Ecology Letters, (2009) 12: 409–419

doi: 10.1111/j.1461-0248.2009.01296.x

LETTER

Plant–pollinator networks: adding the pollinatorÕs perspective

Jordi Bosch,1* Ana M. Martı´n Gonza´lez,1 Anselm Rodrigo1 and David Navarro2 1 CREAF – Ecology Unit, Universitat Auto`noma de

Barcelona, Edifici C, 08193 Bellaterra, Spain 2

Botany Unit, Universitat Auto`noma de Barcelona, Edifici C, 08193 Bellaterra, Spain *Correspondence: E-mail: [email protected]

Abstract Pollination network studies are based on pollinator surveys conducted on focal plants. This plant-centred approach provides insufficient information on flower visitation habits of rare pollinator species, which are the majority in pollinator communities. As a result, pollination networks contain very high proportions of pollinator species linked to a single plant species (extreme specialists), a pattern that contrasts with the widely accepted view that plant–pollinator interactions are mostly generalized. In this study of a Mediterranean scrubland community in NE Spain we supplement data from an intensive field survey with the analysis of pollen loads carried by pollinators. We observed 4265 contacts corresponding to 19 plant and 122 pollinator species. The addition of pollen data unveiled a very significant number of interactions, resulting in important network structural changes. Connectance increased 1.43-fold, mean plant connectivity went from 18.5 to 26.4, and mean pollinator connectivity from 2.9 to 4.1. Extreme specialist pollinator species decreased 0.6-fold, suggesting that ecological specialization is often overestimated in plant–pollinator networks. We expected a greater connectivity increase in rare species, and consequently a decrease in the level of asymmetric specialization. However, new links preferentially attached to already highly connected nodes and, as a result, both nestedness and centralization increased. The addition of pollen data revealed the existence of four clearly defined modules that were not apparent when only field survey data were used. Three of these modules had a strong phenological component. In comparison to other pollination webs, our network had a high proportion of connector links and species. That is, although significant, the four modules were far from isolated. Keywords Apparent specialization, coevolution, generalization, modularity, nestedness, plant– pollinator interactions, pollen analysis, pollination web, sampling effort. Ecology Letters (2009) 12: 409–419

INTRODUCTION

The study of plant–pollinator networks is becoming an increasingly important field of research. Plant–pollinator communities are typically composed of a high number of plant species and an even greater number of pollinator species. For this reason, deciphering the structure of plant– pollinator interactions is important to understand coevolutionary processes in species-rich communities (Bascompte & Jordano 2007). At the same time, a good assessment of the structure of plant–pollinator interactions is essential to evaluate the stability of pollination systems. This is especially important in the face of reported pollinator declines associated with anthropogenic influence (Biesmejer et al.

2006). In the last decade, network analysis has been extensively applied to the study of plant–pollinator interactions (Memmott 1999; Dicks et al. 2002; Olesen & Jordano 2002; Bascompte et al. 2003; Jordano et al. 2003; Va´zquez & Aizen 2004; Santamarı´a & Rodrı´guez-Girone´s 2007; Petanidou et al. 2008). These studies are based on pollinator surveys conducted at the community level, and provide a consistent view of the structure of plant– pollinator interactions: (1) plant–pollinator interaction matrices are sparse (Jordano et al. 2006; Va´zquez & Aizen 2006). That is, connectance (the proportion of all possible interactions in the community that actually occur) is low, suggesting a low degree of generalization at the community level. Connectance decreases with species richness; (2) the  2009 Blackwell Publishing Ltd/CNRS

410 J. Bosch et al.

distribution of connectivity (number of links per species) usually follows a power law or a truncated power law function (Jordano et al. 2003). There are few extreme generalists and a high number of specialists. The frequency of extreme specialists and the connectivity of extreme generalists are much higher than expected if interactions were randomly organized (Va´zquez & Aizen 2004). Connectivity increases slowly with species richness; (3) pollination networks tend to have a centralized structure, with a core of highly connected species to which many peripheral specialist species are connected. Centralization increases with increasing differences among species in connectivity (Wasserman & Faust 1994; de Nooy et al. 2005); (4) plant– pollinator networks are nested. Generalists interact with both generalists and specialists, but few interactions occur between specialists (Bascompte et al. 2003); (5) plant– pollinator networks (at least those above a certain size) are organized in compartments or modules (Olesen et al. 2007). Modules are groups of plant and pollinator species with high levels of within-group connectivity that are poorly connected to species outside the group. The high proportion of species with one or few links in pollination networks is at odds with the generally accepted view that generalization, rather than specialization, is the norm in pollination systems. Most plants are visited, and often pollinated, by a diverse array of pollinators (Herrera 1996; Waser et al. 1996; Fenster et al. 2004; Go´mez & Zamora 2006). Similarly, most pollinators exploit floral resources from a variety of plant species. Most (67–74%) bee species are polylectic, collecting pollen from various plant families (Westrich 1989; Minckley & Roulston 2006). The remaining species (oligoleges or pollen specialists) restrict their pollen collection to one plant family or, sometimes, one genus. Even among oligoleges, females are known to sometimes visit alternative plant species for nectar (Westrich 1989; Cane & Sipes 2006; Minckley & Roulston 2006; Petanidou et al. 2008). Other insect pollinator groups (flies, beetles, wasps, butterflies) are even less specialized (Proctor et al. 1996). The high number of extreme specialists recorded in pollination networks can be explained in two ways. First, generalist pollinator species could, at a local scale, restrict their visits to a single plant species, thus behaving as specialists (ecological vs. evolutionary specialization; Armbruster 2006). Second, pollinator visitation could be insufficiently sampled. In the 1990s, several studies (Cohen et al. 1993; Goldwasser & Roughgarden 1997; Martı´nez et al. 1999) showed that several properties considered typical of food webs were artefacts of incomplete sampling. Pollination network studies are based on plant-centred visual surveys of plant–pollinator contacts. Individual plants are monitored and pollinators landing on their flowers are identified. Pollinator abundance usually follows a right 2009 Blackwell Publishing Ltd/CNRS

Letter

skewed distribution, with a few abundant species and a high number of rare species, to the point that many species are recorded only once (Williams et al. 2001; Petanidou & Potts 2006; Go´mez et al. 2007). These rare pollinator species appear necessarily as extreme specialists when in fact some of them could be wide generalists. A recent study showed striking reductions in apparent extreme specialists when a plant–pollinator community was sampled over 4 years (Petanidou et al. 2008). Sampling bias may thus account for some of the sparseness found in pollination networks, and therefore influence the degree of specialization perceived in pollination networks. Va´zquez & Aizen (2003) identified this problem, and showed a positive correlation between interaction frequency ( f ) and connectivity (s). In a subsequent study (Va´zquez & Aizen 2006), these authors showed that the f –s relationship and the distribution of specialization were robust to reductions of sampling effort (see also Nielsen & Bascompte 2007). It remains to be seen whether increases in sampling effort lead to higher connectivity increases in rare species compared to abundant species, resuting in a flattening of the f –s relationship. Greater increases in connectivity of rare species compared to abundant species may in turn affect the degree of asymmetrical specialization (generalists interacting with specialists), and thus nestedness and centralization. When studying pollination webs, there are two possible ways to increase our capacity to approach connectance levels occurring in nature. First, sampling effort may be increased by investing more time in visual surveys. Second, pollen from the body of pollinators may be identified (Kanstrup & Olesen 2000; Forup et al. 2008). Because pollen grains remain on the body of pollinators for long periods (Courtney et al. 1981), pollen analysis provides a record of extended visitation history, rather than a snapshot of a single interaction. This method is analogous to the use of faecal analysis in seed dispersal systems and the use of stomach contents in food webs. In this study we analyse the structure of plant–pollinator interactions in a Mediterranean scrubland community based on plant-centred visual surveys and pollinator-centred pollen analyses. We ask two basic questions: (1) Is pollen analysis a good method to unveil interactions undetected in plant-centred surveys? (2) If so, does the structure of the pollination network change when interaction resolution is increased via pollen analysis? We expect that the addition of pollen data will provide a more complete view of the interaction network, resulting in higher connectance and connectivity. Because the increase in connectivity is likely to be higher for rare species (poorly sampled in visual surveys), we expect a flattening of the f –s relationship. Nestedness has been shown to either increase (Bascompte et al. 2003) or decrease (Rodrı´guez-Girone´s & Santamarı´a 2006) with increased connectance. In consequence of our prediction of a greater connectivity increase

Letter

for rare (apparent specialist) species, we expect nestedness and centralization to decrease. The addition of pollen data is also likely to affect the degree of modularity. The predicted increase in connectivity could either accentuate or dilute modularity, depending on whether new links (those unveiled by pollen data) are mostly within or between modules.

METHODS

Field surveys

We worked at a Mediterranean scrubland in Garraf Natural Park (Barcelona, NE Spain), 340 m above sea level and 1700 m from the coastline. Field work was conducted from early March to late June, encompassing the main flowering period in the area. No plants were in bloom during the dry summer season (July–August). We counted weekly the number of open flowers in six 50 · 1 m transects and conducted pollinator counts on 19 plant species. These 19 species represented 99.96% of the total number of flowers counted. Fifteen of these species accounted for < 1% of the total number of flowers counted. Pollinator activity periods tend to be short in Mediterranean ecosystems (Bosch et al. 1997; Petanidou & Potts 2006). In an attempt to minimize the number of missed interactions, surveys were conducted twice a week throughout the blooming period. Flower patches were tagged and observed for 4 min, during which time all insects visiting the flowers were visually identified or captured for later identification and pollen analysis. Total field sampling time was 107 h. Pollen analysis

We captured 495 pollinator specimens (mean ± SE: 4.08 ± 0.46 per species). Captured pollinators were individually placed in clean kill-vials. To obtain pollen samples, small cubes of fuchsine-stained gelatin were rubbed over the body of the insect and mounted on glass slides. To avoid pollen contamination, laboratory utensils and work surfaces were cleaned every time. Pollen grains were identified at 400· with the aid of a reference collection from the study area. In slides with sparse pollen grains (200 or fewer), all of them were identified. In densely populated slides, all pollen grains in 10 evenly spaced bands covering approximately half of the cover-slip surface were identified. Pollinator specimens and pollen slides are deposited in the CREAF collection. An individual pollinator visiting a given plant species could pick up heterospecific pollen dropped by another pollinator that had previously visited other plant species. To assess this possibility, we used the above-described methodology to analyse pollen from the body of 18 female

Pollination networks 411

specimens of the bee Hoplitis adunca (Panzer). This species is known to be monolectic on Echium throughout its distribution range, although it may occasionally collect nectar from other species (Westrich 1989; Bosch et al. 2001). At the same time, the local Echium species, Echium vulgare L., is visited by a wide array of pollinators, mostly polylectic bees, beetles and dipterans (J. Bosch, unpublished data). We analysed all pollen grains in these samples. The number of non-Echium pollen grains per specimen averaged 1.56, and ranged from 0 (six specimens) to 5 (one specimen). In most cases in which two or more non-Echium pollen grains were present, they belonged to different species. We thus established the presence of 10 pollen grains from a given plant species in our pollen counts as proof of visitation to that species. A few samples from small pollinators (ants, small beetles) had < 100 pollen grains. In these cases, we included as plant hosts all species with ‡ 10% pollen grains in the sample. Network structure

We observed 122 pollinator species visiting the 19 plants surveyed and we captured at least one specimen of each pollinator species. We built three plant–pollinator matrices with data from field surveys (matrix F), pollen analysis (matrix P), and both types of data (matrix FP). For each network, we calculated the following parameters. Connectance Connectance is the proportion of observed links divided by the number of total possible links. The expected increase in connectance with the addition of pollen data could result from differences in the sampling approach (field survey data vs. pollen data) and ⁄ or from increased sampling effort. To differentiate these two effects we used ESTIMATES ( 2006, Robert K. Colwell) to generate individual-based rarefaction curves of the expected accumulation of interactions using the Coleman method (Gotelli & Colwell 2001) with the three data sets (F, P and FP). Connectivity Connectivity is the number of interaction partners of a species, a measure of its degree of generalization. To characterize connectivity distribution, we used the R package to fit exponential, power law and truncated power law models to the distribution of connectivity of our three matrices (Jordano et al. 2006). We also explored the relationship between f (interaction frequency) and s (connectivity; Va´zquez & Aizen 2006). We used ANCOVA to compare the slopes of the regression lines between log s and log f of the three matrices for plants and pollinators separately. As mentioned, we expected that pollen data would produce a flatter slope. We also expected a lower  2009 Blackwell Publishing Ltd/CNRS

412 J. Bosch et al.

increase in connectivity in species with high s in the F matrix, simply because these were already highly connected, and thus closer to their potential maximum connectivity. Degree centralization We used PAJEK 11.5 (http://pajek.imfm.si/doku.php) to calculate degree centralization (DC), a measure of the level of centralization in the network. Degree centralization measures differences in connectivity among species (Wasserman & Faust 1994; de Nooy et al. 2005). Maximum centralization (DC = 1) is reached in star networks, with a central node linked to the rest of the nodes, which are not linked between themselves (Borgatti & Everett 1999). Nestedness We used the software ANINHADO (Guimara˜es & Guimara˜es 2006) to calculate nestedness based on two metrics: temperature-based relative nestedness (N *; Atmar & Patterson 1993; Bascompte et al. 2003) and nestedness based on overlap and decreasing fill (NODF; Almeida-Neto et al. 2008). Both matrix temperature and NODF were tested for significance against 1000 random matrices built by the Ce model of ANINHADO, which assigns each species a probability of interaction based on its connectivity (Bascompte et al. 2003; Guimara˜es & Guimara˜es 2006). Matrix temperature has been shown to be more sensitive to differences in matrix shape and size than NODF (AlmeidaNeto et al. 2008). However, we provide N* to allow for comparisons with previous studies. Modularity We calculated M, an index of modularity that measures the extent to which species have more links within their modules than expected if linkage was random (Guimera` & Amaral 2005). For each species, we calculated the standardized within-module degree (z) and among-module connectivity (c). Based on z and c values, species can be assigned a network role and classified as peripheral (low z and low c), connectors (low z and high c), module hubs (high z and low c) and network hubs (high z and high c) (see Olesen et al. 2007 for z and c values defining each category). RESULTS

Our visual survey recorded 4265 individual plant–pollinator contacts involving 122 pollinator species and 19 plant species, and representing 351 specific interactions. The analysis of pollen from the body of the 495 pollinator specimens captured provided evidence for 355 interactions. Of these, 205 were common to the F data set and 150 were new. On the other hand, 146 interactions were only recorded in the field surveys. Thus, the combination of field survey and pollen data resulted in a 1.43-fold  2009 Blackwell Publishing Ltd/CNRS

Letter

Table 1 Parameters describing the structure of the Garraf polli-

nation network based on field surveys (F), pollen analysis (P), and field surveys + pollen analyses (FP)

Plant species Animal species Interactions recorded Connectance Mean plant connectivity (sP) Mean animal connectivity (sA) % Extreme animal specialists  Relative nestedness (N *) Nestedness based on overlap and decreasing fill (NODF) Degree centralization (DC) Modularity (M) Number of significant modules

F network

P network

FP network

19 122 351

18 107 355

19 122 501

15.14 18.47

18.43 19.72

21.61 26.37

2.88

3.32

4.11

45.9

29.9

27.9

0.27

0.38

0.49

25.99**

31.69**

38.28**

0.20

0.30

0.30

0.36 NS 0

0.36** 5

0.30** 4

 One-link species, **P < 0.001, NS non-significant

interaction increase (501 interactions). We found no pollen in 63 of the specimens captured, mostly corresponding to small species (ants, small beetles). We found no pollen of one of the surveyed plant species (Iris lutescens Lam.) on any of the captured specimens. Thus, the P matrix was smaller (18 plant species, 107 pollinator species) than the F and FP matrices (Table 1). Some pollinator specimens carried pollen of species not found in our study area (not included in the analyses). Connectance was 15.14 in F, 18.43 in P, and 21.61 in FP (Table 1, Fig. 1). The increased connectance in FP was not solely due to increased sampling effort. Expected interaction richness increased much more rapidly in P than in F or FP (Fig. 2). Average plant connectivity increased from c. 19 in F and P to 26.4 in FP, and average pollinator connectivity from c. 3 to 4.1. The percentage of extreme specialist (onelink) pollinator species decreased from 45.9% in F, to 29.9% in P and 27.9% in FP (Table 1, Fig. 1). This dramatic decrease in specialization did not result in changes in the shape of the connectivity distribution. Pollinator connectivity distribution followed a power law, and plant connectivity distribution a truncated power law in all three matrices (Fig. 3).

Letter

Pollination networks 413

F MATRIX

Figure 2 Individual-based rarefaction curves (mean ± 95% confi-

P MATRIX

dence intervals) showing the expected interaction richness for field survey (F), pollen analysis (P), and field survey + pollen analysis (FP) data.

MODULE 1 MODULE 2 MODULE 3 MODULE 4 MODULE 5

FP MATRIX

MODULE 1 MODULE 2 MODULE 3 MODULE 4

Figure 1 Representation of the Garraf pollination networks

obtained from field surveys (F), pollen analysis (P), and field surveys + pollen analysis (FP). Dots represent pollinator species and squares plant species. No significant modules were found in F.

Connectivity increased with interaction frequency ( f ), for both plants (linear regression, log-transformed data; F matrix: R2 = 0.81, P < 0.0001, slope = 0.454; P matrix: R2 = 0.49, P < 0.0008, slope = 0.456; FP matrix: R2 = 0.76, P < 0.0001, slope = 0.447) and pollinators (F matrix: R2 = 0.66, P < 0.0001, slope = 0.338; P matrix: R2 = 0.25, P < 0.0001, slope = 0.245; FP matrix: R2 = 0.51, P < 0.0001, slope = 0.324). We expected that pollen analysis would result in greater increases in connectivity for rare species (poorly sampled in visual surveys), and

Figure 3 Cumulative distribution of connectivity (number of links

per species) for pollinators and plants in Garraf, based on field surveys (F), pollen analysis (P), and field surveys + pollen analysis (FP). Each diamond may represent more than one species. Plant connectivity distribution follows a truncated power law (F matrix: c = 1.40; P matrix: c = 1.36; FP matrix: c = 1.33), and pollinator connectivity distribution a power-law (F matrix: c = 2.40; P matrix: c = 2.09; FP.08; FP matrix: c = 1.92).

thus a flattening of the f –s relationship. The slopes of the three plant f –s regression lines were almost identical (F2,51 = 0.002; P > 0.9), but differences in slope among pollinator regression lines (lower in P) approached  2009 Blackwell Publishing Ltd/CNRS

414 J. Bosch et al.

significance (F2,360 = 2.78; P > 0.07). We expected species with low s (as obtained in F) to experience greater increase in s (from F to FP). However, both plants and pollinators showed positive, albeit weak, relationships between these two variables (linear regression, log-transformed data; plants: R2 = 0.26, P = 0.027; pollinators: R2 = 0.063, P < 0.005). Because many pollinator species were rare, the number of specimens used to obtain pollen samples varied among species. Pollinator increase in s was positively related to number of specimens sampled (linear regression, logtransformed data: R2 = 0.23, P < 0.0001), suggesting that connectivity could still be increased, at least for rare species. The degree centralization was low in the three matrices and, contrary to our expectations, increased with the addition of pollen data (Table 1, Fig. 1), strengthening differences among species in connectivity. Similarly, nestedness did not decrease with the addition of pollen data. All three matrices were significantly nested and, in fact, NODF and N* tended to increase with the addition of pollen data (Table 1). Thus, despite the overall increase in connectivity, interactions between specialists remained scarce. We found no significant modularity in our F matrix (M = 0.364; Mrandom = 0.366 ± 0.0060; P = 0.40). Conversely, the P matrix yielded five significant modules (M = 0.364; Mrandom = 0.328 ± 0.0061; P < 0.001), and the FP matrix four (M = 0.303; Mrandom = 0.280 ± 0.00488; P < 0.001) (Table 1, Fig. 1). Modules 1 and 2 of FP and P were very similar (included the same plant species and a high proportion of the same pollinator species). Module 3 of FP approximately coincided with modules 3 and 4 of P. Module 4 of FP was highly coincidental with module 5 of P. The first module of FP included five early-blooming (March–April) plant species (Rosmarinus officinalis L., Thymus vulgaris L., Euphorbia flavicoma DC, Ranunculus gramineus L., Biscutella laevigata L.), representing a wide range of flower morphologies and pollen–nectar production (J. Bosch, A. Rodrigo, A. M. Martı´n Gonza´lez, unpublished data). These species were visited by a diverse array of bees, ants and dipterans, and, with the exception of R. gramineus, had high flower densities. The second module included two April-blooming species (Cistus albidus L., Cistus salvifolius L.). Both species had shallow flowers, produced large amounts of pollen, and were primarily visited by beetles. The third module included six late-flowering (May–June) species (Dorycnium hirsutum Ser., Convolvulus althaeoides L., Centaurea paniculata L., Centaurea linifolia L., Leuzea conifera (L.) DC, Galium aparine L.). These species varied in flower morphology and pollen–nectar production, and were visited by a diverse array of bees, beetles, butterflies and ants. The last module included six plant species [I. lutescens, Muscari neglectum Guss. ex Ten., Orobanche lastisquama (F.W. Schultz) Batt., Gladiolus illyricus Koch, Sideritis hirsuta L., Phlomis  2009 Blackwell Publishing Ltd/CNRS

Letter

Table 2 Within- and between-module link gain with the addition

of pollen data in species belonging to the various modules of the Garraf pollination network

Species Within-module links gained Between-module links gained % Within-module link gain* % between-module link gain 

Module 1

Module 2

Module 3

Module 4

45 53

27 12

51 20

18 10

29

13

4

9

40.5 ± 4.9 40.9 ± 8.9 9.5 ± 1.8 19.3 ± 5.5 1.7 ± 0.5

3.9 ± 1.2 4.1 ± 1.0 11.1 ± 4.3

*Per cent of potential within-module links gained with the addition of pollen data (mean ± SE).  Per cent of potential between-module links gained with the addition of pollen data (mean ± SE).

lychnitis L.] with little overlap in flowering period. These species were scarce and patchily distributed (with the exception of S. hirsuta), produced large amounts of nectar (with the exception of I. lutescens), and were visited primarily by bees with long activity periods. Collectively, their flowering periods encompassed the entire season. Interestingly, plant species of module 4 had low connectivity (mean ± SE: 10.83 ± 2.87) compared with plant species in other modules (modules 1, 2 and 3 = 36.60 ± 4.50, 43.50 ± 5.50 and 27.67 ± 4.50, respectively; Kruskal–Wallis H3 = 11.54; P = 0.009), whereas pollinator species of module 4 were among those with highest connectivity (modules 1, 2 and 3 and 4 = 4.03 ± 0.46, 2.80 ± 0.37, 3.58 ± 0.43 and 9.08 ± 1.68, respectively; Kruskal–Wallis H3 = 12.625; P = 0.005). Also interestingly, module 4 occupied a central position within the network, with only one extreme specialist (Fig. 1). Most new links revealed by pollen data involved species of module 1 (Table 2). Overall, most new links were between species belonging to the same module (Table 2), which explains the switch from non-modularity in F to modularity in FP. For each plant and pollinator species, we calculated potential within-module link gain as the difference between the total number of possible within-module interactions and the within-module interactions obtained in the F matrix. We then calculated % within-module link gain as the percentage of potential within-module links actually gained with the addition of pollen data. We followed the same rationale to calculate potential between-module link gain and % betweenmodule link gain for each species. Mean % within-module link gain was highest for species in modules 1 and 2 (Table 2; ANOVA, F3,137 = 11.266, P < 0.0001). Conversely, species in module 4 experienced the highest % between-module link gain (Table 2; ANOVA, F3,137 = 5.488, P = 0.001).

Letter

Figure 4 Distribution of plant and pollinator species according to

their network role (Olesen et al. 2007) in the Garraf FP matrix (field survey + pollen analysis). MH, module hub; NH, network hub; P, peripheral; C, connector. Squares and triangles may represent more than one species.

All module and network hubs (five and two species, respectively) were plant species, and belonged to modules 1, 2 or 3. On the other hand, three of the four connector plant species belonged to module 4. Most (‡ 80%) pollinator species in modules 1, 2 and 3 were peripherals, but the majority (83.3%) of pollinator species in module 4 were connectors (Fig. 4). Most (53.6%) connector species were bees, which were also the only group with similar numbers of connector and peripheral species (15 and 19, respectively). All other pollinator groups had much higher proportions of peripherals. DISCUSSION

When added to a recently published list of 29 pollination matrices (Table 8.1 of Jordano et al. 2006), our F matrix ranks 12th in species number and 8th in connectance. Nine of those 29 matrices have a size similar to ours (potential interactions ranging from 1820 to 2916, compared to 2318 in our study). Connectance of our F matrix (0.1514) ranks second when compared to this subgroup of similar-sized matrices (with a mean connectance of 0.0983). Thus, the sampling effort and sampling frequency (twice a week) employed in our field survey are highly satisfactory by current standards. Yet, pollen analysis revealed a very significant number of interactions undetected in the field survey. Interestingly, the opposite is also true, a nonnegligible fraction of the interactions observed in pollen surveys were not detected in pollen analyses. This is mostly attributable to the fact that we only captured a small sample (c. 12%) of the pollinators observed. In other words, for a given pollinator species, we did not capture specimens on all the plants on which we saw that species. At the same time, some captured specimens yielded no pollen of the species

Pollination networks 415

on which they were caught. These specimens were mostly ants and small beetles with sparse hairiness, typically carrying few pollen grains, but also larger pollinators caught on C. althaeoides and I. lutescens. These two species have wide corolla apertures, so that visitors can easily reach the nectaries without contacting the anthers. Other studies analysing pollen from the body of pollinators also found specimens that carried no pollen from the study community or no pollen at all (Kanstrup & Olesen 2000; Forup et al. 2008). Therefore, pollen analysis should not be regarded as a substitute for visual surveys, but rather as a complementary method. Pollen analysis allows for the detection of links involving rare species and species with low visitation rates, which would require very long observation periods in the field. In addition, because most pollinator specimens carry pollen from more than one species, interaction curves can be saturated with fewer specimens when using pollen analysis. On the other hand, pollinators could pick up heterospecific pollen (see methods and materials), and unlike visual surveys, pollen analysis does not provide a measure of visitation frequency. Finally, preparation of pollen slides and identification of pollen grains is time consuming, and pollen grains of similar species may be difficult to tell apart. Ideally, the two approaches should be combined to provide a complete picture of plant–pollinator interactions. The addition of pollen data to our field survey cut down the number of extreme specialists to almost one half, and increased connectance 1.43-fold. Pollinator connectivity in our FP matrix was 4.11, compared to a mean 2.77 (range: 1.23 to 6.00) in the 29 networks reported in Jordano et al. (2006). Plant connectivity was 26.37 in our FP matrix, compared to a mean 8.02 (range: 2.25 to 22.39) in Jordano et al.Õs table. Connectance of the nine networks with sizes similar to ours averaged 10.07 (range: 7.12 to 19.35) compared to 21.61 in our FP matrix. These results strongly indicate that plant–pollinator networks are usually undersampled. Petanidou et al. (2008) reached a similar conclusion in a recent study in which sampling effort was increased by sampling the same community for 4 years. Connectance of their overall network (4 years pooled together) was 1.23 times higher than the average connectance of individual years. When only species recorded in all 4 years were considered, connectance increase was 2.29-fold. As many as 90% of the species perceived as extreme specialists one year became generalists when data from other years were added (Petanidou et al. 2008). Pollen data revealed an important number of interactions involving rare pollinator species. Because these species are only rarely seen in surveys, it is difficult to gather enough information on their flower visitation habits. It might be argued that the ecological function of these interactions is low, in terms of contribution to pollination and ⁄ or pollen–  2009 Blackwell Publishing Ltd/CNRS

416 J. Bosch et al.

nectar exploitation. However, the proper characterization of floral specialization of rare species is essential both to understand pollination network structure and to design efficient measures to conserve pollinator diversity, given that rare species are the majority in pollinator communities (Williams et al. 2001; Minckley & Roulston 2006; Petanidou & Potts 2006; Go´mez et al. 2007). In addition, the ecological importance of some rare species may be much greater than predicted from their abundance (Power et al. 1996), which may change over time (Goldwasser & Roughgarden 1997). In our study area, the bumblebee Bombus terrestris (L.) was a rare pollinator in 2006 (three individuals recorded), but became a dominant species in the following 2 years (J. Bosch, A. Rodrigo, A. M. Martı´n Gonza´lez, unpublished data). Pollen data also revealed an important number of interactions involving abundant species. Six of the 10 most frequently recorded pollinator species had their s increased by 1–3 links with the addition of pollen data. This result reflects several important traits of pollinator foraging behaviour. First, although emphasis is usually placed on flower fidelity, pollinator foraging behaviour is best described as opportunistic (Waser et al. 1996). Departures from strict flower constancy are common in all major insect pollinator groups (Heinrich 1976; de los Mozos Pascual & Medina Domingo 1991; Goulson et al. 1997; Goulson & Wright 1998), and were often observed in our study area. Bumblebees, honey bees, hoverflies, beeflies and butterflies have been shown to switch hosts when a more rewarding species becomes available (Heinrich 1979; Toft 1984; Kunin 1993; Chittka et al. 1997; Goulson et al. 1997). Second, pollinators, particularly species with multiple generations or with long-lived individuals, often change hosts throughout their activity period (Stephen et al. 1969; Heinrich 1976; Medan et al. 2006). We found a positive relationship between connectivity and duration of activity period in our pollinator community (n = 122, R2 = 0.48, P < 0.001). Third, pollinators may exploit different plant species for different resources (Heinrich 1976; Williams & Tepedino 2003). In our study area it was not uncommon to observe female bees flying from a pollen-rich source (Ranunculus, Cistus) to a nectar-rich source (Rosmarinus, Thymus). Even oligolectic bees are known to visit alternative species for nectar (Westrich 1989; Cane & Sipes 2006; Minckley & Roulston 2006). Finally, different individuals within a species may favour different hosts (individual specialization; Heinrich 1976). Variability in pollinator visitation habits is also evident at a wider temporal (inter-annual) scale. None of the pollinator species in Petanidou et al.Õs (2008) study was completely consistent in plant choice over the four sampling years. Our results and those of Petanidou et al. demonstrate that the high numbers of extreme specialists found in pollination networks are, for the most part, not attributable to ecological specialization.  2009 Blackwell Publishing Ltd/CNRS

Letter

We predicted that rare species (necessarily appearing as specialists; Va´zquez & Aizen 2006) would be those experiencing higher increases of s. The flatter slope of the pollinator f–s regression line in the P matrix compared to the F matrix pointed in that direction, but these differences were diluted in the FP matrix. At the same time, and contrary to prediction, species with low s in the F matrix did not experience greater increase in s. There are two reasons for these results. First, 15 pollinator species yielded no pollen records. These species were rare (and represented by only one specimen in our pollen analyses), thus lowering increase in s for rare species. Second, links gained by pollinators with low s in the F matrix were mostly to abundant, generalist plants. That is, new links followed a rule of preferential attachment to already highly connected nodes (Bascompte & Jordano 2007) and, consequently, asymmetric specialization was not reduced in our FP matrix. This result is congruent with findings that specialist pollinators usually specialize in plants that attract numerous other species (including generalist pollinators) (Minckley & Roulston 2006; Petanidou & Potts 2006). Previous studies have shown that nestedness in mutualistic networks is robust to reductions in sampling effort (Jordano et al. 2006; Nielsen & Bascompte 2006; Va´zquez & Aizen 2006). Our study and that of Petanidou et al. (2008), who found no differences in nestedness between 1-year and 4-year data sets, confirm that nestedness is robust to increases in sampling effort and should thus be regarded as a fundamental property of plant– pollinator networks (Bascompte et al. 2003). Specialization is a successful strategy only when the interaction partner is consistently abundant (Waser et al. 1996). A recent study (Va´zquez et al. 2007) proposes that asymmetry in interaction networks may be explained, at least in part, by the distribution of species abundance. If interactions occurred randomly, abundant species would be highly connected simply because they have frequent encounters, whereas rare species would be less connected simply due to their rareness. Under these circumstances, and because both host and consumer communities are characterized by a predominance of rare species, asymmetric interactions are bound to be common. This scenario is particularly applicable to our highly connected network. In Garraf, s (as obtained in FP) was positively correlated with both flower and pollinator abundance (measured in six 50 · 1 m transects; pollinators: n = 122, R2 = 0.47, P < 0.001; plants: n = 19, R2 = 0.27, P = 0.02; log-transformed data). An important result of our study is the switch from a non-modular to a modular network structure. In an analysis of 51 pollination networks (Olesen et al. 2007) all matrices with > 150 species were modular, whereas all matrices with < 50 species were non-modular. Our network contains 141 species, and falls within the intermediate size range including both modular and non-modular matrices in Olesen et al.Õs

Letter

study. Three of the modules obtained in our FP matrix had a clear seasonal component. This result is consistent with the short activity period (c. 1 month) of most pollinator species, coupled with the high connectivity obtained in our FP matrix. This result is also congruent with other studies in Mediterranean scrubland communities, in which pollinator distribution was mostly driven by flowering phenology, rather than flower traits (Herrera 1988; Bosch et al. 1997). Interestingly, the only two network hubs in the Garraf web were the two plant species of module 2, which bloom in mid-season, further reflecting a strong seasonal component. Pollen data were particularly useful to reveal within-module interactions in module 1. Flower and pollinator availability in the Garraf community follows closely the changing floral market model of Cohen & Shmida (1993). Early-blooming plants in our study area typically produce large amounts of flowers per individual, and flower availability is very high in March and early April. Plants blooming during this period experience very low visitation rates (visits per flower and minute). Instead, later-blooming species produce few flowers, at a time when pollinator numbers increase very significantly. As a result, visitation rates experience a dramatic increase in the second half of the blooming period ( J. Bosch, A. Rodrigo, A. M. Martı´n Gonza´lez, unpublished results). Pollen data were thus particularly useful to increase the connectivity of early-booming plants for which field surveys provide a less complete record of interactions. Two important characteristics of our network, in comparison with the networks analysed in Olesen et al. (2007), are the relative scarcity of species with 0 between-module connectivity (41% vs. 72%) and the high proportion of connector species (23% vs. 11%). In other words, our FP network was significantly modular, but these modules were far from isolated. Between-module connection was mostly achieved through species of module 4, which had an unusually high proportion of connector species (Fig. 4) and occupied a central position within the network (Fig. 1). This Ôtransversal moduleÕ was composed of plants with little blooming overlap that, for the most part, produced large amounts of nectar, had low flower density and a patchy distribution. The main pollinators of this module were highly generalist bees with long activity periods [Rhodanthidium sticticum (Fabricius), Rhodanthidium septemdentatum (Latreille), Lasioglossum bimaculatum (Dours)]. In conclusion, the use of pollen data allowed us to unveil an important number of interactions undetected in plantcentred field surveys. This addition resulted in very significant changes in some fundamental properties of network structure, and provided a more complete view of pollination networks. Importantly, these changes did not only result from increased sampling effort, but were also a consequence of the complementarity of pollen and field survey data. Several of these changes confer greater

Pollination networks 417

robustness to the plant–pollinator community. First, ecological generalization implies that following the extinction of one species, its partners are likely to do well exploiting other species. Second, the modular structure obtained in the FP network hinders the spread of extinction effects throughout the entire web, making cascading processes less likely. Third, increased nestedness strengthens the core of highly connected species, which, coupled with increased connectance of peripheral species, provides a cohesive structure to the network (Jordano et al. 2006). Our results also have important evolutionary consequences. The dramatic increase in connectivity strongly indicates that specialization in plant–pollination webs is routinely overestimated and that, for the most part, pollinators do not behave as ecological specialists, therefore hindering the appearance of tight coevolutionary processes. The observed increased in connectivity is attributable to both inter-year variability (Petanidou et al. 2008), and within-year flexibility in flower choice (our study). A better understanding of pollinator foraging patterns will help determine the relative importance of the various mechanisms explaining generalization in flower visitation, such as incomplete flower constancy, sequential exploitation of different species, contemporaneous exploitation of different species for different floral resources and individual variability. ACKNOWLEDGEMENTS

H. Barril provided invaluable assistance in all phases of the study. C. Romeu and C. Primante contributed to data collection and R. Molowny helped with network analysis. We are grateful to P. Jordano, J. M. Olesen, J. Retana, T. Lewinsohn and an anonymous reviewer, as well as the subject editor (J. Bascompte), for their valuable comments, which greatly improved the manuscript. This study was supported by a CICYT grant (CGL2005-00491) of the Spanish MEC, a 6th Framework Programme grant (project 044346) of the European Commission, and a FPI scholarship to A. M. Martı´n Gonza´lez (BES 2006-13332). REFERENCES Almeida-Neto, M., Guimara˜es, P., Guimara˜es, P.R. Jr, Loyola, D.R. & Ulrich, W. (2008). A consistent metric for nestedness analysis in ecological systems: reconciling concept and measurement. Oikos, 117, 1227–1239. Armbruster, W.S. (2006). Evolutionary and ecological aspects of specialized pollination: views from the Arctic to the Tropics. In: Plant–Pollinator Interactions, from Specialization to Generalization (eds Waser, N.M. & Ollerton, J.). University of Chicago Press, Chicago, IL, pp. 260–282. Atmar, W. & Patterson, B.D. (1993). The measure of order and disorder in the distribution of species in fragmented habitat. Oecologia, 96, 373–382.

 2009 Blackwell Publishing Ltd/CNRS

418 J. Bosch et al.

Bascompte, J. & Jordano, P. (2007). Plant-animal mutualistic networks: the architecture of biodiversity. Annu. Rev. Ecol. Evol. Syst., 38, 567–593. Bascompte, J., Jordano, P., Melia´n, C.J. & Olesen, J.M. (2003). The nested assembly of plant–animal mutualistic networks. Proc. Natl Acad. Sci. USA, 100, 9383–9387. Biesmejer, J.C., Roberts, S.P.M., Reemer, M., Ohlemu¨ller, R., Edwards, M., Peeters, T. et al. (2006). Parallel declines in pollinators and insect-pollinated plants in Britain and the Netherlands. Science, 313, 351–354. Borgatti, S.P. & Everett, M.G. (1999). Models of core ⁄ periphery structures. Soc. Networks, 21, 375–395. Bosch, J., Retana, J. & Cerda´, X. (1997). Flowering phenology, floral traits and pollinator composition in a herbaceous mediterranean plant community. Oecologia, 109, 583–591. Bosch, J., Maeta, Y. & Rust, R. (2001). A phylogenetic analysis of nesting behavior in the genus Osmia (Hymenoptera: Megachilidae). Ann. Entomol. Soc. Am., 94, 617–627. Cane, J.H. & Sipes, S. (2006). Characterizing floral specialization by bees: analytical methods and a revised lexicon for oligolecty. In: Plant–Pollinator Interactions, from Specialization to Generalization (eds Waser, N.M. & Ollerton, J.). University of Chicago Press, Chicago, IL, pp. 99–122. Chittka, L., Gumbert, A. & Kunze, J. (1997). Foraging dynamics of bumblebees: correlates of movements within and between plant species. Behav. Ecol., 8, 239–249. Cohen, D. & Shmida, A. (1993). The evolution of flower display and reward. Evol. Biol., 27, 197–243. Cohen, J.E., Beaver, R.A., Cousins, S.H., De Angelis, D.L., Goldwasser, L., Heong, K.L., et al. (1993). Improving food webs. Ecology, 74, 252–258. Courtney, S.P., Hill, C.J. & Westerman, A. (1981). Pollen carried for long periods by butterflies. Oikos, 38, 260–263. Dicks, L.V., Corbet, S.A. & Pywell, R.F. (2002). Compartimentalization in plant–insect flower visitor webs. J. Anim. Ecol., 71, 32– 43. Fenster, C.B., Armbruster, W.S., Wilson, P., Dudash, M.R. & Thomson, J.D. (2004). Pollination syndromes and floral specialization. Annu. Rev. Ecol. Evol. Syst., 35, 375–403. Forup, M.L., Henson, K.S.E., Craze, P.G. & Memmott, J. (2008). The restoration of ecological interactions: plant–pollinator networks on ancient and restored heathlands. J. Appl. Ecol., 45, 742–752. Goldwasser, L. & Roughgarden, J. (1997). Sampling effects and the estimation of food-web properties. Ecology, 78, 41–54. Go´mez, J.M. & Zamora, R. (2006). Ecological factors that promote the evolution of generalization in pollination systems. In: Plant– Pollinator Interactions, from Specialization to Generalization (eds Waser, N. M. & Ollerton, J.). University of Chicago Press, Chicago, IL, pp. 145–165. Go´mez, J.M., Bosch, J., Perfectti, F., Ferna´ndez, J. & Abdelaziz, M. (2007). Pollinator diversity affects plant reproduction and recruitment: the tradeoffs of generalization. Oecologia, 153, 597– 605. Gotelli, N.J. & Colwell, R.K. (2001). Quantifying biodiversity: procedures and pitfalls in the measurement and comparison of species richness. Ecol. Lett., 4, 379–391. Goulson, D. & Wright, N.P. (1998). Flower constancy in the hoverflies Episyrphus balteatus (Degeer) and Syrphus ribesii (L.) (Syprphidae). Behav. Ecol., 9, 213–219.

 2009 Blackwell Publishing Ltd/CNRS

Letter

Goulson, D., Stout, J.C. & Hawson, S.A. (1997). Can flower constancy in nectaring butterflies be explained by DarwinÕs interference hypothesis? Oecologia, 112, 225–231. Guimara˜es, P.R. & Guimara˜es, P. (2006). Improving the analyses of nestedness for large sets of matrices. Environ. Modell. Softw., 21, 1512–1513. Guimera`, R. & Amaral, L.A. (2005). Functional cartography of complex metabolic networks. Nature, 433, 895–900. Heinrich, B. (1976). The foraging specializations of individual bumblebees. Ecol. Monogr., 46, 105–128. Heinrich, B. (1979). ÔMajoringÕ and ÔminoringÕ by foraging bumblebees, Bombus vagans: an experimental analysis. Ecology, 60, 245– 255. Herrera, J. (1988). Pollination relationships in southern Spanish Mediterranean scrublands. J. Ecol., 76, 274–287. Herrera, C.M. (1996). Floral traits and plant adaptation to insect pollinators: a devilÕs advocate approach. In: Floral Biology, Studies on Floral Evolution in Animal-pollinated Plants (eds Lloyd, D.G. & Barrett, S.C.H.). Chapman & Hall, New York, pp. 65–87. Jordano, P., Bascompte, J. & Olesen, J.M. (2003). Invariant properties in coevolutionary networks of plant-animal interactions. Ecol. Lett., 6, 69–81. Jordano, P., Bascompte, J. & Olesen, J.M. (2006). The ecological consequences of complex topology and nested structure in pollination webs. In: Plant–Pollinator Interactions, from Specialization to Generalization (eds Waser, N.M. & Ollerton, J.). University of Chicago Press, Chicago, IL, pp. 173–199. Kanstrup, J. & Olesen, J.M. (2000). Plant-flower visitor interactions in a neotropical rain forest canopy: community structure and generalization level. Det Norske Videnskaps-Akademi, Ny Serie, 39, 33–41. Kunin, W.E. (1993). Sex and the single mustard: population density and pollinator behaviour effects on seed set. Ecology, 74, 2145– 2160. Martı´nez, N.D., Hawkins, B.A., Dawah, H.A. & Feifarek, B.P. (1999). Effects of sampling effort on characterization of foodweb structure. Ecology, 80, 1044–1055. Medan, D., Basilio, A.M., Devoto, M., Bartoloni, N.J., Torretta, J.P. & Petanidou, T. (2006). Measuring generalization and connectance in temperate, year-long active systems. In: Plant–Pollinator Interactions, from Specialization to Generalization (eds Waser, N. M. & Ollerton, J.). University of Chicago Press, Chicago, IL, pp. 245– 259. Memmott, J. (1999). The structure of a plant-pollinator food web. Ecol. Lett., 2, 276–280. Minckley, R.L. & Roulston, T.H. (2006). Incidental mutualisms and pollen specialization among bees. In: Plant-Pollinator Interactions, from Specialization to Generalization (eds Waser, N. M. & Ollerton, J.). University of Chicago Press, Chicago, pp. 69–98. de los Mozos Pascual, M. & Medina Domingo, L. (1991). Constancia floral en Heliotaurus ruficollis Fabricius, 1781 (Coleoptera : Alleculidae). Elytron, 5, 9–12. Nielsen, A. & Bascompte, J. (2006). Ecological networks, nestedness and sampling effort. J. Ecol., 95, 1134–1141. de Nooy, W., Mrvar, A. & Batagelj, V. (2005). Exploratory Social Network Analysis with Pajek. Cambridge University Press, New York. Olesen, J.M. & Jordano, P. (2002). Geographic patterns in plant– pollinator mutualistic networks. Ecology, 83, 2416–2424.

Letter

Olesen, J.M., Bascompte, J., Dupont, Y.L. & Jordano, P. (2007). The modularity of pollination networks. Proc. Natl Acad. Sci. USA, 104, 19891–19896. Petanidou, T. & Potts, S.G. (2006). Mutual use of resources in Mediterranean plant–pollinator communities: How specialized are pollination webs? In: Plant–Pollinator Interactions, from Specialization to Generalization (eds Waser, N. M. & Ollerton, J.). University of Chicago Press, Chicago, IL, pp. 221–244. Petanidou, T., Kallimanis, A.S., Tzanopoulos, J., Sgardelis, S.P. & Pantis, J.D. (2008). Long-term observation of a pollination network: fluctuation in species and interactions, relative invariance of network structure and implications for estimates of specialization. Ecol. Lett. 11, 564–575. Power, M.E. et al. (1996). Challenges in the quest for keystones. Bioscience, 46, 609–620. Proctor, M., Yeo, P. & Lack, A. (1996). The Natural History of Pollination. Timber Press, Portland, OR. Rodrı´guez-Girone´s, M.A. & Santamarı´a, L. (2006). A new algorithm to calculate the nestedness temperature of presenceabsence matrices. J. Biogeogr., 33, 924–935. Santamarı´a, L. & Rodrı´guez-Girone´s, M.A. (2007). Linkage rules for plant-pollinator networks: trait complementarity or exploitation barriers? PLoS Biol. 5: e31. doi:10.1371/journal. pbio.0050031. Stephen, W.P., Bohart, G.E. & Torchio, P.F. (1969). The Biology and External Morphology of Bees. Agricultural Experiment Station, Oregon State University, Corvallis, OR. Toft, C.A. (1984). Resource shifts in bee flies (Bombyliidae): interactions among species determine choice of resources. Oikos, 43, 104–112. Va´zquez, D.P. & Aizen, M.A. (2003). Null model analyses of specialization in plant–pollinator interactions. Ecology, 84, 2493– 2501.

Pollination networks 419

Va´zquez, D.P. & Aizen, M.A. (2004). Asymmetric specialization: a pervasive feature of plant–pollinator interactions. Ecology, 85, 1251–1257. Va´zquez, D.P. & Aizen, M.A. (2006). Community-wide patterns of specialization in plant–pollinator interactions revealed by null models. In: Plant–Pollinator Interactions, from Specialization to Generalization (eds Waser, N. M. & Ollerton, J.). University of Chicago Press, Chicago, IL, pp. 200–219. Va´zquez, D.P., Melia´n, C.J., Williams, N.M., Blu¨thgen, N, Krasnov, B.R. & Poulin, R. (2007). Species abundance and asymmetric interaction strength in ecological networks. Oikos, 116, 1120– 1127. Waser, N.M., Chittka, L., Price, M.V., Williams, N.M. & Ollerton, J. (1996). Generalization in pollination systems, and why it matters. Ecology, 77, 1043–1060. Wasserman, S. & Faust, K. (1994). Social Network Analysis: Methods and Applications. Cambridge University Press, New York. Westrich, P. (1989). Die Wildbienen Baden-Wu¨rttembergs. Ulmer, Stuttgart. Williams, N.M. & Tepedino, V.J. (2003). Consistent mixing of near and distant resources in foraging bouts by the solitary mason bee Osmia lignaria. Behav. Ecol., 14, 141–149. Williams, N.M., Minckley, R.L. & Silveira, F.A. (2001). Variation in native bee faunas and its implications for detecting community change. Conserv. Ecol., 5, 57–89.

Editor, Jordi Bascompte Manuscript received 13 November 2008 First decision made 11 December 2008 Manuscript accepted 4 February 2009

 2009 Blackwell Publishing Ltd/CNRS