Widely distributed species, such as the perennial plant Arabidopsis lyrata, face a range of environmental conditions across space, creating selective pressures for local evolutionary adaptation. The species' fragmented distribution may reduce gene flow, which could either reduce or increase adaptive potential. The substantial variation in phenotypic traits observed across this species' northwestern European range may reflect a combination of plastic responses to environmental conditions, evolutionary adaptation and nonadaptive genetic differentiation.
We conducted multi-site common garden experiments to study differences in plant performance in core and marginal areas. Plants from eight source populations representing the species' full geographic and altitudinal range in northwestern Europe were planted out in Iceland, Sweden, Scotland and Wales.
We found evidence of both strong plastic responses and apparently adaptive differentiation in performance. Most evidence for local adaptation was found at range margins, with the strongest effects on reproductive output. Both biotic and abiotic factors affected performance, especially at range margins. Performance of most plants was best in the Scottish and Swedish common garden sites, in the core of the species' distribution.
Despite adaptations at range margins, the performance of the species declines at distributional limits, with extreme southern populations looking particularly vulnerable.
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.
Many species occur over large geographical regions, with populations occurring in a wide range of ecologically distinct conditions. As a result, individuals in different parts of a species' range may experience substantially different environmental challenges which can result in plastic phenotypic change and/or different selection pressures and subsequently the evolution of local adaptation (see e.g. Kawecki & Ebert, 2004; Chevin & Lande, 2011). Populations at the margin of a distribution area are useful study systems, as they raise interesting questions both about the nature of evolutionary adaptation to local environments and about the limits to such adaptation (Woodward & Williams, 1987; Jump & Penuelas, 2005). If local adaptation occurs, these marginal populations may harbour valuable genetic variation that allows them to adapt to changing conditions (Jump & Penuelas, 2005; Bridle & Vines, 2007).
Most species have clear distributional ranges which are often interpreted as being set by climatic factors (e.g. Coope, 1995; Sexton et al., 2009), suggesting a limited adaptive potential of species to adapt to the extreme (local) climatic conditions (Lennon et al., 1997; Gaston, 2003; Holt & Keitt, 2005). Thus, the absence (or incompleteness) of local adaptation in marginal environments is itself a matter of interest. The degree of local adaptation may result from the interplay of selective processes imposed by the local environment and nonselective processes such as mutations, genetic drift and the amount of gene flow (Holt & Gomulkiewicz, 1997; Kirkpatrick & Barton, 1997; Galloway & Fenster, 2000; Butlin et al., 2003). Some have suggested that local adaptation in marginal populations is evolutionarily constrained by the lack of selectively relevant genetic variation or appropriate mutations enabling adaptation to extreme environments (Holt & Gomulkiewicz, 1997; Barton, 2001; Holt & Keitt, 2005; Alleaume-Benharira et al., 2006; Colautti et al., 2010). Gene flow between marginal and core populations may increase genetic variation and eventually increase the adaptive potential of marginal populations to respond and adapt to different selection pressures (Barton, 2001; Garant et al., 2007; Sexton et al., 2009). By contrast, others suggest that gene flow severely limits or even prevents local adaptation (Kirkpatrick & Barton, 1997; Case & Taper, 2000; Lenormand, 2002). The larger sizes and a high reproductive output of core populations could lead to a net flow of genes from core to marginal populations (Carey et al., 1995; Gaston, 2003; Angert, 2006), bringing in a flood of genes that are not adapted to the local environment and reducing the likelihood of local adaptations evolving (Case & Taper, 2000; Lenormand, 2002; Etterson, 2004; Goldberg & Lande, 2007). These two competing hypotheses concerning the constraints on evolutionary adaptation at range margins make opposite predictions about the effects of gene flow: gene flow can either promote local adaptation (if adaptation is constrained by the absence of relevant genetic variation) or prevent it (in the case of genetic swamping). This may suggest an intermediate gene flow optimum for local adaptation (Holt et al., 2003).
Reciprocal transplant common garden experiments provide a powerful tool for differentiating adaptive, nonadaptive and plastic changes in plant traits across a species' range (Pigliucci, 2001; Kawecki & Ebert, 2004; Hereford, 2009; Sexton et al., 2009). Greater fitness of locally sourced individuals relative to those transplanted from nonlocal populations suggests that they harbour adaptations to their local environment. Local adaptation has been shown in a range of species, both over long distances (Clausen et al., 1948; Montalvo & Ellstrand, 2000; Leinonen et al., 2009) and over relatively short distances between contrasting habitats and across altitudinal gradients (Knight & Miller, 2004; Sambatti & Rice, 2006; Gonzalo-Turpin & Hazard, 2009). However, most of such common garden trials have limited resolving power as they rely on only one or two sites.
This paper reports on multi-site common garden experiments spanning the range of the perennial plant Arabidopsis lyrata ssp. petraea. In northwestern Europe, A. l. petraea occupies sparsely vegetated habitats in subarctic and subalpine environments, including Iceland and Scandinavia, through a range of fragmented sites in Scotland, down to small and highly isolated sites in Wales and Ireland at the extreme southwestern margins of its range. In field surveys of this species (Vergeer & Kunin, 2011), we have documented substantial geographical variation in a number of traits, including particularly strong differences in life history strategies. Such differences could result from genetic evolution to different environmental conditions across the range (local adaptation) or from nonadaptive genetic differentiation between isolated regional variants, or alternatively may reflect plastic responses to the different environments. Common garden trials can help us to distinguish the relative importance of these different components in explaining local differences.
Earlier work on this system has documented within- and between-population variation, both in local temperatures and in neutral genetic diversity in this species (Kunin et al., 2009). Contrary to predictions, no strong regional differences in within-population genetic diversity across the species' range were observed, but there were substantial differences in between-population differentiation as a function of geographical distance. Indications were found for reduced gene flow and high levels of genetic drift in the highly isolated and fragmented southern populations in Ireland and Wales (Kunin et al., 2009; but see Lesica & Allendorf, 1995). By contrast, low levels of genetic differentiation with distance were observed in the northern, mostly continuous Icelandic populations. Thus, the different levels of habitat (and population) fragmentation observed across the species' distribution appear to be reflected in differences in the strength of inter-population gene flow and spatial genetic differentiation. This suggests that A. l. petraea may be a good test species for examining the role of gene flow in constraining or facilitating local adaptation, as it appears to exhibit much higher gene flow near its northern range margin (Iceland) than at its southern margin (Ireland and Wales).
Here, we investigate whether A. l. petraea populations have adapted to their local conditions and if levels of local adaptation vary across the species' northwestern European range.
Plants from northern regions (Iceland), more central regions (Norway, Sweden and Scotland) and southern regions (Ireland and Wales) were reciprocally planted in four common garden sites across the species' range, and performance was then followed for > 2 yr. If local adaptation occurs, we expect plants taken from populations close (geographically and environmentally) to each common garden site to perform best. We further hypothesized that the extremely isolated populations from the southern range margins would be less well adapted to their local environment compared with the more spatially continuous marginal populations from the northern range margin, as a consequence of the negative genetic effects of fragmentation. Plants from the more central regions would be expected to inhabit more optimal environments for the species than plants from the marginal regions, and would therefore be expected to perform less well when transplanted to northern or southern range extremes.
Materials and Methods
Arabidopsis lyrata ssp. petraea (L.) is a short-lived perennial native to much of northwestern and central Europe. It is an obligate outcrosser, mainly pollinated by insects. Although the small seeds lack mechanisms for long-distance dispersal, the plant flowers over a relatively long period which provides an opportunity for pollen-mediated gene flow among different populations (Clauss & Mitchell-Olds, 2006). The species is restricted to sparsely vegetated natural habitats such as rock-faces, gravel bars and scree slopes. In Iceland, the northern margin of its range (where such habitats are widespread), the species is common with large and often continuous populations spread at low density over large areas (Kristinsson, 2005). In Norway and Sweden, populations are more fragmented but locally still fairly common. In Scotland, populations are even more fragmented. In Wales and Ireland, the southern ranges of its northwestern European distribution, only a few scattered and highly isolated populations exist (National Biodiversity Network gateway database).
In the summer of 2006, four experimental transplant sites (common garden sites) were selected. Two common garden sites were located at the margins of the species' northwestern European distribution; one in northern Iceland, just beyond the northern range margin, and one in Wales near the southern margin. The other two common garden sites were located in Sweden and Scotland, representing more central parts of the distribution. The common garden sites were established in natural habitats in the vicinity of natural populations, but far enough away (> 2 km) to minimize the risk of genetic contamination of the natural populations.
In the summer of 2005, seeds were collected from eight source populations in six countries (Fig. 1): two Icelandic (representing the northern margin), two Norwegian, one Swedish and one Scottish (representing the central part of the distribution), and one Irish and one Welsh (representing the southern margin). In Iceland and Norway, where the species occupies a wide altitudinal range, high- and low-altitude populations were selected to allow for climatic differences within each region (denoted hereafter by the letters ‘h’ and ‘l’ for ‘high’ and ‘low’, respectively). In each population, we measured population size and population density, according to Vergeer & Kunin (2011). From each population, seeds were sampled from 20 haphazardly selected plants (all at least 5 m apart) and germinated in May 2006 in trays containing a mixture of sharp sand and compost (3 : 1). After 6 wk, when the majority of the plants reached the four-leaf stage, three to five seedlings (depending on germination success) from each maternal plant (hereafter called ‘family’) were transplanted into each common garden site (one seedling per 10 × 10 cm). At each site, five blocks accommodating 125–150 seedlings each were planted. Seedlings were allocated so that plants from different families and sources were equally distributed among the different blocks. This experimental design resulted in a total of 2433 plants (three to five seedlings from 20 families of eight source populations in four common garden sites), with 600, 669 and 694 plants in Iceland, Scotland and Wales, respectively. In Sweden, many seedlings died just after germination (before transplanting) as a result of extreme hot and dry weather conditions in the summer of 2006, resulting in only 470 plants being transplanted into the Swedish common garden site. Initial plant size, estimated as the number of leaves multiplied by the length of the longest leaf, was recorded on the day of transplanting.
All 2433 plants were measured at the approximate time of peak flowering (June) and at the time of peak seed set (August), from early June 2006 until late August 2008. From each plant, the number of leaves and length and width of the largest leaf, the number of inflorescences and siliques, the length of the longest silique, and the time to first flowering (summer of 2006, 2007 or 2008) were measured. In addition, visible damage (yes/no) caused by herbivory was recorded. A ‘biomass index’ was estimated by multiplying the total number of leaves by the length of the largest leaf. This biomass index, which was estimated at different life stages, was highly correlated to actual aboveground biomass (Pearson's correlation, R = 0.78; n = 113; P < 0.001; measured on a subset of test plants), enabling us to follow (aboveground) plant growth through time. As an estimate of seed set, we estimated total silique length for each plant by multiplying the total number of siliques by the length of the longest silique (the maximum length was chosen so as not to penalize for immature siliques at the time of survey). In all populations, silique length was highly correlated with the number of seeds produced per silique (Pearson's correlation, R = 0.49; df = 353; see also Vergeer & Kunin, 2011). Average total silique length values were calculated per family and multiplied by the flowering percentage of that specific family to estimate the ‘reproductive output’ per family. In the summer of 2008, all surviving plants (1724 in total) were harvested. In each common garden site, plants from three blocks were used to estimate dry weight root and shoot biomass (oven-dried for 24 h at 70°C). For all traits, the relative performance was calculated by dividing the trait value of each plant at a common garden site by the mean trait value at that site, after Hereford (2009).
Temperature was measured at each common garden site and source population, every 120 min for 2 yr (August 2006 to August 2008), using temperature monitoring probes (Thermocron iButtons Model DS1922L; Maxim Dallas Inc, San Jose, CA, USA). Four iButtons were placed out at each site, buried 1 cm below the soil surface and always within 10 cm of an A. l. petraea plant. Temperature records were analysed by month and by year to produce the average numbers of degree days above 5°C, above 20°C and below 0°C, growing season length, number of days with snow cover, average summer day length and the minimum, maximum and average summer and winter day, night and 24-h temperatures. Summer was defined as the period June–August and winter as the period December–February. Growing season length was defined as the period beginning when the maximum temperature on five consecutive days exceeded 5°C and ending when the maximum temperature on five consecutive days was below 5°C. The number of days with snow cover was defined as the number of days with < 1°C variation in temperature in 24 h and with temperatures constantly below 1°C.
Two soil samples, each consisting of 10 subsamples, were taken haphazardly at each population and common garden site with an auger (depth 10 cm; diameter 3.0 cm) during the June visit of each year. Samples were stored at 4°C until further analysis. In the laboratory, soil samples were homogenized and sieved (4 mm) before extraction and analysed for soil nutrients, organic content, pH and base cations. All analyses were performed according to methods detailed in Vergeer et al. (2008).
To reduce the number of environmental variables, and to avoid multi-collinearity, a principal component analysis (PCA) based on a correlation matrix was carried out on the environmental variables to construct a measure of ‘environmental space’ across the study area (see Kunin et al., 2009), using the ordination program canoco for Windows 4.5 (Ter Braak and Šmilauer 2002). The PCA scores of the first four axes, capturing 83.95% of the variance in the original data set (see Supporting Information Table S1), were used to calculate Euclidean distances in environmental space (‘environmental distances’) between the common garden sites and natural populations (see Kunin et al., 2009). Geographical distances between source populations and common garden sites were also calculated.
The effects of common garden site and plant source on different traits were analysed with linear mixed effect models with common garden site and source effects (and their interaction) included as fixed factors, and with block, nested within common garden site, and family, nested within origin, included as random factors. Dunnett's post hoc tests and pair-wise 95% confidence interval comparisons were used to test for differences between different source populations within a common garden site. Plant size at time of transplantation was included as a covariate when analysing growth and biomass. All biomass measurements were log-transformed. Plant survival, probability of flowering and probability of herbivory were analysed using a binomial distribution. In these models, the importance of each explanatory factor in the minimum adequate model was assessed by comparing this model with a reduced model (with all the terms involving the factor of interest removed) using the likelihood ratio test (Zuur et al., 2009). All analyses were performed in the R (version 2.12.1) statistical and programming environment (R Development Core Team, 2009) using the packages lme4 and lmer.
To test the relative importance of plasticity, local adaptation or differences between sources, we partitioned the variance in our data between effects of common garden sites (as an estimate of phenotypic plasticity), differences between sources (as an estimate of – not necessarily adaptive – trait differences between populations), the interaction between the two components (showing that different populations respond differently to different environments, potentially indicating local adaptation) and variance that remained unexplained. The relative importance of the different variance components was analysed using variance components analyses via the restricted maximum likelihood (REML) method (Logan, 2010) Variance partitioning of this sort is increasingly used to parse the relative contributions of multiple factors and their interactions in explaining observed patterns (see e.g. Parker et al., 2003; Albert et al., 2010, 2011).
The measurement of local adaptation is central to this paper. Here we operationally define local adaptation by a source × common garden site interaction where the local population systematically outperforms nonlocal populations at its home site (e.g. ‘local vs foreign’; see Kawecki & Ebert, 2004). In addition, to quantify the strength and direction of such interactions, we also compared the observed performance (Pobs) of plants of a given origin in a particular site to what would be expected (Pexp) given the mean values overall for those populations and sites, a measure we termed the ‘performance relative to expectation’ (‘PRE’) index. Classical ‘reaction norm’ graphs emphasize either how each population performs across sites or (conversely) how well different populations perform in particular sites (e.g. Kawecki & Ebert, 2004; Hereford, 2009). Our PRE index takes into account both site and source differences in mean performance, thus allowing us to show, for example, whether a particularly stressful environment appears to be less stressful (in relative terms) for individuals from local populations, or whether a generally poorly performing population performs less badly (relatively) when in its home environment. This index thus allows us to visualize the direction and strength of site × source interactions, to see whether they are suggestive of local adaptation. To calculate the PRE index, we first define the expected performance () of a trait value as:
where is the mean performance () of all plants (all origins) in common garden site x, is the mean performance of all plants in all common garden sites and is the mean performance of all plants from origin y in all common garden sites. The PRE index can then be calculated as the relative difference (standardized for expected performance) between the observed and expected performances: . Further details on the PRE index are provided in Notes S1.
Substantial variation in growing season length and temperature was found between the different natural populations and common garden sites. Irish, Welsh and Scottish populations generally experience cool summers and mild winters with little snow cover (Table 1). The Scandinavian and Icelandic populations typically face colder winters, long periods of snow cover and shorter growing season lengths. Relatively warm summers were measured for the Swedish population and the low-altitude Norwegian populations, while the summers for the high-altitude Norwegian and all Icelandic populations were relatively cool. Less variation was observed in abiotic soil conditions: in general, populations from the British Isles were found on more eutrophic soils with higher concentrations of NH4+, whereas most Scandinavian and Icelandic populations grew on soils low in nitrogen (N) (Table 1). PCAs of environmental conditions revealed distinct clusters at a regional scale, with the first axis separating the British Isles from the Scandinavian and Icelandic sites and axis 2 separating Scotland from Ireland and Wales and Iceland from Sweden (Fig. 2). Extreme winter temperatures, snow cover, day length and, although to a lesser extent, soil nutrient variables (mainly carbon (C) : N and NH4+) contributed strongly to the first axis, while degree days above 5°C and summer temperatures were more associated with the second axis (Table S1). Environmental conditions of the common garden sites were most similar to those of the local populations.
Table 1. Environmental conditions and population sizes of natural Arabidopsis lyrata ssp. petraea populations and common gardens
C : N
Shown are latitude and longitude, population size (estimated number of rosettes), population density (rosettes m−2), growing season length in number of days, number of degree days above 5°C (DD5) and below 0°C (DDmin0), number of days of snow cover, average summer month temperature in °C (Smon), average winter month temperature in °C (Wmon), soil pH, the organic content of the soil, soil C : N ratios and soil NH4+ concentrations (μmol kg−1 DW). High- and low-altitude sites within a region are indicated by ‘_h’ and ‘_l’, respectively. For measurements of population size and population density, see Vergeer & Kunin (2011). IC, Iceland; SC, Scotland; SW, Sweden; WA, Wales.
> 10 0000
> 10 0000
Common garden sites
Difference in overall plant performance among common garden sites
Plant performance was generally best in the more central sites: Sweden and Scotland. Strong site effects were observed among different life history traits, such as in survival, reproductive output and herbivory (Table 2). All source populations survived well in the Scottish common garden site (Fig. 3; P-values < 0.01 in Dunnett's post hoc comparisons). In addition, plants from the most distant sources (Iceland and Sweden) performed better or equally well in terms of survival, growth and flowering in Scotland compared with in their local environment (Table 2, Fig. 3). The poorest performance was generally observed in Iceland, shown by relatively low survival rates and small plant sizes (Table 2). In general, large plants were observed in the more central common garden sites (Sweden and Scotland), whereas in the margins (Wales and Iceland), smaller plants were found. Flowering percentages were highest in Sweden, for both local and nonlocal sourced plants (Table 2; P-values < 0.01 in Dunnett's post hoc comparisons). In all common garden sites, time of first flowering was between 0.9 and 1.8 yr after transplantation, with in general longer average times to first flowering in Wales and Sweden (Table 2). Low numbers of siliques per unit biomass were produced in the most southern site in Wales, whereas high numbers of siliques per unit biomass were produced in the northern Icelandic common garden site (Fig. S1). Signs of herbivory were seldom observed at the Icelandic and Scottish common garden sites. By contrast, levels of herbivory were relatively high in Wales and Sweden, with an average of c. 30% of all plants showing signs of herbivory (Table 2).
Table 2. Characteristics of Arabidopsis lyrata ssp. petraea plants from different sources grown at different common garden sites
S × CG
Measurements are from August 2008 (2 yr after transplantation, at the end of the experiment), apart from time to first flowering. Averages and levels of significance are shown (+, P < 0.10; *, P < 0.05; **, P < 0.01; ***, P < 0.001). Common garden sites are indicated by capital letters (IC, Iceland; SC, Scotland; SW, Sweden; WA, Wales), and different source populations by upper/lowercase letters with high- and low-altitude sites indicated by ‘-h’ and ‘-l’, respectively (see Fig. 1 for locations). Local populations are indicated in bold. CG, S and S × CG show the significance levels of the common garden (CG) site effect, the source population (S) effect and the source × common garden site interaction effect, respectively.
There were also consistent differences in the performance of plants from specific populations that held up across multiple common garden sites. Plants from the southern range margin tended to have low survival and low herbivory overall (Welsh plants in particular; Table S2). Plants from lowland sites in Norway (No-l) showed consistently high survival, flowering proportions, herbivory and reproductive output (Table S2), with upland Norway (No-h) and lowland Iceland (Ic-l) populations showing similarly high values for many of these variables.
Local adaptation and phenotypic differences between different source populations
In many cases local populations outperformed nonlocal sources, providing strong evidence for local adaptation. In Iceland and Wales, survival of local source populations was generally among the highest of all populations grown in those sites (Table 2, Fig. 3). This was illustrated by a relative performance of 1.35 for the local population in Iceland (Ic-l) compared with an average value of 0.93 for all nonlocal sources (Table 3). In Wales, lower survival of the nonlocal sources ‘Ir’ and ‘Sw’ was observed compared with the local population (Table S2). Plants from all sources survived well in the Scottish common garden site where no evidence of local adaptation was found (Table 2). Variance partitioning analysis revealed that, although only a small amount of total variance in survival was explained, a relatively large proportion of the explained variance in survival was accounted for by differences among common garden sites (Table 4). This shows that survival and consequently lifetimes vary greatly among the different regions of this species.
Table 3. Relative performance of Arabidopsis lyrata ssp. petraea plants from different source populations grown at different common garden sites, measured in August 2008 (2 yr after transplantation)
S × CG
Relative performance was calculated by dividing the trait value of each plant at a common garden site by the mean trait value at that site, after Hereford (2009). Relative performance of the local source populations is indicated in bold. Values > 1 indicate that sources perform above average and suggest local adaptation. Averages and levels of significance are shown (ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001). For further details, see Table 2.
Table 4. Percentage of total variance for the factors common garden site, source population, common garden site × source and the unexplained variance for the models explaining survival, dry weight total biomass and reproductive output of Arabidopsis lyrata ssp. petraea measured in August 2008, 2 yr after transplantation
Proportion of variance
Common garden sites
Common garden site × source interaction
The percentage of explained variance is indicated in brackets.
Strong patterns of apparent local adaptation were observed for dry weight biomass. In Iceland and Wales the local plants grew to be substantially larger than nonlocal plants (Fig. 3), whereas in Sweden and Scotland no such differences were observed. However, biomass varied greatly among the different common garden sites and a large proportion of the total and explained variance was explained by common garden sites (Table 4), indicating high plasticity for this trait.
In Iceland, most of the Irish and Welsh plants flowered c. 1 yr after transplantation (Table 2), but then often died after flowering without producing new off-shoots (data not shown). After 2 yr hardly any flowering Irish and Welsh plants were recorded in the Icelandic common garden site (Table 2). In Iceland, Sweden and Wales, plants from local populations produced more siliques per unit biomass index as compared with nonlocal plants (Fig. S1). The local Icelandic (Ic-l) plants outperformed all nonlocal populations in terms of reproductive output (Table 3, Table S2), again providing strong evidence for local adaptation. In the Scottish common garden site, all sources performed equally well (Fig. 3, Table S2). For reproductive output, a large proportion of the total explained variance was attributed to common garden site × source interaction effects (Table 4), with an equally large amount of variance explained by differences between common garden sites. Population differences were also observed in the proportion of plants that suffered from herbivory. Relatively high levels of herbivory were reported in the Swedish and Welsh common garden sites (Table 2), but levels of herbivory on the Welsh (Wa) and Irish (Ir) plants were lower than those suffered by plants from most other sources (Table S2).
Local adaptation was also quantified using the PRE index. Highly positive PRE index values were observed for most life history traits for the local population in the Icelandic and Welsh common garden sites (Fig. 4), indicating a higher performance of the local population in these most geographically marginal sites.
Local adaptation related to geographical and environmental distance
The observed differences in performance may be partially explained by the geographical and environmental differences between our sites. In the Icelandic common garden site, the lowest performance in terms of survival, growth and reproductive output was observed for the most distant source populations such as those from Ireland (Ir) and Wales (Wa) (Table S2). Survival probabilities and flowering percentages for plants originating from nearer regions (such as Ic-h, No-l and Sw) were substantially higher (Table 2). In the more central common garden sites in Sweden and Scotland, no such patterns with distance were observed. In these sites, geographical distances from source populations had little value in predicting the performance of transplants (Fig. 5). In Iceland, however, a significant negative relationship between the PRE index for reproductive output and geographical distance was observed (linear models, F1,131 = 9.635; P = 0.002; Fig. 5), showing that plants from geographically more distant sources performed more badly than expected, based on the relative quality of different sites and sources. Geographical and environmental distance were highly correlated (Pearson's correlation, R = 0.88; n = 131; P < 0.001) and similar patterns were observed when the PRE index for reproductive output in Iceland was correlated to environmental distance (linear models, F1,131 = 11.918; P = 0.001), although a lower model fit was observed (Akaike Information Criterion (AIC) values of 712 and 705 for environmental and geographical distance, respectively).
Our results provide evidence that both plasticity and genetic differentiation affected the performance of our transplants. The location of a common garden frequently affected overall performance of plants from all sources, suggesting substantial plasticity in most traits. There were also strong source effects in many traits, suggesting genetic differentiation. Finally, the strong source × site interaction terms showed that different plants performed well in different sites, with the nature of those effects strongly suggestive of local adaptation.
Local adaptation to environmental conditions
It has been suggested that the effect of local adaptation may be stronger with greater geographical distance, because environmental and genetic differences usually increase with geographical distance (Galloway & Fenster, 2000). In our study, the strength of local adaptation increased with geographical as well as environmental distance. This was most apparent in the PRE index (calculated for reproductive output) in Iceland and Wales, where it increased linearly with geographical distance. Trait values that show mean performance within each site, but that do not take into account performance at the other sites, do not always show these linear relationships (Table 2). This emphasizes the importance of the inclusion of both site and source differences in mean performance. Welsh plants performed poorly in Iceland. Environmentally, the Welsh population was predominantly separated from Icelandic populations by factors related to extreme winter temperatures and day length, suggesting that extreme winter temperatures and day length are important factors in determining the performance of this species in these marginal regions. Regional adaptation to differences in temperature and day length has already been shown for a number of species, including A. l. petraea (Riihimaki & Savolainen, 2004; Leinonen et al., 2009, 2011). Further research, such as life history studies in which demographic data are analysed as a function of environmental factors, however, is required to quantify the importance of environmental variables for performance and population growth rates across a species' range.
Local adaptation across the distribution range
It is generally assumed that fitness will be highest towards the core of a species' range (Kirkpatrick & Barton, 1997) both because conditions in such sites may be less stressful and because the resulting larger and more continuous populations may be expected to have the greatest adaptive potential. Our results corroborate these expectations. Generally, plants showed high fitness in terms of growth and reproductive output in the more centrally located regions (Scotland and Sweden) as compared with more marginal regions. Performance in terms of growth and reproductive output of plants from central regions declined when they were transplanted to the range margins. These results indicate that the central Scottish and Swedish sites can be regarded as environmentally core regions for the species.
Plants from southern regions (Wales and Ireland) that were transplanted to northern regions showed low survival rates and an almost complete lack of flowering after 2 yr. These differences may well reflect maladaptation to the extreme harsh climate in the north with extensive periods of frost during the winter. Clinal variation in freezing tolerance has already been observed between German and Scandinavian populations of A. l. petraea (Leinonen et al., 2009). In addition, distinct metabolic phenotypes associated with exposure to cold temperatures were observed in Welsh, Swedish and Icelandic populations (Davey et al., 2009), and a lower tolerance of Welsh plants compared with Icelandic plants to extensive periods of cold was suggested. A contrasting response was observed when Icelandic plants were transplanted to Wales. These plants also showed reduced performance, mainly in terms of much higher herbivory and reduced flowering in the second year, but nonetheless, higher survival rates and greater biomass were observed compared to when they were grown in their home region. These results not only reveal strong local adaptation in the northern range margin but also suggest that northern plants are capable of growing well in milder climates whereas southern plants are not able to thrive in colder climates.
In addition, large differences in life history traits were observed across the range, which were associated with local environmental conditions. Earlier studies (Vergeer et al., 2008; Vergeer & Kunin, 2011) suggest that plants from northern regions have shorter lifespans on average and reproduce earlier compared with plants from southern populations. This appears to be partly a plastic response, as even plants from distant regions showed earlier flowering when grown in these and other regions (e.g. Ir and Wa plants grown in Iceland and Scotland; Table 2, P-values < 0.05 in Dunnett's post hoc comparisons), but at an apparently high cost in terms of survival. However, there are also clearly genetic differences in these life history traits, such that Icelandic and Norwegian plants showed a consistent suite of traits even when grown in other regions. These observations counter common observations of high-latitude/altitude plants which in general live longer and reproduce later than their relatives from more mesic regions (Grime, 2001), but are consistent with transplant studies with high-elevation/latitude plants in which counter-gradient patterns are observed (e.g. see Conover et al., 2009). In another study on A. l. petraea, using plants from a subset of our populations, Norwegian (No-h) plants did not show better subzero tolerances than Irish plants (M. P. Davey, unpublished data), suggesting that northern (e.g. Norwegian) accessions have not evolved much better freezing tolerances compared with their southern (e.g. Irish) relatives. As the northern regions have the coldest winters, plants face an increased risk of mortality, which in general results in shifts in resource allocation favouring early reproduction (Grime, 1977; Sibly & Calow, 1989). Plants in the northern regions may have adapted their life strategy accordingly, focusing on early reproduction rather than investing in growth and longer lifespans.
In Sweden and Wales (the only sites with substantial levels of herbivory), Welsh and Irish plants were harmed less by herbivores compared with plants from most other regions. In natural populations in Wales and Ireland, levels of herbivory are relatively high, whereas in the northern populations (including natural Swedish populations), levels of herbivory are typically low (Vergeer & Kunin, 2011). It is possible that Welsh and Irish plants, in contrast to plants from northern regions, have evolved defence mechanisms against herbivores in response to higher levels of herbivory in their natural environment (e.g. Clauss et al., 2006). This suggests that southern marginal populations are not solely limited by abiotic factors, but that biotic interactions, such as herbivory, may also have an effect (see also Bruelheide & Scheidel, 1999; Wethey, 2002; Vergeer & Kunin, 2011). Interestingly, our observations show parallelisms with what has come to be called the ‘intertidal paradigm’ (Connell, 1972); in rocky intertidal communities, it is often found that the upper limits of an organism's distribution largely depend on its degree of tolerance to abiotic stresses (e.g. desiccation), whereas the lower limits are usually affected by biotic interactions (e.g. competition and predation), and similar arguments at biogeographical scales date back to Darwin (1859, cited in Sexton et al., 2009). Results from other transplant experiments with Boechera stricta, a close relative of Arabidopsis, suggest that adaptation across a range boundary is constrained by counteracting effects of certain abiotic and biotic stress signalling pathways (Siemens et al., 2012). In their review of range margin research, Sexton et al. (2009) identify a serious lack of studies in which both biotic and abiotic factors are measured at contrasting range margins, and Wiens (2011) cites a paucity of information on biotic effects on ranges more generally; our work helps begin to fill this gap. However, in this study, only the presence or absence of herbivory was measured and there is no evidence that this measure of herbivory translates into impacts on fitness.
In addition to strong adaptive differentiation, high levels of plasticity were observed in various traits. Wide variation in biomass, as well as in survival and reproductive output, was observed among the sources in different common garden sites. For example, only small plants were observed in the Icelandic common garden site, whereas plants from all sources grew relatively large in Scotland and Sweden. This suggests that phenotypic plasticity may be at least as important as genetic differentiation for the observed phenotypic differences across this species' range. Substantial variation in plant size, reproductive effort and life history strategies has already been documented in natural A. l. petraea populations across its range (Leinonen et al., 2009; Vergeer & Kunin, 2011), differences that could be ascribed to either plastic responses to local environments or evolutionary adaptations (or indeed both). The use of common garden experiments allows us to partition the relative importance of these two components of phenotypic variation. The majority of variance explained by the models for biomass and survival was attributed to variance among different common garden sites, a measure of plasticity. For reproductive output, however, a relatively large proportion of variance was explained by source population differences, and by the interaction between sources and sites, potentially reflecting the effects of local genetic adaptation. Genetic adaptations and phenotypic plasticity are different strategies to adapt to different environments, with phenotypic plasticity being increasingly considered an important component in the evolution of adaptive responses to environmental differences (e.g. see Gienapp et al., 2008; Chevin et al., 2010). Our data suggest that for this species phenotypic plasticity is an important strategy to cope with environmental heterogeneity.
Limitations for adaptation
The relative performance of plants from the northern and southern margins was substantially better when plants were grown locally than when they were grown in more distant sites. Icelandic plants, although strongly adapted to their local environment, were still able to survive and perform well in other environments; indeed, they displayed higher survival, biomass, and silique production in other common garden sites than they did in their local site. By contrast, the Irish and Welsh populations seem less resilient to environmental changes, displaying systematic decreases in survival at all nonlocal sites, and suffering severely when transplanted to the Icelandic common garden site.
Population size and fragmentation may well explain these observed differences: the Icelandic populations are large and well connected across space. By contrast, the Welsh and Irish populations are extremely isolated, with the Welsh population in particular being very small. In a previous study (Kunin et al., 2009), no strong regional differences in within-population genetic diversity across the species' range were observed between Icelandic, Norwegian, Swedish and British populations (HS = 0.198–0.224, similar to other Scandinavian A. l. petraea populations (Clauss & Mitchell-Olds, 2006; Gaudeul et al., 2007). However, substantial differences in between-population differentiation were found, with low differentiation among populations in Iceland (FST = 0.084) and relatively high genetic differentiation among populations in the British Isles (FST = 0.243). The genetic diversity of the focal populations used in this study varied substantially between the two southern populations (heterozygosity of 0.158 and 0.129 for Ir and Wa, respectively) and all other focal populations (heterozygosity of between 0.201 and 0.233; Kunin et al., 2009), which may help explain their poor performance and low adaptability. The lower genetic diversity and increased genetic similarity within populations observed in the southern populations (Kunin et al., 2009) is consistent with the widely accepted pattern of genetic impoverishment in marginal populations, presumably resulting from increased genetic drift and reduced gene flow (Honnay & Jacquemyn, 2007), and may explain the high levels of genetic differentiation between these populations (Lesica & Allendorf, 1995). In the northern regions, little evidence of genetic differentiation with distance was found (Kunin et al., 2009). Our results suggest that gene flow within a region (between populations from relatively similar environments) may aid adaptation at range margins. These results reinforce recent findings of Sexton et al. (2011), who emphasized the importance of gene flow for local adaptation at range margins and showed that the benefits of gene flow were greatest when it occurred between populations from the same range limit.
Our findings suggest that there is local adaptation at range margins, but that it is limited. Plants from marginal populations all performed better in the core regions. The southern populations have a lower environmental plasticity than the Icelandic populations, possibly as a result of negative effects of fragmentation such as limited gene flow. The extreme fragmentation of suitable habitat at the southern margin makes the prospect of populations tracking environmental change through range shifting a remote one, and their potential for further evolutionary adaptation seems to be limited. We therefore believe that future environmental changes may threaten these small and fragmented populations in the southern range margin.
We thank Diana Bowler, Inga Brereton, Guido Cox, Sara Eriksson, James Hall, Tanaka Kenta, Martijn Slot, Anthony Turner, Steven White and Fredrick Wickström for their help. Two anonymous reviewers and Prof. L. F. Galloway are thanked for useful comments on earlier versions of this manuscript. English Nature, the Countryside Council for Wales, Scottish National Heritage and all landowners are thanked for giving permission to access their land. This study was funded by the Natural Environment Research Council under the Post-Genomics and Proteomics programme (NE/C507837/1).