Scent matters: differential contribution of scent to insect response in flowers with insect vs. wind pollination traits

Abstract

Background and Aims

Growing experimental evidence that floral scent is a key contributor to pollinator attraction supports its investigation as a component of the suite of floral traits that result from pollinator-mediated selection. Yet, the fate of floral scent during the transition out of biotic into abiotic pollination has rarely been tested. In the case of wind pollination, this is due not only to its rarer incidence among flowering plants compared with insect pollination, but also to the scarcity of systems amenable to genus-level comparisons. Thalictrum (Ranunculaceae), with its multiple transitions from insect to wind pollination, offers a unique opportunity to test interspecific changes in floral fragrance and their potential impact on pollinator attraction.

Methods

First, the Thalictrum phylogeny was revised and the ancestral character state of pollination mode was reconstructed. Then, volatile organic compounds (VOCs) that comprise the scent bouquets of flowers from 11 phylogenetically representative wind- and insect-pollinated species were characterized and compared. Finally, to test the hypothesis that scent from insect-pollinated flowers elicits a significantly greater response from potential pollinators than that from wind-pollinated flowers, electroantennograms (EAGs) were performed on Bombus impatiens using whole flower extracts.

Key Results

Phylogenetic reconstruction of the pollination mode recovered 8–10 transitions to wind pollination from an ancestral insect pollination state and two reversals to insect pollination. Biochemical and multivariate analysis showed that compounds are distinct by species and only partially segregate with pollination mode, with no significant phylogenetic signal on scent composition. Floral VOCs from insect-pollinated Thalictrum elicited a larger antennal response from potential insect pollinators than those from wind-pollinated congeners.

Conclusions

An evolutionary scenario is proposed where an ancestral ability of floral fragrance to elicit an insect response through the presence of specific compounds was independently lost during the multiple evolutionary transitions to wind pollination in Thalictrum.

INTRODUCTION

Pollination ‘syndrome’, as originally described, refers to the suite of floral traits influencing pollen flow by biotic or abiotic vectors that originates from the selection pressures exerted by these vectors in the past (Fenster et al., 2004). These floral traits include, but are not limited to, flower colour, position, shape, size, reward (e.g. nectar, pollen or oils) and scent. In spite of being supported by both large- and small-scale analyses of floral traits (Lázaro et al., 2008; Martén-Rodríguez et al., 2009; Reynolds et al., 2009; Rosas-Guerrero et al., 2014; Johnson and Wester, 2017) and experimental studies of pollinators driving floral diversification (Gervasi and Schiestl, 2017), the pollination syndrome concept remains controversial (Ollerton et al., 2009). Mainly, suites of floral traits observed in extant taxa are a reflection of past selective pressures that may or may not predict current pollination mode. Floral scent can be an important contributor to pollination syndrome and, in certain cases, scent composition can be a better predictor of the type of pollinator than morphological floral traits (Gong et al., 2015). Floral volatiles can act as both a long-distance and an immediate cue for pollinators (Klahre et al., 2011). They can also serve as deterrents against secondary flower visitors, such as less effective pollinators (Gegear et al., 2017) or herbivores (Pichersky and Gershenzon, 2002; Turlings and Ton, 2006). Despite growing research on the role of floral scent in mediating plant–pollinator interactions (Levin et al., 2001; Peng et al., 2017), few studies have linked floral scent compounds with pollinator physiology (reviewed in Schiestl, 2010; Schiestl and Johnson, 2013).

That floral scent is a key component mediating pollinator attraction is borne out by a variety of complementary lines of evidence. For example, pollinator choice in hawkmoth vs. hummingbird flowers can be narrowed down to a single scent compound in petunia (Klahre et al., 2011), and floral scent alone is sufficient to elicit differential visitation by bumble-bees vs. hummingbirds between two species of Mimulus (Byers et al., 2014). On the other hand, when pollinators are no longer necessary, as in populations transitioning from an outcrossing to a selfing mating system, floral fragrance is dramatically reduced (Doubleday et al., 2013). Similarly, wind-pollinated plants tend to have lower volatile emissions and less complex floral bouquets than their insect-pollinated counterparts (Wragg and Johnson, 2011; Welsford et al., 2016).

The evolution of wind pollination from animal pollination represents a major evolutionary transition in flowering plants (Barrett, 2008). Approximately 10 % of angiosperm species rely on wind for the transport of pollen between plants, and wind pollination has originated at least 65 times in diverse animal-pollinated lineages (Linder, 1998). While there has been ample discussion on the circumstances favouring the evolution of wind pollination (Cox and Grubb, 1991; Culley et al., 2002; Friedman and Barrett, 2009), the fate of floral scent during this transition is rarely investigated (Wragg and Johnson, 2011; Doubleday et al., 2013; Welsford et al., 2016).

Few genera contain both wind- and insect-pollinated species, making Thalictrum (Ranunculaceae, approx. 190 species) an exemplary system to examine differences in floral scent among closely related species with such different pollination modes. Distinct floral morphologies are found in insect- and wind-pollinated Thalictrum flowers, prompting the question of whether floral scent is also associated with pollination mode. Wind pollination has repeatedly evolved from insect pollination in Thalictrum (Soza et al., 2012), providing multiple opportunities for comparisons of closely related species. Thalictrum flowers are pollinated by unspecialized, pollen-collecting insects, mainly syrphid flies (Syrphidae), beeflies (Bombyliidae) and solitary bees (Kaplan and Mulcahy, 1971; Pellmyr, 1995, and references therein). Since Thalictrum flowers do not produce nectar and lack petals (Leppik, 1964), other floral features such as sepal and stamen colour and size, and possibly floral scent, are likely contributors to pollinator attraction in this genus (Kaplan and Mulcahy, 1971).

Here, we set out to characterize and compare the volatile organic compounds (VOCs) that comprise the scent bouquets of flowers from phylogenetically representative species of insect- and wind-pollinated Thalictrum. A revised Bayesian phylogeny with maximum likelihood reconstruction of pollination mode (wind or insect) provided the evolutionary context. To test the hypothesis that scent from insect-pollinated flowers elicits a significantly greater response from potential pollinators than that from wind-pollinated flowers, we performed electroantennograms (EAGs) on Bombus impatiens with single volatiles and whole flower extracts from both flower types across the phylogeny. Our work contributes towards filling an existing gap by integrating pollination mode, floral scent and pollinator physiology within a phylogenetic context.

MATERIALS AND METHODS

Phylogenetic sampling

For chloroplast DNA (cpDNA)- and nuclear ribosomal DNA (nrDNA)-based phylogenetic reconstruction of the genus, we included 80 of 196 species from 13 of the 14 Thalictrum recognized sections (Tamura, 1995), to represent the breadth of taxonomic classification and geographic distribution. The following 28 species were added since the most recent Thalictrum phylogeny (Soza et al., 2013): T. acutifolium, T. arkansanum, T. calabricum, T. calcicolum, T. chelidonii, T. cirrhosum, T. confine, T. coreanum, T. elegans, T. fargesii, T. foeniculaceum, T. foliolosum, T. galeotti, T. gibbosum, T. grandisepalum, T. lankesteri, T. myriophyllum, T. peltatum, T. pringlei, T. pubigerum, T. rostellatum, T. rotundifolium, T. sachalinense, T. saniculiforme, T. texanum, T. trichopus, T. tuberiferum and T. tuberosum. The six outgroups included in the reconstruction were identified from molecular phylogenies of the Ranunculaceae (Wang and Chen, 2007); three of them were newly added: Semiaquilegia adoxoides, Isopyrum manshuricum and Enemion raddeanum (and a different species of Aquilegia was used).

Molecular sampling

The phylogenetic analysis included data from four new cpDNA regions – ndhA, rbcL, rpl32–trnL and trnL–trnF – in addition to the previously analyzed trnV–ndhC and the nuclear ribosomal external transcribed spacer (ETS) and internal transcribed spacer (ITS) regions. New sequences were obtained from GenBank (Supplementary Data Table S1) and incorporated into the previous alignment (http://purl.org/phylo/treebase/phylows/study/TB2:S13801) (Soza et al., 2013).

Phylogenetic analyses

Sequences were edited in Sequencher version 4.9 (Gene Codes Corporation, Ann Arbor, MI, USA), aligned manually in Mesquite version 3.03 (Maddison and Maddison, 2017) and deposited in TreeBASE (http://purl.org/phylo/treebase/phylows/study/TB2:S22211?x-access-code=33cab2a77a5cf3f752973dd1a5da693e&format=html). Models of evolution were determined for each gene by jModelTest version 2.1 (Posada and Crandall, 1998) under the Akaike information criterion (Akaike, 1974): GTR + I + Γ (ITS, trnL–trnF and trnV–ndhC); GTR + Γ (ndhA, rpl32–trnL and ETS) and GTR + I (rbcL). Four ambiguous nucleotide regions were excluded from the analysis: nucleotides 682–730, 2679–2701, 3559–3590 and 5370–5642.

Bayesian analyses of the concatenated cp + nrDNA data set were conducted in MrBayes version 3.2.3 (Huelsenbeck and Ronquist, 2001) via the CIPRES Science Gateway version 3.3 (Miller et al., 2015), with data partitioned under the selected models for each DNA region. We used default priors and unlinked parameters for nucleotide frequencies, substitution rates and gamma shape across data partitions. Bayesian analyses were conducted with three independent Markov chain Monte Carlo (MCMC) runs (Yang and Rannala, 1997) for 25 million generations for the combined data set. Convergence was determined when the average standard deviation of split frequencies remained <0.01. For the combined data set, the first 2.53 million trees were discarded before convergence and the remaining trees from each run were pooled to construct a 50 % majority rule consensus tree to obtain posterior probabilities (PPs).

Ancestral character state reconstruction

We used the maximum clade credibility consensus tree as a framework for ancestral character state reconstruction of pollination mode, with character states coded as ‘insect’ or ‘wind’. Character states were assigned as previously (Soza et al., 2012) from a combination of field experiments, which were available for three of the study species (Kaplan and Mulcahy, 1971; Melampy and Hayworth, 1980; Steven and Waller, 2004), or using Kaplan and Mulcahy’s (1971) Pollination Index (PI) based on flower morphology as a proxy, which averages character states for seven key floral traits (flower size and colour, anther and stigma length, pistils and stamens exserted or included, and filaments pendulous or erect). We found that this PI accurately predicts pollination mode in six Thalictrum species studied in the field (Melampy and Hayworth, 1980; Davis, 1997; Steven and Waller, 2004; Guzmán, 2005; Humphrey, 2018). Therefore it was applied here to 12 additional taxa, using online resources such as flower photos, herbarium specimens and floral descriptions (Supplementary Data Table S2). Species were coded as ‘wind’ for a PI of <2 or ‘insect’ for a PI >2; the few cases of PI = 2 were assigned based on overall phenotype from flower pictures, with coloured, larger flowers assigned as ‘insect’ and green, smaller flowers as ‘wind’. For simplicity, and because of its limited empirical evidence, we decided not to include a ‘mixed’ pollination state that has been reported in the cryptically dioecious species T. pubescens where, even though mostly insect pollinated, wind pollination contributes to seed set (Kaplan and Mulcahy, 1971; Davis, 1997). Ancestral character states for pollination mode were reconstructed using a maximum likelihood approach as implemented in Mesquite version 3.03 (Maddison and Maddison, 2017).

Plant materials for floral scent collection

Thalictrum plants were grown at the University of Washington (UW) greenhouse, from wild-collected seed or from nurseries. Guided by the expanded phylogeny and character reconstruction presented here (Fig. 1), 11 species were chosen to represent the two pollination modes (insect, wind) across major clades. Voucher specimens for new taxa were deposited in the University of Washington Herbaria (WTU) (Supplementary Data Table S3). For the two dioecious species, both staminate and carpellate inflorescences were sampled and, due to the similarity of the VOC profiles, both sexes were later pooled for species-level analyses.

Fig. 1.

Evolution of pollination mode (insect or wind) in Thalictrum. A revised phylogeny and likelihood reconstruction of pollination mode for Thalictrum and outgroups (see also Supplementary Data Fig. S1). Strongly supported nodes (Bayesian posterior probability >0.95) are indicated by bold branches. Proportional likelihoods (PLs) of ancestral character states are shown as pie charts at each node, in numbers for main clades, and transitions to wind pollination in terminal clades with proportional likelihoods >0.875 are indicated by arrows. The 11 taxa sampled for floral scent are shown in bold. Representative flowers from sampled species: (A) T. thalictroides; (B) T. clavatum; (C) T. filamentosum; (D) T. aquilegiifolium; (E) T. delavayi; (F) T. rochebrunnianum; (G) T. flavum; (H) T. lucidum (photo by Aaron Liston); (I) T. hernandezii, hermaphrodite; (J) T. hernandezii, staminate; (K) T. dioicum, carpelate; (L) T. dioicum, staminate; (M) T. dasycarpum, carpelate; and (N) T. dasycarpum, staminate.

Floral morphology of Thalictrum sampled for floral scent

Clade I insect-pollinated species have upward-facing flowers with relatively large (exceeding the stamens) white or pink petaloid sepals (T. thalictroides; Fig. 1A) or petaloid stamens (with white-pink filaments dilated distally, T. clavatum and T. filamentosum; Fig. 1B, C). Clade II insect-pollinated T. aquilegiifolium has showy stamens with purple filaments (Fig. 1D), while T. delavayi (Fig. 1E) and T. rochebrunnianum (Fig. 1F) have downward-facing flowers with large purple sepals (Fig. 1E, F). Thalictrum flavum and T. lucidum are similar, with relatively small white sepals and white stamen filaments (Fig. 1G, H). All the insect-pollinated species in this study had hermaphroditic flowers. The clade II wind-pollinated species T. hernandezii is andromonoecious (staminate and hermaphroditic flowers on the same plant) (Fig. 1I, J), while T. dioicum (Fig. 1K, L) and T. dasycarpum (Fig. 1M, N) are dioecious (staminate and carpellate flowers on different plants) (Soza et al., 2012, and references therein).

Scent collection

Scent was collected from flowers using dynamic sorption methods, as previously described (Raguso and Pellmyr, 1998; Riffell et al., 2008; Byers et al., 2014). For each of the 11 species, 2–5 floral stems were cut and weighed, and flower counts were obtained from inflorescences. Floral stems were placed in separate 25 mL beakers with miliQ water for each species and enclosed within a nylon oven bag (Reynolds, Richmond, VA, USA). For each sample, a water-only control was included. The closed volatile collection system consisted of a diaphragm pump (Barnant Co., Barrington, IL, USA) that pushed air in through a charcoal filter at a flow rate of 0.11 L min^–1 and out through a volatiles trap. Volatiles traps consisted of 100 mg of Porapak Q adsorbent (mesh size 80–100, Waters Corp., Milford, MA, USA) in 7 mm borosilicate glass tubes plugged with silanized glass wool. Collection proceeded for 24 h to control for the effects of circadian scent emission on floral volatile abundance (Verdonk et al., 2003).

Scent analysis

Volatiles were eluted from the scent traps into 2 mL amber glass vials immediately after the sampling period using 800 μL of HPLC-grade hexane and stored at –80 °C. A 200 μL aliquot was concentrated 10-fold using a flow of nitrogen gas, and 3 μL of this concentrated sample was analysed using an Agilent 7890A gas ghromatograph and a 5975C Network Mass Selective Detector (Agilent Technologies, Palo Alto, CA, USA). Two gas chromatography (GC) columns (J&W Scientific, Folsom, CA, USA) were used: DB5MS (30 m, 0.25 mm, 0.25 μm) and Chiral SilB (30 m, 0.25 mm, 0.25 μm), with helium as the carrier gas at a constant flow of 1 mL min^–1. The initial oven temperature was 45 °C for 4 min, followed by a heating gradient of 10 °C min^–1 to 230 °C, which was then held isothermally for 4 min.

Chromatogram peaks were integrated using ChemStation software (Agilent Technologies) and the top five spectral matches for each peak were found using the National Institute of Standards (NIST) mass spectral library (approx. 120 000 spectra). To verify volatile identities, a 10 ng μL^–1 C7–C30 alkane standard was used to calculate Kovats Indices for all peaks, and synthetic standards for some compounds were run using the same GC-mass spectroscopy (MS) method. A custom software written in Python calculated the Kovats Index for each peak in each samples, compared it with database DB-5 Kovats Indices gathered from the literature and from synthetic standards run in our laboratory (3096 records of plant volatiles), and selected the compound as the peak identity that best matched the top five spectral matches and had a Kovats Index within 10 units of the same match. A range of ±10 Kovats units was chosen because runs of 41 authentic standards on our GC-MS always matched Kovats values in the literature within this range. If no match in our Kovats Index database was found, the top spectral match was retained if its match percentage was ≥70; if not, the peak was discarded. After this process was performed on all samples, data were concatenated to create a single matrix of chemical abundance per compound for each plant species. Peak areas were exported and quantified using the calibration curve of α-pinene to determine mass of volatiles (ng) emitted per gram-minute. Peaks present in the bag-only controls were subtracted from the samples entirely if they were not known floral volatiles; if a peak found in a control was identified as a known floral volatile (e.g. α-pinene or β-ocimene), the mean amount of the volatile found in control runs was removed from each sample. Floral-specific compounds were identified if present in more than two floral samples.

Organ-specific VOC emissions

In order to determine the source of VOCs within a flower, we selected the insect-pollinated species T. thalictroides (Fig. 1) as a representative of clade I with large persistent sepals and a statistically significant EAG effect (Fig. 3). We dissected a total of 18 flowers from seven individuals into separate organs (sepals, stamens and carpels) and used solid phase micro extraction (SPME) to capture scent from the samples. Because dissected floral organs are very small and had very little mass to have detectable emissions, organ types (sepals, stamens or carpels) from all individuals were pooled together in a 10 mL SPME vial (Agilent Technologie). The bottom 2 cm of the vial was submerged in warm water (approx. 40 °C) to facilitate volatilization of compounds, and the SPME fibre was inserted into the vial and exposed for 20 min. The fibre was then removed and inserted into the gas chromatograph. An empty vial was used as a control.

Fig. 3.

Bumble-bee electroantennogram (EAG) recordings in response to Thalictrum floral extracts of insect- and wind-pollinated species. (A) Example EAG recordings from clipped B. impatiens antenna in response to the hexane negative control, wind-pollinated T. dioicum and insect-pollinated T. delavayi. Stimulus duration was 0.5 s. (B) EAG responses to scents from five insect-pollinated species [T. delavayi (T. del); T. flavum (T. fla); T. aquilegiifolium (T.aqu); T. thalictroides (T. tha); and T. rochebrunnianum (T. roc)] and three wind-pollinated species [T. hernandezii (T. her); T. dioicum (T. dio); and T. dasycarpum (T. das)]. Asterisks denote significantly greater responses relative to the negative (hexane) control (P < 0.05). (C) EAG responses across all insect- and wind-pollinated floral scents. Bars are the mean ± s.e.m.

Multivariate analysis of floral compounds

To investigate differences in floral compounds between taxa and in relation to the presumed pollination mode (wind or insect), we performed non-metric multidimensional scaling (NMDS) ordination analysis. NMDS allows visualization of similarities between species in a pre-defined number of dimensions using a non-Eigen values distance matrix, suitable for non-parametric floral scent data. A pairwise distance matrix of scent emission rates (ng emitted g tissue^–1 h^–1) was calculated using the Bray–Curtis dissimilarity metric (Bray and Curtis, 1957) with the ‘Fathom’ toolbox for Matlab (Jones, 2015). Two axes were used to specify the data with guidance from a stress vs. axis graph. NMDS output was rotated by principal component analysis to visualize the largest variance of the NMDS scores along the x- and y-axes. The same results were produced independently with Matlab’s built-in ‘mdscale’ function. Final results were plotted as a biplot according to Thalictrum species and individual scent compounds that elicited EAG responses. One-way ANOSIM was conducted to determine if the data clustered significantly by species or pollination mode, as done in comparable studies of floral scent (Majetic et al., 2008).

Analysis of the phylogenetic signal in VOC composition

To determine if there was a phylogenetic signal in the composition of floral scent, we pruned our latest Thalictrum phylogeny of 80 Thalictrum species to the 11 species sampled here (Fig. 1, species in bold). We calculated the patristic distances among species from the pruned tree in Mesquite v. 3.2 (build 801) (Maddison and Maddison, 2017), and then compared the two data sets (chemical vs. phylogenetic) using a Mantel test (Mantel, 1967) in Matlab 8.0 and Statistics Toolbox 8.1 (The Mathworks, Inc., Natick, MA, USA) using the Mantel script from the ‘Fathom’ toolbox (Jones, 2015). As Mantel tests are known to be susceptible to Type 1 error rates when used to test correlation with phylogenetic distance (Harmon and Glor, 2010), we also calculated Blomberg’s K for each compound using the ‘multiPhylosignal’ function in R package Picante (Blomberg et al., 2003; Kembel et al., 2010). Additionally we calculated multivariate Blomberg’s K for scent composition using the ‘K.mult’ and ‘physignal’ functions in R (Blomberg et al., 2003; Kembel et al., 2010; Adams, 2014; Goolsby, 2016; Adams et al., 2018). In all tests of Blomberg’s K, VOC values were averaged by plant species and fit to a model of Brownian motion.

Electroantennograms

To determine the olfactory responses of a model pollinator to the scent of various Thalictrum spp., as well as responses to individual constituents identified in the scent, we first performed EAGs stimulating the antenna of B. impatiens workers using scent extracts of eight of the Thalictrum study species. Bombus impatiens was used due to their robust antennal responses (Byers et al., 2014) and because Thalictrum are visited by pollen-collecting bees, including Bombidae (Motten, 1986; Pellmyr, 1995, and references therein). We have also observed that Bombus terrestris avidly collect pollen from Thalictrum thalictroides under experimental conditions (Supplementary Data Video), supporting its use as a pollinator model. Each bumble-bee was chilled until anaesthetized, and her whole antenna removed with sharp tenotomy scissors. The tip of the antennae was also removed. The cut antenna was inserted into a glass electrode filled with Spectra 360 conductive gel (Parker Laboratories, Inc., Fairfield, NJ,m USA). The electrodes and antenna(e) were placed in front of a glass tube through which a continuous amount of air was streamed at 100 mL min^–1 flow using a flowmeter (Gilmont Industries, Barnant Co., Barrington, IL, USA) at room temperature. The signal from the EAG was conditioned (Humbug Noise Eliminator; A-M Systems), amplified (1800 A-M Systems, Sequim, WA, USA) and recorded using an RZ2 amplifier (Tucker-Davis Technologies, Alachua, FL, USA). Olfactory stimuli were delivered to the antenna by pulses of clean air from a constant air stream that were diverted through a glass syringe containing a piece of filter paper bearing floral odours. The stimulus was pulsed by means of a solenoid-activated valve controlled by the software OpenEx (Tucker-Davis Technologies). The outlet of the stimulus syringe was positioned 2 cm from and orthogonal to the centre of the antennal flagellum. Stimulus duration was 500 ms, and odour stimuli were delivered five times, with each pulse separated by a 10 s interval. The control solvent for the floral extracts was hexane (Sigma-Aldrich; >99 % purity), whereas mineral oil was used for the single volatiles (Sigma-Aldrich, >99 % purity). The floral olfactory stimuli were presented using an odour cartridge, which consisted of a 2 mL glass syringe (Air-tite, Virginia Beach, VA, USA) and a standard 20-gauge 1-inch needle (PrecisionGlide; Becton, Dickinson and Company, Franklin Lakes, NJ, USA). Two different experimental series were conducted: the first examined the antennal responses to the different Thalictrum species, either insect or wind pollinated, and the second experimental series examined the antennal responses to single odorants that were found in the floral extracts. In the first experimental series, 50 μL of the headspace-collected floral extract of insect-pollinated T. thalictroides, T. flavum, T. rochebrunnianum, T. aquilegiifolium and T. delavayi or the wind-pollinated T. dioicum, T. hernandezii and T. dasycarpum were pipetted onto a piece of filter paper (3MM Whatman) and placed in a glass syringe. In the second experimental series, each compound (Sigma-Aldrich, >99 % purity, except trans-β-ocimene which has >90 % purity) was diluted in mineral oil to a partial pressure of 0.1 Torr, to normalize the airborne concentration presented to the insect antenna(e) during stimulation; 5 μL of benzaldehyde, trans-β-ocimene, β-myrcene, α-farnesene (mixture of isomers) or the mineral oil (no odour) control were pipetted onto a filter paper and placed in a glass syringe. Testing these single odorants allowed us to gauge how the bees respond to constituents found in the floral extracts. Stimuli were randomly presented to the bee, and any degradation of the EAG signal (generally <10 % over the limited time course of the experiment) was adjusted offline by comparing responses to benzaldehyde which was presented multiple times during the experiment. The EAG response (mV) was compared between the floral extracts of the different species and a hexane control, or the single odorants and the mineral oil control, using Kruskal–Wallis tests with multiple comparisons (n = 8 bees).

Phylogenetic analysis of EAG response

In order to assess whether ancestral taxa in Thalictrum were able to elicit a response in Bombus, we tested for a phylogenetic signal in the EAG response using Blomberg’s K and performed an ancestral state reconstruction. The ‘multiPhylosignal’ function in the library ‘Picante’ was used to calculate Blomberg’s K for individual species (Blomberg et al., 2003; Kembel et al., 2010). The ‘contMap’ function in R was used for continuous ancestral state reconstruction under Brownian Motion (Revell, 2012).

Data and scripts

Custom python software used to calculate the Kovats Index can be downloaded at https://github.com/cliffmar/GCMS_and_combine. All data and scripts necessary to perform analysis in Matlab and R are available for download at https://github.com/Jesusthebotanist/Thalictrum-Floral-Scent.

RESULTS

A revised phylogeny of Thalictrum

The addition of 28 previously unsampled taxa provided a net increase of 11 species from a previous Thalictrum phylogeny (Soza et al., 2013). The increased taxonomic sampling, along with the addition of four loci, improved the support for the overall tree topology, especially its backbone, while improving the resolution of intermediate-level clades. This was despite leaving out 17 previously published taxa that were not represented in the new GenBank data, belonging to already well-supported clades.

A major early split in the genus and three of the subclades continues to be strongly supported (Soza et al., 2013, clades I and II and A, B and C, respectively). Increased taxonomic and molecular sampling resulted in the expansion of existing clades, the resolution of polytomies between major groups in clade II (which contains the majority of the species sampled to date) and the formation of two new strongly supported monophyletic groups (Supplementary Data Fig. S1).

Three of the newly sampled species were nested in clade I (T. acutifolium, T. tuberiferum and T. coreanum), while all others fell into clade II. Thalictrum macrocarpum continued to be sister to the rest of clade I, while T. ichangense clustered with T. thalictroides and T. clavatum. Three added species appeared to be early diverging within clade II: T. trichopus was the next sister lineage after T. macrocarpum, with strong support (PP = 1); T. foeniculaceum was next as sister to T. alpinum (PP = 0.99); and T. foliolosum next (in a clade with T. virgatum and T. rhynchocarpum, PP = 1). Thalictrum calabricum, T. grandisepalum, T. myriophyllum, T. sachalinense and T. tuberosum formed a strongly supported clade (PP = 1), expanding from previously defined subclade 1 (Soza et al., 2013) by including previously unresolved taxa. Thalictrum pringlei, T. galeotti, T. pubigerum, T. lankesteri, T. gibbosum and T. peltatum expanded another existing clade, falling alongside other wind-pollinated, andromonoecious species from Central America (clade B, Soza et al., 2013). Thalictrum arkansanum, T. texanum and T. confine fell into another well-supported clade alongside other wind-pollinated, dioecious species from North America (clade C, Soza et al., 2013). Two new, strongly supported clades within clade II were identified by our analysis: one composed of T. chelidonii, T. elegans, T. saniculiforme and T. rotundifolium (PP = 1), and the other containing T. calcicolum, T. cirrhosum and T. fargesii (PP = 1).

Reconstruction of pollination mode on a revised Thalictrum phylogeny

The ancestral character state for Thalictrum was inferred as insect pollination, as previously, but with a much increased proportional likelihood [PL = 0.89, compared with PL = 0.58 (Soza et al., 2012)] (Fig. 1). The ancestral condition for clade I was reconstructed as insect pollination (PL = 0.99), and this was also more strongly supported than before (PL = 0.66, Soza et al., 2012). The ancestral condition for clade II, previously uncertain, was here reconstructed as most probably insect pollination (PL = 0.81). Wind pollination was inferred to have evolved independently at least eight times in the genus (with PL cut-off of 0.5) and as many as ten times (PL cut-off of 0.875). Four main clades were reconstructed as ancestrally wind pollinated within clade II (Fig. 1, arrows). The additional 4–6 transitions to wind pollination were species specific (based on current phylogenetic sampling) or occurred in the ancestor of several species but with low statistical support (PL <0.875). Reversions from wind to insect pollination may have occurred at least twice, in T. flavum and T. lucidum (Fig. 1). Interestingly, these species fall in a category of a relatively intermediate pollination index (2.3 for T. flavum and 2.4 for T. lucidum, Kaplan and Mulcahy, 1971) that is consistent with their intermediate flower morphology: small, white flowers and relatively long stamens with intermediate anther size (Fig. 1G, H).

Floral scent analysis

Volatile organic compound composition and average emission of floral scent were determined for the 11 Thalictrum species. Compounds were similar between sexes in the two dioecious species (Supplementary Data Table S5), and were therefore combined in subsequent species-level analyses (except NMDS; see below). Fifty-one compounds, comprising 22 terpenoids, seven aromatics and 22 aliphatics, were identified across all species (Table 1), with no particular trend in total number of compounds [Sorenson’s Index (SI) for similarity: mean insect-pollinated SI = 0.27 ± 0.12; mean wind-pollinated SI = 0.40 ± 0.12] or in emission rates between insect- and wind-pollinated species (Table 1). However, there were qualitative differences between species. For instance, T. dioicum, a wind-pollinated species, emits a diverse suite of terpenoid and aliphatic compounds that is dominated by the aliphatic (Z)-hex-3-en-1-ol acetate (Table 1; Fig. 2A, peak b). In contrast, one of the most abundant volatiles, especially (but not exclusively) in certain insect-pollinated species, was trans-β-ocimene. The aromatic benzaldehyde was found in the two insect-pollinated species that represent potential evolutionary reversals from wind-pollinated ancestors (T. lucidum and T. flavum). These two species also exclusively shared three terpenoid compounds (citronellyl acetate, neryl acetate and geranyl acetate). Benzyl benzoate was found in clade I species (the clade containing the species with ancestral traits for the genus, Soza et al., 2012), and again in one of the potential evolutionary reversions to insect pollination (T. flavum) (Table 1; Fig. 1).

Table 1.

Fragrance composition of flowers from insect- and wind-pollinated Thalictrum

	KRI	Insect								Wind
		Thalictrum clavatum	Thalictrum thalictroides	Thalictrum filamentosum	Thalictrum aquilegifolium	Thalictrum rochebrunnianum	Thalictrum delavayi	Thalictrum lucidum	Thalictrum flavum	Thalictrum hernandezii	Thalictrum dioicum	Thalictrum dasycarpum
Number of samples		3	3	2	5	3	5	4	2	5	5	5
Average scent emission (ng h–1)		12.50	0.49	120.14	2.16	12.65	27.21	0.42	6.65	18.42	4.33	18.39
Total number of compounds		4	19	3	5	17	10	24	24	15	22	16
Terpenoids
2,6,6-Trimethylbicyclohept-2-ene (α-Pinene)	934	–	2.35 (0.01)	–	–	6.69 (0.03)	–	–	21.45 (0.06)	–	0.80 (0.00)	3.16 (0.02)
2,2-Dimethyl-3- methylidenebicycloheptane (Camphene)	955	–	0.67 (0.01)	–	–	6.65 (0.03)	–	–	–	–	0.20 (0.00)	11.17 (0.06)
6,6-Dimethyl-2- methylidenebicycloheptane (β-Pinene)	979	54.37 (0.11)	4.37 (0.02)	–	–	3.44 (0.02)	–	0.14 (0.00)	17.13 (0.02)	–	0.69 (0.00)	0.94 (0.01)
7-Methyl-3-methylene-1,6-octadiene (β-Myrcene)	991	–	0.25 (0.00)	–	3.13 (0.02)	0.28 (0.00)	–	0.84 (0.00)	0.22 (0.00)	–	0.22 (0.00)	0.06 (0.00)
3,7,7-Trimethylbicyclohept-3-ene (delta-3-Carene)	1010	–	–	–	–	7.33 (0.04)	–	7.11 (0.04)	–	–	–	–
1-Methyl-4-(prop-1-en-2-yl) cyclohex-1-ene (d-Limonene)	1030	–	3.00 (0.02)	18.60 (0.19)	–	18.51 (0.14)	22.02 (0.2)	2.09 (0.01)	0.27 (0.00)	3.69 (0.02)	1.15 (0.01)	0.92 (0.01)
3-Methylene-6-(1-methylethyl) cyclohexene (β-Phellandrene)	1032	–	–	–	–	–	4.99 (0.05)	–	–	–	–	–
1,3,3-Trimethyl-2-oxabicyclooctane (Eucalyptol)	1033	–	–	–	–	–	–	–	–	–	5.25 (0.05)	–
3,7-Dimethyl-1,3,6-octatriene (E-β-Ocimene)	1046	22.78 (0.11)	12.49 (0.00)	–	69.82 (0.13)	1.46 (0.01)	–	7.84 (0.02)	11.85 (0.08)	0.21 (0.00)	10.59 (0.07)	42.19 (0.19)
4-Methyl-1-(1-methylethyl)-1,4- cyclohexadiene (γ-Terpinene)	1060	–	–	–	–	–	–	0.13 (0.00)	–	–	–	–
(Z)-3,7-Dimethyl-2,6-octadien-1-ol (Nerol)	1225	–	–	–	–	–	–	–	6.46 (0.06)	–	–	–
(Z)-3,7-Dimethylocta-2,6-dienal (Neral)	1247	–	–	–	–	–	–	–	0.66 (0.01)	–	–	–
(E)-3,7-Dimethyl-2,6-octadien-1-ol	1250	–	–	–	–	–	–	–	9.02 (0.09)	–	–	–
(E)-3,7-Dimethylocta-2,6-dienal (Geranial)	1269	–	–	–	–	–	–	–	0.69 (0.01)	–	–	–
6-Octen-1-ol, 3,7-dimethyl-, acetate (Citronellyl acetate)	1353	–	–	–	–	–	–	3.58 (0.02)	0.29 (0.00)	–	–	–
(Z)-3,7-Dimethyl-2,6-octadien-1-yl acetate (Neryl acetate)	1362	–	–	–	–	–	–	0.17 (0.00)	0.28 (0.00)	–	–	–
3,7-Dimethyl-2,6-octadien-1-yl acetate (Geranyl acetate)	1381	–	–	–	–	–	–	3.45 (0.02)	2.26 (0.02)	–	–	–
4,11,11-Trimethyl-8- methylidenebicycloundec-4-ene (E-β-Caryophyllene)	1431	0.08(0.00)	–	–	–	5.56 (0.03)	–	6.66 (0.02)	0.95 (0.01)	0.74 (0.01)	–	–
(E)-7,11-Dimethyl-3-methylene- 1,6,10-dodecatriene [(E)-β-Farnesene]	1440	–	–	–	–	–	–	0.11 (0.00)	–	–	–	–
2,6,6,9-Tetramethylcycloundeca- 1,4,8-triene (α-Humulene)	1455	–	1.05 (0.01)	–	–	7.07 (0.04)	–	2.88 (0.01)	–	–	–	–
1,5-Dimethyl-8-(prop-1-en-2-yl) cyclodeca-1,5-diene (Germacrene D)	1482	–	–	–	–	0.42 (0.00)	–	1.98 (0.01)	–	–	–	–
3,7,11-Trimethyl-1,3,6,10- dodecatetraene (E,E-α-Farnesene)	1544	–	44.53 (0.17)	–	–	11.86 (0.06)	2.05 (0.02)	1.20 (0.01)	–	0.58 (0.01)	1.34 (0.01)	7.21 (0.04)
Aromatics
Anisole	918	–	0.98 (0.01)	–	–	–	–	1.99 (0.01)	0.18 (0.00)	–	–	–
Benzaldehyde	961	–	–	–	–	–	–	0.23 (0.00)	0.16 (0.00)	–	0.42 (0.00)	–
1-Phenylethan-1-one	1068	–	0.25 (0.00)	–	–	–	–	–	–	–	–	–
Methyl benzoate	1091	–	4.5 (0.02)	–	1.12 (0.01)	–	0.24 (0.00)	0.13 (0.00)	0.76 (0.01)	–	1.49 (0.01)	–
Ethyl 2-phenylacetate	1245	–	1.8 (0.02)	–	–	–	–	–	–	–	–	–
3-Phenyl-3-pyridin-3-ylprop-2-en-1-ol	1312	–	–	–	–	–	–	–	0.31 (0.00)	–	–	–
Benzyl benzoate	1778	22.56 (0.08)	1.15 (0.01)	–	–	–	–	–	1.79 (0.02)	–	–	–
Aliphatics
Hexanal	798	–	–	–	–	8.3549	–	–	–	–	–	0.27 (0.00)
(Z)-Hex-3-en-1-ol acetate	849	–	10.14 (0.06)	–	20.39 (0.11)	8.8148	–	45.14 (0.12)	23.08 (0.22)	31.48 (0.18)	63.17 (0.11)	22.58 (0.11)
(E)-Hex-2-enal	854	–	–	–	–	–	–	–	–	–	–	0.91 (0.01)
(Z)-Hex-3-en-1-ol	858	–	3.22 (0.02)	–	–	11.9235	0.8851	10.72 (0.01)	0.25 (0.00)	0.43 (0.00)	7.35 (0.04)	6.20 (0.04)
Hexan-1-ol	868	–	–	–	–	–	–	0.29 (0.00)	0.13 (0.00)	–	–	–
Decane	987	–	–	–	–	–	–	–	–	–	–	0.06 (0.00)
Octan-1-ol	1071	–	–	24.44 (0.16)	–	–	–	–	0.15 (0.00)	–	0.28 (0.00)	–
Undecane	1100	–	–	–	–	–	–	–	–	1.94 (0.01)	–	–
Dodec-1-ene	1190	–	0.40 (0.00)	–	–	–	–	–	–	–	1.13 (0.01)	–
Ethyl octanoate	1197	–	–	–	–	–	62.9353	–	–	–	0.29 (0.00)	–
Dodecane	1200	–	–	–	–	–	–	–	–	–	0.18 (0.00)	0.05 (0.00)
Decanal	1202	–	–	–	–	–	–	–	–	1.88 (0.01)	–	0.05 (0.00)
Tridecane	1300	–	–	–	–	0.5511	–	0.24 (0.00)	–	17.58 (0.06)	–	–
Ethyl decanoate	1396	–	–	–	–	–	2.9641	–	–	–	–	–
Tetradecane	1400	–	–	–	5.50 (0.02)	–	–	0.30 (0.00)	–	0.82 (0.00)	0.28 (0.00)	–
Dodecanal	1408	–	–	–	–	–	1.1590	–	–	5.45 (0.03)	–	–
Pentadecane	1500	–	0.82 (0.01)	–	–	–	0.5760	2.02 (0.01)	0.86 (0.01)	20.79 (0.06)	2.43 (0.01)	1.59 (0.01)
Hexadecane	1600	–	0.19 (0.00)	54.95 (0.33)	–	–	–	–	–	0.87 (0.00)	0.32 (0.00)	–
Tetradecanal	1612	–	–	–	–	0.2608	1.7326	–	–	7.74 (0.03)	–	–
Ethyl tetradecanoate	1694	–	–	–	–	–	–	–	–	–	0.57 (0.00)	–
Heptadecane	1700	–	7.29 (0.03)	–	–	0.6182	–	0.74 (0.00)	0.45 (0.00)	5.49 (0.02)	1.62 (0.01)	2.21 (0.02)
Ethyl hexadecanoate	1902	–	–	–	–	–	–	–	–	–	0.15 (0.00)	–

Average relative total floral scent emission rates (ng emitted g^–1 tissue h^–1), Kovats Retention Index (KRI) values and relative abundance of volatile organic compounds (%), organized by biochemical class.

Fig. 2.

Volatile organic compound (VOC) composition of Thalictrum floral scent. (A) GC-MS chromatogram of floral scent extract from a representative wind- (T. dioicum, male individual) and insect- (T. lucidum) pollinated plant. Labelled peaks correspond to: (a) (Z)-hex-3-en-1-ol; (b) (Z)-hex-3-en-1-ol, acetate; (c) E-β-ocimene; (d) caryophyllene; and (e) α-farnesene. (B) Multivariate analysis of floral VOCs in 11 species of Thalictrum, eight insect- and three wind-pollinated species, by non-metric multidimensional scaling (NMDS) with a stress value of 0.18254. Staminate (s) and carpellate (c) samples are indicated for dioecious species. Black arrows represent overlaid bi-plot analysis of predominant compounds along each axis.

There was overall no significant difference in volatile profiles between wind- and insect-pollinated species (R = –0.0376, P = 0.26). Volatile profiles were distinct among species; however, insect-pollinated T. lucidum, for example, had typically more terpene volatiles and at a higher proportion in the scent than male flowers of wind-pollinated T. dioicum (Fig. 2A; Table 1), even though there was considerable variation among individuals within a species (Fig. 2B; Table 1). To better capture the variation generated by the 51 compounds across the 11 species, we conducted a multivariate analysis (NMDS). From this analysis, there was a highly significant difference among species (ANOSIM: R = 0.5378, P = 0.001), with individual species loosely clustered in multidimensional space.

A qualitative assessment of organ-specific VOC composition in T. thalictroides flowers showed differences between organs (Supplementary Data Table S4). Whereas the carpel and sepal were dominated by monoterpenes (100 and 96 %, respectively), the stamens were dominated by aromatic compounds, comprising 83 % of the headspace. Dominant monoterpenes in the carpel and sepal were d-limonene and β-myrcene, and to a lesser extent trans-β-ocimene, all of which are capable of eliciting a significant antennal response (Supplementary Data Fig. S2). Aromatics in the stamen were methyl benzoate and anisole. Sesquiterpenes, which comprised 36 % of the total floral emissions (Table 1), were not identified using the SPME collection method, which may be attributable to their low volatility or the fibre type used in the collection.

Phylogenetic context of VOC composition

To determine if there was phylogenetic signal in the composition of floral scent, we compared the patristic distances among the taxa from a pruned version of our Thalictrum phylogeny (Fig. 1) and compared these taxonomic distances with the Bray–Curtis dissimilarities among VOCs, as calculated from our multivariate analysis (Fig. 2B). The phylogenetic signal was not significant (Mantel test: 0.12; P = 0.26), suggesting that relatedness alone is unlikely to explain patterns in floral scent. This result was corroborated by a test of phylogenetic signal by Blomberg’s K under Brownian motion, whereby K <1 for all individual compounds indicated that no single compound had a strong phylogenetic signal. Similarly, both implementations of multivariate Blomberg’s K resulted in K <1 (K <0.26; P > 0.818). In summary, all analyses coincided in finding that floral scent data do not exhibit phylogenetic signal.

Electroantennograms of floral extracts

To evaluate whether whole-scent floral extracts elicit differential olfactory responses in potential insect pollinators, EAGs were performed on floral fragrances using the model bumble-bee B. impatiens. Thalictrum floral extracts from species with distinct morphologies, representative of the insect and wind pollination modes, elicited highly significantly different antennal responses in B. impatiens (Fig. 3A; Kruskal–Wallis test: χ_8,53 = 31.89, P < 0.0001). Moreover, four of the species with insect pollination floral morphology, T. delavayi, T. flavum, T. aquilegiifolium and T. thalictroides, elicited significantly greater antennal response than the hexane negative control (Kruskal–Wallis test with multiple comparisons: P = 0.0012–0.040); T. rochebrunnianum was the exception (Fig. 3B; P = 0.41). In contrast, the three species with wind pollination floral morphology (T. hernandezii, T. dioicum and T. dasycarpum) were not significantly different from the negative control (Fig. 3B; Kruskal–Wallis test with multiple comparisons: P = 0.30–0.99). Comparison of antennal responses between insect- and wind-pollinated extracts showed that insect-pollinated extracts elicited significantly greater responses (Fig. 3C; Kruskal–Wallis test: χ_1,47 = 12.2, P = 0.00047). When EAG responses to single compounds were examined, weχχ found that β-ocimene elicited a significant antennal response (Kruskal–Wallis test with multiple comparisons: P = 0.047) (Supplementary Data Fig. S2). The aromatic benzaldehyde triggered a significant antennal response (P < 0.0001), and terpenes that elicited strong responses include the sesquiterpene α-farnesene (abundant in insect-pollinated T. thalictroides; P < 0.0001), β-myrcene and d-limonene also elicited significant EAG responses (P = 0.02 and 0.03, respectively).

DISCUSSION

Our results suggest that Thalictrum floral VOCs can elicit a significant response from potential pollinators and that this is likely to be the ancestral condition for the genus. With the subsequent evolution of wind pollination, floral VOCs appear to have diverged, such that they are no longer able to elicit a detectable neurophysiological response from potential pollinators. Interestingly, we find at least one case where insect response was regained in a species that evolved from wind-pollinated ancestors, suggesting that floral scent is fairly labile in Thalictrum. Species had significantly distinct volatile profiles independent of phylogenetic relationships, yet there was no clear separation by pollination mode in multivariate space. Therefore, this system does not appear to have a ‘signature’ bouquet for each pollination mode. We hypothesize that different insect-attracting compounds may have been lost during each independent transition to wind pollination (Figs 2 and 3).

The increased support along the backbone of the revised phylogeny significantly improved our ability to understand the evolution of wind pollination in Thalictrum. We were able to assign insect pollination unambiguously to the ancestor of the genus and to one of the major clades, reconstruct the other major clade as ancestrally insect pollinated with a high proportional likelihood and identify two instances of reversal to insect pollination (Fig. 1). Two new, strongly supported clades were identified, suggesting that the species added in this study significantly increased the breadth of taxonomic coverage within Thalictrum. Taxa from previous analyses that had to be excluded here due to lack of chloroplast data were for the most part nested in large, well-supported clades, and had no significant effect on broad topology. Moreover, the revised phylogeny allowed for a more accurate calculation of taxonomic distances to test for phylogenetic signal, which would have otherwise been skewed by the presence of unresolved polytomies in the backbone of the tree (W. Maddison, pers. comm.).

The actual pollination mode is probably more nuanced than a simple bi-state character. Ambophily, or the combination of insect and wind pollination, potentially plays a role in Thalictrum, especially in species with intermediate flower morphology (and with PIs close to two), as seen in T. pubescens (syn. T. polygamum; Kaplan and Mulcahy, 1971). Another interesting case is T. flavum, which falls in the ‘insect’ category by a small margin based on morphology. This species is embedded in a clade that was reconstructed as ancestrally wind pollinated (PL > 0.875; Fig. 1), representing a potential reversal, and its floral scent significantly triggered an antennal response, while that of wind-pollinated T. dasycarpum in the same clade did not (Fig. 3B). It remains unclear whether a potential combined wind and insect pollination strategy represents a stable or a transitional state (Friedman, 2011). Testing the extent to which Thalictrum spp. are able to take advantage of a combined insect and wind pollination strategy is an interesting avenue of future research that will require evidence from exclusion experiments in the field.

Given that Thalictrum flowers are small and weakly scented, we did not have a strong expectation for floral scent being a dominant character. In fact, the genus has floral exaptations, such as loss of petals, lack of nectar and free, uniovulate carpels that potentially predispose it to the vestigialization of other insect pollination-related traits, facilitating a shift to wind pollination. Nevertheless, just as shown for certain penstemons (Parachnowitsch et al., 2013), and with one exception, there was consistency between the more ‘showy’ floral morphologies and their potential ability to attract pollinators via floral scent.

Out of 51 floral compounds found among the 11 species studied, we presented evidence for at least five single compounds that significantly triggered an antennal response. Our multidimensional floral compound analysis suggests that a few compounds correlate with the separation of species in multidimensional space, and this is consistent with single to a few key compounds affecting differential pollinator attraction in other systems (Klahre et al., 2011; Parachnowitsch et al., 2013; Byers et al., 2014). In addition, we found that compounds present at high levels in the floral scent of certain insect-pollinated Thalictrum (e.g. β-myrcene, d-limonene, β-ocimene and α-farnesene) had significant EAG responses when tested in isolation (Table 1) and are therefore good candidates for future studies. In particular, β-ocimene is common in other bee-pollinated flowers (e.g. Byers et al., 2014) and known to behave as a bee pheromone (Granero et al., 2005; Maisonnasse et al., 2010). Another interesting compound was the aromatic benzaldehyde – a precursor to many oxygenated aromatic compounds (Long et al., 2009) – which was present in high amounts in one wind-pollinated species and the two species with intermediate morphologies representing reversals to insect, or perhaps mixed pollination mode (ambophily), but absent in other insect-pollinated species.

The presence of potentially attractive compounds in wind-pollinated taxa could result from our evidence that the whole genus and the two main clades are ancestrally insect pollinated, or perhaps certain putative wind-pollinated species are in fact ambophilous; a similar scenario has been proposed for wind-pollinated Schiedea (Juergens et al., 2012). Comparable studies in Leucadendron (Welsford et al., 2016), in sedges for the opposite transition (wind to insect pollination; Wragg and Johnson, 2011) and across an analogous transition from outcrossing to selfing, have found fewer compounds and lower emission rates in wind-pollinated or autogamous taxa (Doubleday et al., 2013). In our system, as in the Schiedea study, there was no obvious trend in the number of floral VOC compounds or emission rates between insect- and wind-pollinated Thalictrum.

There was less phylogenetic signal for the entire floral scent dataset than expected under Brownian motion (Blomberg’s K = 0.24). Ancestral state character reconstruction suggests that the ability to elicit a significant response in a potential pollinator evolved at least twice (Fig. 4), once in the clade that contains T. thalictroides (Clade I, Fig. 1) and again in the clade that contains T. aquilegiifolium. The former clade is exclusively insect pollinated, and within the latter clade there were at least two losses in the ability to elicit a significant response in Bombus: in the evolutionary branches leading to T. dasycarpum and T. rochebrunianum. Alternatively, there could have been three independent gains, but this explanation is less parsimonious. Broader taxonomic sampling will be needed to ascertain the number of gains and/or losses in the ability of floral scent to elicit a response from a potential pollinator, and physiological tests in other more prevalent pollinators, such as bee flies, would very informative.

Fig. 4.

Ancestral state reconstruction of bumble-bee antennal responses to whole flower scent extracts of eight Thalictrum species. Electroantennogram response (EAG, in mV) was mapped as a continuous character (from Fig. 3) on a pruned phylogeny of Thalictrum and maximum likelihood reconstruction of insect (white) and wind (black) pollination mode overlaid as pie charts (from Fig. 1). Red stars indicate taxa with floral scent that elicited statistically significant EAG responses in Bombus impatiens relative to a negative hexane control (P < 0.05).

In summary, our results imply that Thalictrum species have distinct floral bouquets that may be more driven by selective pressures for reproduction (pollinators) and/or potentially for survival (as repellents for antagonists; e.g. Parachnowitsch et al., 2013), than by phylogenetic structure. Floral scent bouquets from insect-pollinated species interact more strongly with a potential pollinator than those from wind-pollinated species (Fig. 3), suggesting that the loss of certain VOCs probably accompanied the multiple transitions to wind pollination in this system. The putative loss of key response compounds probably happened independently and differently, so that a ‘scent signature’ could not be detected for species with convergent, wind-driven floral morphologies. Reversions to insect pollination, with a regained ability to attract putative pollinators, although typically less likely, appear possible in Thalictrum. We hope that our results contribute to bridging the gap between evolutionary studies of pollination mode and insect physiology by providing evidence for diverse floral scents differentially affecting the response of a potential pollinator.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Figure S1: a revised phylogeny of Thalictrum (Ranunculaceae). Figure S2: Bombus impatiens antennal responses to single volatiles found in Thalictrum floral scent. Table S1: GenBank accession numbers for species included in the phylogenetic analyses. Table S2: pollination mode data for ancestral character state reconstructions. Table S3: voucher specimen information for taxa sampled for floral scent. Table S4: relative composition of flower organ-specific scent emissions in Thalictrum thalictroides. Table S5: fragrance composition of staminate (male) and carpellate (female) flowers from dioecious (and wind-pollinated) Thalictrum. Video: pollen-collecting activity of Bombus terrestris on T. thalictroides plants in an experimental setting, demonstrating the potential of Bombus sp. as a pollinator proxy.