Range-wide floral trait variation reflects shifts in pollinator assemblages, consistent with pollinator-mediated divergence despite generalized visitation

Abstract

Floral trait evolution mediated by pollinators is important in the diversification of flowering plants, yet few studies have demonstrated the range-wide geographic variation in both floral traits and pollinators which represents a predicted precursor for pollinator-mediated speciation. This study explores whether geographic variation in pollinator interactions underlies the observed patterns of floral divergence both 1) among species of the Castilleja purpurea complex (C. purpurea, C. citrina and C. lindheimeri) and the congener C. sessiliflora, as well as 2) within C. sessiliflora, across its wide geographic range. We sampled floral visitors and floral traits (morphology and color) at 23 populations across a 1900 km-wide study area in 1–3 years, with reproductive fitness (fruit set) data for 18 of these populations. A wide diversity of pollinator functional groups visited the focal species, including bees, butterflies, hawkmoths and hummingbirds, and visitor assemblages varied among species and across geography. We identified relationships between floral traits and visitation by certain pollinator groups, which often aligned with predictions based on pollination syndromes. Despite visitor assemblages being largely generalized across most populations, we found that the observed changes in floral traits were associated with shifts in the relative frequencies of key pollinator functional groups. Hence this study demonstrates that variation in pollinator assemblages across the distributions of taxa may underlie divergence in floral traits and suggests that highly specialized relationships may not be required for early stages of pollinator-mediated floral divergence. Our extensive sampling of 23 populations over multiple years across a large geographic area highlights the value of range-wide studies for characterizing patterns of divergence mediated by ecological interactions.

Introduction

The role of animal pollinators in driving angiosperm diversification is well-studied (Johnson 2006), with evidence from macroevolutionary (Crepet 1984, Lunau 2004, van der Niet and Johnson 2012), microevolutionary (Bradshaw and Schemske 2003, Gervasi and Schiestl 2017) and ecological approaches (Campbell et al. 1997, Fulton and Hodges 1999). Models of pollinator-mediated plant speciation often invoke how geographic variation relates to processes of divergence (Grant 1949, Stebbins 1970, Johnson 2006), as the locally most effective pollinator (Stebbins 1970) is expected to exert selection on floral traits, based on its morphology, physiology and foraging behavior (Schiestl and Johnson 2013). The identity of the most effective pollinator is expected to vary across species distributions (Thompson 2005), along with its ecological context and fitness contributions (Kay and Sargent 2009, van der Niet et al. 2014, Ohashi et al. 2021), resulting in geographic variation that may lead to divergence in reproductive traits. Despite this, many studies of pollinator-mediated selection on floral traits focus on one or a few populations in zones of contact (Bradshaw and Schemske 2003, Campbell and Aldridge 2006, Hopkins and Rausher 2012), resulting in range-wide geographic variation being understudied (Herrera et al. 2006), though attention to geographic mosaics of interactions has increased in recent years (Anderson and Johnson 2008, Boberg et al. 2014, Szenteczki et al. 2021, Ellis et al. 2021, Johnson et al. 2021). Despite interest in connecting microevolutionary population processes to macroevolutionary patterns of pollinator-mediated diversification (van der Niet et al. 2014), few studies have focused on how intraspecific floral variation relates to geographic divergence in pollinators, which is critical for identifying the capacity for future pollinator-mediated speciation (Herrera et al. 2006, Kay and Sargent 2009). Intraspecific studies that explore geographic variation and early stages of divergence are needed to better understand how pollinator identity and shifts between pollinators may or may not lead to reproductive isolation and plant speciation (Waser and Campbell 2004, Kay and Sargent 2009, van der Niet et al. 2014).

In exploring how phenotypic divergence in floral traits may reflect selection mediated by distinct pollinators, pollination syndromes (suites of floral traits associated with specific pollinator functional groups) are thought to represent convergent evolution in plant–pollinator interactions (Faegri and van der Pijl 1971, Fenster et al. 2004, Rosas-Guerrero et al. 2014). Despite criticism of overreliance on the syndrome concept (Waser et al. 1996, Ollerton et al. 2009), pollination syndromes provide a source of a priori hypotheses for associations between certain floral traits and specific pollinator functional groups, which are often supported by detailed observations of floral visitors (Dellinger 2020). While phenotypic divergence at large geographic scales is likely to occur in concert with other sources of ecological variation (Johnson 2006, Strauss and Whittall 2006, Nosil 2012), selection favoring one guild of pollinators can shape both trait divergence and reproductive isolation, given pollinators' direct influence on selection (via reproductive fitness) and gene flow (via pollen movement) of flowering plants (Waser and Campbell 2004). When selection for the most effective pollinator guild confers higher seed set by one group over another, shifts between pollinators can result in reproductive isolation between populations. Nonetheless, recent work argues that ‘adaptive generalization' of broader visitor assemblages could also result in phenotypic divergence (Ohashi et al. 2021), possibly without reproductive isolation, raising questions about the degree of specialization that is required for pollinator-mediated speciation to occur.

Many well-known examples of strong reproductive isolation due to pollinator shifts focus on allopatrically-diverged species experiencing secondary contact at the edges of their ecological tolerances (Campbell et al. 1997, Fulton and Hodges 1999, Bradshaw and Schemske 2003) and thus may reveal how reproductive isolation is maintained, but not necessarily how it first arises. Understanding whether differences in pollinator identity and associated divergence in floral traits are sufficient to give rise to reproductive isolation from a continuum of intraspecific variation remains an important question in evolutionary ecology. To address this, studies must first characterize geographic divergence in floral traits and associated variation in pollinators, both within a taxon and among recently diverged taxa. Characterizing early stages of floral phenotype divergence is critical to inform our understanding of the conditions that may or may not result in reproductive isolation and the role of pollinator shifts therein. Of particular importance are range-wide studies of variation in local pollinators among recently diverged species or pollination ecotypes, as they may represent recent or incipient speciation (Pellmyr 1986, Oyama et al. 2010, van der Niet et al. 2014, Sobel and Streisfeld 2015), a key stage in the process of floral divergence that remains poorly understood (Kay and Sargent 2009) and may precede strong reproductive isolating barriers.

Increasing evidence suggests that strong selection can drive phenotypic divergence rapidly and despite ongoing gene flow, especially in cases of ecological speciation (Nosil 2008, Tavares et al. 2018, Schluter and Rieseberg 2022). Given that strong reproductive isolation (e.g. due to genomic incompatibilities and/or hybrid inviability) is expected to take considerable time in isolation to arise, early or incomplete stages of ecological speciation may be more likely characterized by phenotypic divergence, driven by strong recent selection, with weak reproductive isolation and ongoing gene flow. Evidence for this scenario has been reported in systems with phenotypic divergence in reproductive traits despite little or no genetic differentiation (Nosil and Crespi 2004, Mason and Taylor 2015, Harris et al. 2018), notably including examples of pollinator-mediated selection on floral traits (Streisfeld and Kohn 2005, Whibley et al. 2006, Hopkins et al. 2012, Stankowski et al. 2017), suggesting recent pollinator-mediated divergence may result in a such a pattern.

Recent work on the species of the Castilleja purpurea species complex and its congener C. sessiliflora demonstrated high levels of floral trait variation among and within species, respectively, despite low genetic differentiation across these groups (Wenzell et al. 2021). Castilleja sessiliflora displays geographic variation in floral traits across its wide range, with much of this variation concentrated in inflorescence color: white-green to pale pink inflorescences from north to south, with distinct bright pink and yellow morphs in the southern range extent (Fig. 1). In contrast, the three species of the C. purpurea complex, recently elevated to species status, have relatively small, overlapping ranges and vary primarily in inflorescence color: C. purpurea has purple bracts, C. citrina has yellow bracts, and C. lindheimeri has red-orange bracts (Nesom and Egger 2014). Despite these striking differences in color, the species are characterized by low levels of differentiation at targeted genomic loci, even across narrow clines in floral color (Wenzell et al. 2021), suggesting these species are recently diverged, possibly due to selection on floral color that could be mediated by pollinators. Because these taxa demonstrate a stunning diversity of floral color (Fig. 1) despite little genetic differentiation, they provide an ideal system to examine how geographic variation in floral traits within and among recently diverged species relates to shifts in local pollinators across range-wide scales, to inform our understanding of early stages of potential pollinator-mediated floral divergence.

Figure 1

Geographic range (A) and floral trait divergence within Castilleja sessiliflora (B–F; green range on map) and among species of the C. purpurea complex: C. citrina (G, yellow range), C. purpurea (H, purple range), and C. lindheimeri (I, orange range). Floral trait variation within C. sessiliflora: a grade of white-green inflorescences in the northeastern range extent (B) to pale pink in the southwest (C, D), and two distinct floral morphs in the southern range, with shorter corollas and bright yellow (E, population SMP) or pink (F, SIC) inflorescence color (Wenzell et al. 2021). Photos by K. Wenzell.

In this study, we characterize patterns of pollinator visitation range-wide to investigate whether floral divergence may be driven by selection mediated by pollinators. We address the research question: does a mosaic of pollinator visitation underlie the patterns of phenotypic divergence seen in floral traits for C. sessiliflora and the C. purpurea complex? We assess this question both for among-species floral divergence (i.e. among all four study species: C. purpurea, C. citrina, C. lindheimeri and C. sessiliflora), as well as for within-species floral divergence, focusing on variation across the wide range of C. sessiliflora. If a pollinator mosaic is driving floral divergence, we expect to find evidence for the following predictions: 1) pollinator assemblages and visitation vary among species and across geography, 2) plant fitness (reproductive success) varies among species and across geography and 3) visitation from different pollinator functional groups is associated with variation in floral traits likely to be important for pollination, consistent with potential selection on these traits that could be mediated by pollinators. To test these predictions, we collected data on the composition of floral visitors to 23 natural populations across 1–3 years, in combination with data on floral traits and female fitness (fruit set) across the range of each of the four species. This study unites components of pollinator-mediated selection on floral traits to inform our understanding of how these factors interact across wide geographic scales, thus shaping floral divergence, a potential first step in pollinator-mediated speciation.

Material and methods

Study system

The genus Castilleja (paintbrushes, Orobanchaceae) is hemiparasitic and known for variability in color and morphology of flowers and showy floral bracts, which is often attributed to putative hybridization or the retention of ancestral polymorphisms following rapid radiation (Tank and Olmstead 2008). This study focuses on four perennial species characterized by diverse floral traits: the widespread C. sessiliflora, which is distributed throughout much of central North America, and the species of the more geographically restricted C. purpurea complex: C. purpurea, C. citrina and C. lindheimeri, which occur primarily in Texas and Oklahoma, USA (Fig. 1). Given their geographic proximity and morphological similarity, C. sessiliflora and the C. purpurea complex are expected to be close relatives (D. Tank pers. comm.). These species are largely self-incompatible, and flowers remain open for several days and offer both pollen and nectar rewards (K. Wenzell unpubl.).

Data collection

Floral traits

Floral traits were measured at 23 focal populations and included five morphological traits (corolla length, corolla width, petaloid lip length, stigma exsertion and bract lobe width) and inflorescence color (Royal Horticultural Society (RHS) floral color charts). These traits are expected to influence pollinator attraction, access to floral rewards, and pollination effectiveness, and have been previously studied in these groups (Crosswhite and Crosswhite 1970, Wenzell et al. 2021). Traits were measured from two flowers each of 30 plants in a single year at each population and are described in greater detail in Wenzell et al. (2021). To quantify floral color, RHS color codes were converted to red–green–blue (RGB) values, followed by nonmetric multidimensional scaling (NMDS) of RGB values, using Gower distances in R package vegan (Oksanen et al. 2019), to make color values more easily interpretable. The resulting NMDS axes were associated with each component of RGB color using the envfit function and represent variation in inflorescence color throughout the study (Supporting information). Overall, the NMDS1 axis was characterized by warm colors (red, orange, yellow) at low values and cool colors (purple, pink) at high values. NMDS2 was roughly characterized by darker, reddish colors (purple, red) at high values and paler, greenish colors (green, yellow-green, yellow) at low values.

Sampling of pollinator observations

Pollinator observations were conducted for 1–3 years from 2017 to 2019 at 23 populations distributed across the range of each species (12 populations of C. sessiliflora and 11 populations representing the C. purpurea species complex: four populations of C. purpurea, four populations of C. citrina and three populations of C. lindheimeri; Supporting information), spanning approximately 1900 km. These 23 focal populations were selected to thoroughly sample the species geographic distributions and to capture the range of phenotypic variability of each species (Wenzell et al. 2021). Due to the challenge of sampling such a large geographic area, pollinator observations were conducted over one 24 h period per population per year. While this may limit detection of variability in visitors throughout the season, populations were sampled as close to peak flowering time as possible, and most populations were sampled in multiple years (though see exceptions below), which provides some measure of temporal variability. Additionally, due to the scale of the sampling area, not every population could be sampled in each year: specifically, three C. sessiliflora populations in the center of the range (SDC, SPB and SRS) were sampled only in 2017, and one population of each of the C. purpurea complex species (CHCL, LMN and PTMS) were sampled only in 2019. Furthermore, C. sessiliflora was the only species sampled in 2017 (Table 1). Complete information on sampling years and datasets for each population are given in the Supporting information.

Table 1. Floral visitation summary table. Number of recorded floral visits to each plant species for each pollinator functional group. For each plant species, the number of sampled populations, years sampled, and the number of observation datapoints (defined as observations of pollinator visitation to a given population in a given year for each dataset type, total n = 61) is given, along with total number of visits and average number of visits per datapoint (calculated as total visits/datapoints per species)

	Small/med bee	Large bee	Hawkmoth	Hummingbird	Butterfly	Other	Total visits	Average visits per datapoint
C. citrina	85	268	49	63	53	1	519	47.2
Populations: 4
Years: 2018, 2019
Datapoints: 11
C. lindheimeri	4	0	20	430	115	2	571	71.4
Populations: 3
Years: 2018, 2019
Datapoints: 8
C. purpurea	248	85	409	144	124	42	1052	95.6
Populations: 4
Years: 2018, 2019
Datapoints: 11
C. sessiliflora	698	317	683	0	105	67	1870	60.3
Populations: 12
Years: 2017, 2018, 2019
Datapoints: 31

At each population, two observers recorded floral visits simultaneously during eight 20 min observation periods during daylight hours (evenly spaced throughout periods of pollinator activity) and one 60 min observation period at dusk, which started 15 min before sunset. This sampling resulted in 440 observer minutes recorded per site per year, for a total of 330 h of observation time across all sites and years. Floral visitors were recorded to pollinator functional group (Fenster et al. 2004) and were analyzed as such, though genus (e.g. Bombus sp.) or species (e.g. Hyles lineata) was noted when possible. Pollinator functional groups included: hawkmoths (Sphingidae), hummingbirds (Trochilidae), bumblebees (Bombus sp.), other large bees (approximately 2.5 cm or larger in body length, such as carpenter bees), medium bees (approximately 1.5 cm in length), small bees (approximately < 1 cm in length, including sweat bees such as Lassioglossum sp., and other pollen-collecting bees), bee flies (Bombyliidae), non-Sphingid Lepidoptera categorized as large butterflies (approximately > 4 cm in wing height, including swallowtail butterflies such as Battus philenor and others), small butterflies (approximately < 3 cm) or other moths and flies (Diptera). For analysis, infrequent groups (those recorded visiting < 50 flowers in the total dataset or recorded at only one population) were pooled with more common, functionally similar groups or into a group of ‘other' visitors. This latter category includes mainly bee flies, flies and non-Sphingid moths. Final analyses included the following pollinator functional groups: hawkmoths, hummingbirds, bumblebees and large bees, small and medium bees, butterflies and other visitors.

In all years (2017–2019), floral visitation data were recorded in the following way: a focal patch of flowering Castilleja plants was designated within approximately 1–2 m of the observer (near enough to observe and record visitation by sweat bees or other small insects), and visitation to flowers of these focal plants was recorded during all observation periods, which represents the ‘narrow-view dataset.' The number of open flowers on each focal plant was recorded for each day of observations. However, it was noted that certain pollinator functional groups (especially hummingbirds) were wary of approaching plants close to observers, which had the effect of excluding them from this dataset. To account for this, an additional, second visitation dataset was collected in 2019. For this complementary ‘wide-view dataset', observers recorded floral visits to any Castilleja plants in their field of view that occurred during observation periods. The approximate number of flowering stems of Castilleja in this wide field of view was counted to generate an estimate of the number of open flowers available to pollinators (calculated as the average number of open flowers per focal plant, multiplied by the number of flowering stems in the wide view). Visits to narrow-view focal plants were not included in an observers' wide-view dataset, and care was taken to avoid observers sharing the same wide view, to prevent double-counting of floral visits. Given the nature of this observation method, it is possible that this dataset is biased toward the detection of large pollinators (i.e. hummingbirds, butterflies and large bees) and against small pollinators such as small bees. For this reason, floral visitation data from both observation methods (referred to as ‘dataset types' hereafter) are provided as complementary datasets, and both are included in analyses, with dataset type as a fixed effect in our models to account for the potential impact of these different observation methods.

For both methods, observers recorded the number of flowers visited by a given pollinator during an observation period when a flower was visibly probed and the visitor appeared to contact the anthers and/or stigma. Number of floral visits by each pollinator functional group was pooled across observation periods for a given population-year, and observation periods were an equal amount of time for each population-year. Data were calculated as count (number of floral visits by pollinator group per population-year), visitation rate (number of floral visits relative to the total number of open flowers available in the narrow or wide view, depending on dataset, per hour), and as proportion of all visits (number of floral visits by one pollinator group relative to the total number of visits from all pollinator groups per population-year; Supporting information). Because sampling time (observer hours) was equal across population-years, and to compare relative contribution of different pollinator functional groups across populations, subsequent analyses use proportion data unless otherwise noted. Calculations and analyses were performed in R (www.r-project.org) using tidyverse packages (Wickham et al. 2019).

Plant fitness measurements

To characterize the reproductive fitness of focal populations, we measured fruit to flower ratio in a subset of 18 populations (Supporting information), based on approximately 30 plants per population. Some populations were sampled in multiple years for a total of 20 population-year observations and a total sample size of 591 plants. Fruit set was measured as the total number of filled fruits (or enlarging ovaries if populations were sampled prior to fruit maturation) divided by the number of total flower nodes (1 flower/node) along one or more flowering stems per individual, i.e. the proportion of filled fruits per individual. This fruit set value was used as an estimate of plant fitness.

Variation in key floral traits among species and populations

To establish overall variation of floral traits before proceeding with tests of individual traits, we performed a MANOVA with all seven measured traits using the manova() function in R, testing for multivariate variation among species and among populations nested within species, with individual plant as the sampling unit (n = 684). Next, we identified which individual floral traits varied among species and among populations (nested within species) by comparing mean values of each trait using ANOVA (aov() function). For traits that varied significantly among species, we then used post hoc Tukey HSD tests (TukeyHSD() with 95% CI) to make pairwise comparisons among species, which apply Tukey adjustments to control for multiple testing.

How does pollinator visitation vary among plant species and across geography?

We first assessed how the composition of pollinator assemblages varied among and within species, by testing for multivariate variation in floral visitors (incorporating all pollinator functional groups) using a permutational MANOVA (PERMANOVA) with the adonis2() function in R package vegan (Oksanen et al. 2019). This model assessed proportion of visits from all six pollinator groups (based on a dissimilarity matrix generated using the vegdist() function with Gower distances, which handles zeroes in the dataset) in relation to species, population nested within species, as well as year and dataset type, with 1000 permutations. Because we were also interested in the contribution to pollination of individual pollinator functional groups among our focal species, we then tested for variation of each functional group separately. We assessed whether the proportion of total visits by each pollinator functional group varied among plant species using generalized linear mixed models (GLMMs) in R package glmmTMB (Magnusson et al. 2020), followed by type II Wald χ² tests of the model output, using function Anova() in package car (Fox et al. 2020). Multiple comparison analyses were then performed with function lsmeans() in package lsmeans (Lenth 2018), using pairwise comparisons among plant species with Tukey adjustments, which control for the family-wise error rate with multiple pairwise comparisons. Separate models were run for each pollinator functional group, with proportion of visits calculated as individual datapoints for each population-year sampled and each dataset type (n = 61), with dataset type and observation year included as fixed effects. Proportion of visits for each pollinator type was the response, weighted by total number of visits, with plant species as the predictor, and a beta-binomial error distribution. To assess potential overdispersion in the data, we ran models using both a binomial and beta-binomial error distribution and used package DHARMa (Hartig 2016) to examine residuals and assess evidence for overdispersion in each model. The models with binomial distributions showed greater evidence of overdispersion than those using beta-binomial distributions, and thus the latter were chosen for models of proportion data throughout the study. Our analyses focused on proportion data because these values allow us to compare the relative contributions of pollinator functional groups and the overall composition of pollinator assemblages across populations, species, and geographic ranges. However, because count data may reveal different patterns, we assessed models with count data as well. Results using count data were largely consistent with those using proportion data, so hereafter we present only the proportion data and results, though detailed methods and results of analyses using count data are presented in the Supporting Information.

For each population, we visualized the overall proportion of visits across years, calculated as the total number of visits from each pollinator group, relative to the total number of visits (following Crawley 2015, p. 257; Fig. 2; note that this was done only for visualizations; all statistical models use separate proportion values recorded in each observation year). We performed statistical analyses of geographic variation in pollinator visitation only in the widespread C. sessiliflora, given the more restricted geographic ranges and limited number of sampled populations for the species of the C. purpurea complex. For these tests, we performed GLMMs as described above with proportion of visits of each pollinator type as the response term (separate models for each pollinator group) and population latitude as predictor, along with dataset type and year as fixed effects. Latitude was used as a proxy for geographic location across the range of C. sessiliflora, given the large latitudinal spread of sampled populations, to assess geographic variation within this species.

Figure 2

Geographic mosaic of floral color and floral visitors across the ranges of C. sessiliflora and the C. purpurea complex. Shapes on map show 23 sampled populations of C. sessiliflora (circles), C. purpurea (squares), C. citrina (diamonds), and C. lindheimeri (triangles); fill color is population median floral color (median RGB values; n = 684). Pie charts show overall proportion of visits by pollinator functional group to populations across years. Photos show selected floral visitors: (A) small sweat bee (Halictidae) collecting pollen on C. sessiliflora at SNG; (B) bumblebee Bombus fervidus nectaring at SNG (photo: J. Fant); (C) hawkmoth Hyles lineata in Colorado, near SDC (photo: S. Todd); (D) bumblebee Bombus sonorus on the yellow floral morph at SMP (photo: S. Deans); (E) hawkmoth H. lineata and (F) butterfly (black swallowtail Papilio polyxenes) on the pink floral morph at SIC (photo: S. Deans); (G) bumblebee B. pensylvanicus and (H) small bee on C. citrina at CQL; (I) hummingbird at LRR and (J) pipevine swallowtail butterfly Battus philenor on C. lindheimeri; (K) hawkmoth H. lineata visiting C. purpurea at PTH; (L) pipevine swallowtail butterfly B. philenor and (M) bumblebee Bombus pensylvanicus at PCM. Photos by K. Wenzell unless otherwise noted.

Does plant fitness vary among species and across geography?

To assess whether reproductive success varied among species and across geography, we tested for variation in fruit set among species using GLMMs with individual-level fruit set as the response and species as the predictor. Models were run with a beta-binomial distribution weighted by number of flowers, with population included as a random effect and year as a fixed effect. Additionally, to assess geographic variation in fruit set across the wide range of C. sessiliflora, we performed a GLMM as described above using population latitude (a proxy for geography) as a predictor, and year as a fixed effect.

Are certain floral traits associated with visitation from different pollinator groups?

To investigate how floral traits may relate to visitation from different pollinator functional groups across taxonomic boundaries, we performed multiple regression analyses using pooled floral trait data for all four species. We ran a separate GLMM for each pollinator functional group, with proportion of visits as the response variable and population mean values of all seven floral traits as predictors (scaled using the z transformation with the scale() function in tidyverse), along with dataset type and year as fixed effects. Floral trait values were not measured in each year sampled for pollinator visitation, due to time constraints and because measurements were not expected to vary widely among years in these perennial species, so single-year floral trait measurements are taken as population mean values in these models. Models were performed with beta-binomial distributions, weighted by total number of visits.

Finally, to assess how suites of floral traits may be associated with visitation from certain pollinator groups, we plotted an NMDS ordination of population mean floral trait values (as described above) and used the envfit() function in package vegan (Oksanen et al. 2019) to assess whether proportion of visits from all six pollinator groups were significantly associated with multivariate floral trait variation, based on 1000 permutations.

Results

Floral traits vary among and within species

Our MANOVA revealed significant variation across all floral traits among species (approximate F_3,21,661 = 351.03; p < 0.001) and among populations nested within species (approximate F_19,133,661 = 14.67; p < 0.001). For individual floral traits, all traits varied significantly among species and among populations nested within species (Supporting information), and pairwise multiple comparisons revealed variation in key traits between species pairs (Supporting information). Overall, the corollas of Castilleja sessiliflora were significantly longer than those of the other species, with a prominent floral lip and minimal stigma exertion (Fig. 1, Supporting information). In addition, the bracts subtending these flowers were significantly narrower compared to the other three species and varied in color from green to pale pink from northeast to southwest across the range (Fig. 1B–D), with two distinct populations bearing vivid yellow (Fig. 1E; population SMP) or pink (Fig. 1F; population SIC) inflorescences at the southern range extent. Within the C. purpurea species complex, C. lindheimeri had significantly longer and narrower corollas than either C. purpurea and C. citrina, along with the most exserted stigmas, and its red bracts were the broadest of all taxa. The purple-bracted C. purpurea and yellow-bracted C. citrina showed few morphological differences from each other except for corolla length (shortest in C. citrina of all species) and corolla width, which was widest in C. purpurea (Supporting information).

Visitation by pollinator groups varies among and within species

In total, we recorded 4012 pollinator visits across all sites and years (Table 1). The most common functional groups recorded were hawkmoths (1161 visits), followed by small and medium bees (1035 visits). The next most frequent visitors were bumblebees and large bees (670 visits), hummingbirds (637 visits) and finally butterflies (397 visits). Visits from other functional groups (mainly non-Sphingid moths, bee-flies and flies) made up only a small number of visits (112 visits) and were pooled in the ‘other' category. Our PERMANOVA found significant variation among pollinator assemblages (proportion of visits from all pollinator functional groups) both among species (F_3,35,60 = 8.53; p < 0.001) and among populations nested within species (F_19,35,60 = 2.27; p < 0.001), as well as among dataset types (F_1,35,60 = 7.45; p < 0.001) and years (F_2,35,60 = 2.2; p = 0.04). For individual pollinator functional groups, we found evidence that visitation from hummingbirds (species: χ²_3,53 = 16.9, p < 0.001; year: χ²_2,53 = 0.8, p = 0.67; dataset: χ²_1,53 = 3.4, p = 0.06) and butterflies (species: χ²_3,53 = 11.5, p = 0.009; year: χ²_2,53 = 2.0, p = 0.36; dataset: χ²_1,53 = 2.7, p = 0.1) varied significantly among species (Fig. 3A). Hummingbirds were the most frequent visitor to C. lindheimeri and contributed significantly more visits to this species than to C. citrina and C. purpurea (Supporting information), while zero hummingbirds were observed visiting C. sessiliflora (Table 1). Butterflies visited all species at relatively low levels, though C. purpurea had significantly higher butterfly visitation than C. sessiliflora (Supporting information). Small and medium bees were the most common visitor to C. sessiliflora and second most common to C. purpurea (Table 1, Fig. 3A), but variation among species and years was not significant (species: χ²_3,53 = 5.5, p = 0.14; year: χ²_2,53 = 2.9, p = 0.24), though dataset was (χ²_1,53 = 8.5, p = 0.003). Although large bees were the most frequent visitor to C. citrina and zero large bees were recorded visiting C. lindheimeri, differences in their visitation among species were not significant (species: χ²_3,53 = 3.6, p = 0.3; year: χ²_2,53 = 1.8, p = 0.4; dataset: χ²_1,53 = 4.9, p = 0.02). Similarly, hawkmoths were among the most frequent visitors to C. purpurea and C. sessiliflora, but variation among species was not significant (species: χ²_3,53 = 2.5, p = 0.48; year: χ²_2,53 = 0.1, p = 0.94; dataset: χ²_1,53 = 0.3, p = 0.58), nor was variation among other visitors (species: χ²_3,53 = 5.8, p = 0.12; year: χ²_2,53 = 5.4, p = 0.06; dataset: χ²_1,53 = 2.5, p = 0.11). Year effect was not significant in any model, but dataset was significant for models of all pollinator groups except hawkmoths and other visitors, which is likely due to the difficulty of seeing small/medium bees at longer ranges and the effect of close observation on foraging behavior of hummingbirds, etc.

Figure 3

Floral visitation and fruit set by species (sp, C = C. citrina, L = C. lindheimeri, P = C. purpurea, S = C. sessiliflora). (A) Average proportions of floral visits from each pollinator functional group by species. Center bars: median value per species; upper and lower hinges: first and third quartiles; whiskers: points within 1.5 ⨯ IQR of hinges; large points: outlying points. (B) Fruit set (fitness) by species (fruit-to-flower ratio for individual plants, n = 591). Violin plots show density of datapoints, with species mean (points) ± SE (error bars). Asterisks denote significant variation among species following GLMMs (* 0.05 > p > 0.01; ** 0.01 > p > 0.001; *** 0.001 > p).

By sampling 23 populations across our four focal species ranges, we observed considerable variation in pollinator assemblages among populations within species, and we describe these overall patterns here. We observed that all populations of C. citrina were visited by bumblebees (Fig. 2), but the contribution of additional visitors varied by site. The most southern population (CHCL) was also visited by butterflies, hawkmoths and hummingbirds, while the northernmost population (CMSC) was visited predominately by hawkmoths and small bees. For C. lindheimeri, hummingbirds were the most common visitor to all populations, and the only visitor at one site (LMN), though other populations were also visited by butterflies, hawkmoths and occasional small bees. Notably, we recorded no visitation by bumblebees/large bees to C. lindheimeri during our observation periods. Finally, C. purpurea was visited by a wide diversity of visitors, with at least four of the six pollinator functional groups recorded at every site, and two populations (PMT and PTH) experiencing visitation from all observed pollinator functional groups, which was not observed in any other focal species. Hawkmoths were the most common visitor to C. purpurea in the center of the range (PCM and PTH), while small bees were dominant at the northernmost population (PTMS) and hummingbirds at the most southern population (PMT).

Across its wide range, C. sessiliflora also exhibited a diversity of visitors (as all functional groups but hummingbirds were recorded), though diversity within populations was low overall (Fig. 2). Hawkmoths were the most frequent visitor at most populations in the southern portion of the range, but no hawkmoths were observed at any northern populations, resulting in a significant decrease in hawkmoth visitation from south to north (Fig. 4A; latitude: χ²_1,25 = 11.2, p < 0.001; year: χ²_2,25 = 0.5, p = 0.77; dataset: χ²_1,25 = 0.09, p = 0.76). Small/medium bees were common visitors across the range of C. sessiliflora and accounted for most visits in the northern range, which resulted in the proportion of visits from small/medium bees increasing with latitude (Fig. 4B; latitude: χ²_1,25 = 15.8, p < 0.0001; year: χ²_2,25 = 3.4, p = 0.18; dataset: χ²_1,25 = 5.9, p = 0.01). Two southern populations (SIC and SMP) were previously characterized as having distinctive floral morphs, including shorter corollas and bright inflorescence color (Wenzell et al. 2021), and these populations experienced visitation from a broader array of visitors than other populations of C. sessiliflora (Fig. 2). The distinct bright-pink-flowered population (SIC) was the only population of C. sessiliflora to experience visitation from butterflies and bee-flies (Bombyliidae; included in ‘other visitors'), making its assemblage of floral visitors unique compared to other populations of C. sessiliflora. Additionally, the yellow-flowered population (SMP) saw visitation by bumblebees in all three years studied, although bumblebee visitation was also observed at two northern white-green-flowered populations in 2019. Latitudinal trends were not significant for any other pollinator functional groups, including bumblebees (latitude: χ²_1,25 = 1.9, p = 0.17; year: χ²_2,25 = 1.5, p = 0.46; dataset: χ²_1,25 = 2.3, p = 0.12) and other visitors (latitude: χ²_1,25 = 0.2, p = 0.64; year: χ²_2,25 = 2.2, p = 0.33; dataset: χ²_1,25 = 1.6, p = 0.19). Butterflies (recorded at only one population) and hummingbirds (not recorded at any C. sessiliflora sites) were too infrequent for analysis (Fig. 4C–E). Year was not significant in any model, and dataset effect was significant only for small/medium bees, again likely due to the difficulty observing small bees at longer visual ranges.

Figure 4

Geographic variation in visitation and fruit set across the range of C. sessiliflora. (A–E) Proportion visitation by pollinator functional group (population–year-dataset datapoints, n = 31) or (F) average proportion fruit set (population mean ± SE, n = 317) is plotted against a proxy for geographic position across the species range (population latitude, °N). Fill colors show median floral color per population. Solid trend lines denote a significant relationship (p < 0.05) based on GLMM. (Note: GLMM could not be performed for butterfly data, as all but one population of C. sessiliflora recorded zero butterfly visits. Additionally, no hummingbirds were observed visiting C. sessiliflora.)

Fruit set varies within but not among species

Fruit set did not vary significantly among species (Fig. 3B; χ²_3,583 = 5.5, p = 0.14) or years (χ²_2,583 = 1.6, p = 0.46) when a population random effect was included (suggesting that variation at the population level may outweigh that among species). Within the widespread C. sessiliflora, fruit set varied geographically across the range (Fig. 4F; χ²_1,312 = 55.5, p < 0.0001) and temporally across years (χ²_2,312 = 118.9, p < 0.0001).

Visitation from different pollinator groups is associated with floral traits

We identified significant relationships between floral traits and visitation from different pollinator groups across species (Table 2), summarized here. Visitation from hummingbirds was significantly associated with increased population mean corolla length and shorter average petaloid lips and was marginally associated with greater stigma exsertion. Hawkmoth visitation was significantly associated with wider average corolla tubes and narrower bract lobes. Bumblebee/large bee visitation was significantly negatively associated with NMDS2 axis of floral color, consistent with greater visitation to human-vision yellow-green flowers than to red/purple flowers (Fig. 5A), but was positively associated with stigma exsertion. Visitation from small/medium bees was significantly associated with longer petaloid lips and wider bract lobes, but negatively associated with stigma exsertion (Fig. 5C). For butterflies, visitation was significantly associated with higher values on the NMDS2 color axis, in line with red-purple floral color (Fig. 5B), and marginally associated with shorter corollas. Visitation from other visitors was significantly associated with narrower corollas, wider bract lobes and higher values on the NMDS1 color axis (associated with human-vision purple and pink), though this group of ‘other visitors' includes various functional and taxonomic groups (e.g. bee flies, flies and moths) expected to respond to floral signals in differing ways. Dataset type was significant for all pollinator groups excepts hawkmoths and other visitors, butyear effect was not significant for any pollinator groups (Table 2).

Figure 5

Pollinator visitation by floral trait values. (A) Proportion of large bee visits and (B) butterfly visits in relation to population mean floral color (NMDS2 axis value from NMDS of RGB floral colors; Supporting information). (C) Proportion of small/medium bee visits in relation to population mean stigma exsertion (mm). (D) Ordination of population mean floral trait values (points; centroids and standard deviations labelled by species) with visitation from different pollinator groups with significant associations (p < 0.05) plotted as environmental vectors (arrows). Points show species (sp): C. citrina (diamonds), C. lindheimeri (triangles), C. purpurea (squares), C. sessiliflora (circles), and fill color shows population median floral color. Trend line represents significant relationships (p < 0.05) based on GLMM (A–C, Table 2).

Table 2. Visitation by pollinator group in relation to floral trait values. GLMMs of visitation from different pollinator groups (A–F) according to scaled population-average floral trait values, with dataset type and year included as fixed effects. Coefficient estimate (showing positive or negative relationship, in logit scale), standard error (SE), and z-value (z) are presented for each variable of interest, along with χ², degrees of freedom (df, with residual df = 49 for all models), and p-value (p) for all terms. p < 0.05 and associated variables are in bold

Pollinator group	Variable	Estimate	SE	z	χ^2	df	p
(A) Hummingbirds	Corolla length	6.06	2.33	2.60	6.78	1	0.009
	Corolla width	1.39	1.28	1.09	1.18	1	0.278
	Lip length	−6.67	2.55	−2.61	6.82	1	0.009
	Stigma exsertion	3.88	2.24	1.73	2.99	1	0.084
	Bract lobe width	0.62	1.55	0.40	0.16	1	0.688
	NMDS1 color axis	−0.85	0.71	−1.20	1.43	1	0.231
	NMDS2 color axis	−1.16	1.38	−0.84	0.70	1	0.402
	Year	–	–	–	1.32	2	0.516
	Dataset	–	–	–	7.89	1	0.005
(B) Hawkmoths	Corolla length	−0.17	0.65	−0.26	0.07	1	0.794
	Corolla width	0.94	0.43	2.18	4.75	1	0.029
	Lip length	−1.61	1.14	−1.41	2.00	1	0.158
	Stigma exsertion	0.75	0.72	1.05	1.09	1	0.296
	Bract lobe width	−3.13	0.98	−3.21	10.32	1	0.001
	NMDS1 color axis	−0.53	0.41	−1.31	1.71	1	0.190
	NMDS2 color axis	0.42	0.55	0.77	0.59	1	0.443
	Year	–	–	–	0.77	2	0.682
	Dataset	–	–	–	0.51	1	0.474
(C) Bumblebees/large bees	Corolla length	0.19	1.03	0.18	0.03	1	0.854
	Corolla width	−0.72	0.52	−1.38	1.91	1	0.167
	Lip length	−2.93	1.95	−1.50	2.24	1	0.134
	Stigma exsertion	5.51	2.21	2.49	6.21	1	0.013
	Bract lobe width	−2.36	1.36	−1.74	3.01	1	0.083
	NMDS1 color axis	1.23	0.76	1.62	2.61	1	0.106
	NMDS2 color axis	−5.67	1.92	−2.96	8.78	1	0.003
	Year	–	–	–	2.07	2	0.356
	Dataset	–	–	–	7.62	1	0.006
(D) Small/medium bees	Corolla length	−1.10	0.58	−1.88	3.54	1	0.060
	Corolla width	−0.17	0.33	−0.50	0.25	1	0.616
	Lip length	2.64	0.99	2.67	7.13	1	0.008
	Stigma exsertion	−2.13	0.67	−3.19	10.17	1	0.001
	Bract lobe width	2.62	0.80	3.28	10.77	1	0.001
	NMDS1 color axis	0.56	0.33	1.69	2.85	1	0.091
	NMDS2 color axis	0.02	0.52	0.03	0.00	1	0.976
	Year	–	–	–	3.60	2	0.165
	Dataset	–	–	–	11.20	1	0.001
(E) Butterflies	Corolla length	−3.69	1.89	−1.95	3.80	1	0.051
	Corolla width	−0.31	0.49	−0.64	0.40	1	0.525
	Lip length	2.04	1.69	1.21	1.45	1	0.228
	Stigma exsertion	−1.41	0.91	−1.55	2.39	1	0.122
	Bract lobe width	2.01	1.40	1.43	2.05	1	0.152
	NMDS1 color axis	0.44	0.38	1.16	1.35	1	0.244
	NMDS2 color axis	1.85	0.82	2.26	5.11	1	0.024
	Year	–	–	–	5.49	2	0.064
	Dataset	–	–	–	4.13	1	0.042
(F) Other visitors	Corolla length	−0.22	0.94	−0.23	0.05	1	0.819
	Corolla width	−1.25	0.49	−2.54	6.44	1	0.011
	Lip length	0.95	1.56	0.61	0.37	1	0.544
	Stigma exsertion	−0.72	0.96	−0.75	0.56	1	0.456
	Bract lobe width	2.87	1.37	2.09	4.38	1	0.036
	NMDS1 color axis	1.72	0.45	3.85	14.80	1	< 0.001
	NMDS2 color axis	−0.71	0.77	−0.92	0.85	1	0.357
	Year	–	–	–	5.10	2	0.078
	Dataset	–	–	–	2.68	1	0.101

Based on an ordination of multivariate floral trait variation, we found significant evidence for associations among several pollinator functional groups and suites of floral traits (Fig. 5D), which also aligned with our focal plant species. Proportion of visits from hummingbirds increased towards the flowers of C. lindheimeri in trait space (R² = 0.31, p < 0.001), while butterfly visitation was aligned with flowers of C. purpurea (R² = 0.16, p = 0.003). Visitation from large bees increased in association with the yellow flowers of C. citrina, as well as the yellow-flowered population of C. sessiliflora (R² = 0.11, p = 0.039) and small/medium bee visitation was associated with flowers of C. sessiliflora (R² = 0.11, p = 0.029). Proportion of visits from hawkmoths (R² = 0.06, p = 0.16) and other visitors (R² = 0.06, p = 0.1) were not significantly associated with multivariate floral trait variation.

Discussion

In this study we demonstrate how geographic variation in floral traits reflects differences in local pollinator visitation across the ranges of four species of Castilleja. In doing so, we illuminate how range-wide variation in biotic interactions may contribute to early stages of phenotypic divergence, a condition likely to precede potential pollinator-mediated speciation. By sampling multiple populations over several years, we document a wide spectrum of visitors spanning hummingbirds, Lepidoptera (butterflies and hawkmoths), Hymenoptera (mostly small solitary bees and bumblebees) and Diptera (bee-flies and occasional hoverflies), demonstrating a diversity of visitors consistent with other studies of Castilleja species (Hersch and Roy 2007, Hilpman and Busch 2021). We show that our four focal species were visited broadly by all pollinator groups (with two exceptions: the lack of hummingbird visits to C. sessiliflora and of large bee visits to C. lindheimeri), suggesting generalized visitor assemblages (Ollerton et al. 2007, Armbruster 2017). Despite this broad diversity, the visitor assemblages among populations demonstrated a mosaic of interactions likely to contribute to observed floral trait divergence. Additionally, visitation from specific pollinator functional groups was associated with certain floral traits that often aligned with predictions based on pollination syndromes (Faegri and van der Pijl 1971, Fenster et al. 2004), suggesting local pollinators may play a role in shaping this floral variation. Hence, despite overall generalization of visitors across species, we find support for the hypothesis that a geographic mosaic of pollinators is likely to mediate divergence in floral traits within and among these recently diverged plant species. These results lend support to the notion that ‘adaptive generalization' of visitor assemblages may be more common than previously thought (Ohashi et al. 2021), and that strict specialization may not be necessary for divergence in floral phenotypes to arise. By sampling a wide geographic distribution at multiple populations, we identified considerable variability in floral traits that aligns with variation in pollinator assemblages, demonstrating the potential for pollinators to mediate divergence within this group (Herrera et al. 2006, Kay and Sargent 2009).

Divergence in floral traits and associated pollinators among the species of the Castilleja purpurea complex

We explored whether variation in floral visitors underlies floral divergence among species within the C. purpurea species complex. These species exhibited divergence in floral color across narrow geographic clines despite low genetic differentiation (Wenzell et al. 2021), which is consistent with other examples of pollinator-mediated selection driving floral color transitions (Streisfeld and Kohn 2005, Hopkins et al. 2012, Stankowski et al. 2017). By investigating whether distinct pollinators could be a driver of floral divergence, we found compelling support for this hypothesis in the predominately hummingbird-pollinated C. lindheimeri, and some support but less clear evidence in the largely bee-pollinated C. citrina and in C. purpurea, which hosted a wide diversity of visitor groups, consistent with a generalist pollination mode.

Although the species of the C. purpurea complex were visited by a broad array of pollinator functional groups overall, patterns varied by species. Within the red-bracted C. lindheimeri, which was visited predominately by hummingbirds at all sampled populations, several lines of evidence suggest the potential for floral adaptation to hummingbird pollinators. First, hummingbirds visited C. lindheimeri significantly more frequently than any other species, and visitation by hummingbirds was associated with longer corollas and shorter petaloid lips, floral traits associated with C. lindheimeri flowers (Fig. 5D, Supporting information) and with hummingbird pollination modes (Faegri and van der Pijl 1971, Fenster et al. 2004, Rosas-Guerrero et al. 2014). Castilleja lindheimeri also exhibits red floral pigmentation, another well-documented hummingbird-associated trait, though our model found no association between hummingbird visitation and color values. While initially surprising, this finding is consistent with observations that hummingbirds forage broadly on flowers of varying colors (Waser et al. 2018) and may also reflect the hypothesis that red pigment in hummingbird-pollinated flowers may function as a deterrent to bee pollinators rather than as an attractant to hummingbirds, per se (Castellanos et al. 2004, Schiestl and Johnson 2013). While additional work is needed to directly test for pollinator-mediated selection on floral traits in this system, we present evidence consistent with the hypothesis that C. lindheimeri may exhibit adaptation to pollination by hummingbirds, based on its frequency of hummingbird visitors and its suite of syndrome-aligned floral traits.

In contrast, visitor assemblages of the other species of the C. purpurea complex, C. purpurea and C. citrina, were more generalized. In terms of floral divergence, this species pair appears to differ less in morphological traits compared to the other focal species (Supporting information). Lower floral divergence compared to other focal species may be expected with generalist pollinator assemblages, which may exert contrasting selection pressures on floral traits, resulting in intermediate phenotypes (van der Niet et al. 2014, Ohashi et al. 2021). Nonetheless, some trends in visitation differ between these species. Notably, the yellow-flowered C. citrina experienced visitation from a broad array of pollinator functional groups but was visited most frequently by bumblebees/large bees and was the only species visited by large bees at every population (Fig. 2, 3). We hypothesize that color may play a role in this pattern, given that bumblebee visitation was associated with both lower values on the NMDS2 color axis (which includes mostly human-vision yellow and green inflorescences, Fig. 5A), and an association with yellow-flowered populations in our ordination of floral traits (of both C. citrina and the yellow-flowered population of C. sessiliflora; Fig. 5D). Notably, this same yellow-flowered population of C. sessiliflora (SMP) had greater visitation by bumblebees compared to other nearby C. sessiliflora populations with differing color (Fig. 2). The association between large bee visitation and NMDS2 color value may suggest that large bees are more likely to visit yellow flowers and/or are less likely to visit red flowers (Fig. 5A), and thus could be consistent with the evolution of red floral pigmentation as a deterrent to bee visitors (Castellanos et al. 2004) given the lower sensitivity of bee visual systems to human-vision red spectra (Schiestl and Johnson 2013).

Castilleja purpurea had the broadest diversity of floral visitors of the focal species, and its most common visitor varied from small/medium bees to hawkmoths to hummingbirds across the range (Fig. 2). Despite this diversity, this species experienced frequent visitation from Lepidopteran pollinators, with significantly greater butterfly visitation than other species, and a high number of visits by hawkmoths. We found that butterfly visitation increased in association with the floral traits of the purple-flowered C. purpurea (Fig. 5D) and along the NMDS2 axis of color variation linked to human-vision red-purple flower color (Fig. 5B), which is consistent with innate preference for and spectral sensitivity to human-vision red and purple flowers reported in Lepidoptera (Lunau and Maier 1995, Chittka and Thomson 2001). While a possible relationship between butterflies and purple and bumblebees and yellow flowers in this system remains largely speculative, these findings warrant future work into floral pigmentation and its evolutionary origins in this species complex.

Our findings suggest flower color may not constitute a strong barrier to pollination by different visitor groups but could nonetheless impact their frequency of visitation. Color is considered a weak barrier compared to other syndrome-associated floral traits (Dellinger 2020), though color nonetheless can evolve rapidly and facilitate reproductive isolation through pollinator preference (Schemske and Bradshaw 1999, Bradshaw and Schemske 2003) and reinforcement via pollinator foraging behavior (Hopkins and Rausher 2012). As no clear pollinator shift was observed to explain the color transition between C. purpurea and C. citrina, future work should investigate floral antagonists as potential selective agents on color in this system, given their important role in exerting selection on floral traits (Irwin et al. 2003, Kessler et al. 2010, Jogesh et al. 2017). Taken together, this study provides support for the hypothesis that variable pollinator assemblages likely contribute to floral trait divergence among species of the C. purpurea species complex, though additional ecological drivers seem likely, which warrant further study.

Pollinator mosaics across the range of Castilleja sessiliflora

Compared to its congeners, flowers of C. sessiliflora are clearly differentiated by long corolla tubes, long petaloid lips, and pale floral pigmentation (Fig. 1). Beyond this, C. sessiliflora exhibits intraspecific floral trait variation, including latitudinal variation in inflorescence color across its range and more divergent floral morphs in the south, which vary in color and corolla length (Fig. 1; Wenzell et al. 2021) and which occur in regions adjacent to members of the C. purpurea complex species (see Speculations). In our study, floral trait variation across the distribution of C. sessiliflora corresponded to range-wide variation in floral visitors. Visitation to northern populations came almost exclusively from small/medium bees and less commonly bumblebees, which were previously reported to visit C. sessiliflora in Wisconsin (Crosswhite and Crosswhite 1970). In the southern portion of the range, the most common visitors were hawkmoths Hyles lineata and small/medium bees, along with occasional bumblebees, butterflies and other infrequent visitors (e.g. bee-flies) at a few populations (Fig. 2). Given that hawkmoths were the predominant visitor to many populations in the south, it was noteworthy that no hawkmoths were observed foraging on C. sessiliflora in the northern half of the range, despite reports of them at some study sites (Friends of Nachusa Grasslands 2020) and rare visits to C. sessiliflora observed in previous years (J. Fant unpubl.). Hawkmoth visitation is known to be unreliable in space and time (Campbell et al. 1997, Miller 1981), and it is possible that hawkmoths in the northern region may be active outside our observation window (e.g. later at night) or later in the growing season than the spring-flowering C. sessiliflora. Additionally, because species were sampled unevenly across years (Table 1, Supporting information), we cannot rule out that potential temporal fluctuations in visitor assemblages may influence observed differences among species. However, reported patterns were consistent across multiple populations and regions sampled in multiple years, which lends support to out interpretations.

Within C. sessiliflora, we identified two phenotypically distinct populations that were visited by broad pollinator assemblages that may more closely resemble those of the C. purpurea complex than other populations of C. sessiliflora: the yellow-flowered SMP (Fig. 2D) and pink-flowered SIC (Fig. 2E–F), whose visitors included bumblebees, butterflies and other visitors in addition to hawkmoths and small bees. Previous genetic analyses did not find evidence that these distinct populations represent recent hybrids with the C. purpurea species (Wenzell et al. 2021), suggesting that hybridization with nearby congeners does not explain this divergence in floral phenotypes. We hypothesize that these distinct floral phenotypes may attract (via brighter floral colors) and allow access to (via shorter corollas) a more diverse assemblage of visitors than do more typical C. sessiliflora flowers (Fig. 1B–D), which is supported by these populations' placement closer to flowers of the C. purpurea complex in our ordination of floral traits and associated pollinator functional groups (Fig. 5D). In such a case, these distinct floral morphs may represent pollination ecotypes within C. sessiliflora, which could suggest the potential for early stages of speciation, potentially mediated by geographic mosaics of pollinators (van der Niet et al. 2014). This hypothesis should be investigated further in this system by comparing pollination effectiveness of different functional groups, as well as the capacity for reproductive isolation among these populations.

In the northern portion of the range, visitation was often low (Supporting information) and fruit set was lower on average than in the south. This raises questions about the effectiveness of small/medium bees as pollinators, as they were the predominate visitors to C. sessiliflora in the north. Because small/medium bees, which were only observed foraging on pollen and did not enter corollas to access nectar, may be small enough to do so without contacting stigmas (Fig. 2A), their behavior could potentially result in low pollen deposition and poor pollination. Interestingly, visitation from small/medium bees was associated with lower stigma exsertion (Fig. 5C), which could be selected for to facilitate pollination by small bees foraging on anthers. Visitation by small/medium bees was also associated with longer petaloid lips and wider bract lobes, both of which could serve as landing platforms for bees foraging for pollen at the mouths of corollas.

Despite the prevalence of small/medium bees and the absence of hawkmoths among northern populations, these populations still exhibit clear hawkmoth-associated floral traits, namely long corolla tubes and pale floral pigmentation (Faegri and van der Pijl 1971, Fenster et al. 2004). Though further work is needed, we hypothesize this may reflect factors such as evolutionary constraints on floral traits (Huang and Fenster 2007, Zufall and Rausher 2004), effects of land use change on pollinator populations (Grixti et al. 2009, Young et al. 2017, Durant and Otto 2019) and/or the history of glaciation in the northern Great Plains and recent northward colonization of plant populations, which may have become adapted to pollinator environments in their southern range extent (Clayton and Moran 1982, Waters et al. 2013, Ursenbacher et al. 2015). Additionally, because small/medium bees foraged only on pollen at the mouth of the corolla, they could access this reward regardless of the length of corolla tubes (which limit access only to nectar rewards). Thus, this interaction may not incur a fitness cost that would result in the evolution of a tradeoff between hawkmoth and small bee pollinators (i.e. no selection for shorter corollas, even in the absence of hawkmoths when small bees are the only pollinator), suggesting this scenario could represent an example of ‘adaptive generalization' in floral phenotype (Ohashi et al. 2021). Nonetheless, the long corolla tubes of C. sessiliflora (measuring over 50 mm in length in many cases; Supporting information) may serve as a filter to other groups of visitors that cannot access nectar at the base of the corolla, which could result in the less diverse assemblages and low visitation rates of many northern populations, where small/medium bees were often the only observed visitor (Fig. 2).

Despite their inconsistency across the range of C. sessiliflora and their visitation to other focal species, we hypothesize that hawkmoths may be efficient and effective pollinators of C. sessiliflora when they are present. Both hawkmoth visitation and fruit set decreased in the north of the range, which could suggest hawkmoths are important contributors to reproductive fitness of C. sessiliflora populations (though we acknowledge many other covarying biotic and abiotic factors could also explain these patterns). Hawkmoths Hyles lineata have been reported as infrequent but impactful pollinators in Ipomopsis, where they exert strong selection on flowers in years they visit (Campbell et al. 1997, Campbell and Aldridge 2006), thus influencing floral trait evolution even as inconsistent pollinators. Nonetheless, despite expected associations between long corolla tubes and pollination by hawkmoths, hawkmoth visitation was not associated with corolla length (Table 2). While unexpected, we suspect this reflects the ability of hawkmoths to forage on flowers of all focal species, as their long proboscides can access nectar in both short and long corollas (Johnson et al. 2017). However, this finding does not contradict the hypothesis that longer corollas may improve the efficiency of pollination by hawkmoths, by increasing contact between plant sexual organs and hawkmoths' bodies (Whittall and Hodges 2007). In fact, a pollinator exclusion experiment in this system found evidence that fruit set was greater among plants exposed to nocturnal pollination, but only in long-floral-tubed populations where hawkmoths were present, suggesting both pollinator identity and floral traits mediate increased plant fitness (Wenzell et al. unpubl.). Combined with the results presented here, these findings lead us to hypothesize that the long corollas of C. sessiliflora may reflect an association with hawkmoth pollinators, which warrants further study.

Speculations

Variation in plant–pollinator interactions, revealed through detailed field observations (Dellinger 2020), suggests that complete pollinator shifts may be uncommon (Waser et al. 1996, Ohashi et al. 2021), challenging classic models of pollinator-mediated plant speciation. At the same time, increasing evidence suggests that speciation via selection can occur despite ongoing gene flow (Schluter and Rieseberg 2022), underscoring that much remains unknown about early stages of phenotypic divergence and speciation. Here, we propose that, in these Castilleja complexes, documented divergence is consistent with early stages of speciation with ongoing gene flow. In the Castilleja purpurea complex, we speculate that floral divergence among species reflects selection on floral traits, likely mediated by hummingbird pollinators in C. lindheimeri and a combination of pollinators and other ecological drivers in C. purpurea and C. citrina, despite their largely generalized pollinator assemblages. Additionally, we speculate that the long floral tubes in C. sessiliflora facilitate reproductive isolation by favoring pollination by long-tongued hawkmoths and thereby filtering out potential pollination from groups shared with nearby congeners. The exception to this may be two populations in the southern range extent, characterized by short corollas and bright floral color (Fig. 1), which we hypothesize represent divergent pollination ecotypes. Like the C. purpurea complex, they occur in dry grasslands in the south-central US, where we observed a greater abundance and diversity of pollinator groups (e.g. butterflies, bumblebees and hummingbirds) compared to elsewhere in the C. sessiliflora range. Thus, these coinciding patterns of diversity of floral phenotype and of floral visitors may reflect a geographic mosaic of plant–pollinator interactions, whereby diversity in local pollinators may beget divergence in floral traits. Future work should investigate potential fitness costs of sharing pollinators and hybridization to better characterize reproductive isolation among these recently diverged taxa.

Conclusions

This study characterizes variation in pollinator assemblages at multiple populations across the ranges of C. sessiliflora and the species of the C. purpurea complex and suggests that floral divergence among these taxa may be mediated at least in part by pollinators, despite largely generalized assemblages of visitors. Aided by our thorough sampling of 23 populations across the ranges of our focal species, we found considerable variation in floral visitors among species and throughout their geographic ranges, coinciding with variation in floral traits. Within the C. purpurea species complex, C. purpurea and C. citrina were visited by a diverse array of generalist pollinator functional groups. In contrast, C. lindheimeri was predominately visited by hummingbirds and was characterized by floral traits associated with hummingbird pollination, suggesting that C. lindheimeri may exhibit adaptation to hummingbird pollinators. In the long-floral-tubed C. sessiliflora, floral visitors were structured by geography, with hawkmoths being frequent visitors in the southern half of the range, though absent in the north, where pollen-foraging small and medium bees were dominant. Taken together, these results provide evidence that floral trait variation within and among species may reflect variable pollinator assemblages across their distributions, suggesting that mosaics of largely generalized pollinators may play a role in floral divergence in C. sessiliflora and the species of the C. purpurea complex, though likely in concert with other ecological drivers. Thus, this study provides a robust investigation of geographic variation in floral traits and local pollinators across the distributions of plant species, illuminating how pollinator mosaics may contribute to divergence in floral traits across geography in a recently radiating genus, potentially via pollinator-mediated plant evolution.

Acknowledgements

– The authors thank A. Iler, H. Briggs, K. Byers and M. Neequaye for feedback on previous versions of this manuscript. Additional thanks to A. Iler for input on statistical analyses and to K. Byers for assistance with identification of photographed insects. We also thank D. Tank and M. Egger for insight on the C. purpurea species complex. We are grateful for field assistance from S. Deans, K. Bonefont, A. Cisternas Fuentes, C. Woolridge, A. Gruver, E. Lewis, S. Todd, A. J. Morgan, K. Manion, M. Malone and G. Perez Cartagena and laboratory assistance from J. Zhang.

Funding

– Funding was provided by the Botanical Society of America, American Society of Plant Taxonomists, American Philosophical Society Lewis and Clark Grant, Friends of Nachusa Grasslands, Northwestern University Program in Plant Biology and Conservation and an NSF Graduate Research Fellowship award (DGE-1842165) to KEW. Undergraduate students were supported by an NSF Research Experience for Undergraduates site award to JBF (DBI-461007 and DBI-1757800). Additional support was provided by the Negaunee Institute for Plant Conservation and Action at the Chicago Botanic Garden.

Ethic statement

– All research was conducted following appropriate ethical considerations and with all necessary permits for field work. The following permitting agencies provided permission for research and access to study sites: The Nature Conservancy Chapters of TX, OK and IL, Native Prairies Association of Texas, Cities of Lubbock, TX, Ft. Worth, TX and Tulsa, OK, TX Parks and Wildlife Department, NM Energy, Minerals and Natural Resources Department, MN Department of Natural Resources, IL Nature Preserves Commission, IL Department of Natural Resources, US Fish and Wildlife Services, US Forest Service, US Bureau of Land Management, US Army Corps of Engineers and US National Park Service.

Author contributions

Katherine E. Wenzell: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Funding acquisition (lead); Investigation (lead); Methodology (equal); Visualization (lead); Writing – original draft (lead); Writing – review and editing (lead). Krissa A. Skogen: Conceptualization (equal); Formal analysis (supporting); Funding acquisition (supporting); Investigation (supporting); Methodology (equal); Resources (equal); Supervision (equal); Visualization (supporting); Writing – review and editing (equal). Jeremie B. Fant: Conceptualization (equal); Formal analysis (supporting); Funding acquisition (supporting); Investigation (supporting); Methodology (equal); Resources (equal); Supervision (equal); Visualization (supporting); Writing – review and editing (equal).

Open Research

Data availability statement

Data are available on GitHub: https://github.com/KWenzell/Castilleja_pollinator_mosaics (Wenzell et al. 2022).