Uni-algal, non-axenic cultures of Heterosigma akashiwo (CCMP2393) were grown in L1 medium (without silicate) made with a Long Island Sound seawater base collected from Avery Point, CT, USA (salinity 32) at 18°C with a 14:10 (light:dark) cycle with an irradiance of approximately 100 µmol m-2 s-1 . Cells were acclimated in exponential growth phase to different carbonate chemistries in 1.2 L of L1 media in 2.5-L polycarbonate bottles. To control the carbonate chemistry of the water, the headspace of each bottle was purged continuously with a custom gas mixture of ~21% oxygen, ~79% nitrogen and either 200, 400, 600, 800 or 1000 ppmv CO2 (TechAir, NY).

At the point of harvest, 150 mL (~6 x 106 cells) were filtered on to 5 µm pore size, 25 mm polycarbonate filter and flash frozen in liquid nitrogen. Genetic material from samples was extracted with the RNeasy Mini kit (Qiagen, Valencia, CA) and DNA was removed on-column using the RNase-free DNase Set (Qiagen), yielding total RNA. Total RNA extracts of the triplicate cultures were quantified on a 2100 Bioanalyzer (Agilent, Santa Clara, CA). Libraries were prepared using poly-A pull down with the TruSeq Stranded mRNA Library Prep kit (Illumina, San Diego, CA). Library preparation, barcoding, and sequencing from each library was performed by the JP Sulzberger Columbia University Genome Center (New York, NY).

Sequence reads were de-multiplexed and trimmed to remove sequencing barcodes. Reads were aligned using Bowtie2 (Langmead and Salzberg 2012) to the MMETSP consensus contigs for Heterosigma akashiwo CCMP2393 (https://omictools.com/marine-microbial-eukaryotic-transcriptome-sequencing-project-tool).

Significant differences between physiological parameters by CO2 treatment were assessed with analysis of variance (ANOVA) and Tukey’s honestly significant differences test (aov and TukeyHSD, stats, R). Differential expression of genes in any CO2 treatment compared to modern was determined using the general linear model (GLM) exact test (edgeR, R). Briefly, the read counts were normalized by trimmed mean of M-values (TMM) using the function calcNormFactors, tagwise dispersions were calculated with the function estimateGLMTagwiseDisp, a GLM was fit using glmFit, and log2 fold change (logFC) for each treatment was calculated relative to average expression at modern CO2. P-values from likelihood ratio tests were corrected for multiple testing using the false discovery method (fdr).