Skip to main content

Testing the drivers of the temperature-size covariance using artificial selection

Cite this dataset

Malerba, Martino E.; Marshall, Dustin J. (2019). Testing the drivers of the temperature-size covariance using artificial selection [Dataset]. Dryad.


Body size often declines with increasing temperature. Although there is ample evidence for this effect to be adaptive, it remains unclear whether size shrinking at warmer temperatures is driven by specific properties of being smaller (e.g. surface to volume ratio) or by traits that are correlated with size (e.g. metabolism, growth). We used 290 generations (22 months) of artificial selection on a unicellular phytoplankton species to evolve a 13-fold difference in volume between small-selected and large-selected cells and tested their performance at 22°C (usual temperature), 18°C (-4), and 26°C (+4). Warmer temperatures increased fitness in small-selected individuals and reduced fitness in large-selected ones, indicating changes in size alone are sufficient to mediate temperature-dependent performance. Our results are incompatible with the often-cited geometric argument of warmer temperature intensifying resource limitation. Instead, we find evidence that is consistent with larger cells being more vulnerable to reactive oxygen species (ROS). By engineering cells of different sizes, our results suggest that smaller-celled species are pre-adapted for higher temperatures. We discuss the potential repercussions for global carbon cycles and the biological pump under climate warming.


Materials and Methods

    1. Study species and culturing conditions

As a model species, we chose the green microalgal species Dunaliella tertiolecta (Butcher) because it is cosmopolitan, tolerates a wide range of climates (from tropical to sub-polar), has an intermediate body size compared to other phytoplankton species, and grows well in the laboratory (Guiry 2019). We sourced this species from the Australian National Algae Culture Collection (ANACC; strain code CS-14) and started cultures from multiple cells per independent lineage (i.e. not clonal), cultured in autoclaved F/2 medium (with no silica) from 0.45µm-filtered seawater (Guillard 1975). We kept the environment constant using a temperature-controlled room at 21±1 °C, under a photoperiod of 14-10 h day-night and a light intensity of 150 µM photos m-2 s-1, using low-heat 50 W LED flood lights (Power-liteTM, Nedlands Group, Bedfordale, Australia).

    1. Artificial selection for size

For details on the artificial selection methods, see Malerba et al. (2018c). Briefly, the method relies on larger cells forming a pellet at the bottom of test tubes at lower centrifugal forces compared to smaller cells, which instead will remain in solution (i.e. differential centrifugation). On 25th April 2016, we inoculated 72 lineages with the same ancestral population of D. tertiolecta into aseptic 75 cm2 plastic cell culture flasks (Corning, Canted Neck, Nonpyrogenic). Since then, we selected lineages twice a week, each Monday and Thursday: 30 lineages were large-selected, 30 small-selected and 12 were the control. The selection differential for both larger and smaller cells was approx. 10% shift between cell volume before and after artificial selection. Control cultures experienced identical conditions (including centrifugation) without being size-selected. At the end of selection, all cultures were diluted approx. 3-5 times in fresh F/2 medium. Lineages were not axenic, but we kept bacterial loads to minimal levels by resuspending pelleted cells in autoclaved medium twice a week and by handling samples using sterile materials under a laminar-flow cabinet (Gelman Sciences Australia, CF23S, NATA certified).

For this experiment, we used cells sampled from 12 randomly selected lineages for each of the three size-selection treatments after 290 generations (22 months) of artificial selection. To remove any environmental effects and non-genetic phenotypic differences from artificial selection, before starting trials all cells grew for three generations (a week) under common garden conditions with no centrifugation (neutral selection). Following neutral selection, we measured the mean cell volume for all 36 lineages, using optic light microscopy at 400x after staining cells with lugol’s iodine at 2%. We calculated cell volume from at least 200 cells per culture in Fiji v2.0 (Schindelin et al. 2012) assuming prolate spheroid shape, as recommended for this species by Sun and Liu (2003).

    1. Temperature trials

After neutral selection at 21±1 °C, we diluted samples from the 36 lineages ten-fold, resuspended into fresh F/2 medium and standardized for initial total biovolume (i.e. population density × mean cell volume; unit µm3 µL-1) – which is a better predictor for resource use than population density. We submerged all cultures inside transparent and side-illuminated water baths at a controlled temperature of either 22°C (usual temperature), 18°C (-4), and 26°C (+4; see Fig. 2 for experimental design). This temperature range mirrors usual yearly fluctuations in sub-tropical regions where this species is often found. Each temperature included three independent water baths, each holding 12 lineages (four per size-selection treatment; see Fig. 2). We manipulated the temperature of each water bath using a submergible aquarium heater placed behind the samples so as not to affect light exposure. We ensured that daily temperature fluctuations were within ±1°C. The experiment took place in a temperature-controlled room and samples experienced identical light conditions of 14-10 h day-night photoperiod with a light intensity of ca. 100 µM photos m-2 s-1.

We monitored all samples for mean cell volume, population density and total biovolume at day 0, 3 and 6. We recorded population density (i.e. cells µL-1) with a flow cytometer (FlowCore, BD LSRII; BD Biosciences, Franklin Lakes, NJ, USA) using a blue laser (488 nm) and CountBright absolute counting beads (Thermo Fisher, Waltham MA, USA) as internal standards in each sample. We inferred the mean cell volume (µm3) of a population using a calibration curve between the mean of the cytometric histogram for the forward scatter (after standardizing for the mean of the beads) and the mean cell volume measured using optical light microscopy (R2 = 0.84, F1,106 = 540.9, p < 0.001). Finally, we calculated the total biovolume (µm3 µL-1) by multiplying the population density by the mean cell volume. In total, the dataset included 972 observations (3 artificial selections x 3 temperatures x 3 baths x 4 replicates x 3 times x 3 demographic parameters).

    1. ROS assays

We adopted the methods for quantifying intracellular Reactive Oxygen Species (ROS) from Dao and Beardall (2016). After three generations of neutral selection (i.e. no centrifugation), we standardized six lineages per size-selection treatment to the same total biovolume and washed three times all populations into saline 40mM TRIS-HCl buffer (pH 7 and 35 ppt). Then, we dark-incubated cells in 50µM of fluorescent probe 2’,7’ dichlorodihydrofluorescein diacetate (DCFH-DA) for 90 minutes at 37°C. After resuspending the pellet in buffer, we sonicated cells for 10 minutes. We measured the fluorescence of the supernatant in a spectrophotometer (Hitachi F-7000, Tokyo, Japan) at wavelengths of 485 nm (excitation) and 525 nm (emission). Values were converted into fluorescence units of dichloro-fluorescein (nM) using a 7-point calibration curve (2nd degree polynomial: R2 = 0.999). The buffer was the negative control and dichlorofluorescein (DCF) from 5 to 75 nM was the positive control. We standardized ROS fluorescence in a sample for cell density and for the volume of the cell’s nucleus. Nucleus size was estimated from fluorescent microscopy (excitation at 325-375nm and emission at 435-485nm) with Leica DMi8 at 400x after fixing cells with 2% glutaraldehyde and resuspending the biomass in DAPI at 0.1 µg mL-1. The allometric relationship was: log10 nucleus volume = 0.479× log10 cell volume-0.122.

    1. Data analysis

We calculated three parameters that estimate the species short-term fitness from each time-series. The daily cell production (cells µL-1 day-1) of a lineage indicates rate of change in population densities between day 0 and 3. The daily biovolume production (µm3 µL-1 day-1) is the slope of total biovolume between day 0 and 3. The population carrying capacity (cells µL-1) was the final cell density of each lineage after the time-series reached a stable state (6 days). Finally, we analysed the effects of temperature on mean cell volumes (µm3) between day 0 and 3 among size-selection treatments.

We used linear mixed models to estimate the effects of temperature at each artificial selection treatment on mean cell size and the three fitness parameters. In all models, fixed effects included Temperature (continuous, from 18 to 26) and Artificial Selection Treatment (discrete, either small-selected, large-selected or control). We ensured that treating Temperature as a discrete categorical factor did not change any of the conclusions (see Fig. S1). For mean cell size, we monitored cultures for 6 days: to focus on phenotypic plasticity and exclude evolutionary effects, we only analysed observations at the start of the experiment and after 3 days of growth, with Time (continuous, from 0 to 3) as an additional fixed factor in the model. Initial models included all interaction terms. If not significant, higher-order interaction terms were removed from the model. Final models also included a random intercept for each lineage nested within treatment. We initially included a random slope with temperature, but was later removed from the final models after being consistently selected against by model selection with Akaike Information Criterion (Burnham and Anderson 2002). We calculated probability values for the linear mixed-models using an analysis of deviance with type II Wald chi-square test and Kenward-Roger approximation to calculate the degrees of freedom (see Table S1).

We carried out all analysis and plotting in R v3.5.0 (R Core Team 2018) using packages nlme (Pinheiro et al. 2016), lme4 (Bates et al. 2015), plyr (Wickham 2011), car (Fox and Weisberg 2019) and ggplot2 (Wickham 2009).

Usage notes

This article was accepted for publication on the 13th Nov 2019 in the journal Evolution.

All analyses are carried out using a Mac computer. Please refer to the "READ ME_Instruction.txt" file for more information