V ISITATION FREQUENCIES OF BUMBLEBEES AND SWALLOWTAIL BUTTERFLIES TO FLOWERS AND THE NECTAR SUGAR CONCENTRATION OF R HODODENDRON KAEMPFERI AND R. JAPONICUM IN MOUNTAINS OF CENTRAL J APAN

—Relationships between flower visitor composition and floral traits help us to understand floral evolution. Swallowtail butterflies have been assumed to be the main pollinators of Rhododendron kaempferi and R. japonicum based on their floral shapes, and R. kaempferi was reported to be pollinated by butterflies in southern Japan. In the mountains of central and northern Japan, however, bumblebees are abundant and candidate pollinators of Rhododendron. We found that visitation frequencies of bumblebees were higher than those of swallowtail butterflies to both of the Rhododendron species at three study sites in the mountains of central Japan. R. japonicum pollen adhered to the bodies of both floral visitors. Further, the nectar sugar concentration of R. kaempferi was in the preference range of bumblebees, whereas that of R. japonicum was suited to both bumblebees and butterflies. Taken together, these findings support the hypothesis that bumblebees are potentially also important pollinators of R. kaempferi and R. japonicum in mountains of central Japan, where they are more abundant than swallowtail butterflies.


INTRODUCTION
Relationships between flower visitor composition and floral traits can help us to understand floral evolution (Kevan & Baker 1983;Schemske & Bradshaw 1999). Suites of floral traits have evolved in response to natural selection imposed by different pollinator taxa, and these relationships are referred to as "pollination syndromes" (Johnson & Steiner 2000;Fenster et al. 2004). In recent years, the validity of the traditional pollination syndrome concept on a world scale has been challenged (Ollerton et al. 2009;Rosas-Guerrero et al. 2014). The expected pollinator of some plant species on the basis of their floral traits (shape, colour, scent, size, etc.) is often not realised (Ollerton et al. 2009;Vlašánková et al. 2017). In these cases, although floral traits may have initially evolved in response to the most effective pollinator (Rosas-Guerrero et al. 2014), pollinator transitions can occur (Ollerton et al. 2009), which can result in floral traits that do not predict the current pollinator.
The shrubs R. kaempferi (subgenus Azaleastrum, section Tsutsusi) and R. japonicum (subgenus Pentanthera, section Pentanthera) have medium to large red or red-orange flowers (corolla diameter 4 and 7 cm, respectively), and a long pistil and long stamens that stick out of the corolla.
Rhododendron flowers with this colour and shape have been thought to be pollinated by large butterflies, because butterflies easily recognize reddish colours and when they alight on the corolla, their ventral abdomen is suitably positioned to brush against the long pistil and stamens (Tanaka 1993(Tanaka , 1997(Tanaka , 2001. Indeed, in the Kirishima Mountains in southern Japan, R. kaempferi is frequently visited by the swallowtail butterflies Byasa alcinous and Papilio xuthus (Yokogawa & Hotta 1995). However, swallowtail butterflies are rare in northern and mountainous regions in Japan, whereas bumblebees, the other potential Rhododendron pollinator, are abundant. When the body sizes of bumblebees match the size of Rhododendron flowers, their ventral abdomen touches the anthers and the stigma (Mejias et al. 2002). Whether R. kaempferi and R. japonicum are pollinated by large swallowtail butterflies or by bumblebees in these central mountainous regions of Japan remains unknown.
In this paper, we pose three hypotheses. First, we hypothesize that bumblebees, not swallowtail butterflies, are main flower visitors of R. kaempferi and R. japonicum in the mountains of central Japan. Second, sugar concentration of nectar of R. kaempferi and R. japonicum is high (suited to bumblebee pollination). Third, that due to similar selection pressures by the same pollinators, flower size of R. kaempferi and R. japonicum does not vary between populations.

Study sites
This study was carried out in Nagano Prefecture, central Japan, for 26 days from 29 April to 10 July 2016. This period includes the peak flowering season of both Rhododendron kaempferi and R. japonicum. We investigated R. kaempferi at sites in Ikeda (Ike, 36°23'N, 137°53'E; 633 m a.s.l.) and Norikura (Nor, 36°07'N, 137°37'E; 1,488 m a.s.l.), and R. japonicum at sites in Nor and Utsukushigahara (Uts, 36°14'N, 138°06'E; 1,892 m a.s.l.). In Ike, 50 R. kaempferi plants grew within an area of 50 m × 50 m at the edge of a mixed forest in which Pinus densiflora was the dominant tree species. In Nor, a few dozen R. kaempferi plants grew sporadically over a distance of about 1 km along the edge of a mixed forest in which Quercus crispula and Malus sieboldii were dominant tree species. Also in Nor, 200-300 R. japonicum plants grew sporadically along the forest edge. In Uts, 500-600 R. japonicum plants grew in pastureland within an area of about 200 m × 800 m. We haphazardly selected 15-20 R. kaempferi and R. japonicum plants at each study site for the following investigation.

Visitation frequencies and pollen adhesion on flower-visiting insects
Visitation frequencies of insects, calculated as per hour per flower, were counted for a total of 10 hours for the two plant species between 08:00 and 16:00 local time (LT) at each site. We regarded an insect as a flower visitor if it inserted its mouthpart into at least one flower nectar tube in a patch. The number of R. kaempferi flowers in a patch was 137.8 ± 94.6 (mean ± S.D., N = 10) in Ike and 87.8 To quantify the number of R. japonicum pollen grains that adhered to the bodies of the visiting insects, we placed round seals (8 mm in diameter, adhesive on one side, NICHIBAN ML-151) on the ventral abdomen of P. maackii (N = 1) and on the dorsal and lateral thorax and the ventral abdomen (i.e., three seals per individual) of B. diversus queens (N = 4) in Nor on 15 June 2016. We then counted the number of pollen grains adhering to the seals under a stereoscopic microscope (Nikon SMZ1500). In R. kaempferi, we captured a queen and a worker of B. ardens, and an individual of P. maackii in Ike and confirmed pollen grains of R. kaempferi attached to their bodies with the naked eye.
Although we could not test pollination efficiency directly (e.g., by single visit experiments, or other ways of measuring pollen deposition), we collected visitation data and floral characteristics data (describe below) as proxies.

Daytime and night-time nectar collection quantities
To determine whether R. kaempferi produced more nectar in the daytime or at night, we used 0.5 µL capillary tubes (Drummond Scientific) to remove all the standing crop of nectar from 12 haphazardly selected R. kaempferi flowers in Nor at 16:00 LT on 31 May and then covered each flower with a nylon net to prevent insect access. The next morning, at 07:30 LT, we measured the nectar accumulated during the previous night in nine flowers (three flowers fell off the plant during the night). Then we recovered each of the nine flowers with a nylon net and, at 16:00 LT, measured the nectar volume in six flowers (three flowers fell off the plant during the day) accumulated in the daytime. We compared night-time and daytime nectar collection quantities by the Wilcoxon rank sum test (R version 3.4.0; R core team 2017).

Sugar concentration of nectar
We haphazardly selected 3-20 flowers (each flower from a different plant) per site per species and measured the sugar concentration of the standing crop of nectar at 08:00 LT (three flowers of R. kaempferi in Ike, three flowers of R. kaempferi in Nor, three flowers of R. japonicum in Nor, and 20 flowers of R. japonicum in Uts) by using 2µL capillary tubes (Drummond Scientific) and hand-held refractometer (Bellingham and Stanley). To investigate the pattern of the large variation of sugar concentration in R. japonicum in Uts, we increased the sample size there. In addition, we haphazardly selected three flowers from each of four randomly selected R. japonicum plants in Uts and measured the sugar concentration of the standing crop of nectar between 09:00 and 12:00 LT, and between-plant differences in nectar sugar concentration were tested by using one-way ANOVA (R version 3.4.0; R core team 2017).

Floral size
We haphazardly selected 20 flowers (one from each of 20 plants) per site per plant species (but 15 R. kaempferi J Poll Ecol 21 (5) flowers in Nor), and measured corolla diameter, lengths of the shortest and longest stamen (among five stamens), style length, and nectar tube length. Corolla diameter was defined as the distance between the top centre of the upper petal and the midpoint between the apices of the two lower petals (Fig.  1C). Nectar tube refers to the tubular structure at the base of the upper petal (Fig. 1A). Intraspecific and between-site differences in floral sizes were tested by using an unpaired two-tailed t-test.

Visitation frequencies and pollen adhesion on flower-visiting insects
Two species of butterfly were observed to visit Rhododendron kaempferi in Ike (Parnassius glacialis and Papilio maackii), whilst none were observed to visit R. kaempferi in Nor. P. maackii was also observed to visit R. japonicum in Nor, whilst no butterflies were observed to visit R. japonicum in Uts (Fig. 2). Three species of bumblebee were observed: R. kaempferi was visited by Bombus ardens in Ike and B. diversus in Nor; and R. japonicum was visited by B. ardens, B. hypocrita and B. diversus in Nor and B. hypocrita in Uts.
Bumblebees were more frequent visitors to R. kaempferi and R. japonicum than swallowtail butterflies at all study sites although the difference was only statistically significant for R. japonicum in Nor (Fig. 2).
Our limited observations of 3,200 flowers over 1 hour suggested that bumblebees stayed longer in a flower patch and visited more flowers per hour than butterflies. One individual of P. maackii perched and took nectar from only two flowers within a minute whilst three B. diversus queens stayed for 12, 15, and 18 minutes, visiting about 10 flowers per minute.

Nectar collection quantities during the day and night
In Nor, a total of 0.005 ± 0.004 μL (mean ± S.E., N = 9) of R. kaempferi nectar was extracted after flowers had been bagged overnight (for 15 and a half hours), and a total of 0.289 ± 0.049 μL (N = 6) at the end of the following day (8 and a half hours, W = 0, P < 0.001).

DISCUSSION
Bumblebees were more frequent visitors than swallowtail butterflies to Rhododendron kaempferi and R. japonicum spp., although the difference is only significant in one of four cases. Yokogawa and Hotta (1995) reported that the most frequent flower visitor to R. kaempferi was the swallowtail butterfly Byasa alcinous in the Kirishima Mountains, Kagoshima Prefecture (31°53'N), southern Japan. The species diversity and abundance of swallowtail butterflies are higher in southern Japan (Kiritani 2006), whereas those of bumblebees are higher in northern and mountainous central Japan (Goulson 2003). Therefore, the main pollinator of R. kaempferi may change between mountainous central Japan and southern Japan (over a distance of 777 km) according to local bee and butterfly diversity and abundance.
Although Tanaka (1993Tanaka ( , 1997Tanaka ( , 2001 inferred that floral shape of Rhododendron spp. is adapted to swallowtail butterfly pollination in Japan, pollen grains also adhere to the bodies of the flower-visiting bumblebees, suggesting that they could also be pollinators of R. kaempferi and R. japonicum. When bumblebees hovered in front of the corolla, they changed their position frequently and pollen grains adhered all parts of their bodies. On the other hand, when the swallowtail butterfly P. maackii visited a flower, it landed on the pistil and stamens and sucked nectar with its legs and the ventral side of abdomen being in contact with the stigma and anthers, which extend outward horizontally with slightly curving tips in both R. kaempferi and R. japonicum. This more precise pollen placement may have implications for the relative efficiency of butterflies vs. bumblebees as pollinators, but this requires further investigation. The adaptation of nectar production time has been reported to be a component of a pollination syndrome (Cruden & Hermann-Parker 1979;Itino et al. 1991). R. kaempferi nectar was mainly produced in the daytime, making it suited to daytime pollinators such as swallowtail  The mean nectar sugar concentration of R. kaempferi (51-54%) and R. japonicum (30-45%) suggests an adaptation to pollination by either bumblebees, which prefer nectar with a high sugar concentration (30-55%, Willmer 2011), or butterflies (butterfly-pollinated flowers sugar concentration ranges between 17-40%, Willmer 2011).
However, in Uts, the nectar sugar concentration of R. japonicum varied from 19% to 45% between individual plants. Chwil and Weryszko-Chmielewska (2009) also reported that, in Poland, the sugar concentration of R. japonicum nectar rises ontogenetically from 19% (on the third day of flowering) to 37% (on the fourth day of flowering). Since our sample sizes were small, and did not take into account flower age, the patterns of variation in the nectar sugar concentration warrant a future research. Nagano et al. (2014) and Kuriya et al. (2015) showed that flower size and pollinator size were correlated in Campanula punctata and Prunella vulgaris, respectively, across regions several kilometres apart, reflecting adaptation of flowers to the pollinators distributing each region with different elevations. In contrast to these plant species that are adapted to bumblebee pollination, we expected that flower size of R. kaempferi and R. japonicum that are visited not only by bumblebees but also by swallowtail butterflies does not vary between populations. As a result, it is not surprising that we found no differences in floral size between sites for R. kaempferi. Smaller flowers of R. japonicum at one site may just have been a response to local abiotic conditions. We conclude that both bumblebees and swallowtail butterflies are potential pollinators of R. kaempferi and R. japonicum in the study sites. Further investigation is required to determine the real effective pollinator (bumblebees or butterflies) of R. kaempferi and R. japonicum, and to see whether the pollinator-related traits of R. kaempferi and R. japonicum changes between populations depending on the abundance of and adaptation to the effective pollinator group.