Technical Comments

Comment on “Precipitation drives global variation in natural selection”

See allHide authors and affiliations

Science  26 Jan 2018:
Vol. 359, Issue 6374, eaan5028
DOI: 10.1126/science.aan5028


Siepielski et al. (Reports, 3 March 2017, p. 959) claim that “precipitation drives global variation in natural selection.” This conclusion is based on a meta-analysis of the relationship between climate variables and natural selection measured in wild populations of invertebrates, plants, and vertebrates. Three aspects of this analysis cause concern: (i) lack of within-year climate variables, (ii) low and variable estimates of covariance relationships across taxa, and (iii) a lack of mechanistic explanations for the patterns observed; association is not causation.

Attribution or testing for mechanistic relationships between climate and ecological or evolutionary parameters is a major challenge in ecology and evolution (13). Siepielski et al. (4) combined spatial and temporal studies to test whether variation of standardized directional selection gradients and differentials of a variety of traits relate to climate variables at the location of study. Three aspects of their analysis cause us to doubt their overall conclusion:

1) The use of annual gridded climate data by Siepielski et al. obscures intra-annual climate variation. We question five aspects of the climate data used in this analysis:

(i) Siepielski et al. use annual climate variables that are not relevant to the seasonal periods of the selection experiments or to the temporal scales at which selection acts (5, 6). For example, phenology of egg laying of breeding birds will respond to specific seasonal, not annual, climate variables (7). Thus, annual estimates cannot be meaningfully compared across taxa and selection estimates (Figs. 1 and 2).

Fig. 1 Map of geographic locations of studies included in Siepielski et al. and biogeographic patterns between climate and selection.

(A) Map of the study locations. (B to G) Latitude-variance and mean-variance relationships for temperature (°C) [(B) and (C)], precipitation (mm) [(D) and (E)], and PET [(F) and (G)]. A very strong pattern of higher variance in temperature at high latitude and colder locations is clear, and precipitation and PET are variable among geographic locations. Points are colored by taxa (orange, invertebrates; green, plants; blue, vertebrates). The relative size of the points indicates the standard error of the selection estimates.

(ii) Biogeographic patterns could override mechanistic relationships between climate and selection. Both the mean and standard deviation of climate variables used in this analysis are influenced by the geographic locations of the study (Fig. 1, B, D, and F), because the variance of annual temperature is related to latitude and because seasonal variation is greater in colder and higher (or lower) latitudes versus equatorial and warmer locations. Similarly, the variance in precipitation and potential evapotranspiration (PET) is related to site aridity. Greater variation in annual precipitation and PET occurred at sites in temperate regions, particularly for plant taxa (Fig. 1, D and F), which likely influenced the covariance relationships between precipitation variables and selection in plant taxa (Fig. 2, I to L).

Fig. 2 Taxa-level reanalysis illustrating the high variability in the results among different taxa.

(A to R) Points are the mean proportion of within-study variation in selection; error bars are 95% credible intervals. Climate variables include annual temperature [(A), (B), (G), (H), (M), and (N)], precipitation [(C), (D), (I), (J), (O), and (P)], and potential evapotranspiration [(E), (F), (K), (L), (Q), and (R)]. Color code is the same as in Fig. 1.

(iii) Gridded Climatic Research Unit (CRU) precipitation data are less reliable than temperature data because small-scale variation in landscapes can strongly influence rainfall patterns locally (811). The resolution of climate variables used in this study (0.5° × 0.5°) is much larger than the within-study spatial variation (4). This spatial scale mismatch could substantially influence results and is a caveat of many studies working with gridded data at global scales.

(iv) Variance of climate data is lacking at certain sites for certain climate variables; for example, PET data are lacking for 23 of the 165 study sites. When CRU data are not available for a given geographic location for a period of time, a repeated temporal mean value is used in the data set instead (8). In particular, PET in the CRU data set is calculated from the gridded mean, minimum and maximum temperature, vapor pressure, and cloud cover; thus, missing data in any of these variables will influence PET data availability (12). Care must be taken with gridded data sets to check that data are available and meaningful across all sites and time steps.

(v) The influences of the North Atlantic Oscillation and Oceanic Niño Index, also considered by Siepielski et al., are not consistent in their influences among geographic regions of the world (13). Thus, the use of these geographically heterogeneous indices for analyses is not mechanistically interpretable.

2) Siepielski et al. find low and variable estimates of covariance relationships across taxa. The taxa-level analyses do not support the overall conclusion of precipitation explaining variation in selection. For invertebrates, temperature measures explain considerable variation in selection, yet with low confidence around those estimates; for plants, relationships of precipitation and within-study variation are inconsistent; and for vertebrates, no associations are apparent (Fig. 2). In addition, within-study variation in selection estimates varied among selection measures. These differences should have been better highlighted in the overall findings of the paper, particularly given that invertebrates make up ~80% of all described species on Earth (14).

3) Siepielski et al. do not provide adequate mechanistic explanations for the patterns observed. In our assessment, the observed association presented by Siepielski et al. should not be interpreted as causation. This study does not present sufficient mechanistic explanations for why annual precipitation or temperature at study locations should be drivers of natural selection across taxa and diverse measures of selection.

For example, Study ID 68 (15) tests selection in beak and body size of song sparrows (Melospiza melodia) on Mandarte Island, British Columbia, Canada, from 1975 to 1979. The traits included in this study are overwinter survival and reproductive success. Selection in this system is thus likely related to winter mortality (extreme events) or to spring and early summer climate when breeding occurs.

The authors shared some of our concerns about their analysis, such as the temporal mismatch between climate data and selection measures, the low replication, and the low within-study variation (4). We believe, however, that these words of caution contradict the confidence with which they present their results.

In summary, for relationships between climate and evolutionary or ecological parameters to be mechanistically meaningful, it is essential that caution be taken with temporal and spatial scales of data, implementation of analyses across diverse taxa, and the interpretation of results. Without appropriate caution, studies have the potential to misdirect future research and understanding.

References and Notes

Acknowledgments: We thank A. M. Siepielski and M. B. Morrissey for very productive and generous discussions about their work, J. Hadfield for assistance with the statistical analyses, an anonymous reviewer for helpful suggestions, and M. Tseng and the Biodiversity Discussion Group for calling this paper to our attention. All code and data are archived in the following GitHub repository:
View Abstract

Navigate This Article