SEARCH

SEARCH BY CITATION

Keywords:

  • adaptive differentiation;
  • Arabidopsis thaliana ;
  • climate gradient;
  • clinal variation;
  • elevation;
  • heat and drought;
  • local adaptation;
  • natural variation

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • The extent to which a species' environmental range reflects adaptive differentiation remains an open question. Environmental gradients can lead to adaptive divergence when differences in stressors among sites along the gradient place conflicting demands on the balance of stress responses. The extent to which this is accomplished through stress tolerance vs stress avoidance is also an open question.
  • We present results from a controlled environment study of 48 lineages of Arabidopsis thaliana collected along a gradient in northeastern Spain across which temperatures increase and precipitation decreases with decreasing elevation. We tested the extent to which clinal adaptive divergence in heat and drought is explained through tolerance and avoidance traits by subjecting plants to a dynamic growth chamber cycle of increasing heat and drought stress analogous to low elevation spring in northeastern Spain.
  • Lineages collected at low elevation were the most fit and fitness scaled with elevation of origin. Higher fitness was associated with earlier bolting, greater early allocation to increased numbers of inflorescences, reduction in rosette leaf photosynthesis and earlier fruit ripening.
  • We propose that this is a syndrome of avoidance through early flowering accompanied by restructuring of the organism that adapts A. thaliana to low-elevation Mediterranean climates.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

More than half a century of research on adaptation in plants has shown that local adaptations are relatively common (Clausen et al., 1940; Linhart & Grant, 1996; Leimu & Fischer, 2008). Adaptive differentiation occurs when selection varies and requires conflicting optimization of plant form and function at different sites (Endler, 1986; Kawecki & Ebert, 2004). Especially givenaccelerating anthropogenic climate change (Bradshaw & Holzapfel, 2006; Parmesan, 2006), it is necessary to improve our understanding of the functional bases of local adaptations (Feder, 2007; Mitchell-Olds et al., 2008; Anderson et al., 2011, 2012). One classic and powerful approach to understanding historical and future adaptation is the study of trait divergence along environmental gradients (Endler, 1986) and this approach has been used successfully in studies of clinal variation in plant traits, including leaf physiology and morphology (Martin et al., 2007), above- and below-ground carbon–nutrient balance (Freschet et al., 2010), phenology and architecture (Petrů et al., 2006). In this study we examine how a suite of life-history, physiological and allocation traits provide an integrated adaptive response to increasing heat and drought during the reproductive season in an annual plant.

Range-wide studies of the model plant Arabidopsis thaliana (Brassicaceae) provide an excellent system for clinal studies of geographically varying adaptation and their genetic bases in annual plants (Mitchell-Olds, 2001; Tonsor et al., 2005; Mitchell-Olds & Schmitt, 2006; Wilczek et al., 2009; Fournier-Level et al., 2011, 2013; Ågren & Schemske, 2012; Grillo et al., 2013). A broad suite of differentiated traits is emerging in studies of adaptation across A. thaliana's latitudinal range. These include variable freezing tolerance (Hannah et al., 2006; Zhen & Ungerer, 2008), vernalization requirements (Hopkins et al., 2008), responses to light quality (Stenøien et al., 2002), heat shock protein expression (Tonsor et al., 2008), growth rate (Li et al., 1998), seed dormancy and season of germination (Kronholm et al., 2012; Montesinos-Navarro et al., 2012) and age at onset of flowering (Stinchcombe et al., 2004; Wilczek et al., 2009).

Populations in northeastern (NE) Spain occur across an altitudinal range from near sea level at the Mediterranean coast to near treeline (c. 2200 m above sea level (asl)) in the Pyrenees mountain range. Along this gradient, low-elevation sites are hotter and drier overall compared with high elevations. Low elevations experience temperatures above freezing for most of the autumn and winter, but a relatively short spring reproductive period with rapid warming and drying. By contrast, high elevations experience periodic below-freezing temperatures and snow cover during the winter, but have a relatively prolonged, cooler and wetter spring reproductive period (Montesinos et al., 2009; Montesinos-Navarro et al., 2011).

Importantly, genetic analyses indicate that the populations in this region are genetically distinct from surrounding regions and are probably descended from a common ancestor (Picó et al., 2008). There are two possible evolutionary genetic causes that could result in clinal trait variation. The first is historic colonization of high- and low-elevation sites by genetically distinct ancestors followed by spread from both ends towards mid-elevations and a subsequent isolation by distance-driven clinal pattern. As we do not detect isolation-by-distance among these populations and gene flow is very low (Montesinos et al., 2009), trait–environment covariation must therefore result from a response to a gradient in natural selection (Montesinos-Navarro et al., 2011).

Shifts in the timing of life-history transitions appear to be an important mechanism of adaptation across this climate gradient (Montesinos et al., 2009; Montesinos-Navarro et al., 2011). Temporal duration of seed primary dormancy, sensitivity of seeds to induction of secondary dormancy by high temperatures (Montesinos-Navarro et al., 2012), probability of germinating in autumn vs spring, and age at bolting (Montesinos-Navarro et al., 2011) vary with climate of origin in our study populations. Under the constant cool, moist, mid-elevation conditions used in Montesinos-Navarro et al. (2011), late-bolting high- and mid-elevation plants exhibited the highest seed production.

High temperatures and low water availability are important stresses for virtually all plants (Parmesan, 2006; Wahid et al., 2007) and water availability and temperature have been proposed as determinants of the geographic range limits of A. thaliana (Hoffmann, 2002, 2005). Therefore, in this study we test for adaptive divergence associated with variation in traits hypothesized to play a role in adaptation to hot, dry conditions. The climate gradient from the Mediterranean coast to near treeline in the Pyrenees compresses much of A. thaliana's range-wide climate gradient into a logistically manageable distance (see the 'Results' section). In particular, the coastal conditions are near the southern environmental limit for A. thaliana. The sites of the coastal populations are especially strongly differentiated from the inland, higher altitude by a rapidly warming and drying spring (Montesinos et al., 2009). We therefore focus particularly on clinal variation expressed under conditions of increasing temperature and decreasing water availability during the reproductive period. We can assess functional significance under our experimental conditions by quantifying the link between fitness and clinally varying traits. Additionally, we can use these relationships to generate hypotheses about fitness consequences of functional variation in the field.

Plants facing drought and increasing temperatures during the reproductive period potentially experience two forms of selection: for stress avoidance and/or for stress tolerance. Heat and drought stress during the spring reproductive season might accelerate A. thaliana's developmental program leading to completion of the life cycle before conditions become unsuitable, thus avoiding stress. Alternatively, A. thaliana populations under reproductive season stress might adapt their physiology, allocation strategy and morphology so as to complete the life cycle in spite of stress, thus tolerating it (Grime, 1977). In this study, we investigate a suite of plant characters that are hypothesized to represent aspects of either avoidance or tolerance of spring heat and drought.

Clinal variation in photosynthetic parameters has yet to be investigated in A. thaliana. Variation in photosynthetic parameters might be expected to be associated with adaptation across the NE Spanish climate gradient for several reasons. First, in response to heat, plants may alleviate heat stress by opening stomata and increasing transpiration (Farquhar & Sharkey, 1982). Alternatively, stomata may be closed for increased water-use efficiency (WUE) (Kalisz & Teeri, 1986; Chaves et al., 2002). Clinal variation in WUE might be expected to coincide with the previously observed cline in bolting time along the Spanish climate gradient (Montesinos-Navarro et al., 2011), as age at bolting has been shown to correlate with WUE in A. thaliana (McKay et al., 2003). Finally, clinal variation in photosynthetic rates has previously been observed both within (Arntz & Delph, 2001) and among (Wright et al., 2004) species.

Montesinos-Navarro et al. (2011) observed clinal variation in allocation in which later-bolting high-elevation plants produced larger rosettes but smaller root systems than did earlier-bolting low-elevation plants. In the face of spring heating and drying, two opposing allocation patterns could hypothetically be beneficial. Plants fitting Grime's (1977) definition of ruderals would shift resources from vegetative to reproductive structures in the face of stress. By contrast, stress-tolerant plants might show increased allocation to vegetative structures, potentially allowing continued survival in the spring (Grime, 1977). At the leaf level, changes in dry mass allocation per unit area (specific leaf area; SLA) are known to vary clinally in A. thaliana (Li et al., 1998) and other species, with leaves from higher elevation or colder climates being thicker. By contrast, hot, dry spring conditions might favor plants with low-investment leaves, as these have frequently been associated with shorter life spans and thus, possibly, with stress avoidance (Wright et al., 2004; Levey & Wingler, 2005).

We present results from a laboratory-controlled environment study of multiple lineages collected from across an elevation gradient in NE Spain. We subjected the experimental population to warming and drying during the reproductive period, placing plants under conditions similar to those in the field at the sites of origin of our low-elevation populations. With this study, we accomplish the following aims: we test for a correspondence between the elevation gradient across which the populations were collected and a major climate gradient; we quantify the extent of adaptive divergence at the genotype level by testing in a common environment for correlations between trait values and elevation- and climate-of-origin; we quantify the fitness effects under our experimental conditions of a suite of traits including leaf-level gas exchange, SLA, photochemical quantum efficiency of photosystem II (PSII), dry mass production of roots, rosettes and inflorescences, and the timing of bolting and fruit ripening under conditions of increasing spring heat and drought.

Materials and Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Collections

Seed collected from 16 sites along an elevation gradient in NE Spain (Fig. 1) was propagated by single seed descent for at least two generations in a laboratory-controlled environment to eliminate maternal effects and increase seed stock. Each collection site corresponds with a study population. Three genotypes per population were randomly selected (48 total) for inclusion in the present study.

image

Figure 1. Map of the 16 Arabidopsis thaliana collection sites in northeastern Spain used in this study. Colors indicate elevation going from low to high as follows: green to brown to white (see Table S1 for further information).

Download figure to PowerPoint

Planting

Plants were grown in Ray Leach SC10 164 ml Supercell Conetainers (hereafter pots; www.stuewe.com/products/rayleach.html) and arranged at a density of 24 pots per Ray Leach RL98 Tray (hereafter racks). Racks were placed in matching fiberglass bins. Pots were filled with Turface MVP fritted clay (http://www.turface.com). Nutricote (1.5 ml) was placed 10 cm below the surface level. Nutricote-encapsulated fertilizer releases equal daily quantities of nutrients for 100 d (NPK 13-13-13, Type 100; Arysta Life Science, New York, NY, USA). A 1-cm-wide, 2-cm-deep plug of Sunshine germination mix (http://www.sungro.com) was inserted at the surface of each pot. Before planting, seeds were surface-sterilized via exposure to chlorine gas for 3 h to avoid disease transmission. We planted five seeds per pot at the center of the germination mix plug and later thinned seedlings to one per pot.

Eight pots were planted per genotype. To allow sufficient time for measurement, yet measure all plants at the same age, one replicate of each genotype was planted each day for 8 d into 16 bins (384 pots in total). Planted bins were placed in the dark at 4°C and 100% relative humidity for 4 d to induce germination competency. After cold treatment, bins were transferred into two Conviron PGW36 growth chambers – eight bins each – for the remainder of the experiment.

Growth conditions

We created a dynamic growth chamber cycle in which temperature, day length and water availability changed over time. Our goal here was to provide key seasonal cues through which plants sense and respond to the environment, thereby approximating key aspects of an autumn-germinated, winter annual life history for the experimental plants. Most importantly for this study, during the simulated spring, we imposed a regime of increasing heat and drought stress. By doing this, we sought to determine whether the traits expressed by low- vs high-elevation plants were adaptive under reproductive phase heat and drought stress. Field conditions and therefore selective regimes undoubtedly differ from chamber conditions. Here, we ask whether plants collected along a gradient of seasonal patterns in heat and drought display fitness differences under our growth chamber conditions that are explained by genetically based patterns of trait variation. We use the combined associations among trait value, home climate and fitness to interpret the functional and adaptive significance of our results.

Our growth chamber cycle began with daily temperature ramping evenly from 14°C at lights on to 22°C at lights off and back to the minimum overnight. Daily maximum and minimum temperatures decreased by 2°C every 8 d. On day 25, temperatures fell to a constant 5°C and were constant until day 45, when daily maximum temperature increased to 7°C, and then to 9 and 11°C on days 49 and 53, respectively. On day 57, daily temperatures increased to a maximum of 26°C and a minimum of 18°C. From day 57, for the remainder of the experiment (to day 125), maximum daily temperatures increased by 1°C and minimums by 0.5°C every 4 d to an ultimate daily maximum of 42°C and minimum of 26°C (Supporting Information, Fig. S1a). This produced a gradual increase in the daily temperature range similar to the progression of spring into summer in the low-elevation population sites (Fig. S2).

Initial day length was 12 h, decreasing by 0.25 h every 4 d, reaching a minimum of 10 h on day 33, simulating day length at the winter solstice in NE Spain. From the solstice onward, day length increased at the same rate (0.25 h every 4 d), reaching a maximum of 15 h on day 113 (Fig. S1b).

Water was supplied via an ebb-and-flood system (Earley et al., 2009). Two modes of water control were imposed: water table height and watering frequency. A standpipe in one drain of each bin controlled the water table. Bins were filled to standpipe height and remained filled for 45 min until a drain solenoid opened. The standpipe height was 17.8 cm (7 inches), and watering occurred twice daily until the beginning of simulated spring. In spring, water availability gradually declined to zero by reducing standpipe height to a minimum of 6.35 cm (2.5 inches) and reducing watering frequency to once daily, then every other day, until water supply was permanently discontinued. By weighing empty pots before and after watering at various standpipe heights, we express water availability in g per watering (Fig. S1c).

Light intensity was 150 μM photosynthetically active radiation (PAR) m−2 s−1 until day 45 when light increased by increments of 50 μM PAR m−2 s−1 every 4 d, reaching a maximum of 350 μM PAR m−2 s−1 on day 57.

Tray positions were rotated within each growth chamber every 4 d until day 85, when plants reached an advanced stage of flowering, making rotations impractical.

Trait measures

To measure differences in developmental timing, we recorded dates of germination, bolting and first-ripened fruit. We defined germination date as the day on which cotyledons were first visible, bolting as the date on which the first signs of primordial flower buds were visible, and ripening as the date on which we observed at least one fully yellow or brown fruit. From these life-history transition dates, we derived two life-history traits: the number of days between germination and bolting (time until bolting) and the number of days between bolting and first ripened fruit (bolting until ripening).

We measured leaf traits on day 85 when all but one of the plants was flowering. At midday, two racks d–1 were transferred to a third Conviron PGW36 growth chamber set to 350 μM PAR m−2 s−1 and 32°C, the daily maximum temperature. We conducted gas exchange measurements to assess the capacity of the plants to photosynthesize under heat stress. The plants were acclimated for 30 min before measurements. Measurements were conducted over 8 d consecutively, giving one replicate per genotype d–1. Thus measurements on each plant were conducted at the same age.

For each plant, we selected a recently expanded, fully green rosette leaf. One leaf was placed in the cuvette of an LI-6400 infrared gas analyzer (IRGA; Li-Cor Biosciences Inc., Lincoln, NE, USA) and allowed to equilibrate for 5 min while still attached to the plant. We took four measures of net carbon assimilation (μM CO2 m−2 s−1) and transpiration (mM H2O m−2 s−1) over 1 min. Measurements were averaged for a single record of instantaneous CO2 and H2O exchange per leaf. WUE was calculated as the ratio of net carbon assimilation to transpiration.

We measured maximum PSII quantum efficiency using a PAM 2000 fluorometer (www.walz.com). We chose a second fully expanded rosette leaf per plant and dark-adapted it by placing it in a light-blocking leaf clip for 5 min. Immediately following dark adaptation, the baseline fluorescence (Fo) was measured. We then applied a saturating pulse of white light to determine the maximal fluorescence (Fm). The quantum efficiency of open PSII was calculated as the ratio of variable fluorescence (Fv = Fm − Fo) to Fm (Baker, 2008).

After gas exchange measurement, leaves were excised, flattened under a pane of glass and photographed with an area standard. We measured leaf area using NIH ImageJ 64-bit version 1.44j (http://rsbweb.nih.gov/ij/). Finally, we dried the leaves at 70°C, and weighed them. SLA was calculated as the ratio of fresh leaf area to dry mass.

When all plants had fully senesced (day 125) we partitioned the plants into rosettes, inflorescences and roots and dried them at 70°C for at least 7 d. As a fitness measure, we used the summed fruit length, estimated as follows: inflorescence branches were laid out on a table and the length of only the reproductive (fruit-bearing) portions of each branch were measured with a PlanWheel XL™ (Scalex Corporation, Encinitas, CA, USA). Next we estimated the density of fruits by counting the number of fruits along 10 cm of the primary inflorescence branch. Finally, we measured the length of five fruits to give an average fruit length. Fitness, or summed fruit length, is equal to the reproductive branch length times the fruit density (i.e. fruit number) times the average fruit length. Fruit length is tightly correlated with the number of seeds per fruit (Alonso-Blanco et al., 1999) and fruit number strongly predicts total seed number by itself (Mauricio & Rausher, 1997), so summed fruit length is a good proxy for the total number of seeds. We also counted the number of basal inflorescence branches.

Climate data

We obtained temperature and precipitation data at each of the 16 collection sites from the BIOCLIM dataset described by Hijmans et al. (2005). This data set contains 19 biologically relevant climatic indices (Table S1) derived from monthly precipitation and temperature data for the period 1950–2000 (available at http://www.worldclim.org) and has a resolution of c. 1 km2 per grid cell.

Statistical analysis

We used principal components analysis (PCA) to produce orthogonal axes or principal components (PCs) that explain multivariate climate variance. We conducted the PCA on the correlation matrix of BIOCLIM values and the elevation above sea level at each of the 16 collection sites to produce indices that describe the major climate gradient(s) across which our populations are arrayed. We implemented randomization tests (Peres-Neto et al., 2003) to determine the number of meaningful climate axes (PCs). The distribution of PCs obtainable under the null hypothesis of climate variable independence was constructed by independently permuting the order of each BIOCLIM variable, then calculating PCs on the permuted data, repeated 5000 times. By comparing the actual nth PC to the distribution of nth PCs from the 5000 permutations, we obtained the probability of the actual nth PC value under the null hypothesis (SAS code available at http://www.tonsorlab.pitt.edu). By including elevation in the PCA, we describe the overall multivariate relationship between climate and elevation and test the extent to which elevation can be treated as a proxy for climate.

Before analyses, we first removed variance from each trait accounted for by aspects of the experimental design, that is, rack, measure day, and chamber effects. We also tested for an interaction between chamber and measure day. We used SAS PROC GLIMMIX (version 9.3; SAS Institute Inc., Cary, NC, USA) with rack as a random effect nested in measure day and the other factors as fixed effects. We added the grand mean for each trait to the residuals from this analysis and performed all subsequent analyses on these adjusted trait values.

One major aim of this study was to test for an adaptive cline in the multivariate phenotype associated with climate and elevation. Therefore, we conducted PCA on the correlation matrix of the population means for 12 traits, thereby reducing the dimensionality of trait space to a few orthogonal axes. As in the PCA of climate space, we used randomization (Peres-Neto et al., 2003) to determine the number of meaningful axes in the trait matrix. We then tested whether among-population variation in climate predicted among-population variation in phenotype by regressing population-mean trait PC scores on population-mean climate PC scores.

Principal components analysis produces ranked orthogonal vectors that reflect weighted combinations of variables based on the their covariances. While this is a powerful way to account for covariation and reduce the effective number of dimensions, the intuitive/biological interpretation of the regression of a trait PC onto a climate PC is difficult. Because of this, we also conducted univariate regressions of each trait on climate PC scores for visualization purposes. We also regressed population means for each trait on elevation and compared the predictive power of elevation with that of climate PCs. We used the SAS PROC REG (version 9.3; SAS Institute, 2011) for the trait PC–climate PC, univariate trait–climate and trait–elevation analyses.

Finally, we tested hypotheses about natural selection under simulated low-elevation spring heat and drought using a standard multiple-regression approach to quantifying phenotypic selection (Lande & Arnold, 1983). We transformed all variables (including fitness) to the same scale with a mean of 0 and a standard deviation (SD) of 1, allowing direct comparison of the strength and direction of selection among traits. Selection gradients therefore indicate the number of SDs increase in fitness per SD change in trait grand mean. We tested for both linear and quadratic selection by regressing summed fruit length on both the traits and their squared values in the multiple-regression model. To satisfy the regression assumptions, we used log-transformed summed fruit length as the dependent variable in the estimation of P-values and r-squares. However, we present parameter estimates based on untransformed summed fruit length to facilitate biological understanding. We note that statistical tests for linear selection coefficients may correspond to more complex functions on the nontransformed scale. We have only presented the linear effects in the graphics of untransformed data.

Although multiple regressions provide a view of the relationship between a trait and fitness independent of other correlated traits, correlations among traits remain an important issue in the interpretation of apparent selection. This is in part because multiple regression estimates of selection gradients can suffer from multicollinearity wherein highly correlated traits act to some extent redundantly, inflating the variance explained but causing inaccuracy in our estimates of the effects for those traits (Mitchell-Olds & Shaw, 1987). To aid in our interpretation of selection on and relationships among the traits measured, we used Akaike's information criterion (AIC) to determine the ‘best’ model by rewarding added explanatory power but penalizing the inclusion of additional terms. This provided the simplest model for fitness with the least collinearity and thus, presumably, the best estimates of selection (Shaw & Geyer, 2010). We used the SAS PROC REG (version 9.3; SAS Institute, 2011) for all selection analyses.

To aid interpretation of selection analyses, we also calculate and present the phenotypic Pearson product–moment correlations. We employed a sequential Bonferonni correction to all P-values in the correlation matrix to compensate for the risk of false positives with multiple testing. We used SAS PROC CORR (version 9.3; SAS Institute, 2011) for all correlation analyses.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

All pots contained germinated seeds. A small number of deaths or missing measurements resulted in a multivariate dataset (i.e. all observations with no missing values) of = 366 of the planned = 384 or 95% of the goal.

We detected significant chamber, measure day and chamber × measure day effects for many of the traits analyzed. All results reported in the following sections were obtained with the residuals from this model plus the trait grand mean.

PCA of the climate gradient

Randomization testing of the climate PCA eigenvalues detected two significant PCs (Fig. S3), explaining 75 and 17% of the climate variation among our 16 collection sites (Table S2, Fig. 2). For climate PC1, eigenvector coefficients for all precipitation variables with the exception of precipitation seasonality (coefficient of variation in precipitation) were positive and varied from 0.21 to 0.25. Eigenvector coefficients for all variables that describe temperature variability (e.g. annual temperature range) were positive but were smaller than the precipitation variables, ranging from 0.05 to 0.14. However, eigenvectors of variables that describe temperature during a particular time period (e.g. maximum temperature of the warmest month) were negative, ranging from −0.20 to −0.26 (Table S2, Fig. 2). Variables describing temperature variability (BIO2, BIO3, BIO4 and BIO7) loaded most strongly on climate PC2. Elevation loaded positively (0.24) onto climate PC1 but had an eigenvector of essentially zero on climate PC2 (Table S2, Fig. 2). Climate PC1 and PC2 scores for each population are reported in Table S3.

image

Figure 2. Eigenvector plot of the loadings of 20 climate variables onto the first and second principal components (PCs) of climate space that describe conditions in 16 northeastern NE Spanish Arabidopsis thaliana populations. Each arrow represents a vector of loadings. The direction of each arrow represents the relationship of a variable to climate PC1 and PC2 and the length of the vector represents the strength of that relationship. Clusters of similar variables (e.g. precipitation) are bracketed and labeled to help summarize results.

Download figure to PowerPoint

PCA of trait space

There was a single, significant PC (trait PC1) in population mean trait space, explaining 71% of the variation (Fig. S4). For trait PC1, quantum efficiency, rosette mass, root mass, age at bolting, time from bolting to first ripe fruit, and photosynthetic and transpiration rates had positive loadings ranging from 0.15 to 0.34. Summed fruit length, WUE, inflorescence mass, basal branch number and SLA had negative loadings ranging from −0.26 to −0.33 on trait PC1 (Table S4, Fig. 3). Table S3 contains trait PC1 scores for each population.

image

Figure 3. Eigenvector plot of the loadings of 12 traits onto the first and second principal components (PCs) of trait space for northeastern Spanish Arabidopsis thaliana. Each arrow represents a vector of loadings. The direction of each arrow represents the relationship of a variable to trait PC1 and PC2 and the length of the vector represents the strength of that relationship. SLA, specific leaf area; WUE, water-use efficiency.

Download figure to PowerPoint

Evidence for local adaptation, trait–elevation and trait–climate associations

Multiple regression of trait PC1 score on climate PC1 and PC2 scores explained 36% of the variance in PC1 (= 16, model P-value = 0.05). Climate PC1 showed a significant positive relationship with trait PC1, while climate PC2 was not significant (Fig. 4). The quadratic term was not significant so it was dropped from the model. Finally, we observe that the population BOS (Bossost) is an outlier in terms of its mean phenotypes relative to its elevation. For example, climate PC1 explains 56% of the variance in trait PC1 when BOS is excluded (not shown).

image

Figure 4. Scatterplot and least-squares regression of Arabidopsis thaliana population scores on the first principal component (PC) of trait space (vertical axis) on population scores on PC1 of climate space (horizontal axis). P-value for the slope parameter estimate and r2 of the regression line are also given.

Download figure to PowerPoint

The population mean fitness in our simulated low-elevation environment decreases by 590 mm (0.88 SDs) of summed fruit length per 1000 m asl for the site of population origin. Elevation of origin explained 46% (= 0.004) of among-population variation in summed fruit length (Fig. 5). Population trait means and SDs are presented in Table S5.

image

Figure 5. Scatterplot and least-squares regression of Arabidopsis thaliana population means for each trait (vertical axis) on the elevation of origin (horizontal axis). The line is only shown if < 0.10. Equations for each fitted line, P-values for the slope parameter estimates and r2 statistics are also provided. PSII, photosystem II.

Download figure to PowerPoint

Elevation of origin significantly predicted population means for all traits except SLA (= 0.14) and root dry mass. (= 0.74). Maximum photosynthetic quantum efficiency (Fv/Fm) increased by 103% (r2 = 0.41) per 1000 m asl for the site of population origin. Net photosynthetic rate increased by 194% (r2 = 0.57) and transpiration increased by 267% (r2 = 0.54) per 1000 m asl. Instantaneous WUE decreased by 78% (r2 = 0.41) per 1000 m asl. Rosette dry mass increased by 300 mg (r2 = 0.49) while inflorescence dry mass decreased by 540 mg per 1000 m asl (r2 = 0.39). Age at bolting increased by 7 d and the time from bolting to the first ripened fruit increased by 1.4 d per 1000 m asl, explaining 50 and 22% of among-population variation, respectively. Plants produced eight fewer basal inflorescence branches (r2 = 0.63) per 1000 m asl (Fig. 5). Scores along climate PC1 also significantly predicted population-mean trait values except quantum efficiency, SLA, and root mass; days from bolting to fruit ripening was marginally significant (Table 1). The slopes describing the relationship between climate PC1 and trait values were significantly correlated with the slopes of the trait-on-elevation regressions (= 0.74, = 12, = 0.006). Climate PC2 did not significantly predict any trait variable in univariate regression (results not shown).

Table 1. Univariate results for Arabidopsis thaliana population mean trait values regressed on climate principal component 1 (PC1) scores
TraitsSlope r 2 P-value
  1. Each row represents an independent regression with the dependent variable indicated in the traits column. Slopes, r2 and P-values of the slope parameter estimate are provided for each analysis in columns.

Quantum efficiency0.0020.180.107
Net photosynthesis0.4910.380.011
Transpiration0.670.380.010
Water-use efficiency (WUE)−0.0480.290.031
Specific leaf area (SLA)−2.3530.070.330
Root mass0.2030.000.925
Rosette mass35.1170.370.012
Inflorescence mass−64.1130.300.028
Age at bolting (d)0.7240.290.033
Time from bolting to ripening (d)0.1860.220.066
Number of basal branches−0.9660.510.002
Summed fruit length−663.9150.310.024

Selection under heat and drought during the reproductive phase

We detected significant linear selection on our suite of study traits (Table 2). Quadratic models did not significantly increase explanatory power and are therefore not presented. Most traits were correlated in this study (Table S6). Both the best-fit linear (hereafter ‘best’) model based on AIC and the full linear model explained 69% of the variance in fitness.

Table 2. Full and Akaike's information criterion (AIC)-selected best models of phenotypic selection on northeastern Spanish Arabidopsis thaliana
TraitsFull modelAIC-selected best model
β ± SE P β ± SE P
  1. Summed fruit length (fitness) regressed on phenotypic values of the traits listed in each row. To satisfy regression assumptions, P-values and model r2 from regression with log-transformed summed fruit length are presented. All variables were standardized to mean = 0 and variance = 1 before analysis. Variable coefficients represented by – were excluded from the best-fit model. Selection gradients (β) and standard errors on the parameter estimates (SE) are presented along with P-values for the t-test (h0: β = 0). Significant selection gradients are in bold.

Quantum efficiency−0.05 ± 0.040.30
Net photosynthesis−0.01 ± 0.060.91
Transpiration −0.12 ± 0.06 0.03 −0.11 ± 0.04 0.003
Water-use efficiency−0.03 ± 0.040.38
Specific leaf area−0.06 ± 0.040.09−0.05 ± 0.040.15
Root mass −0.08 ± 0.04 0.02 −0.07 ± 0.03 0.02
Rosette mass0.02 ± 0.050.97
Inflorescence mass 0.62 ± 0.05 < 0.0001 0.61 ± 0.04 < 0.0001
Age at bolting (d) −0.15 ± 0.05 0.002 −0.20 ± 0.04 < 0.0001
Time from bolting to ripening (d)0.05 ± 0.030.07−0.07 ± 0.03 0.01
Number of basal branches0.05 ± 0.040.26
 r2 = 0.69 r2 = 0.69 

Both the full and best models indicated selection for earlier bolting (β = −0.15 and β = −0.20, respectively), with selection for faster fruit ripening detected only in the best model (β = −0.07). In the best model, an SD (6 d) decrease in the age at bolting increases fruit production by 0.20 SDs, c. 145 mm of fruit length. Age at bolting is nearly three times as important to fitness as time from bolting until ripening in both models tested. Age at bolting is the second most powerful predictor of fitness in both models after inflorescence mass and these traits are correlated.

In both models, selection favored lower root mass (β = −0.08 full model and β = −0.07 best model). Rosette dry mass was excluded from the best model and was not significant in the full model. There was strong selection for greater inflorescence mass (β = 0.62 full model and β = 0.61 best model). A single SD increase in inflorescence mass (721 mg) predicts a 0.61 SD increase in summed fruit length (441 mm).

Among the leaf traits, significant selection was detected only for transpiration rate. Selection favored lower transpiration rate (β = −0.12 full model and β = −0.11 best model). A decrease in transpiration rate of one SD (6.5 mmol H2O m−2 s−1) is associated with an increase in fitness of c. 800 mm of fruit length. In the best model, quantum efficiency, net photosynthesis and WUE are dropped, while SLA is retained but remains nonsignificant.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

To test the extent of adaptation to local climate, we grew plants from across an elevation gradient under dynamic growth chamber conditions that produced an accelerated winter annual life cycle and subjected plants to increasing heat and drought stress during the reproductive season, as is observed at low elevations in the field. We identified significant associations between climate and 10 of the 12 traits investigated. Populations from low-elevation coastal Mediterranean sites developed from germination to seed more quickly than those from high elevations. Low-elevation plants had relatively more mass and time invested in inflorescence structures and less in vegetative growth and rosette photosynthesis than high-elevation plants.

The first PC of climate indicated a gradient of increasing heat and drought with decreasing elevation (Fig. 2). Interestingly, most regression analyses conducted with elevation as the independent variable performed better than those using climate PC1 scores. Indeed, elevation explained 54% of the variance in trait PC1 (not shown) while climate PC1 explained only 36% of trait PC1 variation (Fig. 4). One explanation is that while elevation is accurately measured for each collection site, the climate data used are values averaged over 50 yr interpolated from nearby weather station data (Hijmans et al., 2005) and do not account for local and microclimatic site characteristics. There are other factors that may covary with elevation and contribute to clinally varying selection, including soil, atmosphere and light environments. Additionally, there is a possible influence of the biotic community, which could vary in composition and phenology across the elevation gradient.

In this study, relative adaptedness to spring heat and drought scaled with elevation of origin, favoring low-elevation phenotypes under the imposed warming and drying regime (Fig. 5). Our findings form a complement with those of Montesinos-Navarro et al. (2011), who found, using the same study populations and an overlapping but not identical set of genotypes, that higher-elevation plants outproduced low-elevation plants when spring conditions were cool and wet. The clinal trait and fitness variation observed in this study indicate adaptive divergence as a result of differential selection among sites. It is, however, important to point out that in all cases, the direct agents of selection are not known and neither the conditions nor the developmental sequences or phenotypes of the plants in our chambers perfectly match the conditions or phenotypes seen in the field. Our source populations exhibit repeated bouts of germination during favorable conditions from autumn to spring in the field (Picó, 2008). The life cycle can be as long as 9 months or as short as 50–60 d (Picó et al., 2008). Thus there is no one life cycle or set of seasonal conditions that most accurately describes the patterns of phenotype and selection in the field. Despite this variation in the early phases of the life cycle, all plants in low-elevation populations experience spring heat and drought similar to that imposed in our chambers. It is also possible that the characters measured evolved as a result of indirect selection via correlated traits (Lande & Arnold, 1983). With these caveats in mind, we now consider the relationship among traits, elevation and climate to better understand the phenotypic mechanisms underlying adaptive differentiation across the elevation gradient described earlier.

Low-elevation plants exhibited characteristics consistent with avoidance of heat and drought stress during the spring reproductive season, including faster bolting and fruit ripening. Avoiding stress through shortening the vegetative phase and rapidly shifting resources to the reproductive structures is a key to the ruderal plant strategy described by Grime (1977) and has been observed in Arabidopsis thaliana and other species (Chaves et al., 2002; McKay et al., 2003; Griffith & Watson, 2005; Heschel & Riginos, 2005). We observed that genotypes from lower elevations bolted significantly earlier than genotypes from high elevations. This pattern of earlier bolting at hotter, drier low-elevation sites is in accordance with the range-wide pattern of earlier bolting at lower latitudes (Caicedo et al., 2004; Stinchcombe et al., 2004; Lempe et al., 2005; Wilczek et al., 2009). Earlier flowering was associated with functional shifts in the distribution of biomass among plant parts and in physiological rates.

Low-elevation plants’ distribution of dry mass reflects, in part, their earlier flowering time. Low-elevation plants had smaller rosettes and larger inflorescences relative to those from high elevations (Figs 4, 5). Initiation of primary rosette leaves ceases at bolting, because the primary meristem activity transitions to inflorescence production. This ends or slows accumulation of biomass in the rosette, depending on leaf production by axillary meristems in the rosette short-shoot (Bonser & Aarssen, 2001).

The advantage of earlier flowering in the field and its influence on the ratio of inflorescence to rosette may partly be the result of the distinct thermal niches occupied by these organs. At rosette level, radiated heat from the ground and the associated still air layer lead to significantly warmer conditions when compared with air at the inflorescence level above the ground (Geiger, 1950). Thus early flowering may not just ensure earlier reproduction, but also successful carbon gain during warm, dry spring months and increasing carbon uptake capacity while avoiding further self-shading in the rosette. In support of this hypothesis, Earley et al. (2009) showed that, on average, A. thaliana inflorescences contribute a greater proportion of lifetime carbon gain than rosettes. Earley et al. (2009) also found that the inflorescence had greater WUE than the rosette, which may be advantageous during a hot, dry low-elevation spring. Future studies of lifetime carbon gain and water use by the rosette and inflorescence along climate gradients will provide important functional understanding of variation in flowering time.

Low-elevation plants’ greater inflorescence mass may be explained in part by their greater number of basal inflorescence-forming axillary meristems (Fig. 5). This trait may also allow earlier increase in the number of fruits matured, as for n basal inflorescences, a plant will produce n fruits more or less simultaneously at approximately the same age that an single inflorescence will ripen a solitary first fruit.

One final factor explaining the relationship between elevation and the distribution of above-ground dry mass is variation in senescence and reallocation of rosette resources to the inflorescence. It is possible that maximum rosette mass is greater than the final rosette mass observed for early bolting plants. Earlier bolting may lead to earlier rosette senescence. The adaptive role of nutrient and carbon reallocation during senescence may be particularly important in environments where a rapid decrease in water availability can limit the ability of the plant to acquire nutrients from the soil, maximizing the importance of repurposing of stored nutrients.

Leaves of low-elevation plants had significantly lower Fv/Fm, carbon assimilation and transpiration rates and greater WUE than high-elevation plants (Fig. 5). McKay et al. (2003) found that earlier bolting genotypes of Athaliana were less water-use-efficient, as evidenced by decreased carbon isotope discrimination. Low-elevation plants that bolt earlier produce much larger inflorescences both overall (Fig. 5) and relative to the rosette (not shown). Inflorescences can contribute greater lifetime carbon gain while being more water-use-efficient than rosettes (Earley et al., 2009). Thus, low-elevation plants in this study may circumvent the expected tradeoff between drought avoidance and tolerance mechanisms observed in McKay et al. (2003). This result is likely to be dependent on the timing of the measurements relative to the life-history stage and the imposition of stress.

Our low values of gas exchange rates and Fv/Fm among low-elevation plants may indicate senescence of the measured leaves. This is in line with indications that low-elevation conditions produce plants that develop more rapidly than their high-elevation counterparts. The leaves we measured showed no visible sign of senescence at the time measurements were taken. However, recent studies of the molecular and physiological underpinnings of senescence indicate that the process itself begins before visible signs appear (Balazadeh et al., 2008).

We conducted gas exchange measurements in the late afternoon to assess the ability of experimental plants to photosynthesize under heat stress. Our results would not necessarily correlate with measurements taken earlier in either the daily or the developmental cycle. Nevertheless, our measures of gas exchange, WUE and quantum efficiency were all strongly related to the elevation of each population, indicating that measured or correlated unmeasured traits played a significant role in adaptation to the environmental gradient.

There was strong selection for high inflorescence masses (Table 2). This is expected, as the greater the inflorescence size, the greater the number of fruits borne. Additionally, there was selection for earlier flowering and shorter time until fruits ripen, which matches the life histories of low-elevation plants. Multiple regression analysis indicates a much stronger relationship between inflorescence mass and fitness, when controlling for correlated traits such as bolting time. Inflorescence mass is correlated with basal branch number (= 0.52) and rosette dry mass (= −0.62) neither of which are significant in the selection analysis (Table S6).

Inflorescence mass explained only 63% of the variance in fitness in a univariate regression (results not shown). Indeed, while low-elevation plants produced the most fruit overall, they also produced more fruit length per unit inflorescence mass (result not shown), that is, they exhibited greater mass use efficiency in the production of fruits under the conditions of this experiment. Given that inflorescences contribute significantly to lifetime carbon gain and have greater WUE than vegetative rosette tissue (Earley et al., 2009), they are likely to possess adaptive function above and beyond structurally supporting fruit production, further reflecting the advantage of conserving water via stomatal closure in the rosette while photosynthesizing in the inflorescence.

This study demonstrates variation in relative adaptedness of plants from across a climate gradient to heating and drying during the spring reproductive phase. Low-elevation plants from NE Spain were able to maximize seed production given the short reproductive season we imposed because they bolted early and allocated more to reproductive than to vegetative structures and because they ripened fruit more quickly. We propose that this is a syndrome of avoidance through early flowering accompanied by restructuring of the organism that adapts A. thaliana to low-elevation Mediterranean climates.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We are grateful to F.X. Picó for introducing us to this system and for the pleasure of collaboration on fieldwork and collections. We are deeply thankful to T. Elnaccash, A. Montesinos-Navarro and Pitt E&E's PEER discussion group for constructive criticism and commentary during the development of this project and writing of this manuscript, Thanks to T. Helbig, M. Simon and E. York for their contribution to plant maintenance and measurements. Thanks to N. Settles, J. Anderson and two anonymous reviewers for valuable commentary and to K. Garmire for help with GIS. Funding was provided by NSF IOS 0809171 and IOS 1120383.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Please note: Wiley Blackwell are not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

FilenameFormatSizeDescription
nph12485-sup-0001-FigsS1-S4-TablesS1-S6.docxWord document625K

Fig. S1 Experimental conditions Arabidopsis thaliana genotypes were exposed to.

Fig. S2 Field conditions for low elevation populations of NE Spanish Arabidopsis thaliana.

Fig. S3 Eigenvalues from PCA of Arabidopsis thaliana population values for 19 bioclimatic variables.

Fig. S4 Eigenvalues from PCA of Arabidopsis thaliana population means for 12 measured traits.

Table S1 Geographic and climatic information data for the 16 locations in NE Spain where the study genotypes of Arabidopsis thaliana originated

Table S2 Eigenvector coefficients of first principal component resulting from a PCA of the value 19 bioclimatic variables at 16 study populations of Arabidopsis thaliana in NE Spain

Table S3 Principal Component Scores of each population of Arabidopsis thaliana for the first principal component of trait space plus the first and second principal components of climate space

Table S4 Eigenvector coefficients of first principal component resulting from a PCA of the population means for each trait of Arabidopsis thaliana

Table S5 Population-mean trait values for NE Spanish Arabidopsis thaliana

Table S6 Pearson product-moment correlation matrix for NE Spanish Arabidopsis thaliana