Ecological mechanisms of reproductive isolation between diploid and tetraploid Chamerion angustifolium

Authors


Brian C. Husband (fax 519-767-1991; e-mail bhusband@uoguelph.ca).

Summary

1  Theoretical models indicate that coexistence of diploid and polyploid individuals in sympatric populations is unlikely when mating is random among cytotypes and hybrids are inviable. However, coexistence may be facilitated by prezygotic isolating mechanisms that reduce intercytotype mating and increase assortative mating.

2  We examined the ecological factors regulating intercytotype mating in Chamerion (formerly Epilobium) angustifolium by measuring floral morphology, flowering synchrony and insect foraging preferences in a mixed population within a diploid–tetraploid contact zone. We also calculated the minimum rate of between-cytotype mating from estimates of the frequency of triploid offspring.

3  Diploids had significantly shorter and narrower petals as well as shorter styles than tetraploids. Inflorescences were significantly taller in tetraploids than diploids, but the mean number of open flowers per inflorescence did not differ. Diploid and tetraploid flowering periods overlapped by 51%, with diploid stems flowering earlier.

4  In a plot of 20 diploid and 28 tetraploid stems that were flowering simultaneously, only 26% of all bee flights were between flowers on different stems. Of the total flights between inflorescences, only 15% were between different cytotypes. The combined effects of flowering asynchrony and insect foraging reduced the opportunities for intercytotype mating from 49% expected if mating is random to about 2% of the total number of pollinator flights.

5  A computer simulation indicated that the deficiency of pollinator flights between cytotypes was due largely to the spatial structure of cytotypes and limited pollinator flight distances within the observation plot. The frequency of triploid offspring produced during the period when both cytotypes were in flower was 6.6%, similar to the proportion of flights observed between flowers on different cytotypes (4%).

6  The results indicate that flowering phenology and insect behaviour (as influenced by clonal structure) have a significant effect on prezygotic mating isolation and triploid production, and may contribute to the maintenance of mixed cytotype populations in Chamerion angustifolium.

Introduction

Botanists have long recognized the significance of polyploidy in the evolutionary history of flowering plants (Stebbins 1950; Grant 1981; Masterson 1994) and have sought an explanation for its wide taxonomic distribution (Stebbins 1950). As a result, the literature is rich with investigations of the adaptive significance of polyploidy and its ecological, genetic and physiological correlates (reviewed in Stebbins 1950; Levin 1983). More recent research has focused on the population biology of polyploids and specifically the origins of tetraploids and their interactions with their diploid progenitors (Lewis 1980; Thompson & Lumaret 1992). This dynamic perspective has arisen as a result of recent work that suggests that polyploids may arise repeatedly in diploid taxa (Soltis et al. 1989; Wolf et al. 1990; Soltis & Soltis 1993) and that polyploid origins and establishment can be expressed in demographic terms (Fowler & Levin 1984; Felber 1991; Thompson & Lumaret 1992).

A number of theoretical models have been developed that examine the ecological and genetic circumstances governing the interactions between diploids and polyploids in mixed populations (Levin 1975; Fowler & Levin 1984; Felber 1991; Rodríguez 1996). These models indicate that, when mating is random, the rare cytotype will experience a mating disadvantage. This process, called minority cytotype exclusion (Levin 1975), occurs because the rare cytotype is more likely to receive pollen from the majority cytotype than from other plants of its own cytotype, and because the resulting hybrids are usually inviable or sterile. In the absence of recurrent production of individuals of the rare cytotype, such frequency-dependent selection will prevent the coexistence of diploids and polyploids in sympatry.

The strength of frequency-dependent selection against the rare cytotype may be diminished, and hence coexistence enhanced, when prezygotic isolating mechanisms prevent diploid pollen from obtaining access to tetraploid ovules and vice versa (Stebbins 1950; Fowler & Levin 1984; Rodriguez 1996). Such isolation and the resulting elevated intracytotype mating may occur as a result of heterogeneity in the spatial distribution of cytotypes (Husband & Schemske 1998; Van Dijk et al. 1992), asynchronous flowering (Lumaret et al. 1987; Felber 1998; Bretagnolle & Thompson 1996; Husband & Schemske 1997) and self-pollination (Husband & Schemske 1997; Petit et al. 1997). In addition, pollinator constancy and gametophytic selection may affect assortative mating (Grant 1963; Arnold 1997); however, the roles played by these factors in reproductive isolation have not been examined in diploid–tetraploid interactions. Moreover, there are few empirical studies that provide quantitative measures of the realized rate of intercytotype mating (Lumaret & Barrientos 1990) or a comprehensive examination of the ecological and genetic factors regulating the production of triploid hybrids in mixed diploid-tetraploid populations (Petit et al. 1999).

In this study, we examined some of the ecological factors contributing to prezygotic reproductive isolation in mixed populations of diploid and tetraploid individuals of Chamerion (formerly Epilobium) angustifolium. This species exhibits geographical variation in chromosome number throughout its circumpolar distribution. Within North America, individuals are either diploid (2n = 2x = 36) or tetraploid (2n = 4x = 72). In general, diploids occur at higher latitudes than tetraploids, but they co-occur in a contact zone located along the southern border of the boreal forest and in a narrow region along the Rocky Mountains (Mosquin 1966). In sympatric populations within the contact zone, diploids, tetraploids and triploids occur together in a patchy mosaic distribution (Husband & Schemske 1998).

To examine reproductive isolation we focused on two ecological factors: the degree of synchrony in flowering phenology and pollinator visitation patterns, either or both of which may regulate triploid production. Specifically, we investigated the overlap in flowering times of diploids and tetraploids, whether the frequency of pollinator flights between cytotypes is significantly different from the random expectation and to what extent any differences are the result of the microspatial distribution of cytotypes rather than discrimination by insects. Finally, we measured the frequency of triploid offspring to determine the extent of intercytotype mating.

Materials and methods

Study organism

Chamerion (formerly Epilobium) angustifolium (L.) Holub is a perennial, herbaceous plant that is distributed widely throughout the northern hemisphere (Mosquin 1966). It is found in open and disturbed habitats and, although it is capable of clonal expansion through the production of root buds, it can also be a prolific seed producer. The purple-pink flowers, produced on showy multiflowered racemes, are primarily bee-pollinated and produce 100–600 comate seeds. The plant is self-compatible and the self-fertilization rates for a diploid and a tetraploid population have been reported as 0.06 and 0.28, respectively (Husband & Schemske 1995, 1997). Flowering time varies with climate, but in our study site it occurs from mid-July to early September.

Study site

Field work was conducted in 1995 on the Beartooth Pass (Highway 212) near the Clay Butte Road, Wyoming, between Red Lodge and Cooke City, Montana, USA. This region, described as population D23 in Husband & Schemske (1998), was originally described as having both diploid and tetraploid C. angustifolium by Flint (1980). A recent census of chromosomal cytotypes in this population indicated a predominance of tetraploids (55%) over diploids (36%), with relatively few triploids (9%) (D23 mixed site, Husband & Schemske 1998). However, the frequencies of the cytotypes varied significantly among local patches of plants.

Observations of flowering phenology were conducted in a 20 × 20 m area within the population, which was located on a rocky outcrop and comprised 50% diploid, 40% tetraploid and 10% triploid C. angustifolium according to a previous survey (Transect 18 in Husband & Schemske 1998). Although this study location was limited in area, it encompassed the majority of flowering individuals in the population for that year. Pollinator foraging behaviour, flower morphology and estimates of triploid production were estimated in a smaller observation plot (1.8 × 2.3 m) nested within this study area, that consisted of 48 diploid and tetraploid plants, which overlapped in flowering time.

Flow cytometry

DNA content, measured using flow cytometry, was used to determine the ploidy of all 48 plants within the small observation plot. This method was used because it is relatively fast, non-destructive and there is a strong relationship between chromosome number and DNA content (Husband & Schemske 1998). Samples of young leaves were stored on ice, then moved to a refrigerator prior to analysis at the flow cytometry facility at the University of Washington. A 2 × 2 cm section of each leaf was chopped with a fresh razor blade in 1 mL of an extraction buffer (modified from Michaelson et al. 1991) and then filtered through a 30-µm nylon mesh. Nuclei in the supernatant were stained with the DNA specific fluorescent dye DAPI and fluorescence per nucleus was estimated using a flow cytometer. The magnitude of fluorescence is proportional to the mass of DNA. Chicken red blood cells were stained and run as an external standard after every fourth sample (for additional details see Husband & Schemske 1998). For each sample, fluorescence was measured and expressed as a proportion of the mean for the standard of that day (hereafter referred to as relative fluorescence).

Ploidy was assigned to each sample by comparing its relative fluorescence value to the distribution of relative fluorescence for known diploids and tetraploids, which were collected from uniform diploid and tetraploid populations, originally identified as such by Flint (1980) and confirmed by Husband & Schemske (1998). Relative fluorescence for known diploids (n = 12) averaged 1.2 and ranged from 1 to 1.45, while relative fluorescence in tetraploids (n = 10) averaged 2.2 and ranged from 2 to 2.9 (Fig. 1). The relative fluorescence was significantly different between these ploidy levels based on an unpaired t-test (t = 15.13, d.f. = 20, P < 0.00001). The range of DNA content values for triploids only slightly overlapped (5% of cases) that of diploids. To avoid this ambiguity, relative fluorescence values of samples falling within the range for known diploids or tetraploids were classified as diploid or tetraploids, respectively. Samples whose relative fluorescence fell between those of diploids and tetraploids were classified as triploid.

Figure 1.

Distribution of relative fluorescence values for diploid (shaded bars) and tetraploid (solid bars) individuals of C. angustifolium, whose ploidy was know previously. Relative fluorescence is the fluorescence of a DNA-specific dye measured in a sample and expressed as a proportion of the fluorescence in a standard (chicken red blood cells).

Flower morphology

Floral and inflorescence characters were measured for all plants in the observation plot that were identified as diploids and tetraploids. These measurements were used to relate to pollinator visitation frequencies and, as correlates of chromosome number, to allow us to determine ploidy of plants in the larger study area. We measured maximum petal length (lower right petal), maximum petal width (lower right petal), style length (petal insertion to stigma lobe), inflorescence height (ground to bottom and top of inflorescence) and number of open flowers per inflorescence. The number of open flowers was measured on each plant on four different days (16, 17, 19, 20 August) and mean values were calculated; all other traits were measured once.

Flowering phenology

Each week, from 25 July to 7 September, we estimated the number of diploid and tetraploid stems with open flowers in the whole 20 × 20 m area. To do this, cytotypes of all plants in flower were identified as diploid or tetraploid using four morphological criteria (petal length and width, style length and inflorescence height-top) for which the mean values showed significant differences between cytotypes (Table 1). Similar morphological differences have been observed in plants grown in a common garden (B.C. Husband, unpublished data). A discriminant function analysis conducted on plants from the observation plot showed that diploids were correctly classified 90% of the time and tetraploids 92% of the time using these four characters (manova, Wilk's Lambda6,39 = 0.27, F6,39 = 27.8, P < 0.0001).

Table 1.  Comparison of floral and inflorescence traits between diploid (n = 20) and tetraploid (n = 28) plants of C. angustifolium, whose cytotypes were determined using flow cytometry. The data were used both for cytotype determination in the study of phenology and for examining differences that might influence pollinator discrimination. Means were compared using an unpaired t-test. The t-test statistic and level of significance (*  P  < 0.05, * *   P  < 0.01, * * *   P  < 0.001) are reported for each character measured. Values in parentheses are ranges. Number of open flowers is expressed as a mean across 4 days
  DiploidTetraploidt-value
Floral Characters
Petal length (mm)12.1 (8.0–17.4)16.6 (11.2–19.2)7.8***
Petal width (mm)9.0 (5.7–12.2)13.4 (9.6–16.4)8.6***
Style length (mm)9.8 (6.3–16.1)16.4 (11.1–18.8)10.4***
Inflorescence Characters
Height to bottom of inflorescence (cm)48.6 (31–66)68.1 (36–93)4.9***
Height to top of inflorescence (cm)56.1 (35–71)79.0 (38–106)5.7***
Number of open flowers per inflorescence8.9 (2.0–37.0)10.9 (3.0–21.5)1.0

To illustrate the flowering curves graphically, the number of flowering stems per cytotype observed for each census was expressed as a percentage of the maximum number observed flowering in any census period. To compare the flowering curves for diploids and tetraploids, an index of overlap was calculated by, first, expressing the number of flowering stems per cytotype at each census as a percentage of the total flowering stems observed for that cytotype across all census periods, and then for each census date taking the value for the cytotype that had the lower percentage of flowering stems and summing these values across all census dates. This index, which represents the proportion of the flowering curves that overlap, will equal zero when flowering is completely asynchronous and one when there is complete overlap.

Insect visitation

Insect visits to diploid and tetraploid inflorescences were observed over four days (16, 17, 19, 20 August) in the small observation plot. The plot represented one of few locations in the entire population in which the two cytotypes were in close proximity and in flower simultaneously, and that encompassed an area within which pollinator foraging could be monitored effectively. The 4-day observation period represented the time interval when most plants in the plot were flowering and was limited by the narrow overlap in flowering time between cytotypes.

Each day we observed the foraging bouts by the three most common bee species (Bombus melanopygus, Bombus mixtus and Bombus bifarius) that occurred in the plot within a 2-hour period. For each bout, we recorded the sequence of flowering stems visited (stems numbered previously) and the number of flower visits per stem within the plot. A foraging bout comprised all the visits by a single bee each time it entered the plot. Using the ploidy designations for each stem, the visitation frequencies and flight sequences were then compared to the visitation patterns expected with random foraging among cytotypes. Randomly foraging bees should visit diploid and tetraploid plants with frequencies equivalent to their proportions in the plot. Similarly, the proportion of 2x−2x, 2x−4x, 4x−2x and 4x−4x flights should be equal to the product of the appropriate cytotype frequencies in the plot. The percentage of flowering inflorescences in the observation plot that were diploid was 43.5, 42.6, 42.2 and 43.2 for 16, 17, 19 and 20 August, respectively. Because of the uniformity in diploid frequency over this period, we used the mean (43%) as the diploid visitation frequency expected under random foraging. The observed and expected patterns of visitation by pollinators were compared using G goodness-of-fit tests. Gpool statistics described the deviations from random for visitation frequencies pooled across days. Ghet statistics tested for significant variation in visitation patterns among days for each bee species.

Deviations from random expectations may arise not only because of discrimination by insects with respect to cytotype but also because of ecological factors such as spatial structure of cytotypes coupled with limited pollinator flight distances. To distinguish between these options, we also calculated expected visitation frequencies using a computer simulation, which incorporated the positions of each plant in the plot and the distribution of flight distances for each bee species. The distributions of flight distances observed for each bee species were calculated from the pollinator visitation sequences in the observation plot. The simulation consisted of a series of visits to inflorescences, positioned on a two-dimensional grid as they are in the observation plot. For each simulated visit, a random number was drawn from the observed flight distance distribution to determine the next flight distance. The simulation then randomly chose one eligible inflorescence from the total number of stems. The inflorescence visited was recorded and the selection procedure was then repeated. Visitation patterns were generated from 10 000 visits for each bee species. Source code for the simulation is available from the correspondence author upon request.

To relate the pollinator visitation patterns to various plant attributes in C. angustifolium, we conducted multiple regression analyses on two dependent variables: number of visits per inflorescence and mean number of flowers visited per stem. Both measures of pollinator activity were regressed against three independent variables: height of the inflorescence (top), petal length and mean number of open flowers per inflorescence (chosen because they represent different aspects of floral display and were not strongly correlated with each other; product moment correlation < 0.32).

Triploid seed production

Fruit were sampled from the small observation plot and the ploidy of their seeds determined. Data on flowering phenology allowed us to assign each fruit collected to one of three consecutive phases (only diploids flowering, diploids and tetraploids flowering, and only tetraploids flowering), based on the phase of fruit set. The maturing ovaries were marked and wrapped in wire mesh to reduce herbivory, and fruits were collected once they were mature. Seed set (number of full seeds per number of ovules) was estimated for 2–3 fruits per individual and a bulk sample of seeds derived from each flowering phase was sown in an incubator. Although germination rates were not monitored, very few seedlings were produced from the final (tetraploid only) phase so estimates of triploid frequency were made only for the ‘diploid only’ and ‘mixed flowering’ stages. Germinated seeds were transplanted to small pots. Once they were large enough, they were sampled for DNA content and classified as diploid, triploid and tetraploid using flow cytometry. Flow cytometric analyses were conducted at the University of Guelph and involved a slightly different protocol than previously described for the field-collected tissue. Fresh leaf tissue (approximately 2–3 cm2) was finely chopped on ice with a clean razor blade in a chopping buffer (Bino et al. 1992). The homogenate was filtered through a 30-µm nylon mesh and centrifuged at high speed (15 000 r.p.m.) for 10 s and then resuspended in 320 µL of a second buffer (Solution B, Arumuganathan & Earle 1991). Finally, 6.4 µL of propidium iodide solution (5 mg mL−1 H20) and 0.24% DNAase-free RNAase (Boehringer Mannheim, Cat. No. 1119915) was added to each sample. Chicken red blood cells were prepared in the same manner and used as an external standard. The samples of resuspended nuclei were analysed using an EPICS Profile flow cytometer (Coulter Electronics, Haileah, Florida) equipped with an Argon ion laser (25mW, Omnichrome 150) emitting at 488 nm. For each sample, approximately 10 000 nuclei were recorded and displayed as a histogram of fluorescence intensity.

Our estimate of the frequency of triploids was based on offspring that survived to the seedling stage. Because offspring from 2x × 4x matings may not survive to seed maturation, and even those that do may not germinate completely, our estimates of the frequency of triploids represent the minimum number present at the time of fertilization. For this reason, we refer to our estimate as a realized measure of intercytotype mating.

Results

Morphology and phenology

Of the 48 plants in the observation plot, 20 were diploid and 28 were tetraploid (Fig. 2). The two cytotypes differed in floral and inflorescence morphology. Diploids had significantly shorter petals, narrower petals and shorter styles than tetraploids (Table 1). Inflorescences were significantly taller in tetraploids than diploids, but the mean number of open flowers per inflorescence did not differ on any of the four days in which pollinator observations were made. Moreover, in a repeated measures analysis of variance of the number of open flowers per plant, there were no significant differences in flower number among ploidy levels over the 4-day observation period (F = 0.96, P > 0.30).

Figure 2.

Distribution of diploid (○) and tetraploids (□) individuals of C. angustifolium in the insect observation plot.

Diploid plants began flowering and finished flowering one census period earlier than tetraploids. Out of a total of seven censuses, diploids and tetraploids were each observed flowering in five and were flowering together in four (Fig. 3). The measure of overlap, which takes into account the proportion of the total number of stems flowering, was 51%.

Figure 3.

Flowering phenology for diploid (shaded bars) and tetraploid (solid bars) stems assessed on a weekly basis in a mixed population of Chamerion angustifolium. Each week, the number of diploid and tetraploid stems with open flowers was censused. The census number is expressed here as a percentage of the maximum number of flowering stems observed throughout the census period (e.g. the maximum number of diploid flowering stems was observed on 16 August).

Insect visitation

We recorded 238 bee foraging bouts within the observation plot, comprising 2854 flights to inflorescences and 10 836 visits to flowers. The number of flowers visited by B. melanopygus, B. mixtus and B. bifarius was 3423, 4432 and 2981, respectively.

Visitation frequencies to tetraploid stems in the observation plot, pooled across days, were 81, 78 and 63% for B. melanopygus, B. mixtus and B. bifarius, respectively. For all bees combined, mean visitation was 75% compared to the expected 57% based on the mean frequency of tetraploid inflorescences. Each species, and all species combined, visited more tetraploids than expected at random (Fig. 4a; B. melanopygus: n = 908 visits, Gpool= 221.1, P < 0.001, d.f. = 3; B. mixtus: n = 1134 visits, Gpool= 194.9, P < 0.001, d.f. = 3; B. bifarius: Gpool= 7.8, P < 0.01, d.f. = 3; Combined: n = 7854, Gpool= 379.89, P < 0.001 d.f. = 3). The frequency of visits to tetraploid stems was uniform across days for B. melanopygus and for all species combined, but daily visitation frequencies were heterogeneous in the other two species (B. melanopygus: n = 908 visits, Ghet= 10.0, P > 0.05, d.f. = 9; B. mixtus: n = 1134 visits, Ghet= 16.9, P < 0.05 d.f. = 9; B. bifarius: n = 817 visits, Ghet= 27.7, P < 0.01, d.f. = 9; combined: Ghet= 1.24, P > 0.05 d.f. = 9). The heterogeneities may be a result of sampling error since they resulted from 1 of the 4 days having particularly unusual visitation values and very low sample sizes (less than seven foraging bouts).

Figure 4.

Insect visitation frequencies to tetraploid C. angustifolium in the observation plot, expressed as a deviation from random visitation frequencies, for three bee species. (a) Deviation when random expectation is based on the proportion of each cytotype in the observation plot. (b) Deviation when random expectation is based on the spatial arrangement of cytotypes in the plot and the pollinator flight distances observed. All deviations were significantly different from random expectation (see Results).

All bees exhibited sequences of visits that deviated from the random expectation of flights among cytotypes (Fig. 5a; B. melanopygus: n = 844 paired visits, Gpool= 180.3, P < 0.001, d.f. = 3; B. mixtus: n = 1040 paired visits, Gpool= 276.6, P < 0.001, d.f. = 3; B. bifarius: n = 727 paired visits, Gpool= 310.5, P < 0.001, d.f. = 3). Most of the deviation was due to a higher frequency of tetraploid to tetraploid flights and a lower frequency of flights between cytotypes (diploid to tetraploid and tetraploid to diploid) than one would predict from the frequencies of the cytotypes in the observation plot. The difference between observed and expected for diploid-to-diploid (2x−2x) flights was small in all species and the number of flights was less than expected for B. melanopygus and B. mixtus, but greater than expected for B. bifarius. When the data for all three bee species were pooled, the distribution of visitation sequences differed significantly from random expectations (G-test for goodness-of-fit: Gpool= 15338.7, P < 0.001, d.f. = 3); 15.5% of all flights were between different cytotypes, compared to the expectation of 49% and these were evenly distributed between 2x and 4x (8.5%) and 4x−2x (7.0%) sequences. Sixty-seven per cent of flights were between tetraploids, compared to the random expectation of 32%. There was significant heterogeneity in the visitation sequences among days for B. mixtus and B.bifarius but not for B. melanopygus or for all species combined (B. melanopygus: n = 908 visits, Ghet= 16.6, P > 0.05, d.f. = 9; B. mixtus: n = 1134 visits, Ghet= 24.3, P < 0.01 d.f. = 9; B. bifarius: n = 817 visits, Ghet= 31.9, P < 0.001, d.f. = 9; combined: Ghet= 2.6, P > 0.05 d.f. = 9).

Figure 5.

Insect visitation sequences to diploid and tetraploid C. angustifolium in the observation plot. The percentage of all possible pairwise sequences, expressed as the deviation from random expectations, are provided for each of three bee species. (a) Random expectation is based on the proportions of each cytotype in the observation plot. (b) Random expectation is based on the spatial arrangement of cytotypes in the plot and the pollinator flight distances observed.

Using computer simulations, we tested whether cytotype spatial structure and insect flight distances alone could explain the preference for tetraploids. Flight distances within the plot for all species ranged from 8 to 165 cm and were skewed to the right (Fig. 6). The primary effect of incorporating cytotype distribution and flight distances into the calculation of random expectations was to increase the expected frequency of visits to tetraploids and to reduce the predicted frequency of between-cytotype flights. When compared to the observed visitation frequencies, all species nevertheless still showed a disproportionately high number of visits to tetraploids (percentage of visits to tetraploids expressed as a deviation from expected: B. melanopygus, 21%, n = 908; B. mixtus, 15%, n = 1134; B. bifarius, 4%, n = 817), albeit a smaller deviation than when spatial structure was not considered (Fig. 4).

Figure 6.

Distribution of flight distances between stems of C. angustifolium for Bombus mixtus (n = 1040 flights), B. melanopygus (n = 844) and B. bifarius (n = 727).

All three bee species had smaller deviations in flight sequences from random expectations when cytotype structure and flight distance were considered, and in only one species (B. bifarius) was the discrepancy statistically significant (Fig. 5b, G-test for goodness of fit: B. melanopygus, G = 2.2, P > 0.05, d.f. = 3; B. mixtus, G = 6.1, P > 0.05, d.f. = 3; B. bifarius, G = 27.8, P < 0.001, d.f. = 3). In B. bifarius, within-cytotype flights were still more common and between-cytotype flights less common than expected (Fig. 5b).

Bees frequently visited more than one flower on an inflorescence, and usually visited more flowers on tetraploid stems than diploids. For each bee species on each of the four days examined, the mean number of flowers visited was always greater on tetraploids than diploids; these differences were significant on 4, 1 and 2 days for B. melanopygus, B. mixtus and B. bifarius, respectively, based on an unpaired t-test. Averaged across days, the mean number of flowers visited was statistically different for B. melanopygus (mean number of flowers visited: diploids = 2.1, SD = 0.19; tetraploids = 3.7, SD = 0.27; t = − 4.4, P < 0.001, d.f. = 1) and B. bifarius (mean number of flowers visited: diploids = 2.5, SD = 0.32; tetraploids = 3.7, SD = 0.24; t = − 3.0, P < 0.01) but not B. mixtus (mean number of flowers visited: diploids = 2.4, SD = 0.50; tetraploids = 3.5, SD = 0.27; t = − 2.0, P > 0.05). When flower visitation data was combined for all bees, there was a significant difference between the number of flowers visited on diploids ( = 2.6, SD = 1.7, n = 19) and on tetraploids ( = 3.7, SD = 1.1, n = 26, F1,43 = 7.6, P < 0.01).

Multiple regression analysis showed that flower size (petal length), number of open flowers and inflorescence height collectively explained a significant portion of the variation in the number of visits to each inflorescence (R2 = 0.87) and in the number of flowers visited per inflorescence (R2 = 0.83) in C. angustifolium. All three morphological variables explained a significant portion of the number of visits to each inflorescence, but only petal length and number of open flowers influenced the number of flowers visited per inflorescence (Table 2). Except for the lack of a relationship between petal length and number of flowers visited per inflorescence, visitation patterns were similarly related to the three morphological characters when diploids were analysed separately. In tetraploids, the number of open flowers and inflorescence height both explained a significant amount of variation in the number of visits to an inflorescence as well as the number of flowers visited per inflorescence.

Table 2.  Regression coefficients from multiple regression of pollinator visitation variables on three morphological characters: inflorescence height, petal length and number of open flowers. *   P  < 0.05, * * *   P  < 0.001
Dependent variableIndependent variableAll plantsDiploid onlyTetraploid only
No. of visits per inflorescenceInflorescence height0.21***0.22***0.14*
Petal length1.07***1.07***0.79
Number of open flowers0.660.441.19***
No. flowers visited per inflorescenceInflorescence height −0.1 −0.04 −0.02*
Petal length0.18***0.15 −0.04
Number of open flowers0.20***0.19***0.2***

Triploid production

Seed production in diploid and tetraploid plants from the observation plot varied depending on when the flowers were open (Table 3). For diploids, seed set was significantly higher in flowers that were open when only diploids were flowering (seed set = 26%) than in flowers open during the mixed flowering period (seed set = 5.6%; Table 3). For tetraploids, seed set was significantly higher during the period of mixed flowering (seed set = 4.2%) than during the period when only tetraploids were flowering (seed set = 0.1%; Table 3).

Table 3.  Percentage seed-set (SE) in diploid and tetraploid C.angustifolium at three stages in the flowering period. A = only diploids flowering; B = diploids and tetraploids flowering; C = only tetraploids flowering. Values with different superscripts are statistically different (P < 0.05) based on Scheffe's unplanned comparisons
 Flowering Stages
ABC
Diploid seed set26 (2.2)a5.6 (1.9)b
Number of plants sampled890
Number of fruits sampled29300
Tetraploid seed set4.2 (0.9)b0.1 (0.1)c
Number of plants sampled076
Number of fruits sampled02517

Of the 160 seedlings that emerged from seed collected from diploid plants during the diploid-only flowering period, only one was triploid (Table 4). Of the 60 seedlings from diploid or tetraploid parents that flowered during the mixed flowering period that were screened, four were triploid. None of the 18 seedlings from diploid parents were triploid, but four of the 42 seedlings from tetraploid parents were triploid. Overall, the frequency of intercytotype mating (as measured at the seedling stage) during the mixed flowering period was 6.6% (Table 4).

Table 4.  Frequency of triploids produced by diploid and tetraploid C. angustifolium at two stages in the flowering period. A = only diploids flowering; B = diploids and tetraploids flowering. G-test of independence (G = 6.2, P < 0.5) indicates that the incidence of triploids differed among flowering periods. Seed and seedling production by tetraploids in the flowering period when only tetraploids were flowering was too low for analysis of triploid frequency
 Flowering Stage
AB
Number of seedlings
Total16060
Number from diploid parents16018
Number from tetraploid parents042
Frequency of triploids
Total14
Frequency from diploid parents10
Frequency from tetraploid parents04
Percentage of triploids
Total0.606.7
Percentage from diploid parents0.600.0
Percentage from tetraploid parents9.5

Discussion

Throughout much of their North American ranges, diploid and tetraploid cytotypes of Chamerion angustifolium are geographically separated, with the latter generally occurring at lower latitudes and lower altitudes (Mosquin 1966; Flint 1980; Husband & Schemske 1998). However, in the zone of contact diploids and tetraploids can occur in close proximity (Husband & Schemske 1998). The presence of mixed populations of C. angustifolium runs counter to the theoretical assumption that frequency-dependent selection should exclude the cytotype in the minority (Levin 1975; Felber 1991). This study was designed to investigate the role of pre-mating reproductive isolation in regulating the coexistence of diploids and tetraploids in mixed populations. Here we consider the ecological factors influencing mating interactions between diploid and tetraploid cytotypes, their collective effects on reproductive isolation and the implications for the spread and maintenance of polyploids in natural populations.

Within the study locality, diploid and tetraploid individuals had divergent flowering phenologies. Although the duration of flowering was similar for both cytotypes, the flowering phase started approximately 1 week earlier in diploids than tetraploids. This results in an overlap in flowering of only 51%. Such phenological separation may arise because of genetically based differences between the cytotypes as a consequence of chromosome doubling or selection subsequent to the chromosome doubling event. Alternatively, flowering asynchrony may arise if diploids and tetraploids occupy slightly different microenvironments in the population, which affect the timing and rate of growth. We feel that the latter possibility is unlikely to provide the whole explanation as phenological patterns similar to the results here have been observed in glasshouse (Husband & Schemske 1997) and common garden experiments involving allopatric diploid and tetraploid populations (B. C. Husband, unpublished data).

Separation in flowering between polyploids and their diploid progenitors has been documented in other species, both in the field and in common garden experiments, confirming that phenological differences often have a genetic basis (Lewis & Suda 1976; Lewis 1976; Lumaret et al. 1987; Felber 1988; Van Dijk 1991; Bretagnolle & Thompson 1996). Less clear, however, is why tetraploids in C. angustifolium flower later than the diploids. Although it has been suggested that higher DNA content and thus cell size may be associated with slower growth and later flowering (Stebbins 1950), cell size has also been associated with a greater capacity for growth at low temperatures and hence earlier initiation of flowering within each season (Grime et al. 1985; Grime & Campbell 1991; Bretagnolle & Thompson 1996). However, most studies in which polyploids are artificially generated from diploids show that polyploids flower later than their progenitors (reviewed in Ramsey & Schemske 1998). These conflicting predictions mirror the inconsistencies in the order of flowering between polyploids and their diploid progenitors described in empirical studies and highlight the difficulties in predicting the physiological and morphological effects of chromosome doubling.

Reproductive isolation between tetraploid and diploid C. angustifolium was enhanced by a tendency for pollinators to visit a disproportionate number of tetraploid stems and to fly primarily between flowers of the same cytotype rather than between cytotypes. Only 26% of all bee flights were between flowers on different inflorescences, and, of these, only 15% were between different cytotypes. In total, only 4% of all flights between flowers within the pollinator observation plot involved different cytotypes. This suggests that ethological isolation may play an important role in restricting gene flow between cytotypes and that this mechanism may operate even when taxa are located in close proximity. Insect foraging behaviour has been identified as an important mechanism of reproductive isolation (Grant 1949, 1994) and has been invoked as a major limit to natural hybridization between sympatric species (Grant 1952, 1994). However, the general significance of ethological factors for reproductive isolation in plants remains contentious, partly because of the lack of detailed studies of the barriers to hybridization (Arnold 1997). Currently, we are unaware of any investigations highlighting the role of ethological isolation in the evolution and maintenance of polyploidy in plants.

Although the bees we observed exhibited a visitation bias towards tetraploids and cytotype constancy, this behaviour is not necessarily a result of an active preference. Our computer simulations showed that a deficiency of between-cytotype flights would be expected as a result of the clumped distribution of cytotypes in the observation plot and the fact that most bee flights occurred between neighbouring stems. In other words, the non-random foraging behaviour between cytotypes may be a by-product of cytotype spatial structure, caused in part by clonal growth. In one bee species, however, a deficiency of between-cytotype flights remained, even after accounting for cytotype spatial structure, as did a disproportionate number of visits to tetraploids for all species. These non-random visitation patterns, after allowing for spatial structure, may reflect an ability to actively discriminate between cytotypes.

The multiple regression analysis helped to identify the characteristics of individual inflorescences that are associated with high visitation rates. Inflorescences that received the most visitations were consistently taller, and had larger petals and more open flowers compared to under-visited stems. This pattern was apparent when all plants were considered in the analysis, but also when diploids were examined separately. Variation in visitation among tetraploids was also related to inflorescence height and number of open flowers, but not to petal length. Inflorescence height, petal length and flower number were all greater in tetraploids than diploids (though not significantly so for flower number) and therefore may explain why visitation tends to favour tetraploid plants. The morphological characters that explain variation in the number of flowers visited per inflorescence varied depending on which plants were included in the analysis, but the number of open flowers consistently explained a significant portion of the variation. Since diploids and tetraploids did not differ significantly for this character, it is difficult to explain why bees visit more flowers on tetraploids than diploids. One possibility is that bees are responding to other characters which are correlated with flower number and whose means differ between tetraploids and diploids. Further examination of the differences in morphological display and floral rewards between tetraploids and diploids are necessary for a full understanding of the non-random patterns of insect visitation in C. angustifolium.

What are the impacts of flower phenology, insect foraging patterns and cytotype spatial structure on reproductive isolation in tetraploid and diploid Chamerion angustifolium? Our study indicates that each of these ecological factors can, individually, reduce opportunities for intercytotype mating, despite diploid and tetraploid cytotypes being in close proximity. Phenological separation restricted opportunities for intercytotype mating to 50% of the inflorescences produced. Even for individuals that flowered simultaneously, the opportunities for between-stem pollinations were restricted to 26% of foraging flights between flowers. Further, of the insect flights between flowers on different stems, only 15% were between different cytotypes. The cumulative effects of phenology and insect visitation, calculated as the product of all conditional probabilities (0.50 of flowers overlapping × 0.26 flights between stems × 0.15 between cytotype × 100%), suggests that opportunities for intercytotype mating are reduced to 2% of the total flights observed, considerably less than the 49% expected under random mating between the cytotypes. This value is, of course, an estimate of the ecological constraints on intercytotype mating and only a rough approximation of pre-mating reproductive isolation in C. angustifolium. The calculation is incomplete in that it does not consider the differential abilities of pollinators to transfer diploid and tetraploid pollen, nor does it incorporate the effects of pollen carryover or pollen competition on stigmas. Nevertheless, our observations suggest that intercytotype mating is well below random expectations, and therefore that a frequency-dependent mating disadvantage may not be a strong force acting on cytotype frequencies in sympatric populations.

Seed production in the mixed population varied as a function of flowering time. Diploid flowers that opened before any tetraploids had high seed production relative to diploid flowers open during the overlap period. This pattern of reproduction is expected if mating is random during the overlap period and triploids are inviable. However, seed set in tetraploids was high during the overlap period compared to flowers open during the tetraploid-only census. The fact that seed set decreased towards the end of the growing season suggests that reproduction was not influenced by intercytotype mating and triploid inviability in this site. Rather, the decline in seed set is likely to be due to low temperatures at the end of the season that may limit pollinator activity and reduce the likelihood of seed maturation. Because tetraploids flower later, they are more likely to experience low fertilization and possibly higher seed abortion than diploids.

In this study, we treated each stem in the observation plot as an independent unit. This is reasonable from the point of view of phenology, pollinator visitation and intercytotype mating, as each reproductive stem can potentially contribute diploid or tetraploid gametes. Nevertheless, the stems may actually represent ramets from one or a few genets within the observation plot, and thus has two major consequences: one biological and the other statistical. The biological consequence is that pollinations that occur as a result of intracytotype foraging may represent self-pollinations. Previous studies indicate that both cytotypes exhibit high inbreeding depression upon self-fertilization, and that this is particularly marked in diploids (Husband & Schemske 1997). However, because a disproportionate number of the observed flights occur between tetraploid flowers (within and between inflorescences), seed set in tetraploids may be influenced by inbreeding depression, especially during the period when only tetraploids were flowering.

The statistical consequence of studying a small number of genets within the observation plot is that the morphological differences observed between ploidies may largely be a product of sampling error. Furthermore, if the bias is strong our ability to predict ploidy level in the larger phenology plot may be jeopardized. However, the morphological differences measured are consistent with two other studies, one a glasshouse comparison and the other a common garden study, in which tetraploids were found to have taller inflorescences and generally larger flowers. Therefore, we are confident that the diploids and tetraploids measured within the observation plot are representative of the two cytotypes in this population.

Despite the observed ecological barriers to gene flow between diploid and tetraploid C. angustifolium, triploid offspring were still produced. Within the period of flowering overlap between diploids and tetraploids, 6.7% of all offspring examined were triploid. This value is comparable to the frequency of pollinator flights observed in the observation plot between cytotypes (4%) during the overlap period. Triploid production may be elevated somewhat because of the high frequency of tetraploids (57%), relative to the population as a whole (18%). This deviation confirms an observation made in a previous study (Husband & Schemske 1997) that cytotype frequencies are heterogeneous from patch to patch within the population. As a result of the predominance of diploids outside of the observation plot, foraging bees entering the plot are likely to be carrying predominantly diploid pollen, thereby increasing the rate of intercytotype mating. Nevertheless, the direct observation of triploid offspring confirms that gene flow between cytotypes does occur in C. angustifolium. The fact that reproductive isolation between ploidy levels is not complete has also been supported by direct observations in Dactylis glomerata (Lumaret & Barrientos 1990) and inferred from levels of gene flow in Rumex acetosella (den Nijs et al. 1985), Plantago (Van Dijk et al. 1992) and Dactylorhiza (Lord & Richards 1977). What is less clear is what the viability and fertility of these triploids may be in natural populations.

The estimate of triploid offspring production is likely to be an underestimate of the number of 2x−4x matings that actually occurred. We screened for triploids when the seedlings were at least at the 10-leaf stage, so there was sufficient material for the flow cytometric analysis. However, studies based on seed from controlled crosses (Burton & Husband, in press), indicate that seed maturation and germination rates of triploid seed are significantly lower than rates for diploid or tetraploid seed. The high rate of embryo mortality in triploids is likely to create a large disparity between the measured rates of intercytotype mating and the rate actually occurring at the time of fertilization. For this reason, we emphasize that the frequency of triploid offspring estimated in this study represents a ‘realized’ measure of mating between cytotypes, and reflects not only the rates of fertilization but also the relative fitness of triploids in the early life stages.

The patterns of intercytotype mating observed in this study have important implications for our understanding of what maintains the contact zone in C. angustifolium. Hybrid zone models can be classified into two types (Hewitt 1988; Harrison & Rand 1989) based on the nature of selection: those maintained by ecological sorting along some underlying environmental gradient (Endler 1977) and those maintained by genetic incompatibilities between cytotypes and selection against hybrids (Barton & Hewitt 1985). Similar environment-dependent (niche differentiation; Fowler & Levin 1984) and environment-independent (Minority exclusion; Levin 1975; Felber 1991) models have been described in the context of polyploid evolution. Distinguishing the role of these two types of mechanisms in the maintenance of mixed populations of C. angustifolium requires detailed information about the distribution, mating and relative fitnesses of diploid, tetraploid and triploid individuals. However, because of the low observed frequency of intercytotype mating, there appears to be little opportunity for endogenous selection against hybrids. Rather, coexistence of C. angustifolium cytotypes in close proximity may be evolutionarily stable because of the low rates of intercytotype gene flow and limited triploid seed production.

Acknowledgements

We thank R. Thorp for bee identification, H. Kubiw for editorial comments, M. Henry and T. Burton for technical assistance and the Natural Sciences and Engineering Research Council of Canada for financial assistance to B.C.H.

Received 12 August 1999revision accepted 10 January 2000

Ancillary