Plant generalization on pollinators: species property or local phenomenon?

2000). Investigations on the origin, persistence, frequency of occurrence, and evolutionary consequences of generalist pol- lination systems are of interest from both ecological (e.g., in relation to the topology and structure of community-wide plant–pollinator interaction networks; Memmot, 1999; Dicks et al., 2002; Olesen and Jordano, 2002; Bascompte et al.,

2003) and evolutionary viewpoints (e.g., in relation to the role played by pollinators in the evolution of floral traits and di- versification of flowering plants; Herrera, 1996; Waser et al.,

1996; Armbruster et al., 2000; Waser, 2001; Go´ mez, 2002).

Despite increased interest in generalist pollination systems,

two critical aspects have never been concurrently investigated

for any plant species. (1) Is pollinator generalization a species-

level trait or, in other words, do all populations of a generalist

species have roughly similar levels of generalization on pol-

The author thanks Conchita Alonso, Marie C. Anstett, Alfredo Benavente, Dori Ram´ırez, Roc´ıo Requerey, Leo´n Sa´nchez and Antonio T´ıscar for field work assistance; Robert Colwell for making the EstimateS program available; Conchita Alonso, Jordi Bascompte, Jose´ L. Garrido, Antonio Manzaneda, Mo´ nica Medrano, and two reviewers for useful comments or discussion on the manuscript; and the Consejer´ıa de Medio Ambiente, Junta de Andaluc´ıa, for authorizing my research in Cazorla. Fox and Morrow’s (1981) study was inspirational well beyond its obvious help in finding a title for this paper. Research partly funded by grants BOS2000-1122-C03-01 and BOS2003-

03979-C02-01 from Ministerio de Ciencia y Tecnolog´ıa.

² E-mail: herrera@cica.es.
13

linators? (2) Within a population of a generalist plant, are all individuals similarly pollinator-generalized? These questions are related to the broader issue of whether ecological special- ization is an inherent property of a species, an attribute of local populations, or a trait of individual organisms (Fox and Mor- row, 1981). In the case of herbivorous insects, for example, polyphagous species may be made of populations having rel- atively narrow host ranges (Fox and Morrow, 1981; Thomp- son, 1994), and a relatively polyphagous population may be made up of either generalist or specialist individuals (Bernays and Minkenberg, 1997; Bernays and Singer, 2002). Likewise, plant species with generalist pollination systems may be made of populations having relatively specialized pollinator assem- blages, and generalist populations may be composed of vari- able mixtures of generalist and specialist individuals. These different scenarios entail contrasting evolutionary consequenc- es (e.g., Thompson, 1994, 1999; Gomulkiewicz et al., 2000), yet empirical information on these aspects of the ecology of generalist insect-pollinated plants is virtually nonexistent.

In this paper, I will address these questions for Lavandula latifolia Medicus (Lamiaceae), a shrub pollinated by a very diverse insect assemblage (Herrera, 1988). Addressing these questions raises the critical issue of how to properly compare pollinator diversity among populations and among individuals within populations, an aspect that has received surprisingly little attention to date. A secondary objective of this paper is thus to propose a method based on rarefaction procedures (Go- telli and Colwell, 2001) to obtain truly comparable, standard- ized measurements of pollinator diversity. Recent analyses of the degree of specialization of plant–pollinator interactions have frequently used the number of visitor taxa as a measure of generalization (e.g., Waser et al., 1996; Olesen and Jordano,

2002; Kay and Schemske, 2004), yet these studies have gen-

erally proceeded without considering the possible pitfalls in- volved in comparing pollinator diversity estimates based on




raw species counts (but see Ollerton and Cranmer, 2002). In the next section, I briefly introduce some of these pitfalls and justify the approach adopted in this paper.
How to measure plant pollinator generalization?—Plant generalization on pollinators may be variously defined (Her- rera, 1996; Waser et al., 1996; Armbruster et al., 2000; John- son and Steiner, 2000; Olesen, 2000; Go´ mez, 2002; Olesen and Jordano, 2002; Nakano and Washitani, 2003; Kay and Schemske, 2004). Here it is defined in relation to the number of pollinator taxa involved in the interaction, as distinct from morphological or evolutionary generalization. Thus defined, quantifying the degree of plant generalization on pollinators boils down to evaluating pollinator diversity (Ollerton and Cranmer, 2002). This means, on one side, that the task is sub- ject to the same deceptive simplicity, and prone to the same sampling biases and pitfalls, long known to complicate esti- mates of species diversity of plant and animal communities (e.g., Magurran, 1988; Gotelli and Colwell, 2001). Compari- sons based on crude lists of pollinator taxa that do not correct for the influence of differential sampling effort or pollinator abundance may be fatally flawed (Ollerton and Cranmer,

2002), in the same way and for the same reasons as are com- parisons of plant community species richness that neglect the influence of variable number of quadrats or differential abun- dance of individuals (Magurran, 1988; Gotelli and Colwell,

2001). On the positive side, however, approaching the study of pollinator generalization as a particular case of species di- versity assessment has some advantages. Once the set of en- tities on which diversity measurements are to be taken is ex- plicitly defined, then methods devised for measuring the di- versity of ecological communities may be readily imported for use with pollination data.

Generalization being a concept pertaining to the plants, it is the plant’s ‘‘perception’’ of pollinator diversity, not the ecol- ogist’s, that should prevail when measuring pollinator diver- sity. I suggest that a plant’s perception of pollinator diversity should be assessed on the ‘‘population’’ of entities made up of flower visitation events, rather than on the set of biological individuals (pollinators) interacting with the plant. In other words, I propose that the relevant magnitude to quantify pol- linator generalization is the diversity of flower visits, rather than that of floral visitors. The visitor-centered approach, which is closer to the ecologist’s perception of pollinator di- versity, has been adopted frequently (e.g., Herrera, 1988; Mahy et al., 1998; Memmott, 1999; Dicks et al., 2002; Me- le´ndez-Ramirez et al., 2002). Its results, however, will gener- ally provide a distorted picture of a plant’s perception of pol- linator diversity because pollinator taxa differ extraordinarily in one or more of the following parameters: (1) number of flowers visited per time unit, (2) per-visit probability of ef- fecting a pollen transfer event, and (3) average quality of pol- len transfers effected (e.g., Schemske and Horvitz, 1984; Her- rera, 1987, 1989; Thompson and Pellmyr, 1992; Go´ mez and Zamora, 1999; Thompson, 2001). Although it admittedly ne- glects aspects (2) and (3), a visit-centered approach to mea- suring pollinator diversity represents a significant improve- ment over the visitor-centered one in that it at least incorpo- rates information on (1), the number of flowers visited per unit time. Reasons for adopting a visit-centered approach to mea- sure pollinator diversity are essentially the same as those un- derlying the customary use of the proportion of flowers visited by different taxa to quantify the differential strength of inter-

action of one plant with its different pollinators (e.g., Go´ mez and Zamora, 1999; Potts et al., 2001; Kay and Schemske,

2003). Furthermore, the visit-centered approach is conceptu- ally linked to the visitation-rate component of pollinator im- portance, as implied in Stebbins’ (1970, 1974) ‘‘most-effec- tive-pollinator principle’’ and made explicit, among others, by Armbruster (1988), Herrera (1989), and Armbruster et al. (2000).

Provided that (1) information on flower visitation events by pollinators is collected using an adequate sampling protocol based on random independent samples and (2) flower visita- tion events are categorized as to the taxonomic identity of the pollinator involved, then quantitative measurements of polli- nator diversity useful for comparative purposes may be ob- tained by the same procedures used in the study of community diversity. Adoption of this approach will make clear that pol- linator diversity measurements are prone to suffer from the same pitfalls and sampling biases long known to affect com- munity diversity studies (Magurran, 1988; Gotelli and Colwell,

2001). Of all these, the potentially misleading influence of pollinator abundance (from the plant’s perspective, i.e., as re- flected in flower visitation probabilities) on estimates of pol- linator diversity must be singled out, because its influence is both more elusive and more difficult to counteract than that of differential sampling effort. Without adequately accounting for differential pollinator activity or abundance, even compar- isons of pollinator species richnesses based on similar sam- pling efforts (e.g., similar duration of watching intervals, ob- servation days, or number of pollinator censuses) may still largely reflect differences in pollinator abundance rather than pollinator diversity itself. For this reason, in this study I will use sample-based rarefaction curves scaled to number of ‘‘in- dividuals’’ (sensu Gotelli and Colwell, 2001) to estimate pol- linator species richness of Lavandula latifolia populations and individual shrubs. Rarefaction is a statistical method first pro- posed by Sanders (1968) to overcome the problems involved in comparisons of community samples based on different sam- ple sizes. It allows for estimation of the number of species (s) expected in a random sample of n individuals taken from a larger collection made up of N individuals and S species (Krebs, 1989). Species richness was preferred as a measure of pollinator diversity over, e.g., diversity indices, because it is the simplest way to describe diversity, has good discriminant ability (Magurran, 1988; Gotelli and Colwell, 2001), and has been used previously to measure pollinator diversity (e.g., Waser et al., 1996; Ollerton and Cranmer, 2002). It must be noted, however, that focusing on pollinator species richness alone neglects the possible significance of the evenness (or equitability) component of pollinator diversity, an aspect that would deserve consideration in future studies.

2000). Flowering takes place in summer (July–September). Each shrub may produce up to a few thousand flowers per flowering season, with dozens to a few hundreds of flowers simultaneously open on the same plant. Flowers are hermaphroditic, have pale-blue tubular corollas and are self-compatible, but

TABLE 1. Sampling effort and pollinator species richness for 15 La- vandula latifolia shrubs studied in Arroyo Aguaderillos, Sierra de Cazorla, southeastern Spain, in 1991.

Differences among individual shrubs in pollinator diversity were studied in

1991 at a single L. latifolia population growing around Arroyo Aguaderillos, at 1160 m elevation. Pollinator flower-visitation data were collected from 15 flowering shrubs between 20 July and 10 August. The two most distant plants were 30 m apart. Pollinators were censused on these plants from dawn to dusk throughout the study period according to a randomized sampling design. Each census lasted for 5 min, when I closely watched the activity of polli- nators at one of the marked shrubs. All flower visitors were identified to species, and information from previous studies (Herrera, 1987; C. M. Herrera, unpublished observations) was used to ascertain their status as true pollinators. The total number of flowers visited by each pollinator taxon was recorded in each census. Further details on methods can be found in Herrera (1995), in

Sampling effort

Population differences in pollinator diversity were studied in 1996 in 15 populations of L. latifolia distributed over a broad area of the Sierras de Cazorla and Segura region. The two most distant populations were 55 km apart. Populations included the Arroyo Aguaderillos site studied in 1991 and occurred in pine (Pinus nigra or P. pinaster) or oak (Quercus rotundifolia) woodlands. Pollinators were censused at all populations from 24 July to 14

August. At each site, 80–120 pollinator censuses were conducted on a single date on 20 different L. latifolia shrubs, using the same procedures as in 1991

Aguaderillos censuses, except that censuses were of 3-min durations. To in- vestigate annual variation in pollinator diversity and whether population dif- ferences remained consistent across years, five of the 15 populations were censused again in 1997 (29 July to 12 August). Pollinators were censused on the same individual plants with the same methods in both years, the only relevant difference being that in 1997 the censuses for each population were spread over several different dates.

S_RAR100 = rarefaction-estimated species richness, obtained from the flow- er-based rarefaction curve for each shrub (Fig. 1) as the y-value pre- dicted for x = 100 flowers visited. * Non-estimable, because total flow- ers visited <100.

shrubs of Arroyo Aguaderillos in 1991. The number of pol- linator species recorded per plant (S_OBS, Table 1) was much smaller and varied considerably (range 3–16 species per plant; Table 1). Flower visitation also varied widely among plants, ranging from 1.5 ± 4.6 flowers visited/census (mean ± SD) to 28.9 ± 31.6 flowers/census, thus a 20-fold variation (x² =

82.6, df = 14, P K 0.001; Kruskal-Wallis ANOVA). Mean flower visitation and S_OBS were positively, significantly cor- related across plants (r = 0.694, N = 15, P = 0.004), which suggests that plant differences in S_OBS may reflect differential pollinator visitation frequency rather than, or in addition to, true differences in pollinator species richness. Flower-based rarefaction curves computed separately for each plant (Fig. 1) clearly confirm this possibility. The pollinators of some plants with large S_OBS values (e.g., plants 6 and 15, S_OBS > 15 spe- cies) are in fact appreciably less diverse than those of others with lower S_OBS (e.g., plants 3, 12, 13, S_OBS < 13 species) (Fig. 1).

Flower-based rarefaction curves for individual shrubs (Fig.

1) reveal considerable variation among plants in pollinator

species richness. Confidence intervals around each curve have

been omitted from the graph to avoid cluttering, but the con-

fidence belts of the most species-poor plants (2, 5, 9 and 11)

are largely non-overlapping with those of the most species-

rich plants (3, 12, 13). Truly comparable estimates of polli-

nator species richness were obtained from the rarefaction

curves of individual plants as the projected y-value corre-

^{sponding to an abscissa of x = 100 flower visits (S}_RAR100<sup>; Table

1). S</sup>_RAR100 ^{could not be estimated for three plants that were}

quite infrequently visited by pollinators (total number of flow-

ers visited <100; Table 1). For the remaining 12 plants, S_RAR100 ranged from 2.8 to 11.1 species, which denotes that the ex- pected number of pollinator species implicated in the visitation of 100 flowers varied four-fold among shrubs that were <30

m apart. The correlation between S_RAR100 and S_OBS across plants is not statistically significant (r = 0.462, N = 12 plants, P =

Sampling effort

Pollinator species richness

mean flower visitation per census and the effect of this latter variable is thus statistically accounted for (partial r = 0.907, N = 12, P = 0.0001).