Making yourself heard: why well-exposed flowers are an adaptation for bat pollination

Botanists have long recognized that the traits of a flower can often be used to predict its primary pollinator and began formalizing descriptions in the middle of the 20^th century of suites of traits, or pollination syndromes, associated with each pollinator type (Vogel, 1954; van der Pijl, 1961). These traits tend to evolve together during evolutionary shifts to different pollinators, and similar suites of traits will convergently evolve in distantly related taxa (Fenster et al., 2004; Dellinger, 2020). For instance, the syndrome of chiropterophily, or adaptation to pollination by bats, includes flowers with wide, bell-shaped corollas, dull coloration, musty odors, and copious pollen, which are well-exposed relative to the rest of the plant's foliage (Vogel, 1958; von Helversen, 1993; Fleming et al., 2009). The consistent appearance of many of these traits in association with bat pollination strongly suggests an adaptive significance, and in some cases, experimental work confirms this assumption. For instance, wide flowers help to guide bats during visits to ensure consistent pollen placement on specific regions of their heads (Muchhala, 2007), and strong sulfuric scents help to attract bats (von Helversen et al., 2000). Additionally, experiments suggest that copious pollen production is selected for due to the fact that bat fur can carry more pollen than feathers or insect bodies, leading to male–male competition, which favors increased pollen production per flower (Muchhala & Thomson, 2010). However, the adaptive significance of well-exposed flowers remains obscure.

Chiropterophilous plants achieve this increased exposure through various methods. Some woody plants position their flowers on the main trunk or branches (termed ‘cauliflory’), as seen in Crescentia cujete (Diniz et al., 2019). Many epiphytes and lianas hang their flowers below the foliage on long rope-like stems (‘flagelliflory’), such as Mucuna holtonii (von Helversen & von Helversen, 1999) and Weberocereus tunilla (Tschapka et al., 1999). And bat-pollinated herbs and shrubs frequently position their flowers above their foliage (‘styliflory’) via a tall central flowering stalk, such as Aphelandra acanthus (Muchhala et al., 2009) and Werauhia gladioliflora (Tschapka & von Helversen, 2007), or via long floral stems (pedicels), such as Adenocalymma dichilum (Domingos-Melo et al., 2023) and Burmeistera borjensis (Muchhala, 2006). A phylogenetic comparative analysis across c. 24 shifts between bat and hummingbird pollination in the centropogonid clade (Burmeistera, Siphocampylus, and Centropogon) of Campanulaceae demonstrates how consistent this difference in floral exposure is; stem length predicts pollination syndrome closely, with stems of bat-adapted flowers averaging 70% longer than those of hummingbird-adapted ones (Lagomarsino et al., 2017).

What might select for increased exposure in bat-adapted flowers? One possibility involves flight kinematics, in that bats move their wings forward in wide arcs around and in front of their bodies during hovering flight, while hummingbirds keep their wings to the sides and behind the body (Baker, 1961; von Helversen, 1993). A second possibility is that this represents selection to reduce predation risk for the bats, as snakes or predatory mammals (see Hopkins & Hopkins, 1982) would not be able to hide close to the flowers. A final possibility is that the increased exposure serves to increase detection by foraging bats. While these three hypotheses are not mutually exclusive, only the third makes the testable prediction that bats will take longer to find flowers that are not as well exposed.

The hypothesis that increased floral exposure evolves to maximize detection becomes even more plausible considering that New World nectar-feeding bats rely heavily on echolocation to find their flowers (Gonzalez-Terrazas et al., 2016a,b). For an echolocating animal, background clutter echoes can readily mask echoes of target objects (Schnitzler et al., 2003). While insectivorous bats that forage in narrow spaces can overcome this problem by relying on cues from the movement of their prey items or sounds they produce (Arlettaz et al., 2001; Denzinger et al., 2018), nectar-feeding bats need to be able to locate an immobile target. Thus, we would predict that the more plants can separate flowers from clutter echoes, the greater the chance of bats finding these flowers. In support of this idea, it was found that greater amounts of obstruction around Burmeistera flowers significantly decreased bat pollination but had no effect on hummingbird pollination of the same flowers (Muchhala, 2003).

In the present study, we experimentally test the hypothesis that increased exposure will decrease foraging times. We present wild-caught nectar bats with short or long-stemmed flowers in flight cages and time how long it takes them to find the flower. To overcome their well-developed spatial memory (Thiele & Winter, 2005; Carter et al., 2010), we constantly rotate flower position randomly between trials, ensuring each involves a new search. We repeat experiments in simple backgrounds, lacking clutter echoes from foliage, and complex backgrounds, where flowers are presented surrounded by branches and leaves, to determine whether stem length itself influences search time or whether it interacts with background clutter.

This study was conducted from 8 to 24 June 2019, in the private reserve Zingara (3.540°N, 76.605°W), in the Valle de Cauca Department of Colombia. Zingara forms part of the Key Biodiversity Area (KBA) Bosque de San Antonio, and consists of 3.15 ha of cloud forest at 1800–2000 m elevation. To capture nectar-feeding bats, each night we placed four to eight mist nets (of varying lengths, from 2 to 12 m) along potential bat flyways and in front of Burmeistera flowers, which are known to be pollinated by bats (Muchhala & Potts, 2007). All bats were identified, weighed, and immediately released except for individuals of Anoura caudifer (Geoffroy Saint-Hilaire 1818). We also captured several Glossophaga soricina (Pallas, 1766), another nectar-feeding bat, but opted to focus our experiments on the A. caudifer because of its abundance. These were then placed individually in one of three screen tents (3 m² × 2 m high) set up in a field next to the research station, and for the first night, were allowed to habituate to the cages and feed ad libitum from 20% sugar water solutions. Sugar water was placed in 50 ml polypropylene centrifuge tubes, which were affixed to poles in the cage using plastic-covered wire. Bats that did not learn to hover-feed from these tubes within 3 h after capture were released to minimize the risk of starvation (nectar bats have high nightly energy requirements, and need to visit an estimated 100 flowers per night to meet them; Voigt et al., 2006); those that did feed were held for another 2 d for experimental trials.

For the foraging experiments, a single flower was presented to the bat per trial in order to record the time until feeding. We used freshly collected flowers of Burmeistera succulenta H. Karst & Triana (Supporting Information Fig. S1), surrounding the stem (trimmed to c. 2 cm length) in cotton and placing it in a 1.5 ml microcentrifuge tube. Caps were removed from the tubes, and they were filled with water and then covered with a piece of duct tape to hold the flower in place. Plastic-covered wire was then used to fashion new ‘stems’ of two different sizes: 10 or 20 cm. The stem length of Burmeistera flowers can vary from 2.2 to 14.5 cm (N. Muchhala, unpublished); we chose these lengths to maximize differences in our experiment. One end of the wire was spiraled around the base of the centrifuge tube, but not affixed with tape, allowing the same flower to be easily switched between long and short ‘stems’. The other end was spiraled around one of four wood poles set up in the screen tent, such that the flower was positioned either 10 or 20 cm away from the pole. The four 1.5-m-tall poles were arranged in a square pattern, 1 m away from each other. Background clutter was manipulated to present the flowers in two treatments: simple, with no vegetation added to the poles, and complex, with leaves arranged around the tops of the poles (Fig. S2). Specifically, for complex, we placed two fern leaves along the middle of each pole and three Melastomataceae branches (Pleroma heteromallum) at the top (two pointed outwards in a V shape and one pointed upwards; see Fig. 1).

For each of 10 A. caudifer individuals, experimental trials were run in simple backgrounds for the first day and complex backgrounds on the second day. We randomly selected long or short stems for the first trial, and then alternated between these for a total of 20 trials. We noticed that variation in visit time was very large for the initial trials, likely as bats were getting accustomed to the flight cage and the experimental array; thus, we opted to treat the first 10 trials of each day as habituation time and only use the last 10 trials for experimental analyses. For each trial, we refilled the flower with 20% sugar water using a syringe and then randomly selected one of the poles (using a die) to place it on. After the initial trial, we selected one of the three poles that had not been used in the previous trial to avoid repeating the position. We then rolled the die again to randomly select horizontal orientation angle (where 1 = 30°, 2 = 60°, … 6 = 360°) and affixed the wire ‘stem’ to the pole. We positioned the stem roughly parallel to the ground (at the predetermined horizontal angle relative to the screen tent's door), with the tube and flower held upwards at a roughly 75° angle relative to the horizon (mimicking the natural positioning of Burmeistera stems and flowers). Once the flower was set up, we began recording the timing of events, noting the beginning of the trial and each time the bat started or stopped flying (i.e. by perching on the sides or top of the tent). One experimenter tracked bat activity with a headlamp, while the second experimenter recorded events, using a stopwatch cellphone app to take times. We ended the trial when the bat visited the flower (Video S1). At any point in the experiment if the bat remained perched for > 5 min, we gently tapped the screen next to the bat to encourage further flight. Trials in simple and complex backgrounds followed the same procedure, except that in complex backgrounds we also rotated all four poles clockwise by 30° between trials to further reduce bats' reliance on spatial memory while foraging.

For our statistical analyses, we used the duration of the last flight in each trial (from perching until visiting the flower) as the response variable. We used a generalized linear mixed model (GLMM; Bolker et al., 2009) to test the influence of stem length, background type, and their interaction on time until visitation in this last flight. We used the lme4 (Bates et al., 2015) and glmmtmb (Brooks et al., 2017) packages to compare the fit of negative binomial and Poisson error distributions; both AIC and likelihood-ratio tests found the former a better fit. We checked model diagnostics using the packages dharma (Hartig, 2022) and performance (Lüdecke et al., 2021); both found no deviation from the expected distribution of residuals, overdispersion, or significant outlier effects. Thus, for our final analysis, we specified a negative binomial error distribution and a log link function for the model, with stem length (long vs short), background type (simple vs complex), and their interaction as fixed effects, and individual bat identity as a random factor. The trial sequence number was also used as a random factor to control for the fact that bats may improve at the task of finding flowers as the experiment progresses. We calculated 95% confidence intervals around the model estimates using the tidy function from the broom.mixed package (Bolker & Robinson, 2022). As the interaction between background type and stem length was significant, we also performed post hoc contrasts to assess differences between the four combinations of stem length and background type using the Bonferroni P-value correction for multiple comparisons in the package emmeans (Lenth, 2024). We performed analyses using the R statistical software v.4.2.2 (R Development Core Team, 2022) and provided the raw data (Dataset S1) and an annotated R script used for these analyses (Notes S1).

Our initial GLMM analysis showed that stem length had a small but significant effect on the response, in that long stems led to decreased foraging times, while background type showed no effect (see Additive Model, Table 1). A follow-up GLMM that included both factors and an interaction effect (see Interactive Model, Table 1) detected a significant effect of the interaction between stem length and background type. Specifically, there was no clear difference in foraging time for short vs long stems in simple backgrounds (23.1 ± 23.3 SD vs 24.1 ± 22.1 SD; Fig. 2), but nearly double foraging time for short vs long stems in complex backgrounds (33.4 ± 29.8 SD vs 18.1 ± 18.17 SD). Post hoc contrasts are concordant with these results, showing that foraging time in simple vs complex backgrounds are not significantly different for short stems (Z = −2.10, P = 0.14) or for long stems (Z = 1.61, P = 0.42), and that foraging time for short vs long stems is not significantly different in simple backgrounds (Z = −0.28, P = 1.00), but it is significantly different in complex backgrounds (Z = 3.44, P = 0.0024).

The nectar-feeding bats in our flight cage experiments took more time to locate flowers with short floral stems (pedicels) than those with long stems. This demonstrates the importance of well-exposed flowers in decreasing bat foraging times. The fact that short stems led to nearly double the foraging time in complex backgrounds cluttered with vegetation, but have no detectable effect in simple backgrounds without vegetation, demonstrates that it is not stem length per se that is driving this pattern. Rather, long stems only reduce foraging time in cluttered backgrounds, supporting the idea that stem length aids detection because it separates flowers from background foliage.

Our experimental results support an adaptive hypothesis for why bat-pollinated flowers tend to be so well-exposed beyond the rest of the foliage. By reducing bat foraging times, increased exposure directly benefits these flower's mutualists. In turn, better-exposed flowers will be located more quickly and thus likely receive more visits through anthesis, leading to increased pollen export and receipt. And perhaps most importantly, the better exposed that a flower is, the greater the chance it is found at all by pollinators; flowers that remain undetected will, of course, fail to reproduce. The importance of effective cues to maximize detection can also be seen in experiments by von Helversen & von Helversen (1999), which showed that the curved upper petal of M. holtonii flowers serves as an acoustic guide for nectar bats and that removal of this petal decreases visitation from 88% to only 21%.

We suggest that the main reason long stems aid bats in finding flowers, while they do not seem to be as important for hummingbird or insect pollination, has to do with bat reliance on echolocation while foraging. Ears cannot localize the source of sound waves as precisely as eyes can localize the source of light waves, as each ear has a single eardrum, while each eye has multiple photoreceptors, which together provide a two-dimensional image. Thus, any other sources of echoes close to a focal object will obscure it, creating unwanted ‘clutter echoes’ (sensu Schnitzler et al., 2003), which overlap with target echoes. When foraging in clutter, insectivorous bats have been found to shift to increased reliance on vision (Eklöf et al., 2002), and nectar bats will shift to greater reliance on scent (Muchhala & Serrano, 2015), in line with the idea that echolocation becomes a less reliable sensory modality in such situations. Interestingly, rather than moving the flower away from its vegetative parts, another evolutionary approach to reduce clutter echoes found in bat-pollinated cacti is the evolution of wooly hairs around the flowers that absorb ultrasound, thus making the floral echoes more apparent (Simon et al., 2023).

It is possible that other sensory modalities, such as olfaction or vision, may contribute to the observed decrease in foraging time for long stems. For instance, flowers of B. succulenta emit a musty odor; it is possible that greater flower exposure enhances odor plumes (sensu Vickers et al., 2001), via interactions with air movement through foliage, such that long stems aid in emitting odor cues. Additionally, we note that our experiments were conducted in relatively well-lit conditions, due to our headlamps, moonlight, and light pollution from streetlamps; thus, the bats may also have been relying on vision while foraging. It would be illuminating to repeat the experiments in different light levels, from the equivalent light of a full moon to zero light (using infrared cameras to document bat behavior). We predict that the patterns we documented would be even more exaggerated in low-light conditions, such as during a new moon and/or deep in the forest understory, where bats would need to rely solely on echolocation rather than vision. Thus, the benefit of long stems may be due to some combination of enhanced visual, olfactory, or echo cues, which further experiments could help to tease apart.

One potential confounding variable in our experimental design was that we always ran simple-background trials on the first day and complex–background trials on the second day of our experiments. In fact, previous work does suggest that nectar bats are able to learn and improve performance in tasks through time (Muchhala & Serrano, 2015). However, any such learning would be expected to lead to bats finding flowers more quickly; the fact that they took longer to find the flowers in complex backgrounds on the second day suggests that the increased difficulty of the task outweighed any learning effects (and perhaps that the effect size would have been even more pronounced if the order were reversed).

In conclusion, among the three hypotheses outlined in the introduction for the evolution of well-exposed flowers, our results support the idea that exposure increases apparency, making the flowers easier to detect via echolocation. Results do not support the hypothesis that long stems evolved to accommodate bat wing kinematics; bats in our experiments were able to hover effectively in front of the flower for both our long and short-stemmed treatments. Results similarly fail to support the idea that long stems evolve to minimize predation risk, as this does not seem likely to explain why there is a difference in locating the flowers in complex backgrounds and not in simple backgrounds. While it might be argued that the longer time before visits to long-stemmed flowers was due to bats investigating and hovering in front of flowers longer due to perceiving a greater predation risk, anecdotally we did not notice such an effect. The time difference between long and short-stemmed flowers was due to time spent searching among the four poles, not to exploratory flights around the pole with the flower after it was located. Instead, our results suggest that by increasing flower exposure through long stems or other means, plants separate their flowers from background clutter echoes from their own foliage and surrounding foliage that could otherwise obscure the flowers for echolocating animals. We expect strong selection on this trait, as it reduces the amount of time bats need to find these flowers, and perhaps more importantly, increases the chances that a given flower will be found at all. Overall, this provides an adaptive explanation for the well-documented pattern of highly exposed flowers among bat-pollinated plants, as noted in classic descriptions of the chiropterophilous pollination syndrome.

None declared.

NM conceptualized the project, designed the experiments, and wrote the manuscript, JM-H aided in data analysis and interpretation. AZ aided in logistics and acquisition of data. All three coauthors contributed to experimental design and manuscript revisions.