Responses to olfactory signals reflect network structure of flower-visitor interactions

Authors


Correspondence author. E-mail: bluethgen@biozentrum.uni-wuerzburg.de

Summary

1. Network analyses provide insights into the diversity and complexity of ecological interactions and have motivated conclusions about community stability and co-evolution. However, biological traits and mechanisms such as chemical signals regulating the interactions between individual species – the microstructure of a network – are poorly understood.

2. We linked the responses of receivers (flower visitors) towards signals (flower scent) to the structure of a highly diverse natural flower-insect network. For each interaction, we define link temperature – a newly developed metric – as the deviation of the observed interaction strength from neutrality, assuming that animals randomly interact with flowers.

3. Link temperature was positively correlated to the specific visitors’ responses to floral scents, experimentally examined in a mobile olfactometer. Thus, communication between plants and consumers via phytochemical signals reflects a significant part of the microstructure in a complex network. Negative as well as positive responses towards floral scents contributed to these results, where individual experience was important apart from innate behaviour.

4. Our results indicate that: (1) biological mechanisms have a profound impact on the microstructure of complex networks that underlies the outcome of aggregate statistics, and (2) floral scents act as a filter, promoting the visitation of some flower visitors, but also inhibiting the visitation of others.

Introduction

Network patterns such as nestedness, connectance and degree distribution have been described for numerous ecological networks (Bascompte & Jordano 2007). They serve as aggregate statistics for the entire web, i.e. reducing the diversity of species and interactions within a community to a single value or formula. Variation across networks in these metrics may result from the species’ abundances, body sizes, phenology and other factors (Vazquez, Chacoff & Cagnolo 2009b; Vazquez et al. 2009a) or reflects variation in the number of observations per species (Blüthgen et al. 2008) The mechanisms influencing each individual interaction between the species within a network, however, remain largely unexplained (Vazquez, Chacoff & Cagnolo 2009b). For a more thorough understanding of network topology, mechanisms affecting the link-level (interactions between species pairs) need to be considered, focusing on the question how biological traits can influence the interaction strength.

In a qualitative flower-visitor web, Stang, Klinkhamer & Meijden (2006) demonstrated that morphological mismatches represent a feasible mechanistic explanation of the absence of certain interactions related to biological traits. These size-related traits form a clear-cut threshold which divides the potential nectar consumers in those that are able and unable to reach the nectar with their mouthparts (Stang, Klinkhamer & Meijden 2007). However, visitation by pollen consumers and the remaining variation within the ‘allowed’ nectar-mediated interactions in an observed quantitative flower – visitor network must be due to other factors than morphology. Such factors may include primary metabolites, e.g. sugars and amino acids, in nectar (Gardener & Gillman 2002; Petanidou et al. 2006) and pollen (Roulston & Cane 2000), floral display size and form (Glaettli & Barrett 2008), and colour (Haslett 1989, Chittka & Menzel 1992; Whitney et al. 2009). Secondary metabolites in nectar or pollen may increase or decrease certain interactions (Adler 2000; Dobson & Bergström 2000; Raguso 2004; Kessler & Baldwin 2006), and scents are known to play an important role in attraction and repellence either innately or following associative learning (Knudsen et al. 2006; Junker & Blüthgen 2008, 2010, Raguso 2008b; Wright & Schiestl 2009). In this study, we focused on floral scents and their impact on the animals’ behaviour as an explanation for the observed patterns, integrating biological signals and responses of receivers into network analysis to understand the interactions between trophic levels in a diverse community. Additionally, we used data from the literature to explore the role of floral colour and morphology.

Plant-animal interaction networks comprise a continuum from highly generalized to highly specialized assemblages. Flower-visitor networks are on average more specialized than plant-frugivore or nectar-mediated plant-ant relationships (Blüthgen, Menzel & Blüthgen 2006) showing a high degree of species complementarity (flower partitioning) between visitor species. Flower fidelity shown by species or individuals of pollinators either by a high specialization or by short-term specificity is crucial for a successful pollen transfer and may thus be promoted by natural selection (Blüthgen et al. 2007; Wright & Schiestl 2009). Different proximate mechanisms (floral filters) may enhance the visitation of some species and/or prevent the visitation of others and thus lead to high flower partitioning and specialization. For instance, it was shown that secondary metabolites in the nectar of Aloe vryheidensis prevent consumption by non-pollinating honeybees and sunbirds, but do not affect the visitation of pollinating birds (Johnson, Hargreaves & Brown 2006). Moreover, Raguso (2008b) hypothesized that olfactory cues may act as floral filter, but evidence is scant so far. It has been shown that floral scents serve as defence next to their well-known function as innate attractants or recognition cues for pollinators (e.g. Junker & Blüthgen 2008, 2010; Willmer et al. 2009). The distribution of nectar-thieving ants on flowers could be explained by volatiles. Scents from ant-visited flowers did not affect the ants’ behaviour in an olfactometer, but scent from flowers that were not visited by ants were repellent (Junker & Blüthgen 2008). The defensive and attractive properties of floral volatiles, along with the flowers’ necessity to filter out the least beneficial visitors suggest that floral scents are important traits structuring interaction networks. In order to understand the mechanisms that influence the network structure, we used a new metric (link temperature) to quantify the interaction strength of each link – representing the network microstructure – and explored whether it is reflected by the animals’ responses towards olfactory cues.

Material and methods

Study site and organisms

We studied two temporally separated quantitative flower-visitor networks near the University campus of Würzburg, Germany (Fig. 1, Appendix S1 in Supporting Information). The first survey (early June 2008) covered 41 h, the second one (early July 2008) covered 26 h where several people helped to collect flower-visitors within an area of 0·1 ha on a flower-rich fallow land. Two separate networks were established (one for each month). During random walks, all individual insects that visited flowers were recorded as well as the flower species on which they were found. Several individuals per known species and all unknown specimen were collected, sorted and identified to species level where possible with the help of experts (see Acknowledgements).

Figure 1.

 Interaction networks used in this study. Numbers denote animal and plant species (names are shown in Appendix S1 in Supporting Information). Widths of nodes correspond to the total number of observed interactions of each species, width of links to the number of interactions between species pairs. Missing links are only shown as dashed lines if they were tested in the olfactometer trials.

Network metric

As we defined a network as a spatiotemporal entity, summarizing all interactions between insects and flowers recorded in a small area within a short time period, we can assume that all species can potentially interact. This allows us to focus on traits that shape the interactions apart from space and time. We defined a new metric (link temperature Tij) to quantify the deviation of the observed interaction frequency between each species pair from an expected value predicted by a neutral model. This metric extends the network- and species-level indices recently established (H2′ and di′, see Blüthgen, Menzel & Blüthgen 2006). All three metrics focus on the residual deviation of the interaction frequency from the neutral expectation (Blüthgen et al. 2008). If all species interacted randomly with the other trophic level in absence of any constraints, preferences or aversions (neutral model), the expected interaction strength would be eij = inline image, where Ai and Aj are the total number of records of visitor species i and plant species j, respectively and m is the grand total of recorded interactions for all species (Blüthgen et al. 2008). For each possible link, the observed deviation (henceforth termed link temperature Tij) from the neutral model was defined as Tij = inline image, with aij as the observed number of visits of animal species i on plant j. Tij ranges between −1 and 1; negative values correspond to cold links (i.e. fewer observations than expected by the neutral model) and positive ones to hot links. Tij is thus defined from the animal’s perspective using Ai as denominator. Whether link temperatures were significantly hot or cold, was defined using Monte-Carlo statistics by comparing each observed aij with the distribution of interaction frequencies between i and j (α′ij) obtained from the Patefield algorithm (Patefield 1981, Blüthgen, Menzel & Blüthgen 2006) generating 1000 random matrices with a fixed distribution of marginal totals Ai and Aj. The mean α′ij across all randomized networks approximates eij.

Olfactometer trials

Links between relatively common species were haphazardly selected for olfactometer experiments, including significantly hot and cold links. We used a mobile olfactometer to test whether the animals’ responses to floral scents reflect the variation in link temperatures. The system allowed us to conduct bioassays in the field with scents from naturally growing, unpicked flowers and flower-foraging insects that did not live in captivity. Most of the insects were caught during their foraging activities on flowers, most of them (94%) on flower species that were visited more often by this insect species than expected (hot links). In two different types of arenas, these visitors were confronted with floral scents and neutral air (filtered and humidified in charcoal and distilled water). For crawling insects like ants and beetles, a four field arena was used (similar to the one described in detail in Junker & Blüthgen 2008). For flying insects such as bees, bumblebees and hoverflies, a Y-shaped arena was used. Glass plates covered the arenas. The whole apparatus was fitted into an aluminium box making it mobile for an operation in the field. Technical details, dimensions of the arenas, protocol used for the tests and a list of conducted biotests are provided in Appendices S2 and S3 in Supporting Information. Next to an innate foraging behaviour elicited by floral scents, these signals are also often associated with rewards and are therefore learnt by the animals (Cunningham et al. 2004, 2006; Bruce, Wadhams & Woodcock 2005). In order to account for the learning ability, in a third of the trials we compared the responses from (1) individuals that were caught from flowers of the focal species used as scent source in the olfactometer to (2) conspecific individuals that were caught while visiting a flower from a different plant species. The animals’ responses were expressed by the index Rij = inline image, with Nobs = choices for, or time spent in scented fields; Nexp = expected value for each field assuming random choices, i.e. 50% of tested animals or total time; and Ntotal = total number of animals tested or total time span. Like Tij, Rij varies between −1 (avoidance) and 1 (attraction).

Colour and morphology

In order to estimate the influence of floral colour and morphology on link temperatures, we obtained data on basic colours and morphological flower types after Kugler (Kugler 1970) for each plant species in our webs from the internet data base ‘BiolFlor’ (http://www.ufz.de/biolflor/index.jsp; Klotz, Kühn & Durka 2002). We performed Kruskal–Wallis rank sum tests with pooled link temperatures of ants, bees, bumblebees, beetles, hoverflies and butterflies as response variables and colours (Appendix S4) or flower types (Appendix S5) as explanatory variables. This approach enabled us to search for preferences of insect groups regarding certain colours or morphologies. However, note that the resolution of these data is not specific to each link unlike our olfactometer approach. Furthermore, many features are neglected in these categories: the colours neither include information on UV-reflectance nor on subtle differences in the reflectance spectrum. The morphological categories bear limited information on potential barriers such as nectar tube length or other features that prevent certain animals from consuming nectar (see Stang et al. 2006, 2007).

Results

Network metric

In the first network, 303 links between 35 plant and 164 insect species were recorded (connectance = 0·053, 2251 individual interactions) and 170 links between 23 plant and 64 insect species in the second (= 0·115, 1080 individual interactions). Complementary specialization (flower partitioning) of visitor species was pronounced (H2′ = 0·47 and 0·52, respectively), similar to other flower-visitor networks recorded so far (Blüthgen et al. 2007). Correspondingly, many links deviated strongly from neutrality: 47·5% and 47·7% of the realized links were significantly hot, and 20·8% and 31·8 % were significantly cold, respectively (Appendix S1 in Supporting Information, Fig. 1). We observed a strong species turnover between both dates where we recorded the interactions (Appendix S1 in Supporting Information). However, link temperature of those links present in both networks were strongly correlated between the two dates (Pearson’s R2 = 0·40, df = 78, < 0·001).

Olfactometer trials

We performed a total of 59 olfactometer trials including floral scents of 18 plant species and 10 animal species (total = 1557). On average, hot links received positive and cold links negative response indices, whereas neutral links had intermediate values (anova: F2,56 = 9·3, < 0·001, Fig. 2). Animals that were caught from different flowers than the focal ones responded similarly (F2,49 = 3·3, = 0·04, Fig. 2). Individuals caught from the focal species used for the olfactometer trials received mostly (17 of 20 tested) higher Rij values than conspecifics caught from other flowers (paired t-test: t19 = 3·8, < 0·001; Fig. 3). However, because cold links are per se rarely or never realized in a community, we were able to test animals only from three cold links. Four cases of significant attraction corresponded to significantly hot links, nine cases of significant repellence to significantly cold links (Appendix S3 in Supporting Information). For instance, Bombus pascuorum Scop. interacted significantly more often than expected with Ballota nigra L. (Tij = 0·37, < 0·001) but never with Crepis vesicaria L. (Tij = −0·25, < 0·001). This hot link corresponded to significant attraction (Rij = 0·21, = 0·035), the cold link to avoidance (Rij = −0·56, = 0·018). The same flower species often triggers different responses to different visitors. Overall, we found a highly significant correlation between the animals’ response Rij and the link temperature Tij (Pearson’s  = 0·17, df = 57, < 0·001; Fig. 3). For this correlation, we used data from all animal individuals regardless of their experience in order to represent the responses of the whole flower-visitor population. The positive relationship between animals’ response Rij and the link temperature Tij was confirmed within each animal taxon used in the olfactometer trials, namely ants, bees, bumblebees, beetles and hoverflies, albeit significant for beetles only (Appendix S6).

Figure 2.

 Animals’ responses Rij towards floral scents originating from significant hot, neutral and cold link temperatures Tij. Dark grey bars denote to all tests, light grey bars to tests with animals caught from other flowers than from focal ones, white bars denote tests with animals caught from focal flowers. Mean Rij values and 95% confidence intervals (CI) shown, with sample sizes (n) next to each bar. CI for the final bar cover the entire possible range of Rij (−1–1).

Figure 3.

 Correlation between response index Rij and link temperature Tij in each network. For the regression between Rij and Tij, we used mean Rij values of all individuals of a given species irrespective of their previous experience. Arrows denote the effect of immediate experience on the animals’ response to floral scents, pointing from the response of animals not caught from target flowers to those caught from the same flower species used in the trial.

Colour and morphology

Overall, floral colour and morphology had a significant influence on link temperatures (Kruskal–Wallis rank sum test: Χ2 ≥ 28·95, df = 7 and 5 for colour and morphology, respectively, < 0·001). The link temperatures Tij of bees, beetles and butterflies differed across colours (Appendix S4), the link temperatures Tij of all groups of insect groups across flower types (Appendix S5). However, differences were largely confined to a few colours or flower types, while most were associated with similar link temperatures Tij.

Discussion

Network analyses complement the pair-wise treatment and interpretation of biological interactions and draw attention to their broader context in multi-species communities (Proulx, Promislow & Phillips 2005; Ings et al. 2009). Network analyses were often applied to flower-visitor interactions and revealed valuable insights to the structure of these communities (e.g. Olesen et al. 2008, Ings et al. 2009). The common term ‘pollination network’ or ‘mutualistic network’, however, simplifies the variation in the quality of interactions: pollinators are not equally beneficial to each plant species (Stanton 2003; Johnson, Hargreaves & Brown 2006; Reynolds & Fenster 2008), and a considerable proportion of all visitors are not mutualistic, e.g. exploiting floral resources without pollen transfer (e.g. Brody et al. 2008). The mutualistic service provided by a flower visitor is not only dependent on the flower visitors’ identity but also on the focal link (Junker & Blüthgen 2010), an information that is not available for most interactions.

The presence of both mutualists and antagonists forces flowers to display attractive as well as defensive traits (Brown 2002; Irwin, Adler & Brody 2004). Therefore, floral filters – individual traits that invite mutualists but keep antagonists at bay – are needed that may include certain characteristics of nectar (Johnson, Hargreaves & Brown 2006, Hansen et al. 2007), morphological structures (Galen 1999; Galen & Cuba 2001; Stang, Klinkhamer & Meijden 2006, 2007; More, Sersic & Cocucci 2007; Stang et al. 2009), and scent (Dobson & Bergström 2000; Junker & Blüthgen 2008, 2010). Therefore, floral filters that affect several species can not be detected in pair-wise interactions where suites of traits of both partners were viewed as an outcome of co-evolution but in studies where whole communities are considered which is facilitated by network analysis (Stanton 2003; Raguso 2008a).

Our results demonstrate that the animals’ responses to floral scents reflect a considerable proportion of the network structure after neutral effects were accounted for. We found both positive and negative responses towards floral scents, suggesting that next to attraction, repellence may shape the visitation and contribute to the reproductive success of plants (Junker & Blüthgen 2010). In our study, repellent functions of floral scents did not only involve antagonists (Junker & Blüthgen 2008; Kessler, Gase & Baldwin 2008), but also affected the visitation by typical pollinators. For instance, the cold links between Bombus terrestris L. and Apis mellifera L. and Achillea millefolium L. corresponded to negative significant responses Rij (see Appendix S3 in Supporting Information). Furthermore, our results suggest that the insects’ ability to associate rewards with specific floral scents also contribute to the regulation of the microstructure of the network. The trials accounting for the insects’ immediate experience confirm that appetitive learning plays an important role for foraging insects (Raguso 2008b) implying that floral scents often function as recognition traits in addition to their function as innate attractants (Cunningham et al. 2004, 2006). Our experiments are unable to unequivocally differentiate between innate and learned responses. Innate responses (attraction and repellence) would imply that scents are a regulating force of the microstructure of networks, while learned responses (positively or negatively) would represent a reinforcement of decisions based on other factors than scent, e.g. proximately other floral traits like colour or morphology or ultimately the resource quality. However, in both cases volatile cues would influence the network structure. Floral colour and particularly morphology (see Stang et al. 2006, 2007) are also involved, because link temperatures are affected by both factors. For example, nectar-mediated visits by ants and hoverflies where infrequent in lip flowers (i.e. negative link temperatures), where nectar is not accessible to these animals with short mouthparts. However, to compare the importance of different traits on network microstructure, visual, structural and nutritional features should be evaluated for each link in a similar manner as in the present study.

The outcome of the olfactometer trials may be strongly influenced by the selection of insects, i.e. whether they are caught from focal plant species or from non-focal ones. However, the virtual absence of ‘experienced’ animals from cold links and the overrepresentation of ‘experienced’ animals from hot links are an important part of the realized distribution of visitors on flowers in natural communities. Our finding that animals with an immediately prior experience to a floral odour had significantly more positive responses than those without, emphasizes the importance of associative learning and short-term specificity. Long-term experience and flower fidelity may have an even stronger influence on the outcome of olfactometer trials; thus, effects of experience and conditional learning may be underestimated in our approach.

Both individual substances as well as the proportional composition of substances within floral bouquets play a role in flower recognition (Bruce, Wadhams & Woodcock 2005). In the field, Waelti et al. (2008) clearly demonstrated that floral scents of two closely related Silene species strongly contribute to maintain the reproductive isolation of these species due to their role in promoting floral constancy of pollinators. Pollinators, in return, may benefit from flower fidelity by minimizing their handling time (Stanton 2003).

Immobile flowers may use scents in concert with other signals to influence their visitor spectrum, explaining the non-random associations observed in these communities. Although we ignored the morphology of flowers and insects’ mouthparts (see Stang, Klinkhamer & Meijden 2006, 2007; Stang et al. 2009) and additional floral signals such as colour, shape, size and quality of the rewards, we found a correlation between the animals behaviour elicited by scent (expressed as Rij) and the interaction strength (expressed as Tij). The congruence of positive or negative link temperatures and responses, respectively, additionally accentuates the importance of scents in these communities. Thus, floral scents are important biological signals that may regulate network structure due to attraction (Knudsen et al. 2006; Junker & Blüthgen 2010), associative learning (Cunningham et al. 2006) and repellence (Junker & Blüthgen 2008, 2010). The inter- and intraspecific partitioning of visitors on flowers (Palmer, Stanton & Young 2003) shaped by biological filters may facilitate the coexistence of the large number of flower visitors that compete for the floral resources. These mechanisms determine which partners do and do not interact with each other and thereby also a large proportion of network structure.

Acknowledgements

We thank Sophie Batsching for performing some olfactomter trials and Volker Mauss, Christiane Weiner, Michael Werner and Mirko Wölfing for identifying insects and Stefan Dötterl and Afroditi Kantsa for valuable comments. Norbert Schneider provided technical support. Florian Menzel helped with statistical analyses. Anna Hellwig, Mathias Hofmeister, Linda Jung and Daniel Schaub helped to collect field data. The project was supported by the Deutsche Forschungsgemeinschaft (DFG BL960/1-1).

Ancillary