Plant-pollinator interaction sampling
For the plant-centred sampling, we carried out observations of pollinators on flowers between 0700h and 1700h. We did ten minutes of focal observations on each individual flowering plant within the plots (usually four/five days of data collection per month and one period per plot). We maintain a similar sampling effort among sites. In order to include time-dependent variations on plant-pollinator interactions, we recorded all plots at different periods of the day during the sampling period. In addition, we conducted focal observations on a given plant species both in the morning and in the afternoon. Sampling effort depended on species occurrence, phenology, and abundance, resulting in a variable observation time for each species (range, mean ± standard deviation): 10-1780min, =177± 293min per plant species. We recorded the frequency of visitation to flowers and only included in the plant-centred analysis the legitimate interactions (i.e., when the floral visitor contacted the reproductive structures of the flowers). Hereafter, these legitimate visitors are referred as “pollinators”. We collected vouchers for all plant species that were identified, and we lodged them in the CGMS Herbarium (see voucher numbers in supplementary material Table S1). The family names followed the Angiosperm Phylogeny Group (APG IV, 2016), and species names were confirmed in the Plant List database (http://www.theplantlist.org/) and updated/corrected whenever necessary.
For the animal-centred sampling, after a visit was recorded, we captured the insect visitors and placed them in separate clean kill-vials for posterior pollen sampling and identification in the laboratory. For pollen sampling, we rubbed small cubes of fuchsin-stained gelatine on the insect’s body, and then placed the gelatine on glass slides (Dafni, 1992; Bosch et al., 2009). We avoided pollen storage areas such as pollen baskets of bumblebees because these contain pollen unlikely to be available for pollination. We identified pollen grains mainly by comparison to a reference palynological collection we assembled for the study region, and with the aid of literature (Erdtman, 1952). We built the reference collection using the same technique employed for pollen on pollinators, collecting pollen grains from all flowering plants that we found in the study area. In total, we identified 66% of pollen grains at the species level, 30% at the genus level and 3% were morphotyped. We deposited the collected insects at the Zoological Collection of Federal University of Mato Grosso do Sul (ZUFMS: license SISBIO nº 47838-2), and the pollen slides at the CGMS Herbarium. For the animal-centred pollen transport network we considered the number of pollinator individuals carrying a given pollen morphotype as the estimate of interaction frequency (Fig. 2). This was done instead of counting the number of pollen grains on pollinators´ body as different plant species vary greatly in the amount of pollen produced and on how the pollen grains stick to the body of the flower visitor (e.g., presence of pollenkitt), rendering such counts less informative (Maglianesi et al., 2015; Sazatornil et al., 2016; Ramírez-Burbano et al., 2017; de Manincor et al., 2020).