Testing the Distraction Hypothesis: Do extrafloral nectaries reduce ant‐pollinator conflict?

Abstract Ant guards protect plants from herbivores, but can also hinder pollination by damaging reproductive structures and/or repelling pollinators. Natural selection should favour the evolution of plant traits that deter ants from visiting flowers during anthesis, without waiving their defensive services. The Distraction Hypothesis posits that rewarding ants with extrafloral nectar could reduce their visitation of flowers, reducing ant‐pollinator conflict while retaining protection of other structures. We characterised the proportion of flowers occupied by ants and the number of ants per flower in a Mexican ant‐plant, Turnera velutina. We clogged extrafloral nectaries on field plants and observed the effects on patrolling ants, pollinators and ants inside flowers, and quantified the effects on plant fitness. Based on the Distraction Hypothesis, we predicted that preventing extrafloral nectar secretion should result in fewer ants active at extrafloral nectaries, more ants inside flowers and a higher proportion of flowers occupied by ants, leading to ant‐pollinator conflict, with reduced pollinator visitation and reduced plant fitness. Overall ant activity inside flowers was low. Preventing extrafloral nectar secretion through clogging reduced the number of ants patrolling extrafloral nectaries, significantly increased the proportion of flowers occupied by ants from 6.1% to 9.7%, and reduced plant reproductive output through a 12% increase in the probability of fruit abortion. No change in the numbers of ants or pollinators inside flowers was observed. This is the first support for the Distraction Hypothesis obtained under field conditions, showing ecological and plant fitness benefits of the distracting function of extrafloral nectar during anthesis. Synthesis. Our study provides the first field experimental support for the Distraction Hypothesis, suggesting that extrafloral nectaries located close to flowers may bribe ants away from reproductive structures during the crucial pollination period, reducing the probability of ant occupation of flowers, reducing ant‐pollinator conflict and increasing plant reproductive success.

Most ant-plants are angiosperms (Keeler, 2014), and many require the services of animal pollen vectors for seed set (Ballantyne & Willmer, 2012, Dutton et al., 2016, Villamil, Boege, & Stone, 2018, Bentley, 1977b, Torres-Hernández, & Rico-Gray, 2000, Díaz-Castelazo, Rico-Gray, Ortega, & Angeles, 2005, Rico-Gray & Oliveira, 2007, Raine, Willmer, & Stone, 2002, among many other studies documenting animal-pollinated ant-plants), making ants and pollinators likely to co-occur on a given host plant. This raises the possibility of several types of direct and/or indirect conflicts between ants and pollinators. First, an indirect conflict can arise if there is a trade-off between plant allocation of resources to reproduction (which benefits pollinators) versus investment in growth and defence (which benefits ant guards) (Bazzaz, Chiariello, Coley, & Pitelka, 1987). Plants that do not reproduce grow faster and develop larger resourceacquiring and producing organs (roots and leaves) (Frederickson, 2009), leading to indirect conflict between ants and pollinators over plant resources and rewards (Afkhami, Rudgers, & Stachowicz, 2014;Dutton et al., 2016). In extreme cases of ant-pollinator conflict, ants actively increase plant investment towards growth and defence by castrating their host plant through consumption of floral meristems (Frederickson, 2009;Palmer et al., 2010) or mature inflorescences (Izzo & Vasconcelos, 2002). Second, ants may enter flowers and consume floral nectar without providing pollination services, providing no benefit to the plant and potentially reducing the attractiveness of flowers to effective pollinators (Rico-Gray & Oliveira, 2007). Third, ant visits to flowers may reduce pollen viability by depositing antimicrobial substances that decrease pollen germination rates, and hence decrease male fitness for the plant (Dutton & Frederickson, 2012;Wagner, 2000). Finally, ants may attack or intimidate pollinators directly (Villamil et al., 2018;Wagner & Kay, 2002;Willmer et al., 2009), reducing flower visitation rates (Lach, 2008;Ness, 2006) or duration (Villamil et al., 2018). One hundred and forty years ago, Anton Joseph Kerner, an Austro-Hungarian botanist, wrote: Of all the wingless insects it is the widely dispersed ants that are most unwelcome guests to flowers. And yet are they the very ones which have the greatest longing for the nectar, as numberless observations sufficiently show. (Kerner, 1878, p. 21) While ants may be unbidden floral visitors, they are also effective bodyguards (Bentley, 1977a;Chamberlain & Holland, 2009;Rosumek et al., 2009;Trager et al., 2010), which may represent an ecological trade-off for ant-plants. Given that ant guards can have both costs and benefits for different aspects of plant fitness, we expect natural selection to act on ant-plant traits to minimise the negative impacts of ants relative to the protection they provide, ameliorating the negative consequences of ant-pollinator antagonism for plant fitness (Raine et al., 2002). A wide range of mechanisms have been interpreted as achieving this by reducing ant visitation to flowers during anthesis, including physical barriers (Carlson & Harms, 2007;Galen, 1999;Galen & Cuba, 2001;Harley, 1991;Raine et al., 2002;Willmer, 2011), chemical repellents (Agarwal & Rastogi, 2008;Ballantyne & Willmer, 2012;Junker & Blüthgen, 2008;Junker, Chung, & Blüthgen, 2007;Willmer et al., 2009;Willmer & Stone, 1997) or bribes (Kerner, 1878;Martínez-Bauer, Martínez, Murphy, & Burd, 2015;Willmer, 2011). Physical barriers include spiny or hairy surfaces on the outside of the corolla or on floral pedicels that prevent tarsi from gripping and so hinder ant walking (Willmer, 2011), and waxy or sticky plant secretions that prevent ants from climbing (Harley, 1991). Bracts around the calyx can act as a water trap, creating a pool of water or mucilage that prevents ants and other small insects from crawling into the flowers (Carlson & Harms, 2007). The shape of the flower may itself stop ants from entering flowers: pendant, thin and constricted corollas are effective ant-excluding morphologies (Galen, 1999;Galen & Cuba, 2001;Willmer et al., 2009).
Finally, it is possible that high EFN secretion on flowering shoots in myrmecophiles could fulfil both distracting and protective roles, simultaneously keeping ants out of flowers but promoting their patrolling around reproductive tissues to deter herbivores.
While the defensive role of ant recruitment through EFN secretion has been widely demonstrated (Chamberlain & Holland, 2009;Rosumek et al., 2009;Trager et al., 2010), the Distraction Hypothesis has not been adequately tested. To our knowledge, since Kerner proposed it in 1878, only three experimental studies have been performed and all have rejected it (Chamberlain & Holland, 2008;Galen, 2005;Wagner & Kay, 2002). However, none of these studies were carried out in an ecologically realistic setting (a point that we address further in the Discussion).  (Cuautle et al., 2005). Flowers last 1 day, are insect-pollinated (Sosenski, Ramos, Domínguez, Boege, & Fornoni, 2016) and have a yellow, pentamerous, campanulate corolla with nectar easily accessible at the base. Honeybees (Apis mellifera) are the dominant pollinators at La Mancha, accounting for 94% of visits (Sosenski et al., 2016;Villamil et al., 2018).

| Surveys of ants inside flowers
We quantified ant occupancy in flowers of T. velutina by surveying 1,604 flowers across four sites within CICOLMA in November 2014. Flowers at each site were observed and instant counts were recorded every hour throughout the whole anthesis period (08:30-12:30 hr), with one observer at each site. We estimated the proportion of flowers occupied by ants, and the total number of ants across occupied flowers within a site. Flowers were sampled at the same site over multiple days, for 10 site-and-day combinations. Since these are 1-day flowers, we considered each site-day as a replicate (n = 10 site-days), with site-and-day effects incorporated into our statistical modelling (see below).

| Experimental manipulation of EFN secretion
To test the Distraction Hypothesis, we experimentally clogged extrafloral nectaries to prevent nectar secretion and compared ant and pollinator behaviours on paired shoots with and without EFN secretion. This experiment was conducted over 5 days during November 2014. Early on each day of the experiment, a pair of neighbouring, unopened floral buds within a plant were marked as either control or clogged treatments (n = 216 flowers; n = 108 pairs, n = 108 plants).
EFN secretion on clogged treatment leaves was eliminated by sealing the nectary cup with a droplet of transparent acrylic textile paint (Mylin dimensional, Mexico). On control treatment leaves, we applied similarly sized droplets of the same textile paint a couple of millimetres above the gland (Figure 1), controlling for any effects of the acrylic paint itself. Pilot tests confirmed that the paint totally prevented EFN secretion and also that the paint did not deter ants or pollinators. We recorded the frequency and identity of ants (to genera or species level following Zedillo-Avelleyra, 2017) and other insects visiting each flower pair and the associated extrafloral nectaries for 2 min every hour during anthesis (08:30-12:30 hr). Simultaneous observations were performed at each of three sites by different observers. For brevity, we refer to non-ant flower visitors as pollinators, while recognising that the efficacy of visits by all species mentioned in contributing to seed set in T. velutina remains to be demonstrated.
Based on the results from the clogging experiment described above (from now on referred to as the short-term experiment), we conducted a follow-up experiment in which treatment duration and spatial scale were both increased by a factor of 10, using paired branches and focusing on one flower on each control or clogged branch, rather than paired flowers on the same branch. We refer to this experiment from now on as the long-term experiment (see Supporting Information 1 for further details). The extrafloral nectaries of all 10 leaves on the clogged treatment branches were sealed as described above, and the treatment was maintained for 10 days (Figure 1). Our hypothesis was that increasing both the temporal and spatial scales of our treatment would result in a larger experimental effect size. However, a comparison of the results from the short-term and long-term experiments showed that ants respond at a smaller scale (Supporting Information 1: Table S2), and we therefore focus on the results of the short-term experiment and highlight differences in results for the longer term, larger scale experiment where these are relevant to the Distraction Hypothesis. Full results and details regarding the long-term experiment are provided in Supporting Information 1.

| Impacts of EFN secretion on fitness
To quantify the impact of clogging EFN secretion and the Distraction Hypothesis on plant fitness, we collected the fruits resulting from experimental flowers (control and clogged) at which pollinator visitation was observed. We recorded whether those flowers developed into fruits with seeds or whether they were aborted, and counted the number of seeds per fruit. All fruits were collected at least 1 week post-anthesis, at which stage retained fruits can be distinguished from aborted fruits, and developing seeds can be counted distinguishing viable from unviable seeds, even if still immature.

| Surveys of ants in flowers
To test if the proportion of flowers with ants inside them changed over the anthesis period, we fitted a binomial mixed model with time of day as a fixed effect. Flowers of T. velutina last for a single day, and because multiple flowers were sampled on a given site on a given day, we fitted site identity as a random effect to account for differences between site-and-day variation in variables that could influence ant abundance, such as resource availability, ant diversity, or the abundance of ant nests. Tukey post hoc comparisons were used to test differences between hours using the "multcomp" r package (Hothorn, Bretz, & Westfall, 2008).
To test if the number of ants inside occupied flowers changed over the anthesis period, we fitted a Poisson mixed model, using the number of ants inside flowers per site as the response variable and fitted as fixed effects time of day as a linear and as a quadratic term. The number of flowers occupied by ants was fitted as a logtransformed offset to control for ant density in flowers, which is likely to decrease in sites with more flowers occupied by ants, since we recorded counts per site rather than counts per individual flower (see fieldwork methods). Time of day was fitted as a linear and as a quadratic term to investigate the shape of the activity pattern of ants in flowers relationship between the number of ants inside flowers through the day. We fitted site identity as a random effect to account for variation that could influence ant abundance (as detailed above). We also included an observation-level random effect where each data point receives a unique level of a random effect to control for overdispersion (Hinde, 1982). Tukey post hoc comparisons were used to test differences between hours using the "multcomp" r package (Hothorn et al., 2008).

| Ecological consequences of EFN secretion
Five mixed effects models (i-v) were fitted to test the ecological consequences of the Distraction Hypothesis. Because all of these models had the same random effects structure unless otherwise specified, we detail the random effects first and then describe the fixed effects for each model. Flower identity was fitted as a random effect to account for repeated hourly observations. Because this experiment had a paired experimental design, we fitted flower pair identity as a random effect to control for between-pair variation in floral and extrafloral investment. We also included an observationlevel random effect where each data point received a unique level of a random effect to control for overdispersion. We fitted the following models, and have structured our results following the same order: (iv) To test the effect that the total number of ants had on pollinator visitation (regardless of their location in flowers or at extrafloral nectaries), we fitted a Poisson mixed model using number of pollinators as the response variable. As fixed effects we fitted the total number of ants, and treatment to test whether treatment affected pollinator visitation in a way that was unlinked to the number of ants.
(v) To test if the location (inside flowers or at extrafloral nectaries) and number of ants had an effect on pollinator visitation, we fitted a Poisson mixed model. The number of pollinators was fitted as the response variable, while treatment, number of ants in flowers, and number of ants at extrafloral nectaries were fitted as fixed effects.
Data from the long-term experiment were analysed following a similar model structure reported for the ecological consequences models (S.i-v, see Supporting Information).

| Impacts of EFN secretion on plant fitness
(vi) To test the effect of clogging on fruit abortion rates, we fitted a binomial mixed model, with clogging treatment as the fixed effect and pair identity as a random effect.
(vii) For those fruits that developed seeds, we tested the effect of clogging on the number of seeds by fitting a Poisson mixed model.
Clogging was fitted as a fixed effect and as random effects we fitted pair identity and an observation-level random effect to account for overdispersion.

| Exploring the responses and effects of different ant species
We fitted additional models aiming to explore differences between ant species in their response to clogging and in their effects on pol-

| Effect sizes
Cohen d effect sizes for all models were calculated using the likelihood ratio tests (LRT) statistics from each model. To test whether increasing the duration and scale of the clogging treatment by a factor of 10 had a larger effect on the number of ants and pollinators, we estimated the ratio of change in the effect size between the short-and long-term experiment for each type of visitor (See Supporting Information).

| Surveys of ants in flowers
We observed 10 ant species from four subfamilies interacting with T.  (Bolton, 1987;Ettershank, 1966). Feeding preferences also vary among ant species, from opportunistic carnivores such as Pseudomyrmex gracilis  Table 1). The number of ants inside flowers did not vary significantly through daily time and we found no statistical support for any quadratic effect.

| Floral visitors
All but one of 202 visits to T. velutina flowers by non-ant visitors were made by other insects (

| Ecological consequences of EFN removal
Numbers of patrolling ants were significantly affected by EFN treatment (clogged vs. control), ant location and the interaction between these factors ( Figure 3a; Table 1). Ten times more ants were found patrolling extrafloral nectaries (1.49 ± 0.079 ants) than were found inside flowers (0.14 ± 0.02 ants) (Figure 3a;   Figure 3a; Table 1). The percentage of flowers occupied by ants increased significantly from 6.1% under the control treatment to 9.7% when extrafloral nectaries were clogged ( Table 1).
Numbers of flower visitors were not significantly affected by the elimination of EFN secretion (Figure 3a; Table 1), nor was there any significant interaction between visitor numbers and the total number of ants (Table 1) (Table 1).
In all five models used to analyse the short-term experiment (one leaf, 1 day), differences between individual plants (captured by the pair random effect) explained the largest proportion of variation in the numbers of ants and pollinators (Table 1)

Random effects
Variance

| Impacts of EFN secretion on plant fitness
Clogging had a marginally significant effect (p = 0.059) on fruit abortion, increasing by 12% the probability of abortion in flowers associated with leaves in which EFN had been clogged ( Figure 4a, Table 1). Despite the p-value being marginally significant, clogging had a considerable Cohen d effect size (Cohen, 1988) on fruit abortion (Table 2). However, clogging had no effect on the number of seeds per fruit ( Figure 4b, Table 1) with a small Cohen d effect size between treatments (Cohen, 1988) (Table 2).

| Comparison of patterns across spatiotemporal scales
In contrast to our prediction, increasing the duration and spatial scale of the clogging treatment by a factor of 10 did not result in larger effect sizes on ant behaviours (Table S2). In fact, the longterm clogging experiment had less impact on ant patrolling than the short-term clogging experiment, resulting in smaller effect sizes on numbers of ants at extrafloral nectaries, ants inside flowers and on the proportion of flower occupancy by ants (Table S2). The impact of preventing EFN secretion on the number of ants inside flowers changed from positive at a short-term, local scale to negative in the long-term, branch-scale experiment (10 leaves, 10 days) (Table S2).

| Ant species-specific responses to clogging and effects on pollinators
Although ant species explained only 0.21% of the variation in the number of ants inside flowers (Model S.vi in Table S3), there was variation between ant species in responses to clogging ( Figure S3a).
Brachymyrmex sp. ants were the most abundant ants found inside flowers, as shown by the non-zero-overlapping effect ( Figure S3a, Table 3). While effect estimates vary for other ant species, confidence intervals for all taxa other than Brachymyrmex sp. overlap with zero (see Table 3 for rank order of abundance inside flowers and Figure S3a for likelihood of response to clogging; see Supporting Information for further details). Activity by individual ant taxa at extrafloral nectaries had very small effects on pollinator visitation, as shown by the small estimates (model S.vii, Table S4), although their effects were precisely estimated by our models, as indicated by narrow variation around these estimates ( Figure S3b). In contrast, the effects of activity by individual ant taxa inside flowers on pollinator visitation could not be precisely estimated from our data, as indicated by the large variation associated with these estimates (model S.viii, Figure S3c).

| The distraction hypothesis
Plants face a potential trade-off between the benefits they receive from ants patrolling their leaves and flowers and the costs F I G U R E 3 Mean numbers of visitors to flowers of Turnera velutina (mean ± 1 SE) recorded in hourly surveys during 2 min of observation per flower for the short-term experiment. Clogged treatment flowers had secretion of extrafloral nectar (EFN) prevented by clogging the associated extrafloral nectaries. The short-term experiment involved prevention of EFN secretion associated with one flower for 1 day (see Figure 1). Red circles represent ants at extrafloral nectaries; blue triangles represent ant in flowers, and green squares represent pollinators F I G U R E 4 Effects of clogging the extrafloral nectaries on (a) the number of seeds (mean ± SE) produced by Turnera velutina and (b) the probability of fruit abortion (mean ± SE) associated with this activity (Altshuler, 1999;Assunção, Torezan-Silingardi, & Del-Claro, 2014;Dutton et al., 2016). In T. velutina, the presence of the most aggressive ants inside flowers increases the likelihood of pollinators displaying alert behaviours and reduces the time honeybees spend inside the flowers (Villamil et al., 2018 (Wagner & Kay, 2002). These results differ from studies conducted on natural plants (Bentley, 1976;Shenoy, Radhika, Satish, & Borges, 2012;Villamil et al., 2013) and from our findings  (Figure 3), which show that increased EFN results in increased ant visitation. Furthermore, the plastic caps used by Wagner and Kay (2002) to simulate floral (primary) and extrafloral (additional) nectaries were morphologically identical and equally accessible, but neither assumption is met in natural EFN-bearing species (Escalante-Pérez & Heil, 2012;Keeler, 2014). Therefore, no robust conclusions about the Distraction Hypothesis can be drawn from this experimental design.

In 2005, Galen tested the Distraction Hypothesis on
Polemonium viscosum, a plant species without extrafloral nectaries.
Extrafloral nectaries were simulated by trimming the petals, anthers and pistils from some flowers, leaving only the calyx and toral disc that bears the floral nectaries to simulate extrafloral nectaries (Galen, 2005). Control inflorescences contained only intact flowers, while inflorescences with simulated extrafloral nectaries contained intact flowers plus trimmed flowers simulating extrafloral nectaries (Galen, 2005).  (Galen, 1999;Galen & Cuba, 2001). Furthermore, artificial damage (trimming) is a confounding factor because it triggers plant-induced defences (Ballaré, 2011;Heil, 2008;Heil, Koch et al., 2001;Ness, 2003) that strongly affect floral and extrafloral nectar secretion (Heil, 2011(Heil, , 2015Ness, 2003;Radhika, Kost, Bartram, Heil, & Boland, 2008  Our results support the Distraction Hypothesis with predictions 1, 3 and 5 being met. We found that clogging EFN secretion reduced the number of ants patrolling extrafloral nectaries by 30% (prediction 1), increased the likelihood of flower occupation by ants by 3.6% (prediction 3), and increased the likelihood of fruit abortion by 12% (prediction 5). However, we found no significant increase in the number of ants inside flowers (prediction 2), or reduction in pollinator visitation (prediction 4) when extrafloral nectaries were clogged (Tables 1 and 2). Support for prediction 3 (increased flower occupation by ants), and reduction in plant fitness through increased rates of fruit abortion (rather than damage to flowers; Figure 4) are both specific to the Distraction Hypothesis. We therefore conclude that our results represent the first experimental support for this hypothesis obtained under field conditions.

| Fitness consequences
The clogging treatment caused a 12% increase in the probability of fruit abortion, which is not linked to the visitation frequency as the number of pollinators was unchanged. We hypothesise this reduc- Ant patrolling may also affect the plant mating system, affecting the selfing/outcrossing rates, which may lead to fruit abortion due to selective abortion linked to pollen origin or inbreeding depression.
Plants can abort fruits with a higher proportion of selfed seeds, to increase resource allocation to fruits with a higher proportion of outcrossed seeds. Selective fruit abortion linked to pollen origin (selfing vs. outcrossing) has been observed in a wide array of plant species (Huth & Pellmyr, 2000;Marshall & Ellstrand, 1988;Niesenbaum, 1999;Stephenson, 1981).

| Exploring ant species-specific effects
We found 10 ant species interacting with T. velutina, representing a diverse mosaic of partners that may differ in their response to clogging and on their effects on pollinators. Evolution of plant mechanisms that reduce plant-pollinator conflict could be driven by interactions with one or more of these species. We expect plants to evolve phenotypes that favour ant taxa that are both effective guards and that have minimal net negative impacts, including interference with pollinators. Despite the potential for variation in effects across ant taxa, we found that ant species and the interaction between clogging and ant species had a negligible effect on the number of ants found inside flowers (Table S4) and on pollinator visitation (Table S4).
Our findings suggest that ant species vary in their deterrent effect on A. mellifera bees ( Figure S3)

| The spatio-temporal scale of the distraction hypothesis
Based on the relatively small effect sizes of the short-term leaf-scale experiment (Tables 1 and 2), we hypothesised that clogging the glands of only one leaf for 1 day was perhaps too local and short-term a treatment to detect a measurable effect. In the long-term experiment, we therefore increased both the spatial scale and duration of the EFN-removal treatment by a factor of 10, expecting to obtain larger effect sizes overall. However, in contrast to our prediction, the long-term clogging experiments had smaller effect sizes on ant patrolling (Table S2). For example, the short-term clogging experiment had a 13% greater effect size in reducing numbers of ants patrolling extrafloral nectaries, 155% greater effect size increasing the numbers of ants inside flowers, and 132% greater increase of ant occupancy of flowers than the long-term experiment (Figure 3b; Tables 1 and 2).
Hence, we can robustly conclude that clogging the glands of only one leaf for 1 day is not too local and short-term a treatment. In fact, leafday is the scale at which we detected an effect of clogging and our experimental evidence showed that ant foraging behaviour responds to reward availability over this spatio-temporal scale. The non-provision of a whole branch for 10 days is a rather unnatural setting for ants, or may resemble a low-rewarding plant (Lemus Domínguez, 2014).
Our results suggest that in T. velutina EFN-mediated ant distraction is a mutualist management strategy that acts at a local and shortterm scale. This makes adaptive sense because plant structures vary in their vulnerability to herbivores and sensitivity to both benefits and costs of ant guards over similarly local and short-term scales (Bentley, 1977b;Falcão et al., 2014;Villamil, 2017;Willmer & Stone, 1997). From the plant's perspective, protection needs changes at a very small spatial and temporal scale (Bentley, 1977b;Falcão et al., 2014;Villamil, 2017;Willmer & Stone, 1997) because in T. velutina, buds, flowers and fruits indeed occur in close proximity on the same shoot, and develop from bud to young fruit in only 3 days. Flowers are suggested to be the most vulnerable structure due to their soft and exposed water-rich tissues, while buds and fruits are protected by the sepals or the exocarp respectively. Previous work has shown that EFN secretion in T. velutina is greatest at the flower stage, with glands in the associated leaf secreting 10 times more sugar than glands associated with fruit, and 40% more sugar than glands associated with buds (Villamil, 2017 Rhoades, 1979;Stamp, 2003) and the Distraction Hypothesis (for reduction of negative ant-pollinator interactions).
From the ant's perspective, adjustment of foraging patterns at a local scale could maximise net sugar gain (Schilman & Roces, 2006).
The rapid transition from bud to fruit in T. velutina means that secretion by individual glands can vary substantially over consecutive days since EFN secretion varies greatly throughout this transition (Villamil, 2017). Consequently, for the ants, missing the extrafloral nectaries of leaves associated with flowers means missing a bountiful reward.

| Implications for ant and pollinator foraging strategies
We suggest that ants associated with T. velutina learn the location of highly rewarding EFN glands by monitoring variation in rewards within a single day, rather than relying on cues from previous days-a pattern compatible with demonstrated ability of ants to learn spatial and temporal scales of food rewards (Jackson & Morgan, 1993;Jackson & Ratnieks, 2006;Robinson, Jackson, Holcombe, & Ratnieks, 2005). There is also evidence that at least some pollinating insects can respond to similarly local variation in ant activity. Bees in other systems are known to use ant scents to discriminate and avoid heavily patrolled flowers, preventing harassment (Cembrowski, Tan, Thomson, & Frederickson, 2014) and we suggest that bees visiting T. velutina may also use olfactory cues to reduce their visitation of ant-occupied flowers.
The local foraging decisions we propose and the effects of within-plant variation in EFN availability on ants and pollinators should be seen as occurring against a backdrop of significant between-plant variation in EFN rewards. Differences between individual plants, and not between branches or flowers, explained a large part of the variance in both numbers of ants and pollinators at both experimental scales (Table 1, Table S1). It is possible that plant-level variation in nectar availability underlies the positive correlation between numbers of patrolling ants and pollinator visitation observed in the long-term clogging treatment (Table S2)