SEARCH

SEARCH BY CITATION

Keywords:

  • Arabidopsis lyrata;
  • floral display;
  • flowering time;
  • herbivory;
  • phenology;
  • plant–animal interaction;
  • selection analysis

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

To determine whether population differentiation in flowering time is consistent with differences in current selection, we quantified phenotypic selection acting through female reproductive success on flowering phenology and floral display in two Scandinavian populations of the outcrossing, perennial herb Arabidopsis lyrata in two years. One population was located in an alpine environment strongly affected by grazing, whereas the other was close to sea level and only moderately affected by herbivory. Multiple regression models indicated directional selection for early end of flowering in one year in the lowland population, and directional selection for early start of flowering in one year in the alpine population. As expected, there was selection for more inflorescences in the lowland population. However, in the alpine population, plants with many inflorescences were selectively grazed and the number of inflorescences produced was negatively related to female fitness in one year and not significantly related to female fitness in the second year. The results are consistent with the hypothesis that genetic differentiation in flowering phenology between the study populations is adaptive, and indicate that interactions with selective grazers may strongly influence selection on floral display in A. lyrata.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

All plant populations in Scandinavia have colonized their current locations after the last ice age, i.e. during the last 10 000 years. During colonization, populations are likely to become differentiated as a consequence of both random processes and divergent selection. Populations may go through bottlenecks and phases of fragmentation, which influence their genetic structure (e.g. Hewitt, 2000; Lascoux et al., 2004; Dhuyvetter et al., 2005). The traces of history can be detected using neutral molecular markers, but historical processes may also leave traces in the patterns of morphological variation (Lagercrantz & Ryman, 1990). Divergent selection has apparently resulted in the evolution of local adaptation in many species. Latitudinal clines in the timing of growth in forest trees, such as Pinus sylvestris and Betula pendula are well known examples (Eiche, 1966; Mikola, 1982; García-Gil et al., 2003; Li et al., 2003). Variation in flowering time also suggests adaptation to the length of the growing season. For instance, common-garden experiments have demonstrated latitudinal differences in flowering time in Arabidopsis thaliana (Stinchcombe et al., 2004), Lythrum salicaria (Olsson & Ågren, 2002) and two species of Solidago (Weber & Schmid, 1998). Colonizing species have also repeatedly and independently evolved similar adaptations to other abiotic factors, such as serpentine soils, e.g. in populations of Cerastium alpinum (Nyberg Berglund et al., 2004).

Clinal genetically based variation in trait values along environmental gradients suggests a strong role for natural selection. However, conclusive demonstration of local adaptation needs more evidence (Kawecki & Ebert, 2004). Reciprocal transplantation experiments are needed to determine whether local populations have higher fitness than nonlocal populations (Clausen et al., 1948; Stebbins, 1950). The role of migration and genetic drift for patterns of population differentiation in quantitative traits can be examined by comparing patterns of trait divergence to those observed for molecular markers (Prout & Barker, 1989; Karhu et al., 1996; McKay & Latta, 2002). Local adaptation by definition means that population differentiation is generated by natural selection. Thus, current selection on traits of interest is expected to be consistent with patterns of population differentiation (Dudley, 1996; Nagy, 1997; Petit & Thompson, 1998; Geber & Griffen, 2003).

In this study, we compared patterns of selection on flowering phenology and floral display in an alpine and in a coastal, lowland population of the perennial, self-incompatible herb Arabidopsis lyrata ssp. petraea in Scandinavia. In this part of northern Europe, A. lyrata is found in a variety of habitats from coastal lowland populations to alpine areas. The alpine and lowland populations chosen for study are found at sites that differ considerably in the length of the growing season (number of days with mean temperature above 5°C: Spiterstulen < 110 days vs. Stubbsand 150 days; based on data from closely located weather stations; Table 1). Moreover, grazing pressure differs between the two populations. The alpine population is subject to sheep grazing, which results in considerable inflorescence damage, whereas in the lowland population inflorescence damage is caused by hares and insect herbivores and is typically less severe (see below). Common-garden experiments indicate considerable genetic differentiation in flowering phenology among A. lyrata ssp. petraea populations (O. Savolainen, A. Baker, P. Huotari, A. Okuloff, S. Sandring & J. Ågren, unpublished data). In a common-garden experiment that documented both the start and end of flowering, the alpine and the lowland population did not differ in total flower number, but the alpine population began flowering earlier and finished flowering later than the lowland population (S. Sandring & J. Ågren, unpublished data). However, there are no data to test the hypothesis that population differentiation in flowering phenology in A. lyrata corresponds to among-population variation in patterns of selection on flowering time.

Table 1.   Monthly mean temperatures (°C) April–September in 2000/2001 and 2002 and across the period 1961–1990 at the weather stations Sognefjell of the Norwegian Meteorological Institute (1413 m a.s.l., about 22 km west of the Spiterstulen population) and Skagsudde of the Swedish Meteorological and Hydrological Institute (on the coast about 4 km southeast of the Stubbsand population).
 SognefjellSkagsudde
200020021961–1990200120021961–1990
April−3.7−2.2−61.82.91
May1.71.5−06.78.36
June1.95.1412.215.012
July7.27.4615.617.415
August5.310.3514.218.313
September3.04.0111.711.39

We documented spatio-temporal variation in selection by quantifying phenotypic selection acting through female reproductive success on flowering phenology and floral display in the two study populations of A. lyrata in two years. More specifically, we addressed the following questions: (i) Does selection on flowering phenology (first and last day of flowering) differ between populations and years? (ii) Do herbivores influence selection on flowering phenology and floral display? (iii) Are patterns of selection consistent with previously documented differences in flowering time between the alpine and the lowland population selected for study, i.e. is there (a) stronger selection for early start of flowering in the alpine than in the coastal environment and (b) stronger selection for early end of flowering in the coastal environment?

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

The species and study populations

Arabidopsis lyrata ssp. petraea (L.) O'Kane and Al-Shehbaz is a self-incompatible, insect-pollinated, perennial herb (Schierup, 1998; Kärkkäinen et al., 1999) with a disjunct distribution in central and northern Europe (Jalas & Suominen, 1994). It occurs in the mountains of southern Norway up to altitudes beyond 1700 m above sea level (Jonsell et al., 1995), and in Sweden on the eastern coast of the Gulf of Bothnia. Populations of A. lyrata in Norway and Sweden are genetically closely related to each other, with a population divergence estimate (FST) of about 24% at microsatellite loci (M.-H. Muller, A. Niittyvuopio, J. Leppälä & O. Savolainen, unpublished data; M. Gaudeul, H. Stenøien & J. Ågren, unpublished data), and likely originate from the same source of colonization. Northern and central European A. lyrata populations have been found to differ in onset of flowering (Riihimäki & Savolainen, 2004; Riihimäki et al., 2005) and also with respect to morphological characters such as trichome production (Kärkkäinen et al., 2004) and leaf shape (Jonsell et al., 1995), isozyme loci (Jonsell et al., 1995), microsatellite loci (van Treuren et al., 1997) and sequence variation (Savolainen et al., 2000; Wright et al., 2003; Ramos-Onsins et al., 2004).

For this study, we selected one alpine and one lowland, coastal A. lyrata population. The alpine population was located on a bank of the river Visa, just above the tree-line at Spiterstulen, Norway (61°38′N 8°24′E; 1106 m a.s.l.). The lowland, coastal population was located on a gravel shore of the Gulf of Bothnia, at Stubbsand, Sweden (63°58′N 18°17′E). Both populations occurred in highly disturbed habitats with nonclosed vegetation.

Phenotypic selection

We quantified phenotypic selection on number of inflorescences, phenology of flowering (first and last day of flowering), and rosette size in the Spiterstulen and Stubbsand populations in two different years (Spiterstulen 2000 and 2002; Stubbsand 2001 and Stubbsand 2002).

At both sites, plants were marked prior to flowering. In the Spiterstulen population, 150 plants were randomly selected and marked in 2000 with the constraint that individual plants were separated by at least 0.5 m. Of these plants, we included for study the 127 plants that flowered in 2000, and the 96 plants that were retrieved and flowered in 2002. In the smaller and denser Stubbsand population, all flowering plants in 17 permanent 0.5 × 0.5 m2 plots located in a transect across the central part of the population were included in the study (n = 91 in 2001, n = 103 in 2002). The plants in the plots represented about half of the flowering plants in the Stubbsand population. When growing densely, individual plants were distinguished based on variation in leaf morphology.

To document flowering phenology, we recorded flowering status of each plant every 3–4 days (every day at Spiterstulen in 2002) during the flowering period. We defined flowering start as the day when the first flower had opened, and the last day of flowering as the day when the last flower had wilted. For each plant, we recorded the rosette diameter in two directions early in the season (May or early June; rosette diameter was measured in only one direction on plants in the Spiterstulen population in 2000), and from these measures we estimated rosette area using the formula for an ellipse. At fruit maturation, we noted the total number of inflorescences produced, the number of inflorescences consumed by herbivores, and total fruit number for each plant included in the study.

Differences in character values between populations and years were analysed with analysis of variance. Population and year nested within population were treated as fixed effects. Year was nested within population because different years were sampled in the two populations.

We used number of fruits as an estimate of female reproductive success. Number of fruits is highly correlated with number of seeds per plant in both populations [Spiterstulen r = 0.984 (n = 135), Stubbsand r = 0.940 (n = 55)]. Relative fitness was calculated by dividing each absolute value by the population mean (Lande & Arnold, 1983).

Phenotypic selection was analysed using the regression techniques described by Lande & Arnold (1983) and Arnold & Wade (1984). In these analyses, all independent variables were standardized to have a mean of zero and a standard deviation of one. This allows direct comparison of results of selection analyses across traits, populations and years (Lande & Arnold, 1983; Brodie III et al., 1995). Selection differentials describe the total change of a trait mean (Si) or variance (Cii) due to selection. Estimates of selection differentials are derived from a univariate regression analysis, where the trait is the independent variable and relative fitness is the dependent variable. The selection differential estimates total selection on a given trait and represents both direct selection and selection resulting from correlations with other traits subject to selection. Selection gradients are measures of direct selection on a trait (βi, γii), or combination of traits (γij), independent of other correlated traits included in the model. The estimates of selection gradients are obtained from multivariate regression analysis as partial regression coefficients. The selection gradient (βi), describes directional selection on a trait. Nonlinear selection can be estimated by including quadratic terms in the regression model (γii). Because of limited sample sizes, bivariate nonlinear selection gradients (γij), which estimate selection on combinations of traits were not included in the models. The character of nonlinear selection gradients is best examined graphically (Brodie III et al., 1995; Blows & Brooks, 2003). Stabilizing (γii < 0) or disruptive (γii > 0) selection can only be inferred if there is a local maximum or minimum in the fitness surface within the data range (Mitchell-Olds & Shaw, 1987). We used added-variable plots (av.plot in the statistical program R) to investigate the shape of curvature not taken into account by a model containing only linear terms (Rawlings et al., 2001). An added-variable plot for the independent variable i plots the residuals from a regression of the response variable on all other independent variables vs. the residuals from a regression of i on all other independent variables (R Development Core Team, 2005).

To determine whether selection on inflorescence number, flowering phenology and rosette size was influenced by interactions with grazers, we statistically controlled for variation in herbivory by including the proportion of inflorescences damaged by herbivores as a covariate in the selection models.

We used variance inflation factors to examine the effect of multicollinearity on the precision of the estimates of linear selection gradients (cf. Fox & Monette, 1992). Variance inflation factors were smaller than 3 indicating low levels of multicollinearity.

Differences in selection gradients between populations and years were tested with analysis of covariance. Two-way interaction terms were backwards eliminated from the analyses if they did not have a significant effect on relative fitness (Sokal & Rohlf, 1995). As partly the same plants were sampled in both years, samples from different years were not strictly independent, and estimates of among-year variation in selection gradients should therefore be interpreted with caution.

Because of non-normality of the residuals from the selection models, we tested statistical significance of selection differentials, selection gradients and interaction coefficients by estimating 95% confidence intervals with bootstrapping. Data were resampled 4000 times, and the confidence intervals were constructed based on saved coefficients from each bootstrapping (Fox, 2002) using the bias-corrected and accelerated method (Dixon, 2001).

The sampling design differed between the two study populations because plants were originally marked for other purposes. To examine whether among-plot variation in female fitness could explain the patterns of selection observed in the lowland population, we analysed a model which included plot (random factor) as an independent variable in addition to trait values. The estimates of selection gradients were very similar to a model without plot included, which suggests that the documented gradients are not simply an artefact of the sampling scheme. However, including plot as an independent variable in the model to factor out variation in fitness at that spatial scale is problematic because it will also remove some of the covariance between character value and fitness which has a functional basis. We therefore report the results of the model that does not include plot as a factor. To evaluate environmentally caused covariation between character values and fitness more fully requires an experimental approach (Stinchcombe et al., 2002).

All statistical analyses were performed using the free statistical program R (R Development Core Team, 2005).

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Variation in number of inflorescences, flowering phenology and rosette size

Trait values differed both between years and between populations (Table 2). The mean first day of flowering was 21 June 2000 and 5 June 2002 in the alpine Spiterstulen population, and 9 June 2001 and 4 June 2002 in the lowland Stubbsand population (Table 2, Fig. 1). Flowering ended on average later in the Spiterstulen than in the Stubbsand population (Table 2). Plants in the Stubbsand population tended to produce leaf rosettes of similar size, fewer inflorescences, but several times more fruits than plants in the Spiterstulen population (Table 2).

Table 2.   Fruit production, phenology of flowering, number of inflorescences, rosette size, proportion of inflorescences grazed by herbivores in the alpine Spiterstulen and the lowland Stubbsand populations of Arabidopsis lyrata in two years.
 Spiterstulen 2000 (n = 127)Spiterstulen 2002 (n = 96)Stubbsand 2001 (n = 91)Stubbsand 2002 (n = 103)PopulationYear (within population)
F1,414PF2,414P
  1. Means ± SD and F-ratios and associated P-values from an anova of variation among populations and years nested within populations are given. Number of fruits per plant and number of inflorescences were log-transformed, and proportion of inflorescences grazed was arcsine square-root transformed prior to analysis.

No. of fruits per plant6.7 ± 9.511.1 ± 18.432.9 ± 64.933.1 ± 51.856.0< 0.0011.40.25
Flowering start (Julian date)172 ± 5156 ± 6160 ± 9155 ± 1662.7< 0.00179.7< 0.001
Flowering end (Julian date)206 ± 6214 ± 21196 ± 19195 ± 2074.7< 0.0015.80.003
No. of inflorescences6.9 ± 6.610.9 ± 10.15.9 ± 6.95.1 ± 5.225.9< 0.0016.70.001
Rosette size (cm2)13.3 ± 5.111.5 ± 4.611.1 ± 10.415.9 ± 16.61.10.296.10.002
Proportion of grazed inflorescences0.22 ± 0.220.52 ± 0.340.25 ± 0.310.09 ± 0.2046.9< 0.00137.2< 0.001
image

Figure 1.  Histogram of start and end of flowering (Julian date; 5-day intervals) in the alpine Spiterstulen and the lowland Stubbsand population of Arabidopsis lyrata. Means, standard deviations and medians are given.

Download figure to PowerPoint

Phenotypic correlations between number of inflorescences, rosette size and phenology of flowering were moderately strong [median (range) absolute values, 0.29 (0.017–0.75), Table 3]. Plants with many inflorescences tended to start flowering earlier and finish flowering later than plants with few inflorescences, although these correlations were not statistically significant in both populations in both years (Table 3). Flowering start was not correlated with the size of the rosette, but plants with large rosettes produced more inflorescences and finished flowering later than plants with small leaf rosettes (Table 3). Start and end of flowering were positively correlated in the Stubbsand population in 2001 (Table 3), but were not significantly correlated in the Stubbsand population in 2002 or in the Spiterstulen population.

Table 3.   Phenotypic correlations (Pearson coefficients) between the studied traits in the Spiterstulen (year 2000 above the diagonal, 2002 below) and Stubbsand (year 2001 above the diagonal, 2002 below) populations of Arabidopsis lyrata in the two years of study.
 Flowering startFlowering endNo of inflorescencesRosette sizeInflorescence herbivory
  1. *P < 0.05, **P < 0.01.

Spiterstulen
 Flowering start 0.017−0.1210.102−0.009
 Flowering end0.067 0.232**0.398**0.114
 Number of inflorescences−0.282**0.291** 0.732**0.446**
 Rosette size−0.0490.297**0.751** 0.341**
 Inflorescence herbivory−0.203*−0.0820.385**0.312** 
Stubbsand
 Flowering start 0.314**−0.355**−0.1550.244
 Flowering end0.062 0.1850.321**−0.127
 Number of inflorescences−0.420**0.536** 0.822**−0.172
 Rosette size−0.2930.244**0.570** −0.173
 Inflorescence herbivory0.0250.1670.0490.127** 

Herbivory

Herbivore damage to inflorescences tended to be higher in the Spiterstulen population than in the Stubbsand population. On average, 22%, and 52% of the inflorescences were consumed by herbivores (sheep) in the alpine Spiterstulen population in 2000 and 2002, respectively, whereas in the lowland Stubbsand population 22% and 9% of the inflorescences were consumed by herbivores (hares and lepidopteran larvae) in 2001 and 2002 respectively (Table 2). Inflorescence herbivory was positively correlated with rosette size (in the Stubbsand population in only one of two years; Table 3). In the Spiterstulen population, it was also positively correlated with the number of inflorescences (Table 3).

Selection differentials

Most selection differentials were statistically significant in the Stubbsand population and in the Spiterstulen population 2002, whereas no directional selection differential was statistically significant in the Spiterstulen population in 2000 (Table 4). In both populations, significant selection differentials indicated that plants that began flowering early, finished flowering late, produced many inflorescences and large leaf rosettes had higher fitness than other plants (Table 4). The selection differential for inflorescence number was higher in the Stubbsand than in the Spiterstulen population. In the Stubbsand population, also the nonlinear selection differentials for first day of flowering, last day of flowering (2002) and inflorescence number (2001) were statistically significant (and positive; Table 5). Surprisingly, plants with few inflorescences tended to have higher female fitness than plants with many inflorescences in the Spiterstulen population in 2000 (Table 4).

Table 4.   Estimates of linear univariate selection differentials (Si) and multivariate selection gradients (βi) in the alpine Spiterstulen population (2000 n = 127; 2002 n = 96) and the lowland Stubbsand population (2001 n = 91; 2002 n = 103).
 SiBiβi (herbivory in model)
SpiterstulenStubbsandSpiterstulenStubbsandSpiterstulenStubbsand
  1. Selection gradients were calculated with and without proportion of grazed inflorescences included in the models. 95% confidence intervals estimated with bootstrapping are given in brackets. Estimates whose 95% confidence intervals do not include zero are in bold. R2-values are given for the multivariate models.

2000/2001R2 = 0.08R2 = 0.82R2 = 0.24R2 = 0.84
 Flowering start−0.21 (−0.43, 0.03)0.69 (−1.10, −0.45)0.30 (−0.55, −0.05) 0.07 (−0.09, 0.22)0.26 (−0.48, −0.04) 0.14 (−0.02, 0.32)
 Flowering end0.22 (−0.03, 0.49)0.09 (−0.24, 0.57) 0.23 (−0.02, 0.50)0.27 (−0.57, −0.08)0.23 (0.01, 0.48)0.32 (−0.61, −0.13)
 No. inflorescences−0.14 (−0.32, 0.04)1.78 (1.33, 2.11)0.43 (−0.77, −0.03)1.91 (1.17, 2.56)−0.14 (−0.47, 0.34)1.92 (1.25, 2.56)
 Rosette size0.01 (−0.21, 0.21)1.20 (0.39, 1.79) 0.27 (−0.18, 0.62)−0.18 (−0.81, 0.52) 0.28 (−0.13, 0.60)−0.20 (−0.83, 0.44)
 Inflorescence herbivory    0.66 (−0.99, −0.40)0.26 (−0.44, −0.16)
2002R2 = 0.05R2 = 0.73R2 = 0.26R2 = 0.73
 Flowering start0.17 (−0.50, −0.01)0.62 (−0.94, −0.40)−0.12 (−0.32, 0.18)−0.04 (−0.16, 0.07)−0.20 (−0.58, 0.06)−0.04 (−0.16, 0.06)
 Flowering end0.23 (0.03, 0.50)0.68 (0.40, 1.03) 0.16 (−0.05, 0.36)−0.06 (−0.22, 0.16)−0.02 (−0.34, 0.21)−0.03 (−0.20, 0.18)
 No. inflorescences0.35 (0.05, 1.30)1.34 (1.09, 1.71) 0.22 (−0.26, 1.74)1.38 (1.02, 2.09) 0.51 (−0.01, 2.02)1.36 (1.01, 2.10)
 Rosette size0.29 (0.08, 0.68)0.80 (0.32, 1.27) 0.05 (−0.70, 0.40) 0.02 (−0.46, 0.39) 0.16 (−0.41, 0.48) 0.04 (−0.42, 0.41)
 Inflorescence herbivory    0.88 (−1.55, −0.50)0.12 (−0.30, −0.01)
Table 5.   Estimates of nonlinear univariate selection differentials (Ci) multivariate selection gradients (γii) in the alpine Spiterstulen population (2000 n = 127; 2002 n = 96) and the lowland Stubbsand population (2001 n = 91; 2002 n = 103).
 Ciγiiγii (herbivory in model)
SpiterstulenStubbsandSpiterstulenStubbsandSpiterstulenStubbsand
  1. Selection gradients were calculated with and without proportion of grazed inflorescences included in the models. 95% confidence intervals estimated with bootstrapping are given in brackets. Estimates whose 95% confidence intervals do not include zero are in bold.

2000/2001
 Flowering start0.02 (−0.14, 0.16)0.46 (0.26, 0.78)0.02 (−0.17, 0.17)0.09 (0.01, 0.22)0.02 (−0.16, 0.17)0.08 (−0.02, 0.21)
 Flowering end0.04 (−0.09, 0.21)−0.43 (−1.17, 0.26)0.02 (−0.12, 0.20)−0.20 (−0.68, 0.02)0.04 (−0.10, 0.19)0.25 (−0.80, −0.02)
 No. inflorescences−0.09 (−0.30, 0.07)0.32 (0.15, 0.61)−0.03 (−0.34, 0.14)0.42 (0.06, 0.84)0.26 (−0.65, −0.02)0.37 (0.02, 0.84)
 Rosette size−0.08 (−0.23, 0.08)0.14 (−0.34, 0.47)0.05 (−0.11, 0.25)−0.14 (−0.32, 0.28)−0.07 (−0.26, 0.09)−0.15 (−0.33, 0.25)
 Inflorescence herbivory    0.41 (0.08, 0.75)0.11 (−0.02, 0.26)
2002
 Flowering start−0.01 (−0.10, 0.13)0.28 (0.10, 0.55)−0.05 (−0.24, 0.06)0.03 (−0.05, 0.12)−0.02 (−0.20, 0.15)0.03 (−0.06, 0.11)
 Flowering end0.30 (−0.63, −0.05)0.35 (0.05, 0.66)0.30 (−0.66, −0.03)0.05 (−0.13, 0.29)−0.12 (−0.42, 0.16)0.03 (−0.15, 0.26)
 No. inflorescences−0.15 (−0.41, 0.07)−0.01 (−0.17, 0.22)−0.03 (−0.33, 0.41)−0.004 (−0.23, 0.35)−0.26 (−0.63, 0.08)−0.02 (−0.27, 0.29)
 Rosette size−0.09 (−0.36, 0.08)−0.19 (−0.43, 0.11)−0.10 (−0.46, 0.14)0.05 (−0.11, 0.35)−0.04 (−0.32, 0.19)0.05 (−0.10, 0.44)
 Inflorescence herbivory    −0.18 (−0.59, 0.09)0.08 (−0.08, 0.27)

Selection gradients

Multivariate analysis indicated that there was selection for more inflorescences (both years) and early end of flowering (2001) in the lowland Stubbsand population, and that there was directional selection for early start of flowering and few inflorescences in the alpine Spiterstulen population in 2000 (Table 4). The results suggest that in the Stubbsand population, selection on flowering start and rosette size was to a large extent the result of selection on phenotypically correlated characters. On the other hand, directional selection for early flowering start in the Spiterstulen population in 2000 and directional selection for early end of flowering in the Stubbsand population in 2001 were detected in the multivariate, but not in the univariate analyses (Table 4). This suggests that effects of phenotypically correlated characters on female fitness obscured some selection on flowering phenology in the two populations. The positive quadratic selection gradients for number of inflorescences and flowering start in the Stubbsand population in 2001 (Table 5) indicated a significant nonlinearity in the relationships between relative fitness and these characters, but no evidence of disruptive selection was observed in added-variable plots. The negative quadratic selection gradient for end of flowering in the Spiterstulen population in 2000 was a function of two individuals that finished flowering at an intermediate date and produced a very large number of fruits. When these two individuals were excluded from analysis, the quadratic term was not statistically significant (95% confidence interval −0.43, 0.10).

Selection models that included inflorescence damage as a covariate indicated that interactions with herbivores influenced patterns of selection in the alpine Spiterstulen population. Comparisons of models with and without inflorescence herbivory as a covariate suggested that interactions with herbivores could explain a substantial portion of the directional selection for few inflorescences in the Spiterstulen population in 2000, but that they did not influence selection on flowering start (Table 4). In the Stubbsand population, estimates of selection gradients were very similar in the two models indicating that the herbivores did not influence selection on flowering phenology, number of inflorescences or rosette size (Table 4). Fruit number was significantly negatively correlated with inflorescence herbivory except in the Stubbsand population in 2002 (r: Spiterstulen −0.404 and −0.340; Stubbsand −0.251 and −0.03), when the proportion of inflorescences damaged by herbivores was relatively low (Table 2). Models that included inflorescence damage as a covariate explained a markedly larger proportion of the total variance in relative fitness in the Spiterstulen population than models without the covariate (Table 4). In the Stubbsand population the inclusion of inflorescence damage affected the proportion of explained variance only marginally (Table 4).

Spatial and temporal variation in selection gradients

Analysis of covariance and comparisons of 95% confidence intervals of selection gradients obtained with bootstrapping indicated that selection on flowering phenology differed between populations and that selection on floral display differed between years in the alpine Spiterstulen population. The direction of the difference between populations in selection gradient estimates for start and end of flowering was consistent among years, and the bootstrapped 95% confidence interval of estimates of population × trait interaction coefficients did not overlap zero (flowering start 0.003, 0.380; end of flowering −0.584, −0.153). Analysis of covariance indicated a significant number of inflorescences × year within population interaction, and parameter estimates obtained from the model suggested that this was because of among-year variation in selection on number of inflorescences in the alpine population (95% confidence interval for interaction coefficient, Spiterstulen 0.050, 0.639; Stubbsand −0.846, 0.170). No statistically significant among-year variation in selection on flowering phenology, or spatiotemporal variation in selection on rosette size was detected.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

This study has revealed differences in patterns of selection through female reproductive success on flowering phenology and number of inflorescences between an alpine and a lowland population of A. lyrata. Moreover, it has demonstrated that grazing damage to inflorescences significantly influences patterns of selection in the alpine population. Selective grazing of plants with a large floral display could to a large extent explain the negative relationship between number of inflorescences and female fecundity in the alpine population.

Selection for early flowering was detected in the alpine but not in the lowland population, whereas selection for early end of flowering was detected in the lowland population. This is consistent with the hypothesis that genetic differentiation in flowering phenology between the two study populations is the product of altitudinal variation in selection. In a common-garden experiment, the alpine Spiterstulen population started flowering markedly earlier but finished flowering later than the Stubbsand population and several other lowland populations from Sweden (S. Sandring & J. Ågren, unpublished data). A similar concordance between genetically based population differentiation and variation in patterns of selection has been documented for water-use efficiency in Cakile edentula (Dudley, 1996), age and size at first reproduction in Plantago major (Lotz, 1990), and for morphological traits in Gilia capitata (Nagy, 1997) and phenological traits in Arrhenatherum elatius (Petit & Thompson, 1998). However, patterns of selection suggested by phenotypic differentiation among populations are not always observed (e.g. Winn, 1999). This could be because selection varies among years, selection is too weak to be detectable with available sample sizes, or because population differentiation in a given character is due to random processes and history rather than divergent selection.

Among-population variation in selection on flowering time may be a function of climatic differences. The short growing season of the alpine environment should constrain the time available for fruit maturation, which may result in selection for early flowering start (e.g. Kudo, 1993; Olsson & Ågren, 2002). Similarly, among-year variation in weather conditions could potentially explain why selection for early flowering tended to be stronger in the Spiterstulen population in 2000 than in 2002. In the first year, summer temperatures (June–August) were close to 30-year mean temperatures, whereas in the second year the summer was exceptionally warm (Table 1).

Alternatively, spatial and temporal variation in selection on flowering time can be a consequence of differences in seasonal patterns of herbivory or pollination (Widén, 1991; Pilson, 2000; Ashman & Morgan, 2004). In this study, selection for early flowering start in the alpine population and early end of flowering in the lowland population was apparently not a function of interactions with herbivores. Seed predation was rarely observed, and early flowering was associated with high female fitness also in regression models, which controlled statistically for variation in inflorescence damage. The hypothesis that selection for early flowering in the alpine population and early end of flowering in the lowland population is a consequence of seasonal changes in pollination intensity late in the season could be tested with supplemental hand-pollination.

Differences in temperature-sum requirements for the initiation of flowering could explain both why the alpine Spiterstulen population is fast to flower compared with lowland populations in common-garden experiments (A. Okuloff, S. Sandring, J. Ågren & O. Savolainen, unpublished data) and why differences in dates of first flower in the natural populations were relatively small. In 2002, mean flowering start differed by only 1 day (Table 2), indicating that the Spiterstulen population began flowering at a much lower accumulated temperature sum than the Stubbsand population (Table 1). In several plant species, high altitude and northern populations have been found to initiate growth and flowering at lower temperature sums than southern and low-altitude populations in common garden environments (e.g. Olsson & Ågren, 2002).

Floral display may influence interactions both with mutualists and antagonists. A large floral display may increase attractiveness to pollinators, but simultaneously increase the risk of damage from herbivores (Brody, 1992; Ehrlén et al., 2002; Gómez, 2003; Strauss & Irwin, 2004). In the alpine Spiterstulen population, the proportion of inflorescences damaged by sheep was positively correlated with the number of inflorescences produced, and this could to a considerable degree explain why fruit output was negatively related to inflorescence production in 2000. To the best of our knowledge, this is the first example that selective grazing may completely negate the positive effects on female fecundity of producing many inflorescences. In the second year, the positive correlation between inflorescence number and herbivory in the Spiterstulen population did not result in selection for few inflorescences, but the selection gradient was low relative to those observed in the Stubbsand population and was not statistically significant. Between-population differences in selection on inflorescence number were not simply a consequence of differences in the magnitude of inflorescence damage, but rather the result of a difference in the selectivity of the dominating grazers (sheep in the alpine vs. hares and lepidopteran larvae in the lowland population). The proportion of inflorescences damaged by herbivores was similar in the Spiterstulen population in 2000 and in the Stubbsand population in 2001. However, in the Stubbsand population inflorescence herbivory was not correlated with number of inflorescences produced (Table 3). Finally, it is noteworthy that the estimate of the selection gradient for number of inflorescences in the Spiterstulen population in 2000 was negative also in models, which controlled statistically for variation in the proportion of inflorescences damaged by herbivores (Table 4). This may reflect a trade-off between flower and fruit production, between male and female reproductive function, or some other disadvantage associated with a high inflorescence production in this environment.

Selection analysis indicated that in the Stubbsand population, inflorescence number was subject to direct selection, whereas selection differentials for flowering start and rosette size were the result of selection on correlated characters. The univariate regression analyses revealed significant selection differentials for all characters in both years (except end of flowering in 2001), whereas only the selection gradient for number of inflorescences (both years) and end of flowering (first year) were statistically significant in the multivariate analysis (Table 4). As inflorescence number is likely to reflect flower production, it is expected to be under directional positive selection in studies using fruit or seed output in a single year as a measure of fitness. An exception would be when an increase in flower production is associated with a drastic increase in the risk of complete reproductive failure as in the Spiterstulen population (see above). As observed in the Stubbsand population, strong selection for more flowers and inflorescences should result in indirect selection on phenotypically correlated characters (Johnston, 1991; O'Neil, 1997; Olsson, 2004).

The amount of variation explained by the linear regression model used for selection analysis was very low for the alpine Spiterstulen population compared with the lowland Stubbsand population (Table 4). Inflorescence herbivory explained more of the variation in relative fitness than the measured traits together in the Spiterstulen population. However, also models that included herbivory as a covariate had lower R2-values than the corresponding models for the Stubbsand population, suggesting that fruit production was strongly influenced also by other factors in the alpine population.

The median magnitudes of linear and nonlinear selection gradients in the two study populations were similar to those recorded in a recent review of quantitative estimates of phenotypic selection in natural populations by Kingsolver et al. (2001). As expected, directional selection gradients for aspects of plant size (rosette size and inflorescence number) tended to be large in both populations. However, in the Spiterstulen population in 2000, the strength of selection on phenology of flowering was similar to that on size traits (Table 4). There was little evidence for stabilizing or disruptive selection in the study populations. However, sample sizes were limited and the power to detect nonlinear selection therefore modest (Kingsolver et al., 2001).

To conclude, the results of the present study are consistent with the hypothesis that Scandinavian populations of A. lyrata have diverged in response to altitudinal variation in selection on flowering time, and they indicate that interactions with herbivores may strongly influence selection on floral display. We have considered here the effects of character variation on female fecundity in single years. Future studies should ideally assess the effects of plant size and flowering phenology also on survival, flowering frequency and male reproductive success. In addition, experimental studies are needed to examine the potential problem with environmentally caused covariation between character values and fitness (Stinchcombe et al., 2002). We currently explore the adaptive significance and the genetic basis of differences in morphology and flowering phenology between the two study populations with reciprocal transplantation experiments and QTL mapping.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

The authors thank M. Andersson, M. Hyvärinen and G. Løe for assistance in the field and the Sulheim family for facilitating our work at Spiterstulen. The authors acknowledge financial support to M.R. from the Forest Ecology Graduate School and University of Oulu, to S.S. from the Royal Swedish Academy of Science and Helge Ax:on Johnsons Foundation, to O.S. from the Research Council for Biosciences and Environment of Finland, and to J.Å. from Formas and the Swedish Research Council. We thank Johanne Maad, Johan Ehrlén and two anonymous reviewers for valuable comments on the manuscript.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References