Abstract
Diatoms contribute roughly 20% of global primary production, but the factors determining their ability to adapt to global warming are unknown. Here we quantify the capacity for adaptation to warming in the marine diatom Thalassiosira pseudonana. We found that evolutionary rescue under severe warming was slow but adaptation to more realistic scenarios, where temperature increases were moderate or where they fluctuated between benign and severe conditions, was rapid. Adaption to warming was linked to major phenotypic changes in metabolism and elemental composition. Whole genome re-sequencing identified significant genetic divergence both among populations adapted to the different warming regimes and between the evolved and ancestral lineages. Consistent with the phenotypic changes, the most rapidly evolving genes were associated with photosynthetic, transcriptional and translational processes. These results demonstrate that evolution of thermal tolerance in marine diatoms can be rapid, particularly in fluctuating environments, and is underpinned by major genomic and phenotypic divergence.
Introduction
Earth system models predict that global warming will result in significant declines in net primary production by marine phytoplankton throughout the 21st century (up to 20%)1,2 driven by rising temperatures exceeding limits of thermal tolerance and increases in grazing and nutrient limitation in warmer, more stratified oceans3–5. Current models however, do not consider the potential marine phytoplankton to rapidly adapt to environmental changes associated with global warming6–8. Such shortcomings have unknown consequences for projected changes global ocean primary production and arise because the mechanisms that facilitate or constrain the capacity for rapid adaptation to warming in marine phytoplankton are entirely unknown.
Here, we address this fundamental knowledge-gap by carrying out a 300-generation selection experiment, with the model marine diatom, Thalassiosira pseudonana, to assess the potential for, and mechanisms that might facilitate rapid adaptation to warming in this globally important phytoplankton9–11. A key hypothesis we test is that rapidly fluctuating temperatures – an intrinsic feature of natural environments – will play a key role adaptation12–14. Temporary exposure to a benign environment resulting from temperature fluctuations could both accelerate adaptation to severe conditions by increasing population size (a positive demographic effect) or constrain adaptation, by relaxing selection for beneficial mutations that promote persistence in the harsh environment (a negative population genetic effect)1,2. Our experiment was initiated with a single clone (the ancestor), which had an upper thermal limit of 35°C. The ancestor was then distributed among 4 experimental treatments that represent a range of warming scenarios (Fig S1 for experimental set-up): (i) a control at 22°C, which was the long-term culture temperature; (ii) moderate warming at 26°C; (iii) severe warming at 32°C and (iv) a fluctuating thermal regime, which cycled between 22 and 32°C every 3-4 generations. To understand the mechanisms that set the limits of thermal tolerance in marine diatoms and quantify the capacity for adaptation beyond present limits, we re-sequenced the genomes, measured growth rate, photosynthesis, respiration and elemental composition in the ancestor and evolved lineages after approximately 300 generations of selection.
Results & Discussion
Evolution of thermal tolerance
Trajectories of population growth rate (μ, d−1) over the course of the selection experiment differed substantially between the selection regimes (Fig. 1; Table S1-Table S3). In the control lineages, growth rates increased gradually, presumably reflecting continual laboratory adaptation. However, trajectories of growth rate in the warming treatments were markedly different from the control and one another. Lineages selected under severe warming (32°C) exhibited a characteristic pattern of ‘evolutionary rescue’14–17. After an initial increase in the first 3 weeks of the experiment, growth rates declined in the 32°C environment and remained very low (0.24 ± 0.09) for more than one year (∼ 100 generations). After approximately 100 generations, growth rates increased and were statistically indistinguishable from those in the control environment after 300 generations (0.63 ± 0.05 at 32°C, 0.77± 0.11 at 22°C). Under moderate warming (26°C), and in the regime that fluctuated between 22 and 32°C growth rate showed an immediate and sustained increase (2.1 and 1.9-fold faster than the ancestor respectively). These results yield a number of important insights that are pertinent for understanding the evolutionary dynamics of T. pseudonana in response to warming. First, adaptation to severe warming was slow, with evolutionary rescue taking over 1 year to restore growth rates to levels comparable with the ancestor. Second, in the fluctuating environment where populations experienced short bursts of exposure to 32°C followed by periods in the benign (22°C) environment, adaptation to the severe environment was rapid. Consistent with our hypothesis, lineages selected under the fluctuating regime maintained substantially larger population sizes relative to those experiencing severe warming (Fig. S2 and Table S1, S3), suggesting that temporary restoration of benign conditions increased the probability of fixing beneficial mutations required for adaptation to the severe (32°C) environment via a positive demographic effect.
To assess whether adaptation to the various selection regimes changed the thermal tolerance curves in the evolved lineages, we quantified growth rates across a temperature gradient spanning 15°C to 40°C. Populations selected under the moderate (26°C), severe (32°C) and fluctuating (22°C/32°C) warming treatments had higher optimal growth temperatures, Topt, compared to the ancestor and the control (Fig. 1B; Table S1). In the fluctuating treatments, growth rate at high temperatures was traded-off against slow growth at low temperature, suggesting that the severe environment was the dominant driver of selection. By contrast, high temperature tolerance in the moderate warming treatments appeared to incur no cost in performance at low temperature. Despite having a high Topt, the lineages selected under constant severe warming had the slowest growth rates at all measurement temperatures except for 40°C, the most extreme temperature. A key question therefore is: what mechanisms facilitated the rapid evolution of increased thermal tolerance in the lineages selected under moderate and fluctuating warming?
Increased metabolic efficiency facilitates evolution of thermal tolerance
The fraction of photosynthesis that can be allocated to growth tends to decline with warming, owing to the high temperature sensitivity of respiration relative to photosynthesis, suggesting that the upper thermal tolerance of phytoplankton reflect metabolic constraints that limit the efficiency of carbon allocation to growth at high temperature18. To investigate whether changes in metabolic traits could help explain the thermal tolerance curve of the ancestor and evolutionary shifts observed over the selection experiment, we measured responses of gross photosynthesis (P) and respiration (R) to acute gradients in temperature, spanning 4 to 45°C, in the ancestor and all evolved lineages. In contrast to previous work18, the activation energies characterising the temperature sensitivities of P and R in the ancestor were equivalent (e.g. Ea for P = 1.08eV ± .0.18 s.e.m., and Ea for R = 1.07 ± 0.17 s.e.m.), i.e. increases in P and R up to the optima were similar (Fig. 2; Table S6-S8). Optimum temperatures for P were however lower than those for R (Topt for P: 27 °C ± 0.7°C, (± s.e.m.); Topt for R = 29 °C ± 0.36°C, (± s.e.m.)), meaning that above Topt for P, the carbon-use efficiency (CUE = 1 – R/P; i.e., the potential carbon for allocation to growth) declined rapidly. The optimum temperature for growth rate (Topt= 28°C; Fig. 1B) and subsequent decreases in growth at supra-optimal temperatures coincided with declines in the CUE above Topt for P in the ancestor, demonstrating that the temperature dependence of CUE imposes a physiological constraint that shapes the thermal tolerance curve. Thus, we hypothesize that shifts in traits that increase CUE at high temperatures should have played an important role in facilitating adaptation to warming in the evolved lineages of T. pseudonana.
Following 300 generations of selection, we observed substantial shifts in the temperature responses of P and R, both among treatments, as well as between the evolved lineages and the ancestor (Fig. 2; Tables S4 –S8). For both P and R, estimates of Topt in the warming treatments were higher than the ancestor and control, and rates of P and R declined less abruptly at temperatures exceeding Topt. Consequently, CUEs remained high at hotter temperatures in the warming treatments. The mass-specific rates of P and R normalized to a reference temperature Tc= 18° C (P (Tc) and R (Tc), see Eq.6), were significantly down-regulated under moderate, severe and fluctuating warming, relative to the control treatment and the ancestor (Fig. 2 C, D, Tables S7 –S9). Rates of R (Tc) also decreased more than those of P (Tc), resulting in an overall increase in CUE in the lineages selected under moderate and fluctuating warming (Fig. 2E). These results support our hypothesis of fluctuating temperatures facilitating thermal adaptation and show that changes in the traits that increase the CUE, played an important role in the evolution of elevated thermal tolerance in the treatments subjected to warming. It is notable however, that the populations experiencing severe warming had lower growth rates and lower CUEs than the lineages in the moderate and fluctuating warming treatments (Fig. 1B, Fig. 2E, Table S1 and S9, also Fig S3). This suggests that while evolutionary rescue restored growth rates to levels comparable with the control, 300 generations of selection under constant, severe warming was insufficient to evolve CUEs that facilitate growth rates comparable to populations that experienced only brief periods of severe conditions.
Changes in subcellular resource allocation
So how were the lineages selected under the moderate and fluctuating warming regimes able to achieve the shifts in metabolic efficiency that facilitated rapid adaptation to warming? Changes in temperature are known to alter optimal subcellular allocation of resources to various macromolecular classes through phenotypic plasticity (i.e. acclimation) 19-24. Indeed, algae exposed to warming often increase their nitrogen-to-phosphorous ratios (N:P) by down-regulating the density of P-rich ribosomes, relative to N-rich proteins, owing to the increased efficiency of protein synthesis by ribosomes at higher temperatures21-24. We quantified cellular macromolecular composition in the ancestor and the evolved lineages to investigate the biochemical basis for the adaptive shifts in growth and metabolic efficiency, and found marked shifts in elemental composition, both among treatments and between the evolved and ancestral lineages (Fig. 3). Nutrient-use efficiency (expressed as growth rate per pg of nitrogen (N) or phosphorous (P)), the chlorophyll-to-carbon ratio and the quantum efficiency of photosystem II (ΦPSII; the proportion of the total light absorbed by PSII that is used in photochemistry) were highest in the fastest growing populations selected under moderate and fluctuating warming. We also found a positive temperature response in the N:P and C:P ratios, that was consistent, both in direction and magnitude, with the short-term acclimation responses characterised in the ancestor, confirming hypotheses21-23 that the direction of phenotypic plasticity (acclimation) and adaptive shifts in macromolecular composition are convergent (see also Fig S4 and Fig S5, and Tables S10-S11). Together these results suggest that the greater down-regulation of R(Tc) relative to P(Tc), which enabled increased efficiency of C allocation to growth and facilitated rapid adaptation to moderate and fluctuating warming, were in turn tightly coupled to major adjustments in allocation to the key macromolecular classes that determine cellular composition.
Molecular evolution of thermal tolerance
To investigate whether the observed changes in fitness and phenotypes in the evolved lineages were also reflected by consistent patterns of molecular evolution, we resequenced the genomes of the ancestor and the evolved populations to determine the identity and frequency of mutations. Using only non-synonymous single nucleotide polymorphisms (SNPs), we quantified the number of sites that acquired mutations in each population relative to the ancestral reference sequence, and from allele frequencies, the genetic distance of each population from the ancestor and the genetic divergence among populations. Consistent with the fitness trajectories (Fig. 1A), all evolved populations showed significant genetic divergence from the ancestor (Fig. 4A; Table S12). The lineages selected under moderate and fluctuating warming had the greatest genetic distance from the ancestor (Fig. 4A; Fig. S6, Table S13). Furthermore, the populations that experienced fluctuations between severe and benign conditions were significantly divergent from those experiencing constant severe warming. These results are consistent with the demographic data (Fig. 1A; Fig S2), indicating that the temporary restoration of benign conditions in the fluctuating regime accelerated rates of molecular evolution, potentially supplying mutations that were beneficial under the severe conditions. Comparing principal component analyses (PCA) based on combining all the phenotypic data, with the PCA from the non-synonymous SNPs, revealed striking similarities in the patterns of divergence among treatments (Fig. 4A,B, Figs S6, S7, Table S12, S13). These results suggest that the observed changes in metabolic traits and elemental composition that were linked to the evolution of thermal tolerance could also have a basis in some of the underlying genomic changes. To investigate this, we identified the top 20 SNPs most strongly associated with the first and second principal components. We then identified the genes (see Table S14) and protein functions (when annotated in the reference genome) associated with these SNPs. Mutations most strongly associated with the lineages selected under moderate and fluctuating warming clustered in gene families linked to photosynthetic pathways, while the populations evolved under constant severe warming showed a dominance of mutations in genes related to transcriptional and translational processes (Fig.4C). Mutations in transcriptional and translational genes, such as the plastid RNA polymerase subunits (rpoC and rpoC2 – associated with 32°C lineages), have been shown to be linked to stress responses in bacteria28 and may indicate that consistent with their poor growth and physiological performance, even after 300 generations of selection, these lineages were still under severe thermal stress at 32°C. Conversely, the overall larger genetic distances relative to the ancestor in the lineages selected under moderate and fluctuating warming, as well as the predominance of mutations related to photosynthetic pathways, may be indicative of more fundamental metabolic rearrangements associated with the evolution of increased physiological performance at high temperatures in these populations.
Our results shed new light on the adaptive potential and evolutionary dynamics of one of the most abundant and widely distributed eukaryotic marine phytoplankton in response to warming. We found that evolutionary rescue under severe warming was slow (> 1 year), but adaptation to more realistic warming scenarios, where temperature increases were moderate or where they fluctuated between benign and severe conditions, was rapid. The fluctuating environment accelerated adaptation to severe warming because temporary restoration of benign conditions increased population size and therefore the probability of fixing beneficial mutations required for adaptation to the severe environment. Consistent with this demographic effect, the lineages selected under fluctuating warming had the greatest genetic distance from the ancestor after 300 generations, indicating accelerated rates of molecular evolution. Because our experiments were initiated with a single clone, they may in fact be conservative estimates of the evolutionary potential of the highly diverse meta-population of T. pseudonana in the wild5 where adaptation will also be aided by standing genetic variation. Our results also show how rapid, adaptive shifts in thermal tolerance are linked to major phenotypic changes in metabolism and elemental composition as well as significant molecular evolution, particularly in genes associated with photosynthesis, transcription and translation processes. Because the changes in carbon-use efficiency in the warm-adapted lineages arise due to differences the temperature sensitivities of photosynthesis and respiration, which are highly conserved metabolic pathways, our work raises the prospect that at the physiological-level, the mechanisms underpinning rapid thermal adaptation could be universal across the broad diversity of phytoplankton. Our findings could therefore help integrate rapid evolution into models of ocean biogeochemistry and improve projections of marine primary productivity over the 21st century 1,2.
METHODS
Experimental Design
The sequenced strain of Thalassiosira pseudonana (CCMP 1335)11 was obtained from the CCAP culture collection in November 2014. The stock culture was made clonal by serial dilution in a 96-well plate in March 2015 and a single clone was chosen at random to be the ancestor for all treatments in the selection experiment. Pilot experiments with the ancestor revealed that growth rates peaked at 28°C and were negative above 35°C. The ancestor was then distributed among 4 experimental treatments that represent a range of warming scenarios based on its limits of thermal tolerance: (i) a control at 22°C, which was the long-term culture temperature; (ii) moderate warming at 26°C; (iii) severe warming at 32°C and (iv) a fluctuating thermal regime, which cycled between 22 and 32°C every 3-4 generations. Each treatment was replicated 6 times.
Lineages were grown in grown in f/2 Medium (Guillard’s medium for diatoms26) with artificial seawater, under a 12:12 light/dark cycle. Salinity was maintained at 32 (i.e. 32g NaCl L- 1 in 39.5 g L−1 artificial seawater reagents) and pH adjusted to 8.213 ± 0.291 (± s.e.m., averaged across all transfers) prior to each transfer. pH was further measured at the end of each transfer (average across all transfers: 8.171 ± 0.252). Light intensity was at 100 μmol quanta m−2 s−1.Lineages were maintained in semi-continuous batch culture, and transferred during the exponential phase of growth to an inoculum of 100 cells mL−1. This minimizes demographic effects that arise by slow-growing populations going through bottle-necks where mutational supply is limited to a point at which adaptive potential is determined by population size alone, but still allows for population size, and hence mutational supply to play a crucial role in shaping evolutionary trajectories, as faster growing lineages will receive on average more diverse inocula.
Flow cytometry
Abundance and cell size (diameter assuming a spherical shape) were determined with a Accuri C6 flow cytometer (BD Scientific). Cell size was detected using forward scatter (FSC) calibrated using species and beads of known size against photographic images processed in ImageJ as described in27, with a conversion factor of FSC=108058 μm + 4.5665. Although T. pseudonana are cylindrical rather than spherical, this is a useful approximation for relative changes in cell size. Red fluorescence (FL3) was used for chlorophyll fluorescence in order to distinguish autotrophs from bacteria, but was not used here to approximate chlorophyll content, which was instead estimated through acetone extraction (see below). Co-occuring bacteria, e.g. Rhodobacter spp and Marinobacter spp are known to be crucial for nutrient recycling and signaling molecule metabolism in T. pseudonana1,2. We therefore made no attempt to make the cultures axenic as this would have incurred a substantial fitness cost. We quantified the associated bacteria by regularly filtering the cultures between 0.2μm to 2μm, staining with SYBR gold and enumerating bacterial densities and cell size via flow cytometry. Co-occurring bacteria contributed less than 1% of total biomass.
Flow cytometry was further used in combination with a rhodamine stain to measure a proxy for mitochondrial potential – the rate of ATP production by ATP synthase is proportional to mitochondrial potential, which in turn depends on the activity of respiratory proteins embedded in the mitochondrial membrane. Since each unit area of membrane has a number of ATP synthase molecules and the rate at which each of these produces ATP depends on the mitochondrial potential, the overall rate of ATP production per mitochondria depends on both the area of mitochondrial membrane and the mitochondrial potential. Mitochondrial area is a good approximation of the area of mitochondrial membrane, and changes in mitochondrial potential can be measured in live cells using potentiometric dye. Previous work30 has shown that the styrl dye rhodamine 123 (R123) can be used in live cells of phytoplankton to measure mitochondrial potential in real time. For staining, a 10 mg mL−1 stock solution in DMSO was prepared and diluted to a 1μM working solution on the day of measurements. Of this, 20 μL were added to 200 μL of sample and left to incubate in the dark for 45 minutes. To ensure that longer incubation as is inevitable on a 96 well plate did not effect FL1 fluorescence as measured in the accuri c6, samples were randomized across each 96-well plates with the first and the last well containing the same sample. We found no effect of prolonged incubation after an incubation period of 45 minutes.
A Nile Red stain was performed to measure relative lipid contents. Here, the dye was used as a proxy to determine relative quantities of intracellular polar and neutral lipids31 in evolved samples, and in samples from short-term assays. The stock dye was obtained as a powder. The working solution was diluted to 0.1 mg L−1, and 10 μL were added to each 200 μL sample on a 9-well plate (final concentration 15 mM) and left to incubate in the dark for 30 minutes, as pilot trials had shown that after this, fluorescence levels were stable long enough for the time taken to measure one 96 well plate. Samples were randomized on the plate as described above for R123 stains. As Nile Red excites in the same wavelength as chlorophyll (FL3) and chlorophyll derivatives (FL2), samples were measured before and after adding the dye, and the chlorophyll fluorescence subtracted from the fluorescence obtained after staining the sample.
Growth rate trajectories
At the beginning and at the end of each transfer T. pseudonana cells were counted on the flow cytometer as described above and used to estimate specific growth rates (μ d−1) where Nt1 is the density of cells at the end of the transfer, Nt0 is the inoculation density, and Δt is the time passed in days.
Thermal tolerance curves
To characterize the thermal tolerance curves of the ancestor and each of the evolved lineages, an inoculum of 100 cells per mL from the middle of the logarithmic phase of growth was transferred into fresh media at 15°C, 20°C, 25°C, 30°C, 32°C, 35°C, and 40°C. Cell count was then determined daily on a flow cytometer and populations were transferred to fresh media once at each temperature during the middle of the logarithmic phase of growth, before being left to reach stationary phase. Growth rates were estimate from the abundance data using a the logistic growth function where t is time, K is carrying capacity of the population (cell mL−1), r is the maximum growth rate (d−1) and Nt0 is the cell count at t0 (cell mL−1).
Intracellular macromolecular composition
Cellular carbon (C), nitrogen (N), phosphorous (P) and chlorophyll content (total chlorophyll calculated from chlorophyll a and chlorophyll c content) were quantified in the ancestor and each of the evolved lineages. We investigated the effects of temperature on macromolecular composition both over short-term acclimation and long-term evolutionary adaptation. Short-term thermal acclimation was investigated by exposing the evolved lineages to the 15 to 45°C thermal gradient described above for quantifying the thermal tolerance curves for growth rate. Here, lineages were given 1 transfer to acclimate to a given assay temperature in the gradient and then harvested during the logarithmic phase of growth in the second transfer. Long-term evolutionary adaptation to the selection regimes was quantified by harvesting evolved lineages directly from the experiment during the logarithmic phase of growth.
The samples were prepared for C, N, P and chlorophyll measurement by spinning down 50 mL of culture in a centrifuge at 4°C and 3500RPM for 30 minutes. The pellet was then transferred to a 1 mL Eppendorf tube, spun again for 15 minutes at 3500RPM and the remaining supernatant decanted. Chlorophyll content was determined on a spectrophotometer (Jenway 7351) after32, with extraction in 100% methanol, and absorption spectra measured in 10 nm increments from 500 to 700nm, which span the full range of emission peaks for chlorophyll a and chlorophyll c.
Pellets for C and N determination were freeze dried for 24 hours, transferred to zinc capsules, weighed and analyzed using a SerCon Isotope Ratio Mass Spectrometer (CF-IRMS) system (continuous flow mode). Intracellular P content was determined was determined via a colourimetric reaction on a Seal Analytics AA3 segmented flow auto-analyzer. Freeze dried pellets were washed in 0.17M Na2SO4, transferred to scintillation vials and re-suspended in 4 ml 0.017M MnSO4. The samples were transferred to an autoclave (1h, 121°C), shaken vigorously and centrifuged at 2500 r.p.m for 30 minutes, the pellet discarded, and the supernatant brought to 10 ml with MillliQ purified water. The samples were immediately analyzed on the AA3 using the colourimetric molybdate/antimony method33. Using these data we calculated the molar stoichiometric ratios, C:N, C:P, N:P and the Chl:C ratio. Nitrogen-use efficiency (PNUE) and phosphorous-use efficiency (PPUE) were characterized as the ratio of growth rate to cellular nutrient concentration in pg.
Silica content, although not a main trait under investigation here, was determined through preparing pellets as described above. Then, 4ml of 0.2M NaOH were added to the pellet in polyethylene tubes and the pellets vortexed briefly. The samples were then transferred to a heating block at 90°C for ca. 1h. 2 ml of 1M HCl were added, the samples vortexed again and then spun at 3500 rpm for 1h. The supernatant was passed through an 0.5μm filter, pipetted into fresh tubes, topped up to 10ml with MiliQ water and used immediately on the segmented flow auto-analyser, where the colourimetric procedure followed that described in34.
Photochemistry
We characterised a range of photochemical parameters in the ancestor and each of the evolved lineages using fast repetition rate fluorometry (FastPro8, FRRf3, Fast Ocean System Chelsea Technology Group). 500 μL of dilute sample (with a cell count of less than 1000 cells mL−1) were added to 5 ml of fresh culture medium. Samples were then pre-incubated in the dark at the assay temperature for 15 minutes in a water-bath, and another 10 minutes in the fluorometer to make sure that samples were fully dark acclimated and all reaction centers closed. All measurements were made at the selection temperature for the 22, 26, and 32°C lineages, while for the lineages in the fluctuating regime measurements were carried out at both 22 and 32°C. Photochemical traits were measured in response to rapid flashes at increasing light intensities from 0 to1600 μmol m−2 s−1. Flash frequency and rate followed standard protocols for phytoplankton35, with 100 flashes of 1.1 μs at 1 μs intervals. Peak emission wavelengths of the LEDs used for excitations were at 450 nm, 530 nm, and 624 nm. ΦPSII was particularly relevant to our study as it is commonly used to describe the light responses of photosynthetic efficiency. ΦPSII values are used as an indication of the proportion of the total light absorbed that is used in photochemical reactions in PSII (see also Table S10 and Figure S3). Additionally, we determined rP, NPQ, and C (see Fig. 4B and SI Table 13 for PCA on phenotypic traits). These describe the relative rate of photosynthesis in response to irradiance and are obtained as an estimate of electron transport through PSII (rP), the cell’s ability to maintain photochemical function at high light intensities (NPQ, non photochemical quenching) and the proportion of PSII reaction centers in a closed state (C). As measurements were carried out following a light response curve, we were then able to measure these functions both at saturating light intensity and at the light intensity that the samples were grown at in the incubators.
Thermal responses of photosynthesis and respiration
Measurements of photosynthesis and respiration were made on the ancestor and all evolved lineages after 300 generations of selection when in the middle of the logarithmic phase of population growth. Net photosynthesis (NP) was measured as O2 evolution at increasing light intensities in intervals of 50 μmol−1 m−2 s−1 up to 300 μmol−1 m−2 s−1, and then in intervals of 100μmol−1 m−2 s−1 up to 1000 μmol−1 m−2 s−1, followed by 200 μmol steps up to 2000 μmol−1 m−2 s −1. The maximum rate of light saturated photosynthesis was determined by fitting the NP data to a dynamic model of photoinhibition via non-linear least squares regression using the methods outlined in18: where NP(I) is the rate of photosynthesis at light intensity I, NPmax is the maximum rate of net photosynthesis at the optimal light intensity, Iopt, and α controls the rate at which NP(I) increases up to NPmax and R is the rate of respiration (i.e. the rate of O2 flux when I = 0). Gross photosynthesis (P) was estimated as P = NPmax + R. Measurements of O2 flux were measured in a Clark-type oxygen electrode (Hansatech Ltd, King’s Lynn UK Chlorolab2). Aliquots (50 mL) of the populations were concentrated through centrifugation to a density of approximately 8*104 cells mL−1 and acclimatised to the assay temperature for 15 minutes in the dark before measuring metabolic rates. The thermal responses of P and R were quantified by measuring rates over a temperature gradient from 4°C to 45°C (in 3°C increments, with additional measurements at the selection temperatures). Rates of P and R were expressed in units of μg C per μg C using the following equation 36. where b is the metabolic rate (either P or R), 32 * M * (12/44) is used to convert μmol O2 into μg C, and the factor M, is the assimilation quotient of CO2:O2. In our study, NO3- was the source of nitrogen, thus assuming a set of balanced growth equations36: if the C:N ratio is calculated in moles the assimilation ratio of CO2:O2 will be n/n+2 (see36). The estimated M values are given in Table S1 and ranged from ∼ 0.77 to ∼ 0.79. Alongside the biomass measurements for phytoplankton, we also periodically measured bacterial biomass (see methods detailed above), which typically comprised <1% of algal biomass. Consequently, bacterial respiration did not contribute significantly to metabolic rates, and when 2μm filtered samples were run for metabolic rate measurements, the slopes were indistinguishable from water blanks.
DNA extraction for Whole Genome Re-Sequencing
At the beginning (t000) and after 300 generations (t300) of selection in each environment, 250ml of each sample in exponential growth were spun to a pellet in 50ml batches at 4°C and 3500 RPM (rounds per minute). The supernatant was discarded and the pellet stored at −20°C until further use. Prior to DNA-extraction, samples were put through three cycles of thawing and refreezing, as this had shown to increase yield of diatom DNA relative to bacterial DNA in pilot studies. DNA was extracted following a standard cTAB protocol in chloroform:isoamyl and isopropanol (Murray and Thompson, 1980) with a proteinase K step to digest proteins, and an RNAse step to digest RNA. The resulting DNA pellet was quality controlled on 1% agarose gels and fluorometrically on a Qbit system. The extracted genomic DNA was then frozen in TE buffer at −20°C until samples were processed at the Exeter Sequencing Services (University of Exeter, UK). Whole genome re-sequencing (Illumina) was carried out on a HiSeq 2500 platform with paired-end reads. Library preparation was carried out at the Exeter Sequencing Services using Netflex. The resulting reads were assembled against the existing T. pseudonana genome 11. Quality control filtering of sequence data
We generated paired 125-bp reads from genomic DNA from 61 samples (plus three negative controls containing no diatoms) using the Illumina HiSeq. Reads were trimmed and filtered to remove poor quality data using Trimgalore (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) with “–q 30”.
Variant calling: single-nucleotide polymorphisms (SNPs)
After trimming and filtering, remaining sequence reads were then aligned against version 2 of the reference T. pseudonana genome sequence 1,2 (GenBank: GCA_000149405.2) using BWA-mem version 0.7.5a-2 with default settings 38. Alignment statistics were calculated using Qualimap 39–41.This resulted in average aligned sequence depths from 3.4 to 74.5 fold; the average alignment depth for the t0 (ancestor) sample was 18.5 fold. This resulted in a set of 64 BAM-formatted 41 files (see also Table S15).
Each BAM file was converted to mpileup format using SAMtools 41 version 0.1.19-1. The mpileup file contains, for each position in the alignment, the frequency of each base (allele) among the sequence reads aligned at that position; in other words it offers an estimate of the frequencies of each allele among the sequenced population. We considered only those genomic positions that in the t0 (ancestor) sample had a single base (i.e. allele) with a 100 % frequency; in other words we excluded from consideration genomic sites where there was ambiguity about the genetic state in the ancestor population. SNP sites falling within protein-coding exons were classified as synonymous or non-synonymous using the SNPeff tool 42.
We combined the allele frequencies from each mpileup file into a single matrix, which was used for the downstream statistical analyses implemented in R using the packages vegan (2.4-3), mixOmics (6.1-3), and phangorn (2.2-0). Specifically, we created distance matrices on which we ran PERMANOVAs to test for separation of samples (on phenotypic data and SNPs) by treatment. Pairwise contrasts between treatments were examined in PERMDISP followed by TukeyHSD post-hoc tests. Trees were built within the phangorn package through neighbour-joining, and annotation of trees was carried out in the FigTree software (FIGTREE 1.4.3 http://tree.bio.ed.ac.uk/software/figtree/).
To visualise treatment associated PCA-loadings, we used the plotLoadings functions within the mixOmics package, which performs discriminant PCA and then represent the loading weight of each selected variable on each component, i.e. the length of the bar is directly correlated to the location of the loadings on the PCA, with the longest bar indicating the strongest association to a treatment group. This allows us to find treatment associated groups in which the changes in the selected gene or SNPs are maximal 1,2.
Statistical Analyses
Fitness trajectories
The resultant time series of specific growth rates were analyzed using a generalised additive mixed effects model (GAMM) to assess whether the fitness trajectories differed between the selection regimes. We used GAMMs to account for the hierarchical nature of our experimental data. For example, our experimental design yielded replicate fitness trajectories in each selection treatment. This hierarchical structure meant that measurements were non-independent – e.g. measurements from the same replicate will be auto correlated. We account for this by treating replicate as a random effect on the intercept of the model, which models deviations among replicates from the fixed effects as normally distributed with a mean of zero. The most complex models included an effect of treatment (e.g. ‘22’, ‘26’, ‘32’, ‘22-32’) on the intercept (which characterises the median value of the response variable) and also allowed the shape of the time series, which was modeled using a cubic regression spline, to vary among treatments. Treatment effects on the shape and intercept of the seasonal phenology were modeled as fixed effects in the GAMMs. Model selection entailed fitting a range of models to the data, starting with the full model and then a series of reduced models with interaction terms and main effects removed to test hypotheses about the potential differences in the fitness trajectories among treatments. For multi-model selection we computed small sample-size corrected AIC scores (AICc) and then compared between models by calculating delta AICc values and AIC weights using the ‘MuMIn’ package. GAMMs were fitted to the data using the ‘gamm4’ package and were conducted in R (v.3.23).
Thermal responses of growth and metabolism
The thermal responses for growth, photosynthesis and respiration were quantified using a modified version of the Sharpe-Schoolfield equation (see1,2 for the original equations), which assumes that the rate of growth or metabolism is limited by single enzyme atalyzed reaction where b(T), is the rate of metabolism (in μg C μg C−1 d−1) or growth (d−1), k is Boltzmann’s constant (8.62×10−5 eV K−1), Ea is the activation energy (in eV) for the metabolic process, indicative of the steepness of the slope leading to a thermal optimum, T is temperature in Kelvin (K), Eh characterises temperature-induced inactivation of enzyme kinetics above Th where half the enzymes are rendered non-functional and b(Tc) is the rate of metabolism normalised to an arbitrary reference temperature, here Tc = 18°C, where no low or high temperature inactivation is experienced. Equation (6) yields a maximum metabolic rate at an optimum temperature, where metabolic rates are fastest:
Equation (6) differs from the Sharpe-Schoolfield equations in a number of ways. First, we exclude parameters from Eq. (6) used to characterise low-temperature inactivation due to insufficient data to quantify this phenomenon in our analysis. Second, rather than characterise temperature effects below Topt using the Eyring (1935) relation,, we instead use the simpler Boltzmann factor,. This simplification enables an explicit solution for Topt(Eq. 8) and facilitates more direct comparison with previous work on the temperature dependence of metabolism using metabolic theory47–50.
The parameters b(Tc), Ea, Eh, Th, and Topt, represent traits that characterise a unimodal thermal response curve, and we expect them to differ between selection regimes owing to thermal adaptation. To test this hypothesis, we fitted the rate data to Eq. (6) using non-linear mixed effects models in the ‘nlme’ package in R. We analyzed the photosynthesis, respiration and growth data in separate mixed effects models. We also analyzed the thermal responses for ancestor separately from the data for the evolved lineages following the selection experiment. Models included random effects on each of the parameters of Eq. (6) by replicate, and for the analysis with evolved lineages, ‘selection environment’ as a fixed four level factor on each parameter. For the analysis of the metabolism data for the ancestor we included both photosynthesis and respiration data together, with ‘flux’ as a two-level fixed factor on each of the parameters in Eq. (6), to establish whether the thermal responses differed between these fluxes and likely physiological constraints on growth prior to the selection experiment. Model fitting and selection started with the most complex possible model, including fixed and random effects on all parameters. Model selection then proceeded by first removing treatment effects on the parameters individually, then in pairs of two and in pairs of three, and finally, by removing the treatment-effect all four parameters. For multi-model selection we computed small sample-size corrected AIC scores (AICc) and then compared between models by calculating delta AICc values and AIC weights using the ‘MuMIn’ package. When candidate models deviated from the most parsimonious model (that with the lowest AICc score) by less than two AICc units, parameters were averaged across those candidate models. The relative importance of the fixed factors in the averaged model was determined using the sum of their relative weights (see Table S5 and S6).
Carbon-use efficiency (CUE)
Carbon use efficiency, as the potential for carbon allocation to growth, was calculated from the gross photosynthesis (P) and respiration (R) data as CUE = 1 – R/P. For statistical analysis, we used CUE calculated at the temperature of the selection environment. In the fluctuating treatment, we used data for photosynthesis and respiration at 32°C to aid comparison with the populations experiencing 32°C throughout. We fitted a linear mixed effects model to these data, with ‘selection regime’ as a fixed effect and biological replicate as a random effect. Model selection proceeded as described above.
Macromolecular composition and photosynthetic efficiency
The C, N and P content, the C:N, C:P, N:P, Chl:C ratio, size and silicate content, as well as ΦPSII at high and low light, were each analyzed using a separate mixed effects model, where ‘selection regime’ was a fixed effect and replicate was a random effect on the intercept (see Table S11 for details). Model selection was as described above.
We quantified the light response curve of ΦPSII using an exponential decay model where I is the irradiance, a is a normalization constant, and b is the rate constant which characterises how rapidly ΦPSII declines with increasing I. Eq. (8) was fitted to the ΦPSII data using a non-linear mixed model including random effects by biological replicate on each parameter and ‘selection environment’ as a fixed factor. Model output was then used to calculate ΦPSII at the light intensity that samples would have experienced in the incubators (i.e. ∼ 100μmol quanta m−2 s−1 see Fig. 4 G). Model selection then proceeded as described above for P and R data. Detailed model output and model selection data are available in Table S10 where nomenclature and acronyms are as given here (see also Figure S4).
Acknowledgments
This study was funded by a Leverhulme Trust research grant (RPG-2013-335). Whole genome re-sequencing was carried out at Exeter Sequencing Service and Computational core facilities at the University of Exeter, where Dr. Karen Moore, Dr. Audrey Farbos, Paul O’Neill and Dr. Konrad Paszkiewicz lead the handling of the samples. Exeter Squencing Services are supported by Medical Research Council Clinical Infrastructure award (MR/M008924/1), Wellcome Trust Institutional Strategic Support Fund (WT097835MF), Wellcome Trust Multi User Equipment Award (WT101650MA) and BBSRC LOLA award (BB/K003240/1).
Author contributions
GYD conceived and designed the study, analysed data and wrote the manuscript. ES designed and carried out the experimental work, analysed data and wrote the manuscript. AB conceived and designed the study. DS carried out the bioinformatic analyses of the genomic data. All authors contributed to writing the manuscript.
Author information
The authors declare no competing financial interests. Correspondence and requests for materials should be directed to ES or GYD.