Indirect effects arise when one species influences how another species interacts with a third. Pollinator-mediated indirect effects are widespread in many plant communities and are often not restricted to plant species pairs. An analytical framework does not exist yet that allows for the evaluation of indirect effects through shared pollinators in a community context, as well as their consequences for plant fitness. We used network indices describing pollinator sharing to assess the extent to which plant species affect and are affected by others in a pollination network from a species-rich dune community. For 23 plant species, we explore how these indices relate to plant fecundity (seeds/flower) over two years. We further linked plant traits and indices to uncover functional aspects of pollinator-mediated indirect interactions. Species frequently visited by shared pollinators showed higher fecundity and exhibited traits that increase pollinator attraction and generalization. Conversely, species whose shared pollinators frequently visited other plants had lower fecundity and had more specialized traits. Thus, pollinator sharing benefited some species while others suffered reproductive disadvantages, consistent with competition. The framework developed here uses network tools to advance our understanding of how pollinator-mediated indirect interactions influence a species’ relative reproductive success at the community-level.

Fieldwork was carried out in Son Bosc within the protected area inside the s’Albufera Natural Park (39º46’28.1” N, 3º07’45.34” E). The sampling area consisted of a dune ecosystem at sea level in the northern region of Mallorca. We used the same 23 focal species described in Lázaro et al. (2020) for which data on plant-pollinator interactions, plant fitness and population and flower traits were available. The previous study was designed to uncover the link between commonly used network metrics and plant reproductive success (Lázaro et al. 2020), while here we specifically investigate the role of indirect interactions on plant fecundity. Thus, the analyses carried out here are post hoc given the availability of adequate data from Lázaro et al. (2020).

We used observational data on flower visitors to plants recorded in 2016 (38 plant and 119 pollinator species) and 2017 (38 plant and 174 pollinator species), as carried out by Lázaro et al. (2020), during the bloom period of the plant species included in this study (April-July). Insect censuses on focal plants were conducted once per week (15 weeks in 2016, 16 weeks in 2017), during the period of highest pollinator activity (10:00 am-17:00 pm), with each census lasting 5 min. Surveys were conducted on haphazardly selected individuals of all species in bloom each sampling day. Thus, sampling effort was proportional to plant species abundance (Lázaro et al. 2020). We considered only visits in which the insect touched reproductive organs of the flower. We pooled all visits of an insect species to a plant species across all observation periods. Then, we built quantitative networks for each year based on two visitation variables: visitation frequency (number of visits) and visitation rate per flower of each pollinator to each plant species.

Plant fecundity was estimated in 20 and 30 individuals per species in 2016 and 2017 (each year using different individuals), respectively (Lázaro et al. 2020). For each species, we obtained: (1) fruit set, i.e. number of fruits/infructescences per individual divided by the number of flowers/inflorescences, and (2) seed set, estimated as number of seeds per fruit (from one fruit/infructescence per individual chosen randomly). We used seeds per flower as the final fecundity measure, expressed as fruit set (fruits per flower) x seeds per fruit.

We used the trait data measured for each plant species in the field first reported by Lázaro et al. (2020). Briefly, we recorded the following traits: (1) Flower abundance (flowers/m²); average number of open flowers (flower units: flowers or inflorescences depending on the species) of each species counted in five transects of 50 x2 m sampled once every two weeks each year; (2) Flowering length; number of days the focal species was observed in bloom across all temporal surveys each year. (3) Flower shape; zygomorphic vs. actinomorphic flower or inflorescences; (4) Flower size; average largest diameter (width of the flower/inflorescence of each species), measured with digital calliper in 30 randomly chosen individuals per species (one fully-open flower unit randomly chosen per individual); (5) Corolla tube length (30 individuals per species); (6) Nectar volume, measured as the nectar standing crop of 10 randomly chosen individuals per species (one flower unit randomly chosen per individual) by means of microcapillary tubes; (7) Dependence on pollinators (degree of selfing), calculated as 1 – B/C, where B are the seeds per flower produced by 10 branches/individuals prevented from insect visitation (i.e. bagged before flower anthesis) and C are the seeds per flower of 10 control branches/individuals, open to natural pollination.

To estimate indirect interactions via pollinator sharing, we calculated potential indirect effects in the networks (one per year) based on visitation frequency and based on visitation rates per flower between all plant species pairs using the index proposed by Müller et al. (1999), by means of the PAC function in the bipartite R-package (Dormann et al. 2009). We obtained two sets of indirect interaction indices: the first based on visitation frequency (number of visits) and the second based on visitation rates per flower. For the indices based on visitation frequency, we built two plant-plant matrices (one for each year), with column values representing how much a plant species influences the other species, and rows representing how much the plant species is affected by other species. To obtain a single value per species and year, we summed all column values per species (separately for each year), representing the total effect of the species on all other plant species in the network (Acting degree, the sum of the effects of plant j on all other plant species in the network). Similarly, we summed all row values per species, representing the total effect of all species in the network on a particular one (Target degree, the sum of the effects plant i received from all other plant species in the network). We excluded the diagonal, setting these values to zero, as they are estimates of the potential for intraspecific competition. Additionally, to obtain indices without the influence of abundance, we obtained plant-plant matrices based on pollinator visitation rates per flower. We then followed the same procedure to estimate indirect effect indices in the pollination networks with visitation rates per flower as link weight. In other words, how much a plant species affects others by receiving a large fraction of per flower visits of shared pollinators (‘Per flower acting degree’) and how a plant species is affected by others receiving a large fraction of per flower visits of shared pollinators (‘Per flower target degree’).