Vertical stratification of plant–pollinator interactions in a temperate grassland

Field experiments

We conducted two field experiments in a dry grassland near Český Krumlov, in the southern part of the Czech Republic (48°49′28″N 14°18′59″E). The study site is a species rich calcareous grassland on a southwest-facing slope managed by occasional pasture by cows and sheep. The area is state-owned and publicly accessible. No permits were needed for this study.

In the first experiment, we observed visitation of inflorescences of two plant species, Centaurea scabiosa and Inula salicina, at different heights above ground. To avoid confounding factors, e.g. taller plants having a different size of inflorescences than shorter plants, we used inflorescences cut from plants in the local population. We selected inflorescences of a similar size and general appearance and placed them in 15 ml tubes with water. We attached each tube to a bamboo stick of different length and attached the stick to the ground. This way, we manipulated the height of the inflorescence between 5 and 105 cm above ground. We placed the inflorescences along two short transects, each containing seven inflorescences placed 50 cm apart. One transect was surrounded by short and the other by tall and dense vegetation; the transects were ca. 10 m apart. The area of short vegetation was grazed by cows in the spring, while the area of tall vegetation was not managed. Short vegetation was characterized by most plants <10 cm tall; the average height of inflorescences of all plants growing within 50 cm from the transect in all directions was 7.2 cm (SD = 6.00). Tall vegetation was composed of a dense layer of plants reaching ca. 50 cm; plants growing within 50 cm from the transect had flowers on average 50.1 cm above ground (SD = 14.73). We individually placed seven inflorescences of either C. scabiosa or I. salicina in each transect at 5, 15, 25, 45, 65, 85, and 105 cm above ground in a randomized order.

We observed visitation of the inflorescences by insects between 10:30 and 16:00 h, for 30 min in each transect, and identified all visitors at the species level or classified them into taxonomical groups with the highest precision we could achieve without capturing the insects. Both transects were observed simultaneously, one person observed each transect. After the 30 min period, we replaced the inflorescences and randomized the order of their vertical position along each transect and took another set of observations. In total, we measured inflorescence visitation in 20 transects in I. salicina, 10 in short and 10 in tall vegetation, and 16 transects in C. scabiosa, eight in short and eight in tall vegetation. In total, this amounts to 18 h of observations.

In the second experiment, we focused on the effect of inflorescence height for flower visitation rate and its consequences for seed set in S. verticillata at the same site. In this case, we did not cut the inflorescences so that we could test whether flower visitation rate varied between inflorescences within the natural limits of their height above ground and to test whether variation in flower visitation rate translated into differences in seed set; i.e. whether female fitness was affected by the vertical position of the inflorescence. We used the following approach to minimize confounding effects, such as taller plants having more resources, different display size, etc. We selected 17 plants of S. verticillata with multiple ramets of approximately the same size and with the first several flowers open or with buds ready to start flowering on 18 July 2017. We took advantage of the fact that the inflorescences grow on relatively long and flexible stems. We bent one of them close to the ground, where it was attached to a stick so that the bottom of the inflorescence was positioned just above ground. We made sure that the orientation of the inflorescence remained unchanged. The second stem was attached to another stick so that it reached a maximum height and the third inflorescence was positioned at an intermediate height. This way, the only difference between the inflorescences was their vertical position.

We performed observations of flower visitation in individual S. verticillata plants at one of three dates (20 July, 21 July, and 2 August 2017), depending on when they reached the peak of flowering. The three manipulated ramets per plant were observed simultaneously during one 30 min period and filmed using three digital cameras, which gave a total of 25 h of recordings. Afterwards, we measured the height of each inflorescence as a distance of the highest open flower from the ground and counted the number of open flowers. We then watched the recordings and counted and identified all flower visitors. For each visitor, we also counted the number of flowers visited during each inflorescence visit.

We waited for the seeds of S. verticillata to ripen and then harvested them on 14 or 24 August 2017 depending on seed development in individual plants. We counted the number of developed seeds and the maximum potential seed set by multiplying the number of flowers by four, which is the number of seeds the plant can produce per flower. We counted the flowers and seeds in individual whorls within each inflorescence separately to gain data on potential differences in percentage seed set along the inflorescence from the lowest to the highest whorl.

As already mentioned, we identified flower-visiting insects without capturing them. Naturally, we could not identify all individuals to the species level, so we classified some of them into higher taxa or categories, such as ‘small solitary bees.’ The most abundant flower visitors were bumblebees, Bombus spp., some of which are difficult to identify alive. Fortunately, we have extensive collections from the study site, so we know that there are three species, which we could not distinguish from Bombus terrestris, specifically B. lucorum, B. cryptarum, and B. magnus. However, over 85% of individuals of this species group in our collections from this site belong to B. terrestris. Similarly, B. lapidarius could be confused with B. confusus and B. ruderarius, but they have rarely been found on the site. A similar level of uncertainty exists in our identification of B. sylvarum. The number of potential bumblebee misidentifications during the field observations was thus low and unlikely to confound our results.

Data analysis

We tested how the total number of visits and the number and proportion of visits by individual flower visitor taxa depended on inflorescence height using generalized additive models (GAM) to account for the non-linear nature of these relationships. The identity of individual plants, each having three ramets manipulated and observed, was included as a random factor in analyses of data from the experiment with S. verticillata; i.e. generalized additive mixed models (GAMM) were used in this case. Poisson distribution with overdispersion (quasi poisson distribution) was used for the number of visits, while overdispersed binomial (quasibinomial) distribution was used for data on proportions. These analyses were performed using mgcv 1.8-17 package (Wood, 2006) in R 3.2.3 (R Core Team, 2015).

To gain insights into the effects of inflorescence height on plant fitness, we tested how seed set of individual ramets depended on the number of flowers and the inflorescence height using a generalized linear model. Similarly as in analyses of selection gradients (Lande & Arnold, 1983), we standardized both predictors to have zero mean and unit variance. Partial regression coefficients then allowed us to compare whether reproductive performance (seed set) depended more strongly on the number of flowers or inflorescence height.

The effects of inflorescence and vegetation height on visitation by potential pollinators

In the first experiment, using C. scabiosa and I. salicina as focal species, total inflorescence visitation peaked approximately at or slightly above the level of the surrounding vegetation in transects surrounded by both short and tall vegetation. These results are consistent with observations that flower visiting bees tend to fly at a specific height and when they leave one inflorescence, they are more likely to fly to another one at a similar height compared to inflorescences lower or higher above ground (Levin & Kerster, 1973; Gumbert & Kunze, 1999). Flowers positioned above the level of a dense layer of vegetation are probably easier to detect and thus attract more visitors (Gumbert & Kunze, 1999). Height preferences of flower visitors may lead to vertical stratification of the plant–pollinator network if different species vary in their behaviour (Roubik, 1993; Gumbert & Kunze, 1999; Ramalho, 2004).

Different responses of individual species or taxonomic groups of flower visitors to inflorescence height, which we observed, mean that inflorescences at different heights varied in the composition of their flower visitor assemblages (Figs. 4 and 7). Plant–pollinator interactions thus appear stratified along the vertical dimension despite the fact that height differences between inflorescences were in the order of mere decimetres. Previous studies on vertical stratification of pollinator communities were conducted mostly in forests where the importance of the vertical dimension is more obvious. It seems there is generally a major difference in the composition of flower visitor communities between the canopy and the understory in tropical (Roubik, 1993; Nagamitsu et al., 1999; Ramalho, 2004) as well as temperate forests (Ulyshen, Soon & Hanula, 2010), probably related to vertical distribution of flowers preferred by different species (Ramalho, 2004). However, vertical stratification of plant–pollinator interactions has been less studied in grasslands where the vertical distances are limited usually to several decimetres. In one of the few available studies (Gumbert & Kunze, 1999), differentiation between visitor communities on flowers below and above the dominant grass layer was observed in a tropical wetland. Similarly, several species of bees visiting pumpkins in an agricultural landscape preferred flowers at different heights (Hoehn et al., 2008). Our results show that vertical stratification of plant-flower visitor interactions may be important also in common temperate grasslands.

An interesting observation is that small solitary bees were the dominant flower visitors close to the ground in all three plant species. We noticed that unlike other flower visitors, they were frequently flying among plant stems close to the ground even in dense vegetation and visiting flowers hidden there, such as flowers of Rubus sp., which was growing in parts of the study site. The same pattern was observed by Gumbert & Kunze (1999), who suggested that small bees, which are able to manoeuvre in dense vegetation, may benefit from decreased competition for floral resources because most other flower visitors avoid this microhabitat. This seems to be a likely explanation for our results as well. Behaviour of small solitary bees contrasted with the behaviour of larger species, such as bumblebees, which were flying above the layer of dense vegetation. For example, almost all observations of B. lapidarius in tall vegetation were at the height of >40 cm (Fig. 2E).

The effect of flower visitation was thus modified by the height of the surrounding vegetation because most flower visitors avoided flowers within the layer of dense vegetation close to the ground. One bumblebee species, B. terrestris, also visited mostly inflorescences close to the ground in short vegetation (Fig. 2B), but almost completely avoided the area with tall vegetation, similarly to Halictus quadricintus, a solitary bee, which favoured inflorescences >60 cm above ground (Fig. 2C). Previous studies on the effects of the structure of the surrounding vegetation for flower visitation and plant reproductive success are rare and did not provide clear conclusions. For example, Ågren, Fortunel & Ehrlén (2006) manipulated vegetation height and litter presence around individual plants of P. farinosa, and found that litter removal and vegetation pruning increased seed set, especially in short plants. They did not report any data on visitation frequency of flowers in relation to vegetation structure. However, if we assume that higher visitation leads to higher fruit or seed production, our observations of higher visitation of inflorescencences close to the ground in short vegetation exactly mirror these results. In tall vegetation, visitation of both C. scabiosa and I. salicina peaked higher above ground, which fits the results of Sletvold, Grindeland & Ågren (2013), who observed pollinator-mediated selection for taller inflorescences in tall vegetation but not in short vegetation in a deceptive orchid, Dactylorhiza lapponica.

We did not measure the efficiency of different pollinators in the present study, so we cannot infer consequences of the variation of visitation at different heights for the reproductive success of C. scabiosa and I. salicina. However, data from detailed single visit experiments by other authors demonstrated that different flower visitor species vary in pollen deposition by several orders of magnitude (King, Ballantyne & Willmer, 2013). Hence, it is likely that variation in total visitation rate together with the variation in visitor identity with inflorescence height affects reproductive success of plants in our system. We addressed this question in the second experiment with S. verticillata.

Consequences of inflorescence height for seed set and plant fitness

Observations on S. verticillata were constrained by the natural range of inflorescence heights. Unlike previous studies, we collected data both on flower visitation rates by pollinators and on seed set of individual ramets in relation to inflorescence height. Our results are consistent with the hypothesis of positive selection for inflorescence height, which other authors demonstrated in several other plant species (Sletvold, Grindeland & Ågren, 2010, 2013; Jiang & Li, 2017; Trunschke, Sletvold & Ågren, 2017). However, we could also demonstrate that inflorescences positioned higher above ground had higher total flower visitation rates.

Our experimental design allowed us to measure the dependence of seed set on inflorescence height, because we compared similarly looking inflorescences in three ramets per plant (genet), whose height was experimentally adjusted. Vertical position was thus the only apparent difference between the inflorescences. This is important because taller plants usually have higher percentage seed set even when they are hand-pollinated because they have more resources than shorter plants (Andersson, 1996; Červenková & Münzbergová, 2014). Testing for pollinator-mediated selection on inflorescence height thus requires specific experimental designs and is not possible by simply comparing plants of different height. The most frequent approach is to compare selection coefficients for inflorescence height between open-pollinated and hand-pollinated plants (Sletvold, Grindeland & Ågren, 2010; Červenková & Münzbergová, 2014; Jiang & Li, 2017; Trunschke, Sletvold & Ågren, 2017). We used an alternative approach, which allowed us to skip the hand-pollination treatment. We took advantage of the morphology of S. verticillata, which creates multiple closely packed, relatively long, and flexible ramets, which can be easily pinned closer to the ground or straightened up without causing damage. For our observations, we selected three ramets with inflorescences of a similar length and general appearance per plant and randomly adjusted their vertical position, so there was no known confounding factor. So, the observed positive correlation between inflorescence height, visitation rate, and seed set can be interpreted as evidence for pollinator-mediated selection on inflorescence height. Nevertheless, some caution is needed, because vertical position of the inflorescence could also affect water transport, which is more difficult to inflorescences higher above ground. Inflorescences at different heights also presumably experienced different levels of shading. In additional, results of our first experiment, where we observed visitation rate on flowers of C. scabiosa and I. salicina along a wider range of heights, suggest that there may be an optimal height maximising insect pollination depending on the context of the surrounding vegetation.

Flower visitation by different insect species can have different effects on plant fitness not only because different species differ in the number of pollen grains deposited per visit (King, Ballantyne & Willmer, 2013), but also because they differ in the relative frequency of movements between flowers on the same plant and between different plants (Paton, 1993). Our observations showed that honeybees (A. mellifera) visited a majority of flowers within an inflorescence by moving along the inflorescence and probing one flower after another. The three bumblebee species exploited slightly lower percentage of flowers, while small solitary bees and Diptera usually probed only a few flowers per inflorescence (Fig. 7B). Different pollinators thus have a different potential for geitonogamous pollination, because pollen from the previously visited plant is deposited mostly on stigmas of the first few flowers and receipt of foreign pollen exponentially decreases in each successive flower visit within an inflorescence (Thomson & Plowright, 1980; Gerber, 1985; Morris et al., 1994). High levels of geitonogamous pollination may negatively affect plant fitness (Gerber, 1985; Waser & Price, 1991; de Jong, Waser & Klinkhamer, 1993; Ruane et al., 2013). In our population of S. verticillata, we would expect higher level of geitonogamous pollination in inflorescences higher above ground because of shifts in the flower visitor community (see Fig. 7), perhaps also in individual whorls higher within an inflorescence. However, proportional seed set at the scale of entire inflorescences and individual whorls increased with height, as a consequence of higher total visitation rate.

Inflorescences which are more attractive for pollinators may also attract higher numbers of florivores and seed predators (Sletvold & Grindeland, 2008). Seed set is thus driven by a balance between mutualistic and antagonistic interactions with flower visitors (Ehrlén, Käck & Ågren, 2002; Schlinkert et al., 2016). We did not notice any conspicuous evidence of florivory during observations of flower visitors or seed predation when processing seeds of S. verticillata. However, in other plant species, florivory and seed predation can have a large negative effect on plant fitness (Ruane, Rotzin & Congleton, 2014). For example, Schlinkert et al. (2016) found that abundance of both pollinators and florivores increased with plant height and mutualistic and antagonistic interactions had contrasting effects on the number of seeds leading to seet set being independent of plant height. The role of inflorescence height may thus be species-specific and context-dependent, as shown also by our observations of inflorescence visitation in short and tall vegetation.