Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Evolution, 58(7), 2004, pp. 1446–1458 EVOLUTIONARY POTENTIAL OF CHAMAECRISTA FASCICULATA IN RELATION TO CLIMATE CHANGE. I. CLINAL PATTERNS OF SELECTION ALONG AN ENVIRONMENTAL GRADIENT IN THE GREAT PLAINS JULIE R. ETTERSON1 Department of Ecology, Evolution and Behavior, University of Minnesota, St. Paul, Minnesota 55812-3003 Abstract. Climate change will alter natural selection on native plant populations. Little information is available to predict how selection will change in the future and how populations will respond. Insight can be obtained by comparing selection regimes in current environments to selection regimes in environments similar to those predicted for the future. To mimic predicted temporal change in climate, three natural populations of the annual legume Chamaecrista fasciculata were sampled from a climate gradient in the Great Plains and progeny of formal crosses were reciprocally planted back into common gardens across this climate gradient. In each garden, native populations produced significantly more seed than the other populations, providing strong evidence of local adaptation. Phenotypic selection analysis conducted by site showed that plants with slower reproductive development, more leaves, and thicker leaves were favored in the most southern garden. Evidence of clinal variation in selection regimes was also found; selection coefficients were ordered according to the latitude of the common gardens. The adaptive value of native traits was indicated by selection toward the mean of local populations. Repeated clinal patterns in linear and nonlinear selection coefficients among populations and within and between sites were found. To the extent that temporal change in climate into the future will parallel the differences in selection across this spatial gradient, this study suggests that selection regimes will be displaced northward and different trait values will be favored in natural populations. Key words. Climate gradient, clinal variation, global warming, phenotypic plasticity, phenotypic selection. Received January 27, 2004. Climate strongly influences patterns of natural selection (Linhart and Grant 1996). Clines in morphology, phenology, and physiology are commonly found among plant populations that occupy latitudinal and elevational climatic gradients (Clausen et al. 1940; Chapin and Chapin 1981; Lacey 1984, 1988; Reinartz 1984; Shaver et al. 1986; Ehleringer 1988; Neuffer and Bartelheim 1989; Körner et al. 1991; Montagnes and Vitt 1991; Aizen and Woodcock 1992; Sawada et al. 1994; Hurme et al. 1997; Clapham et al. 1998; Weber and Schmid 1998; Gehring and Delph 1999; Jonas and Geber 1999; Clevering et al. 2001). Yet no previous studies have directly linked estimates of clinal selection to clinal patterns in phenotype. Genetically based clinal patterns in phenotype provide strong evidence of adaptive differentiation in response to a cline in natural selection (Mayr 1956). Ongoing climate change may shift clines in selection toward higher latitudes or elevations in some geographic regions. Long-term observational experiments in the field that document temporal changes in selection on wild populations are very rare (but see Reale et al. 2003). Likewise, experiments that compare natural selection in current environments to selection in experimental conditions that mimic those predicted for the future are not common (but see Totland 1999). Characterization of spatial patterns in selection along environmental gradients will illuminate future patterns of selection. Studies of phenotypic selection on natural populations in native environments generally show that selection is weak (20.1 , b , 0.1, Kingsolver et al. 2001), presumably because populations are already well adapted to local conditions. Thus, it may be difficult to detect selection. The power 1 Present address: Department of Biology, University of Minnesota Duluth, 221 Life Science Building, Duluth, Minnesota 55812-2496; E-mail: [email protected]. Accepted March 10, 2004. to detect selection can be enhanced by expanding the phenotypic distribution such that more extreme trait values are present than in the native population (Mitchell-Olds and Shaw 1987; Wade and Kalisz 1990; Schmitt et al. 1999). Such broadening of the phenotypic distribution has been accomplished through the use of mutant and transgenic lines (for a review see Schmitt et al. 1999), physiological manipulation (Dudley and Schmitt 1996), hybridization of ecotypes (Wolff 1988; Jordan 1991; Andersson and Widén 1993; Andersson 1996; Nagy 1997), physical manipulations (Andersson 1982; Nilsson 1988; Anholt 1991; Andersson and Widén 1993; Kingsolver 1996; O’Connell and Johnston 1998; Totland et al. 1998; Sinervo 1999), artificial selection (Ågren and Schemske 1994), and by producing artificial populations composed of individuals from natural populations that differ in average phenotype (Andersson and Widén 1993). Native phenotypes, if well adapted in the home environment, would be favored by selection over those that have been artificially introduced into the population. Due to plasticity, the phenotypic distribution of a population may differ when it is reared in a novel environment and this may, in turn, influence the strength of selection (Lewontin 1974; Mitchell-Olds and Shaw 1987; Wade and Kalisz 1990). Such phenotypic plasticity may be adaptive such that populations express favorable phenotypes that result in partial or full maintenance of fitness in a new environment (Bradshaw 1965; Schlichting 1986; Sultan 1987; van Tienderen 1991; Schlichting and Pigliucci 1998; Donohue et al. 2001). Phenotypic plasticity may evolve in response to the range of environmental heterogeneity that a population experiences. Thus, to the extent that environmental cues in future climates elicit adaptive phenotypic responses, the strength of selection may be moderated and populations maintain a minimum of fitness critical to population persistence, even without substantial genetic change. 1446 q 2004 The Society for the Study of Evolution. All rights reserved. SELECTION ALONG A CLIMATE GRADIENT 1447 FIG. 1. (A) Temperature increases from north to south and precipitation decreases from east to west produce an aridity gradient across the Great Plains of the United States as shown by isoclines of a long-term average growing season soil water index (a for evergreen trees, 1951–1980). a is an integrated measure of seasonal growth-limiting drought stress on plants and takes into account temperature, precipitation, and soil texture (Thompson et al. 2000). Arrows point to the locations of focal populations of Chamaecrista fasciculata in MN, KS, and OK. (B) The 2025–2035 prediction of a growing season soil water index for the state of MN (a* for evergreen trees) based on the Canadian Global Coupled Model (CGCM1, a coupled atmosphere-ocean general circulation model with transient greenhouse and sulfate aerosol forcing; Canadian Climate Center) manipulated by Vegetation/Ecosystem Modeling and Analysis Project (VEMAP2). Experimental garden plot in MN indicated with a star. In an effort to predict how natural selection may be altered with climate change, I conducted a study of phenotypic selection along an aridity gradient in the Great Plains of the United States (Fig. 1A). Patterns of selection are likely to differ spatially along this gradient such that they may serve as a proxy for the temporal trend in selection that northern populations will experience as climate changes. For example, plants from Minnesota (MN) may experience a climate more similar to the current climate of Kansas (KS) or Oklahoma (OK) in the future depending upon the rate of climate change and the time frame of interest. Specific expectations are based on the predictions of global climate models; one such model (CGCM1, Canadian Climate Center, available via http://www. cccma.bc.ec.gc.ca/models/cgcm1.shtml) predicting that plants at one of the focal sites of this study in southeastern MN will experience soil moisture conditions similar to the current conditions of northeastern KS by 2025–2035 (Fig. 1B). Likewise, a population in the northeastern KS may experience water availability conditions similar to southern OK (IPPC 2001). The underlying premise of this work is that differences in selection regimes across this spatial environmental gradient approximate the temporal differences in selection that populations will experience in the future. Surveys of variation among natural field populations of Chamaecrista fasciculata from MN to southern OK (Etterson, unpubl. data) and common-garden studies demonstrated clinal variation and genetic divergence in physiology and morphology (Etterson 2001). Fitness measures from greenhouse drought experiments suggested that populations are adapted to different water-availability conditions (Etterson 2001). Based on these studies, I developed adaptive hypotheses for three traits: (1) selection favoring rapid progression through the life cycle occurs in the short growing season of the north but should be weaker or reversed in sites further south; (2) selection favoring larger plant size occurs in all sites but it should be weaker in the south because of the greater transpirational water losses suffered by plants with more leaf surface area (Etterson 2001); (3) thicker leaves are favored in the more southern sites because smaller but thicker leaves are less subject to water loss but can maintain comparable rates of photosynthesis (Givnish 1979). I conducted a quantitative genetics study using a reciprocal 1448 JULIE R. ETTERSON transplant design among three populations that were sampled from the climate gradient in the Great Plains. Three distinct aspects of evolutionary change were investigated: natural selection on phenotypes (this paper), the underlying genetic basis of phenotypic variance and changes in genetic architecture in different environments (Etterson 2004), and the response to selection when genetic correlations among traits are taken into account (Etterson and Shaw 2001). Short-term evolutionary response (Etterson and Shaw 2001) was predicted using estimates of the additive genetic covariance between traits and fitness (Robertson 1968; Price 1970), a method akin to the genetic regression methods of Rausher (1992). Here the four objectives are: (1) to characterize phenotypic selection in three sites along a climate gradient in the Great Plains using a synthetic population of plants composed of three different populations known to differ in average phenotype; (2) to determine whether patterns of selection are clinal by examining selection on a single population reared in a gradient of environments and by examining differences in selection among populations reared in a single site; (3) to test adaptive hypotheses concerning phenological and morphological traits of C. fasciculata; and (4) to examine plastic responses in climates similar to those predicted for the future. The phenotypic selection analyses presented here exemplify a general approach to understanding the changes in selection that may arise with climate change. MATERIALS AND METHODS Chamaecrista fasciculata Michx. is an annual legume of disturbed prairie, old field, and savanna (McGregor 1986). The northern boundary of its native range in the Great Plains is in central Minnesota; the southern boundary in Mexico is not recorded. Three focal populations were sampled across an aridity gradient from three dry tallgrass prairie fragments in the northern half of the species range: MN (Kellog-Weaver Dunes, Wabasha Co.), KS (Konza Prairie, Riley Co.), and OK (Pontotoc Ridge, Pontotoc Co.). These sites are between 448N and 348N latitude and, from north to south, range 17– 238C in average temperature and 50–60 cm total precipitation during the growing season (see Appendix 1 for other site differences). Crosses Five pods from each of 200 plants in each of the three natural populations were collected (for more details see Etterson 2004). Seeds were hand-scarified, germinated in petri plates, inoculated with Rhizobium spp. strain EL (LiphaTech, Inc., Milwaukee, WI), and raised under a 15:9-h photoperiod in a climate-controlled greenhouse in Pro-Mix potting mix (Premier Brands Inc., Rochelle, NY). Plants were crossed within populations according to a nested paternal half-sib crossing design with independent sets of three dams assigned at random to each sire (North Carolina I; Comstock and Robinson 1948; for more details see Etterson 2004). Planting Design and Measurements Pedigreed offspring were germinated as described above at the University of Minnesota, transplanted into 1 3 2 3 2- in plug trays, inoculated with commercial Rhizobium spp. (strain EL; LiphaTech, Inc.), and grown in the greenhouse for about three weeks. Seedlings from each population were planted into gardens in MN (N 5 3247), KS (N 5 3360), and OK (N 5 3301) according to a randomized block design (four blocks per site). The planting density approximated the average density of C. fasciculata in the native sites in the year the quantitative genetic founders were collected (; 12 plants/m2). To avoid contamination of locally adapted populations by genes from alien populations, planting sites that lacked C. fasciculata but had populations relatively nearby (, 5 km) were chosen. Seeds for the OK site were planted in the greenhouse on 1 May and transplanted into a plowed plot 30 km northwest of the natural source population on 19– 21 May (Robert S. Kerr Environmental Research Center, Ada, OK). Seeds for the KS site were planted on 6 May and transplanted into a plowed field 5 km from the natural population on 25–27 May (Konza Prairie Research Natural Area, Manhatten, KS). Seeds for the MN site were planted on 13 May and transplanted into a plowed agricultural field 110 km northwest of the natural population on 30 May–1 June (University of Minnesota, St. Paul, MN). Previous studies of C. fasciculata have found no evidence of local adaptation at this scale (Galloway and Fenster 2000). The MN site was chosen because the climate was more similar to the native site than closer alternative sites due to the warming effect of the metropolitan area (Baker et al. 1985), but the soil type was quite different (Appendix 1). The objective of this experiment was to assay the three different populations in three different bands of climate broadly defined. To the extent possible, I controlled other aspects of the growth conditions that differed among gardens. Seedlings were provided with the same inoculum because poor performance of nonlocal populations due to a lack of a compatible Rhizobium symbiont would have obscured the climate effect. Garden plots were plowed initially to eliminate competition and weeded as necessary thereafter to reduce the effects of different assemblages of competing species (one native, two agricultural). The OK site was weeded in June; KS and MN sites were weeded in July and September. To reduce the effect of chance differences in local weather at planting times, seedlings at all sites were watered as necessary for a maximum of five days after transplantation at all sites (OK watered once followed by rains, KS site watered twice followed by rain, MN watered three times over the course of five days). To minimize deer herbivory, which differed radically among sites, the OK field was enclosed with a rope fence and the KS field with an electric fence (deer damage in OK 3.1%, KS 1.2%). Summer measurements were taken when the plants were 8–9 weeks old (OK, 20–29 July; KS, 1–10 August; MN, 13– 30 August). I collected and pressed the uppermost fully expanded leaf and recorded leaf number for all plants except those in one block in each of KS and MN sites. Specific leaf area (SLA) is leaf area/dry leaf weight (m2/g). The reproductive stage of the plant was also scored: 0, died before reproduction; 1, vegetative plant; 2, flowering plant without pods; 3, plant with developing green pods; 4, plant with ripening brown pods; 5, plant senescing following reproduction. Autumn measurements were taken to track the natural end 1449 SELECTION ALONG A CLIMATE GRADIENT of the growing season (MN, 23 September–2 October; KS, 5–12 October; OK, 14–21 October). Pods . 2 cm in length were counted. Because seed number per pod differed dramatically among sites and populations, I counted the seed number of three pods per plant and estimated fecundity as average seed count 3 pod number. In some cases, especially MN genotypes in the OK site, pods had dehisced and the plants senesced; in this case I counted the number of pedicels that had remnant pod fragments still attached. In cases where seed counts were not available because pods had already dehisced, I estimated fecundity using the average seed count of the other full-sib replicate within the block or, if that was not available, the average seed count of the full-sib family across blocks. The use of alternative seed counts was necessary for 879 MN, 120 KS, and 59 OK plants across all sites. Data Analysis To assess mean trait differences among populations and sites, ANOVA and MANOVA analyses were performed on all traits (JMP, ver. 3.1, SAS Institute 1995). Block is considered a random effect; population and site are considered fixed effects. Leaf number and specific leaf area were loge transformed. So that selection analyses could be performed for all traits jointly, reproductive stage was analyzed as a continuous trait even though it was discrete and the temporal spacing between the stages may not be even. This trait was not transformed because this did not improve the data structure. I used the multiple regression methods of Lande and Arnold (1983) to characterize the dynamic interface between phenotypes and the environment. The genetic relationship between traits and fitness (e.g., Rausher 1992) and predictions of selection response have been reported previously (Etterson and Shaw 2001). Phenotypic selection analyses were done at two levels. To compare patterns of selection imposed by environmental factors at different latitudes, I pooled data from all plants reared at a particular site to maximize the phenotypic range and enhance detection of selection (site-specific analyses). However, because the populations are genetically divergent and may respond differently to the different environments, selection may differ among the populations within and among garden sites. Therefore, I also performed phenotypic selection analyses for each population and site combination (population-specific analyses). Partial regression coefficients of linear multiple regressions of relative fitness on traits are interpreted as the measure of direct selection, the selection gradient (bi) on trait i (Lande and Arnold 1983). The covariance between relative fitness and trait i, the selection differential (Si), is interpreted as a composite measure of direct linear selection on trait i and indirect linear selection mediated by phenotypic correlations with other traits (Robertson 1966; Price 1970; Lande and Arnold 1983). Estimates of the curvature of selection surfaces, or the adaptive landscape, were obtained from multiple regressions that included as predictors quadratic (gii) and cross-product functions of the traits (gij). The gii-values reflect curvature in the selection surface and indicate stabilizing (disruptive) selection if the peak (valley) of fitness corresponds to an intermediate phenotype (Endler 1986; Mitchell-Olds and Shaw 1987). The gijvalues reflect selection on trait i that varies depending upon the value of trait j (Phillips and Arnold 1989; Brodie et al. 1995). Higher order interactions (. second) were not explored because of difficulty of visualizing four-dimensional data necessary for interpretation of the results. Regression analyses were performed on JMP (ver. 3.1, SAS Institute 1995) and covariances were estimated using SYSTAT (ver. 7.0, SPSS 1997). Depending on the analysis, relative fitness was calculated by dividing individual fecundity by site mean fecundity or population mean fecundity for the site-specific and population-specific analyses, respectively. When relative fitness was left untransformed, the assumption of normality of the residuals was violated in all analyses as indicated by a significant Spearman rank correlation between the residuals and relative fitness (Neter et al. 1996). Consequently, tables report selection differentials and gradients based on analyses that used untransformed relative fitness but P-values are from analyses where relative fitness was loge transformed (Mitchell-Olds and Shaw 1987). All predictor variables were standardized to a mean of zero and standard deviation of one. Correlations among predictor variables for the site- and population-specific analyses are provided in Appendices 2 and 3. To test for differences in selection corresponding to latitude in the site-specific analyses, I pooled data from all gardens and analyzed by ANCOVA with a model that included a categorical term coding for site, the three traits, and the site 3 trait terms (Zar 1999). The finding of significant interaction terms in all ANCOVA analyses was followed by one degree of freedom pairwise contrasts to determine which slopes differed. In the population-specific analyses, patterns of selection that are associated with clines in phenotype were investigated in two ways. First, to examine differences in selection on a single population raised along the environmental gradient, I pooled all data for a population across sites and analyzed with a model that included site as a factor, the traits, and the site 3 trait terms. Secondly, to examine differences in selection among populations within a single site, I pooled all data from a single site across populations and analyzed with a model that included population as a factor, the traits, and the population 3 trait terms. In all ANCOVA analyses, homogeneity of variance was verified by visual inspection of the distribution of residuals. Bivariate nonlinear selection gradients were visualized by three-dimensional mesh plots with predicted relative fitness on the vertical axis and trait values on the horizontal axes. Grid coordinates were calculated with the SigmaPlot interpolation function (intervals 5 15, weight 5 3, SigmaPlot 1999). RESULTS Mortality during the field season was generally low but varied between sites and populations. Plants in MN had the highest mortality rate (6.2%) followed by OK (1.8%) and KS (1.5%). For percent mortality, the rank among the populations was consistent across sites; MN genotypes had the highest mortality in all sites (MN site 7.9%, KS 2%, OK 3.2%), 1450 JULIE R. ETTERSON FIG. 2. Least squares means and standard errors (very small) of four traits (A) fecundity; (B) reproductive stage; (C) leaf number; and (D) specific leaf area measured on three populations (MN, circles; KS, triangles; OK, squares) in three sites. Black squares show the mean for one site averaged across populations and the arrow indicates the direction of phenotypic selection. followed by OK genotypes (MN site 6.2%, KS 1.6%, OK 1.2%), and KS genotypes (MN site 5.0%, KS 0.8%, OK 1.1%). On average, plants exhibited plastic responses to the different environments as indicated by significant site effects (Fig. 2, Table 1). Fecundity was highest in KS, followed by OK and MN (Fig. 2A). Clinal trends corresponding to the latitude were apparent for the rate of phenological development and leaf number. Plants in OK were the fastest to develop, had the fewest leaves, and the smallest specific leaf area (thickest leaves; Fig. 2B–D). Plants in MN had the op- posite response, and plants in KS were intermediate except for specific leaf area. Responses to the sites differed among populations as indicated by significant site 3 population effects for each trait (Table 1). Native populations at each site had significantly higher estimated lifetime fecundity than alien populations, suggesting local adaptation (Fig. 2A). MN plants had the highest seed production at home and were progressively less fecund in the two more southern sites (31% less than KS plants in KS, 67% less than OK plants in OK). KS plants had the highest seed production in KS and were intermediate TABLE 1. Univariate and multivariate analysis of variance F-test statistics for traits measured on three populations of Chamaecrista fasciculate in a reciprocal transplant experiment in field sites in MN, KS, and OK. * P , 0.05; ** P , 0.001; *** P , 0.0001. F-statistics and significance levels Reproductive stage Fecundity Factor Site Block (site) Population Site 3 population Population 3 block (site) df 2, 9, 2, 4, 18, 9 18 18 18 9769 F 49.5*** 3.1* 5.8* 95.7*** 18.1*** df 2, 7, 2, 4, 14, 7 14 14 14 8356 Leaf number F 32.4** 1.2 90.4*** 6.16** 24.17*** df 2, 7, 2, 4, 14, 7 14 14 14 8073 Specific leaf area F 42.9*** 52.5*** 289.6*** 18.7*** 2.9** df 2, 9, 2, 4, 18, 9 18 18 18 9456 F Wilk’s l 23.2** 5.5** 68.8*** 15.2*** 13.2*** 2399.2*** 57.8*** 1149.9*** 568.2*** 17.6*** 1451 SELECTION ALONG A CLIMATE GRADIENT TABLE 2. Site-specific selection analyses. Standardized selection differentials (Si) were estimated as the covariance between relative fitness and measured traits after blocking factors were taken into account (blocking factors not shown). Significance levels are Bonferroni probabilities of Pearson correlation coefficients. Standardized directional selection gradients, bi (SE) were obtained from linear multiple regressions of relative fitness on three traits (all populations within a site included). Whole-model statistics: MN, F6,2290 5 74.82***; KS, F5,2365 5 309.92***; OK, F6,3105 5 179.98***. Variance inflation factors (VIF) , 1.8 for all traits in all analyses. gii, indicating stabilizing or disruptive selection, and gij, indicating joint selection on pairs of traits, were obtained from separate multiple regressions that included quadratic and cross-product terms. Whole model statistics: MN, F11,2287 5 72.89***; KS, F11,2359 5 194.24***; OK, F11,3099 5 110.02***. VIF , 2.9 for all factors except RS 3 RS in MN and KS analyses where VIF . 10. Differences in selection among the sites are indicated by significant site 3 trait interactions from ANCOVA. Subscripts show which b-values differ according to 1-df contrasts with a 5 0.05. Of the 21 contrasts conducted, one would except to erroneously assign significance in 1.05 cases by chance alone. RS, reproductive stage; LN, leaf number; SLA, specific leaf area. † P , 0.10; * P , 0.05; ** P , 0.01; *** P , 0.001. Selection analyses Trait RS LN SLA RS 3 RS LN 3 LN SLA 3 SLA RS 3 LN RS 3 SLA LN 3 SLA S b S b S b gii gii gii gij gij gij ANCOVA MN site KS site OK site 0.529*** 0.737 (0.038)a*** 0.208*** 0.405 (0.038)a*** 20.010 0.094 (0.037)a† 1.000 (0.118)a*** 0.088 (0.023)a 20.030 (0.020)a* 0.356 (0.042)a*** 20.091 (0.039)a*** 20.010 (0.035)a 20.269*** 20.131 (0.019)b*** 0.458*** 0.539 (0.019)b*** 20.245*** 20.127 (0.015)b*** 0.062 (0.016)b*** 0.173 (0.012)b*** 20.003 (0.006)b 20.032 (0.018)b* 20.073 (0.017)a 20.040 (0.017)a** 20.405*** 21.004 (0.026)c*** 0.632*** 0.599 (0.026)b*** 20.335*** 20.232 (0.023)c*** 20.090 (0.029)c*** 0.195 (0.020)c*** 20.004 (0.003)b* 20.044 (0.028)b*** 0.030 (0.026)b 20.063 (0.025)a† in both MN (60% less than MN plants) and OK (11% less than OK plants). OK plants also produced the most seed in their home site and were progressively less fecund in KS and MN (7% less than KS plants in KS, 96% less than MN plants in MN). Low fecundity of all populations at the OK site was likely attributable to the uncharacteristically hot and dry weather during the growing season (Appendix 1). MN plants had the most rapid development and the fewest leaves, regardless of the site (Fig. 2B, C). Although MN genotypes had thinner leaves than other populations overall, they had intermediate leaf thickness in MN and had an anomalous response in KS producing very thin leaves, which could have been a manifestation of extreme competition for light among the relatively small MN plants at this site (Fig. 2D). KS and OK populations developed slowly in MN but differed less in KS or OK (Fig. 2B). These populations showed parallel plastic responses with respect to leaf number and specific leaf area that corresponded to latitude (Fig. 2C, D). Site-Specific Phenotypic Selection Analyses Linear selection. Latitudinal trends in S and b were apparent for each trait in the site-specific selection analyses with all populations pooled (Table 2). The rank order of selection coefficients for each trait corresponds to latitude and shows a cline in selection. Significant site 3 trait factors in ANCOVA and post hoc contrasts show that b differed among the sites for all traits except for leaf number in KS and OK. In MN, plants with shorter phenologies were favored by selection, whereas plants with longer phenologies were favored in KS and to a greater extent in OK. In MN, there was a tendency for thinner leaves to be favored, although neither b (P 5 0.09) nor S was statistically significant. In contrast, there was significant selection for thicker leaves in both KS and OK as estimated by either the S or b. Plants with more leaves had higher fitness in all sites, although the Site 3 trait F2,7763 5 274.84*** F2,7763 5 227.85*** F2,7763 F2,7763 F2,7763 F2,7763 F2,7763 F2,7763 F2,7763 5 5 5 5 5 5 5 30.60*** 70.28*** 42.68*** 1.46 65.08*** 9.25*** 1.32 strength of selection was significantly weaker in MN, where plants were relatively large. The direction of the selection was toward the phenotypic mean of native populations for all traits in each site except for leaf number in MN (arrows in Fig. 2). This either suggests that trait values of the native population are adaptive or that traits other than those measured are under selection and confer high fitness on plants from the local population, causing a spurious relationship between fitness and local values of the chosen traits. To test the latter possibility, the local population, which comprises one-third of the observations, was removed from the analyses and, in all cases, the selection gradients did not differ in sign from the analyses with all the data. This verifies the relationship between fitness and these specific traits (analyses not shown). Quadratic selection. The shape of selection surfaces changed clinally with latitude for five of six traits according to the rank order of gii (Table 2). gii for reproductive stage was positive in MN and KS but negative in OK, and all selection coefficients were significantly different. Estimates of gii for leaf number in MN were weak and negative but were stronger and significantly different in KS and OK. A trend was also evident for leaf thickness for which gii was significantly positive in MN and weaker in KS (ns) and OK, although these coefficients were not significantly different. With the exception of specific leaf area in OK, visualization of these selection surfaces over the range of the observed data did not support the interpretation of stabilizing or disruptive selection because the maximum or minimum of the curve coincided with the edge of the phenotypic distribution rather than with intermediate phenotypes. Clinal patterns of joint selection that favored combinations of traits over and above direct selection alone were found for all pairs of traits as indicated by the rank order of gij (Table 2). In MN, early-developing plants with more and thinner S b S b S b gii gii gii gij gij gij RS LN SLA RS 3 RS LN 3 LN SLA 3 SLA RS 3 LN RS 3 SLA LN 3 SLA Trait 0.027 0.061 (0.045)a,1 0.365*** 0.392*** (0.024)a,1 20.152*** 20.085*** (0.024)a 0.191 (0.320)1 0.038 (0.015) 20.026† (0.016)a 0.044 (0.044)a 20.020 (0.051) 20.050 (0.025) MN site 0.027 0.025 (0.033)ab 0.329*** 0.454*** (0.033)b,1 20.044 20.059* (0.027)b 0.092† (0.275) 0.070*** (0.022)1 20.018** (0.011)b 0.108* (0.036)b 0.008 (0.035) 20.027 (0.027)1 KS site MN population MN site 0.266*** 0.451*** (0.086)a,2 0.349*** 0.365*** (0.070)a,2 20.107 0.005 (0.070) 0.547*** (0.222)a,12 0.146 (0.044) 20.006 (0.027) 0.015 (0.084) 20.060 (0.079) 0.002 (0.059) OK site 20.004 0.029† (0.080)b 0.028*** 0.328*** (0.077)c,1 20.015 20.180† (0.070)b 20.016 (0.009)1 0.011 (0.007) 0.001 (0.001)b 0.001 (0.010)ab 20.002 (0.010) 20.017* (0.009)1 0.001 0.364 (0.268)ab 0.321*** 0.441*** (0.022)b,2 20.020 20.003 (0.018) 1.260 (2.445)ab 0.099*** (0.011)1 20.004 (0.007) 0.496 (0.275) 20.792 (0.578) 0.002 (0.017)1 KS site KS population Selection analyses 20.001 0.048 (0.033)b 0.375*** 0.383*** (0.028)a,2 0.008 20.002 (0.029) 20.009 (0.008)b,1 0.054 (0.017) 20.005 (0.008) 0.003 (0.025) 0.050 (0.037) 0.023 (0.032)b,2 OK site 2.059*** 2.700*** (0.358)a,2 0.234 0.211 (0.376)a,3 20.560 20.609* (0.360) 2.209*** (0.419)a,2 0.011 (0.248)a 20.041 (0.254) 1.286 (0.422) 20.378 (0.394) 20.424 (0.363) MN site 20.011 0.047* (0.026)b 0.256*** 0.394*** (0.030)b,3 20.092** 20.059** (0.022) 0.007 (0.026)b 0.108*** (0.019)b,2 20.002 (0.008) 0.049 (0.027) 20.023 (0.022) 0.035 (0.025)2 KS site OK population 20.024 0.029 (0.029)b 0.444*** 0.458*** (0.028)c,1 0.006 0.004 (0.027) 20.078*** (0.028)b,2 0.124** (0.020)b 0.002 (0.004) 0.048 (0.028) 0.047† (0.023) 0.077* (0.034)2 OK site TABLE 3. Population-specific selection analyses. Standardized selection differentials (Si) for each population in each site estimated as the covariance between relative fitness and measured traits after blocking factors were taken into account (blocking factors not shown). Significance levels for selection differentials are Bonferroni probabilities of Pearson correlation coefficients. Directional selection gradients, bi (SE) were obtained from multiple regressions of relative fitness on three traits. Whole-model statistics (site/ population): MN/MN F5,748 5 78.50***; MN/KS F5,824 5 40.83***; MN/OK F5,704 5 13.61***; KS/MN F5,758 5 53.06***; KS/KS F5,857 5 103.68***; KS/OK F5,732 5 86.54***; OK/MN F6,983 5 6.37***; OK/KS F6,1115 5 32.87***; OK/OK F6,990 5 50.27***. Variance inflation factors (VIF) , 1.6 for all traits in all analyses. Univariate nonlinear selection gradients (gii) and bivariate nonlinear selection gradients (gij) for each population in each field site obtained from separate multiple regressions including quadratic and cross-product terms. Whole model statistics (site/population): MN/MN F11,742 5 37.83***; MN/KS F11,818 5 21.14***; MN/OK F11,698 5 10.93***; KS/MN F11,752 5 28.86***; KS/KS F11,851 5 59.41***; KS/OK F11,726 5 44.73***; OK/MN F12,977 5 4.01***; OK/KS F12,1109 5 17.64***; OK/OK F12,894 5 30.61***. Variance inflation factors (VIF) , 2.9 for all factors except RS 3 RS: MN site MN and KS populations (.10); OK site OK population. RS, reproductive stage (,5.5). Differences in selection for a single population raised in different sites are indicated by significant site 3 trait interactions from ANCOVA. Subscripts letters show which b-values are significantly different according to 1-df contrasts with a 5 0.05. Differences in selection for among populations at a single site are indicated by significant population 3 trait interactions from ANCOVA. Subscript numbers show which b-values are significantly different according to 1-df contrasts, with a 5 0.05. Of the 20 contrasts conducted, one would expect to erroneously assign significance in one case by change alone. LN, leaf number; SLA, specific leaf area. † P , 0.10; * P , 0.05; ** P , 0.01; *** P , 0.001. 1452 JULIE R. ETTERSON F2,3109 5 3.34* F2,3109 5 1.97 F2,3109 5 1.24 F2,3109 5 0.64 F2,3109 5 1.91 F2,3109 5 8.59*** F2,3109 5 0.75 F2,3109 5 112.58*** OK site F2,3109 5 2.33† SELECTION ALONG A CLIMATE GRADIENT 1453 leaves were particularly favored (Fig. 3a, d), whereas in KS and OK, later-developing plants with more and thicker leaves were favored (Fig. 3b, c, e, f). Selection on the rate of development and leaf thickness was significantly different in OK. In KS and to a greater extent in OK (Fig. 3e, f), there was selection for plants with more and thicker leaves but gij did not differ significantly among sites. F2,371 5 6.07** F2,298 5 0.27 F2,2815 5 0.16 F2,2518 5 0.46 LN 3 SLA F2,2445 5 2.77† F2,371 5 0.57 F2,298 5 0.05 F2,2815 5 0.63 F2,2518 5 0.70 RS 3 SLA F2,2445 5 1.61 F2,2371 5 0.96 F2,298 5 0.62 F2,2815 5 2.28† F2,2518 5 2.92* RS 3 LN F2,2445 5 0.49 F2,2371 5 0.68 F2,298 5 0.46 F2,2815 5 0.18 F2,2518 5 4.69** SLA 3 SLA F2,2445 5 0.24 F2,2371 5 3.48* F2,298 5 2.58† F2,2815 5 1.12 F2,2518 5 2.19 LN 3 LN F2,2445 5 3.92* F2,2371 5 0.65 F2,298 5 6.12** F2,2445 5 13.64*** F2,2518 5 1.45 RS 3 RS F2,2815 5 12.88*** F2,2371 5 0.85 F2,2284 5 1.72 F2,2815 5 0.29 F2,2518 5 5.58** SLA F2,2445 5 2.67† F2,2371 5 79.13*** F2,2284 5 64.62*** F2,2445 5 90.56*** F2,2815 5 59.83*** F2,2518 5 203.25*** LN KS site F2,2371 5 0.62 F2,2284 5 14.26*** F2,2445 5 5.65** F2,2815 5 25.13*** F2,2518 5 8.69*** Site 3 trait RS Trait MN population KS population OK population ANCOVA TABLE 3. Extended. MN site Population 3 trait Population-Specific Phenotypic Selection Analyses Linear selection: comparisons of single populations across sites. Clinal trends were apparent in S and b for the rate of phenological development for all populations across sites (Table 2, ANCOVA site 3 trait F-statistics and letter subscripts). The pattern of selection of leaf number was clinal for the OK population, showing progressively stronger selection favoring plants with more leaves from north to south. However, for MN and KS populations, selection favored plants with more leaves to the greatest extent in KS. MN plants with thick leaves were strongly favored in OK but more weakly in MN and KS. Quadratic selection: comparisons of single populations across sites. Clinal patterns in nonlinear selection coefficients were detected in three of five cases where selection differed significantly among populations across sites (Table 2). For the OK population, strong positive gii was estimated for reproductive stage in MN but it was significantly negative in OK. Also for this population, weak positive gii was estimated for leaf number in the MN site that was progressively stronger in southern sites. For the MN population, gii for leaf thickness was significantly more negative in MN than in the two more southerly sites. A significant but nonclinal difference in gii was detected for reproductive stage of the KS population. Joint selection on pairs of traits, gij, differed among sites in only one case and did not show clinal patterns (reproductive stage 3 leaf number, MN population). Linear selection: comparisons among populations at a single site. Clinal patterns of selection were evident when S or b were compared among populations reared in a single site for all nine cases where significant population 3 trait terms were estimated by ANCOVA (Table 2, numerical subscripts). In the MN site, rapid phenology was weakly favored among MN plants but progressively more strongly favored among KS and OK plants. In both the MN and KS sites, larger MN plants were strongly favored, but progressively less so among KS and OK plants. The opposite clinal pattern of selection on leaf number was apparent at the OK site. For 13 of the 15 statistically significant selection gradients estimated in these analyses, b, the measure of direct selection was larger, and more often statistically significant than S, the composite measure of selection that takes into account trait correlations. This indicates that, in most cases, trait correlations are not reinforcing selection but are having antagonistic effects that diffuse the strength of selection. Quadratic selection: comparison among populations at a single site. The pattern among nonlinear coefficients of selection was clinal for all significant population 3 trait tests (Table 2). In the MN site, weak positive gii was estimated for reproductive stage among MN plants which was progressively stronger among KS and OK plants. In contrast, gii 1454 JULIE R. ETTERSON FIG. 3. Interpolated bivariate fitness surfaces showing significant joint selection on reproductive stage and leaf number in (a) MN; (b) KS; and (c) OK; (d) joint selection on specific leaf area and reproductive stage in MN; and joint selection on specific leaf area and leaf number in (e) KS and (f) OK. Negative fitness values are an artifact of the interpolation function. for reproductive stage was negative for all populations in the OK site but significantly more so for the OK population. In the KS site, gii for leaf number was small for MN plants and became progressively larger among KS and OK plants. With the exception of gii for leaf number of MN plants in KS, examination of the curvature of these selection surfaces did not indicate patterns of stabilizing or disruptive selection. Patterns of gij among populations in the KS and OK sites suggested that individuals with more and thicker leaves were favored among MN plants, whereas individuals with fewer and thinner leaves were favored among KS and OK plants. DISCUSSION This study provides strong evidence that climate change will alter selection regimes and challenge the evolutionary potential of C. fasciculata across its range. When northern populations were grown in selective environments that mimicked predicted climate change, they suffered substantial losses in fitness. Each population of this study is currently adapted to its native environment, as indicated by significantly higher seed production of plants native to each site. Native trait values contribute to this local adaptation as evidenced by selection toward the mean of home populations. Clinal patterns in selection across this environmental gradient were repeatedly found in analyses that employed an expanded phenotypic distribution, in analyses of single populations, and in nonlinear aspects of selection. As climate becomes warmer and drier in the Great Plains, selection will favor different phenotypes and maintenance of fitness and population stability will depend on evolutionary change. In principle, populations could respond immediately to climate change, maintaining fitness by adaptive plastic responses (Bradshaw 1965; Schlichting 1986; Sultan 1987; van Tienderen 1991; Schlichting and Pigliucci 1998; Donohue et al. 2001). For leaf number and thickness, the C. fasciculata populations of this study exhibited substantial phenotypic plasticity that could be considered adaptive because it was in the direction of trait values favored by selection. However, even though plasticity is in an adaptive direction here, it does not maintain fitness across environments. Nevertheless, responses to selection on genetically based plasticity may help mitigate fitness losses as climate gradually changes. In this study I presented evidence suggesting that evolutionary response to a northerly displacement of selection regimes could occur. Clines in selection that correspond to the pattern of standing phenotypic variation were found for all traits along this climate gradient in the Great Plains. The findings support the hypothesis that, in the past, environmental gradients induced clinal patterns in selection that ultimately produced genetically based clinal patterns in phenotype through adaptive evolution. A shift in the cline in selection may elicit a similar evolutionary response in the future. Evolution of clinal patterns in phenotypic variation has been documented for a number of invasive species since the time of their arrival on the continent (Lacey 1984, 1988; Reinartz 1984; Weber and Schmid 1998; Huey et al. 2000). 1455 SELECTION ALONG A CLIMATE GRADIENT The primary environmental gradients across the range of this study are climate and photoperiod. Other site-specific environmental factors such as soils, herbivory, and competition are not expected to be clinally arrayed. Consequently, it is likely that these factors contribute to noise rather than pattern in the data. The finding of clinal patterns despite the noise that arises from uncontrolled site-specific factors lends stronger support to the inference that climate and/or photoperiod govern the patterns reported here. Nevertheless, replication of this experiment either spatially or temporally would increase confidence in the results because the current experimental design confounds the effects of climate zone and site. Previous studies of C. fasciculata and other species have demonstrated that selection regimes can vary temporally and over relatively small spatial scales (Kalisz 1986; Gibbs and Grant 1987; Kelly 1992; Gilbert et al. 1996), although other studies have shown temporal and spatial stability of selection for some traits (Rice and Mack 1991). To the extent that the spatial gradient in climate across the Great Plains parallels temporal change in climate into the future, this study suggests that a cline in selection will shift north and different trait values will be favored in natural populations. Traits associated with a longer growing season, such as slower phenological development, may be favored by selection in northern populations. Likewise, selection is likely to favor traits that confer higher fitness in more drought-prone environments, such as thicker leaves. The adaptive hypothesis for one trait in this study was not supported; plants with fewer leaves were not favored in the hotter and drier environment, despite a greater potential for water conservation. Common-garden studies in the greenhouse consistently showed a cline in leaf number with southern plants having fewer leaves (Etterson 2001). The unexpected direction of selection may be attributable to plastic responses of plants grown in OK, where all plants had relatively few leaves. If the phenotypic distribution could have been extended to include plants having more leaves, selection may have favored reduced leaf number. Alternatively, small plants in OK (primarily MN genotypes) may have had low fecundity for reasons other than leaf number. Several other studies have found selection in the opposite direction than expected based on standing ecotypic variation. Bennington and McGraw (1995) found selection for increased plant height within in a locally adapted population of Impatiens pallida characterized by short stature. Likewise, Andersson (1996) found a positive relationship between plant height and fitness within a synthetic hybrid population having an extended phenotypic range for height in a site that is typically occupied by plants of a prostrate ecotype of Crepis tectorum. Broadening of the phenotypic distribution provided a clearer picture of selection corresponding to latitude as illustrated by differences between site- and population-specific analyses. For instance, over an expanded range of phenologies, a strong pattern emerged suggesting that rapid development was favored in the north but the slower development in the south (site-specific analyses). However, within the range of variation expressed by each population in each environment, plants with more rapid phenologies were consistently favored by direct selection (population-specific analysis). Similarly, site-specific analyses showed that selection favored thicker leaves most strongly in the south. In contrast, consideration of selection of only single populations in the gradient of environments would have led to the opposite conclusion because adaptive phenotypic responses restrict the range of phenotypes within each population, thus obscuring the overall selection. Despite substantial plastic responses of the populations in response to the different environments, the phenotypic distribution was extended in at least one direction in all environments in the site-specific analyses. However, in five of nine cases, the distribution was extended in only one direction relative to the local population, which may have reduced the opportunity to detect stabilizing selection on traits. This study has demonstrated a cline in selection that corresponds to a cline in phenotype across the northern extent of the geographical range of C. fasciculata. Adaptive hypotheses concerning clinal patterns for phenology and specific leaf area were supported by observed patterns of genetic differentiation among populations coupled with demonstration of spatial variation in selection. Further support for the adaptive nature of trait values of native populations is indicated by selection toward the phenotypic mean of local populations. The companion to this paper (Etterson 2004) characterizes the genetic architecture of populations across environments that influences the extent to which populations may evolve in response to altered selection regimes with climate change (Etterson and Shaw 2001). ACKNOWLEDGMENTS I thank D. N. Alstad, D. A. Andow, M. B. Davis, C. G. Eckert, R. G. Shaw, J. Willis, and four anonymous reviewers for their valuable comments on previous drafts of this manuscript. I thank the Minnesota and Oklahoma Chapters of the Nature Conservancy for allowing seed collections at KelloggWeaver Dunes and Pontotoc Ridge and the helpful staff at the Konza Prairie Scientific and Natural Area and the Environment Protection Agency’s Office of Research and Development and the Ground Water and Ecosystem Restoration Division in Ada, Oklahoma, who provided essential support throughout this research. I express my deep gratitude to the many people who helped with the greenhouse crosses and field work, especially M. Etterson, D. Otterson, R. Otterson, A. Mertyl, and J. Larson, L. Kinsell, and T. Nguyen. This work was supported by U.S. Environmental Protection Agency STAR fellowship (U 914758-01-2), summer grants from the Minnesota Center for Community Genetics, and DaytonWilkie grants for field research. LITERATURE CITED Ågren, J., and D. W. Schemske. 1994. Evolution of trichome number in a naturalized population of Brassica rapa. Am. Nat. 143: 1–13. Aizen, M. A., and H. Woodcock. 1992. Latitudinal trends in acorn size in eastern North American species of Quercus. Can. J. Bot. 70:1218–1222. Andersson, M. 1982. Female choice selects for extreme tail length in a widowbird. Nature 299:818–820. Andersson, S. 1996. Phenotypic selection on plant height in a segregating hybrid population of Crepis tectorum (Asteraceae). Int. J. Plant Sci. 157:488–492. Andersson, S., and B. Widén. 1993. Pollinator-mediated selection 1456 JULIE R. ETTERSON on floral traits in synthetic populations of Senecio integrifolius (Asteraceae). Oikos 66:72–79. Anholt, B. R. 1991. Measuring selection on a population of damselflies with a manipulated phenotype. Evolution 45:1091–1106. Baker, D. G., E. L. Kuehnast, and J. A. Zendlo. 1985. Climate of Minnesota. XV. Normal temperatures (1951–1980) and their application. Available via http://climate.umn.edu/doc/historical. htm. Bennington, C. C., and J. B. McGraw. 1995. Natural selection and ecotypic differentiation in Impatiens pallida. Ecol. Monogr. 65: 303–323. Bradshaw, A. D. 1965. Evolutionary significance of phenotypic plasticity in plants. Adv. Genet. 13:115–155. Brodie, E. D., III, A. J. Moore, and F. J. Janzen. 1995. Visualizing and quantifying natural selection. Trends Ecol. Evol. 10: 313–318. Chapin, F. S., III, and M. C. Chapin. 1981. Ecotypic differences of growth processes in Carex aquatilis along latitudinal and local gradients. Ecology 62:1000–1009. Clapham, D. H., I. Dormling, I. Ekberg, G. Eriksson, M. Qamaruddin, and D. Vince-Pru. 1998. Latitudinal cline of requirement for far-red light for the photoperiodic control of budset and extension growth in Picea abies (Norway spruce). Physiol. Plant. 102:71–78. Claussen, J., D. D. Keck, and W. M. Hiesey. 1940. Experimental studies on the nature of species. I. The effect of varied environment on western North American plants. Publication 520. Carnegie Institution of Washington, Washington, D.C. Clevering, O. A., H. Brix, and J. Lukavska. 2001. Geographic variation in growth responses in Phragmites australis. Aquat. Bot. 69:89–108. Comstock, R. E., and H. F. Robinson. 1948. The components of genetic variance in populations of biparental progenies and their use in estimating the average degree of dominance. Biometrics 4:254–266. Donohue, K., E. Hammond Pyle, D. Messiqua, M. Shane Heschel, and J. Schmitt. 2001. Adaptive divergence in plasticity in natural populations of Impatiens capensis and its consequences for performance in novel habitats. Evolution 55:692–702. Dudley, S. A., and J. Schmitt. 1996. Testing the adaptive plasticity hypothesis: density-dependent selection on manipulated stem length in Impatiens capensis. Am. Nat. 147:445–465. Ehleringer, J. R. 1988. Changes in leaf characteristics of a species along elevational gradients in the Wasatch Front, Utah. Am. J. Bot. 75:680–689. Endler, J. A. 1986. Natural selection in the wild. Princeton Univ. Press, Princeton, NJ. Etterson, J. R. 2001. Evolutionary potential of the annual legume, Chamaecrista fasciculata in relation to global warming. Ph.D. diss., University of Minnesota, St. Paul, MN. ———. 2004. Evolutionary potential of Chamaecrista fasciculata in relation to climate change: II. Genetic architecture of three populations reciprocally planted along a climate gradient in the Great Plains. Evolution 58:1459–1471. Etterson, J. R., and R. G. Shaw. 2001. Constraint to adaptive evolution in relation to global warming. Science 294:151–154. Galloway, L. F., and C. B. Fenster. 2000. Population differentiation in an annual legume: local adaptation. Evolution 54:1173–1181. Gehring, J. L., and L. F. Delph. 1999. Fine scale genetic structure and clinal variation in Silene aucaulis despite high gene flow. Heredity 82:628–637. Gibbs, H. L., and P. R. Grant. 1987. Oscillating selection on Darwin’s finches. Nature 327:511–513. Gilbert, F., P. Willmer, F. Semida, J. Ghazoul, and S. Zalat. 1996. Spatial variation in selection in a plant-pollinator system in the wadis of Sinai, Egypt. Oecologia 108:479–487. Givnish, T. J. 1979. On the adaptive significance of leaf form. Pp. 375–407 in O. T. Solbrig, S. Jain, G. B. Johnson, and P. H. Raven, eds. Topics in plant population biology. Columbia Univ. Press, New York. Huey, R. B., G. W. Gilchrist, M. L. Carlson, D. Berrigan, and L. Serra. 2000. Rapid evolution of a geographic cline in size in an introduced fly. Science 287:308–309. Hurme, P., T. Repo, O. Savolainen, and T. Paakkonen. 1997. Climatic adaptation of bud set and frost hardiness in Scots pine (Pinus sylvestris). Can. J. For. Res. 27:716–723. Intergovernmental Panel on Climate Change (IPCC). 2001. Climate change 2001: the scientific basis. Contribution of Working Group I to the Third Asessment Report of the Intergovernmental Panel on Climate Change. J. T. Houghton, Y. Ding, D. J. Griggs, M. Noguer, P. van der Linden, X. Dai, and K. Maskell, eds. Cambridge Univ. Press, Cambridge, U.K. Jonas, C. S., and M. A. Geber. 1999. Variation among populations of Clarkia unguiculata (Onagraceae) along altitudinal and latitudinal gradients. Am. J. Bot. 86:333–343. Jordan, N. 1991. Multivariate analysis of selection in experimental populations derived from hybridization of two ecotypes of the annual plant Diodia teres W. (Rubiaceae). Evolution 45: 1760–1772. Kalisz, S. 1986. Variable selection on the timing of germination in Collinsia verna (Scrophulariaceae). Evolution 40:470–491. Kelly, C. A. 1992. Spatial and temporal variation in selection on correlated life-history traits and plant size in Chamaecrista fasciculata. Evolution 46:1658–1673. Kingsolver, J. G. 1996. Experimental manipulation of wing pigment pattern and survival in western white butterflies. Am. Nat. 147: 296–306. Kingsolver, J. G., H. E. Hoekstra, J. M. Hoekstra, D. Berrigan, S. N. Vignieri, C. E. Hill, A. Hoang, P. Gilbert, and P. Beerli. 2001. The strength of phenotypic selection in natural populations. Am. Nat. 157:245–261. Körner, C., G. D. Farquhar, and S. C. Wong. 1991. Carbon isotope discrimination by plants follows latitudinal and altitudinal trends. Oecologia 88:30–40. Lacey, E. P. 1984. Seed mortality in Daucus carota populations: latitudinal effects. Am. J. Bot. 71:1175–1182. ———. 1988. Latitudinal variation in reproductive timing of a short-lived monocarp, Daucus carota (Apiaceae). Ecology 69: 220–232. Lande, R., and S. J. Arnold. 1983. The measurement of selection on correlated characters. Evolution 37:1210–1226. Lewontin, R. C. 1974. The genetic basis of evolutionary change. Columbia Univ. Press, New York. Linhart, Y. B., and M. C. Grant. 1996. Evolutionary significance of local genetic differentiation in plants. Annu. Rev. Ecol. Syst. 27:237–277. Mayr, E. 1956. Geographical character gradients and climatic adaptation. Evolution 10:105–108. McGregor, R. L. 1986. Flora of the Great Plains. Univ. Press of Kansas, Kansas City, KS. Mitchell-Olds, T., and R. G. Shaw. 1987. Regression analysis of natural selection: statistical inference and biological interpretation. Evolution 41:1149–1161. Montagnes, R. J. S., and D. H. Vitt. 1991. Patterns of morphological variation is Meesia triqueta (Bryopsida: Meesiaceae) over an arctic-boreal gradient. Syst. Bot. 16:726–735. Nagy, E. S. 1997. Selection for native characters in hybrids between two locally adapted plant subspecies. Evolution 51:1469–1480. Neter, J. M., H. Kutner, C. J. Nachtsheim, and W. Wasserman. 1996. Applied linear regression models. McGraw-Hill, Columbus, OH. Neuffer, B., and S. Bartelheim. 1989. Gen-ecology of Capsella bursa-pastoris from an altitudinal transsect in the Alps. Oecologia 81:521–527. Nilsson, L. A. 1988. The evolution of flowers with deep corolla tubes. Nature 344:147–149. O’Connell, L. M., and M. O. Johnston. 1998. Male and female pollination success in a deceptive orchid, a selection study. Ecology 79:1246–1260. Phillips, P. C., and S. J. Arnold. 1989. Visualising multivariate selection. Evolution 43:1209–1222. Price, G. R. 1970. Selection and covariance. Nature 227:520–521. Rausher, M. D. 1992. The measurement of selection on quantitative traits: biases due to environmental covariances between traits and fitness. Evolution 46:616–626. Reale, D., A. D. McAdam, S. Boutin, and D. Berteaux. 2003. Ge- 1457 Andropogon gerardi, A. scoparius 42.27 (218.94) 220 gravelly sandy loam 220 gravelly sandy loam 2,900 1 Dominant grasses Data from the National Climatic Data Center, available via http://www5.ncdc.noaa.gov/climatenormals/clim81. 49.25 (212.78) 180 silty clay loam 180 silty clay loam 8,616 Andropogon gerardi, A. scoparius, Sorgastrum nutans 59.69 (7.19) 130 silty loam 62.03 51.16 130 sandy loam 624 Andropogon gerardi, A. scoparius, Koeleria macrantha 61.21 23.21 25.64 (2.43) 61.21 23.21 21.25 21.98 (0.73) 62.03 21.25 17.32 17.20 (20.12) 52.50 16.89 348749 N, 978009 W 348659 N, 968539 W 398129 N, 968369 W 398129 N, 968369 W 448989 N, 938179 W Garden OK Native site Garden KS Native site Garden MN 448319 N, 928009 W Latitude, longitude Average temperature (8C)1 1998 temperature (8C) Precipitation (cm)1 1998 precipitation (cm)1 Growing season (days) Soils Fragment size (acres) Corresponding Editor: J. Willis Native site netic and plastic responses of a northern mammal to climate change. Proc. R. Soc. 270:591–596. Reinartz, J. A. 1984. Life history variation of common mullein (Verbascum thapsus). I. Latitudinal differences in population dynamics and timing of reproduction. J. Ecol. 72:897–912. Rice, K. J., and R. N. Mack. 1991. Ecological genetics of Bromus tectorum. III. The demography of reciprocally sown populations. Oecologia 88:91–101. Robertson, A. 1966. A mathematical model of the culling process in dairy cattle. Anim. Prod. 8:93–108. SAS Institute. 1995. JMP statistics and graphics guide. Ver. 3.1. SAS Institute, Inc., Cary, NC. Sawada, S., Y. Nakajima, M. Tsukuda, K. Sasaki, Y. Hazama, M. Futatsuya, and A. Watanabe. 1994. Ecotypic differentiation of dry matter production processes in relation to survivorship and reproductive potential in Plantago asiatica populations along climatic gradient. Funct. Ecol. 8:400–409. Schlichting, C. D. 1986. The evolution of phenotypic plasticity in plants. Ann. Rev. Ecol. Syst. 17:667–693. Schlichting, C. D., and M. Pigliucci. 1998. Phenotypic evolution: a reaction norm perspective. Sinauer Associates, Sunderland, MA. Schmitt, J., S. A. Dudley, and M. Pigliucci. 1999. Manipulative approaches to testing adaptive plasticity: phytochrome-mediated shade-avoidance responses in plants. Am. Nat. 154:S43–S54. Shaver, G. R., N. Fetcher, and F. S. Chapin III. 1986. Growth and flowering in Eriophorum vaginatum: annual and latitudinal variation. Ecology 67:1524–1535. Sinervo, B. 1999. Mechanistic analyses of natural selection and a refinement of Lack’s and William’s principles. Am. Nat. 154: S26–S42. Sultan, S. E. 1987. Evolutionary implications of phenotypic plasticity in plants. Evol. Biol. 21:127–178. SPSS. 1997. SYSTAT 7.0. Statistics. SPSS, Chicago, IL. ———. 1999. SigmaPlot 5.0. SPSS, Chicago, IL. Thompson, R. S., K. H. Anderson, and P. J. Bartlein. 2000. Atlas of relations between climatic parameters and distributions of important trees and shrubs in North America. U.S. Geological Survey Professional Paper, 1650 A-B. Totland, Ø. 1999. Effects of temperature on performance and phenotypic selection on plant traits in alpine Ranunculus acris. Oecologia 120:242–251. Totland, Ø., H. L. Andersen, T. Bjelland, V. Dahl, W. Eide, S. Houge, T. R. Pedersen, and E. U. Vie. 1998. Variation in pollen limitation among plants and phenotypic selection on floral traits in an early-spring flowering herb. Oikos 82:491–501. van Tienderen, P. 1991. Evolution of generalists and specialists in spatially heterogeneous environments. Evolution 45:1317–1331. Wade, M. J., and S. Kalisz. 1990. The causes of natural selection. Evolution 44:1947–1955. Weber, E., and B. Schmid. 1998. Latitudinal population differentiation in two species of Solidago (Asteraceae) introduced into Europe. Am. J. Bot. 85:1110–1121. Wolff, K. 1988. Genetic analysis of ecological relevant morphological variability in Plantago lanceolata L. 3. Natural selection in an F2 population. Theor. Appl. Genet. 75:772–778. Zar, J. H. 1999. Biostatistical analyses. Prentice Hall, Upper Saddle River, NJ. APPENDIX 1. Latitude and longitude, long-term average (1971–2000) temperature and precipitation during the growing season (April–September), 1998 temperature and precipitation (deviation from normal in parentheses), growing season length, soil substrate, and prairie fragment size of the three sites of population origin and experimental gardens. Climate records for the native and garden sites in KS and OK are from the same weather station (closest). SELECTION ALONG A CLIMATE GRADIENT 1458 JULIE R. ETTERSON APPENDIX 2. Pearson product-moment correlation coefficients from site-specific phenotypic selection analyses. Correlation coefficients indicate the strength of the linear relationships between each pair of predictor variables for each of the three garden plots (all populations within a site included). RS, reproductive stage; LN, leaf number; SLA, specific leaf area; * P , 0.05; ** P , 0.01; *** P , 0.001. Correlation of the predictor variables MN site ri, j i, j RS, LN RS, SLA LN, SLA KS site ri, j 20.186*** 0.013 20.267*** OK site ri, j 20.264*** 0.215 20.066** 20.456*** 0.238*** 20.161*** APPENDIX 3. Pearson product-moment correlation coefficients from population-specific analyses, which summarize the strength of the linear relationships between each pair of predictor variables for each population in each of the three garden plots (populations within sites analyzed individually). RS, reproductive stage; LN, leaf number; SLA, specific leaf area; † P , 0.10; * P , 0.05; ** P , 0.01; *** P , 0.001. Correlation of the predictor variables MN site ri, j i, j RS, LN RS, SLA LN, SLA KS site ri, j OK site ri, j MN population KS population OK population MN population KS population OK population MN population KS population OK population 0.048 20.063† 20.173*** 0.069* 0.085* 20.329*** 20.126*** 0.033 20.389*** 0.172*** 20.041 0.023 0.035 0.076* 0.095* 20.145*** 0.048 0.097** 20.205*** 20.075* 0.110*** 20.102*** 20.053 † 0.041 20.106*** 20.032 20.003