Plant resource-use traits are generally hypothesized to be adaptively differentiated for populations distributed along resource gradients. Although nutrient limitations are expected to select for resource-conservative strategies, water limitations may select for either resource-conservative or -acquisitive strategies. We test whether population differentiation reflects local adaptation for traits associated with resource-use strategies in a desert annual (Helianthus anomalus) distributed along a gradient of positively covarying water and nutrient availability.
We compared quantitative trait variation (QST) with neutral genetic differentiation (FST), in a common garden glasshouse study, for leaf economics spectrum (LES) and related traits: photosynthesis (Amass, Aarea), leaf nitrogen (Nmass, Narea), leaf lifetime (LL), leaf mass per area (LMA), leaf water content (LWC), water-use efficiency (WUE, estimated as δ13C) and days to first flower (DFF).
QST–FST differences support adaptive differentiation for Amass, Nmass, Narea, LWC and DFF. The trait combinations associated with drier and lower fertility sites represent correlated trait evolution consistent with the more resource-acquisitive end of the LES. There was no evidence for adaptive differentiation for Aarea, LMA and WUE.
These results demonstrate that hot dry environments can selectively favor correlated evolution of traits contributing to a resource-acquisitive and earlier reproduction ‘escape’ strategy, despite lower fertility.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Plant traits related to resource use are expected to be adaptively differentiated in habitats differing in resource availability. Low nutrient availability is generally hypothesized to select for resource-conservative strategies associated with slower growth rate (Chapin, 1980; Grime, 1988; Chapin et al., 1993; Arendt, 1997; Wright & Westoby, 2002; Wright et al., 2002; Reich et al., 2003; but see Stanton et al., 2000). Water limitation may also select for resource-conservative strategies and slow growth rate for many life forms, particularly woody perennials and evergreen life forms that tolerate low water potentials (Ludlow, 1989; Dudley, 1996; Etterson, 2004; Wright et al., 2005; Knight et al., 2006). However, annual plants and perennials that go dormant may escape water limitation with a resource-acquisitive strategy and/or accelerated phenology that permits completion of reproduction before water limitation occurs (Ludlow, 1989; Arendt, 1997; Geber & Dawson, 1997; Ackerly et al., 2000; Stanton et al., 2000; McKay et al., 2003; Heschel & Reginos, 2005; Franks et al., 2007; Franks & Weis, 2008; Kigel et al., 2011; Brachi et al., 2012; Ivey & Carr, 2012). We test whether population differentiation in a desert annual (Helianthus anomalus) reflects local adaptation for traits associated more with a resource-acquisitive or resource-conservative strategy in response to a gradient of decreasing water and nutrient availability.
The leaf economics spectrum (LES) is a well-documented pattern of leaf trait correlations that is generally explained in resource economic terms of quantifying leaf income as carbon fixation via photosynthesis and leaf expenditures as metabolic and construction costs (Orians & Solbrig, 1977; Bloom et al., 1985; Reich et al., 1997; Wright et al., 2004). Trait combinations on one end of the spectrum are thought to represent fast-growing species that have the potential for quick returns on investment, with leaves that have resource-acquisitive trait combinations: high photosynthetic rate (Amass,Aarea) to support faster growth, high nitrogen content (Nmass, Narea) to support high photosynthesis, low leaf mass per unit area (LMA) that permits high metabolic rates, and short leaf lifetime (LL) as a result of low LMA (Reich et al., 1997; Wright et al., 2004). The trait combinations at the other end of the spectrum are thought to represent slow-growing species that yield slower returns on investment, with leaves that have more conservative resource traits: low Amass, Aarea, Nmass, Narea, high LMA and long LL (Reich et al., 1997; Wright et al., 2004).While there is some discussion about whether mass- or area-based traits are more important, both are biologically relevant and should be included when possible (Lloyd et al., 2013; Osnas et al., 2013; Westoby et al., 2013). Additional traits such as higher leaf water content (LWC), lower water-use efficiency (WUE), and fewer days to first flower (DFF) are often discussed as traits associated with the resource-acquisitive end of the LES traits spectrum (Cohen, 1970; Geber & Dawson, 1997; Stanton et al., 2000; McKay et al., 2003; Knight et al., 2006; Shipley et al., 2006).
Several different approaches can be used to support the interpretation of different combinations of LES and associated traits as adaptive plant growth strategies (Reich et al., 1997, 1999; Wright et al., 2004; Westoby & Wright, 2006). At the macroevolutionary level, phylogenetically explicit contrasts of species that are largely perennials have demonstrated correlated evolution of a few LES and related trait combinations, providing strong inference for selection on at least some traits and the association of more conservative traits with lower fertility and rainfall sites (Cunningham et al., 1999; Wright & Westoby, 1999; Mediavilla et al., 2008). However, these studies do not identify the relative strength of adaptive differentiation for resource-use traits across habitats, or whether resource limitations select for more resource-conservative or -acquisitive strategies in annuals. At the microevolutionary level, a few within-population studies of predominately annuals have demonstrated phenotypic selection on individual LES and related traits (Stanton et al., 2000; Etterson, 2004; Scheepens et al., 2010; Donovan et al., 2011 and references therein; Galloway & Burgess, 2012; Ivey & Carr, 2012). However, these studies do not identify any consistent direction of phenotypic selection in different resource treatments. Additionally, because response to selection is dependent on heritable variation in the traits targeted by selection (Lande & Arnold, 1983), measurements of phenotypic selection cannot be used to infer adaptive trait evolution. Another microevolutionary approach, performing reciprocal transplants, is the gold standard for demonstrating local adaptation of populations (Clausen et al., 1940; Conner & Hartl, 2004), but trait differentiation in each transplant garden provides only indirect evidence for which traits are most important for local adaptation unless phenotypic selection analyses are additionally included. Common garden studies, including reciprocal transplants, can be used to demonstrate correlations between plant traits and population source site characteristics (e.g. altitude, latitude, climate or resource availability) that infer a response of individual traits to hypothesized selective agents (Billings, 1985; Arntz & Delph, 2001; Christman et al., 2008; Kawakami et al., 2011; Scheepens & Stocklin, 2011; but see Hubner et al., 2013). Correlational analyses (between common garden plant traits and source site characteristics) can be complemented with QST–FST analyses to provide a powerful approach for investigating adaptive evolution of traits in response to gradients (Kawakami et al., 2011; Keller et al., 2011; Dutkowski & Potts, 2012; Frei et al., 2012; Hubner et al., 2013).
QST–FST analyses test for patterns of population differentiation that are indicative of past selection and thus provide evidence of local adaptation. The extent of population differentiation in neutral genetic variation (FST) is calculated as the proportion of total genetic variation that is partitioned among populations (Wright, 1951). FST measured on genetic markers that are selectively neutral provides an estimate of the partitioning of genetic variation that is the result of neutral processes such as drift, mutation and migration. This provides a baseline measurement of population differentiation that can be compared to an analogous parameter, QST, calculated for phenotypic traits (Spitze, 1993; Yang et al., 1996). If QST for a trait is not significantly different from FST, then trait differentiation among populations is not distinguishable from neutral processes and there is no evidence that diversifying selection is causing divergence among populations. If QST is greater than FST, then the trait has diversified more than would be expected by neutral processes alone, providing evidence that populations have responded to diversifying selection and are probably locally adapted. If QST is less than FST, the trait is more similar across populations than expected, suggesting stabilizing selection across the range. Although some aspects of the methodology are currently being debated (Hedrick, 2005; Jost, 2008; Pujol et al., 2008; Edelaar & Bjorklund, 2011; Whitlock, 2011), this is still a powerful method for identifying the signature of past selection on phenotypic traits, allowing tests of hypotheses about the selective pressures driving adaptation. Several QST–FST studies that included a subset of LES and related traits found evidence for selection on some traits but not others (Steane et al., 2006; Ramirez-Valiente et al., 2009; Scheepens et al., 2010; Kawakami et al., 2011; Keller et al., 2011; Frei et al., 2012; Hubner et al., 2013). However, to our knowledge none have focused on the hypothesized response to selection for the suite of LES and related traits across a resource gradient where local adaptation to resource limitations is expected. If different combinations of LES and related traits represent locally adaptive plant growth strategies along a resource gradient, then QST should be greater than FST for multiple LES and related traits, and the correlated trait evolution should favor trait combinations along the multidimensional resource-acquisitive to resource-conservative axis of LES.
Helianthus anomalus Blake, the study species, is an annual sunflower endemic to nutrient-poor desert sand dunes in the southwestern United States. It is a stable homoploid hybrid species derived from Helianthus annuus L. and Helianthus petiolaris Nutt. (Rieseberg, 1991; Schwarzbach & Reiseberg, 2002; Rieseberg et al., 2003). Reciprocal transplant experiments have demonstrated that H. anomalus is adapted to its actively moving sand dune habitat when compared to its ancestral parents grown in that habitat (Donovan et al., 2010). Because H. anomalus grows on active sand dunes, it was originally hypothesized that water limitation was an important selective force driving the evolution of the species (Thompson et al., 1981; Rieseberg, 1991). However, nutrient limitations seem to be as important as or more important than water limitations for H. anomalus productivity in some populations, potentially acting as an additional selective agent (Rosenthal et al., 2005; Ludwig et al., 2006). H. anomalus also appears to be more tolerant of nutrient stress than its ancestral parents based on a lower relative growth rate and higher nutrient-use efficiency, although experiments at the microevolutionary level have demonstrated phenotypic selection for increased Nmass in H. anomalus habitats (Brouillette et al., 2006; Donovan et al., 2007, 2009; Brouillette & Donovan, 2011). Thus, H. anomalus provides a good system for testing adaptive differentiation of resource-related traits.
In this study, we investigate the expectation of adaptive differentiation of H. anomalus along a resource gradient for LES (Amass, Aarea, Nmass, Narea, LMA and LL) and related traits (LWC, WUE and DFF) in a common garden study. Specifically, we test the following hypotheses: population differentiation for trait variation (QST) will be greater than differentiation for neutral genetic variation (FST), providing evidence for diversifying selection; for traits demonstrating evidence of diversifying selection, correlated trait evolution will follow the axis of the LES towards a more resource-acquisitive or resource-conservative strategy; combinations of putatively adaptive traits will be correlated with source site climate and soil fertility characteristics, providing support for these abiotic factors as selective agents driving adaptive differentiation. These results will allow us to determine whether a gradient of positively covarying water and nutrient availability has selected for correlated trait evolution of a more resource-conservative or -acquisitive resource strategy in this desert annual. If a more resource-conservative strategy is associated with drier low fertility sites, then it will not be possible to distinguish between the relative importance of water and nutrient limitations as selective agents. However, if a more resource-acquisitive escape strategy is associated with drier low fertility sites, this would suggest that water limitation has been a more important selective agent than nutrient limitation along this resource gradient.
Materials and Methods
The glasshouse experiment was conducted at the University of Georgia Plant Biology Greenhouses in Athens, GA, USA. Achenes (hereafter ‘seeds’) for the study were collected as maternal half-sibling families from eight natural populations of Helianthus anomalus S.F. Blake in August 2007 (Fig. 1, Supporting Information, Table S1). For germination, seeds were scarified (blunt end removed) on 10 January 2008, soaked overnight in 0.005% solution of fusicoccin to break dormancy, and then germinated on filter paper moistened with deionized water. Seedlings were transplanted into azalea pots, 25 cm in diameter and 19 cm deep, filled with a 3 : 1 sand : Turface mixture (Profile Products, Buffalo Grove, IL, USA) in the glasshouse on 13–14 January 2008. The planned experimental design was a randomized complete block design with a target of 12 families from each of eight populations replicated once in each of three blocks (N =288). However, mortality and poor germination of seeds from White Sands (WHS) and Jericho (JER) populations resulted in an unbalanced design of eight to 14 families and 15–45 individuals per population (Table S1). Plants were watered daily using an automated drip irrigation system and fertilized three times each wk with half-strength Hoagland's solution (Epstein & Bloom, 2005) applied after the automatic irrigation, resulting in high-fertility common garden conditions that were probably higher in nutrient availability than native population sites. Supplemental metal halide lighting was used to extend day length to 14 h and photosynthetically active radiation (PAR) in the glasshouse averaged 16.5 mol m−2 d−1. The average daytime temperature was 27.3°C and the average daytime humidity was 54.9%. The average night-time temperature was 22.9°C.
Photosynthesis on a leaf area basis (Aarea) was measured on the most recently fully expanded leaf on 18–20 February, which were sunny days before flowering. One experimental block was measured each day using a portable photosynthesis system (Li-Cor 6400, Li-Cor Biosciences, Lincoln, NE, USA). Gas exchange cuvette conditions were set to 380 ppm CO2, 2000 μmol m−2 s−1 PAR, 30°C block temperature, and relative humidity slightly above ambient in the glasshouse (c. 50–55%). The morning following photosynthesis measurements, when leaves were maximally hydrated for the day, the gas exchange leaf was excised, weighed, and digitally scanned. The leaf area that had been inside the Li-Cor 6400 chamber and the area of the entire leaf were determined with ImageJ freeware (National Institutes of Health, Bethesda, MD, USA). Gas exchange leaves were then dried to constant weight at 60°C, weighed, and ball mill ground to estimate leaf nitrogen (Nmass; NA1500, Carlo Erba Strumentazione, Milan, Italy) and leaf carbon isotope ratio δ13C (Finnegan, continuous-flow mass spectrometer, Bremen, Germany). Leaf δ13C provides an integrated measure of leaf intercellular CO2 concentration (Ci) over the lifetime of the leaf. Integrated Ci is, in turn, a relative measure of integrated instantaneous WUE, provided leaf temperatures are similar (Farquhar et al., 1989; Ehleringer et al., 1992; Ehleringer, 1993). A higher (less negative) value of leaf δ13C reflects greater WUE. LMA was calculated as the dry mass of the gas exchange leaf divided by leaf area. Leaf photosynthetic rate on a mass basis (Amass) was calculated as Aarea divided by LMA. Leaf nitrogen on an area basis (Narea) was calculated as Nmass multiplied by LMA. LWC was calculated as the difference between fresh and dry leaf mass divided by leaf dry mass (Shipley et al., 2006). The most recently fully expanded leaf opposite the gas exchange leaf was tagged c. 1 month after transplant and tracked for LL, estimated as the time (d) between tagging and leaf color change to 25% of leaf area turned yellow (Brouillette & Donovan, 2011). Tagging of the leaf was done during the vegetative stage of plant growth when leaves were actively being produced so that leaves were of comparable age. The DFF was recorded when at least one ligule was fully extended, making the disk of the inflorescence visible. In addition, 38 other morphological and ecophysiological traits were measured, although they are not the focus of this manuscript. The additional trait list, methods and results are included in the supporting materials (Methods S1, Table S2, Fig. S1).
Source site characterization
Each population source site was characterized for climate and soil fertility. Site climate characteristics (mean annual temperature (MAT) and mean annual precipitation (MAP)) were determined using WorldClim, which interpolates a 50 yr average based on observations from 1950 to 2000 and has resolution to c. 1 km2 (Hijmans et al., 2005). Five soil cores (0–10 cm depth) collected at each site spanned the area in which seeds were collected. Soils were dried at 60°C and analyzed for soil C by Dumas combustion with a CHN analyzer (NA1500, Carlo Erba). For soil N, the CHN analyzer was subsequently modified to remove the CO2 produced via combustion and allow detection of low soil N amounts below the normal detection limit. Soil P and K were analyzed with an inductively coupled plasma optimal emission spectrometer (Thermo Jarrell-Ash Enviro 36, Franklin, MA, USA) after double acid extraction (Mehlich, 1953).
ANOVA, bivariate correlations and principal component analyses (PCAs)
Populations were compared for plant traits and soil fertility characteristics with ANOVA (SAS proc mixed; SAS Institute Inc., Cary, NC, USA). Across all populations, bivariate correlations among traits were determined with Pearson correlations (SAS proc corr). Although the trait data in the original global GLOPNET dataset describing the LES required log transformation in order to meet the assumptions of parametric statistics (Wright et al., 2004), log transformation was not necessary for our data. Five different PCAs (SAS proc princomp) were used to provide population values that summarize: source site climate (MAT, MAP); source site fertility characteristics that differed by population; source site climate and fertility combined (see the 'Results' section); plant traits for which QST > FST; and all nine plant traits regardless of the relationship of QST to FST. Only the first axis of each analysis was used as it captured the bulk of variation in each case. Correlations were then used to assess the relationships among the primary principal components summarizing source site characteristics and plant traits.
Simple sequence repeat (SSR) genotyping and analysis
Leaf samples from each plant were collected and placed in polypropylene tubes containing silica gel desiccant and stored at room temperature until use. DNA was extracted from one individual from each of the 97 maternal families using DNeasy Plant Mini kits (Qiagen). A 12-locus genotype was obtained for each sample using SSR markers located in intergenic regions of the nuclear genome from the Compositae Genome Database (CGP, http://cgpdb.ucdavis.edu). Markers were chosen to be largely unlinked in the parental species H. annuus (Tang et al., 2003) and polymorphic in a test panel of H. anomalus samples: (ORS229, ORS297, ORS511, ORS588, ORS618, ORS844, ORS896-A, ORS896-B, ORS1008-A, ORS1008-B, ORS1017, ORS1141). Fluorescently labeled primers (FAM, NED, HEX, TET) were used to amplify SSRs with a touchdown 58 protocol (Don et al., 1991). Products from the reaction were diluted 1 : 20 and analyzed using capillary gel electrophoresis (ABI 3730xl; Applied Biosystems, Valencia, CA, USA). The size of fragments was determined by comparison with a fluorescently labeled size standard (GS500 LIZ, Applied Biosystems) using Genescan software (Applied Biosystems). Observed and expected heterozygosities were calculated in FSTAT (Goudet, 1995).
Genetic variation in quantitative traits, when partitioned into between-population and within-population components, can be used to calculate population divergence (QST) in a manner analogous to FST calculated from neutral genetic markers:
where σb2 is the genetic variance among populations and σw2 is the genetic variance within populations. When traits are purely additive and neutral, QST = FST (Spitze, 1993). Divergence from this expectation can be used to test for adaptive response to selection.
Variance for all traits was partitioned into population, family nested within population, and block and error components using a completely random model with restricted maximum likelihood estimation (SAS 9.3 proc mixed). The variance component for populations was used as the estimate of among-population variance (σb2). The within-population genetic variance (σw2) was estimated as four times the family within-population variance component because half-siblings were used (Lynch & Walsh, 1998). Owing to high mortality in some populations, our data were unbalanced and thus unsuitable for standard parametric bootstrap methods (which use a χ2 distribution to simulate the distribution around variance components) for estimating confidence intervals around QST (O'Hara & Merila, 2005; Whitlock, 2008; Whitlock & Guillaume, 2009). Additionally, estimation of confidence intervals using Bayesian methods, which do not require balanced data (O'Hara & Merila, 2005), were highly dependent on initial conditions and thus not suitable. Thus, standard error for the mean QST value of each trait was estimated as the standard deviation of QST estimates from 1000 samples bootstrapped over all observations and used to construct confidence intervals using the standard formula (± 1.96 SE) (O'Hara & Merila, 2005; but see Leinonen et al., 2008, p. 11).
FST was calculated using GDA (Genetic Data Analysis software; Weir & Cockerham, 1984; Lewis & Zaykin, 2001), with bootstrapped confidence intervals estimated over 1000 iterations. There remain concerns that markers with high mutation rates, including SSRs, which were used in this study, may increase the rate of type I error when testing for divergent selection. Within-population heterozygosities are increased with highly polymorphic markers, which deflates FST (Hedrick, 1999) and makes it easier to achieve a statistically significant QST > FST. While alternative metrics that are not dependent on within-population heterozygosity have been proposed (e.g. D, Jost, 2008; and G′ST, Hedrick, 2005), only FST is appropriate for comparisons with QST (Edelaar & Bjorklund, 2011). Thus, we acknowledge that some of the traits identified in this study as being under divergent selection may be artifacts of the genetic markers used. However, several traits in this study had QST values much greater than FST, and for these it is likely that QST > FST would remain statistically significant regardless of marker choice. We must also point out that our plants were grown from field-collected seeds. Our analysis may overestimate population divergence in individual traits because variation in the maternal environments could have exaggerated differences between the populations. Additionally, because paternity is unknown, it is likely that some of the families contain full and half siblings. If this is the case, our estimate of σw2 will be inflated, artificially decreasing QST. The inclusion of some full siblings would affect QST similarly across all traits, but differences caused by maternal effects would vary from trait to trait.
Characterization of source site environment
Across all eight of the population sites, the MAP ranged from 161 to 292 mm, and the MAT ranged from 9.6 to 12.5°C. PCA analysis of MAP and MAT (n =8 populations) resulted in a composite variable designated as climate PC1 that captured 87.6% of the variation (loadings of 0.71 and −0.71, respectively). Among the population source sites, the more northern populations, for example (JER, WHS), had higher annual precipitation and lower temperatures (Table S1). The populations differ for precipitation predominately during the late fall and winter months when plants are overwintering as dormant seeds (Fig. S2).
Across all eight of the population sites, the desert sand dune soils had low soil fertility. Soil C and N were extremely low (0.591 ± 0.200% and 0.003 ± 0.001% by mass, respectively). Soil P was 12.3 ± 1.2 ppm and soil K was 16.4 ± 1.6 ppm. There were significant population differences for soil C, P and K (see Table S1 for population means; ANOVA, F >8.47, P <0.001, df 7,32 for all) but not for soil N (F =1.57, P =0.18, df 7,29). PCA analysis (n =8 populations) for the three soil fertility measures that differed by population (C, P and K) resulted in a composite variable designated as ‘soil PC1’ that captured 72.2% of the variation. Higher values of soil PC1 generally represented higher C, P and K, with loadings of 0.65, 0.47 and 0.59, respectively. The more northern populations generally had higher fertility.
When ‘climate PC1’ was plotted against soil PC1, there was a positive correlation (r2 = 0.82, P =0.002). Thus, the climate and soil data (MAP, MAT, soil C, soil P and soil K) were combined in an additional PCA analysis that resulted in a composite variable that captured 74.7% of the variation (loadings of 0.47, −0.47, 0.50, 0.34 and 0.43, respectively), designated as ‘environment PC1’ (Fig. 2, x-axis).
Genetic variation in SSR markers
FST was 0.18, demonstrating moderate population genetic differentiation for an outcrossing herbaceous annual. Total expected heterozygosity (He) across all individuals was 0.621. Populations were moderately diverse, with population-level He ranging from 0.300 (JER) to 0.655 (HAL) (Table S3).
Comparison of neutral genetic and quantitative trait divergence
Comparisons of neutral genetic variation (FST) to quantitative trait variation (QST) indicated that QST was greater than FST for Amass, Nmass, Narea, LWC and DFF, consistent with population differentiation for these traits being the result, in part, of a strong response to direct or indirect divergent selection (Fig. 3). For Aarea and LMA, QST was not significantly different from FST, providing no evidence for adaptive differentiation in response to selection. For leaf WUE estimated from δ13C, QST was less than FST, suggesting stabilizing selection. For LL, the QST could not be estimated because there were essentially no significant differences among families within populations or among populations. For the eight morphological, physiological and life-history traits in this study with estimable QST values, the mean was 0.473, with QST often greater than FST.
Quantitative trait patterns
For H. anomalus the glasshouse-grown plant traits occur at the resource-acquisitive end of the cross-species LES spectrum initially described for field-grown plants (Wright et al., 2004), with high Amass, Nmass and LMA (Fig. 4a). The species-level trait means (± SE) for H. anomalus were as follows: Amass, 897.0 ± 16.24 nmol g−1 s−1; Aarea, 35.60 ± 0.44 μmol m−2 s−1; Nmass, 5.95 ± 0.11%; Narea, 169.3 ± 1.8 mmol m−2; LMA, 40.01 ± 0.43 g m−2; LL, 1.09 ± 0.02 months; LWC, 11.30 ± 0.22 g g−1; δ13C, –31.62 ± 0.06 ‰; and DFF, 55.5 ± 1.5 d.
There were significant population differences for Amass, Aarea, Nmass, LMA, LWC, δ13C and DFF (ANOVA P <0.001, df 7,181–182 for all), but only a trend for Narea (P =0.07, df 7,182) and no difference for LL (P =0.62, df 7,182). Across populations, correlated trait evolution generally followed the main axis of the LES for Amass, Nmass and LMA (Fig. 4b, Table 1; population means are presented in Table S1). Amass was positively correlated with Aarea, Nmass and Narea, as expected, and tended to be negatively correlated with LMA (P =0.06). Nmass was additionally positively correlated with Narea, and negatively correlated with LMA. Contrary to expectation, LL and LWC were not correlated with any other LES traits. Also contrary to expectation, less negative leaf δ13C (i.e. higher WUE) was associated with higher Amass, Aarea, Nmass, Narea and LWC. Later flowering (higher DFF) was associated with lower Nmass and higher LMA but was not correlated with Amass, Aarea, Narea or WUE estimated from leaf δ13C.
Table 1. Trait correlation coefficients (r, bold indicates P < 0.05, df 6) among population means for Helianthus anomalus grown under common garden glasshouse conditions
Amass and Aarea, leaf photosynthetic rate on a mass and an area basis, respectively; Nmass and Narea, leaf nitrogen on a mass and an area basis, respectively; LMA, leaf mass per area; LL, leaf lifetime; LWC, leaf water content; δ13C, leaf carbon isotope ratio; DFF, days to first flower.
Principal component analysis (n =8 populations) for traits that have evidence of population differentiation in response to selection (i.e. QST > FST for Amass, Nmass, Narea, LWC and DFF) resulted in a composite variable designated as ‘plant PC1’ that captures 74.6% of the variation among traits. Higher values represented higher Amass, Nmass, Narea and LWC, and lower DFF, with loadings of 0.51, 0.49, 0.48, 0.36 and −0.37, respectively (Fig. 2, y-axis). PCA (n =8 populations) for all nine plant traits (instead of just those for which QST > FST) resulted in a composite variable that captures 61.1% of the variation among traits. Higher values represented higher Amass, Nmass, Narea and LWC, lower DFF, higher Aarea and leaf δ13C, and lower LMA and LL, with loadings of 0.42, 0.40, 0.40, 0.31, −0.26, 0.31, 0.39, −0.29 and −0.05, respectively.
Relationship between plant traits and source site environment
There was a significant correlation between plant PC1 and the source site characteristics assessed as either climate PC1 (n =8, r =−0.746, P =0.03), soil PC1 (n =8, r =−0.842, P =0.01), or the combined climate and soil variable, environment PC1 (n =8, r =−0.817, P =0.01, Fig. 2). The more resource-acquisitive LES and related traits were associated with drier (lower precipitation, higher temperatures) and lower fertility sites. If all nine plant traits are included in the plant PCA then the r and P values remain similar (r =−0.778, P =0.02; r =−0.847, P =0.01; r =−0.834, P =0.01, respectively).
We compared neutral and genetic variation in the annual H. anomalus to test for adaptive divergence in LES and related traits along a water and nutrient gradient. The FST estimate of 0.18, which was the neutral measure for comparison with QST, suggested a moderate degree of genetic differentiation among populations similar to other wild Helianthus spp. (Ellis et al., 2006; Ellis & Burke, 2007; Gevaert et al., 2013; Mandel et al., 2013). For individual traits, we found evidence for population differentiation consistent with response to diversifying selection (i.e. QST > FST) for Nmass, Narea, DFF, Amass and LWC. Our support for adaptive differentiation of leaf Nmass and Narea is consistent with that found for Quercus suber along a climate gradient (Nmass, Ramirez-Valiente et al., 2009), and among populations and subpopulations for Populus balsamifera (Narea; Keller et al., 2011), but not for Helianthus maximiliani along a latitudinal gradient (Nmass; Kawakami et al., 2011). Our support for adaptive differentiation of DFF is consistent with reports for some species (H. maximiliani, Kawakami et al., 2011; Arabidopsis thaliana using bolting as a proxy, Le Corre, 2005), but not for others (Lythrum salicaria, Chun et al., 2009; Hordeum spontaneum, Hubner et al., 2013). To our knowledge, Amass and LWC have not been previously investigated in QST–FST analyses. For LMA, Aarea and leaf δ13C, our lack of support for adaptive differentiation is also consistent with some but not all reports for other species (Ramirez-Valiente et al., 2009; Kawakami et al., 2011; Keller et al., 2011; Frei et al., 2012). Thus, for each trait examined, there is substantial variation among studies as to whether QST–FST analyses support adaptive differentiation. This suggests either that individual traits are not as important as hypothesized for adaptation to the variety of gradients and landscape heterogeneity examined, or that genetic constraits may prevent them from responding to selection for some species and populations (Donovan et al., 2011). However, multiple traits need to be considered to assess whether adaptive trait divergence contributes more to a resource-acquisitive or a resource-conservative strategy in response to greater resource limitation.
For trait combinations, H. anomalus populations with high Amass, Nmass and low LMA clustered with other annuals and short-lived herbaceous perennials at the resource-acquisitive end of the cross-species LES based on field data (Wright et al., 2004; Fig. 4a). The H. anomalus trait population means covered a relatively small portion of the entire LES range, as expected for closely related populations in a common garden glasshouse study that minimized environmentally induced variation. Despite that small range, correlated trait evolution represented by population differentiation for Amass, Nmass and LMA was consistent with the primary axis of the LES (Fig. 4b, Table 1). Higher Nmass was also associated with the earlier flowering time (lower DFF), indicating a resource-acquisitive earlier flowering strategy. The lack of correlation of LWC with other traits (only nonsignificant positive trends with Amass, Aarea and Narea) does not support the hypothesis that LWC is an important biophysical driver of LES traits in this system (Shipley et al., 2006).
The correlation between H. anomalus traits with a signature of adaptive divergence (plant PC1 for traits with QST > FST) and source site characteristics (environment PC1) supports our hypothesis that climate and/or soil fertility are selective agents acting on plant traits and driving population differentiation (Fig. 2). The more southerly sites have plants with inherently higher Nmass, Narea, DFF and Amass. These sites have lower fertility and a hotter drier climate, with less soil moisture to sustain the growing season as a result of less winter precipitation. Thus, for H. anomalus, the hotter, drier environment of the more southerly sites selectively favored correlated evolution of traits contributing to a resource-acquisitive and earlier reproductive strategy to escape soil moisture depletion predictably occurring late in the growing season, despite lower fertility. This is consistent with the growing literature documenting earlier onset of reproduction as an evolutionary response of annual species to many stresses, including drought, although earlier onset of reproduction is not always accompanied by the greater pre-reproductive growth rates expected for a ‘live fast, die young’ strategy (Arendt, 1997; Geber & Dawson, 1997; Stanton et al., 2000; McKay et al., 2003; Griffith & Watson, 2005; Heschel & Reginos, 2005; Franks et al., 2007; Franks & Weis, 2008; Kigel et al., 2011; Ivey & Carr, 2012).
Although multitrait LES studies on annuals are generally lacking, we can compare the adaptive differentiation of multiple traits for this desert annual to QST–FST studies for several woody species. For the evergreen Quercus suber (Ramirez-Valiente et al., 2009), populations from cooler, drier sites had leaves with lower N and higher LMA, suggesting that water limitation selected for a more resource-conservative strategy. For the deciduous Populus balsamifera, populations from more northerly sites with a shorter, drier growing season had leaves with higher Aarea and Narea, but also a higher LMA, suggesting that water limitation selected for a more resource-acquisitive strategy except for LMA (Soolanayakanahally et al., 2009; Keller et al., 2011). Thus, there is growing evidence at the microevolutionary level of adaptive trait differentiation of multiple traits consistent with the primary axis of the LES in response to diversifying selection, with the direction of the response (more resource-acquisitive or resource-conservative) probably influenced by life history and associated differences in LL, for example, annuals and deciduous perennials vs evergreen perennials. This is consistent with observed macroevolutionary patterns (Wright et al., 2005). However, more studies that provide evidence of adaptive differentiation along environmental gradients will be needed before we can say that microevolutionary studies support this generalization. Additionally, other approaches will be needed to distinguish whether correlated trait evolution is a result of direct selection on individual traits, correlational selection, and/or indirect selection mediated through pleiotropy or linkage that may be reflected as genetic correlations (Chen & Lubberstedt, 2010; Donovan et al., 2011).
In H. anomalus, the difference between QST and FST was greatest for Nmass, Narea and DFF, indicating that these traits have the strongest support for adaptive differentiation. These results suggest several interpretations. First, direct selection may be stronger on leaf Nmass, Narea and DFF than on Amass, LWC and LMA, but that would require the assumption that there was similar heritable variation for all of the traits and that there were no constraints as a result of indirect selection mediated through genetic correlations. Alternatively, the stronger support for adaptive differentiation for these traits might be the result of equally strong or diversifying selection on other traits that did not result in similar population differentiation because of genetic constraints in the form of limited heritable variation or genetic correlations. Quantitative genetic approaches and phenotypic selection analyses would be needed to test these alternative hypotheses. Phenotypic selection analyses carried out in the H. anomalus JER population have demonstrated that leaf Nmass was under direct selection within the context of leaf traits measured in that study (Nmass, δ13C, area, succulence) (Donovan et al., 2009). However, that analysis did not include other LES traits, so there was no power to determine whether Nmass was under direct or indirect selection within the context of other LES traits.
The results for WUE estimated from leaf δ13C were surprising for H. anomalus. First, QST was less than FST, suggesting stabilizing or uniform selection, although this pattern (QST < FST) is not frequently reported in the literature and is difficult to interpret (Scheepens et al., 2010; Frei et al., 2012; Lamy et al., 2012). The analysis of intrinsic WUE, estimated as Ci (approx. Aarea/stomatal conductance), yielded the same result (Fig. S1). Secondly, higher leaf-level WUE (less negative δ13C) was associated with higher Amass, Aarea, Nmass and Narea, which is contrary to the expectation of an association of lower WUE with faster growth. Thirdly, leaf δ13C was not correlated with DFF. Thus, the resource-acquisitive and earlier-flowering strategy associated with drier lower fertility sites did not have lower WUE, as might be expected from theory and empirical studies (Cohen, 1970; Geber & Dawson, 1997; McKay et al., 2003). The trait patterns suggest that despite the lower fertility at the drier sites, these populations achieve a higher leaf N (on mass and area bases) which permits a higher photosynthetic capacity and, thus, both higher photosynthesis and leaf-level WUE.
The population differences for LMA, coupled with the correlation of LMA with Nmass, seem to be at odds with the lack of evidence for adaptive differentiation for LMA. Mathematically, we can consider that FST and QST represent the proportion of total variance attributable to population differences, such that the high within-population variance for LMA decreased QST (Table 2). Taken alone, the QST–FST results for LMA suggest that population differentiation in this trait is indistinguishable from divergence resulting from neutral, rather than adaptive evolutionary processes. However, the correlations observed between LMA and two of the traits with support for adaptive differentiation (DFF and Nmass) suggest that LMA values among and within populations are not random, as would be predicted by genetic drift. An alternative explanation may be that weak direct or indirect selection drove population differentiation for LMA, but the response to selection was too weak to be detected in this study. Studies with low numbers (< 20) of populations have generally had low power to detect selection (O'Hara & Merila, 2005; Goudet & Buchi, 2006; Whitlock, 2008). Robust estimates of heritabilities and the structure of genetic covariance matrices would further our understanding of the role of selection and constraints in the correlated response to selection for LES and related traits (Chapuis et al., 2008; Chenoweth & Blows, 2008; Martin et al., 2008).
Table 2. Components of phenotypic variance of traits within and among eight populations of Helianthus anomalus grown under common garden glasshouse conditions, with total phenotypic variance partitioned using restricted maximum likelihood into variance among populations (Vpop), within-population genetic variance (Vfam; four times the variance among families within populations) and residual variance (Vres)
Amass and Aarea, leaf photosynthetic rate on a mass and an area basis, respectively; Nmass and Narea, leaf nitrogen on a mass and an area basis, respectively; LMA, leaf mass per area; LL, leaf lifetime; LWC, leaf water content; δ13C, leaf carbon isotope ratio; DFF, days to first flower.
We were unable to make an FST–QST comparison for LL in this study, because both within- and among-population variance components were estimated as zero. There may be several explanations for the lack of detectable variance for LL. First, LL is a difficult trait to measure and error may have obscured genetically based variation, although we have used this same methodology to determine that H. anomalus LL is longer than that of its ancestral parental species (Brouillette & Donovan, 2011). DFF might serve as a proxy for LL in some annuals (Luquez et al., 2006; Vasseur et al., 2012), but we found no correlation between DFF and LL. Secondly, there may be minimal heritable genetic variation in H. anomalus. Longer LL could be favored by selection because it increases nitrogen-use efficiency by increasing the length of time nitrogen remains in the plant (Aerts & Chapin, 2000; Brouillette & Donovan, 2011), and traits under extremely strong selection are expected to show low degrees of genetic variance (Geber & Griffen, 2003 and references therein). Low soil fertility might be a strong selective force reducing heritable variation for LL in the H. anomalus populations. However, population differentiation and heritable variation have been documented for LL in A. thaliana (Luquez et al., 2006). Additionally, sexual selection for flower size has been reported to result in a correlated trait response of higher LMA and shorter LL (Delph et al., 2005). Thus, LL deserves further investigation as a putative adaptive trait.
In summary, we found that combining common garden correlational approaches with QST–FST approaches supports the adaptive differentiation of some LES and related traits. A more resource-acquisitive and yet more water-use efficient strategy associated with a higher Nmass and Narea is found in hotter, drier sites, despite lower fertility. A comparison of our results with the literature suggests that while adaptive divergence of traits along stress gradients may generally follow the primary axis of the LES, variation in trait combinations favored by selection is likely to occur as a result of variation in life-history characteristics of the species, the dominance of particular abiotic and biotic components in the stress gradient, and potential genetic constraints within populations.
This research was funded by National Science Foundation grants 0614739 and 1122842 to L.A.D. and a Garden Club of America Award in Desert Studies to L.C.B. We thank C. Darragh for help with the glasshouse work and the laboratory of S.J. Knapp for genotyping support.