Transition from wind pollination to insect pollination in sedges: experimental evidence and functional traits


  • Peter D. Wragg,

    1. School of Biological and Conservation Sciences, University of KwaZulu-Natal, Private Bag X01, Scottsville 3209, South Africa
    2. Present address: Department of Ecology, Evolution, and Behavior, University of Minnesota, 100 Ecology Building, 1987 Upper Buford Circle, St. Paul, MN 55108, USA
    Search for more papers by this author
  • Steven D. Johnson

    1. School of Biological and Conservation Sciences, University of KwaZulu-Natal, Private Bag X01, Scottsville 3209, South Africa
    Search for more papers by this author

Author for correspondence:
Peter D. Wragg
Tel: +1 651 315 2693


  • Transitions from wind pollination to insect pollination were pivotal to the radiation of land plants, yet only a handful are known and the trait shifts required are poorly understood. We tested the hypothesis that a transition to insect pollination took place in the ancestrally wind-pollinated sedges (Cyperaceae) and that floral traits modified during this transition have functional significance.
  • We paired putatively insect-pollinated Cyperus obtusiflorus and Cyperus sphaerocephalus with related, co-flowering, co-occurring wind-pollinated species, and compared pairs in terms of pollination mode and functional roles of floral traits.
  • Experimentally excluding insects reduced seed set by 56–89% in putatively insect-pollinated species but not in intermingled wind-pollinated species. The pollen of putatively insect-pollinated species was less motile in a wind tunnel than that of wind-pollinated species. Bees, beetles and flies preferred inflorescences, and color-matched white or yellow models, of putatively insect-pollinated species over inflorescences, or color-matched brown models, of wind-pollinated species. Floral scents of putatively insect-pollinated species were chemically consistent with those of other insect-pollinated plants, and attracted pollinators; wind-pollinated species were unscented.
  • These results show that a transition from wind pollination to insect pollination occurred in sedges and shed new light on the function of traits involved in this important transition.


Transitions from wind pollination to animal pollination were milestones in the diversification of plants, yet floral trait evolution associated with such transitions remains poorly characterized. The fossil record has revealed glimpses of the evolution of inflorescence morphology and pollen traits associated with transitions to animal pollination in early plants (Hu et al., 2008), but other key floral traits, such as color and scent, cannot be determined from fossils. Extant basal flowering plants and cycads provide some insights into trait evolution associated with early transitions from wind pollination to animal pollination (Pellmyr, 1992; Endress & Doyle, 2009). However, functional and comparative studies of more recent transitions from wind pollination to animal pollination, within plant lineages that have reverted from animal pollination to wind pollination, can yield further insights into the roles of functional traits in this important evolutionary transition.

Animal- and wind-pollinated plants are characterized by distinct suites of floral traits that appear to be adaptive. Animal-pollinated flowers tend to have larger, more colorful perianths and produce fragrance and nectar that attract and reward pollinators, whereas wind-pollinated plants tend to have more flowers with fewer (often single) ovules and smoother, less sticky, and more consistently sized (17–58 μm) pollen that promotes pollen transport by wind (Friedman & Barrett, 2008, 2009b). Animal-pollinated plants tend to be cosexual, allowing each pollinator visit to both remove and deposit pollen, whereas wind-pollinated plants are often dioecious (Friedman & Barrett, 2008), which prevents selfing. However, not all of these trait shifts are required for a transition between pollination modes: some may instead occur after a transition, as an adaptation to enhance efficiency of a new pollination mode.

Transitions from animal pollination to wind pollination have occurred at least 65 times in the flowering plants (Linder, 1998) and phylogenetic analyses suggest that only a few traits – nonsticky pollen and small flowers (Linder, 1998), and perhaps unisexual flowers or plants (Friedman & Barrett, 2008) – are required for this transition. Transitions from wind pollination to animal pollination have been much rarer (Dodd et al., 1999), occurring in the families Caryophyllaceae (Weller et al., 1998) and Moraceae (Datwyler & Weiblen, 2004) and possibly Fagaceae (Manos et al., 2001), Salicaceae (Peeters & Totland, 1999), Joinvilleaceae and Flagellariaceae (Linder & Rudall, 2005). In the ancestrally and predominantly wind-pollinated sedge family Cyperaceae (Linder & Rudall, 2005), a number of transitions to insect pollination have been inferred from shifts in floral color and scent (Goetghebeur, 1998; Magalhães et al., 2005) and observations of insects carrying pollen (Stelleman, 1984; Thomas, 1984). Indeed, it appears that there may have been more transitions from wind pollination to insect pollination in the sedge family than in all other plants combined. However, there has been no experimental evidence that insects contribute substantially to pollination of any sedge lineage. Therefore, Friedman & Barrett (2008) did not include putative transitions from wind pollination to insect pollination in sedges, which lack nectar and often occur in open habitats, in their analysis that concluded that transitions from wind pollination to insect pollination are more likely in nectariferous lineages of closed habitats. Thus, sedges will be a key group for advancing our limited understanding of which trait shifts are required for this rare transition.

Anecdotal observations of insects visiting Cyperus obtusiflorus and Cyperus sphaerocephalus, sedges with showy bracts and slight floral scents (Fig. 1), led us to hypothesize that there has been a transition from wind pollination to exclusive insect pollination, associated with modifications of pollen and floral advertizing traits, in the clade containing these species. We tested the following predictions arising from this hypothesis: (1) excluding insects (but not wind) would reduce seed set in these putatively insect-pollinated species but not in wind-pollinated relatives; (2) pollen of these putatively insect-pollinated species would be less motile in the wind than pollen of wind-pollinated relatives; (3) inflorescences of these putatively insect-pollinated species would be convergent in color with those of other insect-pollinated species, and would be more apparent and attractive to pollinating insects than those of wind-pollinated relatives; and (4) inflorescences of these putatively insect-pollinated species would be convergent in emission rate and chemical composition of scent with those of other insect-pollinated species, and would attract pollinating insects, whereas inflorescences of wind-pollinated relatives would emit fewer compounds and be no more scented than their leaves.

Figure 1.

Mean (± SE) seed set in three sedge species. Exclusion of insect-borne but not wind-borne pollen reduced seed set in Cyperus obtusiflorus (b) and Cyperus sphaerocephalus (c) but not in Pycreus oakfortensis (a). Conversely, the additional exclusion of wind-borne pollen reduced seed set in P. oakfortensis (a) but not in C. obtusiflorus (b) or in C. sphaerocephalus (c). Sample sizes are (florets/spikelets/inflorescences). Insect visitors pictured are (a) leaf beetle, Monolepta cruciata, Chrysomelidae; (b) honeybee, Apis mellifera, Apidae; and (c) monkey beetle, Eriesthis fallax, Scarabaeidae (foreground). Bars, 5 mm.

Materials and Methods

Study species and sites

We studied five perennial sedge species, all herbaceous graminoids, at nine grassland localities in south-eastern Africa (see the Supporting Information Table S1). Cyperus obtusiflorus Vahl has white spikelet scales (hereafter ‘bracts’), whereas C. sphaerocephalus Vahl has yellow bracts. We sought to compare each species with its nearest wind-pollinated relative that was co-flowering in the same habitat patch, to ensure equal opportunities for wind pollination and insect pollination. Phylogenetic studies show that the genus Cyperus is part of a ‘Cyperus clade’ that contains several other genera including Pycreus (Muasya et al., 2009), but the group is not sampled intensively enough, or sufficiently resolved, to allow us to identify sister groups. Hence, we compared each putatively insect-pollinated species with the locally co-flowering species in the Cyperus clade most similar in inflorescence structure yet differing in having inconspicuous brown bracts, suggestive of wind pollination. Thus, in Vernon Crookes Nature Reserve we compared C. obtusiflorus with Pycreus oakfortensis C.B. Clarke; in Summerveld Conservancy we compared C. obtusiflorus with Cyperus tenax Boeck.; and in Kamberg Nature Reserve and Castleburn Resort we compared C. sphaerocephalus with Cyperus semitrifidus Schrad.

The study species produce inflorescences of digitately arranged spikelets (Fig. 1), slightly taller than the grass sward, in early summer. Each spikelet is composed of up to 20 hermaphrodite florets. Florets are protogynous, but stigma receptivity and anthesis overlap within inflorescences. Anthesis occurs in the morning. None of the species produce nectar.

Insect- and wind-exclusion experiments

To determine the contributions of insects and wind to pollen removal, pollen deposition, and seed set, we applied various levels of vector exclusion to inflorescences in bud: (1) open to insects and wind; (2) 1 mm mesh bag to exclude insects but allow passage of wind-borne pollen (Peeters & Totland, 1999); and (3) microfiber bag with pores smaller than sedge pollen to exclude both insects and wind-borne pollen. Thus, comparing (2) with (1) indicated the contribution of insect pollination, and comparing (3) with (2) indicated the contribution of wind pollination. Any pollen deposition or seed set in the microfiber bags would be caused by autonomous selfing or apomixis. Mesh bags reduced the deposition of pollen on silicone grease-coated microscope slides down-wind of wind-pollinated C. tenax by < 30% compared with open control slides, and microfiber bags completely excluded wind-borne pollen (P.D. Wragg, unpublished). To control for effects of the bags on seed set through mechanisms other than pollen exclusion (e.g. altered microclimate or abrasion), we applied the same three exclusion treatments to additional inflorescences that we cross-pollinated.

We applied 15 replicates of these six treatments to intermingled C. obtusiflorus and P. oakfortensis inflorescences in Vernon Crookes Nature Reserve in October 2007, and 16 replicates to C. sphaerocephalus in Kamberg Nature Reserve in December 2007. Both were randomized block experiments. We intended the blocks to account for habitat heterogeneity, but we could not make the blocks spatially discrete because of the erratic distribution of buds. Therefore, blocks were omitted from the analyses but still ensured that treatments were spatially interspersed. Animals destroyed some replicates.

We recorded seed set in six randomly selected spikelets per inflorescence. Seed set mostly reflected cross-pollination, at least in the putatively insect-pollinated species: experimental hand-pollinations revealed self-incompatibility in C. obtusiflorus (fluorescence microscopy showed pollen tubes reaching the ovary in 11/12 crossed florets vs 1/17 selfed florets), C. sphaerocephalus (5/14 crossed vs 0/23 selfed) and perhaps P. oakfortensis (2/2 crossed vs 0/2 selfed; all P.D. Wragg, unpublished). To model presence or absence of a seed in each floret as a function of exclusion treatment and the floret’s position on the spikelet (seed set decreased distally because florets mature from the base to the tip), we used a generalized linear mixed model with binomial error distribution and logit link function (Bolker et al., 2009). Coefficients for the open and full-exclusion (microfiber) treatments were a priori contrasts with the reference insect-exclusion (mesh) treatment. We refitted the model without a reference level to obtain corrected treatment means on the logit scale, and back-transformed these to yield the expected probability of a floret in the center of a spikelet setting seed (floret position was centered on its mean; plotted in Fig. 1). Random effects accounted for the hierarchical structure of florets nested within spikelets and spikelets nested within inflorescences. We fitted the model separately to each of the three species, first using inflorescences that were not hand-pollinated and then using hand-pollinated inflorescences.

We explicitly considered density in these analyses as it has the potential to affect seed set (Ghazoul, 2005). For all three species, there was no significant difference in distance to fourth-nearest neighbor (an index of density: Honig et al., 1992) between the plants allocated to the different exclusion treatments (not shown). Flowering plants of C. obtusiflorus tended to occur further apart (mean ± SD of distance to fourth-nearest neighbor 140 ± 86 cm) than those of P. oakfortensis (58 ± 26 cm); C. sphaerocephalus was intermediate (75 ± 93 cm). To avoid confounding the difference in density with a difference in treatment effects between C. obtusiflorus and P. oakfortensis, we fitted an expanded model to the combined data for both species. To the single-species model above, we added species and its interactions with floret position and treatment; and interactions of distance to fourth-nearest neighbor with species, treatment, and species × treatment, all as fixed effects. We centered distance to fourth-nearest neighbor on the average of the species means for P. oakfortensis and C. obtusiflorus (99 cm) to evaluate the treatment contrasts at a common density.

To determine the contributions of insects and wind to pollen removal and deposition, we counted pollen remaining in stamens and pollen tubes reaching the ovary 1 d after applying the three exclusion treatments to additional inflorescences of C. obtusiflorus and P. oakfortensis (further details in Methods S1).

Pollen motility in a wind tunnel

We measured how effectively wind transports pollen of all five species using a laminar-flow wind tunnel. We counted the number of pollen grains captured on microscope slides coated in sticky fuchsin gel (Beattie, 1971) and placed downwind of inflorescences at anthesis. We used a randomized block design, with each of four blocks containing one run of each species at each of three wind speeds: 5 km h−1 (1.39 m s−1), 10 km h−1 (2.78 m s−1) and 20 km h−1 (5.56 m s−1). For each run, a different inflorescence was exposed to wind for 10 min and pollen was captured on slides placed 10, 30 and 90 cm downwind. Wind speeds, temperature (23°C) and relative humidity (50%) were typical of field conditions during anthesis (P.D. Wragg, pers. obs.).

We modeled the number of pollen grains per slide as a function of species, distance from the inflorescence and wind speed, using a generalized linear mixed model with Poisson error distribution and log link function. An inflorescence random effect accounted for nonindependence of the three slides placed downwind of each inflorescence. A slide random effect nested within the inflorescence effect accounted for overdispersion (Gelman & Hill, 2007, pp. 320–326). We evaluated the effect of species by comparing a full model with a reduced model omitting species using a likelihood ratio χ2 test. We accounted for variation in pollen available on each inflorescence by using the log of the number of available pollen grains, estimated as the number of freshly dehisced stamens multiplied by the mean number of pollen grains per stamen for that species, as an offset. Including available pollen grains as an offset is equivalent to including it as a covariate with its coefficient fixed at 1 (Gelman & Hill, 2007, p. 326); this is preferable to the alternative of dividing the response (pollen grains per slide) by the number of available pollen grains, which would have made a Poisson error distribution inappropriate. We contrasted all species pairs by refitting the model using each species in turn as the reference level.

Floral traits

Color  We characterized floral color by measuring reflectance spectra of bracts of each species over the waveband 300–700 nm using an S2000 reflectance spectrophotometer (Ocean Optics, Dunedin, FL, USA – details in Johnson & Andersson, 2002). To infer which sedge species can be visually distinguished from a green-leaf background by bees, we plotted the spectra as loci on a color hexagon (Chittka, 1992). In this perceptual color space, Euclidean distances between loci are proportional to the ability of bees to distinguish them.

Scent  We performed dynamic headspace scent extraction on inflorescences from up to four populations per species. We sampled single in situ inflorescences at anthesis, while taking control samples from vegetative material (inflorescence buds including subtending leafy bracts) and ambient air. For each sample, we pumped air from an enclosing polyacetate bag through a cartridge containing 1 mg each of Tenax® and CarbotrapTM adsorbents (both Supelco, Bellefonte, PA, USA) at a flow rate of 200 ml min−1 for 1 h. We thermally desorbed each sample and characterized it using GC-MS following the protocol of Shuttleworth & Johnson (2009). We estimated the proportional abundance of each compound using mass spectrometer ion counts and estimated total emission rate by combining total ion count with the average number of ions produced per unit mass of authentic standards of four dominant compounds.

We analysed population means of (1) number of compounds per floral sample, (2) emission rate per inflorescence and (3) emission rate per unit inflorescence dry mass using linear mixed models. Log-transforming emission rates improved normality of residuals. A species random effect accounted for nonindependence of populations of a species. For each of the three responses, we evaluated a contrast between the putatively insect-pollinated species and the wind-pollinated species using a one-tailed t-test. We conservatively allocated the species random effect one degree of freedom for each of the five levels, leaving the fewest possible degrees of freedom (three) for the t-tests (Bolker et al., 2009). We predicted that the floral scents of the two putatively insect-pollinated species would be distinct from each other (two clusters in multivariate space), and that the vegetative scents of these species would instead be more similar to the floral and vegetative scents of the wind-pollinated species (a third cluster). To determine whether clustering the samples into three groups with no a priori assignment of samples to groups would support this prediction, we performed k-means clustering on Bray–Curtis dissimilarities calculated from the presence or absence of compounds in samples. Using the same dissimilarities, we tested for differences in composition between groups of samples using permutational MANOVA (R function adonis: Oksanen et al., 2008).

Pollen  We estimated pollen production by estimating the number of pollen grains in 5–10 pre-dehiscent anthers per species (Protocol 5.1 in Dafni et al., 2005; Methods S1). We used environmental scanning electron micrographs of fresh pollen to measure the length of 10 pollen grains from at least three individuals for each species and characterize surface ornamentation. We examined fresh pollen of each species in a drop of water under a compound light microscope to detect pollenkitt lipids that make pollen sticky.

Insect visitation

We estimated the rate at which insects visited inflorescences by recording visits to groups of inflorescences during times of high insect activity (mostly 1–5 h after sunrise) at five study sites (Table S1). For these observations and the model inflorescence experiments below, we excluded the handful of visits by relatively immobile, nonpollinator insects (Orthoptera and Hemiptera).

Pollinator attraction to floral traits

We conducted Y-maze choice experiments, using methods described by Shuttleworth & Johnson (2009), to determine whether two common insect visitors – solitary bees Allodape sp. (Apidae) and monkey beetles Eriesthis hypocrita (Scarabaeidae) – were attracted to the floral scent of putatively insect-pollinated C. obtusiflorus in the absence of visual cues. For each insect species, we used the number of choices for and against the floral scent by each individual as responses in a generalized linear model with binomial error distribution and logit link function. We tested for a preference for the floral scent by evaluating whether the mean proportion of choices in favor of the floral scent (the ‘intercept’ fixed effect) exceeded 0.5 (corresponding to 0 on the logit scale) using a one-tailed z-test. The beetle model was overdispersed, so we adjusted the standard error of the intercept using a quasi-binomial model (dispersion parameter 1.59).

To test the hypothesis that differences in floral color and scent play a functional role in the greater attractiveness to insects of putatively insect-pollinated C. obtusiflorus than wind-pollinated P. oakfortensis, we measured rates of insect approaches to model inflorescences with factorial color and scent treatments in Vernon Crookes Nature Reserve. We painted spherical, size-matched model inflorescences to match the reflectance spectra of bracts of C. obtusiflorus (white) and P. oakfortensis (brown) (Fig. S4). We approximated the floral scent of C. obtusiflorus using one of the compounds most abundant in its floral scent ((Z)-ocimene, 9% of floral scent, in session 1) or a blend of four of the compounds most abundant in its floral scent ((Z)-ocimene, linalool, benzyl alcohol and 2-phenylethanol, collectively 38% of floral scent, in session 2). We soaked a rubber disk (septum) on each model inflorescence in a dilution of these compounds in hexane that yielded emission rates similar to natural inflorescences. Hexane may affect insect behavior (Roy & Raguso, 1997), so we soaked septa on other model inflorescences in hexane as controls.

Our basic experiment consisted of six model inflorescences, crossing color (white or brown) with scent (none, floral compounds diluted in hexane, or hexane control), plus a cut inflorescence of each of C. obtusiflorus and P. oakfortensis as positive controls. We placed these eight items in random order on the circumference of a circle 1 m in diameter, among intermingled flowering C. obtusiflorus and P. oakfortensis. On each of two consecutive days, six observers each recorded insect approaches to a cluster of four adjacent circles for the 3 h of peak insect activity (384 inflorescences in 48 circles in 12 clusters over 2 d). This constituted one session; there were two sessions 1 week apart, which were identical except for the scent mix (see earlier).

We analysed the effects of color, scent and their interaction on insect approaches to the arrays separately for each session, using a generalized linear mixed model with Poisson error distribution and log link function. The response was the total number of insect approaches to a model inflorescence on a day, because the different insect orders responded similarly to the treatments (not shown). A cluster random effect accounted for observer effects and habitat heterogeneity, and a day fixed effect accounted for differences in insect activity between days within a session.

We tested the prediction that color contributed to greater attractiveness of putatively insect-pollinated C. sphaerocephalus (yellow) than wind-pollinated C. semitrifidus (brown) by counting the number of insects trapped over two combined days on model inflorescences painted to match the reflectance spectra of each species (Fig. S4). We placed 10 pairs of yellow and brown model inflorescences, covered in Tangle-Trap Insect Trap Coating (The Tanglefoot Company, Grand Rapids, MI, USA), at intervals of 10 m along each of four traplines at the Kamberg Nature Reserve. We analysed the effect of color on the number of insects captured using a separate generalized linear mixed model for each insect order, with Poisson error distribution and log link function. Two random effects (model inflorescence pair, nested within trapline) accounted for habitat heterogeneity.

Statistical modeling

We fitted all non-Gaussian generalized linear mixed models by the Laplace approximation using the R function glmer (Bates & Maechler, 2009). We checked for overdispersion by comparing the residual deviance and the sum of squared Pearson residuals to a χ2 distribution with the residual degrees of freedom, assuming each random effect to use one degree of freedom (Bolker et al., 2009). All z, t and χ2 tests were Wald tests because our random effects had few levels, which would have made likelihood ratio tests unreliable; the exception was the likelihood ratio test for species differences in pollen motility, which was appropriate given the larger number of groups in the random effects for that analysis (Bolker et al., 2009).


Insect- and wind-exclusion experiments

Exclusion of insects from wind-pollinated P. oakfortensis did not reduce seed set significantly (56% of florets set seed without insects vs 61% with insects: = 0.644, = 0.519) but additional exclusion of wind reduced seed set enormously (seed set 1%, = 13.133, < 0.001, Fig. 1a). By contrast, exclusion of insects from putatively insect-pollinated C. obtusiflorus reduced seed set significantly (= 4.400, < 0.001) from 24.4% to 2.7% but the additional exclusion of wind did not reduce seed set significantly further (seed set 0.9%, = 1.699, = 0.089, Fig. 1b). Putatively insect-pollinated C. sphaerocephalus showed the same pattern as C. obtusiflorus: exclusion of insects significantly reduced seed set from 41% to 18% (= 3.149, = 0.002) but the additional exclusion of wind did not reduce seed set significantly further (seed set 14%, = 0.648, = 0.517, Fig. 1c). Seed set in hand-pollinated florets was uniformly high (> 75%) in all species and was unaffected by exclusion treatment (> 0.2 for all treatment effects).

An expanded, two-species model indicated that the difference in treatment effects on seed set between intermingled C. obtusiflorus and P. oakfortensis was not caused by the difference in density between them. Dropping the species × treatment × distance to fourth-nearest neighbor interaction did not significantly reduce model fit (χ2 = 0.13, df = 2, = 0.936). However, subsequently dropping the species × treatment interaction significantly reduced model fit (χ2 = 14.98, df = 2, < 0.001), indicating that the species responded differently to the treatments even when evaluated at a common density. The contrasts in this two-species model accounting for density yielded the same treatment inferences as the single-species, density-free models reported earlier.

Exclusion of insects from wind-pollinated P. oakfortensis inflorescences did not significantly reduce pollen removal (mean ± SE, 93 ± 4%) or deposition (pollen tubes reached 56 ± 12% of ovaries) compared with open inflorescences (removal 90 ± 7%, = 1.11, = 0.267; deposition 33 ± 21%, = 0.079, = 0.941), but the additional exclusion of wind substantially reduced pollen removal (66 ± 10%, = 3.14, = 0.002) and deposition (13 ± 13%), although the latter was not significant (= 1.931, = 0.111) (Fig. S2a,d). By contrast, exclusion of insects from putatively insect-pollinated C. obtusiflorus significantly lowered pollen removal (43 ± 6%) and deposition (0%) compared with open inflorescences (removal 87 ± 4%, = 5.70, < 0.001; deposition 57 ± 21%, Mann–Whitney = 99, < 0.001) whereas additional exclusion of wind did not lead to significant further reductions in pollen removal (57 ± 6%, = 0.81, = 0.418) or deposition (0%; Fig. S2c,e).

Pollen motility in a wind tunnel

Species differed significantly in pollen motility (χ2 = 79.4, df = 4, < 0.001): winds from 5 km h−1 to 20 km h−1 transported 84–100% fewer pollen grains of the putatively insect-pollinated species than of the three wind-pollinated species (Fig. 2). All pairwise comparisons of pollen motility between putatively insect-pollinated species and wind-pollinated species were highly significant (< 0.001), relative to a significance level of = 0.005 derived by adjusting the conventional significance level of = 0.05 for multiple comparisons using the Dunn–Sidak procedure (Quinn & Keough, 2002). Species within these two groups did not differ significantly: > 0.05 for all pairwise comparisons among putatively insect-pollinated species and among wind-pollinated species.

Figure 2.

Mean (± SE) pollen deposition on sticky slides placed downwind of an inflorescence in a wind tunnel at 10 km h−1 was much lower for insect-pollinated Cyperus obtusiflorus and Cyperus sphaerocephalus than for three related wind-pollinated sedge species. Results were similar at wind speeds of 5 km h−1 and 20 km h−1.

Floral traits

Color  The reflectance spectra of bracts of the putatively insect-pollinated C. obtusiflorus and C. sphaerocephalus are typical of insect-pollinated bee blue-green and bee green flowers, respectively (Figs 3, S3; Chittka et al., 1994). By contrast, bracts of the wind-pollinated species are achromatic and indistinguishable from the green-leaf background in terms of bee color vision (Fig. 3).

Figure 3.

Bract colors of three individuals from each of three populations of insect-pollinated Cyperus sphaerocephalus, two populations of insect-pollinated Cyperus obtusiflorus, and single populations of three related wind-pollinated sedge species, as loci in the color hexagon model of bee vision (Chittka, 1992). Loci in the central circle, including the green-leaf background and the wind-pollinated sedges, appear achromatic to bees (Chittka et al., 1994). The green-leaf background is represented by leaves of four common grass and forb species of South African grasslands, each species’ locus being a mean of three spectra from Shuttleworth & Johnson (2009).

Scent  The putatively insect-pollinated species emitted significantly more floral volatile compounds (population means ≥ 28.3) than did the wind-pollinated species (population means ≤ 12.8; = 5.803, df = 3, = 0.005; Fig. 4a). The putatively insect-pollinated species also emitted floral volatiles at a higher rate (population means 76.9–916.6 ng per inflorescence h−1) than did the wind-pollinated species (4.6–14.4 ng per inflorescence h−1; = 4.151, df = 3, = 0.013; Fig. 4b). Mass-specific emission rate showed the same pattern (population means 266–3720 ng (g dry mass inflorescence)−1 h−1 for the putatively insect-pollinated species vs 74–252 ng (g dry mass inflorescence)−1 h−1 for the wind-pollinated species; = 3.238, df = 3, = 0.024; Table S2).

Figure 4.

Mean (+SE) number of compounds (a) and emission rate (b) of volatiles produced by five sedge species. Inflorescences of insect-pollinated Cyperus obtusiflorus and Cyperus sphaerocephalus emitted more volatile compounds (a), at a much higher rate (b), at anthesis (floral samples, gray bars) than in bud (vegetative samples, green bars), whereas floral and vegetative samples were similar for related wind-pollinated species. Sample sizes are in parentheses.

The floral samples of the two putatively insect-pollinated species differed in chemical composition from each other and from conspecific vegetative samples, which clustered instead with floral and vegetative samples of the wind-pollinated species (Fig. 5; Table S2). Some monoterpenes (α-terpineol) and sesquiterpenes (α-gurjunene, caryophyllene) occurred in floral samples of wind-pollinated as well as putatively insect-pollinated species, but these also occurred in vegetative samples. Floral samples from the wind-pollinated species did not consistently contain any compounds that were not present at similar abundance in vegetative samples. By contrast, samples from the putatively insect-pollinated species contained many compounds that were exclusively floral and absent from the wind-pollinated species. The putatively insect-pollinated species C. obtusiflorus and C. sphaerocephalus shared benzenoids (1,2-dimethoxybenzene, 2,6-dimethylanisole, 2-phenylethanol, benzyl alcohol, benzyl tiglate, methyl benzoate) and monoterpenes (1,8-cineole, α-pinene, linalool, myrcene, γ-terpineol, (Z)- and (E)-ocimene) that were restricted to their floral scents or were much more abundant in their floral scents than in vegetative samples or samples from wind-pollinated species. Floral samples of C. obtusiflorus were dominated by benzenoids and monoterpenes, and they uniquely contained 3-phenylpropanol, benzaldehyde, benzyl benzoate, cinnamic acetates, cinnamic alcohols and cinnamic aldehydes (benzenoids), and allo-ocimene, (E)-ocimenol and linalool oxides (monoterpenes). Floral samples of C. sphaerocephalus were dominated by monoterpenes, benzenoids, or 2-methylbutanoic acid depending on locality but all contained the benzenoids 1-hydroxy-2-methoxybenzene and benzoic acid and the nitrogenous compound indole, which were not found in other species.

Figure 5.

Chemical composition of scents of inflorescences at anthesis (floral samples) and in bud (vegetative samples) of five sedge species. Grayscale blocks are shaded proportionally to the per cent abundance of a volatile compound (row) in a scent sample (column). Structure is revealed by ordering columns and rows using complete-linkage hierarchical clustering, shown (for columns) by the dendrogram. The clustering is based on Bray–Curtis dissimilarity calculated from the presence or absence of each compound in the floral and vegetative samples of each population. The floral samples from insect-pollinated Cyperus obtusiflorus (three populations) and Cyperus sphaerocephalus (four populations) formed two distinct clusters, whereas vegetative samples from these species were more similar to floral and vegetative samples from three wind-pollinated species (single populations). Compound names corresponding to row numbers are listed in the Supporting Information Table S2.

Three-group k-means clustering recovered the following groups: (1) floral samples of C. obtusiflorus; (2) floral samples of C. sphaerocephalus; (3) vegetative samples of these putatively insect-pollinated species combined with floral and vegetative samples of the wind-pollinated species. The distinctness of these clusters was supported by permutational MANOVA (F2,44 = 44.4, < 0.001, R2 = 0.43).

Pollen  The putatively insect-pollinated species overlapped with the wind-pollinated species in pollen production and pollen grain size. Mean (± SD) pollen production per anther: C. obtusiflorus = 2885 ± 397; C. sphaerocephalus = 2183 ± 197; C. semitrifidus = 856 ± 134; C. tenax = 921 ± 295; P. oakfortensis = 2223 ± 388. All species have three anthers and one ovule per floret. Therefore, tripling pollen production per anther yields the pollen : ovule ratio. Mean (± SD) pollen grain length: C. obtusiflorus = 31.21 ± 0.56 μm; C. sphaerocephalus = 35.03 ± 0.55 μm; C. semitrifidus = 31.50 ± 0.56 μm; C. tenax = 20.61 ± 0.45 μm; P. oakfortensis = 27.50 ± 0.77 μm. All species had similar, smooth pollen ornamentation (Fig. S1). However, pollenkitt was evident under a microscope as yellow globules for the putatively insect-pollinated species but not for any of the wind-pollinated species.

Insect visitation

Pooling across sites, the mean rates of insect visitation to inflorescences of the putatively insect-pollinated C. obtusiflorus (0.28 visits h−1) and C. sphaerocephalus (0.29 visits h−1) were c. 30-fold or more greater than those for visits to inflorescences of wind-pollinated P. oakfortensis (0.01 visits h−1) or C. tenax (zero visits; Table 1). Cyperus obtusiflorus was visited most often by bees, followed by flies and beetles, while C. sphaerocephalus was visited most often by flies, followed by beetles and bees.

Table 1.   Insect visitation rates (mean visits h−1) to inflorescences of four sedge species
  1. Inflorescence-hours, a measure of observation effort, is calculated as number of inflorescences observed × duration of observation, summed over all observation periods. Visitation rates for each species were similar, and therefore pooled, across sites.

Cyperus obtusiflorus316.40.0760.0950.1010.0030.275
Cyperus sphaerocephalus82.50.0970.1090.0850.0000.291
Pycreus oakfortensis541.50.0040.0020.0020.0020.009
Cyperus tenax101.30.0000.0000.0000.0000.000

Pollinator attraction to floral traits

In 75% of olfactometer trials, solitary bees chose the floral scent of putatively insect-pollinated C. obtusiflorus over an air control; this was marginally nonsignificantly higher than 50% (= 1.903, = 0.057, = 16 trials on 4 bees). Monkey beetles significantly preferred the floral scent, choosing it in 78% of trials (= 2.561, = 0.019, = 37 trials on 20 beetles).

In session 1 of the color × scent model inflorescence experiment (one volatile, ocimene), 125 insects (27 beetles, 41 flies, 57 bees) approached the models (Fig. 6a). Insects approached white models 2.8 times more often than brown models (= 2.61, = 0.009). Scent did not significantly affect approach rate (χ2 = 0.87, df = 2, = 0.65). In Session 2 (four volatiles), 123 insects (nine beetles, 12 flies, 100 bees, two butterflies) approached the models (Fig. 6b). Insects approached white models 2.7 times more often than brown models (= 2.78, = 0.005). They approached models with hexane or hexane laced with floral volatiles 2.0 and 1.8 times as often as models without scent, respectively, although the effect of scent was marginally nonsignificant (χ2 = 5.4, df = 2, = 0.066). There were no significant interactions between the effects of color and scent on approach rate (Session 1: χ2 = 1.6, df = 2, = 0.46; Session 2: χ2 = 2.3, df = 2, = 0.32), and insects approached white models and white inflorescences of C. obtusiflorus at comparable rates (Fig. 6). Treatment effects were similar for landings (not shown).

Figure 6.

Mean (± SE) approach rates by insects to model inflorescences with factorial scent and color treatments, and cut inflorescences of two sedge species, Pycreus oakfortensis and Cyperus obtusiflorus. The floral volatiles treatment consisted of (a) (Z)-ocimene or (b) a blend of (Z)-ocimene, linalool, benzyl alcohol and 2-phenylethanol. The floral volatiles were diluted in hexane, so pure hexane was used as a control. Models: circles, floral volatiles in hexane; diamonds, hexane; triangles, no volatiles.

Yellow model inflorescences were preferred over brown ones by beetles (= 7.99, < 0.001) and flies (= 9.12, < 0.001): beetles were trapped 103 times more often on yellow (7.75 per inflorescence) than on brown (0.08 per inflorescence), and flies were trapped 5.2 times more often on yellow (4.75 per inflorescence) than on brown (0.90 per inflorescence).


Our hypothesis that the sedges C. obtusiflorus and C. sphaerocephalus are dependent on pollination by insects was strongly supported, as their inflorescences are visited frequently by insects, which are effective in depositing pollen, their pollen has relatively little motility in wind, and there was a dramatic reduction in seed set when insects were excluded. Given that sedges were ancestrally wind-pollinated, these results firmly establish at least one transition to insect pollination in the family. Cyperus sphaerocephalus was considered a variety of C. obtusiflorus in some earlier taxonomic treatments but relationships within the Cyperus clade remain unresolved (Gordon-Gray, 1995; Muasya et al., 2009), so we cannot determine whether these species derive from one or two transitions to insect pollination.

We can think of three alternative explanations for the dramatic reduction in seed set when insects were excluded from inflorescences of C. obtusiflorus and C. sphaerocephalus, but none are tenable. First, the insect-exclusion bags, which modestly reduced deposition of wind-borne pollen, may have hindered wind pollination of these putatively insect-pollinated species. Second, tethering the inflorescences to stakes to support the bags may have prevented them from oscillating in the wind, which could have inhibited wind pollination (Niklas, 1987). However, neither possibility can account for our results for C. obtusiflorus because identical insect-exclusion bags and tethering simultaneously had no effect on seed set of intermingled plants of wind- pollinated P. oakfortensis. These alternative explanations are also implausible for C. sphaerocephalus, given its similarity to C. obtusiflorus in inflorescence morphology, pollen motility and habitat. Third, the bags may have inhibited pollen tube growth or seed development by altering the inflorescence microclimate. This explanation can be excluded because hand-pollinated florets of C. obtusiflorus and C. sphaerocephalus inside identical bags set full complements of seeds. Thus, we can firmly conclude that C. obtusiflorus and C. sphaerocephalus depend on insects for pollination.

While seed set of the two insect-pollinated species was dramatically reduced by exclusion of insects, it was not zero (Fig. 1b,c). With insects excluded, seed set was slightly higher when wind was admitted than when it was excluded (Fig. 1b,c). These differences in seed set could have been caused by a very low level of wind pollination, or by tiny insects occasionally penetrating the insect-exclusion mesh. However, these differences were not statistically significant so the low but nonzero seed set when insects were excluded may well have resulted entirely from rare self-fertilizations, or contamination with cross-pollen before the exclusion treatments were applied to buds.

Visits to inflorescences of the insect-pollinated species were fairly evenly split between bees, flies and beetles (Table 1). These comprised at least 75 insect taxa (P.D. Wragg, unpublished). Except for bract-feeding weevils (Curculionidae), all were observed to feed on pollen. All visitor species carried sedge pollen and representatives of all three orders effected pollination during single visits (P.D. Wragg, unpublished), indicating a generalized pollination system. Solitary bees (Apidae and Halictidae) and certain beetles (Scarabaeidae and Chrysomelidae), which had the highest visitation rates and carried the largest and purest loads of sedge pollen, were likely the most effective pollinators (P.D. Wragg, unpublished). By contrast, the few recorded visits to wind-pollinated species were primarily by a few slow-moving beetle taxa: the insect fauna visiting the wind-pollinated species appeared to be similar to that visiting nonfloral vegetation (P.D. Wragg, pers. obs.). This is to be expected given that the three wind-pollinated species are visually indistinguishable from the green-leaf background, at least to the bees that are likely major pollinators, as indicated by their placement at the center of the color hexagon (Fig. 3) and confirmed by their perceptual distance from green-leaf samples of < 0.045 Euclidean hexagon units, which is near the threshold of discrimination (Dyer & Chittka, 2004). By contrast, both insect-pollinated species are > 0.25 Euclidean hexagon units from the center of the hexagon and from the green-leaf samples (Fig. 3), well above discrimination thresholds, and thus highly apparent to bees (Dyer & Chittka, 2004).

Floral scent composition in C. obtusiflorus and C. sphaerocephalus is chemically consistent with that of other species with generalized insect pollination systems (Dobson, 2006). The scents of these two species are more complex (mean of 28 compounds per sample, Fig. 4a) than that of Eleocharis elegans, the only other showy, potentially insect-pollinated sedge in which floral scent has been characterized: E. elegans yielded only 16 compounds, all aliphatics and sesquiterpenes (Magalhães et al., 2005). Floral scent appears to be associated with insect pollination in sedges as we recorded very few compounds in our wind-pollinated species (Fig. 5, Table S2) and Magalhães et al. (2005) were unable to detect any scent compounds emitted from the inconspicuous, presumably wind-pollinated, inflorescences of Eleocharis sellowiana. The scent emission rates of C. obtusiflorus and C. sphaerocephalus (Fig. 4b) are comparable with those of other insect-pollinated plants (Raguso et al., 2003; Shuttleworth & Johnson, 2009).

White and yellow colors were sufficient to attract all the major pollinators of the insect-pollinated species to model inflorescences, at visit rates comparable to those in natural populations (c. 0.3 visits h−1), even in the absence of scent (Fig. 6, Table 1). Choice experiments using a Y-maze olfactometer suggested that scent plays a role in attracting insects. However, we were unable to demonstrate this in the field, possibly because our blend of compounds was not similar enough to the floral bouquet or because additional replication would be required (Raguso, 2008a). A high proportion of sedge pollen in the pollen loads of solitary bees and monkey beetles suggests that they showed constancy to the insect-pollinated species (P.D. Wragg, unpublished); this may have been enhanced by floral scent (Raguso, 2008b), particularly given the presence of other similarly colored flowers. Nevertheless, our prediction that color and scent are key functional traits in this transition from wind pollination to insect pollination was supported.

Pollen of the insect-pollinated species is slightly larger than that of some wind-pollinated species (see the Results section) but pollen of all our species is within the size range typical of wind pollination (Friedman & Barrett, 2009b). Contrary to expectation, pollen production, and hence pollen:ovule ratios, tend to be higher in our insect-pollinated species than in our wind-pollinated species, but within the range of wind-pollinated Cyperaceae and Poaceae (Subba Reddi & Reddi, 1986). Some showy sedges such as Kyllinga spp. have micro-echinate pollen ornamentation that may be associated with insect pollination (Tanaka et al., 2004; Nagels et al., 2009; Sannier et al., 2009), but our insect-pollinated species did not (Fig. S1). Thus, the main difference in pollen traits was the presence of evident pollenkitt in our insect-pollinated species compared with its apparent absence in the wind-pollinated species. The amount and distribution of pollenkitt determines pollen stickiness and hence the ease with which pollen is removed from anthers by wind, the degree to which it clumps, and the degree to which it adheres to insects, among other functions (Pacini & Hesse, 2005). Thus, pollenkitt is strongly associated with animal pollination (Hesse, 1979; Hu et al., 2008). The presence of pollenkitt exclusively in our insect-pollinated species probably caused the lower pollen motility in the wind tunnel of these species compared with our wind-pollinated species, as well as lower pollen removal by wind in the field for insect-pollinated C. obtusiflorus than for wind-pollinated P. oakfortensis (Fig. S2).

The transition from insect pollination to wind pollination has been posited to be irreversible, given the loss of apparently complex traits such as scent and nectar (Cox & Grubb, 1991; Culley et al., 2002). However, the example of sedges shows that traits sufficient to mediate a transition from wind pollination to insect pollination can be regained. As shown by our experiments, color is a key functional trait for this transition. Showy white or yellow bracts appear to have evolved independently in the Cyperus clade containing our insect-pollinated species, at least three times within Rhynchospora section Dichromena (Thomas, 1984), and several more times in other sedge lineages (Goetghebeur, 1998; W.W. Thomas, pers. comm.), suggesting that they can evolve relatively easily. Floral scent also appears to have evolved independently in the sedges E. elegans (Magalhães et al., 2005) and Ficinia radiata (S.D. Johnson & P.D. Wragg, unpublished), which are distantly related to the Cyperus clade. Monoterpenes characteristic of C. obtusiflorus and C. sphaerocephalus floral scent occur in trace amounts in samples of the wind-pollinated sedges studied here (Table S2) and are present in defensive roles across the plant kingdom (Harrewijn et al., 1994), for example in essential oils of other wind-pollinated sedges (Kilani et al., 2008), so enhanced emission of these compounds in floral scent may also be a relatively easy evolutionary shift. This study shows that it is not necessary to evolve nectar to make the transition to insect-pollination, at least where sexual parts are close enough together for pollen-collecting insects to contact stigmas.

Another reason the transition from insect pollination to wind pollination has been considered irreversible is the tendency for wind-pollinated species to evolve separation of male and female functions in space (through unisexual flowers or plants) or time (dichogamy, especially protogyny) to avoid inbreeding (Cox & Grubb, 1991; Culley et al., 2002). Separation of the sexes makes pollination by pollen-collecting insects unlikely, because they have no reason to visit female flowers. Thus, unisexuality hinders the transition from wind pollination to animal pollination (Friedman & Barrett, 2008). In such lineages a reward provided by both sexes, such as nectar (Friedman & Barrett, 2008), or a mechanism of deception in one or both of the sexes (as in many cycads, Donaldson, 1997; Proches & Johnson, 2009), may be required to allow the transition from wind pollination to insect pollination. Although shifts from wind pollination to animal pollination are less likely in protogynous than protandrous lineages (Sargent & Otto, 2004), in sedges protogyny rarely appears complete enough at the inflorescence level to prevent the evolution of insect pollination. This is indicated by failure of protogyny to prevent geitonogamous self-pollination in our species (P.D. Wragg, unpublished; self-incompatibility prevented seed set) and in several Carex species (Friedman & Barrett, 2009a).

Our discovery of a transition from wind pollination to insect pollination in Cyperus sedges of open habitats appears to run counter to the tendency for this transition to occur in closed habitats with low wind speeds that inhibit wind pollination (Friedman & Barrett, 2008). While the lack of a detailed phylogeny prevents us from excluding the possibility that this transition took place in a closed habitat, this seems unlikely given that most species in this group occur in open wetlands and grasslands (Gordon-Gray, 1995). Hence, we need to explore other potential ecological drivers of this transition, such as higher efficiency of insect pollination than wind pollination at low density: in this study the insect-pollinated C. obtusiflorus occurred at lower density than co-occurring wind-pollinated species, and this appears to be a general trend (Friedman & Barrett, 2009b).

In conclusion, there do not appear to be any unsurpassable barriers to the transition from wind pollination to insect pollination: any wind-pollinated plant, but especially those with male and female functions proximal in space and time, can shift to insect pollination by evolving showy floral color, floral scent and pollen of low motility. It thus seems reasonable to infer that the first plants to evolve insect pollination possessed some combination of these traits.


For assistance, we thank: C. Bell, L. Bertolli, A. de Wet, I. Dlamini, L. Hebbelmann, J. and M. Wragg, M. Zulu and members of the Johnson laboratory (field work); A. Jürgens, R. Raguso, and F. Schiestl (scent analysis); K. Gordon-Gray (sedge identifications); J. Colville, C. Eardley and B. Grobbelaar (insect identifications); M. Muasya (sedge phylogeny); C. Peter and A. Shuttleworth (color hexagon); G. Carelse and M. Hampton (wind tunnel); and anonymous reviewers. We were supported by the National Research Foundation, the University of KwaZulu-Natal, the South African Association of Botanists, and the Olaf Wirminghaus award.