Conceived and designed the experiments: JS LC RC MD KM PS. Performed the experiments: JS LC RC MD KM JW PS CA OB BB MS BF HS GB MC CO BP MO AS EG UR PS ZF XJ. Analyzed the data: JS LC RC MD KM. Wrote the paper: JS LC KM.
Current address: Department of Policy Advice, National Academy of Sciences Leopoldina, Halle, Germany
The authors have declared that no competing interests exist.
Organisms provide some of the most sensitive indicators of climate change and evolutionary responses are becoming apparent in species with short generation times. Large datasets on genetic polymorphism that can provide an historical benchmark against which to test for recent evolutionary responses are very rare, but an exception is found in the brown-lipped banded snail (
Organisms provide some of the most sensitive indicators of climate change
This illustrates the variety of shell colours (Yellow, Pink, Brown) and banding (0, 1, 5) typically found. Photograph by Robert Cameron.
When exposed to sunlight, the colour of a snail's shell influences the temperature experienced by the animal within
The average land temperature in Europe increased by 1.3°C during the twentieth century, with a particularly steep rise between 1990–2009
A scoring scheme was devised for snail morphs and for the different kinds of habitats in which they are found that would enable us to compare data collected by professional biologists during the twentieth century with data that could be reliably collected by volunteers. The ground colour of the shell is controlled at a single locus (‘C’) and displays three main phenotypes: Brown, Pink or Yellow, in order of decreasing dominance
We scored the habitats in which all
Locations of published samples were usually given in the source as map references or place names which we translated into latitude and longitude. In the Evolution MegaLab (
Data on the frequencies of shell morphs, habitats and locations recorded in the twentieth century, and in just a few cases earlier, were captured from the published literature, from theses and from public and personal archives (R.A.D. Cameron, J.S. Jones and J.J.D. Greenwood). We refer to this dataset as the historical record and it contained data on 6,515 populations of
The Evolution MegaLab operated through a website specially designed for this study. The website displayed a map of all the historical data so that users could re-sample the locations of past records, although most new records were not made at the locations of previous ones. The entire website and supporting training materials were produced in 14 language versions and were used by collaborators in 15 European countries (see author list) to solicit data from the general public. Colour identification guides and videos instructed participants on how to sample, how to identify
We instructed participants to record only mature snails with a lip to the shell because its colour distinguishes
Mean surface climate data per cell on a 0.25 degree grid, as described by Haylock et al
The data were analysed by generalized additive models
A list of independent variables used is given in
ModHist | A 2-category factor that distinguishes samples collected post-2000 (mainly in the EML in 2009) from those collected in the 20th Cent and earlier. |
Year | Range 1909–2009. Fitting Year as well as ModHist enables us to distinguish time effects from other cause of differences between the modern and historical samples. |
Habitat | 4-category factor representing the habitats (woodland, hedgerow, grassland, sand dune) recognized in the EML. Note that the results shown in |
Alt | Altitude of the sample location in m above sea level, derived from a digital elevation model, (GTOPO30) with a horizontal grid spacing of 30 arc seconds (approximately 1 kilometre) |
JanPrecip | Average January precipitation mm/day |
JulyPrecip | Average July precipitation mm/day |
Temp | Temp modeled the effect of temperature as an interaction between January minimum and July maximum temperatures. |
Location | A term including an interaction between latitude and longitude, fitted independently within the modern and the historical samples. This was used to remove the deviance due to samples being made in different places in the historic and modern periods. |
For each phenotype, the model fitted to the probability (p) of the phenotype was: log(p/(1−p)) = ModHist+Year+Habitat+s(Altitude)+s(JanPrecip)+s(JulyPrecip)+s(Temp)+s(Location). Terms with an ‘s’ prefix were smoothed functions rather than linear, to improve fit. In a model with a spatial aspect like this, there is a risk that standard fitting processes will be compromised by spatial autocorrelation. Preliminary analyses of spatial correlograms and variograms indicated that in fact the degree of spatial autocorrelation in the phenotype frequencies was relatively low (and practically nonexistent at scales above 20 km), and similar analysis of the residuals from the fitted model showed negligible levels of spatial correlation in the residuals. Some additional models with fewer terms or with data restricted to the historical period were also run to test for the dependence of overall model results on the presence of particular variables such as temperature and to check for artifacts related to ModHist.
A smaller-scale, more direct analysis was performed on 554 pairs of samples where records were made in the same habitat type within 20 km of each other in both the historical and modern sampling periods. We broke the pairs down into the four habitat types and compared historical and modern samples using paired-sample t-tests on arcsine transformed frequencies.
We investigated the potential relationship between the increase in the frequency of Mid-banded and temperature by testing the hypothesis that Mid-banded would have increased most in places that had warmed the most between the historical and modern periods. There were insufficient samples made at the same locations in the two periods for this test to be performed by direct comparison of the historical and modern data. We therefore used the model to predict how much change in the frequency of Mid-banded had taken place over the 50 years from 1950–2000 at each of 2,827 locations sampled in the Evolution MegaLab. (Temperature change could not be computed for 163 of the 2,990 locations sampled.) These model predictions were used as surrogates for actual frequency change and correlated with the actual recorded changes in temperature at each location. Note that the predictions were compared with temperature change, and not temperature, which was a term in the model. This analysis is therefore going beyond the original modelling, to investigate a possible reason for the model looking as it did, and is not validating the model itself.
Evolutionary change in phenotype frequencies detected in the gam analysis is indicated by significant Year terms in
Phenotype | n | Dev% | ModHist | Year | Habitat | Alt | JanPrecip | JulyPrecip | Temp | Location | ||||
Hedge | Grass | Dune | Overall | Hist | Mod | |||||||||
Yellow | 9505 | 32.8% | ||||||||||||
α | 0.379 | −0.002 | 0.509 | 0.649 | 1.350 | |||||||||
t/F | 2.560 | −2.097 | 14.041 | 15.428 | 20.415 |
|
0.139 | 10.200 | 14.697 | 12.367 | 32.939 | 9.142 | ||
P | 0.011 | 0.036 | <0.001 | <0.001 | <0.001 |
|
0.71 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | ||
Mid-banded | 9325 | 37.2% | ||||||||||||
α | −0.096 | 0.009 | −0.075 | −0.039 | −0.798 | 5.356 | ||||||||
t/F | −0.632 | 7.114 | −1.781 | −0.820 | −9.776 | 46.426 | 4.056 | 26.507 | 6.639 | 15.190 | 36.853 | 17.182 | ||
P | 0.527 | <0.001 | 0.075 | 0.413 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | ||
Unbanded | 9505 | 28.9% | ||||||||||||
α | 0.146 | −0.009 | −0.119 | −0.010 | −1.159 | 7.442 | ||||||||
t/F | 0.972 | −7.938 | −2.886 | −0.214 | −11.836 | 50.956 | 9.769 | 16.023 | 11.536 | 8.347 | 24.313 | 5.436 | ||
P | 0.331 | <0.001 | 0.004 | 0.830 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 |
Statistics shown are coefficients (α) and t-values for unsmoothed terms and F-values for smoothed terms. P-values are approximate. n = sample size (number of populations), Dev% is the percent deviance accounted for by the model. Mid-banded frequency is calculated as a percentage of banded. Coefficient values are not given for the terms entered as smooth functions, because such function cannot be described by a single coefficient.
In this case the model omitting Habitat as an independent variable fits slightly better in terms of deviance than the model including Habitat, so no test statistic or p value can be calculated.
The results of the paired-sample analysis are given in
|
n | Frequency | t | P | ||
Historic | Modern | Change | ||||
|
90 | |||||
Yellow | 0.376 | 0.437 | +0.061 | 1.830 | 0.071 | |
Unbanded | 0.228 | 0.279 | +0.051 | 1.991 | 0.050 | |
Mid-banded | 0.347 | 0.368 | +0.021 | 0.269 | 0.789 | |
|
245 | |||||
Yellow | 0.583 | 0.561 | −0.023 | −1.118 | 0.265 | |
Unbanded | 0.240 | 0.279 | +0.038 | 2.403 | 0.017 | |
Mid-banded | 0.267 | 0.348 | +0.081 | 3.708 | >0.001 | |
|
65 | |||||
Yellow | 0.619 | 0.614 | −0.005 | 0.006 | 0.995 | |
Unbanded | 0.263 | 0.298 | +0.034 | 0.896 | 0.374 | |
Mid-banded | 0.401 | 0.480 | +0.079 | 1.622 | 0.110 | |
Dunes | 52 | |||||
Yellow | 0.534 | 0.672 | +0.138 | 2.875 | 0.006 | |
Unbanded | 0.117 | 0.133 | +0.016 | 0.745 | 0.460 | |
Mid-banded | 0.182 | 0.263 | +0.081 | 2.177 | 0.034 |
n = number of pairs.
The modern dataset displayed the expected cline in the frequency of the Yellow morph (
(a) the modern and historic datasets combined (b) the modern dataset and (c) the historical dataset. The Key shows the division of the frequency of the Yellow morph by intervals of 0.2.
The frequency of Yellow was affected by habitat (
Standard errors are shown. (a) Yellow, all differences between habitats are significant P<0.001; (b) Mid-banded, differences between dune habitat and all others are significant P<0.001. Other differences are not significant; (c) Unbanded. All differences are significant except for that between grassland and woodland. (P<0.001 for all comparisons involving dune, P = 0.023 for woodland v hedge, P = 0.027 for hedge v grassland). All P values in these comparisons were corrected for multiple testing using Bonferroni.
Our dataset on shell polymorphism in
The cline in the frequency of the Yellow morph known from the historical period
This difference between sand dunes and the other habitats which provide more vegetation cover may provide a clue as to why the expected increase in Yellow was not general. Behavioural and physiological adaptation to temperature can buffer evolutionary responses to a warming climate
Two evolutionary changes in banding, controlled by two different, unlinked loci, were detected. The frequency of Unbanded decreased over time in the complete dataset (
Bands are darker than the ground colour of all shell phenotypes and therefore a general decrease in Unbanded (and a corresponding increase in banding) does not support the climate warming hypothesis because this predicts that phenotypes with lower albedo ought to decrease. The increase of Mid-banded, which took place at the expense of morphs with more bands, was in the direction expected from the climate hypothesis and so we applied an additional test. If the climate hypothesis is correct, then the increase in Mid-banded should have been greatest where the temperature increased the most. Our test for this was based upon model predictions and showed that the change in Mid-banded frequency did correlate with local changes in temperature, but that the correlation was negative rather than positive and therefore in the wrong direction to support the hypothesis. The climate hypothesis is therefore not supported and we must look to another explanation for why the frequency of Mid-banded has increased.
Turning to the association between morph and habitat, a strong inference from earlier work was that open habitats have higher frequencies of Yellow and banding than woodland. This observation was made in a mature agricultural region of England
When Cain and Sheppard
By analyzing data on
The comprehensive data collected by volunteers in the Evolution MegaLab allowed us to apply a powerful test of an evolutionary hypothesis linking climate change with polymorphism in the banded snail. That hypothesis was clearly rejected, but to our surprise, the study also produced another unequivocal result, showing that banded morphs in general and Mid-banded morphs in particular increased. These evolutionary trends do not appear to be related to climate warming and may be related to changing predation pressure by birds. Thanks to the enthusiasm of the general public for bird watching, this hypothesis provides another opportunity for investigation using the power of citizen science.
We thank the thousands of members of the public who helped us gather the modern dataset used in this study. We are greatly indebted to Richard Greenwood for building the Evolution MegaLab website and we thank the Institute for Educational Technology at the Open University for hosting it. We thank Yoseph Araya and Ulrike Simonis for transcribing unpublished data and Matthias Glaubrecht for providing the historical data collected by FA Schilder. We thank Steve Jones for advice and he and Jeremy Greenwood for contributing unpublished data. We are grateful to Mark Beaumont for advice during the planning stages of the study and for comments on the manuscript.