Ant guards can increase plant fitness by deterring herbivores, but they may also reduce it by interfering with pollination. While ant impacts on herbivory have been well-studied, much less is known about their impacts on pollinators and associated consequences for plant pollination, particularly pollen transfer dynamics and outcrossing/selfing rates. We used field experiments to quantify the effect of ant guards on pollinator community composition, frequency and duration of flower visits, and cascading effects on outcrossing pollen transfer and pollen exports in Turnera velutina (Passifloraceae). Although ant patrolling did not affect pollinator community composition or visitation frequency, it decreased flower visit duration and the time pollinators spent foraging inside flowers. Such behavioural changes resulted in reduced pollen deposition on stigmas, decreased pollen exports (a proxy for male fitness) and significantly doubled outcross pollen transfer. This study contributes to our understanding of how nonpollinator mutualists can shape plant reproductive processes. We discuss the downstream effects that variation in biotic defences, such as rewards for guarding ants, can have on plant pollen transfer patterns and fitness. In conclusion, guarding ants influence pollen transfer patterns in Turnera velutina, increasing outcrossing in a self-compatible species at the cost of male fitness. We show how non-pollinators, such as defensive ant mutualists, can shape plant reproductive traits and discuss the consequences these interactions may have for plant mating systems.

Study site and system

Field experiments were conducted in coastal sand scrub at Troncones, Guerrero, on the southern Pacific coastline of Mexico (17°47’ N, 101° 44’ W, elevation < 50 m). Turnera velutina (Passifloraceae) is a Mexican endemic shrub (Cuautle and Rico-Gray 2003, Arbo 2005) that establishes a facultative mutualism with ants (Zedillo-Avelleyra 2017), rewarding them with extrafloral nectar (Villamil et al. 2013). Turnera velutina is a self-compatible, herkogamous, polyploid species that requires insect pollinators for seed production (Sosenski et al. 2016). Although it flowers year-round, flowering peaks during summer (Cuautle et al. 2005), and flowers last only one day, having a very short anthesis period (4-6 h) (Sosenski et al. 2016, Villamil et al. 2018). Pollinator rewards are pollen and floral nectar (Sosenski et al. 2016, Villamil et al. 2018). At Troncones, native butterflies are the dominant flower visitors of T. velutina, followed by the introduced honeybee (Apis mellifera); native bees, wasps, and occasionally flies also visit the flowers. All field experiments were conducted during June 2017.

Ant exclusion                                                       

We identified large plants producing six or more flowers per day, surrounded by other T. velutina plants, but not within dense vegetation patches, and with few vegetation links to other plants, so that they could be isolated for the ant exclusion treatment. From within this selection, we haphazardly chose twelve plants (hereafter focal plants) forming six pairs of nearby plants with similar height and foliage abundance. In each plant pair, one focal plant was haphazardly designated as control, with natural levels of ant-guard activity. The second plant had ants excluded from all stems using a sticky barrier (Tanglefoot) (Fig. 1a). Both focal plants in each pair were isolated physically from other plants by trimming or tying back any surrounding vegetation. Exclusion treatments were checked daily and Tanglefoot was reapplied if required. We ensured similar ant patrolling dynamics between plant pairs. All control plants were patrolled by Dorymyrmex bicolor ants with similar instantaneous counts per branch: 10.86 ± 0.77 ants/plant (mean ± se). Plants within the pair were located 2-4 m apart, whilst plant pairs were separated from any other plant pair by at least 10 m (range: 10-150 m). Each focal plant pair was surrounded by 6-10 neighbouring adult plants of T. velutina 2-4 m away to ensure allogamous pollen supply (see details below, Fig. 1a).

Pollen tracking

To assess the effect of ant patrolling on pollen transfer and its consequences for rates of selfing, geitonogamy, and outcrossing, we dyed the anthers and pollen of control and ant-excluded plants using four contrasting dyes: red, blue, green or purple (Fig. 1b) using a focal-plant, focal-flower experimental design. Within a focal plant, a focal flower was selected and its anthers were dyed differently from the anthers in the rest of the flowers from that plant (hereafter referred to as satellite flowers), which were all dyed using a second colour (Fig. 1a). All satellite flowers from a given plant were dyed in the same colour. The focal flower of the other plant in the pair was dyed using a third colour, and all of its satellite flowers using a fourth colour. Colours were assigned independently of treatment to avoid confounding effects. Each pair of focal plants was surrounded by neighbouring T. velutina plants with 10-12 flowers (hereafter referred to as non-focal plants). Pollen from the neighbouring non-focal T. velutina plants was left undyed (naturally yellow-orange). The dyeing treatment was repeated in each of the six plant pairs for five days.

Pollen grains of focal and satellite flowers were dyed once the corolla opened completely (~08:00-08:15 h), but before anther dehiscence. Anthers were individually embedded in a droplet of liquid dye until soaked, and flowers were bagged again until anthers dehisced and the released, dyed pollen grains were dry. The dyes used were methyl violet (purple; Drogueria Cosmopolita, Mexico), Green S (green; SigmaAldrich), safranin (red; SigmaAldrich), and methylene blue (blue; SigmaAldrich). Purple dye was prepared by diluting commercially available methyl violet with ethanol 70% [50:50]; the green dye used was commercially available Green S without dilution. Red (safranin) and blue (methylene blue) were prepared by diluting 1 g of either dye in 15.5 ml of distilled water, and then mixing this solution with ethanol 95% [50:50]. Previous studies showed that dyeing Turnera velutina anthers in these colours effectively dyed pollen grains and had no effect on pollinator visitation (Ochoa Sánchez 2016).

Every morning before anthesis, six flower buds per plant (1 focal + 5 satellite buds) were bagged to exclude visitors. All additional pre-anthesis buds were removed to standardise floral display across focal plants. Once the corollas were fully open (~08:00-08:15 h), anthers were dyed and flowers were re-bagged until the dyed pollen dried (Fig. 1b). To ensure a minimum common supply of allogamous pollen in each experimental array, 10-12 undyed flowers from the neighbouring plants within the array were also bagged before anthesis. All flowers then remained bagged until visitation observations began at 08:40 h. Towards the end of anthesis (11:30 h), pistils from the six focal flowers were collected in Eppendorf tubes and slide-mounted as a glycerine squash (Kearns and Inouye 1993, Ochoa Sánchez 2016) to allow counting of pollen grains received (Fig. 1c), and identification of their origins based on the dye-tracking.

Pollinator visitation: community composition, visitation frequency, duration, and behaviour

We recorded pollinator visitation to all six flowers on control and ant-excluded focal plants. Every focal plant was observed for a total of 40 minutes in two 20-minute periods – one immediately after bag removal (08:40 h) when flowers had a full pollen and nectar load and the second 90 minutes later. We recorded floral visitor identity and behaviour, and the frequency and duration of visits, as detailed below, for 40 h of observations of 360 flowers on 12 plants over five days (N=60 focal flowers), with all experimental plant pairs observed in each of the five days. Because T. velutina has one-day flowers, we considered daily assays as independent samples, whilst accounting for plant identity, pair and day in our models to avoid pseudo-replication. Visitor taxa were identified as potential pollinators if they were observed contacting male and female plant sexual organs. Most statistical models in this section shared a common fixed and random effects structure, even though their error distribution varied depending on the response variable. In general, the models included exclusion treatment, pollinator type and their interaction as fixed effects; plant pair identity and day were included as random effects to account for the non-independence of flowers sampled on the same day or on the same plant across several days. The error distribution of the models varied depending on the response variable and variations to this model structure are specified and detailed in Table 1.

a)     Pollinator community composition

Flower visitors regarded as potential pollinators (hereafter pollinators) were identified to one of five taxonomic categories: Apis mellifera, native bees, butterflies, flies, and wasps. To estimate the relative abundance of each pollinator category, we calculated the percentage of all visitors from each group summed over control and ant-excluded plants. For each taxonomic group, differences in the total number of visitors between control and ant-excluded plants were assessed using a Pearson Chi-squared test. Because Apis mellifera and butterflies jointly accounted for 94% of all visitors (Table S1), only these taxonomic groups were included in all further analyses.

b)     Pollinator visitation frequency and duration

Flower visits were scored each time a pollinator hovered over, landed, and contacted the reproductive organs of a flower. Visit duration was recorded until the pollinator departed. Re-visitation by the same insect was scored as a separate visit. Visitor abundance was estimated as the number of individual visitors per taxon landing on flowers of a particular plant. A visitor that landed, hovered, and landed again in another flower was registered as two visits from one visitor. Ant presence effects on visitation frequency were tested using a Poisson mixed model, which also included an observation-level random effect (OLRE) to cope with overdispersion (Table 1, M1).

c)     Pollinator behaviour

Visit duration was differentiated as inspection and contact behaviours, to test whether the time pollinators spent in these activities differed between ant-patrolled and ant- excluded plants. All pollinator visits were allocated to one of two behavioural categories following Villamil et al. (2018): inspection (approaching a flower  and hovering above it without landing) or contact (landing on the flower). The effect of ant patrolling on the likelihood of inspection behaviour was tested with a binomial mixed model, with presence/absence of inspection behaviour as the response variable; this model included pair identity as a random effect (Table 1, M2). We also used a binomial mixed model to test whether ant patrolling affected pollinator avoidance behaviours, defined as the pollinator failing to land on a flower following inspection behaviour; this model included pair identity as a random effect (Table 1, M3). For every pollinator that displayed an inspection behaviour, we recorded the presence (scored as 1) or absence (scored as 0) of subsequent contact behaviours and fitted this as a binomial response variable.  We used a separate Poisson mixed model to test for an effect of ant presence on the duration of inspection and contact behaviours (visit duration was measured in integers accounting for the number of seconds).  In addition to the exclusion treatment and pollinator type, this model included pollinator behaviour as a fixed effect, and pair identity and an OLRE as additional random effects (Table 1, M4).  We decided a priori to include the full combination of two-way interactions, and deliberately avoided fitting a three-way interaction because their outcomes are usually hard to interpret and visualise.

Pollen transfer dynamics (outcrossing, selfing and geitonogamy)

The effect of ant patrolling on pollen transfer rates was assessed by counting differentially dyed pollen grains on focal flower stigma squash slides under a light microscope. The effect of ant patrolling on stigma pollen load, defined as the total number of pollen grains received per stigma, was tested using a Poisson mixed model which included the exclusion treatment as a fixed effect, and plant identity, pair identity, day and OLRE as random effects (Table 1, M5).

Outcrossing, selfing, and geitonogamous pollen transfer rates were characterised in terms of the proportions of pollen grains on focal stigmas identified by colour as transferred from the same flower (selfing), another flower within the same plant (geitonogamy), or another plant (whether the other focal plant in the same array or un-dyed plants surrounding the array) (Fig. 1a). The proportion of pollen grains from each source (selfing, geitonogamy, or outcrossing) was calculated for each focal stigma by dividing the number of pollen grains from each source by the total number of pollen grains on that stigma (pollen load). Proportional data were transformed to normality using the logit transformation, with infinite numbers resulting from impossible quotients replaced by zeros. The effect of ant patrolling on pollen source was determined using a linear mixed model, fitting the proportion of pollen grains from each source (selfing, outcrossing, or geitonogamy) as the response variable; ant exclusion treatment, mating system (pollen source) and their interaction as fixed effects, and plant identity, pair identity, day and an OLRE as random effects (Table 1, M6).

Pollen transfer dynamics were analysed by further subdividing pollen flow into five categories depending on pollen transfer patterns and pollen source (with pollen-source, pollen-origin categories hereafter abbreviated to PSPO; Fig. 1a). Pollen source refers to whether the pollen comes from selfing, outcrossing or geitonogamy; whilst pollen origin refers to the provider of the pollen and it helps differentiate whether the pollen arrived from a short distance (dyed pollen from the other plant in the pair) or undyed pollen from the undyed, surrounding plants which comes from a greater distance. These categories summarise pollen grains received from, and donated to, every possible pollen source identifiable by colour in this experiment (Fig. 1a), as follows: (i) received/donated by the same flower (selfing), (ii) received from another flower on the same plant (geitonogamy received), (iii) received from the reciprocal focal plant (outcrossing pair received), (iv) donated to the reciprocal focal plant (outcrossing pair donated), or (v) received from another plant of the same species (outcrossing unknown received). The effect of ant presence on pollen flow dynamics was tested using a Poisson mixed model, fitting the number of pollen grains in each of the five PSPO categories as the response variable. This model included the ant exclusion treatment, pollen origin category and their interaction as fixed effects; plant identity, pair identity, day and an OLRE were included as random effects (Table 1, M7).   

Male fitness component

The number of pollen grains donated per flower was used as an estimate of plant male fitness, quantified as the average number of pollen grains donated to focal stigmas. The total number of pollen grains from satellite flowers on the same plant was divided by five, to obtain the mean number of pollen grains donated per satellite flower. Similarly, the total number of pollen grains from the other plant in the same array was divided by six (1 focal flower + 5 satellite flowers). The effect of ants on male fitness was estimated using a Poisson mixed model fitting the number of pollen grains donated per flower as the response variable, the ant exclusion treatment as a fixed effect, and plant identity, pair identity, day and an OLRE as random effects (Table 1, M8).

The effect of ant patrolling on the destination of the pollen grains donated per flower was tested using a Poisson mixed model fitting number of pollen grains as the response variable, the ant exclusion treatment as a fixed effect, and plant identity, pair identity, day and an OLRE as random effects (Table 1, M9). We contrasted the number of pollen grains donated by focal or satellite flowers in each array, but (in contrast to our pollen transfer pattern analysis, M6) ignored pollen from unknown, undyed plants for which the number of contributing flowers could not be determined.

Statistical software and details

Statistical analyses were conducted in R version 3.5 (R Core Team 2016).  All mixed effects models were fitted using the ‘lme4’ R package (Bates et al. 2016), residuals and model assumptions were checked using  the ‘DHARMa’ package (Hartig and Hartig 2017), and post-hoc Tukey comparisons were tested using the ‘multcomp’ R package  (Hothorn et al. 2008). Full details on the fixed and random effect structure of each model are provided in Table 1. To account for overdispersion, when needed, we included an observation-level-random effect in the models (Hinde 1982, Harrison 2014).

Files are saved .csv and can be opened with R, Microsoft Excel or any free-software equivalent.