Winter wren populations show adaptation to local climate

Most studies of evolutionary responses to climate change have focused on phenological responses to warming, and provide only weak evidence for evolutionary adaptation. This could be because phenological changes are more weakly linked to fitness than more direct mechanisms of climate change impacts, such as selective mortality during extreme weather events which have immediate fitness consequences for the individuals involved. Studies examining these other mechanisms may be more likely to show evidence for evolutionary adaptation. To test this, we quantify regional population responses of a small resident passerine (winter wren Troglodytes troglodytes) to a measure of winter severity (number of frost days). Annual population growth rate was consistently negatively correlated with this measure, but the point at which different populations achieved stability (λ = 1) varied across regions and was closely correlated with the historic average number of frost days, providing strong evidence for local adaptation. Despite this, regional variation in abundance remained negatively related to the regional mean number of winter frost days, potentially as a result of a time-lag in the rate of evolutionary response to climate change. As expected from Bergmann's rule, individual wrens were heavier in colder regions, suggesting that local adaptation may be mediated through body size. However, there was no evidence for selective mortality of small individuals in cold years, with annual variation in mean body size uncorrelated with the number of winter frost days, so the extent to which local adaptation occurs through changes in body size, or another mechanism remains uncertain.

Most studies of evolutionary responses to climate change have focused on phenological responses to warming, and provide only weak evidence for evolutionary adaptation. This could be because phenological changes are more weakly linked to fitness than more direct mechanisms of climate change impacts, such as selective mortality during extreme weather events which have immediate fitness consequences for the individuals involved. Studies examining these other mechanisms may be more likely to show evidence for evolutionary adaptation. To test this, we quantify regional population responses of a small resident passerine (winter wren Troglodytes troglodytes) to a measure of winter severity (number of frost days). Annual population growth rate was consistently negatively correlated with this measure, but the point at which different populations achieved stability (λ = 1) varied across regions and was closely correlated with the historic average number of frost days, providing strong evidence for local adaptation. Despite this, regional variation in abundance remained negatively related to the regional mean number of winter frost days, potentially as a result of a time-lag in the rate of evolutionary response to climate change. As expected from Bergmann's rule, individual wrens were heavier in colder regions, suggesting that local adaptation may be mediated through body size. However, there was no evidence for selective mortality of small individuals in cold years, with annual variation in mean body size uncorrelated with the number of winter frost days, so the extent to which local adaptation occurs through changes in body size, or another mechanism remains uncertain.

Background
Species' distributions and range boundaries are widely regarded to be closely related to climate [1][2][3], indicative of limits in the evolutionary potential of species to adapt to climatic days per month across the autumn and winter (October to March) from http://www.metoffice.gov.uk/ climate/uk/summaries/datasets. We calculated the regional mean number of winter frost days (FD C ), as a measure of current climate, by taking the mean of the FD indices (1994-2011) for each region, and the historical regional climate (FD H ), as the regional mean of the FD indices for the World Meteorological Office's international standard reference period (1961-1990).

Biometric data
Wren biometric data were derived from individuals trapped by ringers as part of a national ringing (banding) programme [36]. Body mass (to the nearest 0.1 g) is routinely obtained by weighing individuals using spring or electronic balances, while wing length is measured using the maximum chord method from the carpus to the tip of the longest primary feather (millimetres). An individual's body mass varies dynamically in response to continually changing temperatures and other factors [37]. As we are interested in long-term responses across the population rather than direct plastic phenotypic responses to variation in winter weather conditions, we calculated annual mean weight and wing length from adult individuals caught only during the breeding season (April to June) in each region.

Statistical analysis 2.2.1. Wren abundance and winter frost days
In order to examine the regional variation in annual population growth rate (λ) from 1994 to 2011 in relation to FD, a Poisson generalized linear mixed model (GLMM) was fitted using the glmer function in the lme4 package [38] in R v. 3.1 [39]. Annual counts were modelled as a function of region, FD and the interaction between these terms. Site (BBS square) and year were included as random effects in order to account for differences in coverage between years and the model was offset by the log of the count in the previous year to model population growth, rather than abundance. The count from the previous year was also included in the model to account for potential density-dependence in population growth rate [40]. To test the explanatory power of each of the fixed effects parameters in the model and their interaction, we used Akaike's information criterion (AIC); models with lower AIC were taken to provide a better explanation of the variation in the data. Thus, a large increase in AIC on variable removal ( AIC > 2, following Burnham & Anderson [41]) represents a notable deterioration in the model, providing 'support' for the importance of the variable that should be retained in the model. The output from this model was used to estimate a value for the number of winter frost days when population growth rate in each region (λ) was equal to 1 (i.e. population stability), as a single measure of population sensitivity to winter severity.
To test for evidence of local adaptation, this value was then correlated with FD H as a measure of historic regional climate experienced by the population prior to the study, using a Pearson correlation. To confirm that our results are not an artefact of using this two-step approach, we repeated our analysis modelling the regional variation in annual population growth rate (λ) as a function of FD H , FD and the interaction between these terms. Site, year and region were included as random effects in order to account for differences in coverage between years, previous year's count was included as a fixed effect, and the model was offset by the log of the count in the previous year to model population growth, rather than abundance. Although this second model cannot be readily used to describe the close link between the measure of population sensitivity to winter severity (the number of frost days at which λ = 1) and historic regional climate (FD H ), which is a function of both region-specific intercept and slope values, the output from this alternative model was analogous to the main results (electronic supplementary material, table S1).
In addition to modelling the effect of frost on population growth, we examined whether regional variation in mean abundance was also related to FD, again using a Poisson GLMM with a log link. Counts were modelled as a function of region, with site and year included as random effects in order to simply generate regional abundance estimates. These means were then correlated with the FD C using Pearson correlation, using the same two-step approach described above.

Wren biometrics and winter frost days
Finally, we examined the relationship between wren morphology and climate in an attempt to find a mechanistic link between population sensitivity to the number of winter frost days and regional climate. We first correlated regional variation in body mass and wing length of individuals caught in the breeding . Each data point corresponds to an annual regional mean with annual population growth rate predicted from the results of a GLMM (table 1). Lines correspond with the predictions from the GLMM (table 1). (b) Relationship between the number of frost days when populations are stable (i.e. λ = 1) and the historic regional climate (FD H ) (slope = 0.78 (±0.11 s.e.)). The number of winter frost days when the population is stable is predicted from the results of the GLMM (table 1) as plotted in (a), while the historic regional climate (FD H ) is an overall mean of the winter FD index from 1961 to 2011.
season (April to June) from 1994 to 2011 with FD c to test whether spatial variation in body size is consistent with adaptation to the local climate. We then examined whether annual variation in body mass and wing length was driven by variation in FD using a general linear model (GLM) with normal errors and a log link function, fitted using the glm function in R [39]. The mean annual mass in each region and the equivalent wing length metric were modelled as a function of region, FD and the interaction between these terms. The model was offset by the log of the mass or wing length in the previous year to model proportional change, and annual mean regional abundance was also included as a fixed effect to control for possible density effects on body size. We focus our analyses on body mass as those involving wing length yielded similar results (see the electronic supplementary material).

Results
From 1994 to 2011, both the mean number of frost days across the UK and the national wren population index were highly variable. Wren populations declined in the last years of the study, during three cold winters (electronic supplementary material, figure S2), prior to which the population had increased, coincident with a period of mild winters. There was no significant trend in either the mean number of frost days across the UK (slope = 0.07 (±0.08 s.e.), p = 0.44) or the national wren population index (slope = 0.24 (0.76), p = 0.70), during this time period.
As expected, there was a strong negative relationship between annual population growth rate (1994-2011) and FD across all 10 regions (mean slope = −0.045 (±0.003 s.e.); figure 1a and table 1). Critically, the number of winter frost days at which λ = 1 varied between regions (range = 6.9 (south-west) to 11.5 (east Scotland); figure 1), due to variation in the region-specific intercepts and the region × winter FD interaction. This FD value at stability was strongly correlated with FD H (r = 0.93, n = 10, p < 0.001; figure 1b), with a close to 1 : 1 relationship; wren populations were stable if they were exposed to a winter of equal severity to the long-term average. Consequently, the effect of the winter FD index upon population growth varied significantly with FD H (electronic supplementary material, table S1).
Despite this strong evidence for local adaptation, wrens were less common in colder regions of the UK (r = −0.78, n = 10, p = 0.009; figure 2). These effects were not a consequence of spatial variation in habitat composition between regions, as the same pattern was observed if data were included only from woodland sites, which represent wren core habitat [42] (electronic supplementary material, figure S3).
As expected, wren body mass was negatively correlated with FD c ; wrens became heavier (r = 0.69, n = 10, p = 0.026; figure 3; electronic supplementary material, table S2) and had longer wings (electronic supplementary material, figure S4 and table S2) as FD c increased. However, we found no evidence of a relationship between the annual relative change in mass (  Relationship between mass and the regional mean number of winter frost days (FD C ; slope = 0.12 (±0.04 s.e.)). Each data point corresponds to a regional mean, with mass calculated as a mean of the annual means and the number of winter FD (FD C ) calculated as the overall mean of the winter FD indices from 1994 to 2011. Colours follow figure 1.   be expected if there were differential mortality of small individuals. There was also no evidence that the relationship between body size change and FD varied regionally (p = 0.76, table 2).

Discussion
We identified the expected strong negative effect of winter severity (average number of winter frost days) upon wren populations, which was consistent across regions and matches previous studies [29,43]. The most likely environmental driver for this is reduced survival in cold winters [30,32]. Although the form of this relationship was broadly consistent between regions, there were significant differences which meant that the number of winter frost days at which individual populations achieved stability varied, with northern populations resilient to winters with up to 70% more frost days than southern ones. Crucially, there was a close correlation between historic regional climate (FD H ) and the level of frost days at which each population is stable, indicating that each wren population was closely adapted to its local climate, as measured by the 1961-1990 mean.
Ours is one of only a small number of studies whose results are consistent with local adaptation occurring in population responses to climate (e.g. [18,19,44]). Previous studies of adaption in response to climate change, which have focused on the relationship between temperature and phenology [16][17][18][19]45], have found little evidence for evolutionary adaptation driving population-level responses to warming [16,17]. This could be due to methodological difficulties in identifying such local adaptation [46], or because phenological expression has a relatively weak link to individual fitness and population-level responses [21,22]. It seems likely that we have identified an adaptive response in wrens to local climate because we have been examining variation in an environmental trait that is closely linked to survival, and therefore to selection potential. Where phenological change has been shown to have a close link to fitness, as in the winter moth Operophtera brumata, then evolutionary responses have also been identified [20].
As expected and in accordance with Bergmann's rule, we found that wren body mass was approximately 5% lower in the warmest (south-west) than in the coldest (east Scotland) region. Large individuals are likely to be favoured in colder regions due to the thermal advantage of larger size and their ability to store more fat [25,47,48]. Despite this spatial link between climate and body size, we found no evidence for significant changes in body size to occur in relation to the number of winter frost days, suggesting that differential mortality of small individuals did not occur in cold winters. This may simply result from a high noise-to-signal ratio in the annual biometric data for such a small species (with therefore little variation to measure) collected by multiple individuals; there is evidence from both cliff swallows Petrochelidon pyrrhonota [25] and sand martins Riparia riparia [24] that extreme weather events do result in differential mortality of individuals in relation to body size. Our hypothesis that the observed spatial variation in wren body size may underlie the local adaptation of wrens to climate therefore has some support from our analysis, and from other studies. However, given the lack of a significant interaction between the change in mass and region (table 2), more detailed work examining how the impact of severe weather events on the survival of wrens varies with body size is required to confirm this. The fact that variation in the number of frost days at which wren populations reach stability between regions correlates with 1961-1990 regional climate (FD H ) suggests that at least part of the mechanism underpinning adaptation may operate over a longer timescale than through relatively rapid changes in body size in response to cold winters. This could include variation in behaviour, diet, physiology or other mechanisms [49]. For the spatial gradient in wren body size to exist, there must be a cost to being too large in areas of milder climate that will provide counter-selection in warmer climates (although see [28]). This could result from the greater maintenance costs associated with larger body size [50], larger predation risk due to slower flight speeds and lower agility [51][52][53] or increased resource requirements of nestlings in order to develop large body size [54,55]. Further work is required to identify this cost.
Our finding that wrens were less common in regions of the UK where the number of winter frost days are higher might seem counterintuitive given the evidence for populations being adapted to local climate. However, such an effect may occur through a number of possible mechanisms. Firstly, spatial variation in density may not be limited by winter climate influencing survival, but other demographic parameters, such as spatial variation in reproductive success, which could be linked to climate through variation in food resources or other ecological parameters. Certainly, there is likely to be latitudinal variation in reproductive outputs, which have not yet been examined in wrens, but do occur in other passerines (e.g. [56]), and could result in an apparent correlation between abundance and winter temperature, even if not causal. Secondly, if there is a lag in the time it takes for a population to adapt to a changing climate, which seems likely, then the observed pattern between abundance and mean winter severity could simply be a consequence of that lag. Thus, a reduction in winter severity would lead to a wren population increase until counter-selection stabilized the population at a higher level. Conversely, any increase in winter severity would reduce the population until locally adapted to the new climate, when again, it would stabilize at a lower level.
Evolution is one of the three mechanisms by which a population can persist in response to climate change (the others being dispersal and phenotypic plasticity) [13]. The extent to which populations may adapt to climate change through evolution will depend upon trait heritability, the strength of selection and the rate of environmental change. Although we have not quantified these parameters, we present evidence that is consistent with winter wren populations being locally adapted to their climate, and partial evidence that this is driven by spatial variation in body size. Relatively few studies have identified genetic responses to climate change [57,58]. Wrens may show local adaptation to climate because there is a strong relationship between climate (winter cold) and fitness coupled with a high potential for local evolution to occur through fast generation times and low dispersal. This suggests that although much research on evolutionary adaptation to climate change has focused on phenological change, studies considering the selection of traits, such as body size, which affect individual fitness in response to climate-driven mortality, such as extreme events, may be equally or more useful. In such instances, work to quantify heritability and the strength of selection associated with these traits will be required to quantify the extent to which evolution may help species adapt to future climate change.
Data accessibility. The associated data of this study are openly accessible on the Dryad online data repository at: http:// dx.doi.org/10.5061/dryad.d47c1.