Flower visitation data

Following the methodology proposed by Vaissière et al. [35] in each sample site, we first quantified the total number of flowers in anthesis and flower visitors along two parallel transects (25x1m, each flower within transect was observed for ca. 30 seconds). Data were collected during morning (09h00 to 12h30) and afternoon (12h30 to 16h00), maintaining a minimum time interval of three hours between surveys in a given site (so each site was sampled twice within a single day of the peak of flowering). Environmental conditions were measured in all surveys (temperature between 21 to 37°C, humidity between 32 and 88%, all surveys occurred when there was no precipitation). All insects that touched the reproductive organs of flowers were considered “legitimate flower visitors”. Robbing events (i.e. perforation of the corolla to extract nectar without having contact with the reproductive organs), were also recorded. Whenever an insect was observed, a description of the visitor (Order, size and colour pattern) was also recorded (morphospecies). Afterwards, insects were captured along the transect, and later identified by taxonomists.

While the sampling method was developed specifically to detect pollinators, many species that act as herbivores and seed predators during larval stages visit flowers in their adult stage were detected. Although all these floral visitors may have contributed to pollination to some extent, some have a different main function having negative (crop pests) or positive (biocontrol agents) additional effects. Consequently, in our specific study, flower visitors were divided into three main functional groups based on species biology: pollinators, species that may act as biocontrol agents and species that may act as crop pests (see Table E in S1 File). Pollinators included all bees as well as one syrphid fly species (Palpada vinetorum Fabricius, 1798) that does not act as predator during the larval stage [36]. In the group ‘potential biocontrol agents’, we included wasps (Hymenoptera: Vespidae and Pompilidae) which typically predate larvae of Lepidoptera and Coleoptera [37], and one Syrphidae morphospecies that belongs to a genus that feeds on agricultural pests that affect common bean (Allograpta exotica cf.). Pests included several Lepidoptera, Coleoptera and Hemiptera species which are known to attack common bean or other crop species [28]. Pollinators were further divided into exotic (i.e. honeybee, Apis mellifera) and native pollinators. Wild populations of honeybees occur in the study region and none of the farmers involved in this study managed honeybees. However, given the vast foraging range of this species, we were not able to identify the status of the populations (wild or managed) from which honeybees detected visiting common bean in our study region came from.

As the number of flowers observed varied among plots, and sampling effort is directly proportional to total number of flowers observed, we calculated (i) flower visitor density and (ii) species density, and used these variable in data analyses of productivity. To do that we divided the total number of flower visitors for each group and species by the total number of flowers observed. Only species that were not crop pests were considered in richness calculations. Richness estimations also considered any uncollected morphospecies, which description did not match with the collected species.

Production data

Production data were collected ca. 90 days after the date of planting, during the fruiting period, on the exact same place where flower visitation surveys were done. On each sampling site, 15 crop plants were randomly selected. For each individual selected, all pods produced (including thin pods with no beans, due to lack of ovule fertilization, or with aborted beans) were collected and stored in paper bags. The number of mature beans (> 3mm, following national marketing definition for common bean [38], all other beans were considered aborted) were counted and placed in a 65°C oven until they reached humidity below 14% [39]. Afterwards, the beans were weighed. Overall productivity (yield) depends on number of flowers produced per area unit, and on the number of ovules fertilized per flower. Only the last is influenced by pollinators, and the first is highly dependent on soil fertilization [16]. As this work aimed to evaluate the effect of pollination services, we considered two production metrics: productivity per flower and overall land productivity (yield). As the first metric is independent of the total number of flowers, it better represents the effect of pollination. Number of flowers observed during flower surveys were done is not necessarily correlated with the total number of flowers produced over time in that plot. Therefore, we considered the number of pods (including those undeveloped) to be the best proxy of the number of flowers produced throughout the flowering season. Productivity per flower was, hence, calculated by dividing the total bean weight per plant by the total number of pods. Overall land productivity was calculated by multiplying the average bean weight per plant by the average number of plants per square meter in our study fields (see above) and converted to kg per ha. Overall production deficit was calculated as the ratio between the productivity per flower (of each sampling site) and the maximum value of productivity obtained in the study region, as calculated by Garibaldi et al. [9].

Statistical analyses

Linear mixed-effects models (LMM) were used to evaluate the influence of the density of the different insect groups (i.e. native and exotic pollinators, biocontrol agents and pests) and overall richness (here, species density) on bean productivity. To account for the spatial and temporal aggregation of sampling points data, ‘field’ within ‘farmer’ and ‘year’ were included as random variables. We ran the analyses for the two productivity variables: productivity per flower and overall land productivity. Model assumptions were checked and to account for the typical sigmoidal behaviour of production (i.e. have established maximum and minimum values), and adapted logi transofrmation was applied to normalize residuals, using the maximum value of productivity detected in field (rounded up) as the top asymptote.

To test if higher richness and density of ecosystem service providers (pollinators and biocontrol agents) increased bean productivity and if the effect of density of pollinators was boosted by richness (hypothesis 1), density of native, exotic (i.e. Apis mellifera) pollinators, biocontrol agents and visitor richness were included as fixed terms. In addition, two-way interaction between density of pollinators and richness was also included as fixed terms. To test if pollinator benefits to crop yield increase as N fertilizer input reduces (hypothesis 2), interactions between pollinator density (native and exotic) and N input were also considered. To identify the terms that most contributed to explain bean productivity, a model selection procedure was applied, based on the Akaike Information Criteria corrected for small sample size (AICc) [40]. Whenever several models had equally good predictive power (i.e. several models with ΔAIC below 2), we applied model averaging. Only models with ΔAICc lower than 2 were considered in the calculation of the average estimates, and whenever a variable was not included in a model we assumed an estimate equal to zero.

To check if sites with extreme values of N input (with high and lower N input) were driving the results on the effects of N input on bean production, we ran additional models after removing extreme values (sampling points with 36 or 130 kg N/ha). This sensitivity test also allowed the detection of thresholds above which the effect of nitrogen is no longer observed in this region.

To test if the community composition (abundance of each group and overall richness) of ecosystem service providers and probability of pest occurrence changes with increasing N input, particularly near patches of natural vegetation (hypothesis 3), we used Generalized Linear Mixed Models (GLMM, negative binomial distribution), considering the same random variables mentioned above. First, for each group of ecosystem service providers, we selected the most suitable spatial scale by comparing the AICc values of models with each vegetation cover variable. We then ran another GLMM including the selected vegetation cover variable, distance to natural habitat, the total input of nitrogen and two-way interaction between N input and vegetation cover. To check if landscape management practices that potentially benefit ecosystem service providers also had an effect on crop pests, we repeated a similar procedure using pest densities as the response variable. As density of pests (that forage on flowers) was extremely low, we analyzed this data as presence-absence, using a logistic regression (binomial distribution). All model selection procedures focused on the variables of community composition and also included climatic data (temperature, humidity, and wind speed at the time of the survey) and observation time as covariates. All statistical analyses were performed using R version 3.3.1 using the R packages “lme4” and “MuMIn” [40,41].

Effect of ecosystem service providers and N input on bean production

Flower visitors and N input influenced both metrics of bean productivity, these effects being more accentuated for productivity per flower (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Effect of ecosystem services providers (pollinators and biocontrol agents) and nitrogen application on productivity per flower (g) and overall land productivity (kg ha-1) of common bean (P. vulgaris) farms.

https://doi.org/10.1371/journal.pone.0204460.t001

The positive effect of native pollinators on productivity was independent of flower visitor richness and no effect of biocontrol agents was detected, partly rejecting hypothesis 1. Contrary to our initial expectations, the density of exotic pollinators (i.e. A. mellifera) had a negative effect on productivity (Fig 1, Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Effect of the density of exotic pollinators (i.e. Apis mellifera) on common bean productivity.

Points represent partial residuals (i.e. variability not explained by the other variables included in the model). Model estimates are based on Model 2 from Table 1. Shaded area represents 95% confidence interval.

https://doi.org/10.1371/journal.pone.0204460.g001

In agreement with hypothesis 2, pollinator benefit to productivity was greater when lower levels of N were applied, the effect being even negative at very high levels of N input (Fig 2). When using N input levels recommended for common bean plantations (in rotation with soya) in Cerrado (60 kg ha-1), and when no exotic pollinator is present, productivity per flower obtained in areas with high pollinator densities (0.0165 per flower) more than doubled (143% increase) in comparison with areas with no pollinators (0.82 vs 2.01 g per flower, see equations in Table 1). The effect of native pollinators on overall productivity was similar (i.e. 2464.8 kg.ha^-1 with no native pollinators vs 5299.1 kg.ha^-1, if pollinator density was kept high throughout the field) but no interactive effect with N application was detected. No positive effect of N was detected on overall productivity nor on number of flowers produced, instead a negative effect was detected fro flower productivity (Figure B in S1 File). Also, the positive effect of this nutrient on productivity per flower (Table 1) was no longer detected above 72 kg/ha (Table F in S1 File).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Estimated effect of density of native pollinators on common bean productivity under different levels of nitrogen (N) input.

(N) Nitrogen application (varied from 36.0 and 130.5 kg ha-1 between fields). Model estimates are based on Model 4 from Table 1. Shaded area represents 95% confidence interval.

https://doi.org/10.1371/journal.pone.0204460.g002

Influence of fertilization practices and landscape characteristics on flower visitors

Climate effects on flower visitors were diverse and were taken into account when assessing the effect of nitrogen and landscape (Table G in S1 File). Farms with higher nitrogen input had more visitation by the exotic Apis mellifera (which negatively affects production, see above), lower abundance of biocontrol agents, and lower richness of ecosystem providers and lower probability of pest occurrence (Fig 3). No overall effect on the abundance of native pollinators was detected. For information on species level responses to N input, see Figure C in S1 File. Number of flowers observed in surveys was positively correlated with the abundance of A. mellifera, biocontrol agents, pest occurrence, and overall richness.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Effects of local management practices (nitrogen input and maintenance of native vegetation within agricultural landscape) on pollinators, biocontrol agents and species richness.

The points represent partial residual. The graphs are based on the variables selected on the best models (ΔAICc < 2), having used always the best model that includes most environmental variables (see Table G in S1 File). For the analyses of abundance of native pollinators the vegetation cover scale used was 2000 meters. For the remaining variables, the vegetation cover scale used was 500 meters (see details of vegetation cover scale selection in Table D in S1 File). Shaded area represents 95% confidence interval.

https://doi.org/10.1371/journal.pone.0204460.g003

As for landscape effects, the majority of the responsible variables (biocontrol agents, exotic pollinators, overall richness and pests) were most responsive to changes in vegetation cover at finer scales (500 meters, included in subsequent analyses), while native pollinators were most responsive at larger scales (2000 meters) (see Table D in S1 File). Native pollinator abundance was higher close to the margins with natural habitat, independently of the amount of natural habitat (Fig 3) available in the landscape (i.e. no interaction between cover and distance was detected). While landscapes with greater vegetation cover had fewer native pollinators, they also had more visitor species and higher abundance of biocontrol agents (Fig 3). When nitrogen input was low, the abundance of exotic honeybees declined with vegetation cover, and when N input was high the species abundance increased with vegetation cover.

Effect of ecosystem service providers and N input on bean production

Despite farmers in our study region heavily investing on conventional intensification (chemical inputs and large extensions of monoculture), we detected important benefits of native pollinators on productivity (more than doubled, see Table 1). The strong variation in flower production, here detected will also affect productivity. Therefore, it is essential to maintain both flower production and pollinator density high to achieve optimal results in overall yield. The negative effect of the exotic honeybee on production is likely related to the fact that this species frequently acts as a nectar robber. Such behaviour may decrease flowers attractiveness to legitimate pollinators and even damage reproductive parts of the flower [43], consequently, having a negative effect on common bean production. However, the negative effect of honeybee abundance on native pollinator abundance here detected was not significant (see Fig D in S1 File). Overall, our results indicate that there is potential for improving crop yields through ecological intensification. While common bean was described by Klein et al. [7] as having little dependence on pollinators (i.e. 0–10%), that classification was based on information on a diverse group of species within the genus Phaseolus. At the time of Klein’s publication few studies on pollination efficiency of common bean existed, but those mentioning Phaseolus vulgaris highlight the potential importance of flower visitors for pollination [20, 44, 45], and one detected benefits from pollinators of 13–35% [20].

More recent studies working with different cultivars of common bean also detected important benefits in yield from pollinators in Cameroon [21] (production of fully developed beans more than double) and Kenya [42]. Thus, future studies should consider common bean as a crop with, at least, medium level of dependence on pollinators (i.e. 10–40%). The fact that, in our study, pollinator’s positive effects were most accentuated under lower N input and even became negative above 100 kg ha-1 (Fig 2) suggests an alteration of the plant’s investment strategy (reproductive versus vegetative development) under high N availability [46]. Indeed, in our study system nitrogen input had negative effect on overall flower production (see Figure B in S1 File). Increased N availability may have also changed the quality of resources (e.g. due to changes in essential amino acid, alkaloids, or flower shape) [14]. Such changes could alter flower’s attractiveness to the different flower visitor species affecting pollination effectiveness. For example, in our study, we detected a positive effect of nitrogen input on honeybee abundance (Fig 3) that as discussed above has a negative effect on common bean production. Further surveys and more detailed analyses would be required to evaluate changes in the behaviour of this species (legitimate visitor vs. robber) and on the visitation rates and behaviour of other species of visitors.

The interactive effect between pollinator density and N availability here detected corroborates with the findings of a previous study [17], which found a tendency for greater benefit from pollination under low N availability in canola. However, in their study, the extreme reduction in nitrogen availability used (150 vs. 0 kg ha-1) may have limited pollinator benefits. Fruit formation requires a high level of nutrients and many studies (including our own) show that below certain levels of N, production is constrained (see Fig 2). The positive effect of nitrogen input on productivity could be partially due to a positive effect on flower production (and consequently, on number of pods). However, the effect of this nutrient on flower production was slightly negative in our work (see Figure B in S1 File). The positive effect is hence more likely explained by the high requirement on nutrients for fruit production of our study species. This demand might be higher when cross pollination mediated by insects occurs, since beans produced might have greater nutritious content (see [47] for an example with almond species). However, the fact that no positive effect of N input was detected in the sensitivity tests run only with points with N input higher than 72 kg/ha suggest that farmers are using unnecessarily high levels of nitrogen that do not benefit fruit formation, and reduce the effectiveness of pollinators. Indeed, given that all but one of our study fields had planted soya (and on the field with no soya, another nitrogen fixing plant was planted, Crotelaria sp.) in the last 2 years (see Table B in S1 File), the recommended dosage for common bean plantations in Cerrado is ca. 60 kg ha-1 of nitrogen [33].

As for biocontrol agents, the most plausible reason for the lack of effect on bean productivity is the intensive use of pesticides. Such practices are probably regulating the abundance of pests, which was low in our field sites (2.8% of beans damaged). Further studies are needed to test whether this group of species would effectively maintain common bean crops pests at low abundance if pesticides were applied in lower levels. In other crop systems, pollinator richness improves the benefits of pollinator density (e.g. [48, 49]). The lack of positive effects of richness on productivity in our study system could be because such effect may be diluted in situations of low abundance of species [50], which limits the ability to detect synergistic positive effects between the two variables. Indeed, when comparing with P. vulgaris fields of small scale farmers in Africa [21, 42], the mean flower visitation rates detected in our study were very low (0.004 vs. ca 0.151 visits/flower). Although the flower visitors’ assemblages and sampling efforts of the two study regions likely differ, this striking difference is most likely due to the negative pressure resulting from the strong agricultural intensification in our study area (i.e. large monoculture areas with high chemical input of insecticides, including neonicotinoids, organophosphates, pyrethroids, acetates and carbonates, farmers’ personal communication).

Influence of landscape characteristics and fertilization practices on flower visitors

The positive effect of vegetation cover within 500m radius on biocontrol agents and richness here detected suggests that small fragments of vegetation, adjacent to crop fields are essential to sustain the populations of these species. Such fragments are likely important nesting area for the bees and wasps detected in our study (see [51, 52]). For insects whose larvae can develop within crop fields (e.g. hoverflies), such patches can also be important alternative floral resources when common bean (and other crops) are absent from the landscape [53]. However, when common bean is flowering, those hoverflies depend less on native resources and more on availability of larval resources, possibly preferring to forage further away from natural habitat where they might be more protected from competition with other flower visitors and other pest natural enemies (e.g. wasps). This may explain the positive relationship between the density of biocontrol agents (a group dominated by one morphospecies of hoverfly, that is likely exotic, Allograpta exotica cf.) and distance to natural habitat (see also [54,55]. The positive effects of vegetation cover (2000km radius) on native pollinators could be due to a dilution effect, and reinforce the idea that conservation and restoration of natural habitats in the immediate surroundings of agricultural fields is important for ecological intensification of farming. The creation of hedgerows and flower strips may also bring important benefits facilitating movement of native pollinators [56, 57], and biocontrol agents (e.g. [58]).

Despite the variability of fertilizers application methods between farms, we detected a significant effect of nitrogen on visitor abundance and overall diversity of ecosystem providers. Nitrogen availability can alter amino acid content in pollen and nectar [14, 59], which can affect visitation of species that are attracted by flowers with specific amino acids contents (e.g. of proline, glycine, see [60, 61]). The fact that, in our study, different groups of visitors were affected differently by N availability (e.g. honeybee were positively affected, biocontrol agents negatively affected) suggests different preferences in terms of nectar or pollen quality. Indeed, Petanidou et al. [62] found that different species of flower visitors are differently affected by changes in amino acid content. Further studies evaluating the chemical composition of floral resources of common bean’ flowers would help to clarify their role as regulators of the effect of nitrogen on the attractiveness to the different bee species.