Floral and inflorescence effects on variation in pollen removal and seed production among six legume species


†Author to whom correspondence should be addressed. E-mail: gaku@ees.hokudai.ac.jp


  • 1Most flowering plants display multiple flowers, so that the interaction with pollinators that determines their mating success can vary with both the characteristics of individual flowers and aggregate properties of the entire floral display, especially the number of open flowers (floral display size). These effects are seldom examined in concert and their collective consequences for interspecific differences in reproductive performance have not been considered previously.
  • 2In this paper, we characterize the relation of pollen removal and seed production to differences in floral and inflorescence characteristics among six species of herbaceous legumes (Fabaceae).
  • 3Several aspects of reproductive performance varied significantly among species with either plant traits or aspects of pollinator behaviour that depend on plant traits. Pollinator visitation, as measured by the ratio of pollen removal during 24 h to first-visit removal, varied positively with both nectar production per flower and floral display size. Bumble-bees visited more flowers per inflorescence on species with large floral displays, with no increase in the proportion of flowers visited. Pollen removal during a flower's first visit varied negatively among species with the mean number of flowers visited by bees per inflorescence.
  • 4These results indicate that floral and inflorescence traits act together to influence both pollinator energetics, which affects a plant species’ attractiveness, and the rate of pollen removal, which should affect pollen export. In contrast, neither pollen removal during 24 h, nor female fecundity varied significantly with floral or display characteristics.


Outcrossing animal-pollinated plants reproduce only by attracting pollinators, but they commonly attract more pollinators than would be necessary if a single visit removed all pollen and delivered enough pollen to fertilize all ovules. Although relatively few visits can deliver enough pollen to fertilize all ovules, many more visits may be required to remove a plant's, or even a flower's, pollen (Bell 1985; Spira, Snow & Puterbaugh 1996; Aizen & Basilio 1998; Bell & Cresswell 1998). This schedule of restricted pollen removal results from a variety of floral mechanisms that package or dispense pollen, limiting removal during individual visits (reviewed by Harder et al. 2001). Such mechanisms counteract a general negative relation between pollen removal per pollinator and the proportion of removed pollen grains that is exported to other stigmas on other plants. These diminishing returns on increased pollen removal result from several aspects of animal pollination, including pollinator grooming, layering of pollen on pollinator bodies and self-pollination between flowers on the same plant (reviewed by Harder et al. 2001). Because the pollen-export benefits of restricting pollen removal per pollinator increase with the number of pollinators that remove pollen, selection favours attraction of many pollinators if removal per visit can be restricted (Harder & Thomson 1989; Klinkhamer & de Jong 1993; Harder & Wilson 1994).

A plant species’ attractiveness, and pollination in general, depend on characteristics of both individual flowers and their display in inflorescences (reviewed by Harder et al. 2004). Nectar availability and the number of flowers open at once (floral display size) should be especially important traits influencing the attractiveness of plants to nectar-collecting pollinators. For such pollinators, foraging benefits vary positively with a plant's nectar availability, whereas foraging costs vary negatively with flower number, because expensive flights between plants are less frequent when pollinators visit many flowers per plant. Therefore, more pollinators are attracted to individual plants of a particular species with abundant nectar (e.g. Thomson 1988; Klinkhamer & de Jong 1990) and many open flowers (reviewed by Ohashi & Yahara 2001), although these effects have not been considered together. The relative attractiveness of different plant species also varies positively with their average nectar production (Pleasants 1981; Bosch, Retana & Cerdá 1997), but this interspecific relation has received less attention.

Nectar production and flower number also affect interactions of individual pollinators with a plant's sexual organs directly. The duration of a pollinator's visit to a flower varies positively with nectar availability (e.g. Harder 1986; Thomson 1986; Cresswell 1999; Kudo 2003), which in turn increases pollen removal (Galen & Stanton 1989; Harder 1990a; Young & Stanton 1990; Thøstesen & Olesen 1996; although see Mitchell & Waser 1992; Cresswell 1999) and deposition per flower (Galen & Stanton 1989; Thomson 1986; Thøstesen & Olesen 1996; although see Young & Stanton 1990; Mitchell & Waser 1992). Nectar availability also affects the number of flowers visited on inflorescences (e.g. Hodges 1985; Cibula & Zimmerman 1987; Hodges 1995), which will influence pollination at the plant level (e.g. Zimmerman 1983; Johnson, Peter & Ågren 2004). In general, individual pollinators visit more flowers on large displays (reviewed by Ohashi & Yahara 2001), which tends to increase pollen import (e.g. Ohara & Higashi 1994) and removal. However, pollinator movements within an inflorescence can also increase self-pollination among flowers (geitonogamy: reviewed by Harder & Barrett 1996; Snow et al. 1996; Harder et al. 2001; also see Karron et al. 2004), which can deplete the pollen available for export (pollen discounting: reviewed by Harder & Barrett 1996; Harder et al. 2001). This mating cost of having many flowers open simultaneously and the diminishing returns that accompany increased pollen removal per pollinator favour limited nectar production per flower and an intermediate display size (Harder & Thomson 1989; Harder & Wilson 1994; Iwasa, de Jong & Klinkhamer 1995). In general, the optimal nectar production and flower number for pollen export depend on pollinator abundance (Harder & Wilson 1994; Harder & Barrett 1996): greater nectar production and larger displays are favoured when pollinators visit infrequently, because they guard against pollen remaining undispersed in anthers (Harder et al. 2001).

In this study, we examine six species of legumes (Fabaceae: Papilionoidae) to assess the effects of floral and inflorescence traits on pollinator activity, pollinator behaviour on inflorescences, pollen removal and seed production. All six species are visited predominantly by nectar-feeding bumble-bees, and have similar pea-like flowers and racemose inflorescences. In addition to describing relevant floral and inflorescence characteristics, including self-incompatibility and flowering phenology, we compare several aspects of reproductive performance by these species. To determine the opportunity for nectar availability and flower number to influence pollination outcomes, we consider the relations of pollinator attraction and the number of flowers visited per pollinator to interspecific differences in nectar production and display size. We also assess the relation of pollen removal to nectar production and the number of open flowers. In particular, we focus on whether removal varies negatively with display size among species, as is expected if pollen-dispensing mechanisms restrict pollen removal from individual flowers more in species with large displays as a mechanism of reducing geitonogamy and pollen discounting. Finally, we consider whether fruit production and seed production per fruit vary with nectar availability and the number of open flowers in a similar manner to pollen removal.

Materials and methods

data collection

This study was conducted during 2001 at montane and subalpine sites in the Kananaskis Valley, 60 km west of Calgary, Alberta, Canada. Four legume species, Hedysarum boreale Nutt. var. mackenzii (Richards.) C. L. Hichc., Hedysarum sulphurescens Rhdb., Oxytropis sericea Nutt. and Astragalus striatus Nutt. were studied at an open, montane grassland near Barrier Lake (51°02′ N, 115°02′ W, 1480 m elevation). Astragalus americanus (Hook.) M. E. Jones was studied at a forest edge (at 1800 m elevation), and Oxytropis splendens Dougl. ex Hook. at an open meadow (at 1880 m elevation) in the subalpine near Ribbon Creek (50°56′ N, 115°09′ W).

We measured reproductive phenology and natural seed production for three inflorescences on each of 30 plants of H. boreale (five inflorescences), H. sulphurescens and O. sericea, and 25 plants of A. striatus, A. americanus and O. splendens, which were selected randomly within a c. 20 × 20-m2 area at each site at the beginning of the flowering season and marked with numbered tags. The numbers of floral buds, open flowers (i.e. without wilted petals) and developing fruits of each inflorescence were recorded at 1–3 day intervals throughout the flowering and fruiting periods. Once fruit matured, they were harvested, and the number of mature seeds was counted for each fruit.

In addition to measuring floral phenology, we estimated average floral longevity (L) for each species based on the average number of flowers produced per inflorescence (N), the average duration of flowering by an inflorescence (T) and the average display size (D). If flowers open at a fixed rate, r, then the duration of flowering by an inflorescence equals the time required for all flowers to open (N/r) plus the longevity of the last flower to open, so that T = N/r + L. We did not measure the rate of flower opening, but it can be estimated from the average display size. In particular, if flowers open at a fixed rate, then display size will equilibrate at D = rL during peak flowering, so that r = D/L. Therefore, floral longevity can be estimated by L =T/[1 + (N/D)].

Self-compatibility and the ability of flowers to set seed autonomously were assessed for 20 plants of each species (except 10 plants for O. splendens). Two inflorescences on each plant were bagged with fine mesh before flowering. Flowers that opened subsequently on one inflorescence were hand self-pollinated, whereas the other inflorescence was not disturbed. Mature fruits were harvested and seeds were counted for each fruit.

During flowering we measured aspects of floral morphology, ovule number and nectar production. Floral-tube length and dry mass (after 48 h at 60 °C) were measured for one flower sampled from each of 42–50 randomly selected plants of each species. At the same time, we counted ovules from three to five flowers on 20 randomly sampled inflorescences. To measure floral nectar we caged 25 (O. sericea) or 20 (other species) randomly selected plants on sunny, calm days between 700 and 800 h before bumble-bees became active to exclude visits. During the subsequent hour we measured nectar volume (V: µl) and sugar concentration (C: mass of solute/mass of solution) for three (H. boreale), four (A. striatus) or five (other species) flowers with calibrated capillary tubes and a Bellingham and Stanley refractometer (Tunbridge Wells, UK). Therefore, our measure of nectar availability estimates the reward that pollinators encountered during a flower's first visit of the day. We calculated the sugar content of nectar (mg) as ρCV, where ρ is the density of a sucrose solution with concentration C (CRC 2000).

We consider two measures of pollen removal: removal during a single bumble-bee visit, and total removal during the first 24 h after anthesis. For both measures, pollen removal was calculated from pollen counts for visited or exposed flowers (Pv) and unvisited flowers (or flower buds, P0) within the same inflorescence as (P0 – Pv)/P0. Pollen removal per visit was measured for inflorescences on which we removed previously open flowers and then caged the inflorescence to exclude pollinator visits until new flowers opened. On calm days, inflorescences with more than three open flowers were exposed to bumble-bee visits. After a bee visited an inflorescence, we removed unvisited flowers (or fully developed flower buds) and flowers that received a single visit and stored them separately in vials containing 70% ethanol. Fourteen inflorescences were replicated for each species (seven inflorescences for O. splendens). Pollen removal during 24 h was measured on consecutive sunny, calm days during peak flowering. During early morning, we recorded the positions of three randomly selected flowers on 20 previously bagged inflorescences. Twenty-four hours later, we stored the anthers and any pollen washed from the keel petals of these flowers in vials containing 70% ethanol. Anthers from the three flowers from an inflorescence were stored together, because too few pollen grains remained in individual flowers to allow electronic counting. At the same time, we also collected fully developed flower buds from the selected inflorescences for measurement of pollen production. Pollen was counted electronically with an Elzone 5380 particle analyser (Micromeritics, Norcross, GA) after the vials containing pollen had been sonicated for 15 min to dislodge pollen from the anthers (see Harder 1990a for details).

As a measure of pollinator activity, we considered the ratio of pollen removal during 24 h to pollen removal per visit. If pollinators remove roughly the same number of pollen grains per visit, at least for the first few visits, then this ratio represents the mean number of pollen-removing visits during a flower's first day.

We recorded the number of flowers visited per inflorescence on three to ten consecutive inflorescences by 31–49 bumble-bees for each plant species (20 bees for O. splendens). Bumble-bee visitation was observed during 2 or 3 calm days (mainly from 10.00 to 15.00 hours) during peak flowering for each species.

statistical methods

Our comparison of the reproductive ecology of these six legume species involved two perspectives: species comparisons with individual plants as the unit of replication, and tests of association between species means. Most species comparisons involved single-factor anova (Neter et al. 1996), with species as a fixed factor (SAS release 8·2, proc glm; SAS Institute Inc. 2001). For variables that were measured from several flowers per plant, or for several inflorescences per bee, we used the plant or bee mean as an individual observation. The analysis of total seed production by plants from which pollinators had been excluded involved a two-factor, repeated-measures anova (SAS, proc mixed), with species as a fixed, between-plant factor and pollination treatment (autonomous vs hand self-pollination) as a fixed, within-plant factor. This analysis used restricted maximum likelihood (Jennrich & Schluchter 1986) to characterize the covariance between responses to different pollination treatments by individual plants. A model of heterogeneous compound symmetry was more appropriate than one of independent responses (G3 = 20·31, P < 0·001). Denominator degrees of freedom for F-tests of this analysis were calculated by Kenward & Roger's (1997) technique. We used Tukey's multiple comparisons to distinguish species and/or treatment means that differed significantly (α = 0·05: Neter et al. 1996).

We analysed associations between species means with either Spearman's correlation or linear regression (Neter et al. 1996). Correlation analysis was used for floral traits for which there was no obvious causal relation between variables. Regression analysis was used to analyse response variables (nectar-sugar production, pollinator visitation, pollen removal, fruit and seed set) and employed a stepwise procedure to select independent variables that explained significant proportions of the variation in the dependent variable. Variables were suitably transformed to assure a linear relation between dependent and independent variables. Only the analysis of the mean ratio of pollen removal during 24 h to removal during a flower's first visit (A) found more than one independent variable to influence the dependent variable significantly (see below). To illustrate the independent influences of both variables, we present adjusted values of the removal ratio. Adjustment involved adding the residual (Ai − Âi) for each observation to the ratio predicted by the regression equation for the observed value of one independent variable and the overall mean of the other independent variable.


flowering phenology, floral and inflorescence traits

During 2001, the six legume species flowered from late May until late July. Hedysarum boreale flowered first, from late May until late June. Flowering by O. sericea overlapped with that of H. boreale, although it began flowering later (early June) than H. boreale. H. sulphurescens bloomed from mid-June to early July, and A. striatus bloomed from mid- to late July at the montane site. At the subalpine site, O. splendens bloomed from early until late July, and A. americanus from mid- until late July. The flowering duration of individual inflorescences differed significantly among species (F5,158 = 27·1, P < 0·001), largely because of differences between the genera, as congeneric species had equivalent flowering durations (Table 1a), with Astragalus inflorescences flowering for 7–8 days, Hedysarum for 13–14 days and Oxytropis for 9–10 days. Flowers of all species lasted about two days, except for O. sericea for which flowers lasted about 4 days (Table 1a). In general, floral longevity varied positively among species with average flower mass (Spearman's correlation, rS = 0·829, P < 0·05).

Table 1.  Comparisons of mean (± SE) inflorescence traits, floral traits and pollen removal by bumble-bees among six legume species. Sample size in parentheses. Species with different superscript letters for a given variable differ significantly, based on Tukey's multiple comparisons (P < 0·05). Tukey's comparisons for inflorescence traits are based on ln-transformed data, whereas those for nectar sugar are beased on square-root transformed data
SpeciesAstragalus americanusAstragalus striatusHedysarum borealeHedysarum sulphurescensOxytropis sericeaOxytropis splendens
  • *

    Number of open flowers/total flower number.

  • Estimated from flower production, floral display size and inflorescence flowering period, as described in the text.

  • Pollen removal during the first 24 h/removal during first visit: a measure of pollination activity.

(a) Inflorescence traits
Flowering period (days)7·7 ± 0·3a (25)6·8 ± 0·4a (25)13·1 ± 0·6c (30)13·7 ± 0·6c (30)8·8 ± 0·4a,b(30)10·4 ± 0·8b (25)
Total number of flowers16·0 ± 0·8b (25)15·6 ± 1·1b (25)13·4 ± 0·43a,b (30)49·9 ± 2·4d (30)11·3 ± 0·5a (30)25·7 ± 2·0c (25)
Number of open flowers5·8 ± 0·3b,c(25)5·0 ± 0·2b (25)3·4 ± 0·1a (30)9·0 ± 0·3d (30)9·4 ± 0·4d (30)7·0 ± 0·5c (25)
Flowering overlap*0·36 ± 0·02c (25)0·32 ± 0·02b,c (25)0·26 ± 0·01b (30)0·19 ± 0·01a (30)0·84 ± 0·02d (30)0·29 ± 0·02b (25)
Floral longevity (days)2·031·672·662·174·022·32
Number of flowers visited by individual bumble-bees2·6 ± 0·14b,c (34)2·2 ± 0·11b (34)1·6 ± 0·06a (39)2·7 ± 0·14b,c (45)2·6 ± 0·12b,c (48)3·0 ± 0·18c (20)
(b) Floral traits
Dry flower mass (mg)5·1 ± 0·15b (50)3·6 ± 0·08a (49)10·5 ± 0·25d (42)7·7 ± 0·18c (43)12·6 ± 0·33e (42)4·7 ± 0·10b (50)
Floral-tube length (mm)6·3 ± 0·07b (50)6·2 ± 0·11b (44)6·0 ± 0·14b (42)5·2 ± 0·10a (46)10·1 ± 0·12c (42)6·4 ± 0·10b (50)
Ovule number6·8 ± 0·21b (20)11·6 ± 0·29c (20)6·4 ± 0·16b (20)3·1 ± 0·08a (20)29·1 ± 0·41e (20)18·8 ± 0·54d (20)
Pollen production14330 ± 727a (33)15350 ± 682a (34)67370 ± 4710b (34)50740 ± 2296b (34)18980 ± 1061a (33)18340 ± 936a (33)
Pollen : ovule ratio2199 ± 206a (20)1404 ± 87a (20)10198 ± 1224b (20)16579 ± 1121c (20)585 ± 50a (20)1045 ± 92a (20)
Nectar sugar per flower (mg)0·11 ± 0·020b (20)0·03 ± 0·005a (20)0·13 ± 0·013b,c (20)0·08 ± 0·013b (20)0·17 ± 0·013c (25)0·07 ± 0·013a,b (20)
(c) Pollen removal
Removal during first visit0·58 ± 0·03a,b (13)0·77 ± 0·03c,d (15)0·80 ± 0·04d (14)0·63 ± 0·03b,c (14)0·53 ± 0·05a,b (14)0·40 ± 0·05a (7)
Removal during first 24 h0·66 ± 0·05a,b,c (20)0·54 ± 0·06a,b (20)0·83 ± 0·06c (20)0·82 ± 0·04c (20)0·74 ± 0·06b,c (18)0·46 ± 0·06a (18)
Pollen removal ratio1·150·671·051·301·401·15

Floral traits differed considerably among the six species that we studied (Table 1b: P < 0·001 for all traits considered). Dry flower mass varied almost fourfold from 3·6 mg (A. striatus) to 12·6 mg (O. sericea: Table 1b). All species differed in flower mass, except A. americanus and O. splendens. In contrast to flower mass, floral-tube length varied only twofold among species (Table 1b). Significant differences in tube length resulted because H. sulphurescens (5·2 mm) and O. sericea (10·1 mm) had shorter and longer tubes, respectively, than all other species. Mean flower mass did not vary significantly with mean tube length among species (rS = 0·086, P > 0·1), because of allometric differences in flower structure among genera. In particular, flowers of the Hedysarum species have relatively short tubes, but very large keels compared with the Astragalus and Oxytropis species. The significant variation in average pollen production per flower among species resulted solely because flowers of Hedysarum species produced three times more pollen than Astragalus or Oxytropis flowers (Table 1b). Ovule number per flower varied by almost an order of magnitude from only 3·1 ovules in H. sulphurescens to 29·1 ovules in O. sericea (Table 1b). Only A. americanus and H. boreale had equivalent ovule production per flower. Because Hedysarum flowers produced many pollen grains, but few ovules, Hedysarum species had pollen : ovule ratios more than five times greater than Astragalus or Oxytropis species (Table 1b).

Average nectar production per flower (S) varied fivefold among species (Table 1b: F5,119 = 13·9, P < 0·001). These interspecific differences associate significantly with both flower size (F: ln Ŝ = 0·977 ln[F] − 4·268, R2 = 0·740, F1,4 = 11·4, P < 0·05, based on species means) and floral longevity (L: ln Ŝ = −0·589 − 4·200/L, R2 = 0·787, F1,4 = 14·7, P < 0·025), both of which are correlated with each other. A. striatus produced less sugar per flower than all other species, except O. splendens. In contrast, O. sericea produced more sugar than all other species, except H. boreale. The four species between these extremes produced equivalent amounts of nectar sugar (Table 1b). Nectar presentation per inflorescence, estimated by the product of mean sugar production per flower and the mean number of open flowers, was 0·17 mg for A. striatus, 0·45 mg for H. boreale, 0·52 mg for O. splendens, 0·62 mg for A. americanus, 0·76 mg for H. sulphurescens and 1·59 mg for O. sericea.

The pattern of flower production and display varied considerably among the six legume species (Table 1a: P < 0·001 for all traits considered). Average total flower production per inflorescence ranged from 11·3 in O. sericea to 49·9 in H. sulphurescens (Table 1a). Flower production differed significantly between the Hedysarum species and between the Oxytropis species, but not between the Astragalus species. The number of open flowers per inflorescence varied less between species than total flower number, with three groups of species: H. boreale had small displays (3·4 flowers), H. sulphurescens and O. sericea had the largest displays (≥9 flowers), with O. splendens and the two Astragalus species intermediate (Table 1a). The different rankings of flower production and open flower number resulted from contrasting flowering patterns within inflorescences (Table 1a). O. sericea showed essentially simultaneous flowering (84% overlap among flowers within inflorescences), whereas H. sulphurescens showed sequential flowering (19% overlap). Flowering overlap within inflorescences for other species ranged from 26 to 36%, so that the simultaneous flowering of O. sericea was unusual.

pollinator visitation

All six legume species were visited exclusively by bumble-bees (Bombus spp.). Overwintered queen bees visited the two early flowering species, H. boreale and O. sericea, with most visits by Bombus flavifrons Cresson (65% and 76%, respectively), which has a moderate tongue length. Having a shorter corolla tube, H. boreale also attracted queens of shorter-tongued bee species, B. frigidus Smith (8%), B. bifarius Cresson (7%) and B. moderatus Cresson (7%), as well as some visits by the long-tongued B. californicus Smith (3%). In contrast, other visitors to the long-tubed O. sericea flowers included only long-tongued bees species, B. californicus (20%) and B. nevadensis Cresson (3%). H. sulphurescens, the species with the shortest floral tube, was visited by B. flavifrons (55%), B. bifarius (19%), B. californicus (10%) and B. frigidus (9%), of which approximately 40% were workers. Three late-flowering species, A. americanus, A. striatus and O. splendens, were visited mainly by B. flavifrons (61–74%) and B. bifarius (24–44%), with 45%, 91% and 70% of visits by workers, respectively.

The number of flowers visited by individual bees on inflorescences (V) differed significantly among legume species (F5,214 = 13·92, P < 0·001: Table 1a). These differences reflected a tendency of bees to visit more flowers on species with many open flowers (D: Fig. 1: ln  = 0·498 ln[D] − 0·039, R2 = 0·720, F1,4 = 10·3, P < 0·05, based on species means). However, this tendency was not straightforward, as the number of flowers visited increased in direct proportion (∼0·44) to display size among the four species with the smallest displays, whereas bees visited smaller proportions of open flowers on the two species with the largest floral displays, H. sulphurescens and O. sericea (Fig. 1).

Figure 1.

Relation of the mean (± SE) number of flowers visited by bumble-bees to the mean (± SE) number of open flowers per inflorescence for Astragalus americanus (Asam), A. striatus (Asst), Hedysarum boreale (Hebo), H. sulphurescens (Hesu), Oxytropis sericea (Oxse) and O. splendens (Oxsp). The dotted lines indicate different proportions of flowers visited. The solid line represents the best fit regression based on ln-transformed species means.

pollen removal

Depending on plant species, bees removed 40–80% of a flower's pollen production during their first visit (Fig. 2a), with significant differences between species (F5,71 = 11·63, P < 0·001: Table 1c). Bees removed the smallest proportion of pollen from the two Oxytropis species and the largest proportion from H. boreale. These interspecific differences were not related significantly to any measured floral or inflorescence traits (P > 0·05 in all cases). However, the mean proportion of pollen removed during a flower's first visit (P1) varied negatively among species with the mean number of flowers that bees visited per inflorescence (V: Fig. 2b: ln  = −0·026 − 0·030V3; R2 = 0·883, P < 0·01, based on species means). Most measurements of the pollen removal during first visits involved B. flavifrons queens (except for A. striatus), so that differences in pollen removal reflect intrinsic floral characteristics, rather than pollinator differences.

Figure 2.

Variation in pollen removal among Astragalus americanus (Asam), A. striatus (Asst), Hedysarum boreale (Hebo), H. sulphurescens (Hesu), Oxytropis sericea (Oxse) and O. splendens (Oxsp). (a) The relation of mean (± SE) pollen removal during 24 h to the mean (± SE) amount of pollen removed during first visits to flowers. Dotted lines indicate different ratios of 24-h to first-visit removal (pollen removal ratio), a measure of pollinator activity. (b) The relation of the average (± SE) proportion of a flower's pollen removed during its first visit to the average (± SE) number of flowers visited by bumble-bees. (c) and (d) The relations of the pollen removal ratio to mean sugar production per flower and mean floral display size, respectively. In (c), the pollen removal ratio has been adjusted as though all species had equal floral displays, whereas in (d) the ratio has been adjusted as though all species produced equivalent amounts of nectar (see Materials and methods for details).

During 24 h after anthesis, pollinators removed from 46% (O. splendens) to over 80% (both Hedysarum species) of a flower's pollen, with significant differences between species (F5,110 = 7·48, P < 0·001: Table 1c). These interspecific differences were not related significantly to any measured floral or inflorescence traits, although species with heavier flowers tended to experience higher 24-h removal than small-flowered species (F1,4 = 6·69, P = 0·06, R2 = 0·626). As the flowers of all species last roughly 2 days or more (Table 1a), the daily rate of pollen removal would been sufficient to delete flowers of all species of pollen before they wilted (except perhaps A. striatus). For all species, except A. striatus, pollen removal during 24 h generally equalled or exceeded first-visit removal, with H. sulphurescens and O. sericea having the highest pollen removal ratios (Fig. 2a). These results suggest that flowers on all species, except A. striatus, received at least one pollen-removing visit during their first day of flowering, on average.

The ratio of pollen removal during 24 h to first-visit removal (A), which we interpret as a measure of pollinator activity, varied positively with both a species’ average sugar production per flower (S: Fig. 2c) and its average display size (D: Fig. 2d:  = 1·216 + 0·310ln [S]+ 0·358 ln[D]; R2 = 0·960, P < 0·01, based on species means). Therefore, the low pollen-removal ratio for A. striatus is associated with its combination of low sugar production and small floral displays, whereas the high ratios for H. sulphurescens and O. sericea are associated with high sugar production and large floral displays (Fig. 2c,d). Because pollinator activity varied positively among species with both mean sugar production per flower and mean display size, it also varied positively with mean sugar production per plant ( = 1·314 + 0·327 ln[SD]; R2 = 0·957).

self-incompatibility and seed set

All six legume species exhibited strong self-incompatibility. A two-factor, repeated-measures anova found a significant difference in the overall proportion of ovules setting seed between flowers that could only self-pollinate autonomously and hand self-pollinated flowers on individual plants (F1,104 = 18·88, P < 0·001). On average (± SE), 0·1 ± 0·05% of the ovules set seeds in flowers from which pollinators were excluded when no hand pollinations were done (Table 2), indicating no capacity for autonomous self-pollination. Flowers that experienced only self-pollination had an average (± SE) seed set of 1·6 ± 0·35% (Table 2). All species exhibited at an equivalent difference in seed set between pollination treatments (species–treatment interaction, F5,104 = 2·09, P > 0·05) and overall seed set did not differ significantly among species (F5,104 = 1·79, P > 0·1). These results demonstrate that all six species are overwhelmingly obligate outcrossers.

Table 2.  Comparisons of self-incompatibility, and seed and fruit production under natural conditions among six legume species. Mean ± SE seed : ovule ratio for flowers subjected to pollinator exclusion and exclusion plus self-pollination, and fruit : flower ratio and seed : ovule ratio per fruit for open pollination inflorescences. Sample size in parentheses. Species with different superscript letters for a given variable differ significantly, based on Tukey's multiple comparisons (P < 0·05)
SpeciesSeed set per flowerOpen-pollinated inflorescences
Bagged, but not hand-pollinatedBagged and self-pollinatedFruit setSeed set per fruit
Astragalus americanus 0·000 ± 0·000a (20)0·009 ± 0·004a (20)0·58 ± 0·048d (24) 0·45 ± 0·031c (24)
Astragalus striatus 0·003 ± 0·002a (20)0·008 ± 0·004a (20)0·16 ± 0·033a (25)0·047 ± 0·011a (25)
Hedysarum boreale 0·003 ± 0·002a (20)0·015 ± 0·007a (20)0·34 ± 0·024b,c (30) 0·37 ± 0·020c (30)
Hedysarum sulphurescens 0·000 ± 0·000a (20)0·005 ± 0·003a (20)0·46 ± 0·029c,d (30) 0·25 ± 0·008b (30)
Oxytropis sericea0·0003 ± 0·0003a (20)0·034 ± 0·015a (20)0·35 ± 0·043b,c (29) 0·22 ± 0·033b (29)
Oxytropis splendens0·0005 ± 0·0005a (10)0·027 ± 0·011a (10)0·23 ± 0·032a,b (22) 0·11 ± 0·010a (22)

Fruit set under natural pollination varied significantly among species (F5,154 = 17·09, P < 0·001), ranging from 16 to 58% (Table 2). Mean fruit set did not vary significantly among species with respect to any measured floral or inflorescence trait, or any measure of pollen removal (P > 0·1 in all cases).

Percentage seed set per fruit for naturally pollinated inflorescences ranged from 5 to 45% among species (Table 2), with significant differences between species (F5,154 = 45·16, P < 0·001). Mean seed set per fruit (Sd) varied positively with mean fruit set (Fr) among species (ln Ŝ = 0·861Fr − 0·064, R2 = 0·734, F1,4 = 11·0, P < 0·05, based on species means). However, mean seed set per fruit did not vary significantly among species with any measured floral or inflorescence trait, or any measure of pollen removal (P > 0·1 in all cases).


We detected several interspecific associations of reproductive performance with floral and inflorescence traits. Pollinator visitation, as measured by the ratio of pollen removal during 24 h to first-visit removal, varied positively among species with both nectar production per flower (Fig. 2c) and floral display size (Fig. 2d). In general, bumble-bees visited more flowers per inflorescence for species that displayed more flowers, with no increase in the proportion of flowers visited (Fig. 1). The proportion of available pollen that bees removed during a flower's first visit varied negatively among species with the number of flowers that bees visited per inflorescence (Fig. 2b). In contrast, interspecific differences in 24-h pollen removal, fruit set and seed set per fruit were not associated significantly with floral or inflorescence traits. Interspecific variation in reproductive performance did not parallel phylogenetic relatedness, as performance was typically less similar between congeneric species than between more distantly related species (Tables 1c and 2, Figs 1 and 2). Therefore, the interspecific differences in performance that we observed probably represent contrasting adaptations for enhancing reproduction, especially pollen export, and/or species-specific responses to prevailing ecological conditions such as pollinator abundance.

Two aspects of our results demonstrate that pollen dispersal depends collectively on floral and inflorescence characteristics. First, pollinator activity varied positively among species with both nectar production (Fig. 2c) and floral display size (Fig. 2d), as is expected if bees base their choice of plant species on both foraging benefits and costs. The positive relation of interspecific attractiveness to nectar production has been observed previously (Pleasants 1981; Bosch, Retana & Cerdá 1997) and probably reflects the energetic benefits that bumble-bees realize from feeding on more productive species. In contrast, the positive relation the pollinator activity experienced by plant species and their display size has not been reported previously, although it is consistent with the tendency of bees to visit individual plants with large displays more frequently than plants of the same species with smaller displays (reviewed by Ohashi & Yahara 2001). This role of display size in interspecific choice by pollinators probably arises because pollinators visit more flowers on large displays (Fig. 1: reviewed by Ohashi & Yahara 2001), thereby reducing energy expenditure per flower visit on expensive interplant flights. Given these influences of nectar production and display size on foraging economics, a species’ attractiveness is an aggregate characteristic of individual floral traits and overall flower presentation, and so cannot be understood fully from analysis of floral traits alone (also see Harder et al. 2004). Comparison of 24-h pollen removal (Table 1c) with floral longevity (Table 1a) indicates that all species, except perhaps Astragalus striatus, were sufficiently attractive that pollinators probably removed all of their pollen before their flowers wilted.

The second collective effect of floral and inflorescence functions exposed by our study involves the negative interspecific relation between the average proportion of pollen removed during first visits to flowers and the mean number of flowers that bees visited per inflorescence (Fig. 2b). No proximate explanation involving either pollinator behaviour or plant responses obviously links these outcomes. In particular, it is not clear how pollen removal during a flower's first visit could respond to the number of flowers that an individual bee visited, and it seems unlikely that nectar-collecting bees visit more flowers after they remove relatively little pollen from individual flowers. Instead, limited pollen removal from species on which individual pollinators visit more flowers per inflorescence suggests that the floral mechanisms governing pollen removal per visit may be adjusted evolutionarily to enhance pollen export.

Restricted pollen removal per pollinator visit counteracts a positive relation between the number of grains removed from a flower by an individual pollinator and the proportion of those grains that are lost before reaching a stigma on another plant (Harder & Thomson 1989). For bee-pollinated plants, such a relation probably arises from two processes: geitonogamy (Klinkhamer & de Jong 1993; Harder & Barrett 1996) and bee grooming (Harder & Wilson 1997). Pollen that is deposited geitonogamously cannot be exported to other plants (pollen discounting) and is lost completely from mating for self-incompatible species (Harder & Barrett 1996), such as the legumes that we studied. Because geitonogamy increases with the number of flowers that each pollinator visits per inflorescence (reviewed by Harder et al. 2001; also see Karron et al. 2004), especially when nectar is abundant (Johnson et al. 2004), pollen removal per pollinator should be restricted more for species on which pollinators visit more flowers. In addition, limited pollen removal counteracts the tendency of bees to groom more frequently and more intensively in flight after they remove copious pollen from a plant (Harder 1990b), which displaces pollen from sites on bees’ bodies where it is accessible to stigmas. This tendency is especially likely to vary with the number of flowers that a bee visits per inflorescence for species, such as most of the legumes that we studied (except Astragalus americanus and Hedysarum sulphurescens), on which bees walk between flowers within inflorescences, accumulating pollen with each flower visit without intervening grooming. Hence, adaptive restriction of pollen removal in response to positive relations of both pollen discounting and grooming losses to the number of flowers that pollinators visit could create the interspecific variation in first-visit pollen removal that we observed (Fig. 2b).

The floral mechanisms responsible for the differing restriction of first-visit pollen removal among species are not obvious. Even though all species have papilionaceous flowers, details of their flower structure differ (e.g. Table 1b), perhaps in undetected ways that affect pollen removal. First-visit removal did not vary with nectar production (P > 0·9), which suggests that the effect of nectar availability on visit duration (Harder 1986; Thomson 1986; Cresswell 1999; Kudo 2003) did not influence interspecific differences in pollen removal. This result is not surprising, because nectar production did not vary significantly with corolla-tube length among species (P > 0·5), so that the effects of flower depth on visit duration (Harder 1986) probably counteracted the effects of nectar availability. Whatever the mechanisms responsible for restricting pollen removal, they may respond dynamically to the specific details of pollinator behaviour, as first-visit removal varied significantly with the mean number of flowers that bees visited (Fig. 2b), rather than with floral display size (P > 0·1). Dynamic adjustment of pollen removal to prevailing pollination conditions has been demonstrated for a Lupinus species (Fabaceae: Harder & Wilson 1994), which also has papilionaceous flowers, but does not produce nectar.

Our results indicate that aspects of the pollination process differ in their dependence on floral and inflorescence traits: pollinator activity and first-visit pollen removal varied with measured plant traits, whereas 24-h pollen removal and fruit and seed set per fruit did not. Three factors might explain this difference. First, floral and inflorescence traits may not influence some aspects of reproductive performance. For example, if postpollination processes govern ovule fertilization and/or embryo development, then any influences of plant traits on pollen import will not have corresponding effects on female fecundity (also see Goulson et al. 1998). Whether this explanation applies to fruit and seed set in our study is unclear, as no species realized more than 60% fruit set and seed set per fruit was generally low (<50%; Table 2), which may indicate pollen limitation of fertilization. However, low seed : ovule ratios in the range that we observed have also been attributed to embryo abortion in populations with considerable genetic load (Wiens 1984). The second factor that could cause apparently inconsistent effects of reproductive performance to floral and inflorescence traits is the limited statistical power associated with a comparison of just six species. Despite limited power, our sample of six species did reveal influences on pollinator activity and first-visit pollen removal, suggesting that these particular effects are stronger than any undetected effects on 24-h pollen removal or female fecundity. The final factor that may complicate detection of interspecific associations between floral and inflorescence traits and reproductive performance is the summation over different numbers of visits for each species in the accumulation of 24-h pollen removal and the pollen import responsible for female fecundity. Our observations of first-visit pollen removal (Fig. 2b) indicate that the species dispense different amounts of pollen per visit. In addition, pollen removal during 24 h seems to have involved different numbers of pollinators for the six species, based on the ratio of 24-h pollen removal to first-visit removal (Fig. 2a). Both components of 24-h removal were affected directly or indirectly by floral or inflorescence traits. That these effects were not evident for 24-h removal suggests that they may cancel each other out as species accumulate different numbers of visits. Such an interspecific outcome could arise even if floral and inflorescence traits contribute significantly to variation in pollen removal within individual species, which we did not examine.

Our sample of six species pollinated by bumble-bees illustrates that despite common floral and inflorescence design and similar pollinators, legume species differ significantly in their interaction with pollinators. This functional diversity results partly from contrasting combinations of floral and inflorescence traits within the basic morphological plan of papilionaceous legumes. This conclusion emphasizes that the interaction between plants and their pollinators, and its evolutionary diversification, depends on the collective contributions of floral and display traits to reproductive performance, rather than on the effects of either class of traits alone.


This study was funded by Grant-in-Aid for Scientific Research (no. 15370006) of Japan Society for the Promotion of Science (G.K.) and the Natural Sciences and Engineering Research Council of Canada (L.D.H.).