Hold tight or loosen up? Functional consequences of a shift in anther architecture depend substantially on bee body size

A fundamental question in pollination ecology is how pollinators affect the evolution of different floral forms. Yet functional effects of shifts in floral form for plant and pollinator are frequently unclear. For instance, flowers that conceal pollen within tube-like anthers that are spread apart and move freely (free architecture) or are tightly joined together (joined architecture) have evolved independently across diverse plant families and are geographically widespread. Surprisingly, how their bee pollinators affect the function of both architectures remains unknown. We hypothesised that bee body size would affect foraging success and pollination differently for free and joined anther architectures. Therefore, we modified the anther architecture of a single plant species (Solanum elaeagnifolium) and used a single species of generalist bumble bee (Bombus impatiens), which varies greatly in body size. We found that on free anther architecture, larger bees were better pollinators. More pollen on their bodies was available for pollination and they deposited more pollen on stigmas. Conversely, on joined anther architecture, smaller bees were better pollinators. They collected less pollen into their pollen baskets, had more pollen on their bodies available for pollination, and deposited more pollen on stigmas. While we also found modest evidence that plants benefit more from joined versus free anther architecture, further investigation will likely reveal this also depends on pollinator traits. We discuss potential mechanisms by which pollinator size and anther architecture interact and implications for floral evolution.


Introduction
Pollinators are a primary driver of floral trait evolution and traits including flower colours, scents, and morphologies frequently reflect selection from pollinator behaviour and morphology (Darwin 1877;Barrett 2010;Johnson and Anderson 2010;Schiestl and Johnson 2013). For instance, the flowers of many plant species have evolved ultraviolet-absorbing centres that contrast strongly with the rest of the flower and are thought to reflect selection by pollinators with strong preferences for these colour patterns (Silberglied 1979;van der Kooi et al. 2019). Likewise, a variety of flowering plant taxa have evolved specialised flowers with long nectar spurs, reflecting selection by insect pollinators with long proboscises (Hodges et al. 2003;Vlašánková et al. 2017). Accordingly, floral traits evolving via pollinator-mediated selection affect pollination and foraging success. In particular, pollinator-mediated selection on flowering plants favours the evolution of floral traits that enhance the dispersal of pollen to pollinators and maximise the transfer of pollen to conspecific stigmas, while simultaneously minimising pollen wastage (e.g. pollen lost or consumed as food by the pollinator) (Harder and Wilson 1994;De Kock et al. 2018). Yet while the functional effects of floral traits should depend on how plant and pollinator interact (e.g. Fukuda et al. 2001;Hopkins et al. 2014; Hazlehurt and Karubian 2016), consequences for pollination and foraging success are poorly understood (but see Betts et al. 2015;De Kock et al. 2018;Lichtenberg et al. 2018).
Communicated by Jared Gregory Ali.

3
Modifications of anther architecture (i.e. sizes, degree of fusion, and spatial/functional connections) are common and thought to significantly affect flower function (Endress 2012;Nevard et al. 2021). For example, anthers frequently vary in the degree to which they are joined in the flowers of buzz pollinated plants, yet the functional significance of this variation is imperfectly understood (Faegri 1986;Vogel 1978;Glover et al. 2004;Vallejo-Marín et al. 2022). Such plants are pollinated by bees capable of generating powerful vibrations ('floral buzzing'; performed by an estimated 58% of bee species; Cardinal et al. 2018), which expel pollen from the terminal pores of the tube-like poricidal anthers (Macior 1968;Vallejo-Marín 2019;Brito et al. 2020). Many of these buzz pollinated plant species possess flowers whose anthers are spread apart and capable of relatively independent movement ('free anther architecture') (Glover et al. 2004;Vallejo-Marín et al. 2022). Yet many other buzz pollinated species across taxonomically diverse families have independently evolved a joined anther architecture, in which poricidal anthers are arranged and joined in the centre of the flower, resembling a cone ( Fig. 1; found in at least 21 plant families, see supplementary ;Vogel 1978;Faegri 1986;Glover et al. 2004;De Luca and Vallejo-Marín 2013;Russell et al. 2016). Both joined and free anther architectures are widely distributed among buzz pollinated species, and floral buzzing is expected to affect pollen expulsion differently depending on the degree to which the anthers are joined (Glover et al. 2004;Nevard et al. 2021;Vallejo-Marín et al. 2022). Despite this, the functional significance of both anther architectures for flower and bee remain unclear (see Vallejo-Marín et al. 2022).
The degree to which anthers are joined could affect pollination and foraging success in complementary or opposing ways. For instance, joined anther architecture could simultaneously enhance pollination (benefitting the plant) and pollen collection (benefitting the bee). Given that simulated bee vibrations are propagated more effectively and pollen release is increased when anthers are joined (Nevard et al. 2021;Vallejo-Marin et al. 2022), a bee might be able to collect more of the released pollen. Increased pollen release might also translate to more pollen deposited on the stigma. Alternatively, joined or free anther architecture could enhance either pollination or pollen collection, but not both. For instance, increased pollen release from joined anther architecture might enhance collection by the bee, without resulting in more pollen transferred to conspecific stigmas (e.g. Russell et al. 2021). This would be expected, for instance, if the anther cone more consistently deposited pollen in a readily groomed location on the bee. In contrast, loosely held, sprawling anthers of free anther architectures may more readily distribute pollen to so-called safe sites on the bee, which are accessible to plant stigmas, but protected from bee grooming (Herrera 1987;Harder and Barclay 1994;Huang et al. 2015;Koch et al. 2017;Tong and Huang 2017). Similarly, if free anther architecture releases less pollen (Vallejo-Marin et al. 2022), more pollen may remain in the anthers for subsequent pollinators, resulting in more opportunities for pollination. Reduced pollen release might even entice a given pollinator into spending more time on the flower, thereby enhancing pollen transfer to the stigma.
How anther architecture affects pollination and foraging success also likely depends on pollinator characteristics. Pollination effectiveness is considered to be generally influenced by the physical fit between flower morphology and pollinator body, which can, for instance affect the removal of pollen and contact with floral reproductive structures (Herrera 1987;Minnaar et al. 2018;Moreira-Hernandez and Muchhala 2019;Russell et al. 2021). Given that body size frequently varies both within and among bee species (Cariveau et al. 2016;Cullen et al. 2021), physical fit may strongly influence how a given anther architecture affects pollination and foraging success. For instance, relative to larger bees, smaller bees might be less effective pollinators on flowers with free anther morphology, because their smaller bodies would be less likely to contact the stigma as they move among anthers (Li et al. 2015;Solis-Montero and Vallejo-Marin 2017;Mesquita-Neto et al. 2021). Likewise, given that joined anthers vibrate together (Nevard et al. 2021), relatively larger bees might be less effective pollinators on flowers with joined anther morphology, because their more powerful vibrations (De Luca et al. 2013;2019;Switzer et al. 2019) could enable them to collect more pollen or deplete anthers more completely, leaving less pollen to be transferred to conspecific stigmas by subsequent visitors.
In this laboratory study, we assessed how pollination and foraging success in a plant-pollinator mutualism were influenced by anther architecture and pollinator body size. To control for differences among species, we modified the anther architecture of a single plant species (Solanum elaeagnifolium) and used a single species of bumble bee (Bombus impatiens), which varies substantially in body size, even among individuals within a given colony (e.g. by a factor of 3.2 in size and tenfold in body mass ;Harder 1985;Couvillon et al. 2010;Kelemen et al. 2022). We hypothesised that pollination and foraging success would differ between anther architectures, with foraging bees leaving less pollen within joined anther architecture (the experimentally modified condition), and pollen being deposited more precisely on the bee, resulting in more pollen collection by the bee, and in less pollen ultimately transferred to stigmas. We also hypothesised that bee body size would substantially affect these patterns, with smaller bees leaving more pollen, collecting less pollen, and spending more time on joined anther architecture, and transferring less pollen to stigmas of flowers with free anther architecture (the natural condition).

Experimental subjects
To study how anther architecture affected pollen acquisition by bees and deposition on flowers, we used 67 initially flower naïve workers from 3 captive commercially obtained colonies (Koppert Biological Systems, Howell, MI, USA) of the common eastern bumble bee, Bombus impatiens. Briefly, following Russell et al. (2017a), each colony was maintained on 2 molar solution of sucrose and pulverised honeybee-collected pollen (Koppert Biological Systems) from artificial feeders within enclosed foraging arenas (LWH: 82 × 60 × 60 cm) set to a 14 h:10 h light: dark cycle.
To donate and receive pollen, we used Solanum elaeagnifolium flowers (hereafter, 'Solanum'), which across their natural range are commonly pollinated by bumble bees (Knapp et al. 2017). Flowers used in trials were freshly cut from eight plants grown in a greenhouse (seeds originally collected from plants 2.5 km north-northeast of Portal, Arizona) with supplemental halogen lights to extend day length to a 14:10 h light: dark cycle, and were fertilised weekly (PlantTone, NPK 5:3:3, Espoma, Millville, New Jersey, USA).

Experimental protocol
We modified the naturally free-moving anthers of Solanum flowers into two configurations for trials: free and joined ( Fig. 2) (Vallejo-Marín et al. 2022). In the joined configuration, a small amount of polyvinyl acetate glue was applied to the lateral edges of anthers, which were held together tightly with the aid of a plastic straw while the glue dried over 5 min. Flower styles, which were approximately centred among the free-moving anthers (Fig. 2a), were also positioned at the centre of anther cones (Fig. 2b), mimicking the typical condition in plant species with naturally joined anthers (Fig. 1). In the free configuration, we constructed sham flowers by applying glue to the lateral edge of anthers without pressing them together, to control for potential effects of glue on bee behaviour and pollen removal (Fig. 2a). To prevent desiccation, all freshly cut live flowers were placed into custom water tubes (Russell et al. 2017a).
To test how anther architecture affected pollen acquisition and deposition by bees, we divided flower-naïve bees into two treatments that differed in whether only free or only joined flowers were provided (Fig. 2). We systematically alternated assignment of bees to each treatment to control for effects of time and day on behaviour. To initiate a behavioural trial, we mounted four flowers in a 2 × 2 grid on the arena wall in a cleaned test arena ( Figure S14). From Fig. 2 Profiles of the two types of flowers used in treatments. a The natural "free" configuration of anthers and b the modified "joined" configuration of anthers. In both of the configurations, the style is centrally located among the anthers. The background was digitally removed 1 3 the foraging arena, a single flower-naïve female worker bee was gently captured from the nectar feeder using a 40 dram (148 mL) vial (Bioquip Products, Inc.) and released in the test arena. Before releasing a bee, we visually confirmed the absence of pollen on her body. During a single trial, we allowed a bee to forage twice on each of the four donor flowers for a total of eight visits. Each donor flower was removed from the arena after having received the two foraging visits, using 30 cm long 'jumbo' forceps (BioQuip Products, Inc.) while the bee was in flight. Bees did not exhibit signs of being disturbed by our activity, such as aggressive behaviour or attempts to escape from the arena. Removing a flower did not appear to interrupt visits to the other flowers, as removal took approximately 3 s, less than the length of time between visits to a given flower. Twelve bees made fewer or more than 8 visits, divided equally among joined and free treatments: we included these bees in analyses. Since time spent visiting flowers could have influenced pollen acquisition, collection, and deposition, trials were video recorded to measure the total flower handling time per trial; from three legs on the flower until physical contact with the flower ended (via Avidemux 2.7.6). The average time spent on all flowers in a trial was 2.7 min and a typical flower visit lasted 22 s.
Immediately after a trial, the anthers of the four flowers were pooled, as were the styles, and preserved in 70% ethanol for pollen counting, and the bee was killed and stored in a -20 ℃ freezer for later body size measurements and pollen counting. Bees stopped moving within < 1 min of being stored and containers were handled carefully to limit the possibility of pollen being dislodged. Each bee was only used in a single trial. From analyses, we excluded 7 bees that failed to collect pollen into their pollen baskets, leaving N = 28 and 32 bees within free and joined treatments, respectively.

Pollen counting and body size measurements
To test whether pollen acquired by different body parts affected pollen transfer to flowers, we dissected frozen bees into four parts: the head, thorax (including the fore and midlegs), abdomen, and hind legs (encompassing the pollen baskets, i.e. corbiculae). We submerged and vortexed each body part separately with 70% ethanol and condensed head, thorax, and abdomen samples to 100 μL each and hind leg samples to 1000 μL each using a centrifuge. We counted pollen in three 10 μL aliquots using a haemocytometer (Hausser Scientific, Horsham, PA) at 400 × or 100 × (Leica DM 500) to arrive at an estimate for the total volume. Estimated pollen counts were rounded to the nearest whole number. To count pollen grains on flower stigmas, we acetolysed styles from flowers (following Dafni 1992) pooled by trial, condensed samples to 40 μL, and counted pollen in two aliquots; if we counted zero grains, we counted grains in all four aliquots. To test whether anther architecture affected pollen removal by bees, anthers from flowers pooled by trial were acetolysed and pollen counted in 1000 μL samples as above. To estimate the average amount of pollen removed from flowers, we also counted pollen from the anthers of 20 unvisited flowers (pooled in groups of 4 flowers) as a baseline.
Since patterns of pollen removal, acquisition, collection, and deposition might reflect differences in bee body size, we also measured body size of each test bee (head width at the widest position in mm) using a stereoscope and ImageJ (Rasband and ImageJ 2011; National Institutes of Health, Bethesda, MD, http:// imagej. nih. gov/ ij/) following Russell et al. (2017b).

Data analyses
All data were analysed using R v.4.1.0 (R Development Core Team 2021). We checked for overdispersion, zero inflation, and uniformity for all models using the DHARMa package (Hartig 2018) and log transformed all pollen counts to meet model assumptions.

Does anther architecture, flower handling time, and body size affect pollen remaining within the anthers?
To analyse how the amount of pollen bees left within the anthers was influenced by anther architecture, bee body size, and the amount of time bees handled flowers, we used a generalised linear mixed effects model (GLMM) with a Gaussian distribution using the glmmTMB() function in the glmmTMB package (Magnusson et al. 2018), specifying type II Wald Chi-square (χ 2 ) tests via the Anova() function in the car package (Fox 2015). The response variable was 'pollen in anthers' (number of grains), and the explanatory variables were 'anther architecture treatment' (free vs joined), 'flower handling time' (total time spent handling flowers), and 'body size' (head width). We included 'colony ID' as a random factor. We included trial date as a random effect in initial GLMMs, but as it had no effect, we excluded it from subsequent analyses.

Does anther architecture, flower handling time, and body size affect pollen collection and acquisition by the bee?
Since bumble bees collect pollen into their corbiculae to feed their colonies and this pollen is packed wet (with nectar) and thus largely unavailable for pollination ('collection'), while pollen acquired on the rest of the body remains viable and accessible for pollination ('acquisition') (Parker et al. 2015), we considered both reservoirs of pollen. To analyse how the quantity of pollen accessible for pollination or in the pollen baskets was influenced by anther architecture, bee body size, and the amount of time bees handled flowers, we used two GLMMs with explanatory variables and random factor specified as above. The response variables differed for each GLMM and were 'pollen in the pollen baskets' (number of grains collected in the corbiculae) or 'pollen available for pollination' (number of grains acquired on the body, excluding the corbiculae).
In addition, we analysed how the proportion of donor pollen in the pollen baskets versus on the rest of the body was influenced by the same factors, via a GLMM with a binomial distribution. The response variable was a binomial of pollen type (pollen in the corbiculae versus on the rest of the body) weighted by the total pollen count on the body; explanatory variables were as above. For this model only, we included 'bee ID' within 'colony ID' as random factors.

Does anther architecture, flower handling time, body size, and pollen placement on the bee affect pollen receipt by the stigma?
To analyse how the amount of pollen bees deposited on stigmas was influenced by anther architecture, bee body size, bee body part, and handling time, we first ran a maximal GLMM, with 'anther architecture treatment', 'flower handling time', 'body size', and pollen on the head, thorax, and abdomen as explanatory variables. We treated body parts separately, because the frequency with which different body parts contact flower reproductive parts may depend on the physical fit between plant and bee, which could have been altered by anther architecture treatment. We performed backward elimination using the anova() function in R to examine significance relative to the respective maximal model, finding that all body parts contributed similarly. Thus, for the final GLMM, the response variable was 'pollen on the stigma' and the explanatory variables were 'anther architecture treatment', 'flower handling time', 'body size', and 'pollen available for pollination' (number of grains on the body, but not in the corbiculae). We included 'colony ID' as a random factor.

Anther architecture and body size affected pollen receipt by the stigma
Pollen deposition on the stigma by bees was significantly affected by anther architecture, but the effect depended on the quantity of pollen on the body available for pollination ( Fig. 6a; Table 1, S5; GLMM: effect of architecture X body pollen: χ 2 1 = 5.30, P < 0.021). When bees carried relatively less pollen on their bodies, receipt was less for flowers with joined versus free anthers; the opposite pattern was observed when bees carried relatively more pollen. Overall, bees foraging on free anthers deposited 22% more pollen on stigmas than bees on joined anthers (Fig. 6a, Table 1). Regardless of anther architecture, pollen receipt by the stigma depended significantly on bee body size, with pollen receipt decreasing as body size  1 3 increased (Fig. 6a, b; Table 1, S5; GLMM: effect of body size: χ 2 1 = 4.83, P < 0.028). There was no effect of flower handling time on pollen receipt by the stigma (Fig. 6c; Table S5; GLMM: effect of handling time: χ 2 1 = 1.93, P = 0.165).

Discussion
Our study elucidates how the functional significance of a common and widespread shift in anther architecture (Fig. 1) can be profoundly affected by pollinator body size. This is not unexpected, because pollination effectiveness is often thought to depend substantially on the physical fit between Fig. 4 The quantity of pollen in the pollen baskets (corbiculae) or on the body (excluding corbicular pollen) as influenced by anther architecture and a, c bee body size or b, d total time spent handling flowers. N = 28 and 32 bees for free and joined anther treatments, respec-tively. Plotted lines indicate estimated means and shaded regions indicate standard errors. Asterisks indicate significant differences in the mean pollen in the corbiculae or available for pollination between treatments at P < 0.05 flower and pollinator (Herrera 1987;Minnaar et al. 2018; Moreira-Hernandez and Muchhala 2019). We found that even a modest difference in bee body size (varying by a factor of 1.3) greatly affected benefits to the plant in terms of pollen transfer to the stigma and pollen on the bee body available for pollination (i.e. not held within the pollen baskets), as well as benefits to the bee in terms of pollen collected into the pollen baskets. Consistent with our initial hypothesis, relatively smaller bees were better pollinators for flowers with joined anther architecture, while relatively larger bees were better pollinators for flowers with free anther architecture. Specifically, smaller bees deposited more pollen on stigmas overall, and when foraging on flowers with joined anther architecture, absolutely and proportionally (relative to pollen in the pollen baskets) more pollen was available for pollination on smaller bee bodies. Conversely, larger bees foraging from free anther architecture acquired absolutely and proportionally more pollen available for pollination than smaller bees. Finally, smaller bees collected less pollen into their pollen baskets when foraging from joined, but not free anther architecture. Altogether, our results suggest that to understand the functional effects of floral trait evolution, pollinator characteristics should be taken into consideration.
Multiple mechanisms could account for how bee body size affects the movement of pollen. One possibility is that pollen removal from the anthers by bees, and thus the quantity of pollen that can be transferred to conspecific stigmas or collected by the bee, is a function of bee body size. Indeed, body size can be associated with differences in foraging behaviour (e.g. Jauker et al. 2016;Stout 1999;Russell et al. 2021), including characteristics of floral buzzing predicted to affect pollen removal (e.g. Corbet et al. 2014;De Luca et al. 2013, 2019Switzer et al. 2019). Yet bee body size did not affect pollen removal from the anthers in our study. Instead, the quantity of pollen deposited on the bee (quantified after bee grooming) was affected by body size, with bigger bees overall carrying and collecting more pollen. This result suggests that body size likely affects how much pollen can be intercepted by the bee (affecting pollen wastage; see Vallejo-Marín et al. 2022). However, despite carrying more pollen, larger bees transferred less pollen to flower stigmas, indicating that smaller bees in our study may have been more size matched to flower reproductive anatomy (see Solis-Montero and Vallejo-Marin 2017). Future studies that exploit greater variation in bee body size-and in particular, smaller bees-will be required to determine whether size matching between bee and flower can explain patterns of pollen transfer better than bee body size alone (see also Willmer and Finlayson 2014;Konzmann et al. 2020).
Transfer of pollen to conspecific stigmas is thought to frequently depend on spatial segregation of pollen on the pollinator (e.g. Muchhala and Thomson 2012;Tong and Huang 2017;Minnaar et al. 2018;Russell et al. 2021). While joined anther architecture has been proposed to result in more precise pollen placement on the bee relative to free anther architecture (e.g. Harder and Barclay 1994;Vallejo-Marín et al. 2022), our results were not consistent with these expectations. Pollen was spatially segregated on bee bodies, with more than twice as much pollen available for pollination present on the thorax, versus on the head and abdomen combined (Table 1). However, the proportion of pollen acquired among body parts did not differ among bees foraging on one versus the other anther architecture.
In addition, no particular bee body part contributed disproportionately to pollen transfer to conspecific stigmas for either anther architecture. One possible explanation for this latter pattern is that bees had insufficient time between flower visits to groom pollen more completely. Poorly groomed parts of pollinator bodies (safe sites; e.g. the midline of the dorsal head and thorax and ventral and dorsal abdomen) are often key to pollination (Herrera 1987 ;   Fig. 6 The quantity of pollen transferred to flower stigmas as influenced by anther architecture and a pollen on the body available for pollination (excluding corbicular pollen), b bee body size, and c total time spent handling flowers. N = 28 and 32 bees for free and joined anther treatments, respectively. Plotted lines indicate estimated means and shaded regions indicate standard errors. Asterisks indicate significant differences in the mean pollen transferred to recipient stigmas between treatments at P < 0.05 Buchmann et al. 1990;Huang et al. 2015;Koch et al. 2017;Tong and Huang 2017), but excess and easily groomed pollen might reduce the influence of spatially segregated safe sites on patterns of pollen movement. Perhaps consistent with this, in this study the quantity of pollen on the bee body available for pollination relative to the quantity of pollen transferred to stigmas was large (nearly 9 times greater). Modification of anther architecture could also have had other effects on pollen transfer that we were unable to quantify. For example, joined anther architecture strongly reduces the spatial separation between the anthers and the stigma (herkogamy), and reductions in herkogamy are predicted to increase self-pollination (Webb and Lloyd 1986;Opedal 2018). Future research could, therefore, examine whether shifts in anther architecture alter the likelihood of outcrossing.
Prior studies have predicted that joined anther architecture evolved to more effectively release pollen and reward bees, and experimental work with simulated bee buzzes has found that joined anther architecture increases pollen release (Glover et al. 2004;Vallejo-Marín et al. 2022). In contrast, we find that joined anther architecture overall decreases both pollen release from the anthers (by 34%) and pollen collection by the bee into its pollen baskets (by 23%). Bee behaviour likely at least partially accounts for the discrepancy with earlier work: bees buzzed and released pollen from multiple anthers on visits to flowers with free anther architecture, whereas simulated buzzes were applied to only a single anther. Other possible explanations include that joined anther architecture caused bees to buzz with reduced effectiveness or in suboptimal locations on the anther cone (see De Luca and Vallejo-Marín 2013; Jankauski et al. 2022). In addition, although pollen transfer to the stigma is also decreased (by 18%) for flowers with joined vs free anther architecture, we tentatively suggest that the proportionally greater retention of pollen by joined anthers more than offsets the reduction in pollen transfer, by resulting in more pollen being available for subsequent pollinators to transfer to conspecific stigmas. However, this prediction will require experimental validation and, as above, the relative benefits of joined anther architecture to the plant likely depend on the relative size of its bee pollinators. Finally, while we can only estimate the proportion of wasted pollen (pollen released by the anthers, but neither collected by the bee nor deposited on stigmas), on average, free anther architecture resulted in nearly 3 times as much pollen wasted as joined anther architecture (18.0% versus 6.7% wasted, respectively; Table 1).
In conclusion, assuming our results are broadly representative, even modest differences in bee body size have the potential to drive selection for different anther architectures. Given that interspecific differences in bee body size are often substantial within a given pollinator community (e.g. Solis-Montero and Vallejo-Marin 2017; Mesquita-Neto et al. 2021;Cullen et al. 2021), selection for different anther architectures may be common. Within the context of commonly observed historical and land use-associated reductions in bee body size (e.g. Grab et al. 2019;Nooten and Rehan 2020), our results suggest such changes could have consequences for pollination effectiveness. Accordingly, patterns in the distribution and frequency of plant species with joined and free anther architecture, which offer pollen as a reward, likely reflect the degree to which physical characteristics of the bee community vary, such as occurs in the context of floral traits associated with nectar rewards (e.g. Harder 1985;Kaiser-Bunbury et al. 2014;Klumpers et al. 2019;Sponsler et al. 2022). Furthermore, given that the degree to which anthers may be joined can be less discrete than in our study (see Faegri 1986;Russell et al. 2016;Vallejo-Marín et al. 2022), effects on pollination and foraging success may be variable. Finally, while the present study focuses on how pollinator body size influences the movement of pollen, behaviour and especially learning are key to affecting how pollinators interact with different flower morphologies (e.g. Laverty 1994;Papaj et al. 2017;Russell et al. 2021). A more complete understanding of the functional effects of shifts in anther architecture will, therefore, require addressing the influence of both pollinator morphology and behaviour simultaneously.