Multivariate trait analysis reveals diatom plasticity constrained to a reduced set of biological axes
Culture maintenance and growthTwelve strains of Thalassiosira spp. were obtained from the Provasoli-Guillard National Centre of Marine Phytoplankton (NCMA, https://ncma.bigelow.org/), and one strain from the Australian National Culture Collection, representing 7 species in total (Supplementary Table 1). Cultures were maintained in polystyrene tissue culture flasks in artificial seawater with f/2 media [37] at 20 °C, with 60 µmolm−2s−1 of light on a 12:12 light cycle.Three strains originally identified as Thalassiosira sp. in the NCMA collection were further classified to the species level using sequencing of the ITS2 gene region (Supplementary Table 1): CCMP1055 as T. auguste-lineata (84.64% similarity; [38]) and CCMP2929 as T. weisflogii (98.37% similarity to Strain 1587 used in our study; [39]). Strain CCMP1059 was tentatively identified as Cyclotella striata (94.17% identity match to clone ZX28-3-40; [40]) also from order Thalassiosirales, but this assignment requires further investigation.Experimental set upExperimental cultures (200 mL) were grown in 250 mL polystyrene tissue culture flasks in triplicate, at a starting concentration of 2500 cells ml−1. All 13 strains were grown in a “standard” environment (identical to maintenance conditions) with 9 phenotypic traits measured to describe the initial trait-scape. Five strains (1010, 1059, 2929, 3264, and 3367) were grown in two additional environments in triplicate: a high temperature and light treatment (HT: 30 °C, 200 µmol photons m−2s−1 of light, 12:12 light:dark), and a low nutrient treatment (LN: f/400 media with an adjusted N:P ratio of 10:1 achieved by reducing the nitrate concentration from 4.4 to 1.8 µM, 60 µmol photons m−2s−1 of light, 12:12 light:dark). Cultures for the two additional treatments were inoculated with 10,000 cells ml−1 (LN) and 5,000 cells ml−1 (HT) in anticipation of limited growth.Growth was tracked daily using in vivo fluorescence as a proxy for cell density [41]. One mL aliquots of experimental cultures were measured for chlorophyll-a fluorescence using a plate reader (TECAN Infinite M1000 Pro, Männedorf, Switzerland) using 455/680 nm excitation/emission spectra. Phenotypic traits were measured at mid-late exponential phase, assessed by visually examining in vivo fluorescence growth curves. In the case of the low nutrient treatment, where growth was limited to 3–5 days, cultures were harvested in early stationary phase. Duration of growth for each experiment is summarised in Supplementary Table 2.Phenotypic trait measurement methodsPhenotypic traits were selected to capture different commonly measured base physiological functions, and to include traits that are used in biogeochemical models. We also selected traits that demonstrated independence and orthogonality (i.e., not all co-varying), based on pilot studies, in order to successfully define the multivariate trait-scape [42].Growth rateGrowth rates for each time step were calculated from the daily in vivo fluorescence measurements according to the calculation:$$mu = frac{{{{{{{{{mathrm{ln}}}}}}}}left( {F_2} right)-{{{{{{{mathrm{ln}}}}}}}}left( {F_1} right)}}{{t_2 – t_1}}$$Maximum growth rates were determined by the average growth over 2–4 consecutive steps depending on the duration of exponential growth.Flow cytometry traitsFor flow cytometry trait measures (growth rate, size, chlorophyll a content, lipid content), 1 mL aliquots of experimental culture were fixed with EM grade paraformaldehyde (0.8% final concentration, Electron Microscopy Sciences, Ft Washington, PA) in 1.6 mL cryopreservation tubes (CryoPure, Sarstedt), frozen in liquid nitrogen, then stored at −80 °C prior to analysis. All measures were performed using a Cytoflex LX (Beckman Coulter, CA, USA).Cell counts and sizeCell counts were done by gating the diatom population using chlorophyll a (488 nm excitation, 690/50 nm detector) and forward scatter channel thresholds. Cell size was estimated using forward scatter values calibrated against spherical beads (2, 4, 6, 10, 15 µM diameters; Invitrogen, CA). This resulted in a conversion equation of equivalent spherical diameter (ESD) = (FSC + 194636)/75775, which was used to assess relative changes in cell size [43].Chlorophyll a contentChlorophyll a (Chl-a) fluorescence of the gated diatom population was quantified using 488 nm excitation, 690/50 nm detection. A standard bead (Cytoflex Daily QC Fluorospheres; Beckman Coulter) was used to calibrate the performance of the instrument and ensure comparable measures across samples. Chlorophyll values were divided by ESD to account for cell size differences.Side scatter/granularitySide scatter is an indicator of the internal complexity of a cell or “granularity”. This trait is measured in tandem with other flow cytometry measures and was included as a phenotypic trait. The interpretation of this trait is not straight forward, but is independent of other flow cytometry traits measured and has been used in other flow cytometry studies of microalgae [44]. This trait was divided by ESD to account for cell size differences.Neutral lipidsRelative neutral lipid content was determined using the fluorescent stain BODIPY™ 505/515 (Thermo Fisher, MA, USA) which is commonly used to assess neutral lipid content in phytoplankton [45,46,47]. Background fluorescence (488 nm excitation, 525/40 nm detector) of PFA-fixed cells was measured in tandem with the size, chlorophyll a, and side scatter. After this, 10 µL of BODIPY stain (2 mg mL−1 in DMSO) was added to each sample, resulting in a final BODIPY concentration of 2 μg mL−1. Samples were incubated for 10 min in the dark before being read again on the flow cytometer. Neutral lipid content was defined as the difference in median fluorescence per cell between the pre- and post-stained sample. This value was then divided by the ESD size to account for size-related effects.Photophysiological traitsPhotophysiological measures were taken by conducting a rapid light curve [48] with a water PAM (Water-PAM; Walz GmbH, Effeltrich, Germany) using 1 mL of experimental culture diluted in artificial seawater. The rapid light curve protocol exposes the culture to 8 steps of increasing irradiance for 10 seconds each, measuring the photophysiological response at each step. Maximum electron transport rate (ETRmax), Ik (half saturation irradience), and alpha (the photosynthetic rate during the light-limited linear region) were calculated using the regression fit function in the PAM WinControl software. Photophysiology measurements were taken between 4–5 h after the start of the photoperiod.Reactive oxygen speciesThe development of reactive oxygen species (ROS) was measured using the fluorescent probe 2’,7’-dichlorodihydrofluorescein diacetate (H2DCFDA; Thermo Fisher, MA, USA) which has been used in a number of phytoplankton studies [49,50,51]. Two 1 mL aliquots of experimental culture were transferred to a 48 well tissue culture plate; 2 µL of stain (2.5 mg mL−1 H2DCFDA was made in DMSO) was added to one aliquot, with the other acting as a blank. The plates were sealed (Breathe-Easy, Diversified Biotech) and incubated in the dark at growth temperature (20 or 30 °C) for 2 h. Incubation was done in the dark because of the effects of light on the dye itself, therefore the effects of the excess light treatment were not captured in this trait. Fluorescence of H2DCFDA was read using a plate reader with 488 nm excitation 525 nm emission (TECAN Infinite M1000 Pro, Männedorf, Switzerland). ROS concentration was estimated as the difference in fluorescence units per cell between the stained and unstained aliquots of each culture. This metric was also divided by ESD size to account for size effects.Taxonomic confirmation of strainsDNA from stock cultures (10 mL) was extracted using a DNeasy PowerSoil kit (QIAGEN Inc., CA, USA) and checked for quality with a NanopDrop™ 2000 (ThermoFIsher Scientific, MA, USA), before amplification and sequencing at the Australian Genome Research Facility (AGRF, Sydney, Australia). PCR conditions and primers used were those developed by Chappell et al. [52] for the ITS region: forward primer: 5ʹ-RCGAAYTGCAGAACCTCG-3ʹ, reverse primer: 5ʹ-TACTYAATCTGAGATYCA-3ʹ.Bioinformatics processing was conducted using Geneious Prime (Version 2020.0.5; Biomatters Ltd.). Strain sequences were compared to GenBank using the BLAST function to confirm species identity. Nucleotide sequences were aligned using the MUSCLE alignment [53], followed by Bayesian inference analysis using MrBayes [54] to generate a phylogenetic tree. The out-group for the tree was a strain of Chaetoceros atlanticus isolate TPV2 1146 obtained from GenBank. Percentage similarity between strains according to the alignment was used as a metric of genetic relatedness.Statistical analysisWe assessed the multivariate phenotypes for the Thalassiosira strains using principal component analysis (PCA). The input variables were the 9 independent trait measurements made on each replicate culture (n = 36, 3 biological replicates per strain). Trait data was standardized (mean = 0, SD = 1) for each trait prior to PCA analysis to account for differences in the units and scale of measurements. The resulting PCA plot was defined as the ‘trait-scape’.Hierarchical clustering analysis was performed on the 9-trait dataset used to assess similarity in multivariate phenotypes between each replicate for each strain (n = 3 per strain).To compare genetic vs. phenotypic similarity, percentage similarity between strains was correlated against the distance between strain centroids (multivariate means) within the trait-scape. Distances between multivariate means (centroids) were calculated using the equation:$${{{{{{{mathrm{distance}}}}}}}} = sqrt {left( {{{{{{{{mathrm{{Delta}}}}}}}PC}}1.{{{{{{{mathrm{a}}}}}}}}} right)^2,+,left( {{{{{{{{mathrm{{Delta}}}}}}}PC}}2.{{{{{{{mathrm{b}}}}}}}}} right)^2}$$ΔPC1 is the difference in PC1 co-ordinates between the two strains, a is the % variance explained by PC1, ΔPC2 is the difference in PC2 co-ordinates between the two strains, b is the % variance explained by PC2.To assess whether a trait-scape generated using fewer input traits (4 rather than 9) was representative of the full, 9-trait plot, we conducted PCA using 4 input traits, and then assessed whether the inter-strain distances (distances between centroids) within the plot were correlated using linear regression. This provided a quantitative assessment of whether the strains were in the same relative positions to each other within the trait-scape.Covariation of traitsTo compare the pairwise relationships between traits across the strains, correlation matrices were made using data collected in the standard environment, and for the HT and LN environments.Phenotypic plasticityThe change in phenotypes in the new environments were assessed firstly by conducting PCA on the full dataset, including trait data from the 13 strains grown in the standard environment, plus the 5 strains grown in the two additional environments. This generated an “expanded trait-scape”. In addition, correlation matrices were generated for the new environments’ trait dataset to assess differences in trait-trait relationships between the ‘standard’ and “expanded” datasets.Relative changes in trait values for each trait in the new environments were calculated as follows:$$ {{{{{{{mathrm{Relative}}}}}}}},{{{{{{{mathrm{change}}}}}}}} \ = frac{{{{{{{{{mathrm{trait}}}}}}}},{{{{{{{mathrm{value}}}}}}}},{{{{{{{mathrm{new}}}}}}}},{{{{{{{mathrm{environment}}}}}}}} – overline {{{{{{{mathrm{x}}}}}}}} ,,{{{{{{{mathrm{trait}}}}}}}},{{{{{{{mathrm{value}}}}}}}},{{{{{{{mathrm{standard}}}}}}}},{{{{{{{mathrm{environment}}}}}}}}}}{{overline {{{{{{{mathrm{x}}}}}}}} ,,{{{{{{{mathrm{trait}}}}}}}},{{{{{{{mathrm{value}}}}}}}},{{{{{{{mathrm{standard}}}}}}}},{{{{{{{mathrm{environment}}}}}}}}}}$$We used PCA to assess whether the relative changes in trait values were consistent between strains in the two different environments. i.e., was the relative change in whole phenotype consistent. If the changes were consistent across strains, we expected to see clustering in the PCA based on treatment.Statistical softwareStatistical analyses were performed in R [55], Matlab, and Microsoft Excel. Hierarchical clustering analysis with multiscale bootstrap resampling (1000 replicates) on trait values from biological replicates was done with the ‘pvclust’ package in R [56] using Euclidean distance and the average (UPGMA) method. Principal component analysis was used to generate the multivariate trait-scape was done using the “vegan package” in R [57]. The contributions of each trait to the PC axes (loadings) were extracted using the “factoextra” package in R [58]. Trait correlation matrices were generated using the “corrplot” package in R [59]. More
