Variable selection in Platanthera bifolia (Orchidaceae): phenotypic selection differed between sex functions in a drought year

Authors


Johanne Maad, Department of Biology, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway.
Tel.: +47 73550733; fax: +47 73596100;
e-mail: johanne.maad@bio.ntnu.no

Abstract

We estimated selection on three morphological characters in the hermaphroditic, hawkmoth-pollinated orchid Platanthera bifolia and explored selection surfaces through male and female function. The work was carried out in northern Sweden during two flowering seasons (1994 and 1995) in one natural population and one season (1995) in another natural population. Fitness was estimated as number of pollinia removed (male function) and number of fruits produced (female function). We detected directional selection towards larger inflorescence size (flower number) through both sex functions in both populations in 1995. In 1994, with an unusually dry growing season, 78% of the individuals failed to set any fruit, and there was selection for larger inflorescences only through male function. In this year, there was selection towards longer flower spurs, which could be a direct or indirect effect of spurs being shortened by drought. The results demonstrate that selection patterns may vary temporally and spatially, and that the ‘male function hypothesis’ may be applicable as female function is more resource dependent than male function.

Introduction

The relative importance of the male parent in flowering plant evolution has been recognized in the ‘male function hypothesis’ (Burd & Callahan, 2000; see also Janzen, 1977; Willson, 1979). This stems from the fact that reproductive traits influence not only seed production (female function) but also siring success (male function) and may influence the two sex functions in different ways (Bell, 1985; Campbell, 1989; Young & Stanton, 1990; Stanton et al., 1991; Queller, 1997; Maad, 2000). The extent to which male and female functions relate differently to reproductive characters influences functional gender of individuals and is believed to have impact on sex allocation and evolution of mating systems (Zhang & Wang, 1994; Elle, 1999; Burd & Callahan, 2000; Campbell, 2000; Elle & Meagher, 2000). The male function hypothesis predicts that, in plants that are not pollen limited, a portion of flowers will export pollen but not produce seeds (Burd & Callahan, 2000). This enables stronger selection through male function than through female function on pollinator attraction characters such as flower number and size (Bell, 1985; Campbell, 1989; Burd & Callahan, 2000). Conversely, in pollen-limited plants, selection may act more strongly on pollinator attraction characters through female than male function (Wilson et al., 1994; Johnson, 1996). However, Stanton (1994) found evidence through theoretical modelling that the male function hypothesis may be relevant even at intermediate pollination levels.

Shape and strength of selection on floral characters have often been found to vary temporally within or between years, and spatially within or among populations (e.g. Kalisz, 1986; Campbell, 1991; Eckhart, 1991; Gómez & Zamora, 2000; Galen, 2000; Totland, 2001). A varying selection pattern may be because of direct and indirect effects of abiotic and biotic factors, such as weather conditions, resources, herbivores and pollinators (e.g. Bennington & McGraw, 1995; Wilson, 1995; Totland & Eide, 1999; Caruso, 2000, 2001; Herrera, 2000). Pollinator and resource limitation of seed set have been found to vary within and among seasons and populations in several plants species (Burd, 1994). Female fitness should generally be more constrained by resources than male fitness (cf. the male function hypothesis; Burd & Callahan, 2000), and selection on characters such as flower number and size should therefore differ more between the sex functions during years when seed set is resource limited. There are, as far as we know, no studies that have investigated how environmental factors, like water availability, may affect selection differently through the two sex functions in natural plant populations.

The present paper documents selection that differs between sex function and varies among years and populations. The study plant chosen was the hawkmoth-pollinated orchid Platanthera bifolia. Selection in a natural population of this plant has previously been found to be strong for larger inflorescence sizes through male and female function (Maad, 2000). In the present study, we aimed at finding out if patterns of selection may vary among populations and years, and differ between sex functions. The specific questions asked are: (i) Is fruit set pollen limited? (ii) Does the shape or strength of selection on morphological characters differ temporally and spatially? (iii) Is selection stronger on pollinator attraction characters through male than female function when the population is resource limited (cf. the male function hypothesis)?

Materials and methods

Plant species

Platanthera bifolia (L.) L. C. Rich. is a terrestrial orchid with a wide Eurasian distribution (Hultén & Fries, 1986). The flowering period of P. bifolia in North Europe is June–July, and the plant occurs in a variety of forest and meadow habitats (Davies et al., 1983; Hultén & Fries, 1986). The plant is a polycarpic perennial, but most individuals do not flower in two successive years. Typically, it produces two oval leaves at the base of a single 20–60 cm high flower-bearing stem. The inflorescence spike displays 10–20 white, nocturnally fragrant, c. 1.5–2 cm wide flowers. Each flower has a slender 2–4 cm long spur as a backward extension of the lip petal. The spur contains nectar as reward for nocturnal insect pollinators, particularly moths of the families Sphingidae and Noctuidae (Nilsson, 1983). The column has two two-lobed pollinia (packages of pollen), and each pollinium is attached by a short stalk (caudicle) to a sticky viscidium exposed laterally in front of the spur entrance. The relatively short distance between the two viscidia allows them to attach to the base of the proboscis of the pollinator (Darwin, 1862; Nilsson, 1985). When a pollinium-bearing moth inserts its proboscis deep into the spur, parts of the carried pollinia become deposited on the stigma, which is situated around the spur mouth (Darwin, 1862; Nilsson, 1988).

The study areas

Information on all study populations is listed in Table 1. Four P. bifolia populations were selected to investigate if fruit set is pollen limited. Three of the four populations were located in the province of Västerbotten, northern Sweden, and one in the province of Uppland, central Sweden. Hand pollinations were performed in 1993, 1994 and 1997 (Table 1). All four study sites were coniferous forests [Pinus sylvestris L. and/or Picea abies (L.) Karst.] and the P. bifolia individuals (c. 50–100 inflorescences per population) were not highly exposed to sun and wind.

Table 1.  Information on the seven Swedish study populations of Platanthera bifolia.
PopulationProvincePositionStudy yearsType of study
  1. *Data from Hönsarvsberget included in ancova of the effect of population and year on selection pattern (see Materials and methods).

ErsmarksbergetVästerbotten63°56′N, 20°17′E1993Pollen limitation
KulbäckslidenVästerbotten64°10′N, 19°35′E1994Pollen limitation
BalbergetVästerbotten63°56′N, 19°08′E1994Pollen limitation
HammarskogUppland59°47′N, 18°35′E1997Pollen limitation
LillbergetVästerbotten64°00′N, 19°27′E1994, 1995Selection
Yttre HembergetVästerbotten63°54′N, 19°54′E1995Selection
HönsarvsbergetDalarna60°31′N, 15°33′E1993–1995Selection*; see Maad (2000)

Two populations located in Västerbotten (northern Sweden), were chosen for the selection study (Table 1). Lillberget was censused in both 1994 and 1995. The population consisted of 151 and 182 flowering individuals in 1994 and 1995, respectively, of which 37 individuals flowered both years. After exclusion of damaged individuals, 108 and 101 individuals were included in the analyses of selection in 1994 and 1995 respectively. The study area was an open logged Pinus sylvestris forest and the P. bifolia individuals were exposed to strong sun and wind. The other selection study population was located at Yttre Hemberget. This population was censused in 1995 and consisted of 95 flowering individuals, of which 84 nondamaged individuals were used in the selection analyses. This site was less exposed to wind and sun than the Lillberget population. The study area was dominated by spruce–pine (Picea abies, Pinus sylvestris) forest mixed with open, herb-dominated patches, where the majority of P. bifolia individuals were growing.

Selection in the Hönsarvsberget population (located in the province of Dalarna, central Sweden), was estimated by Maad (2000). To generalize results, present data from Lillberget and Yttre Hemberget was analysed together with the old data from Hönsarvsberget (ancova, see below).

To be able to relate plant characters and fitness components to environmental factors, we reported weather data from three local weather stations (Table 2). In the flowering season of 1993 the weather was wet in the provinces of Dalarna (central Sweden) and Västerbotten (northern Sweden), but the temperature was quite normal. The summer of 1994 was unusually dry and hot in most parts of Sweden (SMHI, 1994), and many areas suffered severely from drought (J. Maad and R. Alexandersson, personal observation). In Västerbotten the precipitation in July 1994 was c. 1/6 of the normal and the mean temperature was more than 2 °C higher than normal – at that date the hottest July of the 20th century (SMHI, 1994). In Dalarna the precipitation in July this year was c. 1/3 of normal and the temperature was more than 4 °C higher than normal. In 1995, the flowering season was quite normal with regard to temperature and precipitation in both provinces. In 1997 the July weather in the province of Uppland (central Sweden) was warmer and somewhat dryer than normal.

Table 2.  Weather data from three Swedish weather stations (SMHI, 1993, 1994, 1995, 1997). Normal values are means from 1961 to 1990.
Weather stationProvinceJuly mean temperature (°C)July precipitation (mm)
1993199419951997Normal1993199419951997Normal
  1. *Normal mean from 1965 to 1990.

FalunDalarna15.320.416.216.0107245676
Umeå flygplatsVästerbotten15.117.414.215.2*11185553*
UppsalaUppland19.216.5 5375

Pollinators have not been seen in the study sites located in Västerbotten but the hawkmoth Hyloicus pinastri (L.) has previously been observed pollinating P. bifolia in another population of Västerbotten (T. Nilsson and R. Alexandersson, unpublished data). This pollinator species has also been observed in Hammarskog, Hönsarvsberget, and other populations in central Sweden (J. Maad and L. G. Reinhammar, unpublished data; Maad, 2000).

Supplemental hand pollinations

To investigate whether fruit set of P. bifolia is pollen limited or not, supplemental hand pollinations were performed in four populations. In all four populations a subsample of 10–20 individuals was randomly chosen for the supplemental hand-pollination treatment. Cross-pollen (from individuals >2 m away) was deposited on stigmas of all flowers of the chosen individuals. In Hammarskog, all study individuals (treated and controls) were cut to leave 10 flowers, while all flowers were left in the other three populations.

Traits investigated

To estimate pollinator-mediated selection through male and female function, a number of potentially explanatory morphological variables and fitness measures were censused on the study individuals of the two populations chosen for the selection study (Lillberget and Yttre Hemberget). There were no individuals with more than one inflorescence. Flower number was the total number of flowers opened. Plant height was the distance from ground to the top of the highest flower. Stalk length was the distance between ground and the base of the ovary of the lowest opened flower. Both were measured to the nearest 5 mm when the uppermost flower of the plant had opened. Spike length was calculated as the difference between plant height and stalk length. Spur length was the distance between the spur tip and the spur mouth (at the inner border of the stigma) and was measured to the nearest 0.1 mm on the second lowermost flower (Maad, 2000).

Plant fitness was estimated as pollination success in terms of pollinium removal and fruit set. The number of pollinia removed was determined by examining all flowers at the end of flowering. The number of fruits set was recorded c. 6 weeks after flowering (early September). Absolute fitness was determined as total number of pollinia removed (male component) and total number of fruits set (female component) per inflorescence (individual). Relative fitness (male and female component) was the absolute fitness divided by mean of absolute fitness of all censused individual in that population and year (e.g. Endler, 1986). Percentage of pollinium removal (pollinium removal to flower ratio × 50; as there are two pollinia per flower) and percentage of fruit set (fruit to flower ratio × 100) was calculated for each individual.

Statistical analyses

The supplemental hand-pollination data were analysed separate for each population. Percentage of fruit set was compared between control and hand-pollinated individuals by using Mann–Whitney U-test (Conover, 1980).

Character mean and standard deviations were calculated for each population and year. Confidence intervals (CIs) of mean values were estimated by using a bias-corrected bootstrap method (see below). By using CIs, we compared mean values of characters in the 2 years (1994 and 1995) censused in Lillberget, and the two populations (Lillberget and Yttre Hemberget) censused in 1995. If two CIs were nonoverlapping the variables were considered as significantly different. The opportunity for selection was estimated as the variance in male and female relative fitness (Brodie et al., 1995). CIs (generated with a bias-corrected bootstrap method, see below) were used to determine if variances differed between years or populations.

To study which characters were undergoing selection, we estimated selection gradients by using multiple regression analysis (Lande & Arnold, 1983). The selection gradients of characters are coefficients in a regression of fitness on characters measured as deviations from their respective mean values (Phillips & Arnold, 1989). In the present study, flower number and stalk length represent aspects of plant size, and spur length represents floral morphology. These variables were not strongly intercorrelated (0.08 < r < 0.41). Plant size has been found to be under strong phenotypic selection in many plant species (e.g. Campbell, 1989; Johnston, 1991; Fritz & Nilsson, 1996; Maad, 2000). However, there are far fewer documented cases of selection on flower depth (Nilsson, 1988; Schemske & Horvitz, 1989; Steiner & Whitehead, 1990; Maad, 2000) and selection on this character has rarely been found to be strong (Alexandersson & Johnson, 2002).

Preliminary analyses of the data set revealed that there were neither curvature (e.g. standardizing and disruptive; Phillips & Arnold, 1989) nor correlational (i.e. on character combinations; Sinervo & Svensson, 2002) selection present on the three chosen characters. As directional gradients are usually estimated without including quadratic terms (curvature and correlational) due to multivariate non-normality of data (Lande & Arnold, 1983; Brodie et al., 1995), we did not include any quadratic terms in the final selection analyses.

To find general trends of selection in P. bifolia and to see if selection varies temporally and/or spatially we combined the data from Yttre Hemberget and Lillberget with data from Hönsarvsberget (Maad, 2000). The data was analysed with ancova to estimate the effects (main effects and interactions) of population, year, and characters on male and female absolute fitness per inflorescence. To avoid dependence among years within a population, individuals that were represented more than once were included only the first year of flowering (Maad, 2000). Highly insignificant variables (P > 0.50) were removed from the final analyses.

As most variables were non-normally distributed, CIs of mean values, variances and regression coefficients (selection gradients) were estimated by using a bias-correcting bootstrap method (Dixon, 1993). This method is refined from the percentile bootstrap method and is useful especially if the data is skewed and a few data points have a large influence on the estimate. We generated 10 000 bootstrap samples from the original data set. Each bootstrap sample was obtained by re-sampling by replacement and had the same number of observations as the original data set. The statistics (mean values, regression coefficients, etc.) were estimated for each bootstrap sample to generate a bootstrap distribution. Then we determined the fraction (F) of bootstrap estimates that were smaller than the observed value (estimated statistics in the original data set), and calculated the probit transform of F: z0 = Φ−1(F) where Φ−1 is the normal inverse cumulative distribution. The bias-corrected percentiles (lower and upper) of a 95% CI were then calculated as: Pl = Φ(2z0 − 1.96) and Pu = Φ (2z0 + 1.96), where Φ is the normal cumulative distribution function. Finally, the bounds were given by the values in the bootstrap distribution that matched the calculated percentiles Pl and Pu. The P-values of the selection gradients were estimated by using the same method.

Probability values in the ancovas (on the combined data set) were estimated by using a randomization method (Crowley, 1992). P-values from such a method are more accurate than those from the corresponding parametric test when data are from a non-normal distribution, and relate to the probabilities of getting similar or higher F-values by chance. We generated 10 000 permutation samples without replacement. The dependent variable (male or female fitness) was separated from the independent variables (population, year and characters) and, for each permutation sample, the fitness values were assigned randomly to the independent variable values. Each permutation sample had the same number of observations as the original data set. The F-statistics were estimated for all permutation samples. Finally, each P-value was estimated as the proportion of permutation samples with an F-value equal to or greater than the original F-value.

Results

Supplemental hand cross-pollination increased fruit set by 1.6 to four times in four Swedish populations (Table 3). This increase in fruit set was largest in the three Västerbotten populations, northern Sweden.

Table 3.  The effect of supplemental cross-pollinations (treatment) on fruit set in four Swedish populations of Platanthera bifolia. Test refers to Mann–Whitney U-test (see Materials and methods).
PopulationMean % fruit set (N)Z
ControlTreated
  1. *P < 0.01, **P < 0.001.

Ersmarksberget32 (34)65 (17)2.7*
Kulbäcksliden11 (54)60 (10)4.3**
Balberget13 (78)76 (20)5.9**
Hammarskog62 (15)98 (14)3.6**

Table 4 shows descriptive statistics of characters and fitness measures in P. bifolia in Lillberget and Yttre Hemberget. Mean values of traits differed between years in the Lillberget population: in the dry season of 1994, flower number was lower, spur length was shorter, and pollinium removal and fruit set were lower than in 1995. Mean values of characters also differed between the two populations in 1995: number of flowers was lower, spur length was shorter, and pollinium removal and fruit set were lower in the Yttre Hemberget than in the Lillberget population. The proportion of individuals failing to export any pollinium was 35% in Lillberget 1994, 2% in Lillberget 1995, and 15% in Yttre Hemberget 1995. The proportion of individuals failing to produce any fruit was 78% in Lillberget 1994, 13% in Lillberget 1995, and 20% in Yttre Hemberget 1995.

Table 4.  Mean and standard deviations (SD) of characters and fitness measures in two Platanthera bifolia populations. Confidence intervals (95% CI) for mean values were estimated with a bias-corrected bootstrap method (see Materials and methods). Nonoverlapping CIs indicate significantly different mean values. Number of observations for all variables was 108, 101 and 84 in Lillberget 1994, 1995 and Yttre Hemberget 1995 respectively.
VariableLillbergetYttre Hemberget

1995
Mean19941995
(95% CI)SDMean(95% CI)SDMean(95% CI)SD
Number of flowers12.0(11.2, 12.8)4.314.5(13.6, 15.6)5.110.2(9.4, 11.1)4.0
Spike length8.7(8.1, 9.3)3.38.5(7.9, 9.1)3.27.5(6.8, 8.2)3.2
Stalk length28.9(28.1, 29.7)4.427.9(26.9, 28.9)5.226.5(25.4, 27.6)4.8
Spur length28.3(27.7, 28.8)2.930.7(30.1, 31.2)2.829.3(28.6, 29.9)3.1
Pollinium removal per inflorescence3.41(2.66, 4.24)4.2111.97(9.69, 12.42)7.015.67(4.59, 6.80)5.13
Fruits set per inflorescence0.47(0.30, 0.66)0.987.52(6.53, 8.58)5.294.51(3.69, 5.36)3.97
% Pollinium removal13.4(10.5, 16.4)15.337.6(34.0, 41.3)19.225.0(20.9, 29.1)19.0
% Fruit set4.2(2.6, 6.2)9.651.1(45.0, 56.9)31.441.7(35.0, 48.4)31.8

The variances of relative fitness components (opportunity for selection) are shown in Table 5. Variances differed significantly between the two seasons (1994 and 1995) in Lillberget, 1994 having much higher variances, especially that of female fitness. There was no significant difference in variance in male and female fitness between Yttre Hemberget and Lillberget 1995.

Table 5.  Variances (σ2) of relative fitness per inflorescence in two populations of Platanthera bifolia. Confidence intervals (95% CI) were estimated by using a bias-corrected bootstrap method (see Materials and methods). Nonoverlapping confidence intervals indicate significant different variances. Number of observations was 108, 101, and 84 in Lillberget 1994, 1995 and Yttre Hemberget 1995 respectively.
Fitness componentLillbergetYttre Hemberget

1995
 19941995
σ2(95% CI)σ2(95% CI)σ2(95% CI)
Male relative fitness1.52(0.96, 2.40)0.41(0.27, 0.61)0.82(0.54, 1.17) 
Female relative fitness4.31(2.61, 6.25)0.49(0.37, 0.65)0.77(0.56, 1.05) 

The results from the selection analyses are reported in Table 6. There was strong selection for more flowers per plant in all populations and years (Lillberget 1994, 1995, and Yttre Hemberget 1995) via male fitness per inflorescence. Through female fitness per inflorescence, there was generally strong selection for larger number of flowers, but nonsignificant in Lillberget 1994. There was weak positive selection on stalk length through female fitness in Lillberget 1995. In Lillberget 1994, there was positive selection on spur length via female fitness, indicating that individuals with longer spurs had higher fruit set.

Table 6.  Analysis of directional phenotypic selection through male and female relative fitness on three characters in Platanthera bifolia in two populations. All parameters are standardized to unit variance. Confidence intervals (95% CI) and P-values of β′ were estimated by using a bias-corrected bootstrap method (see Materials and methods).
Population and yearCharacterMale functionFemale function
β(95% CI)Pβ(95% CI)P
  1. n.s.: P > 0.10.

Lillberget 1994 (n = 108)Flower number0.459(0.233, 0.699)0.0010.310(−0.139, 0.784)n.s.
Stalk length−0.141(−0.385, 0.075)n.s.−0.218(−0.648, 0.165)n.s.
Spur length0.192(−0.015, 0.422)0.070.506(0.056, 1.025)0.03
Lillberget 1995 (n = 101)Flower number0.331(0.163, 0.481)0.0010.298(0.139, 0.432)0.001
Stalk length0.065(−0.035, 0.166)n.s.0.188(0.065, 0.308)0.02
Spur length0.063(−0.063, 0.169)n.s.0.056(−0.059, 0.178)n.s.
Yttre Hemberget 1995 (n = 84)Flower number0.579(0.376, 0.768)0.0010.501(0.315, 0.686)0.001
Stalk length0.027(−0.128, 0.175)n.s.0.105(−0.063, 0.282)n.s.
Spur length0.034(−0.124, 0.235)n.s.0.060(−0.105, 0.236)n.s.

The results of the ancovas on the merged data set, also including data from Maad (2000), are reported in Table 7. Mean male and female absolute fitness varied between populations as indicated by significant main effects of population. However, there was no significant main effect of year on either male or female fitness, but an interaction of year and population had significant effect on female function indicating that mean fruit set varied among years in one of the populations. The general trend was that flower number had the strongest impact on both male and female fitness. However, the strength of selection on flower number varied among years and populations as indicated by significant interaction effects. Stalk length had weaker, although significant, main effect on both male and female fitness, as well as an interaction effect with year through male function. Spur length had also influence on fitness but was only significant for female function.

Table 7. ancova of character impact on absolute male and female fitness in three populations and three years of Platanthera bifolia of a large data set including three populations: Lillberget (1994, 1995) and Yttre Hemberget (1995), as well as Hönsarvsberget (1993–1995) from Maad (2000). Highly insignificant (P > 0.5) items were excluded from the final analysis. Significances were estimated by using a re-sampling method (see Materials and methods).
SourceMale functionFemale function
dfMSFdfMSF
  1. *P < 0.05, **P < 0.01, ***P < 0.001, †P = 0.07.

Population2117.63.55*234.83.76*
Year239.01.1926.30.69
Year × population1384.141.57***
Flower number12876.286.83***11393.6150.79***
Stalk length1325.29.82**1116.712.63***
Spur length1104.43.15†1136.114.73***
Flower number × population2737.622.27***2370.940.13***
Flower number × year2178.05.37**285.79.27***
Stalk length × year2141.34.26*
Spur length × year230.30.92
Error56533.1 5669.2 

Discussion

Selection on plant morphology

In the present study, there were generally steep directional selection gradients towards larger display size (flower number) through both sex functions. However, in Lillberget 1994 there was significant selection for such larger size only through male fitness per inflorescence. Not surprisingly, flower number was the character that contributed the most to the variation in fitness in the analyses combining present data with data from Maad (2000). These analyses also revealed that the strength of selection on this character varied spatially and temporally. Strong selection towards larger plant size has been documented to occur in many other flowering plants (e.g. Campbell, 1989; Dudash, 1991; Fritz & Nilsson, 1996; Ollerton & Lack, 1998). Floral display may have both direct and indirect effects on reproductive success in plants because flower number sets an upper limit to the amount of pollen removed and number of fruits produced, and it may influence pollinator visitation (Gómez, 2000). Stalk length also contributes to the conspicuousness of the plant (Donnelly et al., 1998; Ehrlén et al., 2002), but had little impact on fitness in the present study. Selection for longer stalks was present only in Lillberget 1995 through female function. In the combined analyses including also the data from Maad (2000), stalk length was found to have a rather weak, but significant, main effect on both male and female fitness. Selection on stalk length through male function was also shown to vary temporally.

Strong selection has rarely been documented on flower depth (Schemske & Horvitz, 1989; Herrera, 1993; Alexandersson & Johnson, 2002). In the present study of P. bifolia, at Lillberget 1994 there was selection for increased spur length via female fitness. However, there was no such selection present in 1995 in either of the two populations. In the combined analysis, also including data from Maad (2000), there was a significant main effect of spur length on fruit set only. This indicates that although the effect of spur length generally is weak, it has some importance at least for the female function (Nilsson, 1988). In the Lillberget population, mean spur length was c. 2.5 mm shorter in 1994 than in 1995. The spur length in Hönsarvsberget during the drought year was, similarly, also considerably shorter than in the normal wet years (Maad, 2000). This may have caused the observed selection towards longer spur in the drought year, if short-spurred individuals had too short spurs relative to the moth's tongue for efficient pollen import. Alternatively, if both fruit development and a turgor-dependent character such as spur length are affected by drought, the observed relationship between spur length and female fitness may be indirect via turgor.

Pollen vs. resource limitation of fruit set

Hand cross-pollination increased fruit set especially in the Västerbotten populations. Therefore, we conclude that fruit set of P. bifolia is generally pollen limited within season in the area of the present selection study. The combined selection analysis (including data from Maad, 2000) showed that mean fruit set varies naturally between populations. Few pollinators of P. bifolia have been observed in Västerbotten, northern Sweden, compared with areas in central Sweden (J. Maad and L. G. Reinhammar, unpublished data; T. Nilsson and R. Alexandersson, unpublished data), and fruit set is generally lower in northern than in central parts of Sweden (Table 4; Maad, 2000; R. Alexandersson and J. Maad, unpublished data).

As the fruit set was not 100% or nearly so in the hand-pollinated individuals of the Västerbotten populations, it is probable that fruit set is resource dependent within season if enough pollen is supplied. Fitness of a plant can be limited by both pollen and resources (e.g. Ehrlén, 2002). The reason that nearly 100% of the hand-pollinated flowers set fruit in the population of central Sweden may be that the inflorescences were cut to leave 10 flowers and all individuals had resources to mature 10 fruits. Fruit set has previously been shown to be mainly pollinator limited, but somewhat nutrient limited in small but not in large plants of P. bifolia (Mattila & Kuitunen, 2000). However, the effect of extra water has not been tested. Pollinator and resource limitation of seed or fruit set have been found to vary within and among seasons and populations in several plants species (Burd, 1994; Thomson, 2001), which is probably also the case for P. bifolia.

Lifetime male and female fitness must be limited by resources to some extent. The amount of resources invested in reproduction one year may have an impact on future growth and reproduction (Ackerman & Montalvo, 1990; Ehrlén & Eriksson, 1995; Willems & Dorland, 2000; Hautekeete et al., 2001). In the drought year of the present study, both male and female absolute fitness was lower compared with the other study year. This indicates a cost also for male (as well as female) function or, alternatively, poor pollinator service during flowering. The flowering duration of individuals was shortened by drought (R. Alexandersson, personal observation), which may have led to fewer pollinator visits. Alternatively, there may have been few pollinators this season. Pollinium removal correlates highly with pollen receipt in P. bifolia, and should be a very good measure of visitation frequency (J. Maad, unpublished data). In 1995, in the present study and in all three years in the other P. bifolia study (Maad, 2000), mean percentage of fruit set was consistently somewhat higher than percentage of pollinium removal. However, in the drought year of the present study, percentage of fruit set was c. one-third of the percentage of pollinium removal, a fact that indicates that fruit set was limited by other resources than pollen. We believe that fruit set in Lillberget 1994 was limited by water supplies, although other factors such as poor pollinator service may also have lowered fruit set. To specifically test the effect of water availability on male and female fitness in natural populations, experimental addition of water during drought years and normal wet years would be needed.

Patterns of selection and environmental conditions

Patterns of selection on reproductive characters have been shown to change when plants are subject to stress (Galen, 2000; Stanton et al., 2000). Selection on pollinator-attraction traits may be promoted by benign weather conditions but may be relaxed when weather conditions constrain fitness (Totland & Eide, 1999; Totland, 2001). Similar interactions have been documented between herbivores and pollinators: selection on reproductive characters was strongest in the absence of herbivores and the presence of pollinators (Herrera, 2000; Herrera et al., 2002). Other studies have detected stronger selection on certain characters in stressful environments (Bennington & McGraw, 1995; Stanton et al., 2000). In the present study, selection on spur length was significant only during a dry season, a situation that has previously been documented in Hönsarvsberget, central Sweden (Maad, 2000).

In the dry season of 1994, we observed significant selection for more flowers through male function but not through female function. However, the magnitudes of these selection gradients did not differ. The inaccuracy in estimating male fitness by using pollinium removal instead of siring success may, in part, also have generated a differing selection through male and female functions when a large fraction of removed pollen may have been wasted on flowers which had no resources to mature fruits, especially at the end of the flowering season when the drought was more severe (Wilson et al., 1994).

A higher cost of female than male function in P. bifolia may explain the significant selection on flower number through male function only in the dry year when fruit set was probably limited by water availability. Lack of physical resources, such as water, generally restricts seed and fruit production more than pollen export (Sutherland & Delph, 1984; Bell, 1985), although male function also represents a significant cost to the plant in low-nutrient environments (Eckhart & Chapin, 1997). This enables stronger selection through male than female fitness on pollinator attraction characters when fruit set is resource limited as predicted by the male function hypothesis. Stronger selection through male than female function on floral display would, in the long run, cause plants to produce more flowers than they can mature fruits. In this case, when sex-specific selection occurred in a year with very low fruit set compared with normal wet years, this sex-specific selection will probably have little evolutionary effect in the population as the plant is a polycarpic perennial.

Acknowledgments

U. Krynitz, A.R. Waites and U. Gunnarsson assisted in field. W.S. Armbruster, C.B. Fenster, J. Ågren, L.A. Nilsson, C. Eckhert, S. Andersson, B. Oxelman, K. Bremer, M. Thulin, B. Bremer, and five anonymous reviewers gave comments on previous versions of this manuscript. This work was supported by J. Kempes Minnes Stipendiefond and Gunnar and Ruth Björkman's Fund for Botanical Research in northern Sweden to R.A.

Ancillary