Search tips
Search criteria 


Logo of annbotAboutAuthor GuidelinesEditorial BoardAnnals of Botany
Ann Bot. 2009 September; 104(4): 689–701.
Published online 2009 June 9. doi:  10.1093/aob/mcp140
PMCID: PMC2729625

Implications of a long-term, pollinator-mediated selection on floral traits in a generalist herb


Background and Aims

The phenotypic selection of a diverse insect assemblage was studied on a generalist plant species (Paeonia broteroi) in ten flowering seasons, with tests for whether visitor preferences for plants with larger flowers eventually translated into significant differences among plants in visitation rates, seed production, seed mass, seed germination and seedling survival.


Selection gradients were used to assess if selection on flower size contributed to explain differences in visitation rates, seed production and seed mass. First, independent analyses were carried out for each season; then for the ten season as a whole. Seedling emergence and survival were assessed by generalized linear models.

Key Results

Directional selection was found on flower size through visitation rates and seed production, and stabilizing selection through seed mass. Thus, larger flowers were more visited, and produced more, but lighter seeds, than smaller flowers. The results suggest a conflicting selection on flower size through seed number and size. Floral integration found in the study populations was larger than that in populations of a distant region. Finally, seed size did not influence seedling emergence and survival; thus, any advantages of seed size may be constrained under natural conditions before plants become reproductive individuals.


Plants with larger flowers may be benefited by producing more lighter seeds than fewer heavier ones, as they may contribute disproportionately to the seed bank, and have better chances that any descendant could eventually recruit. However, it seems unlikely that differences in flower size and integration found among populations in different regions could have been originated by rapid evolutionary change. First, because of the conflicting selection described; second, because of the remarkably low seedling survival found under natural conditions. Consequently, the influence of pollinator selection alone does not seem to explain differences in flower size and integration.

Key words: Paeonia broteroi, long-term selection, conflicting selection, flower size, seed production, generalist pollination


Natural selection constitutes one of the major forces driving evolutionary change in living organisms, and several studies have concluded that it can be quite strong under natural conditions (e.g. Endler, 1986; Hoëkstra et al., 2001). Consequently, and given enough time, this process may result in evolutionary change (adaptation and speciation), although it may have very diverse constraints.

Among those constraints, one of the most recurrent ones is related to the fact that agents originating selection may largely vary at different spatial and temporal scales, and this variation contributes to the inconsistency in the selective pressures. For instance, some models developed to explain adaptation in plant–pollinator interactions predict that becoming a generalist or a specialist plant species may largely depend upon the spatio-temporal variation in the abundance and efficiency of the different pollen vectors (e.g. Waser et al., 1996). These models suggest that, if selection is consistent across plant generations (i.e. the same vectors provide the same pollination services), becoming a specialist implies better fitness gains than becoming a generalist. Contrarily, if pollination services are provided across generations by a highly variable (in abundance and/or efficiency) vector assemblage, becoming a generalist increases fitness gains. Such spatio-temporal variation of the pollinator assemblage is commonly reported in most pollination studies carried out in natural plant populations (e.g. Herrera, 1989; Schemske and Horvitz, 1989).

The influence of the spatial scale on evolutionary processes has been made clear in the last decade (Thompson, 1994, 2005), and particularly interesting from the view point of plant–animal interactions is the way in which the spatial variation in the scenarios where such interactions occur, and hence in their response, may originate different evolutionary outcomes adapted to local conditions across a geographical range (e.g. Gómez and Zamora, 2000; Fedriani et al., 2004; Rey et al., 2006; Alcántara et al., 2007). However, the number of studies addressing the outcome of selective processes under field conditions over large temporal scales is still limited (but see Herrera, 1989, 1998, 2000a; Waser and Price, 1989; Valone and Kaspari, 2005), most likely due to practical reasons. Yet, it has been demonstrated that temporal variation in the composition and efficiency of the agents originating selection may give biased estimates of selection in short-term studies (e.g. Schemske and Horvitz, 1989). Consequently, long-term studies may offer a more realistic picture of the biological processes taking place in natural populations, and hence a more successful and biologically meaningful interpretation of them.

In recent decades, many studies on the interaction between plants and pollinators have focused on intra-specific analyses of the pollinator preferences for individual plants based upon floral traits (including variation in flower arrangement, form or colour; and with both observational and experimental approaches), to evaluate, in turn, their consequences for plant fitness. These studies have benefited from the availability of statistical techniques allowing the estimation of direct and indirect selection (e.g. selection gradients and selection differentials; Lande and Arnold, 1983). Following this phenotypic selection approach, numerous studies have shown pollinator-mediated selection on floral traits in a large number of species (e.g. Galen, 1989; Gómez, 2000, among others) and, in some cases, that the likelihood of evolutionary change may be driven by such selective processes (e.g. Galen, 1996). Other studies, however, have not found evidence of selection on any particular floral trait, or other factors were more important for plant fitness than pollinators and/or herbivores, becoming better predictors of potential evolutionary change (e.g. Herrera, 1993, 2001). However, irrespective of their findings, conclusions of most of these studies may be limited by the (generally) short time scale used to evaluate selective processes.

In this paper, the phenotypic selection exerted by a very diverse insect assemblage on a highly generalist plant species, Paeonia broteroi (see Sánchez-Lafuente et al., 1999), is studied for ten consecutive flowering seasons (1997–2006), testing whether pollinator preferences for particular phenotypes eventually turn into significant differences among plants in several estimates of fitness. In a previous work, Sánchez-Lafuente (2002) found that plants of this species in population located in two distant regions, differed in floral integration and in a number of floral traits, particularly in size, and suggested that such differences could have been originated by the interaction with the local pollinator fauna. Although selection on different traits does not necessarily imply an evolutionary change (given the number of genetic, ecological or environmental factors that may limit that change), a precondition for this possibility is the existence of a consistent selection of pollinators on floral and/or reproductive traits, which eventually could translate into fitness differences among individual plants.

Specifically, an attempt is made to answer the following questions: Do pollinators preferentially visit plants with larger flowers? Is pollinator diversity and flower size related to seed production and seed size? Is flower size under selection? How strong and consistent is selection over time? Is seed size related to germination and seedling survival?


Study species and sites

Paeonia broteroi Boiss. & Reut. (Paeoniaceae) is a perennial herb that in the Iberian Peninsula typically grows in the understorey of oak and pine woods. Plants consist of one-to-several leafy stems sprouting every season from tuberous roots. The number of flowers may vary between one and eight, but in the study area the modal number of flowers per plant is one. Morphologically, flowers are actinomorphic and large (approx. 6–12 cm in diameter when fully open), with five to eight dark pink petals and nectaries outside the perianth. The number of stamens is approx. 170 and the number of apocarpous carpels ranges between one and five. Functionally, flowers are hermaphroditic, and protogynous (stigmas may be receptive for approx. 1 d before anthers dehisce); then male and female phases overlap for 3–4 d. Stigmas remain receptive for approx. 4–5 d if unpollinated. Flowering extends from mid-April to early June, and seed maturation takes approx. 10–16 weeks. Detailed descriptions of the natural history and pollination biology of the species may be consulted elsewhere (e.g. Sánchez-Lafuente et al., 1999; Herrera, 2000b; Sánchez-Lafuente, 2002).

This study was conducted in Sierra de Jaén, a mountainous area in southern Spain, for ten consecutive flowering seasons (1997–2006). The two largest populations in this range (between 100 and 120 plants) were used: Cañada de las Hazadillas (Hazadillas) and Llanos de Navalopo (Navalopo), both located in a mixed pine and oak forest between 1000 and 1200 m a.s.l., and separated by approx. 3 km. A total of 81 single-flowered plants were tagged in 1997 (41 in Hazadillas and 40 in Navalopo), and the same plants were followed for the ten seasons. The selected plants had a scattered distribution over the study populations (minimum distance between the closest plants was approx. 6 m.). Their initial status was standardized by tagging only those producing one flower and showing no signs of herbivory on vegetative parts or flower buds.

Pollinator censuses and characterization of floral traits

Visitor censuses were conducted in all the study seasons, during the flowering period and on all flowering plants tagged. Censuses lasted for 3 min flower−1 and were conducted from 0900 to 1600 h from the day after anthesis until stigmas were no longer receptive. Each day a plant was randomly chosen for the first count; then the rest were sequentially monitored several times a day. During each census, the number and species (or genus, if the species was unknown) of visitors in each focal flower were noted. Overall, 12 350 censuses were conducted. Although detailed information on insect visitation and behaviour is available for seven of the ten seasons included, and although potential pollinators may vary in efficiency (at pollen export and seed production; Sánchez-Lafuente et al., 1999; A. M. Sánchez-Lafuente et al., unpubl. res.), for the purpose of this study visitation rates of all insects will be considered pooled (as number of visit per flower per hour), because we are interested in the realized selection on plant traits by the whole and diverse visitor assemblage, and its consequences for plant fitness in terms of seed production, seed mass, seed germination and seedling survival.

Every season, a number of floral traits of each tagged plant, related to floral advertisement and sexual functions were measured. These were: petal size (a measure of flower size, as the average length of two random petals per flower); number of anthers/flower (counted when flowers began to senesce, to avoid interfering with insect visitation); number of carpels/flower; carpel size (averaged from two carpels per flower, if available, and measured from carpel top to basal disc); number of ovules/carpel, number of ovules/flower and number of seeds/flower (all estimated after fruit collection). Measurements were made with an electronic calliper to the nearest ± 0·01 mm. All seeds collected were individually weighed to the nearest ± 0·1 mg. A total of 5845 seeds were weighed, 2275 from Hazadillas and 3570 from Navalopo. Plant size, as the height of the flowering stem, was also recorded for each plant, site and season, and estimated in centimetres to the nearest ± 1 cm.

Although the number of flowers per plant may vary between seasons, in the study area plants only produced as many as two flowers per season. Only 16 plants (n = 12 in Hazadillas and n = 4 in Navalopo) in the ten seasons analysed eventually produced more than one flower. Of these, five produced two flowers in three of ten seasons; four produced two flowers in two seasons, and seven produced two flowers in one season. Given this small sample size, the consequences of differences in flower production were not analysed, and all plants producing more than one flower per season were removed from the analyses.

Seed germination and seedling survival

To test germination ability, seeds collected in 2000, 2003 and 2004 were sown in plots located in their original population. Plants were considered as blocks, and their seeds randomly sown on patches over the selected plot. Overall 1423 seeds were assigned one of six categories, according to their mass: <250, 250–500, 500–750, 750–1000, 1000–1250 and >1250 mg. In 2000, seeds were sown from 37 plants from Hazadillas (191 seeds) and 35 from Navalopo (343 seeds), in 2003 from 29 and 32 plants (163 and 316 seeds), respectively, and from 31 and 36 plants (109 and 301 seeds) in 2004. Each patch was covered by a wire mesh until germination, to avoid seed losses due to unwanted biotic or abiotic effects (e.g. rodents, wild boars, heavy rain, etc.). Plots were visited every 3 months until germination, then every 15–20 d until seedling seasonal senescence. After germination all cages were removed, allowing seedlings to be exposed to natural environmental conditions. Seedling survival was analysed from data obtained from seeds collected and sown in 2000 (i.e. seedling survival after 5 years in the field).

Spatial distribution of plants

In 2006, using a portable GPS device (Garmin eTrex Vista Hcx), the boundaries of each study site were registered from four corners of a rectangular area (5575 m2 in Hazadillas and 7350 m2 in Navalopo), inside which all the tagged plants were located. A reference grid (each grid cell was 25 m2) was calculated within each rectangular area, and the location of all tagged plants within each grid cell was registered. Finally, all GPS locations were converted to Cartesian co-ordinates referred to each rectangular area.

Data analyses

First, to find out whether pollinator diversity was related to seed production, an analysis was carried out at plant level. The Shannon–Wiener (e.g. Magurran, 1988; Oksanen et al., 2008) diversity index was calculated for all pollinators visiting each plant, from direct counts in censuses over seven seasons (2000–2006). Thus, one pollinator diversity value was obtained for each plant and season. A general linear model was used to test the influence of pollinator diversity on seed production, including season and site as fixed factors, and the total number of ovules per flower as a continuous predictor, to account for its effect on seed production.

The overall contribution of the different floral traits to visitation rates (visits per plant h−1), seed production and seed mass, was analysed using mixed models, with plant and floral traits as fixed factors and season as a random factor. The effect of plant size on the variation of the traits considered was first assessed by multivariate analysis of variance (MANOVA). Since no relationship was found for any site or season, plant size was not used as a covariate.

Analysis of selection on plant traits, which significantly helped to explain visitation rates, seed production and seed mass (as fitness estimates), were carried out using selection gradients models (Lande and Arnold, 1983), an estimation of the direct effects of selection on a given trait, independently of any indirect effects caused by selection on the traits with which this trait is correlated (Aspi et al., 2003; Hersch and Phillips, 2004). Two different approaches were used: (1) in order to look for heterogeneity in selection over time, selection gradients and their confidence intervals were calculated for each of the years in the survey, with their corresponding plant traits and fitness estimates; (2) a life-time fitness (i.e. accumulated fitness over the ten seasons) was calculated for each estimate as: total visitation rates (total number of visits per plant h−1), total number of seeds produced and total mass of all seeds produced. In this analysis, plant traits were averaged over the ten seasons. In both approaches, variables contributing significantly to explain differences in fitness (according to the first analysis) were used as predictors. Furthermore, models included quadratic terms in order to detect non-linear relationships. In both approaches, coefficients express variation in standard deviation units, since fitness values were transformed to relative by dividing each measure by the mean absolute fitness in the population (Lande and Arnold, 1983) and predictors were standardized. Although the sample sizes used in this study may be considered small to estimate accurate selection gradients (see Kingsolver et al., 2001; Hersch and Phillips, 2004; but see Knapczyk and Conner, 2007) a larger sample size could not be used given the patchy distribution of plants in the study range (occurring mostly isolated or in groups of a few individuals; only the populations used are relatively large, between 100 and 120 plants).

Selection gradients measure the plausible change in a phenotypic trait resulting from direct selection, while holding selection on correlated traits constant. However, since genetic correlations among floral traits may be likely, an eventual evolutionary response of a trait due to selection may be conditioned by its correlation with other traits. The magnitude of the floral phenotypic integration may be considered as a surrogate for genetic correlations among traits (Cheverud, 1988; Waitt and Levin, 1998), and may be useful to evaluate whether eventual responses to selection may occur on a selected trait independently of other traits. Thus, a high floral integration may indicate that traits do not vary independently of each other, which may impose limitations to an eventual evolutionary response if a conflicting selection is acting on different correlated traits. The magnitude of floral integration (INT) was calculated for the measured floral traits, averaged over the ten seasons, as the variance of the eigenvalues (λi where i varies between 1 and the number of traits) of the character correlation matrix [INT = var (λi); Wagner, 1984; Cheverud et al., 1989; Herrera et al., 2002]. Standard errors and confidence intervals of the integration index were obtained by bootstrapping (n = 5000 repetitions). Floral integration was considered significant if the confidence interval did not include zero. The magnitude of integration was corrected for sample size as suggested by Wagner (1984).

Whether certain plants consistently had larger flowers, received more visits and produced more and heavier seeds was tested over the ten seasons analysed. In other words, we wanted to test if some plants may consistently perform better than others over time, based upon their flower size and fitness estimates. For this purpose, first the Kendall coefficient of concordance (W; see Legendre, 2005) was used to assess whether the rank of each individual plant was maintained over time for each fitness estimator (i.e. significant values of W indicate that the rank of each individual plant is not independent across seasons). Secondly, tests were carried out to find out if the rank of each individual plant was the same over time for all fitness estimators. For this purpose, a matrix was created in which columns represented the average rank of each plant for each season and fitness estimator, and rows represented the plants. The Euclidean distances were calculated between rows (i.e. plants) and used partitioning around medoids (PAM; Maechler et al., 2005) to calculate plant clustering. PAM computes a variable number (k) of representative objects, called medoids, that can be defined as objects of a cluster whose average dissimilarity to all the data in the cluster is minimal. After finding the set of medoids, each observation in the dataset is assigned to its nearest medoid. These analyses were carried out independently for each study site.

Two more fitness estimates were considered after seed dispersal. First, germination likelihood (for seeds produced in 2000, 2003 and 2004) was assessed as a binomial variable (germinated vs. not germinated) and analysed by generalized mixed linear models. Plant, site and seed mass class (see above) were used as fixed factors, and sowing season as a random factor. Seed mass was used as a factor rather than as a continuous predictor, because seeds were not individually identified in the sowing patches. A similar approach was used to assess differences in seedling survival. In this case, survival after 5 years in the field was estimated from germination data of seeds sown in 2000. The response was modelled as binomial (alive vs. dead) in a generalized linear model with the same factors indicated above; however, season was disregarded since only data for 2000 were used.

To find out if the spatial distribution of flower sizes, visitation rates, seed production and seed mass occurred at random in each study site, tests were carried out using Moran's I. This is a weighted correlation coefficient used to detect departures from spatial randomness, indicating spatial patterns. This should indicate if there were particularly suitable patches in the study sites where plants might develop larger flowers, receive more visits or produce more and/ heavier seeds. This analysis requires measuring a spatial weights matrix, reflecting distance between neighbours. The matrix was obtained from GPS locations converted to Cartesian co-ordinates (see above). A one-tailed test was used, in order to detect specifically if the spatial distribution of the variables was more clustered than could be expected by chance.

All analyses had been carried out using the R 2·8·0 environment for statistical computing (R Development Core Team, 2008; freely available online), with NLME (Pinheiro et al., 2006), IRR (Gamer et al., 2007), CLUSTER (Maechler et al., 2005), BOOT (Davison and Hinkley, 1997; Canty and Ripley, 2008); VEGAN (Oksanen et al., 2008) and SPDEP (Bivand, 2008) packages.


Visitor assemblage and trait variation among plants, seasons and sites

More than 60 different species of visitors were detected at least once in P. broteroi over the ten seasons considered. Among these, 45·46 % were Hymenoptera (mainly bees, both social and solitary, including Apis mellifera, Bombus terrestris and Xylocopa violacea as the most important species), followed by Coleoptera (34·09 % of the species registered). Diptera (Fam. Syrphidae) and Hemiptera were also represented in censuses, with 9·09 % and 6·82 %, respectively, of the species recorded. Finally, some Orthoptera were also observed, although rarely during censuses (4·54 % of the species).

Pollinator diversity over seven seasons ranged 0·26–3·42 in Hazadillas and 0·61–3·82 in Navalopo. Diversity was slightly lower in Hazadillas (mean ± s.d.; 2·01 ± 075) than in Navalopo (mean ± s.d.; 2·19 ± 0·64; (F1,449 = 4·04, P < 0·05), and no overall differences were found among seasons (F6,449 = 0·71, P = 0·64). Pollinator diversity had an overall influence on seed production (F1,449 = 29·48, P << 0·001; coefficient ± s.d.; 0·13 ± 0·09), once the effect of the number of ovules/flower, which could affect seed production, was taken into account (F1,449 = 11·40, P < 0·001). Slightly significant differences in seed production were also found between sites (F1,449 = 4·72, P < 0·04) and among seasons (F6,449 = 2·53, P < 0·02). However, the influence of pollinator diversity on seed production was consistent between sites (site × diversity; F1,449 = 2·95, P = 0·08) and among seasons (season × diversity; F6,449 = 1·66, P = 0·13).

The seasonal variation in mean ± standard deviation values of the floral traits over the ten seasons is presented in Fig. 1, and variance partitioning of traits among plants, seasons and sites in Fig. 2. Most variation was found among plants, and only a small fraction occurred among seasons or sites. Traits related to maternal fecundity were less subjected to variations among seasons or sites than plant size and traits related to flower advertisement. While season (MANOVA: F9,454 = 3·31, Wilks' lambda = 0·72, P < 0·001) and site (MANOVA: F1,454 = 17·52, Wilks' lambda = 0·847, P < 0·001) both had a significant effect on the variation of all traits, no significant relationship was found between plant size and any other trait (MANOVA: F1,454 = 1·89, Wilks' lambda = 0·97, P = 0·09). This result was consistent across sites and seasons (MANOVA: plant size × site × season; F9,454 = 0·91, Wilks' lambda = 0·91, P = 0·64).

Fig. 1.
Seasonal variation (mean ± s.d.) of the floral traits measured over ten seasons.
Fig. 2.
Variance decomposition of the floral traits measured among plants, seasons and sites.

Differences among plants in fitness estimators and selection gradients on floral traits

Significant differences were found among plants in visits h−1, number of seeds and seed mass (Table 1). When analysing which of the measured traits could contribute to such differences, it was found that petal size was the only trait related to floral advertisement that explained visitation rates and seed production (Table 1). Among traits related to maternal fecundity, the number of carpels/flower and ovules/flower influenced seed production, but only in Navalopo (Table 1). Finally, none of the traits contributed to explain seed mass in any site (Table 1). From these results, the following hypothesis was tested: that petal size could be subjected to selection by pollinators, and that pollinator selection of flowers, based upon their size, could eventually generate selection gradients on this trait that could, in turn, originate differences among plants in several estimates of fitness.

Table 1.
Results of mixed linear models predicting plant visitation rates (visits h−1), number of seeds and seed mass as a function of plant traits in each site: (A) Hazadillas and (B) Navalopo

Directional selection gradients on petal size through visitation rates were positive in all seasons and statistically significant in five (in Hazadillas) and six (in Navalopo) of the ten seasons considered in each population (Fig. 3). Besides, negative values for some of the CI that included zero were very small, compared with the positive ones. None of the non-directional selection gradients was significant, and there was a higher inconsistency in the sign of non-directional versus directional selection gradients. The magnitude of the selection gradients were not related to mean petal size (data for 1997–2006) or pollinator diversity (2000–2006), but differences among seasons (2000–2006) in relative abundance of bees in the pollinator assemblage were positively related to the directional selection gradients in Hazadillas (Spearman rank correlation; r = 0·86, P < 0·02, n = 7), although only marginally in Navalopo (r = 0·74, P = 0·08, n = 7). A similar result was found for selection gradients through seed number, which were positive and significant in five (Hazadillas) and four (Navalopo) of ten seasons in each study site. Only in 2005 and 2006 were directional selection gradients negative, but not significant. Again, none of the non-directional selection gradients was significant, and they exhibited a higher inconsistency in the sign. The results were quite different when selection gradients were estimated through seed mass. In this case, none of the directional selection gradients was significant, while six (Hazadillas) and four (Navalopo) stabilizing gradients were significant out of the ten in each population. Overall, in three of the ten seasons analysed (both populations pooled), selection gradients on petal size were significant according to the tree fitness estimators (Fig. 3), although in most seasons one or two of the estimates was responsible for a significant gradient. When all seasons were considered together, a consistent, directional selection on petal size through total visits and total seed production was found in both study sites (Table 2). Furthermore, a stabilizing selection on petal size through total seed mass was also found in Hazadillas (this same trend was detected, although not significant, in Navalopo; Table 2). This result suggests a strong selection by pollinators on plants which, on average in the long term, produce larger flowers, which in turn may translate into a higher seed production (Fig. 4). However, selection through seed mass of a comparable intensity may also be acting on plants producing flowers of an intermediate size, which may produce heavier seeds than larger flowers.

Fig. 3.
Standardized selection gradients on petal size through three different fitness estimates, in both study sites (Hazadillas and Navalopo) over ten seasons. Fitness estimated through (A) visitation rates, (B) seed number, and (C) seed mass. Closed circles ...
Fig. 4.
Relationship between mean petal size (averaged over ten consecutive seasons) and three estimators of life-time fitness (total visitation rates, total seed production and total seed mass) in the two study sites. See also Table Table22.
Table 2.
Selection gradients (directional β and non-directional γ) on petal size through three different life-time (i.e. for ten consecutive flowering seasons) fitness estimates

Floral integration was significant in both populations (INT ± s.e.; Hazadillas: 1·15 ± 0·55, 95 % CI = 0·65–1·66; Navalopo: 1·82 ± 0·54, 95 % CI = 1·04- 2·84), suggesting a consistent covariation within populations among floral traits. The percentage of INT with respect to the maximum possible for five traits was 23·00 % in Hazadillas and 36·40 % in Navalopo. For comparative purposes, floral integration was also calculated for the same traits measured in 1994–1995 in three distant (>100 km) P. broteroi populations in another mountain range. INT ± s.e. were 0·81 ± 0·23 (95 % CI = 0·59–1·29), 0·75 ± 0·13 (95 % CI = 0·52–0·98) and 0·75 ± 0·15 (95 % CI = 0·59–0·80) in each of the three populations, respectively, accounting for 16·20 %, 15·00 % and 15·00 % of the maximum possible INT for five traits. These results of floral integration at a wider geographical range indicate that INT is considerably high in the study populations.

Seed germination and seedling survival

Under field conditions, most P. broteroi seeds germinate in the second spring after sowing, although some may delay germination until the third or even the fourth spring (A. M. Sánchez-Lafuente et al., unpubl. res.). Thus, germination success was analysed from data collected at the end of the fourth spring after sowing, testing the effect of individual differences among genotypes, sites and seed mass classes as fixed effects, and including season as a random factor.

Significant differences were found among plants (F1,1230 = 8·59, P < 0·001) and seasons (likelihood ratio = 43·94, P < 0·001) in germination likelihood, and a significant effect of seed mass was also observed (F5,1230 = 30·24, P < 0·001; see Table 4A). However, post-hoc tests indentified that such differences only originated in the low germination rate of the class grouping the smallest seeds, compared with the rest. The interaction between plant and seed mass was also significant (plant × seed mass class; F5,1230 = 4·70, P < 0·001), suggesting that germination likelihood of seeds of different mass differ among genotypes. However, no overall differences in germination likelihood were found between sites (F1,1230 = 2·51, P = 0·11), while the effect of the site did not influence the relationship between plants and seed mass (plant × seed mass class × site; F5,1230 = 1·16, P = 0·33).

Table 4.
(A) Mean values of seed germination likelihood related to size class in the two sites and for the three seasons analysed. (B) Mean value of seedling survival likelihood from seeds sown in 2000, related to size class in the two study sites

Regarding survival, only a small fraction of the seedlings emerged from seeds sown in 2000 were alive 5 years after germination (see Table 4B), and the survival likelihood did not differ according to any of the factors analysed. Thus, no differences were found between plants, sites or seed mass classes in survival likelihood after 5 years in the field. A marginal difference was detected in the interaction between plants and sites (plant × site; F1,246 = 3·19, P = 0·08). However, since only one season was analysed it cannot be evaluated whether such a trend may be consistent for different seasons.

Seasonal plant rankings related to fitness estimators

Plant rankings according to pollinator diversity, visitation rates, seed production and seed mass were not independent across seasons or sites, except for diversity in Hazadillas and seed mass in Navalopo (Table 3). Thus, the potential advantage of particular genotypes according to their flower size, visitation rates and seed production, does not seem to vary among seasons. The marginal difference found between sites, related to the effect of pollinator diversity on seed production, may be due to the inconsistent rank of the plants in Hazadillas. On the other side, the independence of ranks found for seed mass at Navalopo (and the low critical value for the same variable found at Hazadillas) suggests that investment in seed mass may be highly variable among genotypes, and largely dependent on seasonal differences in environmental conditions.

Table 3.
Values of the Kendall coefficient of concordance (W), testing whether the rank of each individual plant in each site is maintained across seasons for insect diversity and three variables related to fitness

When fitness estimators were considered as a whole, partitioning analyses extracted a number of plant groups according their mean rank across seasons (Fig. 5). Thus, seven clusters of similar plants over time were found for Hazadillas and six for Navalopo. Components on which clusters were selected were related to a gradient of visitation rates and seed production (component in abscissae), and a gradient on flower size and seed mass (component in ordinates). Cluster distribution suggested that, although most plants exhibit a great variation in relative fitness through the different estimators, certain plants in each population consistently had larger flowers, received more visits and produced more seeds than others over time (Fig. 4).

Fig. 5.
Results of the PAM analyses testing for plant clustering in the two study sites, according to the rank of each plant over ten seasons, related to three estimates of life-time fitness (total visitation rates, total seed production and total seed mass). ...

Spatial autocorrelation analyses

The spatial autocorrelation coefficients for plants in the study populations are presented in the Appendix. The only significant results indicate that, in 1997 and 1998, and for plants in Hazadillas, there was a greater dissimilarity (negative correlation) among plants, regarding flower size, than that expected by chance. The same was found for seed production in Navalopo in 2000. Yet, the coefficients obtained for these significant cases are negligible. Consequently, there is overwhelming evidence that the spatial distribution of flower sizes, visitation rates and seed production in the study populations occur at random.


Selection gradients, phenotypic integration and changes in flower size

Using data of ten consecutive flowering seasons, and despite the very diverse pollinator assemblage visiting flowers of P. broteroi, and its unspecialized pollination system, strong selection gradients, acting on petal size through different estimators of fitness, were found in two close populations of this plant species sharing the same pollinator assemblage. In half of the seasons analysed, significant gradients (directional or non-directional) acting on petal size through visitation rates, seed number and seed mass were found. Similar results were found when gradients were calculated on petal size averaged over the ten seasons, using total visitation rates, seed production and seed mass as estimators of life-time fitness. Taken together, the results suggest that, even assuming inter-seasonal variations in floral and seed traits, and in the relative abundance of the pollinator assemblage (A. M. Sánchez-Lafuente and R. Parra, unpub. res.), plants whose flowers present longer petals are overall more attractive to insects in the long term, receive more visits (particularly in Hazadillas) and produce more seeds than others. Thus, despite the unsophisticated pollination system of P. broteroi, pollinators may have a variable, but overall significant, potential to originate differences in fitness among flowers of different phenotypes. While in other species with generalist pollination systems the role of pollinators may be obscured by the influence on fitness of traits related to fecundity (e.g. floral display; Gómez, 2000), in this few-flowered species the role of pollinator visitations may be crucial for fitness associated with plant reproduction.

It was found that floral integration in Hazadillas was 1·58 times lower than that in Navalopo, which had a more diverse pollinator assemblage. It was also found that integration in the study populations was higher than that reported for the same species in populations located in a distant region with a different pollinator assemblage (see Sánchez-Lafuente, 2002). Thus, the composition and selection patterns of the pollinator assemblage could also originate different magnitudes of floral integration. Furthermore, the magnitude of floral integration presented in this study is even higher than that reported by Herrera et al. (2002) and Ordano et al. (2008) in their analyses of different insect-pollinated species with contrasting differences in floral phenotype and pollinating agents. This high floral integration suggests that an eventual modification of the floral phenotype is unlikely to occur through independent variation of different floral traits. In turn, a high integration may facilitate or constrain an eventual morphological change, depending on the sign and magnitude of the selection acting on different floral traits. However, the magnitude of the whole floral integration does not necessarily correlate with the strength of selection on certain floral traits. For instance, a lower whole floral integration may be detected in cases of selection on intrafloral integration (Ordano et al., 2008). But alternatively, differences in the magnitude of integration may not be originated by the interaction with pollinators, but by developmental constraints and/or genetic drift (e.g. Herrera, 2001; Herrera et al., 2002). This last possibility cannot be discarded for P. broteroi, which in its range mostly occurs in small populations or patches, or as a continuum of isolated individuals.

In a previous study (Sánchez-Lafuente, 2002) the possibility was pointed out that differences in pollinator type, size and behaviour at the plants, could be responsible for the differences in flower size and integration found among these P. broteroi populations and others located in the distant region mentioned. This could be possible through a selection mosaic, at a broad scale, in which plant traits, interactors and fecundity traits may differ more among regions than between populations within regions (e.g. Rey et al., 2006). In such a selection mosaic, and despite plants may be involved in similar interactions, regional variation in abundance of biotic agents, in the abiotic environment, or in concurrent interactions of any of the partners with another species, may yield locally adapted outcomes of the interactions, and promote a local differentiation of any traits subjected to selection. The long-term directional selection found on petal size, the comparatively higher breeding success (in terms of seed production) of plants with larger petals, and the higher floral integration found, suggest that pollinators of the study populations may have the ability to promote an overall modification of the floral phenotype in this plant species at a local scale (including advertisement-related and reproductive traits, assuming they may be heritable; e.g. Robertson et al., 1993; Campbell, 1997; Mazer et al., 1999). Interestingly enough, larger flowers may become a limiting factor for reproductive success if they exclude a number of small visitor species as efficient pollinators (Sánchez-Lafuente, 2002).

Limitations to the evolutionary change in flower size

One of the suggested limitations to an eventual evolutionary response to selection is the influence of the environmental conditions on the expression of phenotypic traits (e.g. Rausher, 1992). Although the spatial correlation analysis suggests that the distribution of plants regarding flower size, visitation rates, seed production and seed mass, occurs at random in the study populations, the spatial distribution of any particularly suitable spots for plant growth and/or reproduction is unexplored. Thus, the possibility cannot be totally discarded that some genotypes might grow on favourable microsites, and produce flowers with larger petals influenced by particular conditions (e.g. favourable microclimate or resource availability). However, some studies have shown that flower morphology is not necessarily influenced by the substrate type (Herrera, 1993), while others indicate that the influence of plant location versus traits on different fitness estimators may yield opposite results (e.g. Herrera, 1995; Sanchez-Lafuente et al., 2005). We suggest that differences among plants in flower size due exclusively to the influence of microsite conditions may be unlikely. First, no significant relationship was found between plant size and any floral trait. If favourable conditions of certain microsites may influence plant growth and development, it is reasonable to think that their influence should show at different stages. Hence, larger plants would be expected to develop flowers with larger petals, more ovules, carpels or pollen grains, but this was not the case. Secondly, it was found that plant scores for petal size were consistent across seasons, but not so for seed mass, suggesting that the latter variable may be more influenced by environmental factors.

Another limitation to an eventual evolutionary change could originate in conflicting selection acting on the same, or correlated, traits. In both populations, a significant directional selection on petal size through seed production was found, but a significant stabilizing selection on the same trait through seed mass (only marginally in Navalopo) was also found. This appears as a selection conflict on petal size through a trade-off between seed number and size (flowers may benefit from higher visitation rates and seed production by having longer petals, while the heaviest seeds are produced by flowers with petals of an intermediate size, which in turn are less visited and produce less seeds). However, this conflict seemed solved when no overall differences were found in germination success (except for the lowest mass class) or seedling survival related to seed mass in any study site. While a large number of studies have shown that seed mass may be an important factor accounting for differences between seeds in their ability to germinate and produce seedlings with higher chances to recruit (e.g. Martin and Lee, 1993; Khurana and Singh, 2004; Baraloto et al., 2005, among others; see also Moles and Westoby, 2006), others have found that the size effect may be constrained by factors related to the abiotic or biotic environment in which dispersion and germination take place (e.g. Gómez, 2004; Rey et al., 2004). Among the latter, current data suggest that herbivory may be crucial to explain adult and seedling survival in this species (Sánchez-Lafuente, 2007; A. M. Sánchez-Lafuente, unpubl. res.), particularly in combination with abiotic factors related to the extreme summer conditions (high temperatures and droughts) in Mediterranean areas (e.g. Angelopoulos et al., 1996; Kitajima and Fenner, 2000; Rey et al., 2004; Manzaneda et al., 2005; Ramírez et al., 2006). Thus, a combination of biotic and abiotic environmental conditions at a local scale may be responsible for this outcome in terms of germination and survival. However, this result points to another likely limitation to the evolutionary change, originated by the demographic patterns of the study species. An extremely limited seedling survival was found under natural conditions in both study populations, suggesting that P. broteroi might rely mainly upon adult survival to maintain population size, while an eventual population increase through seedling establishment could be limited to periods of extremely suitable environmental conditions. Similar results seem to be found in the close relative Paeonia cambessedesii (A. Traveset, Spanish Research Council, CSIC, pers. comm.), whose seedlings may grow very slowly under natural conditions, although they may reach a reproductive stage very quickly in favourable controlled environments. Since the seeds sown to test the germination and survival ability were obtained from plants not subjected to any experimental manipulation (e.g. plant or flower size, autogamy vs. xenogamy, pollination treatment, etc.), the same distribution of seed sizes, and of germination and survival likelihood, is expected in the present sample compared with those found for naturally dispersed seeds in the populations (A. M. Sánchez-Lafuente et al., unpubl. res.).

From a demographic perspective, these results also support findings suggesting that individual differences in the abilities to exploit variable environmental conditions may lead particular genotypes to become the main contributors to the maintenance and development of plant populations in the wild. The present analyses indicated that a number of plants in each site might do better than others by consistently producing flowers with longer petals, receiving more visits and producing more mature seeds than others in the long term, despite the significant maternal effects accounting for differences among these variables (but see Totland, 2004). Similar results have been pointed out in studies on different species and at different stages of the life cycle. For instance, Alcántara et al. (1997) found that individual Olea europaea var. sylvestris trees whose seeds were dispersed more by frugivorous birds were roughly the same in two consecutive seasons; thus they were likely to become the main contributors to the seed bank. Also, in a 6-year study with Lavandula latifolia, Herrera (2000a) showed that seedlings from seeds of certain mother plants did better than others at germination and survival, and were more likely to recruit into the population.


In this study, a strong directional selection was detected operating on the same phenotypic trait (petal size), after ten consecutive seasons, and through different fitness estimators (insect visitation rates and total seed production), while seed mass did not influence seed germination or seedling survival. If natural selection is expected to act increasing offspring number under environmental unpredictability (Simons, 2007; but see Kisdi, 2007), the results suggest that plants having larger flowers are benefited by producing more lighter seeds than fewer heavier seeds, because seed mass does not confer any advantages for germination and survival. If there is a selection process on flower size through visitation rates and seed production, and assuming that floral traits were heritable, plants with larger flowers may contribute disproportionately to the seed pool, and may have higher chances that any of their descendants could be represented in future generations.

However, and according to the selection process described here, the chance that differences in flower size among different populations and regions have originated by rapid change may be unlikely. First, a conflicting selection on flower size through a trade-off between seed number and size was detected; but secondly, and most importantly, the results indicate a very low seedling survival under natural conditions, which may imply a very large generation time. Consequently, despite the magnitude of the selection found, the influence of pollinators on the floral phenotype does not seem to fully explain differences among populations and regions in floral form and integration.


We thank Pedro Rey, Julio Alcántara, Don Levin and two anonymous reviewers, for helpful comments on earlier versions of this paper, and the Consejería de Medio Ambiente (Junta de Andalucía) for providing permissions for field work in the study areas. While writing this paper A.M.S.-L. was supported by grant BOS2003-00292, from the Spanish Ministerio de Ciencia y Tecnología, and R.P. was supported by grant CGL2006-02860 from the Spanish Ministerio de Educación y Ciencia.


Results of the autocorrelation analyses testing the spatial association between plants in visitation rate (visits h−1), petal size, seed production and seed mass in the two sites (Hazadillas and Navalopo). Significant values are presented in bold.

Visits h−1
Petal size
No. of seeds
Seed mass


  • Alcántara JM, Rey PJ, Valera F, Sánchez-Lafuente AM, Gutierrez JE. Habitat alteration and plant intra-specific competition for seed dispersers: an example with Olea europaea var. sylvestris. Oikos. 1997;79:291–300.
  • Alcántara JM, Rey PJ, Manzaneda AJ, Boulay R, Ramírez JM, Fedriani JM. Geographic variation in the adaptive landscape for seed size at dispersal in the myrmecochorous Helleborus foetidus. Evolutionary Ecology. 2007;21:411–430.
  • Angelopoulos K, Dichio B, Xiloyannis C. Inhibition of photosynthesis in olive trees (Olea europaea L.) during water stress and rewatering. Journal of Experimental Botany. 1996;47:1093–1100.
  • Aspi J, Jäkäläniemi A, Tuomi J, Siikamäki P. Multilevel phenotypic selection on morphological characters in a metapopulation of Silene tatarica. Evolution. 2003;57:509–517. [PubMed]
  • Bivand R. SPDEP: Spatial dependence: weighting schemes, statistics and models. 2008. R package version 0.4-20. .
  • Baraloto C, Forget PM, Goldberg DE. Seed mass, seedling size and neotropical tree seedling establishment. Journal of Ecology. 2005;93:1156–1166.
  • Campbell DR. Genetic correlation between biomass allocation to male and female functions in a natural population of Ipomopsis aggregata. Heredity. 1997;79:606–614.
  • Canty A, Ripley B. boot: Bootstrap R (S-Plus) Functions. 2008. R package version 1.2-34. .
  • Cheverud JM. A comparison of genetic and phenotypic correlations. Evolution. 1988;42:958–968.
  • Cheverud JM, Wagner GP, Dow MM. Methods for the comparative analysis of variation patterns. Systematic Zoology. 1989;38:201–213.
  • Davison AC, Hinkley DV. Bootstrap methods and their applications. Cambridge: Cambridge University Press; 1997.
  • Endler JA. Natural selection in the wild. Princeton, NJ: Princeton University Press; 1986.
  • Fedriani JM, Rey PJ, Garrido JL, et al. Geographical variation in the potential of mice to constrain an ant-seed dispersal mutualism. Oikos. 2004;105:181–191.
  • Galen C. Measuring pollinator-mediated selection on morphometric floral traits: bumblebees and the alpine skypilot, Polemonium viscosum. Evolution. 1989;43:882–890.
  • Galen C. Rates of floral evolution: adaptation to bumble bee pollination in an alpine wildflower. Polemonium viscosum. Evolution. 1996;50:120–125.
  • Gamer M, Lemon J, Fellows I. IRR: Various coefficients of interrater reliability and agreement. 2007. R package version 0.70. .
  • Gómez JM. Phenotypic selection and response to selection in Lobularia maritima: importance of direct and correlational components of natural selection. Journal of Evolutionary Biology. 2000;13:689–699.
  • Gómez JM. Bigger is not always better: conflicting selective pressures on seed size in Quercus ilex. Evolution. 2004;58:71–80. [PubMed]
  • Gómez JM, Zamora R. Spatial variation in the selective scenarios of Hormathophylla spinosa (Cruciferae) American Naturalist. 2000;155:657–668. [PubMed]
  • Herrera CM. Pollinator abundance, morphology, and flower visitation rate: analysis of the ‘quantity’ component in a plant-pollinator system. Oecologia. 1989;80:241–248.
  • Herrera CM. Selection on floral morphology and environmental determinants of fecundity in a hawk moth-pollinated violet. Ecological Monographs. 1993;63:251–275.
  • Herrera CM. Microclimate and individual variation in pollinators: flowering plants are more than their flowers. Ecology. 1995;76:1516–1524.
  • Herrera CM. Long-term dynamics of Mediterranean frugivorous birds and fleshy fruits: a 12-yr study. Ecological Monographs. 1998;68:511–538.
  • Herrera CM. Individual differences in progeny viability in Lavandula latifolia: a long-term field study. Ecology. 2000;a 81:3036–3047.
  • Herrera CM. Measuring the effects of pollinators and herbivores: evidence for non-additivity in a perennial herb. Ecology. 2000;b 81:2170–2176.
  • Herrera CM. Deconstructing a floral phenotype: do pollinators select for corolla integration in Lavandula latifolia? Journal of Evolutionary Biology. 2001;14:574–584.
  • Herrera CM, Cerdá X, García MB, et al. Floral integration, phenotypic covariance structure and pollinator variation in bumblebee-pollinated Helleborus foetidus. Journal of Agricultural, Biological & Environmental Statistics. 2002;15:108–121.
  • Hersch EI, Phillips PC. Power and potential bias in field studies of natural selection. Evolution. 2004;58:479–485. [PubMed]
  • Hoëkstra E, Hoëkstra JM, Berrigan D, et al. Strength and tempo of directional selection in the wild. Proceedings of the National Academy of Sciences of the USA. 2001;98:9157–9160. [PubMed]
  • Khurana E, Singh JS. Germination and seedling growth of five tree species from tropical dry forest in relation to water stress: impact of seed size. Journal of Tropical Ecology. 2004;20:385–396.
  • Kingsolver JG, Hoëkstra HE, Hoëkstra JM, et al. The strength of phenotypic selection in natural populations. American Naturalist. 2001;157:245–261. [PubMed]
  • Kisdi E. No direct selection to increase offspring number of bet-hedging strategies in large populations: Simons' model revisited. Journal of Evolutionary Biology. 2007;20:2072–2074. [PubMed]
  • Kitajima K, Fenner M. Ecology of seedling regeneration. In: Fenner M, editor. Seeds: the ecology of regeneration in plant communities. Wallingford: CABI International; 2000. pp. 331–359.
  • Knapczyk FN, Conner JK. Estimates of the average strength in natural selection are not inflated by sampling error or publication bias. American Naturalist. 2007;170:501–508. [PubMed]
  • Lande R, Arnold SJ. The measurement of selection on correlated characters. Evolution. 1983;37:1210–1226.
  • Legendre P. Species associations: Kendall coefficient of concordance revisited. Journal of Agricultural, Biological & Environmental Statistics. 2005;10:226–245.
  • Maechler M, Rousseeuw P, Struyf A, Hubert M. Cluster analysis: basics and extensions: an R library. 2005. .
  • Magurran AE. Ecological diversity and its measurement. Princeton, NJ: Princeton University Press; 1988.
  • Manzaneda AJ, Sperens U, Garcia MB. Effects of microsite disturbances and herbivory on seedling performance in the perennial herb Helleborus foetidus (Ranunculaceae) Plant Ecology. 2005;179:73–82.
  • Martin ME, Lee D. Self pollination and resource availability affect ovule abortion in Cassia fasciculata (Caesalpiniaceae) Oecologia. 1993;94:503–509.
  • Mazer SJ, Delesalle VA, Neal PR. Responses of floral traits to selection on primary sexual investment in Spergularia marina: the battle between the sexes. Evolution. 1999;53:717–731.
  • Moles AT, Westoby M. Seed size and plant strategy across the whole life cycle. Oikos. 2006;113:91–105.
  • Oksanen J, Kindt R, Legendre P, et al. Vegan: Community Ecology Package. 2008. R package version 1. 15–0. .
  • Ordano M, Fornoni J, Boege K, Dominguez CA. The adaptive value of phenotypic floral integration. New Phytologist. 2008;179:1183–1192. [PubMed]
  • Pinheiro J, Bates D, Debroy S, Sarkar D. NLME: linear and non-linear mixed effects models. R package version 3.1-75. 2006. .
  • R Development Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2008. ISBN 3-900051-07-0; .
  • Rausher MD. The measurement of selection on quantitative traits: biases due to environmental covariates between traits and fitness. Evolution. 1992;46:616–626.
  • Ramírez JM, Rey PJ, Alcántara JM, Sánchez-Lafuente AM. Altitude and woody cover control recruitment of Helleborus foetidus in a Mediterranean mountain area. Ecography. 2006;29:375–384.
  • Rey PJ, Alcántara JM, Valera F, et al. Seedling establishment in Olea europaea: seed size and microhabitat affect growth and survival. Ecoscience. 2004;11:310–320.
  • Rey PJ, Herrera CM, Guitián J, et al. The geographic mosaic in predispersal interactions and selection on Helleborus foetidus (Ranunculaceae) Journal of Evolutionary Biology. 2006;19:21–34. [PubMed]
  • Robertson AW, Diaz A, MacNair MR. The quantitative genetics of floral characters in Mimulus guttatus. Heredity. 1993;72:300–311.
  • Sánchez-Lafuente AM. Floral variation in the generalist perennial herb Paeonia broteroi (Paeoniaceae): differences between regions with different pollinators and herbivores. American Journal of Botany. 2002;89:1260–1269. [PubMed]
  • Sánchez-Lafuente AM. Sex allocation under simulated herbivory in the generalist perennial herb Paeonia broteroi (Paeoniaceae) Plant Systematics and Evolution. 2007;265:59–70. [PubMed]
  • Sánchez-Lafuente AM, Rey PJ, Alcántara J, Valera F. Breeding system and the role of floral visitors in seed production of a ‘few-flowered’ perennial herb, Paeonia broteroi Boiss. & Reut. (Paeoniaceae) Ecoscience. 1999;6:163–172.
  • Sánchez-Lafuente AM, Guitián J, Medrano M, Herrera CM, Rey PJ, Cerdá X. Plant traits, environmental factors, and pollinator visitation in winter-flowering Helleborus foetidus (Ranunculaceae) Annals of Botany. 2005;96:845–852. [PubMed]
  • Schemske DW, Horvitz C. Temporal variation in selection on a floral character. Evolution. 1989;43:461–465.
  • Simons AM. Selection for increased allocation to offspring number under environmental unpredictability. Journal of Evolutionary Biology. 2007;20:813–817. [PubMed]
  • Thompson JN. The coevolutionary process. Chicago, IL: University of Chicago Press; 1994.
  • Thompson JN. The geographic mosaic of coevolution. Chicago, IL: University of Chicago Press; 2005.
  • Totland O. No evidence for a role of pollinator discrimination in causing selection on flower size through female reproduction. Oikos. 2004;106:558–564.
  • Valone TJ, Kaspari M. Interactions between granivorous and omnivorous ants in a desert grassland: results from a long-term experiment. Ecological Entomology. 2005;30:116–121.
  • Wagner GP. On the eigenvalue distribution of genetic and phenotypic dispersion matrices: evidence for a nonrandom organization of quantitative character variation. Journal of Mathematical Biology. 1984;21:77–95.
  • Waitt DE, Levin DA. Genetic and phenotypic correlations in plants: a botanical test of Cheveruds's conjecture. Evolution. 1998;80:310–319.
  • Waser NM, Price MV. Optimal outcrossing in Ipomopsis aggregata: seed set and offspring fitness. Evolution. 1989;43:1097–1109.
  • Waser NM, Chittka L, Price MV, Williams NM, Ollerton J. Generalization in pollination systems, and why it matters. Ecology. 1996;77:1043–1070.

Articles from Annals of Botany are provided here courtesy of Oxford University Press