Most plant-pollinator interactions involve the plant producing signals that manipulate the pollinator to bring it to the plant. These signals can work across many different sensory modalities, and may include color (Gumbert et al. 1999, Glover and Whitney 2010, Dyer et al. 2012), shape and pattern (Leonard and Papaj 2011, Lawson et al. 2017a), symmetry (Rodríguez et al. 2004, Krishna and Keasar 2018), scent (Miyake et al. 1998, Theis 2006, Raguso 2008, Lawson et al. 2018), temperature (Dyer et al. 2006, Whitney et al. 2008, Harrap et al. 2017, Rands and Harrap 2021), texture (Kevan and Lane 1985, Goyret and Raguso 2006, Whitney et al. 2009, Goyret 2010), electrostatic charge (Clarke et al. 2013, 2017). Recently, humidity has been suggested to be another sensory modality which could be used for signaling to pollinators. There is good evidence that flowers produce humidity that is elevated when compared to the immediate environment, having been demonstrated to occur in a number of flower species (Corbet et al. 1979, von Arx et al. 2012, Nordström et al. 2017, Harrap et al. 2020). Floral humidity is created by a combination of nectar evaporation and floral transpiration (Corbet et al. 1979, Azad et al. 2007, von Arx et al. 2012, Harrap et al. 2020) although the contribution of these two influences may vary between species, and appears to be different in its intensity from the humidity produced by nonfloral vegetation (Harrap and Rands 2022). A sample of the floral headspace of 42 species found that 30 (71%) produce floral humidity of an intensity greater than would be expected from any conflating environmental humidity sources (Harrap et al. 2020). The intensity of floral humidity produced by flowers, represented the average peak difference in relative humidity in the flower species' headspace, compared to the background, was close to zero in some species (such as Nicotiana tabacum L. [Solanales: Solanaceae]) but reached up to 3.71% (in Calystegia silvatica (Kit.) Griseb. [Solanales: Convolvulaceae]).

Floral humidity occurs widely and varies between species (Harrap et al. 2020), and does not appear to be limited to species visited by a particular group of pollinators (Harrap et al. 2020). Elevated floral humidity intensity has been observed in flowers pollinated primarily by moths (von Arx et al. 2012), flies (Nordström et al. 2017), and bees (Corbet et al. 1979). Sensitivity to environmental (nonfloral) humidity is well reported in insects (Havukkala and Kennedy 1984; McCall and Primack 1992; Kwon and Saeed 2003; Peat and Goulson 2005; Liu et al. 2007, 2014; Enjin 2017), and it is possible that these variations in floral humidity could function as a foraging cue for pollinators, especially since many different taxa of insect have been shown to be able to detect humidity differences similar to those seen between flowers and their surrounding environment (Harrap et al. 2021). Evidence for floral humidity acting as a signal currently comes from two pollinator species. The hawkmoth Hyles lineata Fabricius (Lepidoptera: Sphingidae) was demonstrated (von Arx et al. 2012, von Arx 2013) to show a preference to artificial flowers producing floral humidity comparable to that produced by Oenothera caespitosa Gillies ex Hook. & Arn. (Myrtales: Onagraceae) (which is naturally pollinated by H. lineata), and the bumblebee Bombus terrestris (L.) (Hymenoptera: Apidae) is capable of learning and differentiating between artificial flowers that mimic natural humidity patterns (Harrap et al. 2021). Other existing evidence of the capacity of pollinators to respond to floral humidity is limited (von Arx 2013), with nonexperimental observations that flies may use floral humidity in addition to other floral display traits produced within Indian alpine environments (Nordström et al. 2017). Overall, given that floral humidity is present in many flower species, and can potentially be detected by insect visitors, it is therefore likely that floral humidity will provide at least one component of the complex multisensory advertisement that a flower produces to attract its pollinating visitors (Raguso 2004, Leonard et al. 2011).

Signals that are more easily detectable due to contrast with the background will enhance the learning response and promote constancy (Spaethe et al. 2001, Dyer and Chittka 2004a, Heuschen et al. 2005, Chittka and Spaethe 2007). Pollination interactions exist against a background of noise due to the local climate (Wilson et al. 2015, Lawson et al. 2017b, Lawson and Rands 2019), which can reduce signal detectability and learning responses (Chittka et al. 1999, Dyer and Chittka 2004a, Kaczorowski et al. 2012, Lawson et al. 2017b). Environmental humidity is highly heterogenous (Lionello et al. 2006, Alarcón et al. 2008, Tichy and Kallina 2014, Enjin 2017) and is therefore likely to be a significant source of noise for floral humidity signals, but this effect has not yet been explored. If this is the case, environmental humidity may significantly affect the detectability of floral humidity, potentially constraining the contexts in which it can be used as a salient communication pathway between plants and their pollinators. In this study, we consider whether the ability of a pollinator to use floral humidity could be dependent upon the environment that the signal is presented within, which in turn tells us whether floral humidity is important regardless of environmental or weather conditions, or whether changes in the environment can impact on the effectiveness of communication using this sensory pathway.

In this study, we use differential conditioning techniques to explore the detectability of an elevated floral humidity signal comparable to natural flowers when presented to bumblebees Bombus terrestris audax (Harris) against different levels of signaling noise comparable with native conditions. This species is suitable for these experiments, as B. terrestris foragers have been used in similar differential conditioning experiments to those described here, showing a capacity to detect floral humidity and associate nectar with both humid and dry artificial flowers (Harrap et al. 2021). If there is no effect of noise on detectability, making floral humidity a robust signal, we would expect that the bees' learning responses would not be affected by the environmental conditions.

Methods

This study used differential conditioning techniques to test the robustness of a constant floral humidity signal generated at 2% against ambient lab conditions against varying levels of background humidity. Signal detectability was determined by observing the learning response of foraging bees to a sucrose reward associated with relative humidity differences between humid and dry artificial flowers. Observations were made when each of the flower variants was associated with the reward, and the other was nonrewarding. Trials were repeated in Low, Ambient, Medium, and High relative atmospheric humidity conditions to alter the contrast of humid flowers with both the background and with dry flowers.

Bee Trials

Differential conditioning techniques were used to test for the ability of bees to learn the presence of a sucrose reward associated with both humid and dry flower variants in each of the environmental treatment groups as an indicator of floral humidity signal detectability. Bees came from three different nests, which were used on a rotating basis. Bees were assigned to one of the four environmental humidity treatments described above. Within each of the environments, each bee was also allocated to one of three test groups, which were 1) Control, 2) Humid rewards, and 3) Dry rewards. In the Control, all flowers were of the dry variant with dry sponges, meaning there was no humidity difference between rewarding and nonrewarding flowers. In the Humid group, the rewarding flowers were humid variants, and nonrewarding flowers were dry variants. In the Dry group, the rewarding flowers were dry flower variants, and the nonrewarding flowers were humid variants. The feeding wells of flowers held a 25 ml of 30% sucrose solution or of water depending on if they were rewarding or nonrewarding, respectively. A summary of the twelve total test groups is shown in Table 1. Twelve bees were used for each test group, and no bee was used more than once, meaning that 144 bees were used in total.

During a learning trial, individual foragers were released alone into the arena, where they would encounter four rewarding and four nonrewarding flowers. The humidity produced by the internal components of the flowers varied according to the test group that the bee was in, and variants were recognizable to the experimenter by the odd or even numbers on the sides. Once released into the flight arena, the bee was allowed to forage on any flower. Observations were made on two foraging behaviors: 1) ‘land’, defined as the bee making physical contact with the top of the flower surface at any location even if the bee continued flying, and 2) ‘feed’, defined as the bee extending its proboscis into the feeding well. Consequently, each visit could be categorized either as 1) ‘correct’ where a bee would land and feed from a rewarding flower, or land without feeding from a nonrewarding flower, and 2) ‘incorrect’ where a bee would land and feed from a nonrewarding flower or land without feeding from a rewarding flower. These classifications reflect those used in previous conditioning studies (e.g., Lawson et al. 2018), and take the perspective that optimal behavior for the bee is to maximize its rate of energetic gain by collecting the sugar solution when it detects that it is on a rewarding flower, and minimizes the time it spends on a flower that it is able to classify as unrewarding. We acknowledge here the assumption that it is necessary for the bee to be in the contact with the flower to be able to accurately identify it as being humid or dry, and note that some variant of this information may be detectable whilst the bee is in flight. During each foraging bout, each visit was observed and immediately noted by the experimenter and recorded as either correct or incorrect. At the end of a foraging bout the bee was allowed to return to the nest, and then released back into the arena at the start of the next bout. This was repeated until the bee had completed a total of 70 individual visits to flowers. After each visit, the flower that the bee had landed on was repositioned whether the bee had fed or not to prevent the bee from learning the locations of rewarding and nonrewarding flowers. This was done whilst the bee was still in the arena, to avoid short-term learning of position during a single foraging trip – this meant that the behavior of the bee was conducted with constant environmental disturbance. When a bee departed, the flower was removed by the experimenter through one of the six doors. If the bee had fed from the feeding well, the well was topped up with a fresh 25 ml drop of sucrose or water as appropriate. The flower was replaced at a different location in the arena using the doors while the bee was visiting another flower. Any return visits made before a flower had been repositioned and refilled by the experimenter were not counted.

Table 1.

Summary of the treatment conditions in each environment and test group

At the end of each foraging bout, all flowers were removed and cleaned as described previously before being returned before the start of the next bout, and all feeding wells cleaned and refilled as described previously. Atmospheric relative humidity conditions were also managed at the end of each bout as described previously. A foraging bout would only begin once all flowers were cleaned and refilled, and the arena humidity was within the defined parameters.

Results

The number of correct responses made during the ten-visit blocks was dependent upon both environmental treatment, rewarding flower type, and the time interval, with significant three-way interactions (Table 2). In all of the environments, the bees faced with control flowers (where both rewarded and unrewarded flowers were dry, the blue solid lines in Fig. 1, and see also the ‘control’ panel in Supp  Fig. 1 [online only] (nvac049_suppl_supplementary_material.pdf)) remained close to the ‘random’ choice line (the thick grey dotted line on the panels of Fig. 1), suggesting that they were unable to discern the reward before landing and probing the feeding wells.

Table 2.

Summary of fit of fixed effects, for a linear three-way mixed-effects model of how bee flower choice behavior changes over time. Details of contrasts are given in the supplementary information (Supp  Tables 1 (nvac049_suppl_supplementary_material.pdf),  2 (nvac049_suppl_supplementary_material.pdf) and Supp  Fig. 1 [online only] (nvac049_suppl_supplementary_material.pdf)). Bold p values occur where p < 0.05

Fig. 1.

Mean correct visits (±SE) made by Bombus terrestris in a ten-visit block, for bees presented with a) low, b) ambient, c) medium, or d) high humidity environments. Blue solid lines represent the control treatment, orange dashed lines represent the rewarding humid flower treatment, and black dotted lines represent the rewarding dry flower treatment. The grey, dashed horizontal lines at 5 visits presents the mean number of correct visits that would be expected if the bees were randomly choosing which flower they visited (See online version for colored lines).

When the bees were challenged with having to differentiate between humid and dry flowers, they demonstrated that they were able to learn the rewarding flower type in most of the environments, shown by the increase in correct choices over time when either the dry (black dotted line) or humid (orange dashed line) flowers were rewarding, with both flower types being learnt at similar rates, although the dry rewarding flowers appeared to be learnt faster in all the environments. However, in the high humidity environment, the bees appeared unable to learn that humid flowers were rewarding (Fig. 1d), whilst in the medium humidity environment (with humidity levels above that experienced by the bees outside of the experimental period), the bees appeared to learn that humid flowers could be rewarding at a lower rate than they showed for dry flowers (Fig. 1c). This was confirmed by the significant interactions in Table 2, which mostly involved the high humidity environment (see also the high humidity panel in Supp  Fig. 1 [online only] (nvac049_suppl_supplementary_material.pdf)). Contrasts ( Supp Tables 1 and 2 [online only] (nvac049_suppl_supplementary_material.pdf)) suggest that the bees were unable to learn humid flowers as their ‘learning’ rate could not be distinguished from the behavior shown by control bees; in all other environments, dry or humid rewarding flowers were learnt at the same rate (Supp  Table 1 [online only] (nvac049_suppl_supplementary_material.pdf)). Bees appeared to learn dry rewarding flowers at similar rates regardless of environmental conditions (Table 2); they were slightly better at the task in medium and low humidity environments than in ambient and high humidity environments (Supp  Table 2 (nvac049_suppl_supplementary_material.pdf) and  Fig. 1 [online only] (nvac049_suppl_supplementary_material.pdf)), but given that the humidity in the ambient environment (35–45% RH) is lower than humidity in the medium environment (72.5% RH), this very small difference is most likely to be due to chance.

Discussion

Signals produced by plants to attract pollinators may be vulnerable to being masked by background noise created by climatic heterogeneity (Candolin 2003, Bradbury and Vehrencamp 2011). If floral humidity signals produced by flowers (von Arx et al. 2012, Harrap et al. 2020) are not robust to environmental humidity fluctuations, then their use in pollination interactions may be limited. This study explored the detectability to bumblebees of a floral humidity signal generated at 2% contrast to ambient conditions when presented against varying levels of background noise in a series of differential conditioning experiments, mimicking the levels of floral humidity that would be experienced by a visiting bumblebee. If floral humidity is robust to environmental fluctuations, it was expected that there would be no effect of increasing noise on the learning response in bees. We found that our artificial humidity signal was detectable and robust to noise as learning was observed across all environments, to varying degrees. Learning was faster when elevated floral humidity was presented as a nonrewarding stimulus, or at greater contrast to the background in an environment with low humidity. The results also show that the capacity of elevated floral humidity that could be used as a positive stimulus by bees could be constrained by background humidity; lower environmental humidity led to learning occurring, while at high humidity there was no response.

We found that in high humidity environments, the bees had trouble learning that flowers presenting humid signals were rewarding. It would be tempting to explain this by suggesting that the high humidity environment may have enhanced the ‘wetness’ of the dry flowers, making it difficult to differentiate between the dry and humid forms. However, this is unlikely given that the bees were able to learn about dry rewarding flowers when in the humid environment. Similarly, the low humidity environments appeared to have no impact on bees being able to differentiate between dry and humid flowers, suggesting that any masking that might have been caused by interactions between the floral humidity and the environment was minimal. In the context of high humidity or recent rainfall pollinators may still use floral humidity to accurately determine profitability, but process humidity as a negative stimulus as has been demonstrated by pollinators switching to flowers with concentrated nectar that would be relatively drier (Willmer 1986, Raguso and Willis 2005, Contreras et al. 2013). In this case elevated floral humidity would be a by-product of environmental interference of rewards and no longer be a floral signal. Our observations in high humidity support this view as bees used humidity differences to determine rewards, but were unable to learn a positive association to humid variants. Observations by Harrap et al. (2020, 2021) that floral humidity is produced at varying and sometimes negligible levels, and that bumblebees learnt to associate rewards with both dry and artificial flowers led to the suggestion that floral humidity may in some cases be used instead to confer the identity of a flower to pollinators instead of reward status. The fact that bees learned to associate humidity with both nonrewarding and rewarding flowers supports the interpretation that in some cases floral humidity may be used to confer identity instead of reward status (Harrap et al. 2021), although this is less likely to be the case in extreme conditions, and more likely in conditions of intermediate humidity where osmotic demands are not strongly weighted in either direction. An alternative interpretation may be that bees can be conditioned to learn floral humidity as either a positive or nonrewarding stimulus in ambient conditions simply because they hold the capacity for either response as is required to respond appropriately to humid flowers in varying climatic conditions. In summary, while floral humidity clearly holds capacity to be a useful component in the multimodal display, the relative value of its presence or absence to plants as a signaling strategy depends on the background humidity at a given time.

For a signal to be detectable it must be generated at a level of contrast against the background which meets or exceeds the minimum response threshold of the receiver's sensory channels, at which point a response is induced (Wiley 2006). Masking occurs when background noise is elevated to an intensity at which the reduction in contrast between a signal and the background means signals are no longer distinguishable, but a robust signal persists amidst noise. Robustness and the noise required for masking are therefore determined by the investment in signal generation by the sender and in sophisticated sensory equipment by the receiver (Wiley 2006). The current study shows that a floral humidity signal produced at 2% contrast to ambient conditions was robust as it was detectable by foraging bees when background noise was elevated to both 72.5–79.5% and 92.5–97.5% RH, where it is expected that signaling contrast was reduced due to the lowering of the osmotic gradient between humid flowers and the surrounding air preventing evaporation (Corbet 1990, Azad et al. 2007, von Arx et al. 2012). The results indicate that despite the reduction in the osmotic gradient, the effect of the environment was not sufficient to completely eliminate the signal from humid artificial flowers, which must have still been producing a floral RH signature elevated above background conditions in the Medium and Humid environments to be detected. However, the actual RH value of humid flowers was not measured after the initial setup, so the exact extent to which signal contrast was reduced below 2%, if at all, cannot be identified. The robustness of floral humidity will also be partly determined by the sensitivity of the insect hygroreceptor system. In honeybees Apis mellifera L. (Hymenoptera: Apidae), firing activity of dry and moist cells is increased by 1 impulse s^–1 by instantaneous humidity changes of –2.5% RH and +3.3% RH respectively (Tichy and Kallina 2014). Corresponding observations in cockroaches Periplaneta americana (L.) (Blattodea: Blattidae) were –1.6% RH and + 2% RH, and were –3.3% RH and +2.5% RH in stick insect Carausius morosus (De Sinéty) (Phasmida: Phasmatidae) (Tichy and Kallina 2014). Sensitivity to the rate of change of humidity is much greater. Changes of 1 impulse s^–1 were induced in honeybees by altering the rate of humidity change by –0.18% RH s^–1 and +0.19% RH s^–1 (Tichy and Kallina 2014). Corresponding observations in cockroaches were –0.2% RH s^–1 and +0.1% RH s^–1, and were –0.4% RH s^–1 and +0.4 RH s^–1 in stick insects (Tichy and Kallina 2014). Furthermore, observations of low sensory resolution in dry or ambient conditions and heightened sensitivity of responses as conditions approach humid extremes have been recorded in the case of mosquitoes Culex quinquefasciatus Say (Diptera: Culicidae), mealworm beetles Tenebrio molitor L. (Coleoptera: Tenebrionidae) and human lice Pediculus humanus L. (Psocodea: Pediculidae) (Thomson 1938, Pielou and Gunn 1940, Wigglesworth 1941). A response to instantaneous humidity appears unlikely to explain the detectability below 2% unless sensitivities in bumblebees are also heightened with increasing environmental humidity. However, greater resolution to qualitative change suggests that bees could have responded to even a heavily reduced rate of change of humidity as they entered the floral headspace of humid flowers, encountering a gradient of increasing intensity with proximity to the surface (Harrap et al. 2020).

Where contrast is increased against a background due to a decrease in noise or intensity of production, detectability is enhanced (Lunau et al. 1996) which increases the likelihood of a response in the receiver (Wiley 2006). When background noise was reduced to 20–25% RH, a steeper osmotic gradient would have been established between floral humidity and the background, enhancing evaporative effects and providing a steeper contrast against the resulting signal. The observations of faster learning responses in both groups in this environment support previous studies that emphasize the role of contrasting floral signals in promoting learning and floral constancy. Bumblebees took longer to make decisions and identify rewarding flowers amongst nonrewarding alternatives when presented at low contrast (Spaethe et al. 2001, Chittka and Spaethe 2007, Dyer et al. 2008), and Goulson (2000) reported that foraging bumblebees took twice as long to locate flowers that were presented alongside similarly colored alternatives than when alone. Furthermore, increasing contrast through color distance was shown to enhance the speed of developing constancy and the accuracy of decisions (Spaethe et al. 2001, Dyer and Chittka 2004b). Reduced contrast of olfactory signals had detrimental effects of learning time of bumblebees (Lawson et al. 2017a) and reduced efficiency of reward location in hawkmoths Manduca sexta (L.) (Lepidoptera: Sphingidae) (Cardé and Willis 2008). The observations presented here may also be partly understood by considering the variable sensitivity of hygroreceptor cells, where experiments on the honey demonstrated greater sensitivity when amplitudes of humidity oscillation were increased (Yokohari et al. 1982, Tichy and Kallina 2014). An increased signal contrast may have induced a similar shift in hygroreceptor sensitivity, promoting a faster neural response in low conditions than in other environments. In light of the literature presented, it is not surprising that elevating signal contrast in the current study promoted a faster onset of learning floral humidity differences, whether as a positive or negative association.

The local climate dictates the behavioral responses towards individual flowers and stimuli by regulating demands and preferences of foragers, and by changing floral reward status (Corbet 1990, Chown et al. 2011). The results of the current study show that environmental humidity affected the learning response of bees to rewards associated with humid flowers, accelerating learning in dry conditions but mitigating the response in high humidity. In the natural habitat, relative foraging preferences are likely determined by homeostatic water balances induced partly by environmental humidity (Prange 1996), with aridity promoting water deficit both of colonies and foragers (Willmer 1986, Human et al. 2006, Ellis et al. 2008). For example, bees initiated water collection trips (Willmer 1988) and hawkmoths switched to preferences for more dilute nectar rewards in arid conditions (Contreras et al. 2013). As osmotic demand becomes more weighted, the perceived value of floral humidity signals may increase (von Arx et al. 2012) and floral humidity may become a significant attractant (Willmer 1986). For example, the diluted nectar by the desert dwelling Echium wildpretii H.Pearson ex Hook.f. (Boraginales: Boraginaceae) is thought to function as a water reward to foraging honeybees (Olesen 1988). Conversely, in high humidity energetic demands become more weighted and preferences shift to more concentrated nectar sources (Raguso and Willis 2005). Mason bees switched to concentrated nectar sources in higher humidity (Willmer 1986). In extreme conditions, high water burden may reduce the perceived profitability of humid flowers further still (Bertsch 1984, Willmer 2011). In the natural environment, climatic humidity also affects the reward status of nectar sources. Aridity lowers reward volume by reducing the secretion (Bertsch 1983, Nicolson 1993) and accelerating the evaporation of nectar, which may also reduce palatability by increasing viscosity (Corbet et al. 1979, Plowright 1987, Willmer 2011). Therefore, the perceived profitability of humid flowers is likely to increase in arid conditions. On the other hand, high humidity may cause excessive dilution of nectar which reduces energetic content (Corbet and Delfosse 1984, Cnaani et al. 2006, Willmer 2011). Dilution may be induced by physical contamination of nectar during rainfall, which is, notably, a likely context in which pollinators will experience high humidity extremes (Lawson and Rands 2019). As bee-pollinated flowers tend to hold higher sugar concentrations, diluted nectar is a likely deterrent indicating unfavorable rewards and so the perceived profitability of humid flowers will decrease (Willmer 2011). In both instances, the mechanistic link between nectar and humidity (von Arx 2013, Harrap and Rands 2022) may provide the means by which transient profitability status is assessed. The effects of climate on pollinator preferences and reward status may mean that whilst in arid conditions a floral humidity signal encourages visitation and enhances learning responses towards humid flowers, the same signal in a humid environment may discourage visitation and inhibit floral humidity being learnt as a positive stimulus. Our results appear to mimic this response which is surprising because the bees were only briefly exposed to different humidity conditions when entering the arena to forage. It is unlikely they were exposed long enough for significant shifts in individual osmotic demands to develop before returning to the colony, which itself remained in ambient conditions so would not have contributed to the response. Furthermore, the reward status of artificial flowers was controlled and nectar rewards were not exposed to conditions long enough for equilibration to occur as in natural flowers. Our observations may therefore reflect an innate shift in the perceived profitability of elevated floral humidity induced by the environment which was induced in bees despite no change in actual profitability of humid flowers. In the natural environment, such a shift in innate preference may be useful for quickly avoiding unprofitable flowers and maintaining osmotic balance as conditions change without incurring the time costs associated with learning. It would be interesting to explore how bees and other pollinators responded to natural differences in environmental humidity in the wild, as the natural responses of pollinators may well be dependent upon conditions which could be vastly altered by changes in rainfall and temperature induced by climate change.