Floral phenology of an Andean bellflower and pollination by buff‐tailed sicklebill hummingbird

Abstract The Andean bellflowers comprise an explosive radiation correlated with shifts to specialized pollination. One diverse clade has evolved with extremely curved floral tubes and is predicted to be pollinated exclusively by one of two parapatric species of sicklebill hummingbirds (Eutoxeres). In this study, we focused on the floral biology of Centropogon granulosus, a bellflower thought to be specialized for pollination by Eutoxeres condamini, in a montane cloud forest site in southeastern Peru. Using camera traps and a pollination exclusion experiment, we documented E. condamini as the sole pollinator of C. granulosus. Visitation by E. condamini was necessary for fruit development. Flowering rates were unequivocally linear and conformed to the “steady‐state” phenological type. Over the course of >1800 h of monitoring, we recorded 12 E. condamini visits totaling 42 s, indicating traplining behavior. As predicted by its curved flowers, C. granulosus is exclusively pollinated by buff‐tailed sicklebill within our study area. We present evidence for the congruence of phenology and visitation as a driver of specialization in this highly diverse clade of Andean bellflowers.

One diverse clade of Centropogon, the "eucentropogonids" (38 spp.), evolved after a single unique shift to pollination by sicklebill hummingbirds (Eutoxeres, Lagomarsino et al., 2017). Almost all members of this clade are predicted, on the basis of their strongly curved corolla tubes, to be pollinated by sicklebills (Stein, 1987). In this study, we focus on the pollination of Centropogon granulosus C. Presl by buff-tailed sicklebill (Eutoxeres condamini).
Centropogon granulosus is an understory vine with abruptly curved, bright red to orange tubular flowers (Figure 1). This species is both the most widespread and variable eucentropogonid, occurring from southern Nicaragua to Bolivia. The species examined here conforms to Centropogon granulosus subsp. granulosus (sensu Stein, 1987). Although other eucentropogonid species are found in this region (Stein, 1987), we focus on C. granulosus as it has been previously studied in Costa Rica with respect to pollination by another sicklebill species, Eutoxeres aquila (Stiles, 1985), the only congener of E. condamini. Moreover, C. granulosus is locally abundant, providing a tractable system for study.
Eutoxeres is comprised of two parapatric species of sicklebill hummingbirds that together, adhere to the geographic distribution of the eucentropogonids (Abrahamczyk et al., 2017). White-tipped sicklebill (Eutoxeres aquila) occurs from Costa Rica to northern Peru, whereas buff-tailed sicklebill (E. condamini) occurs from northern Peru to Bolivia (Hinkelmann & Boesman, 2020). Previous studies have supported white-tipped sicklebill as a specialized pollinator of eucentropogonids and some Heliconia spp. with curved corolla tubes (Gill, 1987;Maglianesi et al., 2015;Morrison & Mendenhall, 2020;Stiles, 1985). Its bill curvature matches the curved corollas of these plants more than other co-occurring hermits (Maglianesi et al., 2014;Sonne et al., 2019). Further, its local abundance is correlated with the seasonal flowering of C. granulosus in Costa Rica (Stiles, 1985). In contrast, very little is known of its southern congener, E. condamini. Like E. aquila, its curved bill appears to be suited to feed from eucentropogonids. Currently, there is only a single written record of visitation to a eucentropogonid (Centropogon gamosepalus Zahlbr., Stein, 1987), and further details on the extent of mutualism have not yet been studied (e.g., effects of visitation on fruit set and seed production).
Furthermore, because this pollination system is thought to be specialized, we expect additional aspects of the pollination syndrome, specifically phenology, to reflect adaptation to Eutoxeres behavior. In addition to the seasonal flowering trends documented by Stiles (1985), phenological patterns at finer temporal scales (i.e., days) might also conform to the daily foraging habits of Eutoxeres.
Phenological patterns at this scale have been previously categorized by Gentry (1974): for example, "big bang" species produce many flowers simultaneously over several days, whereas "steady-state" species produce only a few flowers per day over a number of weeks.
Considering that E. aquila is a trapliner (Stiles, 1985;but see Sargent et al., 2021), the "phenological type" of Gentry (1974) that eucentropogonids are likely to exhibit is steady-state flowering, consistent with low, but regular, daily visitation rates by pollinators. Moreover, we expect steady-state flowering to provide insufficient daily nectar to territorial hummingbirds so that these plants would, therefore, only be visited by traplining species (Kessler et al., 2020).
Although the phenological types of some centropogonids have been described qualitatively, (e.g., Colwell et al., 1974;Weiss, 1996), the "phenological type" framework of Gentry (1974) considers two continuous variables, flowering duration (L) and rate (r). We propose that to categorize phenological types, the anthesis rate (r) should be examined for linearity, where we expect steady-state species to exhibit a constant daily flowering rate, whereas "cornucopia" and "big bang" species would flower nonlinearly (Gentry, 1974). To this end, the average deviation from linearity metric (Kroll et al., 2000) will be useful in developing a reproducible, quantitative framework for assigning Gentry's (1974) phenological types (see Methods).
The goal of this study is to test the hypothesis that eucentropogonids are uniquely specialized for pollination by sicklebill hummingbirds, specifically the less well-known buff-tailed sicklebill, by examining the floral phenology and pollination of C. granulosus. Specifically, we ask: (1) Is buff-tailed sicklebill a visitor to, and the sole pollinator of C. granulosus? (2) Does sicklebill visitation affect the reproductive success of C. granulosus? and (3) Is the phenological type of C. granulosus consistent with adaptation to the presumed foraging mode of buff-tailed sicklebill, that is, does C. granulosus exhibit steady-state flowering? F I G U R E 1 Presumed pollination niches within hummingbird-adapted Centropogon. K tot is total curvature in degrees (see: Boehm et al., 2022)

| Pollinator observations
We deployed five camera traps (Hyperfire HC600; Reconyx Inc.) near Centropogon granulosus vines located in a previous survey of the area . Cameras were mounted onto nearby trees using a bungee cord, typically at a distance of 1-2 m from (and at a height equivalent to) the inflorescence. Camera traps were checked for new captures every 12 h. If no floral visitors were recorded within 3 days, the camera traps were moved to different C. granulosus individuals. Where floral visitors were recorded, we attempted targeted video recording to better document visitation behavior. Camera traps were active continuously from August 17 to September 20, 2017 (Table S1). Some flowering stages were not completely observed due to herbivory or weather. Similarly, monitoring of some flowers began with the current stage partially completed. This type of data is "right censored", that is, the true durations of these stages are greater than was observed (Allison, 2014). To account for censoring, we fit parametric survival functions (Allison, 2014) to the stage duration data.

| Pollinator exclusion and floral development
This allowed an estimation of the median duration (m X n ) for each stage (X n ), that is, the number of days elapsed in stage X n before the daily probability of transitioning to stage X n+1 surpassed 50%.
To reconstruct floral development from the censored dataset, we used the median stage durations and 95% confidence intervals (CI) estimated from the survival analysis above. For each treatment, we cumulatively summed the median stage durations to approximate the number of days elapsed between stages A and G. We accounted for error propagation, that is, the uncertainty of each m X 1 , ⋯, m X n in influencing the 95% CI of m X 1 + ⋯ + m X n + m X n+1 , by summing the 95% CIs in quadrature (Ku, 1966).

| Phenological type
To characterize the phenological type of C. granulosus, we used broom v.0.7.6 (Robinson et al., 2021) to fit linear models to the number of flowers produced through time. A separate model was fit for each inflorescence that produced at least five flowers (n = 5 for each treatment). The slope of each linear regression was interpreted as the anthesis rate. To assess linearity, we used lin.eval v.0.1.2 (Shrivastav, 2019) to fit linear and polynomial (>1°) curves to the anthesis rate. This method uses the average deviation from linearity (Kroll et al., 2000), to determine if non-linear fits have significantly lower residuals than a linear regression.

| Floral visitors
Camera trap recordings and in situ observations confirm buff-tailed sicklebill (E. condamini) as a visitor to Centropogon granulosus flowers (Figures 3,S4). Visitation tended to occur from 5:20 to 10:40 in the morning (n = 9), and 12:40 to 16:30 in the afternoon (n = 3), though these patterns may have been affected by our activity in the area.
Given that Eutoxeres is active within an ~11-h daily window, the total monitoring effort was ~1870 h (5 camera traps × 34 days × 11 h per day). Within this time, we recorded 12 visits to six C. granulosus individuals, totaling 42 s of E. condamini observation ( Figure S12, Table   S3). Ten of the 12 records were single, brief visits (≤3 s) that occurred once in the day-two additional records were made when a second visitation was observed on the same day. A total of seven flowers were probed from six C. granulosus individuals, that is, a second visit was recorded to an inflorescence as flowers opened sequentially.
E. condamini feeds both by perching on the lignified inflorescence (n = 3) and hovering (n = 9). We also recorded two instances of sicklebills approaching and inspecting inflorescences without open flowers. Wedge-billed hummingbird (Schistes geoffroyi) was also recorded nectar robbing C. granulosus by piercing the corolla tube at the base. Over the course of 2 days, a camera trap recorded five visits per day to the same inflorescence (Table S4). Further details of S. geoffroyi behavior can be found in Boehm (2018). No other hummingbirds were recorded visiting these flowers.
Reviewing still frames from the video recording reveals an interesting feeding problem posed by the sharply curved flowers of C. granulosus. The hook shape forces the hummingbird to hover below the corolla opening and tilt its head backward so that it is facing skyward (dorsal head flexion), at which point, it can insert the tip of its bill into the flower aperture ( Figure S4). The remainder of the bill is further inserted by tilting the head back to a forward-facing position while lifting itself to eye level with the corolla opening. Due to the unique morphology and orientation of C. granulosus flowers, this hovering maneuver is likely performed only by Eutoxeres. We note that once the bill is inserted, the throat and crown are covered by the ventral and dorsal corolla lobes, respectively.
Two additional vertebrates, a murid (Muridae) and a long-nosed bat (Glossophaginae) were recorded near the inflorescences but not observed to interact with the plant directly ( Figures S5, S6). We note this because it is unknown how the fleshy berries are dispersed, though we documented signs of frugivory ( Figure S7).
Numerous invertebrates occupied or visited the flowers of C. granulosus in this study. As found in previous studies, we observed ants (Stein, 1992), mites (Naskrecki & Colwell, 1998), and dipterids (Weiss, 1996) in or on the flowers of this species. We observed unidentified Arachnids inside of the floral tubes, and note that Anelosimus spiders (Araneae) are known to build webs scaffolded by Centropogon coccineus (Hook.) Regel ex B.D. Jacks. (Nentwig & Christenson, 1986). We also recorded a larval lepidopteran inhabiting a flower (Figure S8), and a stingless bee (Meliponini) collecting pollen from the anther scale ( Figure S9).

F I G U R E 3 Visitation of Centropogon granulosus by Buff-tailed Sicklebill (Eutoxeres condamini)
shape-nearly all floral curvature is developed here. Stages E and F are defined by the staminate and pistillate phases of anthesis, respectively. Following anthesis, the flowers wilt (Stage G) and produce berries (Stage H, Table S2).
These eight stages were used to compare developmental differences between control and pollinator-excluded flowers (Figure 4).
Between treatments, the progression of floral development is comparable from stages A (bud development) to E (anthesis). However, control flowers spend 24.2 ± 4.47 days (median ± 95% CI) developing berries, while no hummingbird-excluded flowers produced berries.
The upper limit of flower production for a single inflorescence has not been determined, though we counted 68 flower abscission scars on the peduncle of an individual not included in this study ( Figure S10).
Linear models accurately described flowering rate ( Figure 5): all anthesis rates were fit better by linear models than polynomials (p < .05). Flowering rate (

| Buff-tailed sicklebill is a pollinator of C. granulosus
As predicted from its extreme bill curvature, buff-tailed sicklebill (E. condamini) is a visitor to C. granulosus, and these visits are necessary for developing fruit. No other hummingbirds were observed legitimately probing these flowers. Covering flowers with wire cages excluded hummingbirds while allowing invertebrates to access the flowers freely-however, none of these flowers produced fruits.
Therefore, we conclude that buff-tailed sicklebill is the sole pollinator of C. granulosus.
While probing for nectar, the face of E. condamini is inserted into the corolla tube so that the crown and throat are covered by the petal lobes. This is facilitated by the exceptionally inflated corolla opening characteristic of the eucentropogonids (Lagomarsino et al., 2017). While narrow corolla apertures are thought to promote specialization (Temeles et al., 2002), the evolution of curvature might relax selection for corolla width. Conversely, because E. condamini tilts its head backward during bill insertion, it may not be able to see the corolla opening; thus, a narrow corolla width could negatively affect pollination if the barriers to accessing nectar are too high (Rico- Westerkamp, 1990).

In contrast to previous accounts of sicklebill visitation to
Centropogon (Stein, 1987;Stiles, 1985), we observed hovering in addition to perching. While floral orientation in some hummingbirdpollinated plants may have evolved to exclude non-hovering visitors (R. Colwell, pers. comm.), hovering is one of the most energetically expensive modes of locomotion (Suarez & Gass, 2002) and is avoided when perches are available (Westerkamp, 1990).
Recent work has found that short-billed hummingbird species have repeatedly evolved large claws that improve their ability to perch (R. Colwell, pers. comm.). Conversely, long-billed species tend to hover to feed, supporting the idea that long (and sometimes curved) tubular flowers evolve in response to selection for pollinator F I G U R E 4 Developmental trajectories for the flowers of Centropogon granulosus and effects of pollinator exclusion (green) versus the control treatment (orange). The bars represent the median duration spent in each stage. 95% of CIs are estimates of when a stage could end. At stage E, where some individuals begin, others will have already finished and moved on to stage F. No fruits were produced by plants with pollinators excluded specialization (Temeles et al., 2019). We speculate that the inflorescences of C. granulosus are lignified primarily to support and orient flowers and are only opportunistically used by sicklebills as perches.
This is because open flowers tend to face away from the stem on long pedicels (Figure 2). This is in contrast to E. condamini visits to nearby Heliconia, which has flowers oriented so that the aperture is aligned with the perch (i.e., floral bract, Figure S11). Whether floral orientation promotes specialization in the eucentropogonids is an understudied aspect of pollination in this clade.

| Steady-state flowering and traplining
Because hummingbird species generally adhere to a single foraging mode (Feinsinger & Colwell, 1978;Stiles, 1985; but see Sargent et al., 2021), phenological types may be effective filters of the local pollinator community, further promoting floral specialization in the eucentropogonids. As with floral shape, phenological types are thought to evolve either via competition for pollination or selection against interspecific pollen transfer (Kessler et al., 2020;Primack, 1985;Rathcke & Lacey, 1985). Therefore, accurately assigning phenological types in the context of pollinator foraging modes will be a key to examining the evolution of this trait in the centropogonids and assessing the role of phenology in pollinator shifts.
Centropogon granulosus exhibits a linear flowering rate befitting the "steady-state" phenological type described by Gentry (1974) as "[the production of] a few flowers a day over an extended period of time (usually a month or more)". It is one of several phenological modes representing an axis of niche partitioning that is thought to contribute to tropical plant diversity (Gentry, 1974;Kessler et al., 2020). Indeed, most hummingbird species exhibit foraging behavior that is adapted either to steady-state or "cornucopia" flowering (sensu Gentry, 1974), with few species able or willing to visit plants of both types (Kessler et al., 2020 (Stiles, 1985), the fine-scale daily movements of Eutoxeres (and Hermits generally) have not yet been studied-at present, comparative analyses are constrained by our limited knowledge of these rarely seen pollinators.
Finally, while steady-state flowering is not solely indicative of specialization in Eutoxeres, we speculate that it is a component of the iterative process by which specialization evolves. That is, steadystate flowering may have first co-evolved with traplining hummingbirds (Rombaut et al., 2022), which excluded visitation by species under stabilizing selection for territoriality. Among the steady-state species, floral morphology continued to evolve, further partitioning the steady-state species between grades of curvature ( Figure 1).

ACK N OWLED G M ENTS
Access to field sites was made possible by the Servicio Nacional

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data and R scripts are available in the Dryad repository https:// doi.org/10.5061/dryad.ns1rn 8pwb. These materials are also available as an RStudio Project at: https://github.com/mannf red/centr opogon_eutox eres.