Research

Pollen aggregation by viscin threads in varies with pollinator

Yun-Peng Song1*, Zhi-Huan Huang2* and Shuang-Quan Huang1 1Institute of Evolution and Ecology, School of Life Sciences, Central Normal University, Wuhan 430079, China; 2Guangxi Institute of Botany, Guangxi Zhuang Autonomous Region and Chinese Academy of Sciences, Guilin 541006, China

Summary Author for correspondence: Pollen grains can be dispersed singly or variously aggregated in groups. Whether the evolu- Shuang-Quan Huang tion of pollen aggregation is driven by the pollinator remains unexplored. We hypothesize that Tel: +86 27 67867221 an extensive pollen aggregation is favored under a scarcity of pollinators. Email: [email protected] Variation in pollen aggregation by viscin threads in 13 Rhododendron was mea- Received: 13 May 2018 sured as it is related to pollen removal in a visit. Visitation rates of functional pollinator groups Accepted: 13 July 2018 that vary in their grooming behavior were investigated in each species. Pollen deposited on stigmas in the field was also sampled. New Phytologist (2019) 221: 1150–1159 Seven Rhododendron species were infrequently pollinated by low-intensity grooming ani- doi: 10.1111/nph.15391 mals, including birds, butterflies and moths. The other six species were more frequently polli- nated by bees with a high intensity of pollen grooming. Bird- and Lepidoptera-pollinated Key words: floral traits, pollen aggregation, species produced longer pollen-connecting threads that connected more pollen grains. Phylo- pollen dosing strategy, pollen grooming genetically independent contrast analysis of the 13 species showed that pollinator visitation behavior, pollen-thread tangles, pollinator frequency was negatively related to amounts of pollen removal per visit but not to stigmatic functional groups, Rhododendron. pollen loads. The finding of interspecific patterns in pollen removal related to pollinator visitation fre- quency suggests pollinator-mediated selection on pollen packaging strategies, supporting the hypothesis of floral evolution via pollen export.

of their bodies while moving between flowers (in-flight groom- Introduction ing; Harder, 1998; Tong & Huang, 2018). By contrast, birds Pollinators and often confront a conflict of interest con- and Lepidoptera usually visit flowers to forage for nectar and do cerning the fate of pollen. From the perspective of pollinators, not groom pollen so regularly, lacking comb-like structures such selection for enhancing foraging efficiency would be favored, as bees have evolved (Harder, 1998). Pollen grains on beaks or whereas for plants, selection for reducing pollen loss would be feathers are likely to be available for deposition during subse- favored. Models show that animal-pollinated plants benefit from quent visits to receptive flowers, with little pollen loss during presenting pollen in small doses to a large number of pollinators transfer (Muchhala & Thomson, 2010). The optimal pollen pre- as a means of alleviating diminishing returns on male investment sentation strategy is expected to be governed by the frequency of (Harder & Thomson, 1989). The amount of pollen placed on a pollinator visits (Harder & Thomson, 1989), and the capacities pollinator does not generally correlate linearly with the number of those pollinators to remove and transfer pollen (Harder & of grains deposited on stigmas. Compared with small pollen Wilson, 1994; Thomson et al., 2000; Castellanos et al., 2006; doses placed on pollinators, large pollen doses export proportion- Muchhala & Thomson, 2010). ately smaller quantities of pollen to stigmas, i.e. diminishing The modes of anther dehiscence and pollen packaging among returns on dose size investment (Harder & Thomson, 1989). anthers greatly influence patterns of pollen presentation (Harder Castellanos et al. (2003) compared pollen transfer efficiency & Thomson, 1989; Thomson et al., 2000). Pollen grains of most between bee- and bird-pollinated Penstemon species, and argued species are monads, i.e. single mature pollen grains, but different that different modes of pollen presentation between these groups forms of pollen aggregation, involving tetrads, polyads, pollen were favored by very different intensities of grooming behavior threads and pollinia, have independently evolved among flower- between the two functional pollinator groups, leading to diver- ing plants (Harder & Johnson, 2008). Pollen grains in orchids gent pollen presenting strategies (Castellanos et al., 2006). Bees are grouped into cohesive masses. Such a pollen presentation frequently groom pollen grains off the stigma-contacting surfaces strategy has been considered to rescue the last chance for pollen to be dispersed given that orchids often experience pollinator *Y-P.S. and Z-H.H. contributed equally to this work. scarcity, facilitating the removal of large numbers of pollen grains

1150 New Phytologist (2019) 221: 1150–1159 Ó 2018 The Authors www.newphytologist.com New Phytologist Ó 2018 New Phytologist Trust New Phytologist Research 1151 in a single pollinator visit. Pollen grains are united by viscin variable size from an individual anther. Based on floral traits, it threads in only two families, and Onagraceae (Cruden has been proposed that diverse Rhododendron species may be & Jensen, 1979; Kress, 1981; Knox & McConchie, 1986; Hesse pollinated by bats (in the tropics), birds, Lepidoptera or et al., 2000). These sporopollenin-containing threads that are bees (Stevens, 1976). Early observations of pollinators long, thin, flexible ropes on pollen surface, connecting single pol- indicated that bees (particularly bumblebees) were effective len grains or pollen tetrads (Hesse, 1984) allow aggregation of pollinators of Rhododendron aureum (Kudo et al., 2011), clusters of pollen grains of varying sizes. However, pollen removal Rhododendron cyanocarpum and Rhododendron delavayi (Ma et al., or receipt has not been measured in species with pollen grains 2010), Rhododendron eriocarpum and Rhododendron indicum united by viscin threads (Harder & Johnson, 2008), in which (Tagane et al., 2008), and Rhododendron ponticum (Stout et al., reproductive strategies remain unexplored empirically. 2006; Stout, 2007), but recent investigations involving pollinator Large masses of pollen attached to pollinator bodies are benefi- exclusion experiments showed that birds and butterflies con- cial for plants only if the pollinators do not collect pollen as a tributed more to pollination in larger-flowered species (Epps reward; otherwise, passive loss of pollen increases with large pol- et al., 2015; Huang et al., 2017). len loads on pollinators. For example, the most evolutionarily We sampled 13 Rhododendron species (Table 1) at Cangshan derived forms of pollen aggregation, i.e. pollinia in Orchidaceae Mountain in the East Himalayas, Province, southwestern and Apocynaceae (Harder & Johnson, 2008), are rarely groomed China (25°400N, 100°060E) during the flowering period from off during a pollinator visit, but remain on the body to contact late March to early June 2012, 2013, 2016 and 2017. stigmas. If pollen grains are intensively groomed by pollinators, small pollen loads would be favored, limiting removal loss Pollinator species and visitation frequency (Harder & Wilson, 1994). However, if pollinator visitation rates are low, small loads involve a high risk of removal failure (Harder To compare pollinator species and visitation frequency in the 13 & Routley, 2006). In Rhododendron (Ericaceae), anecdotal obser- species, we observed floral visitors to at least three flowering indi- vations show that large-flowered species usually have long viscin viduals for > 10 h in each species. We made at least 20 9 30-min threads and are pollinated by birds and Lepidoptera, which rarely surveys, all between 08:00 h and 18:00 h local time during 2–3d initiate active grooming during foraging bouts, while small- of fine weather. Different kinds of visitors to tagged inflores- flowered species have short viscin threads and are pollinated by cences and numbers of flowers on these inflorescences were bees (Huang et al., 2017). recorded (Huang et al., 2017). Visitors that contacted stigmas To test the hypothesis of pollinator-mediated selection on flo- and anthers and had pollen attached to their bodies were ral evolution, empirical study of pollen removal as a measure of regarded as potential pollinators. Diurnal visitation rates were reproductive success is needed. Although a shift of func- recorded as visits per flower per 0.5 h (Fang et al., 2012). Noctur- tional pollinator groups has been considered as an important nal visitation was difficult to observe, so we used an indirect driver for evolution of floral traits among related species method to assess moth visitation to estimate nocturnal visitation (Schemske & Bradshaw, 1999; Fenster et al., 2004; Whittall & by observing moth scale deposition on stigmas under a micro- Hodges, 2007; Harder & Johnson, 2009), very few comparative scope (Singer & Cocucci, 1997; Alonso-Pedano & Ortega-Baes, studies have attempted to estimate pollen removal, leaving the 2012). At least 30 flowers approaching senescence from 10 plants hypothesis poorly tested empirically. To explore whether degrees per species were randomly collected during two fine days. We of pollen aggregation are driven by pollinator selection, we com- pared pollinator functional groups (with different intensities of pollen grooming) and their visitation rates among 13 Table 1 Comparisons of four pollination-related variables (mean SE) between the 13 Rhododendron species pollinated by two functional Rhododendron species. Pollen grains are connected via viscin groups, rarely- or non-grooming (bird and Lepidoptera) and frequently- threads in these species, facilitating an estimate of pollen removal grooming (bee) pollinators under GLMs per visit. A phylogeny constructed for these species and field observations of pollinator visits from the Cangshan Mountains Pollination- Birds and v2 over several years permit us to perform interspecific comparisons related variables Lepidoptera Bees only P with respect to variation in pollen aggregation and pollinator Visitation 0.36 0.04 0.78 0.06 37.61 < 0.001 interactions. frequency, visits per flower per half hour Materials and Methods PTT length, mm 10.77 0.37 3.72 0.21 250.29 < 0.001 Pollen grains per 3723.8 218.2 1168.3 87.4 122.1 < 0.001 Rhododendron species have showy flowers with colored dots or PTT patches (nectar guides) on the upper petals. Rhododendron flowers Stigmatic pollen 2436.4 196.8 1623.1 102.5 21.4 < 0.001 have poricidal anthers, each with two large, round apical pores. loads Pollen grains, which are typically fused into tetrads, are further Estimated visits to 0.7 0.0 1.8 0.1 37.0 < 0.001 connected by viscin threads (Hesse, 1984). We refer to ‘pollen- transfer pollen onto stigmas thread tangles’ (PTT), comprising numerous pollen grains and their entangled viscin threads, which are removed as a unit of PTT, pollen-thread tangles.

Ó 2018 The Authors New Phytologist (2019) 221: 1150–1159 New Phytologist Ó 2018 New Phytologist Trust www.newphytologist.com New 1152 Research Phytologist

considered each stigma with moth scales to represent a single (IBM, Armonk, NY, USA) and ORIGIN 8.0 (Originlab, Washing- moth visit. This calculation of frequency of scale deposition may ton, DC, USA), respectively. underestimate moth visitation frequency. To examine the relationship between visitation frequency and variation in pollen aggregation among 13 species, we calculated Pearson correlations between mean visitation rates (data were log Size variation in pollen-thread tangles e transformed before analysis) and pollen grain number per PTT, To estimate amounts of pollen removal per visit in the 13 species, and stigmatic pollen loads. To further estimate pollinator roles in we imitated pollinator contact with freshly opened anthers by a the evolution of pollen aggregation, we conducted a phylogenetic single visit to pick up the PTTs in which pollen grains were comparative analysis of the 13 Rhododendron species. We selected counted later. In each species, we randomly collected one stamen four DNA markers: ITS, matK, rbcL and PsbA-trnH, and used (a relatively long stamen if their length is obviously different) that Cassiope pectinata, Cassiope selaginoides, Enkianthus deflexus and was just releasing the PTT from one flower on each of 15 plants. Enkianthus perulatus as outgroups. We downloaded DNA A small piece of double-sided sticky tape on the lower surface of sequences of these four markers of the 17 species from NCBI a clean slide was held above a stamen so that the anther tip con- (https://www.ncbi.nlm.nih.gov/). GenBank accessions are listed tacted the tape, and the filament was pulled gently away allowing in Table S2. Four DNA markers were separately aligned in the PTT to emerge slowly from the anther pore (Supporting MAFFT 7.0 (Katoh & Toh, 2010) using the default options and Information Fig. S1). Then we inverted the slide to let the PTT gaps, and ambiguous regions of the alignments were adjusted lie passively flat on the slide, under little tension. The length of manually with GENEIOUS 11.0 (Biomatters, Auckland, New the PTT was measured with a digital caliper to the nearest 0.01 Zealand). Matrixes of four DNA regions were concatenated using mm. To estimate pollen grain number per PTT, we examined SEQUENCEMATRIX v.1.8 (Vaidya et al., 2011). three segments of a thread, each 0.5 mm long. We counted the The Bayesian inference (BI) method was used for phylogenetic pollen tetrads in each segment under a microscope at 40 9 or construction by MRBAYES v.3.2.6 (Ronquist & Huelsenbeck, 100 9 magnification. The mean number of pollen grains 2003). The concatenated dataset was partitioned by region. We attached to the PTT was calculated as four times the number of used JMODELTEST v.2.0 (Darriba et al. 2012) to calculate the best- pollen tetrads based on the average value of the three segments, fitting DNA substitution model for each region using the by taking the full length into account from 15 anthers in each Bayesian information criterion (BIC). The best-fit model of each species. DNA region is listed in Table S3. Markov Chain Monte Carlo To estimate how many pollinator visits per flower are needed (MCMC) analyses were run in MRBAYES with 10 million genera- to transfer pollen onto the stigma, we counted pollen grains tions with each run comprising four incrementally heated chains. deposited on the stigmas of open-pollinated flowers from wild The MCMC started from a random tree and sampled every 1000 plants in each species. At least 30 stigmas from two or three late- generations. The first 25% of the generations were discarded as phase flowers from 15 plants of each species were randomly col- burn-in. The remaining generations were used to generate a lected and stored in FAA solution. Stigmas were softened in 8 M majority-rule consensus tree. Both BI and JMODELTEST analyses NaOH for c. 10 h in the lab. Before observation under a light were performed at the CIPRES Science Gateway (http:// microscope, each stigma was squashed on a slide with a cover slip. www.phylo.org). Pollen tetrads per stigma were counted. The minimum number Phylogenetically independent contrast (PIC) analyses were of visits required to transfer pollen in each species was calculated implemented in MESQUITE v.2.75 (Maddison & Maddison, as the mean stigmatic pollen load divided by the mean number 2010) with the phenotypic diversity analysis program (PDAP) of pollen grains per PTT. package (Midford et al., 2005). The outgroups from the BI tree were pruned before analyses. We calculated Felsenstein’s con- trasts correlation between each possible pair of traits on the tree, Data analysis including pollen number per PTT, stigmatic pollen loads and All data are summarized as means standard errors (SE). We mean visit rate of each species (Felsenstein, 1985). used generalized linear models (GLMs) to compare: (1) plant species visited at least partially by low-grooming birds and Lepi- Results doptera vs plant species visited only by high-grooming bees; (2) then within the low-groomers, a second orthogonal comparison Pollination mechanism of the species visited by birds vs the species visited by Lepi- doptera. Response variables were visitation frequency (normal In all 13 studied Rhododendron species, pollen tetrads within an distribution and identity link function), length of the PTT (nor- anther were connected by viscin threads. As flowers opened in the mal distribution and identity link function), number of pollen field, clusters of pollen grains connected by viscin threads were grains per PTT (Poisson distribution and log link function), usually presented at the apical anther pores, ready for pickup by number of stigmatic pollen grains (Poisson distribution and log animals (Fig. 1). Clusters of PTT with pollen tetrads were link function) and the estimated number of visits required to deposited on birds’ beaks, heads and/or throats when they probed transfer pollen onto stigmas (normal distribution and log link the corolla tubes to drink nectar, or on the wings and/or function). All analyses and graphs were performed in SPSS 19.0 abdomen of Lepidoptera such as long-tongued butterflies or

New Phytologist (2019) 221: 1150–1159 Ó 2018 The Authors www.newphytologist.com New Phytologist Ó 2018 New Phytologist Trust New Phytologist Research 1153

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

Fig. 1 Flowers, floral visitors and pollen-thread tangles (PTT) released from anther pores under a microscope in Rhododendron species: R. delavayi (a, d, g), R. decorum (e, h) and R. trichocladum (f, i). (a) A sunbird collecting nectar from red-flowered R. delavayi. (b) A bumblebee Bombus funeraris collecting nectar from white-flowered R. edgeworthii where the bee was too small to contact the stigma. (c) A butterfly Araschnia levana (Nymphalidae) visiting R. yunnanense; note that the anthers contact the wings. (g) A fragment of PTT from R. delavayi (d); (h) shows a single PTT being pulled out of an anther of R. decorum with many tetrads adhering to it (e); (i) illustrates the PTT status in a newly opened flower of R. trichocladum (f). Arrows, viscin threads connecting pollen tetrads. Bars, 100 lm. hawkmoths (Fig. 1a,c). The PTT with pollen tetrads were pas- on the bees’ ventral surfaces, they groomed the clusters and sively placed on the bird and Lepidoptera bodies, and were not packed them into corbiculae. The degree of adhesion to pollina- seen to be removed to other body sites; these visitors did not reg- tor bodies did not prevent the clusters from being transferred to ularly groom Rhododendron pollen in our field observations. The stigmas when direct contact occurred, although grains packed Red-whiskered Bulbul (Pycnonotus jocosus monticola, Pycnonoti- into corbicular pellets were unavailable to stigmas. dae) was observed to clean its beak, discarding PTT with pollen tetrads of R. cyanocarpum. Bees (most species were bumblebees) Functional pollinator groups and visitation frequency often collected nectar from Rhododendron flowers, but they some- times deliberately collected pollen, and facilitated pollen release Birds were observed drinking nectar from four species by buzzing the anthers. As multiple groups of PTT accumulated (R. delavayi, R. cyanocarpum, Rhododendron neriiflorum and

Ó 2018 The Authors New Phytologist (2019) 221: 1150–1159 New Phytologist Ó 2018 New Phytologist Trust www.newphytologist.com New 1154 Research Phytologist

Rhododendron lacteum) including two sunbird species, Aethopyga We recognized three functional pollinator groups: birds, gouldia and Aethopyga ignicauda (Nectariniidae), and other Lepidoptera (butterflies and moths) and bees. We categorized passerine species, Phylloscopus affinis (Phylloscopidae), P. jocosus the 13 plant species into three groups depending on whether monticola (Pycnonotidae) and Heterophasia melanoleuca they were visited by birds and bees, Lepidoptera and bees, or (Leiothrichidae). On Cangshan Mountain, we observed bees only. The three-pollinator classification can be agglomer- Rhododendron yunnanense visited only by butterflies. In two ated into two groups: those that attracted only bees formed species, moth scales were deposited on stigmas: the frequency the ‘high-grooming’ group; and those that attracted Lepi- and number of scales were 0.3, 1.0 0.4 for Rhododen- doptera or birds in addition to bees formed the ‘low- dron decorum and 0.5, 6.8 2.0 for Rhododendron edgeworthii, grooming’ group. The three pollinator groups had different respectively. For the remaining six Rhododendron species, only visitation frequency (Wald v2 = 42.40, df = 2, P < 0.001) bees were observed to visit the flowers (Table S1; Fig. 2a). Bees across the 13 plant species (Fig. 2a). The visitation frequency were observed visiting flowers in all studied species. Five bumble- of birds (0.28 0.03 visits per flower per half hour) was bee species and a honeybee species were the six most frequent bee lower (P = 0.038) than that of Lepidoptera (0.47 0.07), visitors (Table S1). We observed that bees were too small to con- and the visitation frequency of bees (0.78 0.06) was the tact the stigma while collecting nectar in Rhododendron species highest (Table 1; both P < 0.001, sequential Bonferroni dif- with relatively large flowers (Fig. 1b). ference; Fig. 2a). Therefore, the 13 Rhododendron species were

(a) (c)

(b) (d)

Fig. 2 Variation in length of pollen-thread tangles (PTT) (a), pollen grain number per PTT (b), pollinator visits per flower per half hour (c), and minimum visits to transfer pollen on stigmas (d) among 13 Rhododendron species of three groups categorized as bird- (squares), Lepidoptera- (butterfly and/or moth, triangles) and bee-pollinated (circles). Values are means SE. See Fig. 3 or Supporting Information Table S1 for complete species names corresponding to the three-letter codes.

New Phytologist (2019) 221: 1150–1159 Ó 2018 The Authors www.newphytologist.com New Phytologist Ó 2018 New Phytologist Trust New Phytologist Research 1155

Fig. 3 Phylogenetic relationship of 13 Rhododendron species, with Cassiope pectinata, C. selaginoides, Enkianthus deflexus and E. perulatus as outgroups. Numbers beside branches indicate Bayesian inference (BI) posterior probability (percent) support. Mean values are shown at the tips as bar charts: (a) visitation frequency of functional pollinator group; (b) pollen removal per visit (pollen grains per pollen- thread tangles (PTT)); and (c) stigmatic pollen loads. Bar length indicates relative (a) (b) (c) mean value. Four species names in red, three in blue and six in black are correspondingly pollinated by birds, Lepidoptera and bees. categorized into three types of flowers: bird flowers (four Stigmatic pollen loads on open-pollinated flowers were species) in which birds acted as pollinators, Lepidoptera flow- significantly higher in the low-grooming group than in the ers (three species) in which butterflies and/or moths acted as high-grooming group (Table 1). The numbers of pollen grains pollinators, and bee flowers (six species) in which only bee deposited on stigmas were significantly higher (P < 0.001) in visitors were observed during the field studies. Lepidoptera flowers (3263.8 364.6) than in bird (1665.8 133.2) flowers. The estimated minimum numbers of visits by pollinators required to transfer pollen was significantly higher in Variation in pollen-thread tangles associated with pollinator bee than in bird or Lepidoptera flowers (Fig. 2d; Table 1), sug- groups gesting that bee flowers required more pollinator visits than the In the 13 Rhododendron species, all pollen tetrads were connected other types of flowers, which experienced lower pollinator visita- by tangled elastic threads as in R. delavayi (Fig. 1e,h) for which tion rates. one PTT could be pulled out through each of two pores in the Pollinator visitation rates were negatively correlated with the anthers. PTT length and pollen grains per PTT were significantly numbers of pollen grains per PTT (Pearson correlation r = 0.699, higher in the ‘low-grooming’ (bird and Lepidoptera) than in the P = 0.009) but not with stigmatic pollen loads (P = 0.088) ‘high-grooming’ group (bee pollinators; Table 1). For example, among the 13 Rhododendron species. The relationships do not PTT units in bird-pollinated R. delavayi (16.89 0.50 mm) and change under the PIC analysis (Fig. 3; Table 2), in which Felsen- Lepidoptera-pollinated R. decorum (13.98 0.69 mm) were over stein’s contrasts of pollinator visitation rates and pollen grains per 50 times longer than those in bee-pollinated R. racemosum (0.27 PTT were significantly related among species (r = 0.610, P = 0.02 mm); correspondingly in the three species one unit of PTT contained 4045.8 278.1, 7873.9 312.2 and 218.5 23.4 pollen grains (Fig. 2b). Table 2 Pearson’s correlation (lower left) and P-values (upper right) Within the low-grooming pollinator group, Lepidoptera flow- between Felsenstein’s contrasts of pollinator frequency, pollen removal per ers had shorter (9.08 0.57 mm) PTT than bird flowers (12.04 visit and stigmatic pollen loads on the phylogenetic tree involving 13 0.44 mm, P < 0.001), but they had approximately similar Rhododendron species numbers of pollen grains per PTT (4080.2 430.9 vs 3414.5 = Visitation rates of Pollen Stigmatic 155.3, P 0.07; Fig. 2b). Pearson correlation analysis among the functional pollinators removal pollen loads 13 Rhododendron species indicated that PTT length and pollen grains per PTT were strongly correlated (r = 0.835, P < 0.001). Visitation rates of 0.027 0.242 These results indicate that among the 13 Rhododendron species functional pollinators larger doses of pollen removed per visit (PTT) would be favored Pollen removal 0.610 < 0.001 Stigmatic pollen loads 0.349 0.853 in bird and Lepidoptera flowers compared with bee flowers (Table 1). P-values < 0.05 are in bold.

Ó 2018 The Authors New Phytologist (2019) 221: 1150–1159 New Phytologist Ó 2018 New Phytologist Trust www.newphytologist.com New 1156 Research Phytologist

0.027), but pollinator visitation rates and stigmatic pollen loads pollinator visit frequency and stigmatic pollen loads of open- were not correlated (P = 0.242; Fig. 4). Sizes of pollen doses pollinated flowers could be due to species presenting small pollen (PTTs, pollen removal per visit) were strongly correlated with doses (PTTs) being compensated by more frequent visits. stigmatic pollen loads (Table 2), indicating that large doses rarely Documentation of the fate of pollen is essential for the estima- involved pollen removal loss in these species. tion of male fitness in plants. Pollen collected for food or con- sumed by pollinators accounts for most pollen loss during pollen transfer (Harder & Routley, 2006), diminishing plant reproduc- Discussion tive success. Bees are often considered to be an important group Our measurements of PTT among 13 Rhododendron species indi- of pollinators for diverse flowering plants, but they may function cated large variation in sizes of PTT, ranging from only 220 pol- as a group of ‘ugly pollen collectors’ (sensu Thomson & Thom- len grains per dose to over 7800. The corresponding amounts of son, 1992) because they may achieve high pollen removal from pollen removal were observed to be negatively related to pollina- anthers but low pollen deposition on stigmas in particular plant tor visitation rates. The relationship was confirmed by the phylo- genetic comparative analyses (Fig. 4; Table 2). This is consistent with the pollen presentation theory that predicts that species vis- ited by infrequent pollinators with low intensities of grooming exhibit stronger pollen aggregation than bee-pollinated species. Intense grooming by frequent visitors, here corbiculate bees, is likely to cause diminishing returns of male fitness on pollen pro- duction, thereby selecting for more doses and lower degrees of pollen aggregation (Harder & Thomson, 1989). Floral evolution driven by shifts of functional pollinator groups from bees to hummingbirds has been suggested in several genera of diverse lineages in flowering plants, such as Aquilegia (Whittall & Hodges, 2007), Iochroma (Smith et al., 2008), Mimulus (Schemske & Bradshaw, 1999) and Penstemon (Wilson et al., 2007). These interspecific comparisons based on established phylogeny could illustrate an evolutionary trajectory in floral traits if the role of those traits in plant reproductive success was examined. Measuring pollen removalistheessentialfirststepinassessingpollenexportand outcross paternity to estimate male fitness (Harder & Routley, 2006). Although an estimate of male fitness might facilitate a better understanding of floral evolution (Harder & Wilson, 1994; Much- hala et al., 2010), empirical studies measuring effects of floral traits on pollen removal are not easy in many plants in which pollen grains are separate. We found that interspecific patterns in the amount of pollen removed as a unit from anthers related to pollina- tor functional groups among 13 Rhododendron species (Fig. 3). An investigation of pollinator visitation rates in 11 species of Costus in neotropical forests showed that bee visits were relatively more fre- quent (3.2 visits per flower per half hour) than hummingbird visits (0.5) on each plant species (Kay & Schemske, 2010). When polli- nator visits are frequent, selection may favor smaller clusters of PTT, which may be less likely to trigger grooming behavior and more likely to evade grooming if the behavior does occur. Consis- tent with these predictions, our interspecific comparisons showed that pollinator visits were less frequent in the seven species with the ‘low-grooming’ (bird and Lepidoptera) pollinators with longer PTT and more pollen grains per PTT than the six species that were prin- cipally pollinated by bees, the ‘high-grooming’ group. One may expect that pollen deposited on stigmas in bee flowers would require more pollinator visits than that in infrequently visited flow- ers. In support of this prediction, visitation frequency from field Fig. 4 Felsenstein’s contrasts of visitation frequency of functional observations (Fig. 2c) and the estimated minimum numbers of visits pollinators and pollen removal per visit (a) and stigmatic pollen loads (b) by pollinators were significantly higher in bee than in bird or Lepi- based on the phylogenetically independent contrast (PIC) analysis among doptera flowers (Fig. 2d). The absence of a relationship between 13 Rhododendron species.

New Phytologist (2019) 221: 1150–1159 Ó 2018 The Authors www.newphytologist.com New Phytologist Ó 2018 New Phytologist Trust New Phytologist Research 1157 groups. In Penstemon, Castellanos et al. (2006) showed that accessible to ineffective pollinators or pollen thieves. The viscin restrictive anther openings could be viewed as a character in the threads connecting pollen grains may reduce pollen wash-out ‘bee syndrome’. Our work is analogous in that we suggest that caused by rain, but pollen in rainwater presumably loses viability small PTT clusters, resulting from low strength of pollen aggre- quickly. Interestingly, our previous study of pollen longevity in gation, can be seen as another adaptive response to counteract the water in 80 spring flowering plants showed that pollen from the potential waste of pollen resulting from frequent collections by two Rhododendron species had the longest life-span both in dry corbiculate bees. Both of these novel characters reflect a conflict conditions and in water (Mao & Huang, 2009). Pollen resistance interest for pollen between bee pollinators and patterns of pollen to water damage may allow Rhododendron species to present pol- presentation in plants (Thomson & Wilson, 2008). len extruded from anthers without protection from rain damage. Large bees were reported to be effective pollinators in numerous Unlike bird or Lepidoptera foraging behavior, which involves Rhododendron species (reviewed in Huang et al., 2017) given that little grooming, bees usually groom pollen from their body into clusters of viscin threads connecting many pollen grains covered corbiculae where it is unavailable for pollen transfer to the next their bodies. For example, R. cyanocarpum in southwestern China flower (Harder, 1998). Empirical studies have shown that polli- was considered to be pollinated by bumblebees (Ma et al., 2015). nators with low grooming behavior were comparatively efficient In large-flowered Rhododendron species, however, the morphologi- in pollen transfer (Castellanos et al., 2006; Muchhala & Thom- cal misfit between small pollinators and the spatial separation of son, 2010). Pollen packaging strategies in Rhododendron species sexual organs from the nectar reward may prevent stigmas from may correspond to different frequencies of pollinator visits and contacting pollinators, greatly reducing pollination precision (Wil- pollinator grooming intensities. This dichotomy between large son & Thomson, 1991; Armbruster et al., 2011). Although the doses of pollen in species with infrequent visits and smaller doses seven Rhododendron species categorized as bird and Lepidoptera in species with more frequent visits corresponds well with pollen flowers were also visited by bumblebees in the field, we observed presentation theory (Harder & Thomson, 1989; Thomson et al., that the bees carrying multiple PTTs rarely deposited pollen onto 2000). We observed that all pollen could be removed by one visit stigmas, regardless of whether they were collecting nectar or in R. decorum, which had the largest PTT among the 13 species, pollen. Recent examinations of effective pollinators involving pol- and which was probably moth-pollinated at low visitation rates linator exclusion experiments demonstrated that butterflies in (Fig. 1e,h). Rhododendron calendulaceum in southwestern Virginia, USA (Epps Basal groups in Ericaceae related to Rhododendron do not pro- et al., 2015) and passerine birds in 10 large-flowered duce viscin threads to connect pollen grains, suggesting that Rhododendron species in southwestern China (Huang et al., 2017) extensive pollen aggregation is likely a derived trait. Shifts of contributed a major pollination role, rather than bees. Our catego- functional pollinator groups occurred at least four times in the 13 rization of three flower types invites more field observations of flo- Rhododendron species we sampled in Cangshan Mountains. Shifts ral visitors. However, the general pattern of frequent visits by from bee to Lepidoptera flowers (i.e. R. yunnanense), from bee to heavily grooming bees and infrequent visits by weakly grooming bird flowers (R. delavayi) or from bird to Lepidoptera flowers visitors was consistent with our observations over the years (Fig. 2; (R. decorum) were all associated with an increase of pollen aggre- Huang, 2015; Huang et al., 2017). Large amounts of pollen gation (Fig. 3; all P < 0.001 by simple t-tests). We found that a placed on bats could facilitate pollen transfer, increasing male negative relationship between the strength of pollen aggregation reproductive success of bat-pollinated plants, in that fur holds and pollinator visitation characteristic visitation frequencies and many pollen grains that are little groomed (Muchhala & Thom- pollen grooming behavior. To our knowledge, this is the first son, 2010; Muchhala et al., 2010). Consistent with the notion that study to quantitatively estimate interspecific variation in pollen pollen transfer efficiency would be high if pollinators lost little removal, showing that it relates to pollinator interactions. Studies pollen during pollen removal, we found that interspecific variation have shown that quantitative variation in floral traits is usually in the number of pollen grains per PTT was strongly correlated associated with the strength of pollinator selection (Harder & with stigmatic pollen loads (Table 2). Selection for larger PTT Johnson, 2009). For example, pollinator visitation frequency has with more pollen in Rhododendron species could increase male fit- been demonstrated as a key factor affecting floral longevity (Ash- ness in that pollen removal per visit increases linearly with size of man & Schoen, 1994). Further studies on intraspecific variation PTT in plants pollinated by non-grooming animals. in PTT sizes and geographic differentiation in functional pollina- Bumblebees usually buzz when collecting pollen from tor groups among populations would provide additional evidence Rhododendron flowers, but pollen release with PTT is not for the hypothesis that pollen dosing strategy is associated with attributable to bee sonication, given that PTT connected pollen pollinator characteristics. To illustrate the evolution of floral grains are ready to spill out of the anther pore or have already traits driving male fitness in plants, a perspective of plant groups been extruded at the anther tip in newly opened flowers. King & in which male reproductive success and shifts of pollinators are Buchmann (1995) proposed that a centrifugal force generated by easily quantified would be very fruitful. the natural vibration mode of the stamens was sufficient to throw the PTT toward the apical pore and out of the anther. Such an Acknowledgements exposed pollen presentation mode may involve a high risk of pol- len loss caused by pollen collectors or physical conditions. The authors thank lab members Wei Guo, Kai Hao, Ze-Yu Without mechanical protection, the PTT with pollen are Tong, Bo Wang and Di Wu for their help in the field study, and

Ó 2018 The Authors New Phytologist (2019) 221: 1150–1159 New Phytologist Ó 2018 New Phytologist Trust www.newphytologist.com New 1158 Research Phytologist

Ying-Ze Xiong for help with data analysis. James Thomson Harder LD, Thomson JD. 1989. Evolutionary options for maximizing pollen dispersal of animal-pollinated plants. American Naturalist 133: kindly commented on an early draft of the manuscript; Sally – Corbet, Paul Wilson, Jana Vamosi, Wen-Bin Yu and two anony- 323 344. Harder LD, Wilson WG. 1994. Floral evolution and male reproductive success: mous reviewers improved a later version. This work was sup- optimal dispensing schedules for pollen dispersal by animal-pollinated plants. ported by the National Science Foundation of China (NSFC Evolutionary Ecology 8: 542–559. grant nos. U1402267, 31730012 and 31700196). Hesse M. 1984. An exine architecture model for viscin threads. Grana 23:69–75. Hesse M, Vogel S, Halbritter H. 2000. Thread-forming structures in angiosperm anthers: their diverse role in pollination ecology. Plant Systematics and Evolution Author contributions 222: 281–292. Huang ZH. 2015. Pollination system and floral divergence in Rhododendron S.Q-H. and Y-P.S. planned and designed the research. Z-H.H. (Ericaceae). PhD Thesis, Wuhan University, Wuhan, China. and Y-P.S. performed experiments, conducted fieldwork, Y-P.S. Huang ZH, Song YP, Huang SQ. 2017. Evidence for passerine bird pollination and S-Q.H. analyzed data, etc. Y-P.S., Z-H.H. and S-Q.H. wrote in Rhododendron species. AoB Plants 9: plx062. the manuscript. Y-P.S. and Z-H.H. contributed equally to this Katoh K, Toh H. 2010. Parallelization of the MAFFT multiple sequence alignment program. Bioinformatics 26: 1899–1900. work. Kay KM, Schemske DW. 2010. Pollinator assemblages and visitation rates for 11 species of neotropical Costus (Costaceae). Biotropica 35: 198–207. King MJ, Buchmann SL. 1995. Bumble bee-initiated vibration release ORCID mechanism of Rhododendron pollen. American Journal of Botany 82: 1407– Shuang-Quan Huang http://orcid.org/0000-0003-4540-1935 1411. Knox RB, McConchie CA. 1986. Structure and functions of compound pollen. In: Blackmore S, Ferguson IK, eds. Pollen and spores: form and function. London, UK: Academic Press for the Linnean Society, 265–282. Kress WJ. 1981. Sibling competition and evolution of pollen unit, ovule number, References and pollen vector in angiosperms. Systematic Botany 6: 101–112. Alonso-Pedano M, Ortega-Baes P. 2012. Generalized and complementary Kudo G, Hirao AS, Kawai Y. 2011. Pollination efficiency of bumblebee queens pollination system in the Andean cactus Echinopsis schickendantzii. Plant and workers in the alpine Rhododendron aureum. International Journal of Systematics and Evolution 298: 1671–1677. Plant Sciences 172:70–77. Armbruster WS, Gong YB, Huang SQ. 2011. Are pollination “syndromes” Ma YP, Wu ZK, Dong K, Zhang CQ, Sun WB, Marczewski T. 2015. Pollination predictive? Asian Dalechampia fit neotropical models. American Naturalist 178: biology of Rhododendron cyanocarpum (Ericaceae), an alpine species endemic to 135–143. NW Yunnan, China. Journal of Systematics and Evolution 53:63–71. Ashman TL, Schoen DJ. 1994. How long should flowers live? Nature 371: 788– Ma YP, Zhang CQ, Zhang JL, Yang JB. 2010. Natural hybridization between 791. Rhododendron delavayi and R. cyanocarpum (Ericaceae), from morphological, Castellanos MC, Wilson P, Keller SJ, Wolfe AD, Thomson JD. 2006. Anther molecular and reproductive evidence. Journal of Integrative Plant Biology 52: evolution: pollen presentation strategies when pollinators differ. American 844–851. Naturalist 167: 288–296. Maddison WP, Maddison DR. 2010. Mesquite: a modular system for evolutionary Castellanos MC, Wilson P, Thomson JD. 2003. Pollen transfer by analysis. 2.73. [WWW document] URL http://mesquiteproject.org. hummingbirds and bumblebees, and the divergence of pollination modes in Mao YY, Huang SQ. 2009. Pollen resistance to water in 80 angiosperm species: Penstemon. Evolution 57: 2742–2752. flower structures protect rain susceptible pollen. New Phytologist 183: 892–899. Cruden RW, Jensen KG. 1979. Viscin threads, pollination efficiency and low Midford PE Jr, Garland T, Maddison WP. 2005. PDAP package of Mesquite, pollen-ovule ratios. American Journal of Botany 66: 875–879. version 1.07. Darriba D, Taboada GL, Doallo R, Posada D. 2012. jModelTest 2: more Muchhala N, Brown Z, Armbruster WS, Potts MD. 2010. Competition drives models, new heuristics and parallel computing. Nature Methods 9: 772. specialization in pollination systems through costs to male fitness. American Epps MJ, Allison SE, Wolfe LM. 2015. Reproduction in flame azalea Naturalist 176: 732–743. (Rhododendron calendulaceum, Ericaceae): a rare case of insect wing pollination. Muchhala N, Thomson JD. 2010. Fur versus feathers: pollen delivery by bats and American Naturalist 186: 294–301. hummingbirds and consequences for pollen production. American Naturalist Fang Q, Chen YZ, Huang SQ. 2012. Generalist passerine pollination of a 175: 717–726. winter-flowering fruit tree in central China. Annals of Botany 109: 379– Ronquist F, Huelsenbeck JP. 2003. MrBayes 3: Bayesian phylogenetic inference 384. under mixed models. Bioinformatics 19: 1572–1574. Felsenstein J. 1985. Phylogenies and the comparative method. American Schemske DW, Bradshaw HD. 1999. Pollinator preference and the evolution of Naturalist 125:1–15. floral traits in monkeyflowers (Mimulus). Proceedings of the National Academy of Fenster CB, Armbruster WS, Wilson P, Dudash MR, Thomson JD. 2004. Sciences, USA 96: 11910–11915. Pollination syndromes and floral specialization. Annual Review of Ecology, Singer R, Cocucci AA. 1997. Eye attached hemipollinaria in the hawkmoth and Evolution & Systematics 35: 375–403. settling moth pollination of Habenaria (Orchidaceae): a study on functional Harder LD. 1998. Pollen-size comparisons among animal-pollinated morphology in 5 species from subtropical South America. Botanica Acta 110: angiosperms with different pollination characteristics. Biological Journal of the 328–337. Linnean Society 64: 513–525. Smith SD, Ane C, Baum DA. 2008. The role of pollinator shifts in the floral Harder LD, Johnson SD. 2008. Function and evolution of aggregated pollen in diversification of Iochroma (Solanaceae). Evolution 62: 793–806. angiosperms. International Journal of Plant Sciences 169:59–78. Stevens PF. 1976. The altitudinal and geographical distribution of flower types in Harder LD, Johnson SD. 2009. Darwin’s beautiful contrivances: Rhododendron section Vireya, especially in the Papuasian species, and their evolutionary and functional evidence for floral adaptation. New Phytologist significance. Botanical Journal of the Linnean Society 72:1–33. 183: 530–545. Stout JC. 2007. Pollination of invasive Rhododendron ponticum (Ericaceae) in Harder LD, Routley MB. 2006. Pollen and ovule fates and reproductive Ireland. Apidologie 38: 198–206. performance by flowering plants. In: Harder LD, Barrett SCH, eds. Stout JC, Parnell JAN, Arroyo J, Crowe TP. 2006. Pollination ecology and seed Ecology and evolution of flowers. Oxford, UK: Oxford University Press, production of Rhododendron ponticum in native and exotic habitats. Biodiversity 61–80. & Conservation 15: 755–777.

New Phytologist (2019) 221: 1150–1159 Ó 2018 The Authors www.newphytologist.com New Phytologist Ó 2018 New Phytologist Trust New Phytologist Research 1159

Tagane S, Hiramatsu M, Okubo HJ. 2008. Hybridization and asymmetric introgression between Rhododendron eriocarpum and R. indicum on Yakushima Supporting Information Island, southwest Japan. Journal of Plant Research 121: 387–395. Thomson JD, Thomson BA. 1992. Pollen presentation and viability schedules in Additional Supporting Information may be found online in the animal pollinated plants: consequences for reproductive success. In: Wyatt R, Supporting Information section at the end of the article: ed. Ecology and evolution of plant reproduction. New York, NY, USA: Chapman & Hall, 1–24. Fig. S1 An illustration of measuring sizes of pollen aggregation Thomson J, Wilson P. 2008. Explaining evolutionary shifts between bee and in Rhododendron species. hummingbird pollination: convergence, divergence, and directionality. International Journal of Plant Sciences 169:23–38. Thomson JD, Wilson P, Valenzuela M, Malzone M. 2000. Pollen presentation Table S1 Pollinator functional groups in the 13 Rhododendron and pollination syndromes, with special reference to Penstemon. Plant Species species Biology 15:11–29. Tong ZY, Huang SQ. 2018. Safe sites of pollen placement: a conflict of interest Table S2 GenBank accessions of four markers for 13 studied between plants and bees? Oecologia 186: 163–171. Vaidya G, Lohman DJ, Meier R. 2011. SequenceMatrix: concatenation software Rhododendron species and four outgroup species for the fast assembly of multi-gene datasets with character set and codon information. Cladistics 27: 171–180. Table S3 Summary statistics of DNA datasets and best-fit model Whittall JB, Hodges SA. 2007. Pollinator shifts drive increasingly long nectar of BIC for BI spurs in columbine flowers. Nature 447: 406–409. Wilson P, Thomson JD. 1991. Heterogeneity among floral visitors leads to discordance between removal and deposition of pollen. Ecology 72: 1503–1507. Please note: Wiley Blackwell are not responsible for the content Wilson P, Wolfe AD, Armbruster WS, Thomson JD. 2007. Constrained lability or functionality of any Supporting Information supplied by the in floral evolution: counting convergent origins of hummingbird pollination in authors. Any queries (other than missing material) should be Penstemon and Keckiella. New Phytologist 176: 883–890. directed to the New Phytologist Central Office.

New Phytologist is an electronic (online-only) journal owned by the New Phytologist Trust, a not-for-profit organization dedicated to the promotion of plant science, facilitating projects from symposia to free access for our Tansley reviews and Tansley insights.

Regular papers, Letters, Research reviews, Rapid reports and both Modelling/Theory and Methods papers are encouraged. We are committed to rapid processing, from online submission through to publication ‘as ready’ via Early View – our average time to decision is <26 days. There are no page or colour charges and a PDF version will be provided for each article.

The journal is available online at Wiley Online Library. Visit www.newphytologist.com to search the articles and register for table of contents email alerts.

If you have any questions, do get in touch with Central Office (np-centraloffi[email protected]) or, if it is more convenient, our USA Office (np-usaoffi[email protected])

For submission instructions, subscription and all the latest information visit www.newphytologist.com

Ó 2018 The Authors New Phytologist (2019) 221: 1150–1159 New Phytologist Ó 2018 New Phytologist Trust www.newphytologist.com