Evolution of thermal performance curves: A meta-analysis of selection experiments

Abstract

Temperatures are increasing due to global changes, putting biodiversity at risk. Organisms are faced with a limited set of options to cope with this situation: adapt, disperse or die. We here focus on the first possibility, more specifically, on evolutionary adaptations to temperature. Ectotherms are usually characterized by a hump-shaped relationship between fitness and temperature, a non-linear reaction norm that is referred to as thermal performance curve (TPC). To understand and predict impacts of global change, we need to know whether and how such TPCs evolve. Therefore, we performed a systematic literature search and a statistical meta-analysis focusing on experimental evolution and artificial selection studies. This focus allows us to directly quantify relative fitness responses to temperature selection by calculating fitness differences between TPCs from ancestral and derived populations after thermal selection. Out of 7561 publications screened, we found 47 studies corresponding to our search criteria representing taxa across the tree of life, from bacteria, to plants and vertebrates. We show that, independently of species identity, the studies we found report a positive response to temperature selection. Considering entire TPC shapes, adaptation to higher temperatures traded off with fitness at lower temperatures, leading to niche shifts. Effects were generally stronger in unicellular organisms. By contrast, we do not find statistical support for the often discussed "Hotter is better" hypothesis. While our meta-analysis provides evidence for adaptive potential of TPCs across organisms, it also highlights that more experimental work is needed, especially for under-represented taxa, such as plants and non-model systems.

Keywords: adaptation; climate change; experimental evolution; global change; thermal niche.

FIGURE 1

Simplified alternative hypotheses for evolutionary changes in TPCs when faced with temperature change (here only increase in temperature for simplicity; blue: Ancestors; red: Derived populations adapted to increased temperatures). (a) the ‘Hotter is Better’ hypothesis (Angilletta et al., 2010) predicts that the maximal thermal performance of a genotype increases with temperature. (b) Alternatively, TPCs could simply shift without a change in shape. (c) Increased maximal fitness from the ‘Hotter is better’ scenario could trade off with TPC width (generalist‐specialist trade‐off). Finally, (d) shows the extreme case of ‘Hotter is wider’ (Knies et al., 2009) which implies a lack of constrains or trade‐offs on TPC shape. Figure adapted from Knies et al. (2009).

FIGURE 2

Flowchart representing the total number of papers obtained after the systematic search, along with criteria used for selecting relevant papers for the meta‐analysis. The search covered all papers available in ISI web of science with our search criteria up to and including the year 2021 (the last update to the search was conducted in April 2022 but excluded entries from 2022).

FIGURE 3

Overview of all the organisms included in the meta‐analysis represented by different colours. The stacked bars represent the number of studies and the number of TPCs associated with each organism. Quantitative information can be found in Table S2.

FIGURE 4

Overall response to thermal selection. (a) Relative fitness, that is, fitness of the selection lines relative to the corresponding control or ancestor assayed at the same temperature as experienced during selection (relative assay temperature of zero). Each point corresponds to one experimental populations (data subset with 230 relative fitness estimates; 40 papers and 28 species), and the error bars represent the associated standard error. Different colours correspond to different species as shown in Figure 3. Horizontal lines visualize median posterior model predictions and shaded areas are 95% compatibility intervals. Here, the best model (Table S8) retains an effect of the type of genetic variation used in the experiments: Relative fitness is higher for studies that relied on de novo mutations (red) than those that relied on selection from standing genetic variation (blue). (b) Visualization of the ‘species ID’ random effect.

FIGURE 5

Response to selection in experiments with two assay temperatures (data subset with 122 data points in 61 TPCs from 18 papers and 15 species). The plot shows relative fitness as a function of absolute relative assay temperature, that is, the absolute difference between selection and assay temperature. Each combination of two points connected by a dashed line corresponds to one TPC (reaction norm) and the error bars represent the associated standard error. Different colours correspond to different species as shown in Figure 3. The horizontal lines visualizes the median posterior model prediction and the shaded area is the 95% compatibility interval. Here, the best model (Table S10) is the null model and the prediction interval overlaps consistently with 0, indicating no effect of selection.

FIGURE 6

Evolution of the TPC shape (data subset of studies that included more than two assay temperatures: 600 data points in 159 TPCs from 21 papers and 13 species). The plot shows relative fitness as a function of relative assay temperature, that is, the difference between selection and assay temperature, for studies that assayed selected lines at three and more temperatures allowing us to infer changes in TPC shape. Each combination of points connected by a dashed line corresponds to one TPC and the error bars represent the associated standard error. Different colours correspond to different species as shown in Figure 3. The solid lines visualize the median posterior model predictions and the shaded areas are the 95% compatibility interval. Here, the best model (Tables S12 and S13) is a cubic model that includes an effect of the type of genetic variation used in the experiments (standing genetic variation vs. de novo mutations; represented in blue and red, respectively). The inset shows only the statistical model prediction for better visibility of the prediction intervals. Note that we also find an effect of relative selection temperature (‘sign’) which we detail in Figure S6. For simplicity, we here only show results for selection at a higher temperature compared to the control (panel C in Figure S6). Generally, fitness tends to be gained at the selection temperature and slightly above and lost below the selection temperature and for very high temperatures.

FIGURE 7

Evaluation of the ‘Hotter is better’ hypothesis (data subset with 27 data points from 11 papers and 13 species). The plot shows relative fitness taking the maximal fitness value of the selected and ancestor/control lines. Each points corresponds to one comparison and the error bars represent the associated standard error. Different colours correspond to different species as shown in Figure 3. The solid line visualizes the median posterior model prediction, and the shaded area is the 95% compatibility interval.