frequencies of vibrating bees to all plant 172 species in the community from which they were observed to collect pollen. We thus identified the 173 vibrating bees to genus or species level using the keys by Eardley (1994) and Eardley & Brooks (1989)  174 The plant species that were visited by the vibrating bees were identified to species level using 175 Manning and Goldblatt (2012). Further, some of our analyses required the total number of visits 176 made by all bee species to all pollen-offering plant species in each community, particularly for the 177 calculation of d' and link temperature (see below). For these calculations, only the sum of all the 178 interactions in the community is relevant and the identities of bee and plant species do not matter. 179 We thus did not identify these additional bee and plant species to species level, however, we 180 identified these additional bee species as morphospecies in the field (through capture, behaviour, 181 and photographs), and we identified plant species to genus level, to assist with our observations in 182 the field. The number of bee species visiting a plant species (i.e., ecological pollination specialization -sensu 186 Armbruster (2017)) can be described by multiple metrics, and we calculated three metrics that 187 captured this within each community (Table 1). The first metric is interaction partner richness, which 188 measures the raw number of bee species that visits C. hyacinthoides. For instance, if five species visit 189 C. hyacinthoides, then the interaction partner richness would be 5. The second metric is interaction 190 partner diversity, calculated using Hill numbers of the Shannon diversity index (Jost 2007). This 191 metric calculates the number of bee species that visits C. hyacinthoides and weights it with the 192 interaction frequency of each pollinator species, thus accounting for interaction evenness (Kemp et 193 al. 2019). Thus, if C. hyacinthoides is visited by many insect species, but only a few of these species 194 have high visitation rates to C. hyacinthoides, this metric will indicate that C. hyacinthoides is 195 effectively visited by few pollinator species. For instance, if one pollinator species makes ten visits to 196 C. hyacinthoides and four species each make one visit, the interaction partner diversity would be 197 2.70. This metric thus adjusts for uneven visitation by pollinators. Finally, we calculated the 198 specialization index d' which indicates the specialization of C. hyacinthoides in relation to the 199 availability of bee species in the community (Blüthgen et al. 2006). This metric ranges from 0 to 1, 200 where high values indicate that a plant species is selectively visited by few insect species, and low 201 values indicate that a plant species is either visited by many insect species or by the common insect 202 species in the community. Ultimately, this metric shows how strongly a plant species deviates from a 203 random sampling of the available pollinators. For instance, if a bee species that visits C. 204 hyacinthoides also visits all plant species in the community at similar frequencies, then C. 205 hyacinthoides will have a low d' value because it is not selectively visited by the bee species. 206 However, if a bee species makes many visits to C. hyacinthoides and few visits to other plant species 207 in the community, then C. hyacinthoides will have a high d' value because the bee species is 208 selectively visiting it rather than other available plant species. 209 To assess whether ecological specialization of C. hyacinthoides, as measured by the three metrics 210 Visits were summed across plant species to give the total number of visits made by these bee 233 species within a community, and we used this as proxy for the abundances of vibrating bees. 234 We conducted three models with the specialization metrics (i.e., interaction partner richness, 235 interaction partner diversity, and d') as respective response variables. Vibrating bee abundances and 236 community composition (as represented by PC1 from the PCA above) as calculated above were used 237 as explanatory variables. Specifically, Poisson regression was used to test the influence of vibrating 238 bee abundances (natural log-transformed to improve model fit) and community composition on 239 interaction partner richness. Further, we conducted two linear regressions to assess the influence of Additionally, we recorded flower colour by measuring reflectance spectra indoors at a 45° angle to 253 the petal surface with an OceanOptics USB4000 Spectrometer (Ocean Optics, Dunedin, FL, USA) 254 calibrated with a diffuse reflectance WS-2 white standard. Multiple measurements were taken per 255 species (mean = 14, range = 6-30), and these were averaged to obtain a single spectrogram per 256 species. We modelled these spectra in honeybee vision using Chittka's hexagon model (Chittka 257 1992), assuming a D65 illumination and a standard green background, in the 'pavo' package (Maia et  We recorded visits from 180 insect morphospecies, of which 66 morphospecies were bees. Only 299 three bee species visited C. hyacinthoides all of which used vibrations to extract pollen from this 300 species. Two of these bee species were present in all six communities, and only one of these visited 301 C. hyacinthoides in all six communities. 302 The number of bee species visiting C. hyacinthoides, measured as interaction partner richness and 303 diversity, was not influenced by the availability of other pollen (Table 2) Our stepwise model selection procedure showed that colour group was the only important predictor 313 of link temperature (Table 3) with bee-blue-green reflective petals and avoided those that primarily reflect bee-UV-blue or bee-422 green. In contrast to our results, previous work has shown that some bee species have innate 423 preferences for bee-UV-blue (Giurfa et al. 1995) and for bee-green (Giurfa et al. 1995).  Tables and figures   561  562   Table 1. Summary of the metrics used to assess the variation the pollination interactions of Cyanella 563 hyacinthoides. 564

Metric Definition Citation Interaction partner richness
The number of bee species that a plant species interacts with.

Interaction partner diversity
The number of bee species that a plant species interacts weighted by the interaction frequencies of each bee species. Thus, if a plant species is visited by many bee species in low frequencies and visited by one bee species in high frequencies, the interaction partner diversity metrics will be close to 1, indicating that the plant species is specialized.

d' (interaction specialization)
Specialization of a plant species in relation to the availability of bee species in the community. This metric shows how strongly a plant species deviates from a random sampling of the available pollinators. Ranges from 0 (random, generalist visitation) to 1 (selective, specialized visitation).

Visitation rate
The number of visits by bees to a thousand flowers of a plant species in a 20-minute period. This is calculated as the sum of all visits by bees to a plant species across all observation periods, divided by the sum of flowers observed of that plant species, multiplied by 1000.

Link temperature
Link temperature quantifies whether the observed interaction frequency deviates from the expected interaction frequency based on a model of neutral interaction. Calculated for each plant-bee species pair separately. Ranges from -1 (avoidance) to 1 (preference).