EXPERIMENTAL EVIDENCE FOR PREDOMINANT NOCTURNAL POLLINATION DESPITE MORE FREQUENT DIURNAL VISITATION IN ABRONIA

Different suites of floral traits are associated with historical selection by particular functional groups of pollinators, but contemporary floral phenotypes are not necessarily good predictors of a plant’s effective pollinators. To determine the extent to which plant species specialize on particular functional groups of pollinators, it is important to quantify visitation rates for the full spectrum of flower visitors as well as to experimentally assess the contributions of each functional group to plant reproduction. We assessed whether attracting both diurnal and nocturnal flower visitors corresponded to pollination generalization or specialization in the Pacific coastal dune endemic Abronia umbellata var. umbellata. In multiple populations over two years, we observed flower visitors during the day and at night to assess visitation rates by different insect groups and conducted pollinator exclusion experiments to assess the contributions of diurnal and nocturnal visitors to seed production. Flower visitation rates were 8.67 times higher during the day than at night, but nocturnal visitation resulted in significantly higher seed set, suggesting that nocturnal noctuid and sphingid moths are the chief pollinators. Most diurnal visitors were honey bees, with tongues too short to reach A. umbellata nectar or contact stigmas and effect pollination. The prevalence of honey bees, combined with the lack of successful seed production resulting from diurnal pollination, suggests that honey bees are pollen thieves that collect pollen but do not deposit it on stigmas. Our results underscore the need to experimentally assess the contributions of different groups of flower visitors to plant reproduction.


INTRODUCTION
The remarkable phenotypic diversity of flowering plants is thought to be due largely to selection exerted on flower morphology and development by pollen vectors (Fenster et al. 2004). One of the most compelling pieces of evidence for the importance of pollinator-mediated selection is the nonrandom association of floral traits among species that differ in the types of animals that mediate cross-pollination, a phenomenon generally referred to as pollination syndromes (van der Pijl 1961). For example, plants that have historically experienced selection exerted predominantly by nocturnal moths typically have medium to long narrow corolla tubes that contain relatively dilute nectar, are white or pale in colour, lack nectar guides, and emit a sweet fragrance in the evening (Willmer 2011).
While pollination syndromes can provide insight into the historical pollinator-mediated selection that helped to shape contemporary floral phenotypes, they do not necessarily predict which flower visitors are currently effective pollinators (Ollerton et al. 2009), nor can they be used to infer the degree of specialization of a plant's current pollinator fauna (Waser et al. 1996). The number of functional groups observed visiting a plant species is often a better indicator of that plant's degree of specialization than the number of visiting taxa alone (Fenster et al. 2004;Armbruster 2017), but it remains critical to distinguish between visitors that effect pollination and those that do not, especially because many animals that visit flowers perform little to no pollination (Hargreaves et al. 2009;Irwin et al. 2010).
The important distinction between floral visitation and effective pollination makes experimental approaches critical for determining the importance of particular visitors or visitor guilds to pollination and plant fitness. Different visitor guilds often visit specific plant taxa at different times of day, thus temporal pollinator exclusion can be used to determine the relative importance of temporally-divergent visitors (Brunet & Holmquist 2009;Walter 2010;Bustamante et al. 2010). Combining such experiments with observations of flower visitors can determine which functional groups are the most effective pollinators. Such experiments should be replicated across populations and years, given that many studies have demonstrated pronounced spatiotemporal variation in pollinator faunas (Herrera 1988;Wolfe & Barrett 1988; Eckert 2002). However, spatiotemporal replication is logistically demanding and difficult to achieve in many systems. We are aware of only four studies that include multiple years of study at more than one site (Morse & Fritz 1983;Fleming et al. 2001;Holland & Fleming 2002;Bustamante et al. 2010), just three of which combined temporal exclusion experiments with pollinator observations. The Pacific coast dune plant Abronia umbellata Lam. var. umbellata (Nyctaginaceae) possesses some floral traits seemingly specialized for moth attraction at night, but other traits that are attractive to diurnal insects. It provides an excellent opportunity to study the extent to which floral traits associated with attracting visitors from multiple functional groups translate into a generalized vs. specialized pollinator fauna. Abronia umbellata var. umbellata is self-incompatible (SI) and obligately outcrossing, in contrast to the selfcompatible and highly autogamous var. breviflora (Doubleday et al. 2013). Abronia umbellata var. umbellata (hereafter SI A. umbellata) exhibits some traits that typify the mothpollination syndrome: it has reverse herkogamous narrow, tubular flowers with anthers close to the mouth of the floral tube and stigmas recessed deeply within (Fig. 1B) and, in the evening, emits a sweet fragrance containing benzenoid compounds typical of moth-pollinated species (Doubleday et al. 2013). However, unlike canonically white "moth flowers," the flowers are pink-purple with contrasting white "eyespots" encircling the floral tube opening (Fig. 1A), suggesting attraction of diurnal visitors. The flowers are uniovulate and borne on umbellate inflorescences, and fruits are tough, winged diclesia, a type of anthocarp that appears to be   Tillett (1967) speculated that crepuscular and nocturnal noctuid and sphingid moths are likely the most important pollinators.
This study has two main objectives. First, we document patterns of visitation by different insect guilds to flowers during the day and at night. Second, we use temporal pollinator exclusion experiments to determine which visitors are effective pollinators by directly quantifying the relative importance of nocturnal vs. diurnal flower visitors to seed set.
We assess SI A. umbellata's pollination ecology in multiple natural populations in two years.

Pollinator observations
At each of four sites on the Pacific coast of California, USA (Tab. 1) we estimated the rates of diurnal and nocturnal insect visitation to individual SI A. umbellata flowers during standardized 3-or 5-min observation periods. For each period, we randomly selected a patch of inflorescences within a circular plot of ~ 2 m diameter and simultaneously observed all flowers without regard to individual plants (plants are prostrate and intermingle such that individuals that cannot be separated without damaging them). We recorded each flower visitor's identity (to the lowest taxonomic level possible in the field), the number of umbels visited by each flower visitor, and the number of flowers probed on each umbel visited. Our presence, standing ~ 1 m from the plot, did not seem to influence flower visitor behaviour. We used red headlamps (Petzl® TIKKA PLUS2®) for evening (sunset -0200 h) observations because nocturnal insects are relatively insensitive to red light (Briscoe & Chittka 2001). We conducted observations across all times of day and night except between midnight and 0500 h (Fig. 2). We saw insects probing flowers only between 1000 and 2200 h, with peaks in flower visitation between 1100 and 1600 h and 2000 and 2200 h (Fig. 2). The timing of these flower visitation peaks, combined with a lull in visitation between 1600 and 2000 h (Fig. 2), sunrises between 0545 and 0630 h, and sunsets between 2000 and 2030 h made 0600 and 1800 h reasonable dividing times  between day and night. We made 492 observation periods (26.97 h) between 0600 and 1800 h ("day") and 871 (52.03 h) between 1800 and 0600 h ("night"). There were two reasons why we conducted more observations at night: daylight was important for other sampling activities, so we could not devote as much daytime effort to pollinator observations, and we were determined to learn the taxonomic identities of nocturnal moth pollinators, even though they were relatively rare. Because of the logistic challenges involved in studying multiple sites separated by hundreds of km, while conducting experiments lasting 7-10 d at some sites, sampling effort was unbalanced across sites and sites were visited on different dates (Tab. 1 & 2). Pollinator faunas often fluctuate seasonally (Herrera 1988), and sampling date may have affected which flower visitors were present at a given site. Whenever possible, we collected a sample of flower-visiting insects and identified these to the lowest taxonomic level possible given the expertise available to us (Tab. 3). We calculated total visitation for each insect group as visits per flower per hour.

Temporal pollinator exclusion experiment
At four sites over two years we quantified the relative importance of diurnal vs. nocturnal pollination by excluding pollinators from individual umbels at different times of day. The experiment involved one site in 2010 (CMNA) two others in 2011 (CMGA, CSPA) and a third in both years (CCOA; Tab. 1). Individual umbels were allocated to one of three treatments: (i) exposed to pollination at night but enclosed in a wire cage covered with fine bridal veil during the day (0600-1800 h); (ii) exposed to pollination during the day but enclosed at night (1800-0600 h); or (iii) exposed to pollination always (i.e. never enclosed). In 2011, we added an additional treatment: (iv) enclosed always. Any seed produced by these umbels would have to result from either autonomous selfing and/or pollinators accessing enclosed flowers. Only a very small number of continuously enclosed umbels set seed: one at CCOA (mean proportion of flowers setting seed ± 1SE = 0.0018 ± 0.0033 seeds/ovule), two at CMGA (0.010 ± 0.0081) and seven at CSPA (0.029 ± 0.012) suggesting that autonomous selfing and/or the failure of the enclosures to exclude pollinators were infrequent. It is possible that stigmas frequently received self-pollen, but fertilization rarely occurred because of strong genetic self-incompatibility (Doubleday et al. 2013). Experimental crosses on both enclosed and exposed plants did not reveal any negative effects of enclosure on seed set (Doubleday 2012).
We randomly assigned ≥ 30 umbels (only one umbel per plant) just at anthesis (1-2 flowers beginning to open) to each of the treatment groups. Because A. umbellata infructescences shatter at maturity, we bagged them with bridal veil to capture all mature diclesia. For each umbel, we assessed whether each flower produced a seed by counting the number of flowers on each umbel and the number of filled seeds produced. Failure to set seed included flowers from which a diclesium never developed, and those from which a diclesium developed but did not contain a filled seed. At our study sites, only about 30% of developed diclesia contain seeds (mean ± 1SD: 29.52 ± 30.00%; L. A. D. Doubleday and C. G. Eckert, unpublished data). Diclesia routinely expand but contain no seeds under pollinator-free glasshouse conditions (L. A. D. Doubleday and C. G. Eckert, unpublished data), indicating that successful pollination is not a prerequisite for fruit expansion.

Statistical analyses
We used R (version 3.1.3, R Core Team 2017) for all statistical analyses. Because flower visitation occurred in only 6% of observation periods, visitation rates were zero-inflated. Accordingly, we tested for a difference in visitation rate between night and day using randomization tests. We computed the mean difference between day and night visitation rates, randomized the data without respect to when an observation was conducted using the "sample()" function in R, and calculated mean visitation rates for day and night for each of 10,000 randomizations. The proportion of differences calculated from randomized data that were equal to or greater than the observed difference is equivalent to a P value (Prand). We report these as approximate because each randomization run returns a slightly different value.
We tested for differences in seed set among temporal enclosure treatments by fitting generalized linear models (GLMs, glm() function in R) with binomial error structure (logit link function) to variation in seed set (the number of flowers making a seed vs. the number not making a seed). Because we studied different populations and performed a different set of treatments in each year, data for 2010 and 2011 were analyzed separately. Hence, we evaluated site and enclosure treatment and their interaction as potential predictors. We used quasi-likelihood estimation because data were overdispersed, and performed likelihood ratio tests to evaluate the significance of each term in the model using F tests following Buckley (2015). We used the lsmeans() and contrast() functions in the lsmeans R package (version 2.25, Lenth 2016) to perform post-hoc, pairwise contrasts among enclosure treatments.

Flower visitation
Visitation was 8.67-times higher during day than night observation periods for populations and years combined (Tab. When we excluded honey bee visits from the analysis (pooling years and populations), mean night visitation (mean ± 1 SE: 0.0030 ± 0.0011) was higher than day visitation (0.0008 ± 0.0009) but not significantly so (Prand ~ 0.13). The differences for individual sites analysed separately were also not significant (all Prand > 0.11). When observations were pooled by year with honey bee visits excluded, night visitation was more frequent than day visitation for 2010 (Prand ~ 0.037) but not 2011 (Prand ~ 0.18).
We caught 81 visitors that probed SI A. umbellata flowers and identified 70 to species, nine to genus, and two to order (Tab. 3). The 49 day visitors represented at least seven species (though 33 visitors were honey bees) and three orders. The 32 night visitors, all of which were Lepidoptera, represented at least seven species from four families. Noctuid moths were most common (50% of night visitors), with three of four species collected from only one population and the fourth species collected from two populations.

Temporal pollinator exclusion
Seed set was generally very low, even among openpollinated umbels always exposed to pollinators (mean ± 1 SE: 0.071 ± 0.012 seeds/ovule, Fig. 3). In 2011, umbels exposed to pollinators only at night set 3.8-times more seed than those exposed only during the day and did not differ from umbels exposed all the time (Fig. 3, Tab. 4). Seed set was somewhat higher at sites CSPA and CMGA than CCOA, but there was no difference in treatment effects between sites. Results from 2010 suggest 3.4-times higher seed set by nightexposed than day-exposed umbels (Fig. 3), but the difference among treatments was not quite significant (P = 0.073, Tab. 4). Again, there was variation among sites in mean seed set (CCOA > CMNA), but not in the effects of enclosure treatment on seed set. At CCOA for 2010 and 2011 combined, seed set did not vary among treatments or between years, nor did the treatment effect vary between years (Tab. 4).

DISCUSSION
Self-incompatible, outcrossing A. umbellata var. umbellata possesses some traits suggestive of historical selection by nocturnal moths but does not completely conform to the classic moth-pollination syndrome. Our pollinator observations suggest that several functional groups of insects visit the plant during the day and at night, with much higher rates of flower visitation during the day in some populations. However, our temporal pollinator exclusion experiments suggest that nocturnal visitors are more effective pollinators than diurnal visitors. In 2011, night-pollinated umbels had significantly higher seed set (3.8-times higher) than day-pollinated umbels, and the difference in 2010 was of similar magnitude, but not statistically significant (Fig. 3). Taken together, these results suggest predominant nocturnal pollination despite more frequent visitation during the day.
Of the taxa we observed visiting flowers at night, noctuid moths were most common, but we also observed sphingid, pyralid, and geometrid moths. Average tongue lengths were available for members of three of the four moth families we collected from SI A. umbellata flowers: 25-33mm for Hyles lineata, the sphingid species we observed; 10-20 mm for most temperate noctuids; and 4-9 mm for a different pyralid species than the one we observed (Willmer 2011). Stigmas are recessed 14.07 ± 1.01 mm (mean ± 1 SD, C. G. Eckert, unpublished data) from the floral face in SI A. umbellata, suggesting that sphingids and many noctuids would be Many species of Abronia may be predominantly moth pollinated, as the floral display of most species is typified by umbels of fragrant, tubular flowers with recessed stigmas, requiring pollinators with long, narrow tongues. Jabis et al. (2011) suggested that the sphingid moth Hyles lineata was the most effective potential pollinator of A. alpina, but that diurnal taxa also contributed substantially to pollination. Saunders and Sipes (2006) suggested that several species of noctuid and sphingid moths were likely the most important pollinators of A. ammophila, but that butterflies were also likely to affect some pollination. Similarly, Williamson et al. (1994) speculated that noctuid and sphingid moths were key pollinators of A. macrocarpa as multiple moth species were observed visiting flowers and bearing A. macrocarpa pollen in the field.
During the day, we observed much higher frequencies of visitation by honey bees than by other taxa and, and when performing pollen supplementations as part of another experiment, we found it increasingly difficult to obtain sufficient donor pollen from open flowers, which we suspect was due to pollen theft by honey bees (L. A. D. Doubleday, personal observation). The combination of recessed stigmas in flowers of A. umbellata, the relatively short glossae of A. mellifera (mean ± 1 SD = 5.15 ± 0.47mm, n = 28; L. A. D. Doubleday, unpublished data), and our observation of low TABLE 4. Analyses of variation in seed production by umbels of Abronia umbellata after experimental isolation from pollinators. Cells present F and P values from likelihood ratio tests comparing generalized linear models fit with binomial errors to the number of flowers setting seed vs. not setting seed. Contrasts among treatment means are shown when the effect of treatment was significant (A = umbels always exposed to pollinators, N = exposed only during the night, D = exposed only during the day, X = never exposed). All four treatments were applied in 2011, whereas only three were used in 2010 (A, N, D). For the analysis of CCOA including both years, the full model did not fit the data better than a null model (F5,132 = 1.08, P = 0.38). NS = not significant at P < 0.05. Least squares means by treatment and site are in Fig. 2 NS seed set among day-pollinated umbels makes it extremely unlikely that honey bees vector substantial outcross pollen. It is also unlikely that they are successfully extracting nectar from flowers, because available nectar is deeply recessed within the flower at the base of the floral tube (L. A. D. Doubleday, personal observation). This, and the close proximity of dehiscing anthers to the mouth of the floral tube suggests that these visitors are gathering pollen as pollen thieves, but further study would be required to confirm this as we were unable to assess pollen loads on honey bees foraging on SI A. umbellata flowers. It is likely that pollen theft has negative effects on plant fitness: for example, Hargreaves et al. (2010) demonstrated that pollen theft by bees decreases reproductive success in Aloe maculata. Bees are the most commonly reported pollen thieves (Hargreaves et al. 2009).
Introduced pollinators like honey bees have significant, but varied, effects on native ecosystems (Goulson 2003). It would be premature to generalize about the effects of honey bees across diverse native ecosystems, but honey bees have been shown to dominate the spectrum of flower-visiting taxa in certain fragmented habitats in the Neotropics (Aizen & Feinsinger 1994), and meta-analysis has shown that habitat loss has negative effects on the abundance of unmanaged bee species but not on honey bee abundance (Winfree et al. 2009). Honey bees are ineffective pollinators of some native plants but effective pollinators of others (reviewed by Butz Huryn 1997) and introduced honey bees may compete for nesting cavities with native bees, birds, and mammals, but definitive studies are lacking (reviewed by Goulson 2003). Adding hives of Africanized honey bees to habitats in French Guiana reduced visitation to certain plants by native stingless bees, due to honey bees outcompeting native bees for limited floral resources, and removing the honey bee hives reversed the trend (Roubik 1978). Most studies of the effects of honey bees on native pollinators focus on introduced vs. native bees, and less is known about the effects of honey bees on moth pollinators and moth-pollinated plants. The presence of honey bees in Pacific coastal dune systems is likely to have a negative effect on SI A. umbellata's reproductive success, because pervasive pollen theft would increase pollen limitation and A. umbellata's seed set is partly limited by pollen (L. A. D. Doubleday & C. G. Eckert, unpublished data).