Michael L. Arnold Department of Genetics, University of Georgia, Life Sciences Building, Athens, GA 30602–7223, USA. Tel.: +1 706 5421407; fax: +1 706 5423910; e-mail: email@example.com
Pollinator preference may influence the origin and dynamics of plant hybrid zones. Natural hybrid populations between the red-flowered Iris fulva and the blue-flowered Iris brevicaulis are found in southern Louisiana. The genetic structure of these populations reflects a lack of intermediate genotypes. We observed pollinator behaviour in an experimental array with five plants each of I. fulva, I. brevicaulis, their F1, and the first backcross generation in each direction, to obtain data on flower type preferences and transitions between flower types. The most abundant visitors were Ruby-throated Hummingbirds (Archilochus colubris) and workers of the bumblebee Bombus pennsylvanicus. Hummingbirds visited I. fulva twice as often as I. brevicaulis and visited hybrids at intermediate frequencies. Bumblebee workers preferred the purple-flowered F1s and visited plants of I. fulva and the backcross to I. fulva more often than I. brevicaulis and its backcross. Overall, F1 flowers were visited most frequently. Both hummingbirds and bumblebees visited nearest neighbours in almost 80% of the interplant movements. This meant that a majority of movements were between different flower types, rather than between plants of the same type. Findings from the present study suggest that pollinator preference is not a major causal factor for the lack of intermediate genotypes in natural iris hybrid populations. Instead, pollinator behaviour in our array promoted mixed mating between flower types belonging to different pollination syndromes. However, owing to predominant nearest-neighbour visitation, the spatial distribution of parental and hybrid genotypes (in concert with pollinator behaviour) will have a strong influence on mating patterns and thus the genotypic structure and evolution of Louisiana iris hybrid zones.
This paper focuses on the behaviour of pollinators that visit the Louisiana iris species Iris fulva Ker-Gawler and Iris brevicaulis Raf. Natural hybridization between these species (and a third species, Iris hexagona Walter) takes place in southern Louisiana ( Viosca, 1935; Riley, 1938), and numerous hybrid populations have been found (e.g. Arnold et al., 1991 ). Iris fulva grows in shallow water on the edges of bayous, creeks and ditches ( Viosca, 1935; Cruzan & Arnold, 1993). It flowers in March–April, carries red flowers on tall, erect stalks and is pollinated by both hummingbirds and bumblebees (S. K. Emms & M. L. Arnold, unpublished data). Iris brevicaulis flowers in April–May and grows in dryer habitats (i.e. pasture edges and oak forests) relative to I. fulva ( Viosca, 1935; Cruzan & Arnold, 1993). The flowering stalk of I. brevicaulis plants is short and seldom extends above the foliage. The sky-blue flowers with yellow nectar guides are thought to be mainly visited by bumblebees ( Viosca, 1935). F1 hybrids between the two species have purple flowers and varying amounts of yellow in the nectar guide. F2 and backcross individuals show a wide variety of flower colours, ranging from the parental and F1 colours to pale pink, burgundy, dark blue, and deep purple, with or without a nectar guide (M. L. Arnold, personal observation).
Our objective in this paper is to assess the role of pollinator preference and behaviour in the structuring of Louisiana iris hybrid zones. We observed pollinator behaviour in a controlled situation, with equal numbers of flowers of the two parental species and three hybrid classes. Our goal was to identify: (1) which pollinator groups visit I. fulva, I. brevicaulis and their hybrids, (2) differences in flower type preference within and among pollinator groups, (3) spatial patterns of transition between plants in the array, and (4) transition frequencies between and within flower types. Knowledge of relative transition frequencies between types can be used to predict pollen deposition patterns and thus subsequent offspring formation. The types of offspring formed affect the genotypic structure and the form of selection that acts in hybrid zones in general ( Arnold, 1997). The current study thus allows a prediction of the expected range of genotypes in Louisiana iris hybrid zones based solely on pollinator behaviour. We can compare these expectations to known genotypic structures in contemporary hybrid populations to assess the relative role of pollinator selection in the genetic structuring of these contact zones.
Materials and methods
We used I. fulva (If), I. brevicaulis (Ib), their reciprocal F1s (F1(f) and F1(b), the subscript refers to the paternal parent), and first generation backcrosses to each species, Bf and Bb (Bf is F1(f) × If, Bb is F1(b) × Ib). Rhizomes of the parental species were collected from native populations in Louisiana and have been maintained in the Department of Botany greenhouses at the University of Georgia. Crosses between the species were performed to yield the F1 generation. Flowers on the resulting F1 plants were pollinated with a mixture of pollen from at least five genotypes of a parental species to yield the Bf and Bb individuals. In the autumn of 1997, rhizomes of at least 50 genotypes in each group were taken and repotted in 15-cm pots. The plants were transported to a site near Labadieville, Assumption Parish, in south-central Louisiana in March 1998. They were placed in a fenced pasture plot in their pots and were watered daily. The plants were large and often produced more than one flowering stalk, either simultaneously or consecutively. All plants were freely accessible to pollinators, and were often visited by hummingbirds and various insects.
We used a diamond-shaped array of 25 plants (five plants each of If, Ib, F1, Bf and Bb; distance between plants 80 cm). For the F1 type, we alternately used three F1(f) and two F1(b) or two F1(f) and three F1(b) plants, so that the two types of F1s were equally represented over the total observation period. The two F1-types look the same and did not differ in visitation frequency, and thus data from F1(f) and F1(b) individuals were pooled. The positions of each of the types were determined by slanting a 5 × 5 square over a 60° angle to obtain a hexagonal array. Each type was represented only once in each row and column, and plants in the centre of the array had six neighbours. We used whole plants in their pots, which mimic the natural situation more closely than cut stalks.
Plants used for the observations had one fresh flower that had opened that morning. All other flowers were removed and buds were covered with brown paper cones. Flowering stalks of these iris species usually carry just one or two open flowers at a time. For most plants we used one flower for not more than 5 h of observation. In a few cases (mainly I. brevicaulis), we used plants more often because there were insufficient new plants flowering. In total, we used 19 genotypes of I. fulva, 12 of I. brevicaulis, 10 of F1(f), 10 of F1(b), 20 of Bf and 17 of Bb.
Observations took place on 5, 6, 8, 9 and 11 May 1998, between 08.15 and 17.00 h local time (Central Daylight Time), for a total of 20 h. The time of arrival of a pollinator was noted to the nearest minute, and movements were recorded on audiotape. Iris flowers consist of three pollination units (Fægri & van der Pijl, 1979), each comprising one anther and two stigmatic lobes. We recorded the number of pollination units visited, and whether the unit visit was legitimate (anther or stigmatic lobes touched). We recorded the position of the plants that were approached or visited. Because the position of each plant was known, we could translate the position recordings to the plant type present on that position. The total duration of the array visit was determined to the nearest second. After each hour, the plants were moved to new positions and each genotype’s position was recorded. The relative positioning of types within the array was slightly changed after the first 10 h of observation, but types were still represented only once in each row and column of the square. Plant types were rotated among positions after each hour to prevent pollinators from learning the position of the different phenotypes.
The array was positioned in a pasture, ≈20 m away from where the irises were kept. The surrounding area contains a natural population of I. fulva. Plants of both I. hexagona and F1 hybrids between I. fulva and I. hexagona have been transplanted into this area in previous years ( Arnold et al., 1993 ; Hodges et al., 1996 ; S. K. Emms & M. L. Arnold, unpublished data). At the time the observations were performed, the only iris plants with flowers were those used for the present study, because the potted plants developed slightly slower than the wild irises. When the potted I. brevicaulis plants started to flower, the wild I. fulva plants at the study site had already ceased flowering. However, the area surrounding the array provided many alternative food plants, such as white clover (Trifolium repens) for bumblebees and thistles (Cirsium sp.) for hummingbirds and bumblebees. The iris plants that were not used in the array were also freely available for visitation. In that way potential visitors could gain experience with all iris flower types.
In the remainder of this paper, we use ‘approach’ for all instances where a pollinator came close to a flower. The flower could then be rejected or visited. For insect approaches, we distinguished rejection with and without touching the flower. We further analysed the subset of complete array visits in which we had observed the arrival, all movements within the array and the voluntary departure of the focal animal. This excluded approaches by hummingbirds that were interrupted by the attack of another hummingbird, and visitation sequences in which we had missed some part. For these complete array visits we calculated the average duration, the mean number of flowers approached and visited, the mean fraction of flowers rejected, and the mean number of flowers approached and visited per unit time. We used t-tests to analyse differences in these variables between the two pollinator groups (hummingbirds and bumblebee workers) for which we had sufficient data. For the fraction of flowers rejected, data were arcsine-square root transformed prior to testing.
For all further analyses we used data from all observed array visits in which at least one flower was visited. Pollinator preferences within pollinator groups (deviations from equal visitation to all flower types) were tested with χ2-tests on 5 × 1 contingency tables. Differences in preference among pollinator groups (hummingbirds, bumblebee workers and bumblebee queens) were tested with a χ2-test on a 5 × 3 contingency table. χ2-tests were also used for determining whether pollinators differentially rejected flower types. Differences between flower types within pollinator groups were tested with pairwise χ2-tests, and the resulting P-values were corrected for multiple comparisons in a sequential Bonferroni procedure ( Holm, 1979; Rice, 1989).
Spatial flight patterns were analysed for the two largest pollinator groups (bumblebee workers and hummingbirds) by determining the distribution of distances flown between flowers, and testing this against the expected frequencies of these distance classes based on their frequency of occurrence in the array. We performed separate analyses for all approached flowers and for visited flowers only. The spatial pattern of arrivals (first flower approached) was studied for flights where the arrival of the pollinator had been observed.
We analysed for bumblebee workers and hummingbirds separately whether pollinators showed constancy (i.e. the tendency to visit flowers of the same type consecutively) by comparing the frequency of possible same-type transitions within each distance class with the observed transitions between plants of the same type. The differences were tested using the binomial distribution.
Pollinators and pollinator behaviour
In 20 h of observation, we recorded 240 approaches to the array, 228 of which resulted in at least one visit to a flower (Table 1). The most numerous visitors were workers of the bumblebee Bombus pennsylvanicus and female Ruby-throated Hummingbirds (Archilochus colubris). Additional approaches were made by queen bumblebees, male Ruby-throated Hummingbirds, female carpenter bees (Xylocopa sp.), and butterflies of at least three unidentified species, most frequently by a skipper (Hesperiidae). All flower visits by this skipper species were illegitimate, since the animals probed for nectar at the base of the sepals, thereby avoiding contact with anthers or stigmatic lobes. Nectar robbing was also observed in hummingbirds. Ten flower visits were completely illegitimate, and in 18 instances the hummingbird visitor robbed nectar but also made legitimate visits to one or more pollination units. Ib and Bb flowers were most frequently subject to nectar robbing by hummingbirds; of the 38 flowers robbed, 15 were of the Ib type and 17 of the Bb type. The remaining flowers belonged to F1 (n=5) and Bf (n=1).
Table 1. Visits to the array by each of five pollinator groups during 20 h of observation. For complete array visits only (all movements of a pollinator recorded and at least one flower legitimately visited), we calculated the number of flowers approached, the number of flowers visited, the rejection rate per array visit, and the flower approach and visitation rates. All means are given with ±1 SD. Different superscript letters indicate significant differences between hummingbird females and bumblebee workers.
Array visits by the two most numerous groups of pollinators, bumblebee workers and hummingbird females, differed significantly in many respects. Bumblebee workers made longer array visits (t125=6.981, P < 0.0001), but visited fewer flowers per array visit (t125=–2.361, P=0.0198), because they rejected a larger proportion of flowers approached (t125=12.196, P < 0.0001). Bumblebee workers were much slower than hummingbirds, both in number of flowers approached per minute (t125=–9.066, P < 0.0001) and in number of flowers visited per minute (t125=–11.380, P < 0.0001).
Pollinators often visited more than one pollination unit on a flower. On average, bumblebee workers visited 1.86 units, bumblebee queens 1.63 units, and hummingbirds 1.91 units per flower visited. We found only slight differences in number of units visited among flower types (data not shown).
Figure 1 gives the distribution of flower approaches and resulting visits and rejections over plant types for each of three groups of pollinators. Hummingbird males did not differ from females, and the two were grouped into one category for subsequent analyses. Patterns of approaches to flower types were significantly different from equality for all three pollinator groups (χ2=50.395, d.f.=8, P < 0.0001). Hummingbirds preferentially approached flowers on the I. fulva side of the spectrum (χ2=57.381, d.f.=4, P < 0.0001). Bumblebee workers had a peak in approaches to Bf and F1 flowers (χ2=78.750, d.f.=4, P < 0.0001), while bumblebee queens avoided I. fulva flowers completely (χ2=27.437, d.f.=4, P < 0.0001).
The rejection rate of approached flowers differed significantly among pollinator groups (χ2=358.66, d.f.=2, P < 0.0001), as was also shown in Table 1 for complete visits only. Bumblebee workers rejected 47.4% of the flowers they approached, compared to 23.3% for bumblebee queens and 10.6% for hummingbirds. Bumblebees (overall χ2=32.364, d.f.=4, P < 0.0001) had lower rejection rates on F1 flowers and a higher rate on flowers of I. brevicaulis ( Fig. 1A). Bumblebee queens (overall χ2=19.856, d.f.=3, P=0.0002) had the highest rejection rate on Bf flowers ( Fig. 1B). Hummingbirds showed significant overall heterogeneity in fraction rejected (χ2=12.074, d.f.=4, P=0.0168). Rejection rates for hummingbirds appeared to be lowest on If and F1 flowers, but none of the pairwise comparisons was significant.
Like the number of approaches, the number of flowers visited per flower type was significantly different among pollinator groups (χ2=74.112, d.f.=8, P < 0.0001). Bumblebee workers preferentially visited F1 and Bf flowers (χ2=97.493, d.f.=4, P < 0.0001), while If, Bb and especially Ib flowers were visited less than expected. Bumblebee queens (χ2=30.684, d.f.=4, P < 0.0001) strongly preferred Ib, Bb and F1 flowers and did not visit I. fulva. Hummingbirds visited If, Bf and F1 flowers more than Bb and Ib flowers. When the numbers of visits to each flower type during the whole observation period are combined for all pollinators ( Fig. 1), F1 flowers received most visits, followed by Bf and If. Bb and Ib had the lowest number of visits. Variance in visitation rates to individual genotypes did not differ among types: the visitation rates of the pure species and the F1s were as variable as those of the backcross plants (data not shown).
To determine whether certain phenotypes were differentially attracting pollinators to the array, we compared the distribution of first flowers approached with that of all flower visits by a pollinator group ( Fig. 1). In bumblebee workers, the first approach preferences were not significantly different from overall preference (χ2=6.360, d.f.=4, P=0.1738; Table 2), but hummingbirds approached F1 more and Ib flowers less often than expected from overall preferences (χ2=10.037, d.f.=4, P=0.0398). Results were similar for first flower visited (Table 2).
Table 2. Frequency of flower types for first flower approached and first flower visited. For frequencies of flower types in all approaches and visits, see Fig. 1.
Spatial visitation patterns
In our 5 × 5 array, a total of 625 movements are possible, including those in which the pollinator returns to the flower it just visited. These movements can be grouped into 15 different distance classes (Table 3). We tested the distribution of interplant distances of each of the pollinator groups against the frequency distribution of all possible interplant flights. We found a highly significant deviation for both groups ( Fig. 2). When flights between approached flowers were considered, bumblebee workers (χ2=2780.025, d.f.=14, P < 0.0001; Fig. 2) and hummingbirds (χ2=2254.116, d.f.=14, P < 0.0001; Fig. 2) showed a preponderance of nearest neighbour (distance class A) flights (≈80% of total interplant movements). For flights between accepted flowers only, the pattern for hummingbirds remained largely the same (χ2= 10000801.588, d.f.=14, P < 0.0001; Fig. 2). For bumblebee workers, the frequency of longer distances increased at the expense of nearest neighbour flights, but the bias towards shorter distances was still highly significant (χ2=582.715, d.f.=14, P < 0.0001; Fig. 2). The increase in distance for bumblebee workers can be explained by their higher rejection rate, which resulted in longer distances covered before a flower was accepted.
Table 3. The 15 distance classes in the array, with their distance in metres (d) and their frequency of occurrence (n).
Transitions among flower types
The predominance of nearest neighbour movements had consequences for the analysis of transitions between flower types. The availability of types varied among distance classes as a result of the design. For instance, the probability of encountering a flower of the same type in flights to nearest neighbours was only 4.5% (averaged over the first 10 h and last 10 h of observations, which had slightly different flower type distributions), compared to 125/625=20% for all possible movements in the array. If we want to test for constancy (i.e. a deviation from random movement in favour of within-type transitions) we need to regard distance classes separately and use the expected values for the distance class under study. Because the array layout in the first 10 h differed from that in the latter half of the observations, we analysed these two periods separately as well. The results of these analyses are shown in Table 4. In the first 10 h, none of the observed frequencies of same-type transitions was significantly different from expectation. In hours 11–20 we did find significantly more same-type transitions than expected for bumblebee workers in distance class D and for hummingbirds in distance classes A and B. The latter finding is especially important, because these two distance classes comprised the great majority of visits (83.0%). However, the increase in frequency was quite small (hummingbirds class A: 5.4% vs. 8.9%), indicating that the inclination towards constancy was low. After Bonferroni correction for multiple comparisons, only the P-value for hummingbirds in hours 11–20 remained significant (corrected P=0.0364).
Table 4. Occurrence of sequential visits to the same flower type (within-type transitions) for each set of observation hours (1–10 and 11–20) within each distance class (A through G) for bumblebee workers and hummingbirds. We used the binomial distribution to test whether the observed frequency of within-type transitions was significantly larger than expected from random movement. The one-sided binomial P-value is given after the observed numbers, with significant values in bold type.
Pollinator choice in hybrid populations
All types of flowers, both parental species and hybrids, were visited by all pollinator groups except bumblebee queens. Our results suggest that reproductive isolation by pollinator choice is weak in Louisiana iris hybrid zones, especially once hybrids are present. We thus found no evidence for pollinator-mediated selection acting against the F1 hybrids in our experimental arrays. Instead, the F1 individuals were apparently favoured relative to their parents and the backcross progeny. Pollinator visitation studies involving two species and their F1 hybrids have also been performed for Mimulus ( Sutherland & Vickery, 1993), Ipomopsis ( Campbell et al., 1997 ; Meléndez-Ackerman et al., 1997 ; Meléndez-Ackerman & Campbell, 1998), Silene ( Goulson & Jerrim, 1997), Raphanus ( Lee & Snow, 1998) and Baptisia ( Leebens-Mack & Milligan, 1998). In all of these studies, F1 hybrids were visited at either intermediate rates relative to their parents, or as often as the most visited parent. In a study on Mimulus species, hybridization occurred between the hummingbird-pollinated species M. cardinalis and the bumblebee-pollinated M. lewisii. In arrays with both species and their F1 hybrid, hummingbirds visited M. cardinalis more than M. lewisii, and visitation rates to hybrids were intermediate ( Sutherland & Vickery, 1993). Bumblebees completely avoided M. cardinalis, but did not distinguish between M. lewisii and the F1 hybrid. This suggests that differences in floral colour and shape do influence pollinator choice, but that these differences will not be sufficient to guarantee reproductive isolation (but see Hiesey et al., 1971 ; Bradshaw et al., 1998 ), and even less so once F1s have been formed. Pollinator preferences for specific character syndromes appear to be less stable than previously thought ( Waser et al., 1996 ; Goulson & Jerrim, 1997; Waser, 1998), and the availability and quality of rewards seem to be more important in determining floral choice ( Pleasants & Waser, 1985; Waser, 1998). Iris fulva × I. brevicaulis F1 hybrids had the highest nectar production rates of all flower types involved in the present experiment (R. A. Wesselingh, unpublished data), which may in part explain their attractiveness to pollinators.
Pollinator movements and genotypic distributions in iris hybrid zones
The findings from the present analysis suggest that the spatial arrangement of genotypes may influence patterns of pollen transfer. Almost 80% of the movements by hummingbirds and bumblebee workers in our array were between neighbouring plants. Aggregation of similar genotypes may thus lead to positive assortative mating. For example, the hybrid zone studied by Cruzan & Arnold (1993, 1994) showed a steep genotypic cline on the boundary between a forest with mainly I. fulva-like plants and a pasture with predominantly I. brevicaulis-like genotypes. Even when plants are flowering simultaneously, their spatial distribution will likely limit pollen flow between different flower types. Ecological associations ( Cruzan & Arnold, 1993) cause further spatial separation of genotypes and increase the likelihood of positive assortative mating. Alternatively, the abutment of habitats in a mosaic landscape ( Howard, 1982, 1986; Harrison, 1986) could result in alternate genotypes being nearest neighbours, increasing the probability of negative assortative mating (i.e. preferential mating between alternate genotypes).
Analyses of the genetic structure of I. fulva ×I. brevicaulis hybrid zones ( Cruzan & Arnold, 1993, 1994) have uncovered a strongly bimodal genotypic distribution in adult plants, with an almost complete lack of intermediate genotypes. The deficit of intermediate, recombinant genotypes was reflected in significant nuclear and cytonuclear disequilibria estimates. Significant nuclear and cytonuclear disequilibria were also detected in seeds produced during a period of overlap in flowering of I. fulva-like and I. brevicaulis-like plants ( Cruzan & Arnold, 1994). However, there was significantly less disequilibrium in the seed population relative to the parental plants. This indicates that pollen transfer and fertilization occurred between plants at either end of the genotypic distribution. Pollinators were thus transferring pollen between plants with different pollination syndromes. The presence of intermediate hybrid genotypes in the seeds, which are absent in the adult plants, apparently reflects selection against hybrid establishment, rather than the influence of pollinator behaviour.
Our analysis utilized equal numbers of individuals of the parental and hybrid classes. The most likely scenario for the formation of hybrid populations in Louisiana irises involves one or just a few F1s established in a large population containing one or both parental species ( Arnold et al., 1993 ; Hodges et al., 1996 ). If pollinators favour (or do not select against) the F1 individuals, as found in the present study, the evolution of the hybrid zone will proceed. A low frequency of F1 flowers may lead to a decrease in visitation rates ( Nagy, 1997), but this is not a general rule ( Epperson & Clegg, 1987; Lee & Snow, 1998). Furthermore, if nectar reward is lower in the more common flower types (e.g. the parental species in a newly formed hybrid zone), pollinators will switch more often to the alternate, rare flower type ( Smithson & Macnair, 1997).
We recorded a total of 1622 flower visits in 20 h of observation. This gives an average of 3.24 visits per flower per hour (range 0–17). It is likely that the visitation rates to our array are higher than would be observed in the wild, although we have no information on visitation rates in natural iris populations. A high visitation rate can lead to a depletion of rewards, and pollinators may turn away from some flowers in search of more rewarding ones. However, irises produce copious amounts of nectar (between 4 and 8 μL h– 1; R. A. Wesselingh, unpublished data), and a high production rate can maintain a flower’s attractiveness even when visitation rates are high. Furthermore, some bumblebee workers foraged only for pollen on I. fulva flowers, which left the nectar reward intact for subsequent visitors. Finally, visitation rates increased with time, with over twice as many visits on day 4 than on day 1, but rejection rates of bumblebee workers did not increase, as would be expected if flowers were unrewarding. Thus, we have no indication that reward depletion played a role in determining pollinator preferences.
Overall preferences recorded in this paper resemble those from a previous study of I. fulva, I. hexagona and their F1 hybrids (S. K. Emms & M. L. Arnold, unpublished data). This study was performed at the same site in 1996 during the earlier, natural flowering season of I. fulva and I. hexagona. The preferences of each of the pollinator groups (hummingbirds, bumblebee workers, bumblebee queens) were comparable to what we found (replacing I. hexagona for I. brevicaulis), but the relative frequency of pollinators differed due to the difference in seasonal timing. In the 1996 study, hummingbird visits were most abundant at the site, followed by bumblebee queens and bumblebee workers. We observed almost equal numbers of approaches by hummingbirds and bumblebee workers, and only a few bumblebee queens. The higher number of bumblebee workers in our study resulted in an overall preference for F1 flowers. Iris brevicaulis × I. fulva F1 hybrids resemble I. fulva flowers more than I. hexagona×I. fulva F1s (R. A. Wesselingh, personal observation). Emms and Arnold (unpublished data) found lower visitation rates by hummingbirds on F1 hybrids. The difference between their results and those of the present study may thus be due to the difference in F1 flower morphology.
Transitions among flower types
We did not test for flower constancy sensuWaser (1986), but we found an excess of same-type transitions in distance classes A and B for hummingbirds in the second half of our observation period (Table 4). The tendency to visit flowers of the same type thus seemed to be present, but it was by no means a consistent pattern. Absence of constancy was also found for hummingbirds in arrays of Ipomopsis aggregata, I. tenuituba and intermediate hybrids ( Meléndez-Ackerman et al., 1997 ). Similarly, addition of F1 hybrids to mixed arrays of Baptisia leucophaea and B. sphaerocarpa reduced the level of constancy so that a previously identified preference for B. sphaerocarpa disappeared ( Leebens-Mack & Milligan, 1998). In a natural population with different flower colour morphs of Ipomoea purpurea, Brown & Clegg (1984) found high constancy in nearest-neighbour flights. Non-nearest-neighbour flights, however, which comprised two-thirds of the total number of visits, had much lower and mostly nonsignificant constancy values. It appears that visitation sequences by pollinators are not regulated by a fixed degree of constancy, but rather are a result of a flexible response to the number of flower types/colours and their relative abundance. In well-mixed populations of several flower types, constancy is abandoned for a strategy that is more time- and energy-efficient: nearest-neighbour visitation with only a low degree of within-type flights.
Flower morphology and pollen transfer
Both hummingbirds and bumblebee workers appear to take up and deposit pollen in all flower types. The stigmatic lobes in iris flowers are located at the end of the stylar arms, right at the entrance of the channel formed by a stylar arm and a sepal. All flower visitors of sufficient size would come into contact with the stigmatic surface upon entering the flower, if the lobes have unfolded. Although the narrow channel of the Iris fulva flowers did not normally allow bumblebees to reach the nectar, the workers often foraged for pollen in the upper part of the channel. The morphology of F1 flowers approached that of I. fulva flowers, but with a wider channel that allowed bumblebees to crawl all the way down to reach the nectar. Only I. brevicaulis-like flowers had a wide enough channel to allow passage by bumblebee queens. Hummingbirds would use their feet to push down the sepal of these flowers and enter the otherwise closed flower of I. brevicaulis with their forehead pushed against the top of the stylar arm, where the stigmatic lobes are located. The actual efficiency of pollen uptake and deposition by each of the pollinator groups is unknown, but flower morphology and pollinator behaviour make it plausible that at least some pollen is transferred by both hummingbirds and bumblebees from and to all flower types studied.
Factors affecting the genetic structure of I. fulva×I. brevicaulis hybrid zones
Future studies will address the effects of pollinator choice in situations that will include flowering phenology of the plant types used in this study. We plan to quantify pollinator visitation through time in a seminatural mixed population and estimate paternity for seed progeny. This will enable us to further assess the relative importance of pre- and postpollination processes in shaping the genotypic distributions in Iris hybrid zones.
We thank Wayne Talbot for kindly granting us permission to work on his property, Andy Tull and Mark Zimmerman for taking care of the irises in the greenhouse, Jacob Vogel for his assistance in the field, and Amy Bouck, Mark Bulger, John Burke, Kent Holsinger, Jill Johnston, Ed Kentner, James Mallet, Joe Williams, and anonymous reviewers for useful comments on previous versions of the manuscript. This research was supported by the American Iris Society Foundation and by National Science Foundation grant DEB 9703853 to M.L.A.
Present address: Unité d’Écologie et de Biogéographie, Université catholique de Louvain, Croix du Sud 4-5, B-1348 Louvain-la-Neuve, Belgium.